Merge pull request #29557 from fengyuentau:4x/hal_rvv/split_opt

hal_rvv core: Cover more data types for split with native RVV intrinsics #29557 Covering 16-bit, 32-bit and 64-bit split. ### Pull Request Readiness Checklist See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request - [x] I agree to contribute to the project under Apache 2 License. - [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV - [x] The PR is proposed to the proper branch - [x] There is a reference to the original bug report and related work - [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable Patch to opencv_extra has the same branch name. - [x] The feature is well documented and sample code can be built with the project CMake
Merge pull request #29566 from fengyuentau:4x/hal_rvv/copy_mask_opt
2026-07-22 03:43:03 +04:00 · 2026-07-21 10:16:38 +03:00 · 2026-07-21 10:15:07 +03:00 · 2026-07-20 23:43:41 +08:00 · 2026-07-20 12:26:12 +03:00 · 2026-07-20 00:22:38 +08:00
3332 changed files with 1277985 additions and 290499 deletions
@@ -1,4 +1,4 @@
-name: 5.x
+name: 4.x

 on:
  schedule:
@@ -9,5 +9,6 @@ jobs:
  CodeQL:
    uses: opencv/ci-gha-workflow/.github/workflows/OCV-CodeQL.yaml@main
    with:
-      target_branch: '5.x'
+      target_branch: '4.x'
      workflow_branch: main
+
@@ -0,0 +1,61 @@
+name: PR:4.x
+
+on:
+  pull_request:
+    branches:
+      - 4.x
+
+jobs:
+
+  Linux:
+    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-Linux.yaml@main
+    with:
+      workflow_branch: main
+
+  Linux-no-HAL:
+    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-Linux-NoHAL.yaml@main
+
+  Linux-Apline:
+    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-Linux-Alpine.yaml@main
+
+  Windows:
+    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-Windows.yaml@main
+    with:
+      workflow_branch: main
+
+  Ubuntu2404-ARM64:
+    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-4.x-ARM64.yaml@main
+
+  Ubuntu2404-ARM64-Debug:
+    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-4.x-ARM64-Debug.yaml@main
+
+  Ubuntu2004-x64-OpenVINO:
+    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-4.x-U20-OpenVINO.yaml@main
+
+  Ubuntu2004-x64-CUDA:
+    if: "${{ contains(github.event.pull_request.labels.*.name, 'category: dnn') }} || ${{ contains(github.event.pull_request.labels.*.name, 'category: dnn (onnx)') }}"
+    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-4.x-U20-Cuda.yaml@main
+
+  macOS-ARM64:
+    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-4.x-macOS-ARM64.yaml@main
+
+  macOS-x64:
+    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-4.x-macOS-x86_64.yaml@main
+
+  macOS-ARM64-Vulkan:
+    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-4.x-macOS-ARM64-Vulkan.yaml@main
+
+  iOS:
+    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-4.x-iOS.yaml@main
+
+  Android-SDK:
+    uses: opencv/ci-gha-workflow/.github/workflows/OCV-4.x-Android-SDK.yaml@main
+
+  TIM-VX:
+    uses: opencv/ci-gha-workflow/.github/workflows/OCV-timvx-backend-tests-4.x.yml@main
+
+  docs:
+    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-4.x-docs.yaml@main
+
+  Linux-RISC-V-Clang:
+    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-4.x-RISCV.yaml@main
@@ -1,62 +0,0 @@
-name: PR:5.x
-
-on:
-  pull_request:
-    branches:
-      - 5.x
-
-jobs:
-  Linux:
-    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-Linux.yaml@main
-    with:
-      workflow_branch: 'main'
-
-  Linux-no-HAL:
-    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-Linux-NoHAL.yaml@main
-
-  Windows:
-    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-Windows.yaml@main
-    with:
-      workflow_branch: main
-
-  Ubuntu2404-ARM64:
-    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-5.x-ARM64.yaml@main
-
-  Ubuntu2404-ARM64-Debug:
-    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-5.x-ARM64-Debug.yaml@main
-
-  Ubuntu2004-x64-OpenVINO:
-    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-5.x-U20-OpenVINO.yaml@main
-
-  Ubuntu2004-x64-CUDA:
-    if: "${{ contains(github.event.pull_request.labels.*.name, 'category: dnn') }} || ${{ contains(github.event.pull_request.labels.*.name, 'category: dnn (onnx)') }}"
-    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-5.x-U20-Cuda.yaml@main
-
-# Vulkan configuration disabled as Vulkan backend for DNN does not support int/int64 for now
-# Details: https://github.com/opencv/opencv/issues/25110
-#  Windows10-x64-Vulkan:
-#    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-5.x-W10-Vulkan.yaml@main
-
-  macOS-ARM64:
-    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-5.x-macOS-ARM64.yaml@main
-
-#  macOS-ARM64-Vulkan:
-#    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-5.x-macOS-ARM64-Vulkan.yaml@main
-
-  macOS-x64:
-    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-5.x-macOS-x86_64.yaml@main
-
-  iOS:
-    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-5.x-iOS.yaml@main
-
-  Android:
-    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-5.x-Android.yaml@main
-
-  TIM-VX:
-    uses: opencv/ci-gha-workflow/.github/workflows/OCV-timvx-backend-tests-4.x.yml@main
-
-  docs:
-    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-5.x-docs.yaml@main
-
-  Linux-RISC-V-Clang:
-    uses: opencv/ci-gha-workflow/.github/workflows/OCV-PR-5.x-RISCV.yaml@main
@@ -1,49 +0,0 @@
-# ----------------------------------------------------------------------------
-#  CMake file for opencv_lapack. See root CMakeLists.txt
-#
-# ----------------------------------------------------------------------------
-project(clapack)
-
-# TODO: extract it from sources somehow
-set(CLAPACK_VERSION "3.9.0" PARENT_SCOPE)
-
-include_directories("${CMAKE_CURRENT_SOURCE_DIR}/include")
-
-# The .cpp files:
-file(GLOB lapack_srcs src/*.c)
-file(GLOB runtime_srcs runtime/*.c)
-file(GLOB lib_hdrs include/*.h)
-
-# ----------------------------------------------------------------------------------
-# 				Define the library target:
-# ----------------------------------------------------------------------------------
-
-set(the_target "libclapack")
-
-add_library(${the_target} STATIC ${lapack_srcs} ${runtime_srcs} ${lib_hdrs})
-
-ocv_warnings_disable(CMAKE_C_FLAGS -Wno-parentheses -Wno-uninitialized -Wno-array-bounds
-    -Wno-implicit-function-declaration -Wno-unused -Wunused-parameter -Wstringop-truncation
-    -Wtautological-negation-compare) # gcc/clang warnings
-ocv_warnings_disable(CMAKE_C_FLAGS /wd4244 /wd4554 /wd4723 /wd4819) # visual studio warnings
-
-set_target_properties(${the_target}
-  PROPERTIES OUTPUT_NAME ${the_target}
-  DEBUG_POSTFIX "${OPENCV_DEBUG_POSTFIX}"
-  COMPILE_PDB_NAME ${the_target}
-  COMPILE_PDB_NAME_DEBUG "${the_target}${OPENCV_DEBUG_POSTFIX}"
-  ARCHIVE_OUTPUT_DIRECTORY ${3P_LIBRARY_OUTPUT_PATH}
-  )
-
-set(CLAPACK_INCLUDE_DIR "${CMAKE_CURRENT_SOURCE_DIR}/include" PARENT_SCOPE)
-set(CLAPACK_LIBRARIES ${the_target} PARENT_SCOPE)
-
-if(ENABLE_SOLUTION_FOLDERS)
-  set_target_properties(${the_target} PROPERTIES FOLDER "3rdparty")
-endif()
-
-if(NOT BUILD_SHARED_LIBS)
-  ocv_install_target(${the_target} EXPORT OpenCVModules ARCHIVE DESTINATION ${OPENCV_3P_LIB_INSTALL_PATH} COMPONENT dev)
-endif()
-
-ocv_install_3rdparty_licenses(clapack lapack_LICENSE)
@@ -1,102 +0,0 @@
-#ifndef __CBLAS_H__
-#define __CBLAS_H__
-
-/* most of the stuff is in lapacke.h */
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-typedef struct lapack_complex
-{
-    float r, i;
-} lapack_complex;
-
-typedef struct lapack_doublecomplex
-{
-    double r, i;
-} lapack_doublecomplex;
-
-typedef enum {CblasRowMajor=101, CblasColMajor=102} CBLAS_LAYOUT;
-typedef enum {CblasNoTrans=111, CblasTrans=112, CblasConjTrans=113} CBLAS_TRANSPOSE;
-
-void cblas_xerbla(const CBLAS_LAYOUT layout, int info,
-                  const char *rout, const char *form, ...);
-
-void cblas_sgemm(CBLAS_LAYOUT layout, CBLAS_TRANSPOSE TransA,
-                 CBLAS_TRANSPOSE TransB, const int M, const int N,
-                 const int K, const float alpha, const float *A,
-                 const int lda, const float *B, const int ldb,
-                 const float beta, float *C, const int ldc);
-
-void cblas_dgemm(CBLAS_LAYOUT layout, CBLAS_TRANSPOSE TransA,
-              CBLAS_TRANSPOSE TransB, const int M, const int N,
-              const int K, const double alpha, const double *A,
-              const int lda, const double *B, const int ldb,
-              const double beta, double *C, const int ldc);
-
-void cblas_cgemm(CBLAS_LAYOUT layout, CBLAS_TRANSPOSE TransA,
-               CBLAS_TRANSPOSE TransB, const int M, const int N,
-               const int K, const void *alpha, const void *A,
-               const int lda, const void *B, const int ldb,
-               const void *beta, void *C, const int ldc);
-
-void cblas_zgemm(CBLAS_LAYOUT layout, CBLAS_TRANSPOSE TransA,
-                CBLAS_TRANSPOSE TransB, const int M, const int N,
-                const int K, const void *alpha, const void *A,
-                const int lda, const void *B, const int ldb,
-                const void *beta, void *C, const int ldc);
-
-int xerbla_(char *, int *);
-int lsame_(char *, char *);
-double slamch_(char* cmach);
-double slamc3_(float *a, float *b);
-double dlamch_(char* cmach);
-double dlamc3_(double *a, double *b);
-
-int dgels_(char *trans, int *m, int *n, int *nrhs, double *a,
-	int *lda, double *b, int *ldb, double *work, int *lwork, int *info);
-
-int dgesv_(int *n, int *nrhs, double *a, int *lda, int *ipiv,
-	double *b, int *ldb, int *info);
-
-int dgetrf_(int *m, int *n, double *a, int *lda, int *ipiv,
-	int *info);
-
-int dposv_(char *uplo, int *n, int *nrhs, double *a, int *
-	lda, double *b, int *ldb, int *info);
-
-int dpotrf_(char *uplo, int *n, double *a, int *lda, int *
-	info);
-
-int sgels_(char *trans, int *m, int *n, int *nrhs, float *a,
-	int *lda, float *b, int *ldb, float *work, int *lwork, int *info);
-
-int sgeev_(char *jobvl, char *jobvr, int *n, float *a, int *
-	lda, float *wr, float *wi, float *vl, int *ldvl, float *vr, int *
-	ldvr, float *work, int *lwork, int *info);
-
-int sgeqrf_(int *m, int *n, float *a, int *lda, float *tau,
-	float *work, int *lwork, int *info);
-
-int sgesv_(int *n, int *nrhs, float *a, int *lda, int *ipiv,
-	float *b, int *ldb, int *info);
-
-int sgetrf_(int *m, int *n, float *a, int *lda, int *ipiv,
-	int *info);
-
-int sposv_(char *uplo, int *n, int *nrhs, float *a, int *
-	lda, float *b, int *ldb, int *info);
-
-int spotrf_(char *uplo, int *n, float *a, int *lda, int *
-	info);
-
-int sgesdd_(char *jobz, int *m, int *n, float *a, int *lda,
-	float *s, float *u, int *ldu, float *vt, int *ldvt, float *work,
-	int *lwork, int *iwork, int *info);
-
-#ifdef __cplusplus
-}
-#endif
-
-#endif /* __CBLAS_H__ */
@@ -1,129 +0,0 @@
-/* f2c.h  --  Standard Fortran to C header file */
-
-/**  barf  [ba:rf]  2.  "He suggested using FORTRAN, and everybody barfed."
-
-	- From The Shogakukan DICTIONARY OF NEW ENGLISH (Second edition) */
-
-#ifndef __F2C_H__
-#define __F2C_H__
-
-#include <assert.h>
-#include <math.h>
-#include <ctype.h>
-#include <stdlib.h>
-#include <string.h>
-#include <stdio.h>
-
-#include "cblas.h"
-#include "lapack.h"
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-#undef complex
-
-typedef int integer;
-typedef unsigned int uinteger;
-typedef char *address;
-typedef short int shortint;
-typedef float real;
-typedef double doublereal;
-typedef lapack_complex complex;
-typedef lapack_doublecomplex doublecomplex;
-typedef int logical;
-typedef short int shortlogical;
-typedef char logical1;
-typedef char integer1;
-
-#define TRUE_ (1)
-#define FALSE_ (0)
-
-#ifndef abs
-#define abs(x) ((x) >= 0 ? (x) : -(x))
-#endif
-#define dabs(x) (double)abs(x)
-#ifndef min
-#define min(a,b) ((a) <= (b) ? (a) : (b))
-#endif
-#ifndef max
-#define max(a,b) ((a) >= (b) ? (a) : (b))
-#endif
-#define dmin(a,b) (double)min(a,b)
-#define dmax(a,b) (double)max(a,b)
-#define bit_test(a,b)	((a) >> (b) & 1)
-#define bit_clear(a,b)	((a) & ~((uinteger)1 << (b)))
-#define bit_set(a,b)	((a) |  ((uinteger)1 << (b)))
-
-static __inline double r_lg10(float *x)
-{
-    return 0.43429448190325182765*log(*x);
-}
-
-static __inline double d_lg10(double *x)
-{
-    return 0.43429448190325182765*log(*x);
-}
-
-static __inline double d_sign(double *a, double *b)
-{
-    double x = fabs(*a);
-    return *b >= 0 ? x : -x;
-}
-
-static __inline double r_sign(float *a, float *b)
-{
-    double x = fabs((double)*a);
-    return *b >= 0 ? x : -x;
-}
-
-static __inline int i_nint(float *x)
-{
-    return (int)(*x >= 0 ? floor(*x + .5) : -floor(.5 - *x));
-}
-
-int pow_ii(int *ap, int *bp);
-double pow_di(double *ap, int *bp);
-static __inline double pow_ri(float *ap, int *bp)
-{
-    double apd = *ap;
-    return pow_di(&apd, bp);
-}
-static __inline double pow_dd(double *ap, double *bp)
-{
-    return pow(*ap, *bp);
-}
-
-static __inline void d_cnjg(doublecomplex *r, doublecomplex *z)
-{
-	double zi = z->i;
-	r->r = z->r;
-	r->i = -zi;
-}
-
-static __inline void r_cnjg(complex *r, complex *z)
-{
-	float zi = z->i;
-	r->r = z->r;
-	r->i = -zi;
-}
-
-static __inline int s_copy(char *a, char *b, int maxlen)
-{
-    strncpy(a, b, maxlen);
-    a[maxlen] = '\0';
-    return 0;
-}
-
-int s_cat(char *lp, char **rpp, int* rnp, int *np);
-int s_cmp(char *a0, char *b0);
-static __inline int i_len(char* s)
-{
-    return (int)strlen(s);
-}
-
-#ifdef __cplusplus
-}
-#endif
-
-#endif
@@ -1,386 +0,0 @@
-// this is auto-generated header for Lapack subset
-#ifndef __CLAPACK_H__
-#define __CLAPACK_H__
-
-#include "cblas.h"
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-int cgemm_(char *transa, char *transb, int *m, int *n, int *
-	k, lapack_complex *alpha, lapack_complex *a, int *lda, lapack_complex *b, int *ldb,
-	lapack_complex *beta, lapack_complex *c__, int *ldc);
-
-int daxpy_(int *n, double *da, double *dx, int *incx, double
-	*dy, int *incy);
-
-int dbdsdc_(char *uplo, char *compq, int *n, double *d__,
-	double *e, double *u, int *ldu, double *vt, int *ldvt, double *q, int
-	*iq, double *work, int *iwork, int *info);
-
-int dbdsqr_(char *uplo, int *n, int *ncvt, int *nru, int *
-	ncc, double *d__, double *e, double *vt, int *ldvt, double *u, int *
-	ldu, double *c__, int *ldc, double *work, int *info);
-
-int dcombssq_(double *v1, double *v2);
-
-int dcopy_(int *n, double *dx, int *incx, double *dy, int *
-	incy);
-
-double ddot_(int *n, double *dx, int *incx, double *dy, int *incy);
-
-int dgebak_(char *job, char *side, int *n, int *ilo, int *
-	ihi, double *scale, int *m, double *v, int *ldv, int *info);
-
-int dgebal_(char *job, int *n, double *a, int *lda, int *ilo,
-	 int *ihi, double *scale, int *info);
-
-int dgebd2_(int *m, int *n, double *a, int *lda, double *d__,
-	 double *e, double *tauq, double *taup, double *work, int *info);
-
-int dgebrd_(int *m, int *n, double *a, int *lda, double *d__,
-	 double *e, double *tauq, double *taup, double *work, int *lwork, int
-	*info);
-
-int dgeev_(char *jobvl, char *jobvr, int *n, double *a, int *
-	lda, double *wr, double *wi, double *vl, int *ldvl, double *vr, int *
-	ldvr, double *work, int *lwork, int *info);
-
-int dgehd2_(int *n, int *ilo, int *ihi, double *a, int *lda,
-	double *tau, double *work, int *info);
-
-int dgehrd_(int *n, int *ilo, int *ihi, double *a, int *lda,
-	double *tau, double *work, int *lwork, int *info);
-
-int dgelq2_(int *m, int *n, double *a, int *lda, double *tau,
-	 double *work, int *info);
-
-int dgelqf_(int *m, int *n, double *a, int *lda, double *tau,
-	 double *work, int *lwork, int *info);
-
-int dgemm_(char *transa, char *transb, int *m, int *n, int *
-	k, double *alpha, double *a, int *lda, double *b, int *ldb, double *
-	beta, double *c__, int *ldc);
-
-int dgemv_(char *trans, int *m, int *n, double *alpha,
-	double *a, int *lda, double *x, int *incx, double *beta, double *y,
-	int *incy);
-
-int dgeqr2_(int *m, int *n, double *a, int *lda, double *tau,
-	 double *work, int *info);
-
-int dgeqrf_(int *m, int *n, double *a, int *lda, double *tau,
-	 double *work, int *lwork, int *info);
-
-int dger_(int *m, int *n, double *alpha, double *x, int *
-	incx, double *y, int *incy, double *a, int *lda);
-
-int dgesdd_(char *jobz, int *m, int *n, double *a, int *lda,
-	double *s, double *u, int *ldu, double *vt, int *ldvt, double *work,
-	int *lwork, int *iwork, int *info);
-
-int dhseqr_(char *job, char *compz, int *n, int *ilo, int *
-	ihi, double *h__, int *ldh, double *wr, double *wi, double *z__, int *
-	ldz, double *work, int *lwork, int *info);
-
-int disnan_(double *din);
-
-// "small" is a macro defined in Windows headers: https://stackoverflow.com/a/27794577
-#ifdef small
-#undef small
-#endif
-
-int dlabad_(double *small, double *large);
-
-int dlabrd_(int *m, int *n, int *nb, double *a, int *lda,
-	double *d__, double *e, double *tauq, double *taup, double *x, int *
-	ldx, double *y, int *ldy);
-
-int dlacpy_(char *uplo, int *m, int *n, double *a, int *lda,
-	double *b, int *ldb);
-
-int dladiv1_(double *a, double *b, double *c__, double *d__,
-	double *p, double *q);
-
-double dladiv2_(double *a, double *b, double *c__, double *d__, double *r__,
-	double *t);
-
-int dladiv_(double *a, double *b, double *c__, double *d__,
-	double *p, double *q);
-
-int dlaed6_(int *kniter, int *orgati, double *rho, double *
-	d__, double *z__, double *finit, double *tau, int *info);
-
-int dlaexc_(int *wantq, int *n, double *t, int *ldt, double *
-	q, int *ldq, int *j1, int *n1, int *n2, double *work, int *info);
-
-int dlahqr_(int *wantt, int *wantz, int *n, int *ilo, int *
-	ihi, double *h__, int *ldh, double *wr, double *wi, int *iloz, int *
-	ihiz, double *z__, int *ldz, int *info);
-
-int dlahr2_(int *n, int *k, int *nb, double *a, int *lda,
-	double *tau, double *t, int *ldt, double *y, int *ldy);
-
-int dlaisnan_(double *din1, double *din2);
-
-int dlaln2_(int *ltrans, int *na, int *nw, double *smin,
-	double *ca, double *a, int *lda, double *d1, double *d2, double *b,
-	int *ldb, double *wr, double *wi, double *x, int *ldx, double *scale,
-	double *xnorm, int *info);
-
-int dlamrg_(int *n1, int *n2, double *a, int *dtrd1, int *
-	dtrd2, int *index);
-
-double dlange_(char *norm, int *m, int *n, double *a, int *lda, double *work);
-
-double dlanst_(char *norm, int *n, double *d__, double *e);
-
-int dlanv2_(double *a, double *b, double *c__, double *d__,
-	double *rt1r, double *rt1i, double *rt2r, double *rt2i, double *cs,
-	double *sn);
-
-double dlapy2_(double *x, double *y);
-
-int dlaqr0_(int *wantt, int *wantz, int *n, int *ilo, int *
-	ihi, double *h__, int *ldh, double *wr, double *wi, int *iloz, int *
-	ihiz, double *z__, int *ldz, double *work, int *lwork, int *info);
-
-int dlaqr1_(int *n, double *h__, int *ldh, double *sr1,
-	double *si1, double *sr2, double *si2, double *v);
-
-int dlaqr2_(int *wantt, int *wantz, int *n, int *ktop, int *
-	kbot, int *nw, double *h__, int *ldh, int *iloz, int *ihiz, double *
-	z__, int *ldz, int *ns, int *nd, double *sr, double *si, double *v,
-	int *ldv, int *nh, double *t, int *ldt, int *nv, double *wv, int *
-	ldwv, double *work, int *lwork);
-
-int dlaqr3_(int *wantt, int *wantz, int *n, int *ktop, int *
-	kbot, int *nw, double *h__, int *ldh, int *iloz, int *ihiz, double *
-	z__, int *ldz, int *ns, int *nd, double *sr, double *si, double *v,
-	int *ldv, int *nh, double *t, int *ldt, int *nv, double *wv, int *
-	ldwv, double *work, int *lwork);
-
-int dlaqr4_(int *wantt, int *wantz, int *n, int *ilo, int *
-	ihi, double *h__, int *ldh, double *wr, double *wi, int *iloz, int *
-	ihiz, double *z__, int *ldz, double *work, int *lwork, int *info);
-
-int dlaqr5_(int *wantt, int *wantz, int *kacc22, int *n, int
-	*ktop, int *kbot, int *nshfts, double *sr, double *si, double *h__,
-	int *ldh, int *iloz, int *ihiz, double *z__, int *ldz, double *v, int
-	*ldv, double *u, int *ldu, int *nv, double *wv, int *ldwv, int *nh,
-	double *wh, int *ldwh);
-
-int dlarf_(char *side, int *m, int *n, double *v, int *incv,
-	double *tau, double *c__, int *ldc, double *work);
-
-int dlarfb_(char *side, char *trans, char *direct, char *
-	storev, int *m, int *n, int *k, double *v, int *ldv, double *t, int *
-	ldt, double *c__, int *ldc, double *work, int *ldwork);
-
-int dlarfg_(int *n, double *alpha, double *x, int *incx,
-	double *tau);
-
-int dlarft_(char *direct, char *storev, int *n, int *k,
-	double *v, int *ldv, double *tau, double *t, int *ldt);
-
-int dlarfx_(char *side, int *m, int *n, double *v, double *
-	tau, double *c__, int *ldc, double *work);
-
-int dlartg_(double *f, double *g, double *cs, double *sn,
-	double *r__);
-
-int dlas2_(double *f, double *g, double *h__, double *ssmin,
-	double *ssmax);
-
-int dlascl_(char *type__, int *kl, int *ku, double *cfrom,
-	double *cto, int *m, int *n, double *a, int *lda, int *info);
-
-int dlasd0_(int *n, int *sqre, double *d__, double *e,
-	double *u, int *ldu, double *vt, int *ldvt, int *smlsiz, int *iwork,
-	double *work, int *info);
-
-int dlasd1_(int *nl, int *nr, int *sqre, double *d__, double
-	*alpha, double *beta, double *u, int *ldu, double *vt, int *ldvt, int
-	*idxq, int *iwork, double *work, int *info);
-
-int dlasd2_(int *nl, int *nr, int *sqre, int *k, double *d__,
-	 double *z__, double *alpha, double *beta, double *u, int *ldu,
-	double *vt, int *ldvt, double *dsigma, double *u2, int *ldu2, double *
-	vt2, int *ldvt2, int *idxp, int *idx, int *idxc, int *idxq, int *
-	coltyp, int *info);
-
-int dlasd3_(int *nl, int *nr, int *sqre, int *k, double *d__,
-	 double *q, int *ldq, double *dsigma, double *u, int *ldu, double *u2,
-	 int *ldu2, double *vt, int *ldvt, double *vt2, int *ldvt2, int *idxc,
-	 int *ctot, double *z__, int *info);
-
-int dlasd4_(int *n, int *i__, double *d__, double *z__,
-	double *delta, double *rho, double *sigma, double *work, int *info);
-
-int dlasd5_(int *i__, double *d__, double *z__, double *
-	delta, double *rho, double *dsigma, double *work);
-
-int dlasd6_(int *icompq, int *nl, int *nr, int *sqre, double
-	*d__, double *vf, double *vl, double *alpha, double *beta, int *idxq,
-	int *perm, int *givptr, int *givcol, int *ldgcol, double *givnum, int
-	*ldgnum, double *poles, double *difl, double *difr, double *z__, int *
-	k, double *c__, double *s, double *work, int *iwork, int *info);
-
-int dlasd7_(int *icompq, int *nl, int *nr, int *sqre, int *k,
-	 double *d__, double *z__, double *zw, double *vf, double *vfw,
-	double *vl, double *vlw, double *alpha, double *beta, double *dsigma,
-	int *idx, int *idxp, int *idxq, int *perm, int *givptr, int *givcol,
-	int *ldgcol, double *givnum, int *ldgnum, double *c__, double *s, int
-	*info);
-
-int dlasd8_(int *icompq, int *k, double *d__, double *z__,
-	double *vf, double *vl, double *difl, double *difr, int *lddifr,
-	double *dsigma, double *work, int *info);
-
-int dlasda_(int *icompq, int *smlsiz, int *n, int *sqre,
-	double *d__, double *e, double *u, int *ldu, double *vt, int *k,
-	double *difl, double *difr, double *z__, double *poles, int *givptr,
-	int *givcol, int *ldgcol, int *perm, double *givnum, double *c__,
-	double *s, double *work, int *iwork, int *info);
-
-int dlasdq_(char *uplo, int *sqre, int *n, int *ncvt, int *
-	nru, int *ncc, double *d__, double *e, double *vt, int *ldvt, double *
-	u, int *ldu, double *c__, int *ldc, double *work, int *info);
-
-int dlasdt_(int *n, int *lvl, int *nd, int *inode, int *
-	ndiml, int *ndimr, int *msub);
-
-int dlaset_(char *uplo, int *m, int *n, double *alpha,
-	double *beta, double *a, int *lda);
-
-int dlasq1_(int *n, double *d__, double *e, double *work,
-	int *info);
-
-int dlasq2_(int *n, double *z__, int *info);
-
-int dlasq3_(int *i0, int *n0, double *z__, int *pp, double *
-	dmin__, double *sigma, double *desig, double *qmax, int *nfail, int *
-	iter, int *ndiv, int *ieee, int *ttype, double *dmin1, double *dmin2,
-	double *dn, double *dn1, double *dn2, double *g, double *tau);
-
-int dlasq4_(int *i0, int *n0, double *z__, int *pp, int *
-	n0in, double *dmin__, double *dmin1, double *dmin2, double *dn,
-	double *dn1, double *dn2, double *tau, int *ttype, double *g);
-
-int dlasq5_(int *i0, int *n0, double *z__, int *pp, double *
-	tau, double *sigma, double *dmin__, double *dmin1, double *dmin2,
-	double *dn, double *dnm1, double *dnm2, int *ieee, double *eps);
-
-int dlasq6_(int *i0, int *n0, double *z__, int *pp, double *
-	dmin__, double *dmin1, double *dmin2, double *dn, double *dnm1,
-	double *dnm2);
-
-int dlasr_(char *side, char *pivot, char *direct, int *m,
-	int *n, double *c__, double *s, double *a, int *lda);
-
-int dlasrt_(char *id, int *n, double *d__, int *info);
-
-int dlassq_(int *n, double *x, int *incx, double *scale,
-	double *sumsq);
-
-int dlasv2_(double *f, double *g, double *h__, double *ssmin,
-	 double *ssmax, double *snr, double *csr, double *snl, double *csl);
-
-int dlasy2_(int *ltranl, int *ltranr, int *isgn, int *n1,
-	int *n2, double *tl, int *ldtl, double *tr, int *ldtr, double *b, int
-	*ldb, double *scale, double *x, int *ldx, double *xnorm, int *info);
-
-double dnrm2_(int *n, double *x, int *incx);
-
-int dorg2r_(int *m, int *n, int *k, double *a, int *lda,
-	double *tau, double *work, int *info);
-
-int dorgbr_(char *vect, int *m, int *n, int *k, double *a,
-	int *lda, double *tau, double *work, int *lwork, int *info);
-
-int dorghr_(int *n, int *ilo, int *ihi, double *a, int *lda,
-	double *tau, double *work, int *lwork, int *info);
-
-int dorgl2_(int *m, int *n, int *k, double *a, int *lda,
-	double *tau, double *work, int *info);
-
-int dorglq_(int *m, int *n, int *k, double *a, int *lda,
-	double *tau, double *work, int *lwork, int *info);
-
-int dorgqr_(int *m, int *n, int *k, double *a, int *lda,
-	double *tau, double *work, int *lwork, int *info);
-
-int dorm2r_(char *side, char *trans, int *m, int *n, int *k,
-	double *a, int *lda, double *tau, double *c__, int *ldc, double *work,
-	 int *info);
-
-int dormbr_(char *vect, char *side, char *trans, int *m, int
-	*n, int *k, double *a, int *lda, double *tau, double *c__, int *ldc,
-	double *work, int *lwork, int *info);
-
-int dormhr_(char *side, char *trans, int *m, int *n, int *
-	ilo, int *ihi, double *a, int *lda, double *tau, double *c__, int *
-	ldc, double *work, int *lwork, int *info);
-
-int dorml2_(char *side, char *trans, int *m, int *n, int *k,
-	double *a, int *lda, double *tau, double *c__, int *ldc, double *work,
-	 int *info);
-
-int dormlq_(char *side, char *trans, int *m, int *n, int *k,
-	double *a, int *lda, double *tau, double *c__, int *ldc, double *work,
-	 int *lwork, int *info);
-
-int dormqr_(char *side, char *trans, int *m, int *n, int *k,
-	double *a, int *lda, double *tau, double *c__, int *ldc, double *work,
-	 int *lwork, int *info);
-
-int drot_(int *n, double *dx, int *incx, double *dy, int *
-	incy, double *c__, double *s);
-
-int dscal_(int *n, double *da, double *dx, int *incx);
-
-int dswap_(int *n, double *dx, int *incx, double *dy, int *
-	incy);
-
-int dtrevc3_(char *side, char *howmny, int *select, int *n,
-	double *t, int *ldt, double *vl, int *ldvl, double *vr, int *ldvr,
-	int *mm, int *m, double *work, int *lwork, int *info);
-
-int dtrexc_(char *compq, int *n, double *t, int *ldt, double
-	*q, int *ldq, int *ifst, int *ilst, double *work, int *info);
-
-int dtrmm_(char *side, char *uplo, char *transa, char *diag,
-	int *m, int *n, double *alpha, double *a, int *lda, double *b, int *
-	ldb);
-
-int dtrmv_(char *uplo, char *trans, char *diag, int *n,
-	double *a, int *lda, double *x, int *incx);
-
-int idamax_(int *n, double *dx, int *incx);
-
-int ieeeck_(int *ispec, float *zero, float *one);
-
-int iladlc_(int *m, int *n, double *a, int *lda);
-
-int iladlr_(int *m, int *n, double *a, int *lda);
-
-int ilaenv_(int *ispec, char *name__, char *opts, int *n1, int *n2, int *n3,
-	int *n4);
-
-int iparmq_(int *ispec, char *name__, char *opts, int *n, int *ilo, int *ihi,
-	int *lwork);
-
-int sgemm_(char *transa, char *transb, int *m, int *n, int *
-	k, float *alpha, float *a, int *lda, float *b, int *ldb, float *beta,
-	float *c__, int *ldc);
-
-int zgemm_(char *transa, char *transb, int *m, int *n, int *
-	k, lapack_doublecomplex *alpha, lapack_doublecomplex *a, int *lda, lapack_doublecomplex *b,
-	 int *ldb, lapack_doublecomplex *beta, lapack_doublecomplex *c__, int *ldc);
-
-#ifdef __cplusplus
-}
-#endif
-
-#endif
@@ -1,48 +0,0 @@
-Copyright (c) 1992-2017 The University of Tennessee and The University
-                        of Tennessee Research Foundation.  All rights
-                        reserved.
-Copyright (c) 2000-2017 The University of California Berkeley. All
-                        rights reserved.
-Copyright (c) 2006-2017 The University of Colorado Denver.  All rights
-                        reserved.
-
-$COPYRIGHT$
-
-Additional copyrights may follow
-
-$HEADER$
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions are
-met:
-
- Redistributions of source code must retain the above copyright
-  notice, this list of conditions and the following disclaimer.
-
- Redistributions in binary form must reproduce the above copyright
-  notice, this list of conditions and the following disclaimer listed
-  in this license in the documentation and/or other materials
-  provided with the distribution.
-
- Neither the name of the copyright holders nor the names of its
-  contributors may be used to endorse or promote products derived from
-  this software without specific prior written permission.
-
-The copyright holders provide no reassurances that the source code
-provided does not infringe any patent, copyright, or any other
-intellectual property rights of third parties.  The copyright holders
-disclaim any liability to any recipient for claims brought against
-recipient by any third party for infringement of that parties
-intellectual property rights.
-
-THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
@@ -1,272 +0,0 @@
-appdoc = """
-    This is generator of CLapack subset.
-    The usage:
-
-    1. Make sure you have the special version of f2c installed.
-       Grab it from https://github.com/vpisarev/f2c/tree/for_lapack.
-    2. Download fresh version of Lapack from
-       https://github.com/Reference-LAPACK/lapack.
-       You may choose some specific version or the latest snapshot.
-    3. If necessary, edit "roots" and "banlist" variables in this script, specify the needed and unneeded functions
-    4. From within a working directory run
-
-       $ python3 <opencv_root>/3rdparty/clapack/make_clapack.py <lapack_root>
-       or
-       $ F2C=<path_to_custom_f2c> python3 <opencv_root>/3rdparty/clapack/make_clapack.py <lapack_root>
-
-       it will generate "new_clapack" directory with "include" and "src" subdirectories.
-    5. erase opencv/3rdparty/clapack/src and replace it with new_clapack/src.
-    6. copy new_clapack/include/lapack.h to opencv/3rdparty/clapack/include.
-    7. optionally, edit opencv/3rdparty/clapack/CMakeLists.txt and update CLAPACK_VERSION as needed.
-
-    This is it. Now build it and enjoy.
-"""
-
-import glob, re, os, shutil, subprocess, sys
-
-roots = ["cgemm_", "dgemm_", "sgemm_", "zgemm_",
-         "dgeev_", "dgesdd_", 
-         #"dsyevr_",
-         #"dgesv_", "dgetrf_", "dposv_", "dpotrf_", "dgels_", "dgeqrf_",
-         #"sgesv_", "sgetrf_", "sposv_", "spotrf_", "sgels_", "sgeqrf_"
-         ]
-banlist = ["slamch_", "slamc3_", "dlamch_", "dlamc3_", "lsame_", "xerbla_"]
-
-if len(sys.argv) < 2:
-    print(appdoc)
-    sys.exit(0)
-
-lapack_root = sys.argv[1]
-dst_path = "."
-
-def error(msg):
-    print ("error: " + msg)
-    sys.exit(0)
-
-def file2fun(fname):
-    return (os.path.basename(fname)[:-2]).upper()
-
-def print_graph(m):
-    for (k, neighbors) in sorted(m.items()):
-        print (k + " : " + ", ".join(sorted(list(neighbors))))
-
-blas_path = os.path.join(lapack_root, "BLAS/SRC")
-lapack_path = os.path.join(lapack_root, "SRC")
-
-roots = [f[:-1].upper() for f in roots]
-banlist = [f[:-1].upper() for f in banlist]
-
-def fun2file(func):
-    filename = func.lower() + ".f"
-    blas_loc = blas_path + "/" + filename
-    lapack_loc = lapack_path + "/" + filename
-    if os.path.exists(blas_loc):
-        return blas_loc
-    elif os.path.exists(lapack_loc):
-        return lapack_loc
-    else:
-        error("neither %s nor %s exist" % (blas_loc, lapack_loc))
-
-all_files = glob.glob(blas_path + "/*.f") + glob.glob(lapack_path + "/*.f")
-all_funcs = [file2fun(fname) for fname in all_files]
-all_funcs_set = set(all_funcs).difference(set(banlist))
-all_funcs = sorted(list(all_funcs_set))
-
-func_deps = {}
-
-#print all_funcs
-
-words_regexp = re.compile(r'\w+')
-
-def scan_deps(func):
-    global func_deps
-    if func in func_deps:
-        return
-    func_deps[func] = set([]) # to avoid possibly infinite recursion
-    f = open(fun2file(func), 'rt')
-    deps = []
-    external_mode = False
-    for l in f.readlines():
-        if l.startswith('*'):
-            continue
-        l = l.strip().upper()
-        if l.startswith('EXTERNAL '):
-            external_mode = True
-        elif l.startswith('$') and external_mode:
-            pass
-        else:
-            external_mode = False
-        if not external_mode:
-            continue
-        for w in words_regexp.findall(l):
-            if w in all_funcs_set:
-                deps.append(w)
-    f.close()
-    # remove func from its dependencies
-    deps = set(deps).difference(set([func]))
-    func_deps[func] = deps
-    for d in deps:
-        scan_deps(d)
-
-for r in roots:
-    scan_deps(r)
-
-selected_funcs = sorted(func_deps.keys())
-print ("total files before amalgamation: %d" % len(selected_funcs))
-
-inv_deps = {}
-for func in selected_funcs:
-    inv_deps[func] = set([])
-
-for (func, deps) in func_deps.items():
-    for d in deps:
-        inv_deps[d] = inv_deps[d].union(set([func]))
-
-#print_graph(inv_deps)
-
-func_home = {}
-for func in selected_funcs:
-    func_home[func] = func
-
-def get_home0(func, func0):
-    used_by = inv_deps[func]
-    if len(used_by) == 1:
-        p = list(used_by)[0]
-        if p != func and p != func0:
-            return get_home0(p, func0)
-        return func
-    return func
-
-# try to merge some files
-for func in selected_funcs:
-    func_home[func] = get_home0(func, func)
-
-# try to merge some files even more
-for iters in range(100):
-    homes_changed = False
-    for (func, used_by) in inv_deps.items():
-        p0 = func_home[func]
-        n = len(used_by)
-        if n == 1:
-            p = list(used_by)[0]
-            p1 = func_home[p]
-            if p1 != p0:
-                func_home[func] = p1
-                homes_changed = True
-            continue
-        elif n > 1:
-            phomes = set([])
-            for p in used_by:
-                phomes.add(func_home[p])
-            if len(phomes) == 1:
-                p1 = list(phomes)[0]
-                if p1 != p0:
-                    func_home[func] = p1
-                    homes_changed = True
-    if not homes_changed:
-        break
-
-res_files = {}
-for (func, h) in func_home.items():
-    elems = res_files.get(h, set([]))
-    elems.add(func)
-    res_files[h] = elems
-
-print ("total files after amalgamation: %d" % len(res_files))
-#print_graph(res_files)
-
-outdir = os.path.join(dst_path, "new_clapack")
-outdir_src = os.path.join(outdir, "src")
-outdir_inc = os.path.join(outdir, "include")
-
-shutil.rmtree(outdir, ignore_errors=True)
-try:
-    os.makedirs(outdir_src)
-except os.error:
-    pass
-try:
-    os.makedirs(outdir_inc)
-except os.error:
-    pass
-
-f2c_appname = os.getenv("F2C", default="f2c")
-print ("f2c used: %s" % f2c_appname)
-
-f2c_getver_cmd = f2c_appname + " -v"
-
-verstr = subprocess.check_output(f2c_getver_cmd.split(' ')).decode("utf-8")
-if "for_lapack" not in verstr:
-    error("invalid version of f2c\n" + appdoc)
-
-f2c_flags = "-ctypes -localconst -no-proto"
-f2c_cmd0 = f2c_appname + " " + f2c_flags
-f2c_cmd1 = f2c_appname + " -hdr none " + f2c_flags
-
-lapack_protos = {}
-extract_fn_regexp = re.compile(r'.+?(\w+)\s*\(')
-
-def extract_proto(func, csrc):
-    global lapack_protos
-    cname = func.lower() + "_"
-    cfname = func.lower() + ".c"
-    regexp_str = r'\n(?:/\* Subroutine \*/\s*)?\w+\s+\w+\s*\((?:.|\n)+?\)[\s\n]*\{'
-    proto_regexp = re.compile(regexp_str)
-    ps = proto_regexp.findall(csrc)
-    for p in ps:
-        n = p.find("*/")
-        if n < 0:
-            n = 0
-        else:
-            n += 2
-        p = p[n:-1].strip() + ";"
-        fns = extract_fn_regexp.findall(p)
-        if len(fns) != 1:
-            error("prototype of function (%s) when analyzing %s cannot be parsed" % (p, cfname))
-        fn = fns[0]
-        if fn not in lapack_protos:
-            p = re.sub(r'\bcomplex\b', 'lapack_complex', p)
-            p = re.sub(r'\bdoublecomplex\b', 'lapack_doublecomplex', p)
-            lapack_protos[fn] = p
-
-for (filename, funcs) in sorted(res_files.items()):
-    out = ""
-    f2c_cmd = f2c_cmd0
-    for func in sorted(list(funcs)):
-        ffilename = fun2file(func)
-        print ("running " + f2c_cmd + " on " + ffilename +  " ...")
-        ffile = open(ffilename, 'rt')
-        delta_out = subprocess.check_output(f2c_cmd.split(' '), stdin=ffile).decode("utf-8")
-        # remove trailing whitespaces
-        delta_out = '\n'.join([l.rstrip() for l in delta_out.split('\n')])
-        extract_proto(func, delta_out)
-        out += delta_out
-        ffile.close()
-        f2c_cmd = f2c_cmd1
-    outname = os.path.join(outdir_src, filename.lower() + ".c")
-    outfile = open(outname, 'wt')
-    outfile.write(out)
-    outfile.close()
-
-proto_hdr = """// this is auto-generated header for Lapack subset
-#ifndef __CLAPACK_H__
-#define __CLAPACK_H__
-
-#include "cblas.h"
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-%s
-
-#ifdef __cplusplus
-}
-#endif
-
-#endif
-""" % "\n\n".join([p for (n, p) in sorted(lapack_protos.items())])
-
-proto_hdr_fname = os.path.join(outdir_inc, "lapack.h")
-f = open(proto_hdr_fname, 'wt')
-f.write(proto_hdr)
-f.close()
@@ -1,289 +0,0 @@
-#include "f2c.h"
-#include <stdarg.h>
-
-void cblas_cgemm(const CBLAS_LAYOUT layout, const CBLAS_TRANSPOSE TransA,
-                 const CBLAS_TRANSPOSE TransB, const int M, const int N,
-                 const int K, const void *alpha, const void  *A,
-                 const int lda, const void  *B, const int ldb,
-                 const void *beta, void  *C, const int ldc)
-{
-   char TA, TB;
-
-   if( layout == CblasColMajor )
-   {
-      if(TransA == CblasTrans) TA='T';
-      else if ( TransA == CblasConjTrans ) TA='C';
-      else if ( TransA == CblasNoTrans )   TA='N';
-      else
-      {
-         cblas_xerbla(layout, 2, "cblas_cgemm", "Illegal TransA setting, %d\n", TransA);
-         return;
-      }
-
-      if(TransB == CblasTrans) TB='T';
-      else if ( TransB == CblasConjTrans ) TB='C';
-      else if ( TransB == CblasNoTrans )   TB='N';
-      else
-      {
-         cblas_xerbla(layout, 3, "cblas_cgemm", "Illegal TransB setting, %d\n", TransB);
-         return;
-      }
-
-      cgemm_(&TA, &TB, (int*)&M, (int*)&N, (int*)&K, (complex*)alpha, (complex*)A, (int*)&lda,
-             (complex*)B, (int*)&ldb, (complex*)beta, (complex*)C, (int*)&ldc);
-   }
-   else if (layout == CblasRowMajor)
-   {
-      if(TransA == CblasTrans) TB='T';
-      else if ( TransA == CblasConjTrans ) TB='C';
-      else if ( TransA == CblasNoTrans )   TB='N';
-      else
-      {
-         cblas_xerbla(layout, 2, "cblas_cgemm", "Illegal TransA setting, %d\n", TransA);
-         return;
-      }
-      if(TransB == CblasTrans) TA='T';
-      else if ( TransB == CblasConjTrans ) TA='C';
-      else if ( TransB == CblasNoTrans )   TA='N';
-      else
-      {
-         cblas_xerbla(layout, 2, "cblas_cgemm", "Illegal TransB setting, %d\n", TransB);
-         return;
-      }
-
-      cgemm_(&TA, &TB, (int*)&N, (int*)&M, (int*)&K, (complex*)alpha, (complex*)B, (int*)&ldb,
-             (complex*)A, (int*)&lda, (complex*)beta, (complex*)C, (int*)&ldc);
-   }
-   else cblas_xerbla(layout, 1, "cblas_cgemm", "Illegal layout setting, %d\n", layout);
-}
-
-void cblas_dgemm(const CBLAS_LAYOUT layout, const CBLAS_TRANSPOSE TransA,
-                 const CBLAS_TRANSPOSE TransB, const int M, const int N,
-                 const int K, const double alpha, const double *A,
-                 const int lda, const double  *B, const int ldb,
-                 const double beta, double  *C, const int ldc)
-{
-   char TA, TB;
-
-   if( layout == CblasColMajor )
-   {
-      if(TransA == CblasTrans) TA='T';
-      else if ( TransA == CblasConjTrans ) TA='C';
-      else if ( TransA == CblasNoTrans )   TA='N';
-      else
-      {
-         cblas_xerbla(layout, 2, "cblas_dgemm", "Illegal TransA setting, %d\n", TransA);
-         return;
-      }
-
-      if(TransB == CblasTrans) TB='T';
-      else if ( TransB == CblasConjTrans ) TB='C';
-      else if ( TransB == CblasNoTrans )   TB='N';
-      else
-      {
-         cblas_xerbla(layout, 3, "cblas_dgemm", "Illegal TransB setting, %d\n", TransB);
-         return;
-      }
-
-      dgemm_(&TA, &TB, (int*)&M, (int*)&N, (int*)&K, (double*)&alpha, (double*)A, (int*)&lda,
-             (double*)B, (int*)&ldb, (double*)&beta, (double*)C, (int*)&ldc);
-   }
-   else if (layout == CblasRowMajor)
-   {
-      if(TransA == CblasTrans) TB='T';
-      else if ( TransA == CblasConjTrans ) TB='C';
-      else if ( TransA == CblasNoTrans )   TB='N';
-      else
-      {
-         cblas_xerbla(layout, 2, "cblas_dgemm", "Illegal TransA setting, %d\n", TransA);
-         return;
-      }
-      if(TransB == CblasTrans) TA='T';
-      else if ( TransB == CblasConjTrans ) TA='C';
-      else if ( TransB == CblasNoTrans )   TA='N';
-      else
-      {
-         cblas_xerbla(layout, 2, "cblas_dgemm", "Illegal TransB setting, %d\n", TransB);
-         return;
-      }
-
-      dgemm_(&TA, &TB, (int*)&N, (int*)&M, (int*)&K, (double*)&alpha, (double*)B, (int*)&ldb,
-             (double*)A, (int*)&lda, (double*)&beta, (double*)C, (int*)&ldc);
-   }
-   else cblas_xerbla(layout, 1, "cblas_dgemm", "Illegal layout setting, %d\n", layout);
-}
-
-
-void cblas_sgemm(const CBLAS_LAYOUT layout, const CBLAS_TRANSPOSE TransA,
-                 const CBLAS_TRANSPOSE TransB, const int M, const int N,
-                 const int K, const float alpha, const float  *A,
-                 const int lda, const float  *B, const int ldb,
-                 const float beta, float  *C, const int ldc)
-{
-   char TA, TB;
-
-   if( layout == CblasColMajor )
-   {
-      if(TransA == CblasTrans) TA='T';
-      else if ( TransA == CblasConjTrans ) TA='C';
-      else if ( TransA == CblasNoTrans )   TA='N';
-      else
-      {
-         cblas_xerbla(layout, 2, "cblas_sgemm", "Illegal TransA setting, %d\n", TransA);
-         return;
-      }
-
-      if(TransB == CblasTrans) TB='T';
-      else if ( TransB == CblasConjTrans ) TB='C';
-      else if ( TransB == CblasNoTrans )   TB='N';
-      else
-      {
-         cblas_xerbla(layout, 3, "cblas_sgemm", "Illegal TransB setting, %d\n", TransB);
-         return;
-      }
-
-      sgemm_(&TA, &TB, (int*)&M, (int*)&N, (int*)&K, (float*)&alpha, (float*)A, (int*)&lda,
-             (float*)B, (int*)&ldb, (float*)&beta, (float*)C, (int*)&ldc);
-   }
-   else if (layout == CblasRowMajor)
-   {
-      if(TransA == CblasTrans) TB='T';
-      else if ( TransA == CblasConjTrans ) TB='C';
-      else if ( TransA == CblasNoTrans )   TB='N';
-      else
-      {
-         cblas_xerbla(layout, 2, "cblas_sgemm", "Illegal TransA setting, %d\n", TransA);
-         return;
-      }
-      if(TransB == CblasTrans) TA='T';
-      else if ( TransB == CblasConjTrans ) TA='C';
-      else if ( TransB == CblasNoTrans )   TA='N';
-      else
-      {
-         cblas_xerbla(layout, 2, "cblas_sgemm", "Illegal TransB setting, %d\n", TransB);
-         return;
-      }
-
-      sgemm_(&TA, &TB, (int*)&N, (int*)&M, (int*)&K, (float*)&alpha, (float*)B, (int*)&ldb,
-             (float*)A, (int*)&lda, (float*)&beta, (float*)C, (int*)&ldc);
-   }
-   else cblas_xerbla(layout, 1, "cblas_sgemm", "Illegal layout setting, %d\n", layout);
-}
-
-void cblas_zgemm(const CBLAS_LAYOUT layout, const CBLAS_TRANSPOSE TransA,
-                 const CBLAS_TRANSPOSE TransB, const int M, const int N,
-                 const int K, const void *alpha, const void  *A,
-                 const int lda, const void  *B, const int ldb,
-                 const void *beta, void  *C, const int ldc)
-{
-   char TA, TB;
-
-   if( layout == CblasColMajor )
-   {
-      if(TransA == CblasTrans) TA='T';
-      else if ( TransA == CblasConjTrans ) TA='C';
-      else if ( TransA == CblasNoTrans )   TA='N';
-      else
-      {
-         cblas_xerbla(layout, 2, "cblas_zgemm", "Illegal TransA setting, %d\n", TransA);
-         return;
-      }
-
-      if(TransB == CblasTrans) TB='T';
-      else if ( TransB == CblasConjTrans ) TB='C';
-      else if ( TransB == CblasNoTrans )   TB='N';
-      else
-      {
-         cblas_xerbla(layout, 3, "cblas_zgemm", "Illegal TransB setting, %d\n", TransB);
-         return;
-      }
-
-      zgemm_(&TA, &TB, (int*)&M, (int*)&N, (int*)&K, (doublecomplex*)alpha, (doublecomplex*)A, (int*)&lda,
-             (doublecomplex*)B, (int*)&ldb, (doublecomplex*)beta, (doublecomplex*)C, (int*)&ldc);
-   }
-   else if (layout == CblasRowMajor)
-   {
-      if(TransA == CblasTrans) TB='T';
-      else if ( TransA == CblasConjTrans ) TB='C';
-      else if ( TransA == CblasNoTrans )   TB='N';
-      else
-      {
-         cblas_xerbla(layout, 2, "cblas_zgemm", "Illegal TransA setting, %d\n", TransA);
-         return;
-      }
-      if(TransB == CblasTrans) TA='T';
-      else if ( TransB == CblasConjTrans ) TA='C';
-      else if ( TransB == CblasNoTrans )   TA='N';
-      else
-      {
-         cblas_xerbla(layout, 2, "cblas_zgemm", "Illegal TransB setting, %d\n", TransB);
-         return;
-      }
-
-      zgemm_(&TA, &TB, (int*)&N, (int*)&M, (int*)&K, (doublecomplex*)alpha, (doublecomplex*)B, (int*)&ldb,
-             (doublecomplex*)A, (int*)&lda, (doublecomplex*)beta, (doublecomplex*)C, (int*)&ldc);
-   }
-   else cblas_xerbla(layout, 1, "cblas_zgemm", "Illegal layout setting, %d\n", layout);
-}
-
-void cblas_xerbla(const CBLAS_LAYOUT layout, int info, const char *rout, const char *form, ...)
-{
-   extern int RowMajorStrg;
-   char empty[1] = "";
-   va_list argptr;
-
-   va_start(argptr, form);
-
-   if (layout == CblasRowMajor)
-   {
-      if (strstr(rout,"gemm") != 0)
-      {
-         if      (info == 5 ) info =  4;
-         else if (info == 4 ) info =  5;
-         else if (info == 11) info =  9;
-         else if (info == 9 ) info = 11;
-      }
-      else if (strstr(rout,"symm") != 0 || strstr(rout,"hemm") != 0)
-      {
-         if      (info == 5 ) info =  4;
-         else if (info == 4 ) info =  5;
-      }
-      else if (strstr(rout,"trmm") != 0 || strstr(rout,"trsm") != 0)
-      {
-         if      (info == 7 ) info =  6;
-         else if (info == 6 ) info =  7;
-      }
-      else if (strstr(rout,"gemv") != 0)
-      {
-         if      (info == 4)  info = 3;
-         else if (info == 3)  info = 4;
-      }
-      else if (strstr(rout,"gbmv") != 0)
-      {
-         if      (info == 4)  info = 3;
-         else if (info == 3)  info = 4;
-         else if (info == 6)  info = 5;
-         else if (info == 5)  info = 6;
-      }
-      else if (strstr(rout,"ger") != 0)
-      {
-         if      (info == 3) info = 2;
-         else if (info == 2) info = 3;
-         else if (info == 8) info = 6;
-         else if (info == 6) info = 8;
-      }
-      else if ( (strstr(rout,"her2") != 0 || strstr(rout,"hpr2") != 0)
-                 && strstr(rout,"her2k") == 0 )
-      {
-         if      (info == 8) info = 6;
-         else if (info == 6) info = 8;
-      }
-   }
-   if (info)
-      fprintf(stderr, "Parameter %d to routine %s was incorrect\n", info, rout);
-   vfprintf(stderr, form, argptr);
-   va_end(argptr);
-   if (info && !info)
-      xerbla_(empty, &info); /* Force link of our F77 error handler */
-   exit(-1);
-}
@@ -1,72 +0,0 @@
-#include "f2c.h"
-#include <float.h>
-#include <stdio.h>
-
-/* *********************************************************************** */
-
-double dlamc3_(double *a, double *b)
-{
-    /*  -- LAPACK auxiliary routine (version 3.1) -- */
-    /*     Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
-    /*     November 2006 */
-
-    /*     .. Scalar Arguments .. */
-    /*     .. */
-
-    /*  Purpose */
-    /*  ======= */
-
-    /*  DLAMC3  is intended to force  A  and  B  to be stored prior to doing */
-    /*  the addition of  A  and  B ,  for use in situations where optimizers */
-    /*  might hold one of these in a register. */
-
-    /*  Arguments */
-    /*  ========= */
-
-    /*  A       (input) DOUBLE PRECISION */
-    /*  B       (input) DOUBLE PRECISION */
-    /*          The values A and B. */
-
-    /* ===================================================================== */
-
-    /*     .. Executable Statements .. */
-
-    double ret_val = *a + *b;
-
-    return ret_val;
-
-    /*     End of DLAMC3 */
-
-} /* dlamc3_ */
-
-
-/* simpler version of dlamch for the case of IEEE754-compliant FPU module by Piotr Luszczek S.
-   taken from http://www.mail-archive.com/numpy-discussion@lists.sourceforge.net/msg02448.html */
-
-#ifndef DBL_DIGITS
-#define DBL_DIGITS 53
-#endif
-
-static const unsigned char lapack_dlamch_tab0[] =
-{
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-    0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 2, 0, 0, 0, 0, 0, 0, 3, 4, 5, 6, 7, 0, 8, 9, 0, 10, 0,
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 2, 0, 0, 0, 0, 0, 0, 3, 4, 5, 6, 7, 0, 8, 9,
-    0, 10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
-};
-
-const double lapack_dlamch_tab1[] =
-{
-    0, FLT_RADIX, DBL_EPSILON, DBL_MAX_EXP, DBL_MIN_EXP, DBL_DIGITS, DBL_MAX,
-    DBL_EPSILON*FLT_RADIX, 1, DBL_MIN*(1 + DBL_EPSILON), DBL_MIN
-};
-
-double dlamch_(char* cmach)
-{
-    return lapack_dlamch_tab1[lapack_dlamch_tab0[(unsigned char)cmach[0]]];
-}
@@ -1,96 +0,0 @@
-#include "f2c.h"
-
-static const int CLAPACK_NOT_IMPLEMENTED = -1024;
-
-int sgesdd_(char *jobz, int *m, int *n, float *a, int *lda,
-	float *s, float *u, int *ldu, float *vt, int *ldvt, float *work,
-	int *lwork, int *iwork, int *info)
-{
-    *info = CLAPACK_NOT_IMPLEMENTED;
-    return 0;
-}
-
-int dgels_(char *trans, int *m, int *n, int *nrhs, double *a,
-	int *lda, double *b, int *ldb, double *work, int *lwork, int *info)
-{
-    *info = CLAPACK_NOT_IMPLEMENTED;
-    return 0;
-}
-
-int dgesv_(int *n, int *nrhs, double *a, int *lda, int *ipiv,
-	double *b, int *ldb, int *info)
-{
-    *info = CLAPACK_NOT_IMPLEMENTED;
-    return 0;
-}
-
-int dgetrf_(int *m, int *n, double *a, int *lda, int *ipiv,
-	int *info)
-{
-    *info = CLAPACK_NOT_IMPLEMENTED;
-    return 0;
-}
-
-int dposv_(char *uplo, int *n, int *nrhs, double *a, int *
-	lda, double *b, int *ldb, int *info)
-{
-    *info = CLAPACK_NOT_IMPLEMENTED;
-    return 0;
-}
-
-int dpotrf_(char *uplo, int *n, double *a, int *lda, int *
-	info)
-{
-    *info = CLAPACK_NOT_IMPLEMENTED;
-    return 0;
-}
-
-int sgels_(char *trans, int *m, int *n, int *nrhs, float *a,
-	int *lda, float *b, int *ldb, float *work, int *lwork, int *info)
-{
-    *info = CLAPACK_NOT_IMPLEMENTED;
-    return 0;
-}
-
-int sgeev_(char *jobvl, char *jobvr, int *n, float *a, int *
-	lda, float *wr, float *wi, float *vl, int *ldvl, float *vr, int *
-	ldvr, float *work, int *lwork, int *info)
-{
-    *info = CLAPACK_NOT_IMPLEMENTED;
-    return 0;
-}
-
-int sgeqrf_(int *m, int *n, float *a, int *lda, float *tau,
-	float *work, int *lwork, int *info)
-{
-    *info = CLAPACK_NOT_IMPLEMENTED;
-    return 0;
-}
-
-int sgesv_(int *n, int *nrhs, float *a, int *lda, int *ipiv,
-	float *b, int *ldb, int *info)
-{
-    *info = CLAPACK_NOT_IMPLEMENTED;
-    return 0;
-}        
-
-int sgetrf_(int *m, int *n, float *a, int *lda, int *ipiv,
-	int *info)
-{
-    *info = CLAPACK_NOT_IMPLEMENTED;
-    return 0;
-}
-
-int sposv_(char *uplo, int *n, int *nrhs, float *a, int *
-	lda, float *b, int *ldb, int *info)
-{
-    *info = CLAPACK_NOT_IMPLEMENTED;
-    return 0;
-}        
-
-int spotrf_(char *uplo, int *n, float *a, int *lda, int *
-	info)
-{
-    *info = CLAPACK_NOT_IMPLEMENTED;
-    return 0;
-}
@@ -1,25 +0,0 @@
-#include "f2c.h"
-
-static const unsigned char lapack_toupper_tab[] =
-{
-    0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23,
-    24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45,
-    46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67,
-    68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89,
-    90, 91, 92, 93, 94, 95, 96, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79,
-    80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 123, 124, 125, 126, 127, 128, 129, 130, 131,
-    132, 133, 134, 135, 136, 137, 138, 139, 140, 141, 142, 143, 144, 145, 146, 147, 148, 149,
-    150, 151, 152, 153, 154, 155, 156, 157, 158, 159, 160, 161, 162, 163, 164, 165, 166, 167,
-    168, 169, 170, 171, 172, 173, 174, 175, 176, 177, 178, 179, 180, 181, 182, 183, 184, 185,
-    186, 187, 188, 189, 190, 191, 192, 193, 194, 195, 196, 197, 198, 199, 200, 201, 202, 203,
-    204, 205, 206, 207, 208, 209, 210, 211, 212, 213, 214, 215, 216, 217, 218, 219, 220, 221,
-    222, 223, 224, 225, 226, 227, 228, 229, 230, 231, 232, 233, 234, 235, 236, 237, 238, 239,
-    240, 241, 242, 243, 244, 245, 246, 247, 248, 249, 250, 251, 252, 253, 254, 255
-};
-
-#define lapack_toupper(c) ((char)lapack_toupper_tab[(unsigned char)(c)])
-
-int lsame_(char *ca, char *cb)
-{
-    return lapack_toupper(ca[0]) == lapack_toupper(cb[0]);
-}
@@ -1,27 +0,0 @@
-#include "f2c.h"
-
-double pow_di(double *ap, int *bp)
-{
-    double p = 1;
-    double x = *ap;
-    int n = *bp;
-
-    if(n != 0)
-    {
-        if(n < 0)
-        {
-            n = -n;
-            x = 1/x;
-        }
-        unsigned u = (unsigned)n;
-        for(;;)
-        {
-            if((u & 1) != 0)
-                p *= x;
-            if((u >>= 1) == 0)
-                break;
-            x *= x;
-        }
-    }
-    return p;
-}
@@ -1,25 +0,0 @@
-#include "f2c.h"
-
-int pow_ii(int *ap, int *bp)
-{
-    int p;
-    int x = *ap;
-    int n = *bp;
-
-    if (n <= 0) {
-        if (n == 0 || x == 1)
-            return 1;
-        return x != -1 ? 0 : (n & 1) ? -1 : 1;
-    }
-    unsigned u = (unsigned)n;
-    for(p = 1; ; )
-    {
-        if(u & 01)
-            p *= x;
-        if(u >>= 1)
-            x *= x;
-        else
-            break;
-    }
-    return p;
-}
@@ -1,22 +0,0 @@
-/* Unless compiled with -DNO_OVERWRITE, this variant of s_cat allows the
- * target of a concatenation to appear on its right-hand side (contrary
- * to the Fortran 77 Standard, but in accordance with Fortran 90).
- */
-
-#include "f2c.h"
-
-int s_cat(char *lp, char **rpp, int* rnp, int *np)
-{
-    int i, L = 0;
-    int n = *np;
-
-    for(i = 0; i < n; i++) {
-        int ni = rnp[i];
-        if(ni > 0) {
-            memcpy(lp + L, rpp[i], ni);
-            L += ni;
-        }
-    }
-    lp[L] = '\0';
-    return 0;
-}
@@ -1,40 +0,0 @@
-#include "f2c.h"
-
-/* compare two strings */
-int s_cmp(char *a0, char *b0)
-{
-    int la = (int)strlen(a0);
-    int lb = (int)strlen(b0);
-    unsigned char *a, *aend, *b, *bend;
-    a = (unsigned char *)a0;
-    b = (unsigned char *)b0;
-    aend = a + la;
-    bend = b + lb;
-
-    if(la <= lb)
-    	{
-    	while(a < aend)
-    		if(*a != *b)
-    			return( *a - *b );
-    		else
-    			{ ++a; ++b; }
-
-    	while(b < bend)
-    		if(*b != ' ')
-    			return( ' ' - *b );
-    		else	++b;
-    	}
-    else
-    	{
-    	while(b < bend)
-    		if(*a == *b)
-    			{ ++a; ++b; }
-    		else
-    			return( *a - *b );
-    	while(a < aend)
-    		if(*a != ' ')
-    			return(*a - ' ');
-    		else	++a;
-    	}
-    return(0);
-}
@@ -1,71 +0,0 @@
-#include "f2c.h"
-#include <float.h>
-#include <stdio.h>
-
-/* *********************************************************************** */
-
-double slamc3_(float *a, float *b)
-{
-    /*  -- LAPACK auxiliary routine (version 3.1) -- */
-    /*     Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
-    /*     November 2006 */
-
-    /*     .. Scalar Arguments .. */
-    /*     .. */
-
-    /*  Purpose */
-    /*  ======= */
-
-    /*  SLAMC3  is intended to force  A  and  B  to be stored prior to doing */
-    /*  the addition of  A  and  B ,  for use in situations where optimizers */
-    /*  might hold one of these in a register. */
-
-    /*  Arguments */
-    /*  ========= */
-
-    /*  A       (input) REAL */
-    /*  B       (input) REAL */
-    /*          The values A and B. */
-
-    /* ===================================================================== */
-
-    /*     .. Executable Statements .. */
-
-    float ret_val = *a + *b;
-
-    return ret_val;
-
-    /*     End of SLAMC3 */
-
-} /* slamc3_ */
-
-/* simpler version of slamch for the case of IEEE754-compliant FPU module by Piotr Luszczek S.
- taken from http://www.mail-archive.com/numpy-discussion@lists.sourceforge.net/msg02448.html */
-
-#ifndef FLT_DIGITS
-#define FLT_DIGITS 24
-#endif
-
-static const unsigned char lapack_slamch_tab0[] =
-{
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-    0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 2, 0, 0, 0, 0, 0, 0, 3, 4, 5, 6, 7, 0, 8, 9, 0, 10, 0,
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 2, 0, 0, 0, 0, 0, 0, 3, 4, 5, 6, 7, 0, 8, 9,
-    0, 10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
-};
-
-const double lapack_slamch_tab1[] =
-{
-    0, FLT_RADIX, FLT_EPSILON, FLT_MAX_EXP, FLT_MIN_EXP, FLT_DIGITS, FLT_MAX,
-    FLT_EPSILON*FLT_RADIX, 1, FLT_MIN*(1 + FLT_EPSILON), FLT_MIN
-};
-
-double slamch_(char* cmach)
-{
-    return lapack_slamch_tab1[lapack_slamch_tab0[(unsigned char)cmach[0]]];
-}
@@ -1,19 +0,0 @@
-/* xerbla.f -- translated by f2c (version 20061008).
-   You must link the resulting object file with libf2c:
-	on Microsoft Windows system, link with libf2c.lib;
-	on Linux or Unix systems, link with .../path/to/libf2c.a -lm
-	or, if you install libf2c.a in a standard place, with -lf2c -lm
-	-- in that order, at the end of the command line, as in
-		cc *.o -lf2c -lm
-	Source for libf2c is in /netlib/f2c/libf2c.zip, e.g.,
-
-		http://www.netlib.org/f2c/libf2c.zip
-*/
-
-#include "f2c.h"
-
-/* Subroutine */ int xerbla_(char *srname, int *info)
-{
-    printf("** On entry to %s, parameter number %2i had an illegal value\n", srname, *info);
-    return 0;
-} /* xerbla_ */
@@ -1,752 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b CGEMM
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE CGEMM(TRANSA,TRANSB,M,N,K,ALPHA,A,LDA,B,LDB,BETA,C,LDC)
-//
-//      .. Scalar Arguments ..
-//      COMPLEX ALPHA,BETA
-//      INTEGER K,LDA,LDB,LDC,M,N
-//      CHARACTER TRANSA,TRANSB
-//      ..
-//      .. Array Arguments ..
-//      COMPLEX A(LDA,*),B(LDB,*),C(LDC,*)
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> CGEMM  performs one of the matrix-matrix operations
-//>
-//>    C := alpha*op( A )*op( B ) + beta*C,
-//>
-//> where  op( X ) is one of
-//>
-//>    op( X ) = X   or   op( X ) = X**T   or   op( X ) = X**H,
-//>
-//> alpha and beta are scalars, and A, B and C are matrices, with op( A )
-//> an m by k matrix,  op( B )  a  k by n matrix and  C an m by n matrix.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] TRANSA
-//> \verbatim
-//>          TRANSA is CHARACTER*1
-//>           On entry, TRANSA specifies the form of op( A ) to be used in
-//>           the matrix multiplication as follows:
-//>
-//>              TRANSA = 'N' or 'n',  op( A ) = A.
-//>
-//>              TRANSA = 'T' or 't',  op( A ) = A**T.
-//>
-//>              TRANSA = 'C' or 'c',  op( A ) = A**H.
-//> \endverbatim
-//>
-//> \param[in] TRANSB
-//> \verbatim
-//>          TRANSB is CHARACTER*1
-//>           On entry, TRANSB specifies the form of op( B ) to be used in
-//>           the matrix multiplication as follows:
-//>
-//>              TRANSB = 'N' or 'n',  op( B ) = B.
-//>
-//>              TRANSB = 'T' or 't',  op( B ) = B**T.
-//>
-//>              TRANSB = 'C' or 'c',  op( B ) = B**H.
-//> \endverbatim
-//>
-//> \param[in] M
-//> \verbatim
-//>          M is INTEGER
-//>           On entry,  M  specifies  the number  of rows  of the  matrix
-//>           op( A )  and of the  matrix  C.  M  must  be at least  zero.
-//> \endverbatim
-//>
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>           On entry,  N  specifies the number  of columns of the matrix
-//>           op( B ) and the number of columns of the matrix C. N must be
-//>           at least zero.
-//> \endverbatim
-//>
-//> \param[in] K
-//> \verbatim
-//>          K is INTEGER
-//>           On entry,  K  specifies  the number of columns of the matrix
-//>           op( A ) and the number of rows of the matrix op( B ). K must
-//>           be at least  zero.
-//> \endverbatim
-//>
-//> \param[in] ALPHA
-//> \verbatim
-//>          ALPHA is COMPLEX
-//>           On entry, ALPHA specifies the scalar alpha.
-//> \endverbatim
-//>
-//> \param[in] A
-//> \verbatim
-//>          A is COMPLEX array, dimension ( LDA, ka ), where ka is
-//>           k  when  TRANSA = 'N' or 'n',  and is  m  otherwise.
-//>           Before entry with  TRANSA = 'N' or 'n',  the leading  m by k
-//>           part of the array  A  must contain the matrix  A,  otherwise
-//>           the leading  k by m  part of the array  A  must contain  the
-//>           matrix A.
-//> \endverbatim
-//>
-//> \param[in] LDA
-//> \verbatim
-//>          LDA is INTEGER
-//>           On entry, LDA specifies the first dimension of A as declared
-//>           in the calling (sub) program. When  TRANSA = 'N' or 'n' then
-//>           LDA must be at least  max( 1, m ), otherwise  LDA must be at
-//>           least  max( 1, k ).
-//> \endverbatim
-//>
-//> \param[in] B
-//> \verbatim
-//>          B is COMPLEX array, dimension ( LDB, kb ), where kb is
-//>           n  when  TRANSB = 'N' or 'n',  and is  k  otherwise.
-//>           Before entry with  TRANSB = 'N' or 'n',  the leading  k by n
-//>           part of the array  B  must contain the matrix  B,  otherwise
-//>           the leading  n by k  part of the array  B  must contain  the
-//>           matrix B.
-//> \endverbatim
-//>
-//> \param[in] LDB
-//> \verbatim
-//>          LDB is INTEGER
-//>           On entry, LDB specifies the first dimension of B as declared
-//>           in the calling (sub) program. When  TRANSB = 'N' or 'n' then
-//>           LDB must be at least  max( 1, k ), otherwise  LDB must be at
-//>           least  max( 1, n ).
-//> \endverbatim
-//>
-//> \param[in] BETA
-//> \verbatim
-//>          BETA is COMPLEX
-//>           On entry,  BETA  specifies the scalar  beta.  When  BETA  is
-//>           supplied as zero then C need not be set on input.
-//> \endverbatim
-//>
-//> \param[in,out] C
-//> \verbatim
-//>          C is COMPLEX array, dimension ( LDC, N )
-//>           Before entry, the leading  m by n  part of the array  C must
-//>           contain the matrix  C,  except when  beta  is zero, in which
-//>           case C need not be set on entry.
-//>           On exit, the array  C  is overwritten by the  m by n  matrix
-//>           ( alpha*op( A )*op( B ) + beta*C ).
-//> \endverbatim
-//>
-//> \param[in] LDC
-//> \verbatim
-//>          LDC is INTEGER
-//>           On entry, LDC specifies the first dimension of C as declared
-//>           in  the  calling  (sub)  program.   LDC  must  be  at  least
-//>           max( 1, m ).
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date December 2016
-//
-//> \ingroup complex_blas_level3
-//
-//> \par Further Details:
-// =====================
-//>
-//> \verbatim
-//>
-//>  Level 3 Blas routine.
-//>
-//>  -- Written on 8-February-1989.
-//>     Jack Dongarra, Argonne National Laboratory.
-//>     Iain Duff, AERE Harwell.
-//>     Jeremy Du Croz, Numerical Algorithms Group Ltd.
-//>     Sven Hammarling, Numerical Algorithms Group Ltd.
-//> \endverbatim
-//>
-// =====================================================================
-/* Subroutine */ int cgemm_(char *transa, char *transb, int *m, int *n, int *
-	k, complex *alpha, complex *a, int *lda, complex *b, int *ldb,
-	complex *beta, complex *c__, int *ldc)
-{
-    // Table of constant values
-    complex c_b1 = {1.f,0.f};
-    complex c_b2 = {0.f,0.f};
-
-    // System generated locals
-    int a_dim1, a_offset, b_dim1, b_offset, c_dim1, c_offset, i__1, i__2,
-	    i__3, i__4, i__5, i__6;
-    complex q__1, q__2, q__3, q__4;
-
-    // Local variables
-    int i__, j, l, info;
-    int nota, notb;
-    complex temp;
-    int conja, conjb;
-    int ncola;
-    extern int lsame_(char *, char *);
-    int nrowa, nrowb;
-    extern /* Subroutine */ int xerbla_(char *, int *);
-
-    //
-    // -- Reference BLAS level3 routine (version 3.7.0) --
-    // -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    December 2016
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. External Functions ..
-    //    ..
-    //    .. External Subroutines ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    //    .. Local Scalars ..
-    //    ..
-    //    .. Parameters ..
-    //    ..
-    //
-    //    Set  NOTA  and  NOTB  as  true if  A  and  B  respectively are not
-    //    conjugated or transposed, set  CONJA and CONJB  as true if  A  and
-    //    B  respectively are to be  transposed but  not conjugated  and set
-    //    NROWA, NCOLA and  NROWB  as the number of rows and  columns  of  A
-    //    and the number of rows of  B  respectively.
-    //
-    // Parameter adjustments
-    a_dim1 = *lda;
-    a_offset = 1 + a_dim1;
-    a -= a_offset;
-    b_dim1 = *ldb;
-    b_offset = 1 + b_dim1;
-    b -= b_offset;
-    c_dim1 = *ldc;
-    c_offset = 1 + c_dim1;
-    c__ -= c_offset;
-
-    // Function Body
-    nota = lsame_(transa, "N");
-    notb = lsame_(transb, "N");
-    conja = lsame_(transa, "C");
-    conjb = lsame_(transb, "C");
-    if (nota) {
-	nrowa = *m;
-	ncola = *k;
-    } else {
-	nrowa = *k;
-	ncola = *m;
-    }
-    if (notb) {
-	nrowb = *k;
-    } else {
-	nrowb = *n;
-    }
-    //
-    //    Test the input parameters.
-    //
-    info = 0;
-    if (! nota && ! conja && ! lsame_(transa, "T")) {
-	info = 1;
-    } else if (! notb && ! conjb && ! lsame_(transb, "T")) {
-	info = 2;
-    } else if (*m < 0) {
-	info = 3;
-    } else if (*n < 0) {
-	info = 4;
-    } else if (*k < 0) {
-	info = 5;
-    } else if (*lda < max(1,nrowa)) {
-	info = 8;
-    } else if (*ldb < max(1,nrowb)) {
-	info = 10;
-    } else if (*ldc < max(1,*m)) {
-	info = 13;
-    }
-    if (info != 0) {
-	xerbla_("CGEMM ", &info);
-	return 0;
-    }
-    //
-    //    Quick return if possible.
-    //
-    if (*m == 0 || *n == 0 || (alpha->r == 0.f && alpha->i == 0.f || *k == 0)
-	    && (beta->r == 1.f && beta->i == 0.f)) {
-	return 0;
-    }
-    //
-    //    And when  alpha.eq.zero.
-    //
-    if (alpha->r == 0.f && alpha->i == 0.f) {
-	if (beta->r == 0.f && beta->i == 0.f) {
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    i__3 = i__ + j * c_dim1;
-		    c__[i__3].r = 0.f, c__[i__3].i = 0.f;
-// L10:
-		}
-// L20:
-	    }
-	} else {
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    i__3 = i__ + j * c_dim1;
-		    i__4 = i__ + j * c_dim1;
-		    q__1.r = beta->r * c__[i__4].r - beta->i * c__[i__4].i,
-			    q__1.i = beta->r * c__[i__4].i + beta->i * c__[
-			    i__4].r;
-		    c__[i__3].r = q__1.r, c__[i__3].i = q__1.i;
-// L30:
-		}
-// L40:
-	    }
-	}
-	return 0;
-    }
-    //
-    //    Start the operations.
-    //
-    if (notb) {
-	if (nota) {
-	    //
-	    //          Form  C := alpha*A*B + beta*C.
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		if (beta->r == 0.f && beta->i == 0.f) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			i__3 = i__ + j * c_dim1;
-			c__[i__3].r = 0.f, c__[i__3].i = 0.f;
-// L50:
-		    }
-		} else if (beta->r != 1.f || beta->i != 0.f) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			i__3 = i__ + j * c_dim1;
-			i__4 = i__ + j * c_dim1;
-			q__1.r = beta->r * c__[i__4].r - beta->i * c__[i__4]
-				.i, q__1.i = beta->r * c__[i__4].i + beta->i *
-				 c__[i__4].r;
-			c__[i__3].r = q__1.r, c__[i__3].i = q__1.i;
-// L60:
-		    }
-		}
-		i__2 = *k;
-		for (l = 1; l <= i__2; ++l) {
-		    i__3 = l + j * b_dim1;
-		    q__1.r = alpha->r * b[i__3].r - alpha->i * b[i__3].i,
-			    q__1.i = alpha->r * b[i__3].i + alpha->i * b[i__3]
-			    .r;
-		    temp.r = q__1.r, temp.i = q__1.i;
-		    i__3 = *m;
-		    for (i__ = 1; i__ <= i__3; ++i__) {
-			i__4 = i__ + j * c_dim1;
-			i__5 = i__ + j * c_dim1;
-			i__6 = i__ + l * a_dim1;
-			q__2.r = temp.r * a[i__6].r - temp.i * a[i__6].i,
-				q__2.i = temp.r * a[i__6].i + temp.i * a[i__6]
-				.r;
-			q__1.r = c__[i__5].r + q__2.r, q__1.i = c__[i__5].i +
-				q__2.i;
-			c__[i__4].r = q__1.r, c__[i__4].i = q__1.i;
-// L70:
-		    }
-// L80:
-		}
-// L90:
-	    }
-	} else if (conja) {
-	    //
-	    //          Form  C := alpha*A**H*B + beta*C.
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    temp.r = 0.f, temp.i = 0.f;
-		    i__3 = *k;
-		    for (l = 1; l <= i__3; ++l) {
-			r_cnjg(&q__3, &a[l + i__ * a_dim1]);
-			i__4 = l + j * b_dim1;
-			q__2.r = q__3.r * b[i__4].r - q__3.i * b[i__4].i,
-				q__2.i = q__3.r * b[i__4].i + q__3.i * b[i__4]
-				.r;
-			q__1.r = temp.r + q__2.r, q__1.i = temp.i + q__2.i;
-			temp.r = q__1.r, temp.i = q__1.i;
-// L100:
-		    }
-		    if (beta->r == 0.f && beta->i == 0.f) {
-			i__3 = i__ + j * c_dim1;
-			q__1.r = alpha->r * temp.r - alpha->i * temp.i,
-				q__1.i = alpha->r * temp.i + alpha->i *
-				temp.r;
-			c__[i__3].r = q__1.r, c__[i__3].i = q__1.i;
-		    } else {
-			i__3 = i__ + j * c_dim1;
-			q__2.r = alpha->r * temp.r - alpha->i * temp.i,
-				q__2.i = alpha->r * temp.i + alpha->i *
-				temp.r;
-			i__4 = i__ + j * c_dim1;
-			q__3.r = beta->r * c__[i__4].r - beta->i * c__[i__4]
-				.i, q__3.i = beta->r * c__[i__4].i + beta->i *
-				 c__[i__4].r;
-			q__1.r = q__2.r + q__3.r, q__1.i = q__2.i + q__3.i;
-			c__[i__3].r = q__1.r, c__[i__3].i = q__1.i;
-		    }
-// L110:
-		}
-// L120:
-	    }
-	} else {
-	    //
-	    //          Form  C := alpha*A**T*B + beta*C
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    temp.r = 0.f, temp.i = 0.f;
-		    i__3 = *k;
-		    for (l = 1; l <= i__3; ++l) {
-			i__4 = l + i__ * a_dim1;
-			i__5 = l + j * b_dim1;
-			q__2.r = a[i__4].r * b[i__5].r - a[i__4].i * b[i__5]
-				.i, q__2.i = a[i__4].r * b[i__5].i + a[i__4]
-				.i * b[i__5].r;
-			q__1.r = temp.r + q__2.r, q__1.i = temp.i + q__2.i;
-			temp.r = q__1.r, temp.i = q__1.i;
-// L130:
-		    }
-		    if (beta->r == 0.f && beta->i == 0.f) {
-			i__3 = i__ + j * c_dim1;
-			q__1.r = alpha->r * temp.r - alpha->i * temp.i,
-				q__1.i = alpha->r * temp.i + alpha->i *
-				temp.r;
-			c__[i__3].r = q__1.r, c__[i__3].i = q__1.i;
-		    } else {
-			i__3 = i__ + j * c_dim1;
-			q__2.r = alpha->r * temp.r - alpha->i * temp.i,
-				q__2.i = alpha->r * temp.i + alpha->i *
-				temp.r;
-			i__4 = i__ + j * c_dim1;
-			q__3.r = beta->r * c__[i__4].r - beta->i * c__[i__4]
-				.i, q__3.i = beta->r * c__[i__4].i + beta->i *
-				 c__[i__4].r;
-			q__1.r = q__2.r + q__3.r, q__1.i = q__2.i + q__3.i;
-			c__[i__3].r = q__1.r, c__[i__3].i = q__1.i;
-		    }
-// L140:
-		}
-// L150:
-	    }
-	}
-    } else if (nota) {
-	if (conjb) {
-	    //
-	    //          Form  C := alpha*A*B**H + beta*C.
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		if (beta->r == 0.f && beta->i == 0.f) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			i__3 = i__ + j * c_dim1;
-			c__[i__3].r = 0.f, c__[i__3].i = 0.f;
-// L160:
-		    }
-		} else if (beta->r != 1.f || beta->i != 0.f) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			i__3 = i__ + j * c_dim1;
-			i__4 = i__ + j * c_dim1;
-			q__1.r = beta->r * c__[i__4].r - beta->i * c__[i__4]
-				.i, q__1.i = beta->r * c__[i__4].i + beta->i *
-				 c__[i__4].r;
-			c__[i__3].r = q__1.r, c__[i__3].i = q__1.i;
-// L170:
-		    }
-		}
-		i__2 = *k;
-		for (l = 1; l <= i__2; ++l) {
-		    r_cnjg(&q__2, &b[j + l * b_dim1]);
-		    q__1.r = alpha->r * q__2.r - alpha->i * q__2.i, q__1.i =
-			    alpha->r * q__2.i + alpha->i * q__2.r;
-		    temp.r = q__1.r, temp.i = q__1.i;
-		    i__3 = *m;
-		    for (i__ = 1; i__ <= i__3; ++i__) {
-			i__4 = i__ + j * c_dim1;
-			i__5 = i__ + j * c_dim1;
-			i__6 = i__ + l * a_dim1;
-			q__2.r = temp.r * a[i__6].r - temp.i * a[i__6].i,
-				q__2.i = temp.r * a[i__6].i + temp.i * a[i__6]
-				.r;
-			q__1.r = c__[i__5].r + q__2.r, q__1.i = c__[i__5].i +
-				q__2.i;
-			c__[i__4].r = q__1.r, c__[i__4].i = q__1.i;
-// L180:
-		    }
-// L190:
-		}
-// L200:
-	    }
-	} else {
-	    //
-	    //          Form  C := alpha*A*B**T + beta*C
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		if (beta->r == 0.f && beta->i == 0.f) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			i__3 = i__ + j * c_dim1;
-			c__[i__3].r = 0.f, c__[i__3].i = 0.f;
-// L210:
-		    }
-		} else if (beta->r != 1.f || beta->i != 0.f) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			i__3 = i__ + j * c_dim1;
-			i__4 = i__ + j * c_dim1;
-			q__1.r = beta->r * c__[i__4].r - beta->i * c__[i__4]
-				.i, q__1.i = beta->r * c__[i__4].i + beta->i *
-				 c__[i__4].r;
-			c__[i__3].r = q__1.r, c__[i__3].i = q__1.i;
-// L220:
-		    }
-		}
-		i__2 = *k;
-		for (l = 1; l <= i__2; ++l) {
-		    i__3 = j + l * b_dim1;
-		    q__1.r = alpha->r * b[i__3].r - alpha->i * b[i__3].i,
-			    q__1.i = alpha->r * b[i__3].i + alpha->i * b[i__3]
-			    .r;
-		    temp.r = q__1.r, temp.i = q__1.i;
-		    i__3 = *m;
-		    for (i__ = 1; i__ <= i__3; ++i__) {
-			i__4 = i__ + j * c_dim1;
-			i__5 = i__ + j * c_dim1;
-			i__6 = i__ + l * a_dim1;
-			q__2.r = temp.r * a[i__6].r - temp.i * a[i__6].i,
-				q__2.i = temp.r * a[i__6].i + temp.i * a[i__6]
-				.r;
-			q__1.r = c__[i__5].r + q__2.r, q__1.i = c__[i__5].i +
-				q__2.i;
-			c__[i__4].r = q__1.r, c__[i__4].i = q__1.i;
-// L230:
-		    }
-// L240:
-		}
-// L250:
-	    }
-	}
-    } else if (conja) {
-	if (conjb) {
-	    //
-	    //          Form  C := alpha*A**H*B**H + beta*C.
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    temp.r = 0.f, temp.i = 0.f;
-		    i__3 = *k;
-		    for (l = 1; l <= i__3; ++l) {
-			r_cnjg(&q__3, &a[l + i__ * a_dim1]);
-			r_cnjg(&q__4, &b[j + l * b_dim1]);
-			q__2.r = q__3.r * q__4.r - q__3.i * q__4.i, q__2.i =
-				q__3.r * q__4.i + q__3.i * q__4.r;
-			q__1.r = temp.r + q__2.r, q__1.i = temp.i + q__2.i;
-			temp.r = q__1.r, temp.i = q__1.i;
-// L260:
-		    }
-		    if (beta->r == 0.f && beta->i == 0.f) {
-			i__3 = i__ + j * c_dim1;
-			q__1.r = alpha->r * temp.r - alpha->i * temp.i,
-				q__1.i = alpha->r * temp.i + alpha->i *
-				temp.r;
-			c__[i__3].r = q__1.r, c__[i__3].i = q__1.i;
-		    } else {
-			i__3 = i__ + j * c_dim1;
-			q__2.r = alpha->r * temp.r - alpha->i * temp.i,
-				q__2.i = alpha->r * temp.i + alpha->i *
-				temp.r;
-			i__4 = i__ + j * c_dim1;
-			q__3.r = beta->r * c__[i__4].r - beta->i * c__[i__4]
-				.i, q__3.i = beta->r * c__[i__4].i + beta->i *
-				 c__[i__4].r;
-			q__1.r = q__2.r + q__3.r, q__1.i = q__2.i + q__3.i;
-			c__[i__3].r = q__1.r, c__[i__3].i = q__1.i;
-		    }
-// L270:
-		}
-// L280:
-	    }
-	} else {
-	    //
-	    //          Form  C := alpha*A**H*B**T + beta*C
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    temp.r = 0.f, temp.i = 0.f;
-		    i__3 = *k;
-		    for (l = 1; l <= i__3; ++l) {
-			r_cnjg(&q__3, &a[l + i__ * a_dim1]);
-			i__4 = j + l * b_dim1;
-			q__2.r = q__3.r * b[i__4].r - q__3.i * b[i__4].i,
-				q__2.i = q__3.r * b[i__4].i + q__3.i * b[i__4]
-				.r;
-			q__1.r = temp.r + q__2.r, q__1.i = temp.i + q__2.i;
-			temp.r = q__1.r, temp.i = q__1.i;
-// L290:
-		    }
-		    if (beta->r == 0.f && beta->i == 0.f) {
-			i__3 = i__ + j * c_dim1;
-			q__1.r = alpha->r * temp.r - alpha->i * temp.i,
-				q__1.i = alpha->r * temp.i + alpha->i *
-				temp.r;
-			c__[i__3].r = q__1.r, c__[i__3].i = q__1.i;
-		    } else {
-			i__3 = i__ + j * c_dim1;
-			q__2.r = alpha->r * temp.r - alpha->i * temp.i,
-				q__2.i = alpha->r * temp.i + alpha->i *
-				temp.r;
-			i__4 = i__ + j * c_dim1;
-			q__3.r = beta->r * c__[i__4].r - beta->i * c__[i__4]
-				.i, q__3.i = beta->r * c__[i__4].i + beta->i *
-				 c__[i__4].r;
-			q__1.r = q__2.r + q__3.r, q__1.i = q__2.i + q__3.i;
-			c__[i__3].r = q__1.r, c__[i__3].i = q__1.i;
-		    }
-// L300:
-		}
-// L310:
-	    }
-	}
-    } else {
-	if (conjb) {
-	    //
-	    //          Form  C := alpha*A**T*B**H + beta*C
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    temp.r = 0.f, temp.i = 0.f;
-		    i__3 = *k;
-		    for (l = 1; l <= i__3; ++l) {
-			i__4 = l + i__ * a_dim1;
-			r_cnjg(&q__3, &b[j + l * b_dim1]);
-			q__2.r = a[i__4].r * q__3.r - a[i__4].i * q__3.i,
-				q__2.i = a[i__4].r * q__3.i + a[i__4].i *
-				q__3.r;
-			q__1.r = temp.r + q__2.r, q__1.i = temp.i + q__2.i;
-			temp.r = q__1.r, temp.i = q__1.i;
-// L320:
-		    }
-		    if (beta->r == 0.f && beta->i == 0.f) {
-			i__3 = i__ + j * c_dim1;
-			q__1.r = alpha->r * temp.r - alpha->i * temp.i,
-				q__1.i = alpha->r * temp.i + alpha->i *
-				temp.r;
-			c__[i__3].r = q__1.r, c__[i__3].i = q__1.i;
-		    } else {
-			i__3 = i__ + j * c_dim1;
-			q__2.r = alpha->r * temp.r - alpha->i * temp.i,
-				q__2.i = alpha->r * temp.i + alpha->i *
-				temp.r;
-			i__4 = i__ + j * c_dim1;
-			q__3.r = beta->r * c__[i__4].r - beta->i * c__[i__4]
-				.i, q__3.i = beta->r * c__[i__4].i + beta->i *
-				 c__[i__4].r;
-			q__1.r = q__2.r + q__3.r, q__1.i = q__2.i + q__3.i;
-			c__[i__3].r = q__1.r, c__[i__3].i = q__1.i;
-		    }
-// L330:
-		}
-// L340:
-	    }
-	} else {
-	    //
-	    //          Form  C := alpha*A**T*B**T + beta*C
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    temp.r = 0.f, temp.i = 0.f;
-		    i__3 = *k;
-		    for (l = 1; l <= i__3; ++l) {
-			i__4 = l + i__ * a_dim1;
-			i__5 = j + l * b_dim1;
-			q__2.r = a[i__4].r * b[i__5].r - a[i__4].i * b[i__5]
-				.i, q__2.i = a[i__4].r * b[i__5].i + a[i__4]
-				.i * b[i__5].r;
-			q__1.r = temp.r + q__2.r, q__1.i = temp.i + q__2.i;
-			temp.r = q__1.r, temp.i = q__1.i;
-// L350:
-		    }
-		    if (beta->r == 0.f && beta->i == 0.f) {
-			i__3 = i__ + j * c_dim1;
-			q__1.r = alpha->r * temp.r - alpha->i * temp.i,
-				q__1.i = alpha->r * temp.i + alpha->i *
-				temp.r;
-			c__[i__3].r = q__1.r, c__[i__3].i = q__1.i;
-		    } else {
-			i__3 = i__ + j * c_dim1;
-			q__2.r = alpha->r * temp.r - alpha->i * temp.i,
-				q__2.i = alpha->r * temp.i + alpha->i *
-				temp.r;
-			i__4 = i__ + j * c_dim1;
-			q__3.r = beta->r * c__[i__4].r - beta->i * c__[i__4]
-				.i, q__3.i = beta->r * c__[i__4].i + beta->i *
-				 c__[i__4].r;
-			q__1.r = q__2.r + q__3.r, q__1.i = q__2.i + q__3.i;
-			c__[i__3].r = q__1.r, c__[i__3].i = q__1.i;
-		    }
-// L360:
-		}
-// L370:
-	    }
-	}
-    }
-    return 0;
-    //
-    //    End of CGEMM .
-    //
-} // cgemm_
-
@@ -1,171 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b DCOPY
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE DCOPY(N,DX,INCX,DY,INCY)
-//
-//      .. Scalar Arguments ..
-//      INTEGER INCX,INCY,N
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION DX(*),DY(*)
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//>    DCOPY copies a vector, x, to a vector, y.
-//>    uses unrolled loops for increments equal to 1.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>         number of elements in input vector(s)
-//> \endverbatim
-//>
-//> \param[in] DX
-//> \verbatim
-//>          DX is DOUBLE PRECISION array, dimension ( 1 + ( N - 1 )*abs( INCX ) )
-//> \endverbatim
-//>
-//> \param[in] INCX
-//> \verbatim
-//>          INCX is INTEGER
-//>         storage spacing between elements of DX
-//> \endverbatim
-//>
-//> \param[out] DY
-//> \verbatim
-//>          DY is DOUBLE PRECISION array, dimension ( 1 + ( N - 1 )*abs( INCY ) )
-//> \endverbatim
-//>
-//> \param[in] INCY
-//> \verbatim
-//>          INCY is INTEGER
-//>         storage spacing between elements of DY
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date November 2017
-//
-//> \ingroup double_blas_level1
-//
-//> \par Further Details:
-// =====================
-//>
-//> \verbatim
-//>
-//>     jack dongarra, linpack, 3/11/78.
-//>     modified 12/3/93, array(1) declarations changed to array(*)
-//> \endverbatim
-//>
-// =====================================================================
-/* Subroutine */ int dcopy_(int *n, double *dx, int *incx, double *dy, int *
-	incy)
-{
-    // System generated locals
-    int i__1;
-
-    // Local variables
-    int i__, m, ix, iy, mp1;
-
-    //
-    // -- Reference BLAS level1 routine (version 3.8.0) --
-    // -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    November 2017
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. Local Scalars ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    // Parameter adjustments
-    --dy;
-    --dx;
-
-    // Function Body
-    if (*n <= 0) {
-	return 0;
-    }
-    if (*incx == 1 && *incy == 1) {
-	//
-	//       code for both increments equal to 1
-	//
-	//
-	//       clean-up loop
-	//
-	m = *n % 7;
-	if (m != 0) {
-	    i__1 = m;
-	    for (i__ = 1; i__ <= i__1; ++i__) {
-		dy[i__] = dx[i__];
-	    }
-	    if (*n < 7) {
-		return 0;
-	    }
-	}
-	mp1 = m + 1;
-	i__1 = *n;
-	for (i__ = mp1; i__ <= i__1; i__ += 7) {
-	    dy[i__] = dx[i__];
-	    dy[i__ + 1] = dx[i__ + 1];
-	    dy[i__ + 2] = dx[i__ + 2];
-	    dy[i__ + 3] = dx[i__ + 3];
-	    dy[i__ + 4] = dx[i__ + 4];
-	    dy[i__ + 5] = dx[i__ + 5];
-	    dy[i__ + 6] = dx[i__ + 6];
-	}
-    } else {
-	//
-	//       code for unequal increments or equal increments
-	//         not equal to 1
-	//
-	ix = 1;
-	iy = 1;
-	if (*incx < 0) {
-	    ix = (-(*n) + 1) * *incx + 1;
-	}
-	if (*incy < 0) {
-	    iy = (-(*n) + 1) * *incy + 1;
-	}
-	i__1 = *n;
-	for (i__ = 1; i__ <= i__1; ++i__) {
-	    dy[iy] = dx[ix];
-	    ix += *incx;
-	    iy += *incy;
-	}
-    }
-    return 0;
-} // dcopy_
-
@@ -1,172 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b DDOT
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-// Definition:
-// ===========
-//
-//      DOUBLE PRECISION FUNCTION DDOT(N,DX,INCX,DY,INCY)
-//
-//      .. Scalar Arguments ..
-//      INTEGER INCX,INCY,N
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION DX(*),DY(*)
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//>    DDOT forms the dot product of two vectors.
-//>    uses unrolled loops for increments equal to one.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>         number of elements in input vector(s)
-//> \endverbatim
-//>
-//> \param[in] DX
-//> \verbatim
-//>          DX is DOUBLE PRECISION array, dimension ( 1 + ( N - 1 )*abs( INCX ) )
-//> \endverbatim
-//>
-//> \param[in] INCX
-//> \verbatim
-//>          INCX is INTEGER
-//>         storage spacing between elements of DX
-//> \endverbatim
-//>
-//> \param[in] DY
-//> \verbatim
-//>          DY is DOUBLE PRECISION array, dimension ( 1 + ( N - 1 )*abs( INCY ) )
-//> \endverbatim
-//>
-//> \param[in] INCY
-//> \verbatim
-//>          INCY is INTEGER
-//>         storage spacing between elements of DY
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date November 2017
-//
-//> \ingroup double_blas_level1
-//
-//> \par Further Details:
-// =====================
-//>
-//> \verbatim
-//>
-//>     jack dongarra, linpack, 3/11/78.
-//>     modified 12/3/93, array(1) declarations changed to array(*)
-//> \endverbatim
-//>
-// =====================================================================
-double ddot_(int *n, double *dx, int *incx, double *dy, int *incy)
-{
-    // System generated locals
-    int i__1;
-    double ret_val;
-
-    // Local variables
-    int i__, m, ix, iy, mp1;
-    double dtemp;
-
-    //
-    // -- Reference BLAS level1 routine (version 3.8.0) --
-    // -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    November 2017
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. Local Scalars ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    // Parameter adjustments
-    --dy;
-    --dx;
-
-    // Function Body
-    ret_val = 0.;
-    dtemp = 0.;
-    if (*n <= 0) {
-	return ret_val;
-    }
-    if (*incx == 1 && *incy == 1) {
-	//
-	//       code for both increments equal to 1
-	//
-	//
-	//       clean-up loop
-	//
-	m = *n % 5;
-	if (m != 0) {
-	    i__1 = m;
-	    for (i__ = 1; i__ <= i__1; ++i__) {
-		dtemp += dx[i__] * dy[i__];
-	    }
-	    if (*n < 5) {
-		ret_val = dtemp;
-		return ret_val;
-	    }
-	}
-	mp1 = m + 1;
-	i__1 = *n;
-	for (i__ = mp1; i__ <= i__1; i__ += 5) {
-	    dtemp = dtemp + dx[i__] * dy[i__] + dx[i__ + 1] * dy[i__ + 1] +
-		    dx[i__ + 2] * dy[i__ + 2] + dx[i__ + 3] * dy[i__ + 3] +
-		    dx[i__ + 4] * dy[i__ + 4];
-	}
-    } else {
-	//
-	//       code for unequal increments or equal increments
-	//         not equal to 1
-	//
-	ix = 1;
-	iy = 1;
-	if (*incx < 0) {
-	    ix = (-(*n) + 1) * *incx + 1;
-	}
-	if (*incy < 0) {
-	    iy = (-(*n) + 1) * *incy + 1;
-	}
-	i__1 = *n;
-	for (i__ = 1; i__ <= i__1; ++i__) {
-	    dtemp += dx[ix] * dy[iy];
-	    ix += *incx;
-	    iy += *incy;
-	}
-    }
-    ret_val = dtemp;
-    return ret_val;
-} // ddot_
-
@@ -1,444 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b DGEMM
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE DGEMM(TRANSA,TRANSB,M,N,K,ALPHA,A,LDA,B,LDB,BETA,C,LDC)
-//
-//      .. Scalar Arguments ..
-//      DOUBLE PRECISION ALPHA,BETA
-//      INTEGER K,LDA,LDB,LDC,M,N
-//      CHARACTER TRANSA,TRANSB
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION A(LDA,*),B(LDB,*),C(LDC,*)
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> DGEMM  performs one of the matrix-matrix operations
-//>
-//>    C := alpha*op( A )*op( B ) + beta*C,
-//>
-//> where  op( X ) is one of
-//>
-//>    op( X ) = X   or   op( X ) = X**T,
-//>
-//> alpha and beta are scalars, and A, B and C are matrices, with op( A )
-//> an m by k matrix,  op( B )  a  k by n matrix and  C an m by n matrix.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] TRANSA
-//> \verbatim
-//>          TRANSA is CHARACTER*1
-//>           On entry, TRANSA specifies the form of op( A ) to be used in
-//>           the matrix multiplication as follows:
-//>
-//>              TRANSA = 'N' or 'n',  op( A ) = A.
-//>
-//>              TRANSA = 'T' or 't',  op( A ) = A**T.
-//>
-//>              TRANSA = 'C' or 'c',  op( A ) = A**T.
-//> \endverbatim
-//>
-//> \param[in] TRANSB
-//> \verbatim
-//>          TRANSB is CHARACTER*1
-//>           On entry, TRANSB specifies the form of op( B ) to be used in
-//>           the matrix multiplication as follows:
-//>
-//>              TRANSB = 'N' or 'n',  op( B ) = B.
-//>
-//>              TRANSB = 'T' or 't',  op( B ) = B**T.
-//>
-//>              TRANSB = 'C' or 'c',  op( B ) = B**T.
-//> \endverbatim
-//>
-//> \param[in] M
-//> \verbatim
-//>          M is INTEGER
-//>           On entry,  M  specifies  the number  of rows  of the  matrix
-//>           op( A )  and of the  matrix  C.  M  must  be at least  zero.
-//> \endverbatim
-//>
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>           On entry,  N  specifies the number  of columns of the matrix
-//>           op( B ) and the number of columns of the matrix C. N must be
-//>           at least zero.
-//> \endverbatim
-//>
-//> \param[in] K
-//> \verbatim
-//>          K is INTEGER
-//>           On entry,  K  specifies  the number of columns of the matrix
-//>           op( A ) and the number of rows of the matrix op( B ). K must
-//>           be at least  zero.
-//> \endverbatim
-//>
-//> \param[in] ALPHA
-//> \verbatim
-//>          ALPHA is DOUBLE PRECISION.
-//>           On entry, ALPHA specifies the scalar alpha.
-//> \endverbatim
-//>
-//> \param[in] A
-//> \verbatim
-//>          A is DOUBLE PRECISION array, dimension ( LDA, ka ), where ka is
-//>           k  when  TRANSA = 'N' or 'n',  and is  m  otherwise.
-//>           Before entry with  TRANSA = 'N' or 'n',  the leading  m by k
-//>           part of the array  A  must contain the matrix  A,  otherwise
-//>           the leading  k by m  part of the array  A  must contain  the
-//>           matrix A.
-//> \endverbatim
-//>
-//> \param[in] LDA
-//> \verbatim
-//>          LDA is INTEGER
-//>           On entry, LDA specifies the first dimension of A as declared
-//>           in the calling (sub) program. When  TRANSA = 'N' or 'n' then
-//>           LDA must be at least  max( 1, m ), otherwise  LDA must be at
-//>           least  max( 1, k ).
-//> \endverbatim
-//>
-//> \param[in] B
-//> \verbatim
-//>          B is DOUBLE PRECISION array, dimension ( LDB, kb ), where kb is
-//>           n  when  TRANSB = 'N' or 'n',  and is  k  otherwise.
-//>           Before entry with  TRANSB = 'N' or 'n',  the leading  k by n
-//>           part of the array  B  must contain the matrix  B,  otherwise
-//>           the leading  n by k  part of the array  B  must contain  the
-//>           matrix B.
-//> \endverbatim
-//>
-//> \param[in] LDB
-//> \verbatim
-//>          LDB is INTEGER
-//>           On entry, LDB specifies the first dimension of B as declared
-//>           in the calling (sub) program. When  TRANSB = 'N' or 'n' then
-//>           LDB must be at least  max( 1, k ), otherwise  LDB must be at
-//>           least  max( 1, n ).
-//> \endverbatim
-//>
-//> \param[in] BETA
-//> \verbatim
-//>          BETA is DOUBLE PRECISION.
-//>           On entry,  BETA  specifies the scalar  beta.  When  BETA  is
-//>           supplied as zero then C need not be set on input.
-//> \endverbatim
-//>
-//> \param[in,out] C
-//> \verbatim
-//>          C is DOUBLE PRECISION array, dimension ( LDC, N )
-//>           Before entry, the leading  m by n  part of the array  C must
-//>           contain the matrix  C,  except when  beta  is zero, in which
-//>           case C need not be set on entry.
-//>           On exit, the array  C  is overwritten by the  m by n  matrix
-//>           ( alpha*op( A )*op( B ) + beta*C ).
-//> \endverbatim
-//>
-//> \param[in] LDC
-//> \verbatim
-//>          LDC is INTEGER
-//>           On entry, LDC specifies the first dimension of C as declared
-//>           in  the  calling  (sub)  program.   LDC  must  be  at  least
-//>           max( 1, m ).
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date December 2016
-//
-//> \ingroup double_blas_level3
-//
-//> \par Further Details:
-// =====================
-//>
-//> \verbatim
-//>
-//>  Level 3 Blas routine.
-//>
-//>  -- Written on 8-February-1989.
-//>     Jack Dongarra, Argonne National Laboratory.
-//>     Iain Duff, AERE Harwell.
-//>     Jeremy Du Croz, Numerical Algorithms Group Ltd.
-//>     Sven Hammarling, Numerical Algorithms Group Ltd.
-//> \endverbatim
-//>
-// =====================================================================
-/* Subroutine */ int dgemm_(char *transa, char *transb, int *m, int *n, int *
-	k, double *alpha, double *a, int *lda, double *b, int *ldb, double *
-	beta, double *c__, int *ldc)
-{
-    // System generated locals
-    int a_dim1, a_offset, b_dim1, b_offset, c_dim1, c_offset, i__1, i__2,
-	    i__3;
-
-    // Local variables
-    int i__, j, l, info;
-    int nota, notb;
-    double temp;
-    int ncola;
-    extern int lsame_(char *, char *);
-    int nrowa, nrowb;
-    extern /* Subroutine */ int xerbla_(char *, int *);
-
-    //
-    // -- Reference BLAS level3 routine (version 3.7.0) --
-    // -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    December 2016
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. External Functions ..
-    //    ..
-    //    .. External Subroutines ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    //    .. Local Scalars ..
-    //    ..
-    //    .. Parameters ..
-    //    ..
-    //
-    //    Set  NOTA  and  NOTB  as  true if  A  and  B  respectively are not
-    //    transposed and set  NROWA, NCOLA and  NROWB  as the number of rows
-    //    and  columns of  A  and the  number of  rows  of  B  respectively.
-    //
-    // Parameter adjustments
-    a_dim1 = *lda;
-    a_offset = 1 + a_dim1;
-    a -= a_offset;
-    b_dim1 = *ldb;
-    b_offset = 1 + b_dim1;
-    b -= b_offset;
-    c_dim1 = *ldc;
-    c_offset = 1 + c_dim1;
-    c__ -= c_offset;
-
-    // Function Body
-    nota = lsame_(transa, "N");
-    notb = lsame_(transb, "N");
-    if (nota) {
-	nrowa = *m;
-	ncola = *k;
-    } else {
-	nrowa = *k;
-	ncola = *m;
-    }
-    if (notb) {
-	nrowb = *k;
-    } else {
-	nrowb = *n;
-    }
-    //
-    //    Test the input parameters.
-    //
-    info = 0;
-    if (! nota && ! lsame_(transa, "C") && ! lsame_(transa, "T")) {
-	info = 1;
-    } else if (! notb && ! lsame_(transb, "C") && ! lsame_(transb, "T")) {
-	info = 2;
-    } else if (*m < 0) {
-	info = 3;
-    } else if (*n < 0) {
-	info = 4;
-    } else if (*k < 0) {
-	info = 5;
-    } else if (*lda < max(1,nrowa)) {
-	info = 8;
-    } else if (*ldb < max(1,nrowb)) {
-	info = 10;
-    } else if (*ldc < max(1,*m)) {
-	info = 13;
-    }
-    if (info != 0) {
-	xerbla_("DGEMM ", &info);
-	return 0;
-    }
-    //
-    //    Quick return if possible.
-    //
-    if (*m == 0 || *n == 0 || (*alpha == 0. || *k == 0) && *beta == 1.) {
-	return 0;
-    }
-    //
-    //    And if  alpha.eq.zero.
-    //
-    if (*alpha == 0.) {
-	if (*beta == 0.) {
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    c__[i__ + j * c_dim1] = 0.;
-// L10:
-		}
-// L20:
-	    }
-	} else {
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    c__[i__ + j * c_dim1] = *beta * c__[i__ + j * c_dim1];
-// L30:
-		}
-// L40:
-	    }
-	}
-	return 0;
-    }
-    //
-    //    Start the operations.
-    //
-    if (notb) {
-	if (nota) {
-	    //
-	    //          Form  C := alpha*A*B + beta*C.
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		if (*beta == 0.) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			c__[i__ + j * c_dim1] = 0.;
-// L50:
-		    }
-		} else if (*beta != 1.) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			c__[i__ + j * c_dim1] = *beta * c__[i__ + j * c_dim1];
-// L60:
-		    }
-		}
-		i__2 = *k;
-		for (l = 1; l <= i__2; ++l) {
-		    temp = *alpha * b[l + j * b_dim1];
-		    i__3 = *m;
-		    for (i__ = 1; i__ <= i__3; ++i__) {
-			c__[i__ + j * c_dim1] += temp * a[i__ + l * a_dim1];
-// L70:
-		    }
-// L80:
-		}
-// L90:
-	    }
-	} else {
-	    //
-	    //          Form  C := alpha*A**T*B + beta*C
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    temp = 0.;
-		    i__3 = *k;
-		    for (l = 1; l <= i__3; ++l) {
-			temp += a[l + i__ * a_dim1] * b[l + j * b_dim1];
-// L100:
-		    }
-		    if (*beta == 0.) {
-			c__[i__ + j * c_dim1] = *alpha * temp;
-		    } else {
-			c__[i__ + j * c_dim1] = *alpha * temp + *beta * c__[
-				i__ + j * c_dim1];
-		    }
-// L110:
-		}
-// L120:
-	    }
-	}
-    } else {
-	if (nota) {
-	    //
-	    //          Form  C := alpha*A*B**T + beta*C
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		if (*beta == 0.) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			c__[i__ + j * c_dim1] = 0.;
-// L130:
-		    }
-		} else if (*beta != 1.) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			c__[i__ + j * c_dim1] = *beta * c__[i__ + j * c_dim1];
-// L140:
-		    }
-		}
-		i__2 = *k;
-		for (l = 1; l <= i__2; ++l) {
-		    temp = *alpha * b[j + l * b_dim1];
-		    i__3 = *m;
-		    for (i__ = 1; i__ <= i__3; ++i__) {
-			c__[i__ + j * c_dim1] += temp * a[i__ + l * a_dim1];
-// L150:
-		    }
-// L160:
-		}
-// L170:
-	    }
-	} else {
-	    //
-	    //          Form  C := alpha*A**T*B**T + beta*C
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    temp = 0.;
-		    i__3 = *k;
-		    for (l = 1; l <= i__3; ++l) {
-			temp += a[l + i__ * a_dim1] * b[j + l * b_dim1];
-// L180:
-		    }
-		    if (*beta == 0.) {
-			c__[i__ + j * c_dim1] = *alpha * temp;
-		    } else {
-			c__[i__ + j * c_dim1] = *alpha * temp + *beta * c__[
-				i__ + j * c_dim1];
-		    }
-// L190:
-		}
-// L200:
-	    }
-	}
-    }
-    return 0;
-    //
-    //    End of DGEMM .
-    //
-} // dgemm_
-
@@ -1,370 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b DGEMV
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE DGEMV(TRANS,M,N,ALPHA,A,LDA,X,INCX,BETA,Y,INCY)
-//
-//      .. Scalar Arguments ..
-//      DOUBLE PRECISION ALPHA,BETA
-//      INTEGER INCX,INCY,LDA,M,N
-//      CHARACTER TRANS
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION A(LDA,*),X(*),Y(*)
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> DGEMV  performs one of the matrix-vector operations
-//>
-//>    y := alpha*A*x + beta*y,   or   y := alpha*A**T*x + beta*y,
-//>
-//> where alpha and beta are scalars, x and y are vectors and A is an
-//> m by n matrix.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] TRANS
-//> \verbatim
-//>          TRANS is CHARACTER*1
-//>           On entry, TRANS specifies the operation to be performed as
-//>           follows:
-//>
-//>              TRANS = 'N' or 'n'   y := alpha*A*x + beta*y.
-//>
-//>              TRANS = 'T' or 't'   y := alpha*A**T*x + beta*y.
-//>
-//>              TRANS = 'C' or 'c'   y := alpha*A**T*x + beta*y.
-//> \endverbatim
-//>
-//> \param[in] M
-//> \verbatim
-//>          M is INTEGER
-//>           On entry, M specifies the number of rows of the matrix A.
-//>           M must be at least zero.
-//> \endverbatim
-//>
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>           On entry, N specifies the number of columns of the matrix A.
-//>           N must be at least zero.
-//> \endverbatim
-//>
-//> \param[in] ALPHA
-//> \verbatim
-//>          ALPHA is DOUBLE PRECISION.
-//>           On entry, ALPHA specifies the scalar alpha.
-//> \endverbatim
-//>
-//> \param[in] A
-//> \verbatim
-//>          A is DOUBLE PRECISION array, dimension ( LDA, N )
-//>           Before entry, the leading m by n part of the array A must
-//>           contain the matrix of coefficients.
-//> \endverbatim
-//>
-//> \param[in] LDA
-//> \verbatim
-//>          LDA is INTEGER
-//>           On entry, LDA specifies the first dimension of A as declared
-//>           in the calling (sub) program. LDA must be at least
-//>           max( 1, m ).
-//> \endverbatim
-//>
-//> \param[in] X
-//> \verbatim
-//>          X is DOUBLE PRECISION array, dimension at least
-//>           ( 1 + ( n - 1 )*abs( INCX ) ) when TRANS = 'N' or 'n'
-//>           and at least
-//>           ( 1 + ( m - 1 )*abs( INCX ) ) otherwise.
-//>           Before entry, the incremented array X must contain the
-//>           vector x.
-//> \endverbatim
-//>
-//> \param[in] INCX
-//> \verbatim
-//>          INCX is INTEGER
-//>           On entry, INCX specifies the increment for the elements of
-//>           X. INCX must not be zero.
-//> \endverbatim
-//>
-//> \param[in] BETA
-//> \verbatim
-//>          BETA is DOUBLE PRECISION.
-//>           On entry, BETA specifies the scalar beta. When BETA is
-//>           supplied as zero then Y need not be set on input.
-//> \endverbatim
-//>
-//> \param[in,out] Y
-//> \verbatim
-//>          Y is DOUBLE PRECISION array, dimension at least
-//>           ( 1 + ( m - 1 )*abs( INCY ) ) when TRANS = 'N' or 'n'
-//>           and at least
-//>           ( 1 + ( n - 1 )*abs( INCY ) ) otherwise.
-//>           Before entry with BETA non-zero, the incremented array Y
-//>           must contain the vector y. On exit, Y is overwritten by the
-//>           updated vector y.
-//> \endverbatim
-//>
-//> \param[in] INCY
-//> \verbatim
-//>          INCY is INTEGER
-//>           On entry, INCY specifies the increment for the elements of
-//>           Y. INCY must not be zero.
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date December 2016
-//
-//> \ingroup double_blas_level2
-//
-//> \par Further Details:
-// =====================
-//>
-//> \verbatim
-//>
-//>  Level 2 Blas routine.
-//>  The vector and matrix arguments are not referenced when N = 0, or M = 0
-//>
-//>  -- Written on 22-October-1986.
-//>     Jack Dongarra, Argonne National Lab.
-//>     Jeremy Du Croz, Nag Central Office.
-//>     Sven Hammarling, Nag Central Office.
-//>     Richard Hanson, Sandia National Labs.
-//> \endverbatim
-//>
-// =====================================================================
-/* Subroutine */ int dgemv_(char *trans, int *m, int *n, double *alpha,
-	double *a, int *lda, double *x, int *incx, double *beta, double *y,
-	int *incy)
-{
-    // System generated locals
-    int a_dim1, a_offset, i__1, i__2;
-
-    // Local variables
-    int i__, j, ix, iy, jx, jy, kx, ky, info;
-    double temp;
-    int lenx, leny;
-    extern int lsame_(char *, char *);
-    extern /* Subroutine */ int xerbla_(char *, int *);
-
-    //
-    // -- Reference BLAS level2 routine (version 3.7.0) --
-    // -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    December 2016
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. Parameters ..
-    //    ..
-    //    .. Local Scalars ..
-    //    ..
-    //    .. External Functions ..
-    //    ..
-    //    .. External Subroutines ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    //
-    //    Test the input parameters.
-    //
-    // Parameter adjustments
-    a_dim1 = *lda;
-    a_offset = 1 + a_dim1;
-    a -= a_offset;
-    --x;
-    --y;
-
-    // Function Body
-    info = 0;
-    if (! lsame_(trans, "N") && ! lsame_(trans, "T") && ! lsame_(trans, "C"))
-	    {
-	info = 1;
-    } else if (*m < 0) {
-	info = 2;
-    } else if (*n < 0) {
-	info = 3;
-    } else if (*lda < max(1,*m)) {
-	info = 6;
-    } else if (*incx == 0) {
-	info = 8;
-    } else if (*incy == 0) {
-	info = 11;
-    }
-    if (info != 0) {
-	xerbla_("DGEMV ", &info);
-	return 0;
-    }
-    //
-    //    Quick return if possible.
-    //
-    if (*m == 0 || *n == 0 || *alpha == 0. && *beta == 1.) {
-	return 0;
-    }
-    //
-    //    Set  LENX  and  LENY, the lengths of the vectors x and y, and set
-    //    up the start points in  X  and  Y.
-    //
-    if (lsame_(trans, "N")) {
-	lenx = *n;
-	leny = *m;
-    } else {
-	lenx = *m;
-	leny = *n;
-    }
-    if (*incx > 0) {
-	kx = 1;
-    } else {
-	kx = 1 - (lenx - 1) * *incx;
-    }
-    if (*incy > 0) {
-	ky = 1;
-    } else {
-	ky = 1 - (leny - 1) * *incy;
-    }
-    //
-    //    Start the operations. In this version the elements of A are
-    //    accessed sequentially with one pass through A.
-    //
-    //    First form  y := beta*y.
-    //
-    if (*beta != 1.) {
-	if (*incy == 1) {
-	    if (*beta == 0.) {
-		i__1 = leny;
-		for (i__ = 1; i__ <= i__1; ++i__) {
-		    y[i__] = 0.;
-// L10:
-		}
-	    } else {
-		i__1 = leny;
-		for (i__ = 1; i__ <= i__1; ++i__) {
-		    y[i__] = *beta * y[i__];
-// L20:
-		}
-	    }
-	} else {
-	    iy = ky;
-	    if (*beta == 0.) {
-		i__1 = leny;
-		for (i__ = 1; i__ <= i__1; ++i__) {
-		    y[iy] = 0.;
-		    iy += *incy;
-// L30:
-		}
-	    } else {
-		i__1 = leny;
-		for (i__ = 1; i__ <= i__1; ++i__) {
-		    y[iy] = *beta * y[iy];
-		    iy += *incy;
-// L40:
-		}
-	    }
-	}
-    }
-    if (*alpha == 0.) {
-	return 0;
-    }
-    if (lsame_(trans, "N")) {
-	//
-	//       Form  y := alpha*A*x + y.
-	//
-	jx = kx;
-	if (*incy == 1) {
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		temp = *alpha * x[jx];
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    y[i__] += temp * a[i__ + j * a_dim1];
-// L50:
-		}
-		jx += *incx;
-// L60:
-	    }
-	} else {
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		temp = *alpha * x[jx];
-		iy = ky;
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    y[iy] += temp * a[i__ + j * a_dim1];
-		    iy += *incy;
-// L70:
-		}
-		jx += *incx;
-// L80:
-	    }
-	}
-    } else {
-	//
-	//       Form  y := alpha*A**T*x + y.
-	//
-	jy = ky;
-	if (*incx == 1) {
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		temp = 0.;
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    temp += a[i__ + j * a_dim1] * x[i__];
-// L90:
-		}
-		y[jy] += *alpha * temp;
-		jy += *incy;
-// L100:
-	    }
-	} else {
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		temp = 0.;
-		ix = kx;
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    temp += a[i__ + j * a_dim1] * x[ix];
-		    ix += *incx;
-// L110:
-		}
-		y[jy] += *alpha * temp;
-		jy += *incy;
-// L120:
-	    }
-	}
-    }
-    return 0;
-    //
-    //    End of DGEMV .
-    //
-} // dgemv_
-
@@ -1,186 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b DISNAN tests input for NaN.
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-//> \htmlonly
-//> Download DISNAN + dependencies
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/disnan.f">
-//> [TGZ]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/disnan.f">
-//> [ZIP]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/disnan.f">
-//> [TXT]</a>
-//> \endhtmlonly
-//
-// Definition:
-// ===========
-//
-//      LOGICAL FUNCTION DISNAN( DIN )
-//
-//      .. Scalar Arguments ..
-//      DOUBLE PRECISION, INTENT(IN) :: DIN
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> DISNAN returns .TRUE. if its argument is NaN, and .FALSE.
-//> otherwise.  To be replaced by the Fortran 2003 intrinsic in the
-//> future.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] DIN
-//> \verbatim
-//>          DIN is DOUBLE PRECISION
-//>          Input to test for NaN.
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date June 2017
-//
-//> \ingroup OTHERauxiliary
-//
-// =====================================================================
-int disnan_(double *din)
-{
-    // System generated locals
-    int ret_val;
-
-    // Local variables
-    extern int dlaisnan_(double *, double *);
-
-    //
-    // -- LAPACK auxiliary routine (version 3.7.1) --
-    // -- LAPACK is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    June 2017
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    // .. External Functions ..
-    // ..
-    // .. Executable Statements ..
-    ret_val = dlaisnan_(din, din);
-    return ret_val;
-} // disnan_
-
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-//> \brief \b DLAISNAN tests input for NaN by comparing two arguments for inequality.
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-//> \htmlonly
-//> Download DLAISNAN + dependencies
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dlaisnan.f">
-//> [TGZ]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dlaisnan.f">
-//> [ZIP]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dlaisnan.f">
-//> [TXT]</a>
-//> \endhtmlonly
-//
-// Definition:
-// ===========
-//
-//      LOGICAL FUNCTION DLAISNAN( DIN1, DIN2 )
-//
-//      .. Scalar Arguments ..
-//      DOUBLE PRECISION, INTENT(IN) :: DIN1, DIN2
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> This routine is not for general use.  It exists solely to avoid
-//> over-optimization in DISNAN.
-//>
-//> DLAISNAN checks for NaNs by comparing its two arguments for
-//> inequality.  NaN is the only floating-point value where NaN != NaN
-//> returns .TRUE.  To check for NaNs, pass the same variable as both
-//> arguments.
-//>
-//> A compiler must assume that the two arguments are
-//> not the same variable, and the test will not be optimized away.
-//> Interprocedural or whole-program optimization may delete this
-//> test.  The ISNAN functions will be replaced by the correct
-//> Fortran 03 intrinsic once the intrinsic is widely available.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] DIN1
-//> \verbatim
-//>          DIN1 is DOUBLE PRECISION
-//> \endverbatim
-//>
-//> \param[in] DIN2
-//> \verbatim
-//>          DIN2 is DOUBLE PRECISION
-//>          Two numbers to compare for inequality.
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date June 2017
-//
-//> \ingroup OTHERauxiliary
-//
-// =====================================================================
-int dlaisnan_(double *din1, double *din2)
-{
-    // System generated locals
-    int ret_val;
-
-    //
-    // -- LAPACK auxiliary routine (version 3.7.1) --
-    // -- LAPACK is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    June 2017
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    // .. Executable Statements ..
-    ret_val = *din1 != *din2;
-    return ret_val;
-} // dlaisnan_
-
@@ -1,184 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b DLACPY copies all or part of one two-dimensional array to another.
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-//> \htmlonly
-//> Download DLACPY + dependencies
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dlacpy.f">
-//> [TGZ]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dlacpy.f">
-//> [ZIP]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dlacpy.f">
-//> [TXT]</a>
-//> \endhtmlonly
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE DLACPY( UPLO, M, N, A, LDA, B, LDB )
-//
-//      .. Scalar Arguments ..
-//      CHARACTER          UPLO
-//      INTEGER            LDA, LDB, M, N
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION   A( LDA, * ), B( LDB, * )
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> DLACPY copies all or part of a two-dimensional matrix A to another
-//> matrix B.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] UPLO
-//> \verbatim
-//>          UPLO is CHARACTER*1
-//>          Specifies the part of the matrix A to be copied to B.
-//>          = 'U':      Upper triangular part
-//>          = 'L':      Lower triangular part
-//>          Otherwise:  All of the matrix A
-//> \endverbatim
-//>
-//> \param[in] M
-//> \verbatim
-//>          M is INTEGER
-//>          The number of rows of the matrix A.  M >= 0.
-//> \endverbatim
-//>
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>          The number of columns of the matrix A.  N >= 0.
-//> \endverbatim
-//>
-//> \param[in] A
-//> \verbatim
-//>          A is DOUBLE PRECISION array, dimension (LDA,N)
-//>          The m by n matrix A.  If UPLO = 'U', only the upper triangle
-//>          or trapezoid is accessed; if UPLO = 'L', only the lower
-//>          triangle or trapezoid is accessed.
-//> \endverbatim
-//>
-//> \param[in] LDA
-//> \verbatim
-//>          LDA is INTEGER
-//>          The leading dimension of the array A.  LDA >= max(1,M).
-//> \endverbatim
-//>
-//> \param[out] B
-//> \verbatim
-//>          B is DOUBLE PRECISION array, dimension (LDB,N)
-//>          On exit, B = A in the locations specified by UPLO.
-//> \endverbatim
-//>
-//> \param[in] LDB
-//> \verbatim
-//>          LDB is INTEGER
-//>          The leading dimension of the array B.  LDB >= max(1,M).
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date December 2016
-//
-//> \ingroup OTHERauxiliary
-//
-// =====================================================================
-/* Subroutine */ int dlacpy_(char *uplo, int *m, int *n, double *a, int *lda,
-	double *b, int *ldb)
-{
-    // System generated locals
-    int a_dim1, a_offset, b_dim1, b_offset, i__1, i__2;
-
-    // Local variables
-    int i__, j;
-    extern int lsame_(char *, char *);
-
-    //
-    // -- LAPACK auxiliary routine (version 3.7.0) --
-    // -- LAPACK is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    December 2016
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. Local Scalars ..
-    //    ..
-    //    .. External Functions ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    //    .. Executable Statements ..
-    //
-    // Parameter adjustments
-    a_dim1 = *lda;
-    a_offset = 1 + a_dim1;
-    a -= a_offset;
-    b_dim1 = *ldb;
-    b_offset = 1 + b_dim1;
-    b -= b_offset;
-
-    // Function Body
-    if (lsame_(uplo, "U")) {
-	i__1 = *n;
-	for (j = 1; j <= i__1; ++j) {
-	    i__2 = min(j,*m);
-	    for (i__ = 1; i__ <= i__2; ++i__) {
-		b[i__ + j * b_dim1] = a[i__ + j * a_dim1];
-// L10:
-	    }
-// L20:
-	}
-    } else if (lsame_(uplo, "L")) {
-	i__1 = *n;
-	for (j = 1; j <= i__1; ++j) {
-	    i__2 = *m;
-	    for (i__ = j; i__ <= i__2; ++i__) {
-		b[i__ + j * b_dim1] = a[i__ + j * a_dim1];
-// L30:
-	    }
-// L40:
-	}
-    } else {
-	i__1 = *n;
-	for (j = 1; j <= i__1; ++j) {
-	    i__2 = *m;
-	    for (i__ = 1; i__ <= i__2; ++i__) {
-		b[i__ + j * b_dim1] = a[i__ + j * a_dim1];
-// L50:
-	    }
-// L60:
-	}
-    }
-    return 0;
-    //
-    //    End of DLACPY
-    //
-} // dlacpy_
-
@@ -1,367 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b DCOMBSSQ adds two scaled sum of squares quantities.
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE DCOMBSSQ( V1, V2 )
-//
-//      .. Array Arguments ..
-//      DOUBLE PRECISION   V1( 2 ), V2( 2 )
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> DCOMBSSQ adds two scaled sum of squares quantities, V1 := V1 + V2.
-//> That is,
-//>
-//>    V1_scale**2 * V1_sumsq := V1_scale**2 * V1_sumsq
-//>                            + V2_scale**2 * V2_sumsq
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in,out] V1
-//> \verbatim
-//>          V1 is DOUBLE PRECISION array, dimension (2).
-//>          The first scaled sum.
-//>          V1(1) = V1_scale, V1(2) = V1_sumsq.
-//> \endverbatim
-//>
-//> \param[in] V2
-//> \verbatim
-//>          V2 is DOUBLE PRECISION array, dimension (2).
-//>          The second scaled sum.
-//>          V2(1) = V2_scale, V2(2) = V2_sumsq.
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date November 2018
-//
-//> \ingroup OTHERauxiliary
-//
-// =====================================================================
-/* Subroutine */ int dcombssq_(double *v1, double *v2)
-{
-    // System generated locals
-    double d__1;
-
-    //
-    // -- LAPACK auxiliary routine (version 3.7.0) --
-    // -- LAPACK is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    November 2018
-    //
-    //    .. Array Arguments ..
-    //    ..
-    //
-    //=====================================================================
-    //
-    //    .. Parameters ..
-    //    ..
-    //    .. Executable Statements ..
-    //
-    // Parameter adjustments
-    --v2;
-    --v1;
-
-    // Function Body
-    if (v1[1] >= v2[1]) {
-	if (v1[1] != 0.) {
-	    // Computing 2nd power
-	    d__1 = v2[1] / v1[1];
-	    v1[2] += d__1 * d__1 * v2[2];
-	}
-    } else {
-	// Computing 2nd power
-	d__1 = v1[1] / v2[1];
-	v1[2] = v2[2] + d__1 * d__1 * v1[2];
-	v1[1] = v2[1];
-    }
-    return 0;
-    //
-    //    End of DCOMBSSQ
-    //
-} // dcombssq_
-
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-//> \brief \b DLANGE returns the value of the 1-norm, Frobenius norm, infinity-norm, or the largest absolute value of any element of a general rectangular matrix.
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-//> \htmlonly
-//> Download DLANGE + dependencies
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dlange.f">
-//> [TGZ]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dlange.f">
-//> [ZIP]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dlange.f">
-//> [TXT]</a>
-//> \endhtmlonly
-//
-// Definition:
-// ===========
-//
-//      DOUBLE PRECISION FUNCTION DLANGE( NORM, M, N, A, LDA, WORK )
-//
-//      .. Scalar Arguments ..
-//      CHARACTER          NORM
-//      INTEGER            LDA, M, N
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION   A( LDA, * ), WORK( * )
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> DLANGE  returns the value of the one norm,  or the Frobenius norm, or
-//> the  infinity norm,  or the  element of  largest absolute value  of a
-//> real matrix A.
-//> \endverbatim
-//>
-//> \return DLANGE
-//> \verbatim
-//>
-//>    DLANGE = ( max(abs(A(i,j))), NORM = 'M' or 'm'
-//>             (
-//>             ( norm1(A),         NORM = '1', 'O' or 'o'
-//>             (
-//>             ( normI(A),         NORM = 'I' or 'i'
-//>             (
-//>             ( normF(A),         NORM = 'F', 'f', 'E' or 'e'
-//>
-//> where  norm1  denotes the  one norm of a matrix (maximum column sum),
-//> normI  denotes the  infinity norm  of a matrix  (maximum row sum) and
-//> normF  denotes the  Frobenius norm of a matrix (square root of sum of
-//> squares).  Note that  max(abs(A(i,j)))  is not a consistent matrix norm.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] NORM
-//> \verbatim
-//>          NORM is CHARACTER*1
-//>          Specifies the value to be returned in DLANGE as described
-//>          above.
-//> \endverbatim
-//>
-//> \param[in] M
-//> \verbatim
-//>          M is INTEGER
-//>          The number of rows of the matrix A.  M >= 0.  When M = 0,
-//>          DLANGE is set to zero.
-//> \endverbatim
-//>
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>          The number of columns of the matrix A.  N >= 0.  When N = 0,
-//>          DLANGE is set to zero.
-//> \endverbatim
-//>
-//> \param[in] A
-//> \verbatim
-//>          A is DOUBLE PRECISION array, dimension (LDA,N)
-//>          The m by n matrix A.
-//> \endverbatim
-//>
-//> \param[in] LDA
-//> \verbatim
-//>          LDA is INTEGER
-//>          The leading dimension of the array A.  LDA >= max(M,1).
-//> \endverbatim
-//>
-//> \param[out] WORK
-//> \verbatim
-//>          WORK is DOUBLE PRECISION array, dimension (MAX(1,LWORK)),
-//>          where LWORK >= M when NORM = 'I'; otherwise, WORK is not
-//>          referenced.
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date December 2016
-//
-//> \ingroup doubleGEauxiliary
-//
-// =====================================================================
-double dlange_(char *norm, int *m, int *n, double *a, int *lda, double *work)
-{
-    // Table of constant values
-    int c__1 = 1;
-
-    // System generated locals
-    int a_dim1, a_offset, i__1, i__2;
-    double ret_val, d__1;
-
-    // Local variables
-    extern /* Subroutine */ int dcombssq_(double *, double *);
-    int i__, j;
-    double sum, ssq[2], temp;
-    extern int lsame_(char *, char *);
-    double value;
-    extern int disnan_(double *);
-    extern /* Subroutine */ int dlassq_(int *, double *, int *, double *,
-	    double *);
-    double colssq[2];
-
-    //
-    // -- LAPACK auxiliary routine (version 3.7.0) --
-    // -- LAPACK is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    December 2016
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    //=====================================================================
-    //
-    //    .. Parameters ..
-    //    ..
-    //    .. Local Scalars ..
-    //    ..
-    //    .. Local Arrays ..
-    //    ..
-    //    .. External Subroutines ..
-    //    ..
-    //    .. External Functions ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    //    .. Executable Statements ..
-    //
-    // Parameter adjustments
-    a_dim1 = *lda;
-    a_offset = 1 + a_dim1;
-    a -= a_offset;
-    --work;
-
-    // Function Body
-    if (min(*m,*n) == 0) {
-	value = 0.;
-    } else if (lsame_(norm, "M")) {
-	//
-	//       Find max(abs(A(i,j))).
-	//
-	value = 0.;
-	i__1 = *n;
-	for (j = 1; j <= i__1; ++j) {
-	    i__2 = *m;
-	    for (i__ = 1; i__ <= i__2; ++i__) {
-		temp = (d__1 = a[i__ + j * a_dim1], abs(d__1));
-		if (value < temp || disnan_(&temp)) {
-		    value = temp;
-		}
-// L10:
-	    }
-// L20:
-	}
-    } else if (lsame_(norm, "O") || *(unsigned char *)norm == '1') {
-	//
-	//       Find norm1(A).
-	//
-	value = 0.;
-	i__1 = *n;
-	for (j = 1; j <= i__1; ++j) {
-	    sum = 0.;
-	    i__2 = *m;
-	    for (i__ = 1; i__ <= i__2; ++i__) {
-		sum += (d__1 = a[i__ + j * a_dim1], abs(d__1));
-// L30:
-	    }
-	    if (value < sum || disnan_(&sum)) {
-		value = sum;
-	    }
-// L40:
-	}
-    } else if (lsame_(norm, "I")) {
-	//
-	//       Find normI(A).
-	//
-	i__1 = *m;
-	for (i__ = 1; i__ <= i__1; ++i__) {
-	    work[i__] = 0.;
-// L50:
-	}
-	i__1 = *n;
-	for (j = 1; j <= i__1; ++j) {
-	    i__2 = *m;
-	    for (i__ = 1; i__ <= i__2; ++i__) {
-		work[i__] += (d__1 = a[i__ + j * a_dim1], abs(d__1));
-// L60:
-	    }
-// L70:
-	}
-	value = 0.;
-	i__1 = *m;
-	for (i__ = 1; i__ <= i__1; ++i__) {
-	    temp = work[i__];
-	    if (value < temp || disnan_(&temp)) {
-		value = temp;
-	    }
-// L80:
-	}
-    } else if (lsame_(norm, "F") || lsame_(norm, "E")) {
-	//
-	//       Find normF(A).
-	//       SSQ(1) is scale
-	//       SSQ(2) is sum-of-squares
-	//       For better accuracy, sum each column separately.
-	//
-	ssq[0] = 0.;
-	ssq[1] = 1.;
-	i__1 = *n;
-	for (j = 1; j <= i__1; ++j) {
-	    colssq[0] = 0.;
-	    colssq[1] = 1.;
-	    dlassq_(m, &a[j * a_dim1 + 1], &c__1, colssq, &colssq[1]);
-	    dcombssq_(ssq, colssq);
-// L90:
-	}
-	value = ssq[0] * sqrt(ssq[1]);
-    }
-    ret_val = value;
-    return ret_val;
-    //
-    //    End of DLANGE
-    //
-} // dlange_
-
@@ -1,125 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b DLAPY2 returns sqrt(x2+y2).
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-//> \htmlonly
-//> Download DLAPY2 + dependencies
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dlapy2.f">
-//> [TGZ]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dlapy2.f">
-//> [ZIP]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dlapy2.f">
-//> [TXT]</a>
-//> \endhtmlonly
-//
-// Definition:
-// ===========
-//
-//      DOUBLE PRECISION FUNCTION DLAPY2( X, Y )
-//
-//      .. Scalar Arguments ..
-//      DOUBLE PRECISION   X, Y
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> DLAPY2 returns sqrt(x**2+y**2), taking care not to cause unnecessary
-//> overflow.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] X
-//> \verbatim
-//>          X is DOUBLE PRECISION
-//> \endverbatim
-//>
-//> \param[in] Y
-//> \verbatim
-//>          Y is DOUBLE PRECISION
-//>          X and Y specify the values x and y.
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date June 2017
-//
-//> \ingroup OTHERauxiliary
-//
-// =====================================================================
-double dlapy2_(double *x, double *y)
-{
-    // System generated locals
-    double ret_val, d__1;
-
-    // Local variables
-    int x_is_nan__, y_is_nan__;
-    double w, z__, xabs, yabs;
-    extern int disnan_(double *);
-
-    //
-    // -- LAPACK auxiliary routine (version 3.7.1) --
-    // -- LAPACK is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    June 2017
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. Parameters ..
-    //    ..
-    //    .. Local Scalars ..
-    //    ..
-    //    .. External Functions ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    //    .. Executable Statements ..
-    //
-    x_is_nan__ = disnan_(x);
-    y_is_nan__ = disnan_(y);
-    if (x_is_nan__) {
-	ret_val = *x;
-    }
-    if (y_is_nan__) {
-	ret_val = *y;
-    }
-    if (! (x_is_nan__ || y_is_nan__)) {
-	xabs = abs(*x);
-	yabs = abs(*y);
-	w = max(xabs,yabs);
-	z__ = min(xabs,yabs);
-	if (z__ == 0.) {
-	    ret_val = w;
-	} else {
-	    // Computing 2nd power
-	    d__1 = z__ / w;
-	    ret_val = w * sqrt(d__1 * d__1 + 1.);
-	}
-    }
-    return ret_val;
-    //
-    //    End of DLAPY2
-    //
-} // dlapy2_
-
@@ -1,768 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b DGER
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE DGER(M,N,ALPHA,X,INCX,Y,INCY,A,LDA)
-//
-//      .. Scalar Arguments ..
-//      DOUBLE PRECISION ALPHA
-//      INTEGER INCX,INCY,LDA,M,N
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION A(LDA,*),X(*),Y(*)
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> DGER   performs the rank 1 operation
-//>
-//>    A := alpha*x*y**T + A,
-//>
-//> where alpha is a scalar, x is an m element vector, y is an n element
-//> vector and A is an m by n matrix.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] M
-//> \verbatim
-//>          M is INTEGER
-//>           On entry, M specifies the number of rows of the matrix A.
-//>           M must be at least zero.
-//> \endverbatim
-//>
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>           On entry, N specifies the number of columns of the matrix A.
-//>           N must be at least zero.
-//> \endverbatim
-//>
-//> \param[in] ALPHA
-//> \verbatim
-//>          ALPHA is DOUBLE PRECISION.
-//>           On entry, ALPHA specifies the scalar alpha.
-//> \endverbatim
-//>
-//> \param[in] X
-//> \verbatim
-//>          X is DOUBLE PRECISION array, dimension at least
-//>           ( 1 + ( m - 1 )*abs( INCX ) ).
-//>           Before entry, the incremented array X must contain the m
-//>           element vector x.
-//> \endverbatim
-//>
-//> \param[in] INCX
-//> \verbatim
-//>          INCX is INTEGER
-//>           On entry, INCX specifies the increment for the elements of
-//>           X. INCX must not be zero.
-//> \endverbatim
-//>
-//> \param[in] Y
-//> \verbatim
-//>          Y is DOUBLE PRECISION array, dimension at least
-//>           ( 1 + ( n - 1 )*abs( INCY ) ).
-//>           Before entry, the incremented array Y must contain the n
-//>           element vector y.
-//> \endverbatim
-//>
-//> \param[in] INCY
-//> \verbatim
-//>          INCY is INTEGER
-//>           On entry, INCY specifies the increment for the elements of
-//>           Y. INCY must not be zero.
-//> \endverbatim
-//>
-//> \param[in,out] A
-//> \verbatim
-//>          A is DOUBLE PRECISION array, dimension ( LDA, N )
-//>           Before entry, the leading m by n part of the array A must
-//>           contain the matrix of coefficients. On exit, A is
-//>           overwritten by the updated matrix.
-//> \endverbatim
-//>
-//> \param[in] LDA
-//> \verbatim
-//>          LDA is INTEGER
-//>           On entry, LDA specifies the first dimension of A as declared
-//>           in the calling (sub) program. LDA must be at least
-//>           max( 1, m ).
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date December 2016
-//
-//> \ingroup double_blas_level2
-//
-//> \par Further Details:
-// =====================
-//>
-//> \verbatim
-//>
-//>  Level 2 Blas routine.
-//>
-//>  -- Written on 22-October-1986.
-//>     Jack Dongarra, Argonne National Lab.
-//>     Jeremy Du Croz, Nag Central Office.
-//>     Sven Hammarling, Nag Central Office.
-//>     Richard Hanson, Sandia National Labs.
-//> \endverbatim
-//>
-// =====================================================================
-/* Subroutine */ int dger_(int *m, int *n, double *alpha, double *x, int *
-	incx, double *y, int *incy, double *a, int *lda)
-{
-    // System generated locals
-    int a_dim1, a_offset, i__1, i__2;
-
-    // Local variables
-    int i__, j, ix, jy, kx, info;
-    double temp;
-    extern /* Subroutine */ int xerbla_(char *, int *);
-
-    //
-    // -- Reference BLAS level2 routine (version 3.7.0) --
-    // -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    December 2016
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. Parameters ..
-    //    ..
-    //    .. Local Scalars ..
-    //    ..
-    //    .. External Subroutines ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    //
-    //    Test the input parameters.
-    //
-    // Parameter adjustments
-    --x;
-    --y;
-    a_dim1 = *lda;
-    a_offset = 1 + a_dim1;
-    a -= a_offset;
-
-    // Function Body
-    info = 0;
-    if (*m < 0) {
-	info = 1;
-    } else if (*n < 0) {
-	info = 2;
-    } else if (*incx == 0) {
-	info = 5;
-    } else if (*incy == 0) {
-	info = 7;
-    } else if (*lda < max(1,*m)) {
-	info = 9;
-    }
-    if (info != 0) {
-	xerbla_("DGER  ", &info);
-	return 0;
-    }
-    //
-    //    Quick return if possible.
-    //
-    if (*m == 0 || *n == 0 || *alpha == 0.) {
-	return 0;
-    }
-    //
-    //    Start the operations. In this version the elements of A are
-    //    accessed sequentially with one pass through A.
-    //
-    if (*incy > 0) {
-	jy = 1;
-    } else {
-	jy = 1 - (*n - 1) * *incy;
-    }
-    if (*incx == 1) {
-	i__1 = *n;
-	for (j = 1; j <= i__1; ++j) {
-	    if (y[jy] != 0.) {
-		temp = *alpha * y[jy];
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    a[i__ + j * a_dim1] += x[i__] * temp;
-// L10:
-		}
-	    }
-	    jy += *incy;
-// L20:
-	}
-    } else {
-	if (*incx > 0) {
-	    kx = 1;
-	} else {
-	    kx = 1 - (*m - 1) * *incx;
-	}
-	i__1 = *n;
-	for (j = 1; j <= i__1; ++j) {
-	    if (y[jy] != 0.) {
-		temp = *alpha * y[jy];
-		ix = kx;
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    a[i__ + j * a_dim1] += x[ix] * temp;
-		    ix += *incx;
-// L30:
-		}
-	    }
-	    jy += *incy;
-// L40:
-	}
-    }
-    return 0;
-    //
-    //    End of DGER  .
-    //
-} // dger_
-
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-//> \brief \b DLARF applies an elementary reflector to a general rectangular matrix.
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-//> \htmlonly
-//> Download DLARF + dependencies
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dlarf.f">
-//> [TGZ]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dlarf.f">
-//> [ZIP]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dlarf.f">
-//> [TXT]</a>
-//> \endhtmlonly
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE DLARF( SIDE, M, N, V, INCV, TAU, C, LDC, WORK )
-//
-//      .. Scalar Arguments ..
-//      CHARACTER          SIDE
-//      INTEGER            INCV, LDC, M, N
-//      DOUBLE PRECISION   TAU
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION   C( LDC, * ), V( * ), WORK( * )
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> DLARF applies a real elementary reflector H to a real m by n matrix
-//> C, from either the left or the right. H is represented in the form
-//>
-//>       H = I - tau * v * v**T
-//>
-//> where tau is a real scalar and v is a real vector.
-//>
-//> If tau = 0, then H is taken to be the unit matrix.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] SIDE
-//> \verbatim
-//>          SIDE is CHARACTER*1
-//>          = 'L': form  H * C
-//>          = 'R': form  C * H
-//> \endverbatim
-//>
-//> \param[in] M
-//> \verbatim
-//>          M is INTEGER
-//>          The number of rows of the matrix C.
-//> \endverbatim
-//>
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>          The number of columns of the matrix C.
-//> \endverbatim
-//>
-//> \param[in] V
-//> \verbatim
-//>          V is DOUBLE PRECISION array, dimension
-//>                     (1 + (M-1)*abs(INCV)) if SIDE = 'L'
-//>                  or (1 + (N-1)*abs(INCV)) if SIDE = 'R'
-//>          The vector v in the representation of H. V is not used if
-//>          TAU = 0.
-//> \endverbatim
-//>
-//> \param[in] INCV
-//> \verbatim
-//>          INCV is INTEGER
-//>          The increment between elements of v. INCV <> 0.
-//> \endverbatim
-//>
-//> \param[in] TAU
-//> \verbatim
-//>          TAU is DOUBLE PRECISION
-//>          The value tau in the representation of H.
-//> \endverbatim
-//>
-//> \param[in,out] C
-//> \verbatim
-//>          C is DOUBLE PRECISION array, dimension (LDC,N)
-//>          On entry, the m by n matrix C.
-//>          On exit, C is overwritten by the matrix H * C if SIDE = 'L',
-//>          or C * H if SIDE = 'R'.
-//> \endverbatim
-//>
-//> \param[in] LDC
-//> \verbatim
-//>          LDC is INTEGER
-//>          The leading dimension of the array C. LDC >= max(1,M).
-//> \endverbatim
-//>
-//> \param[out] WORK
-//> \verbatim
-//>          WORK is DOUBLE PRECISION array, dimension
-//>                         (N) if SIDE = 'L'
-//>                      or (M) if SIDE = 'R'
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date December 2016
-//
-//> \ingroup doubleOTHERauxiliary
-//
-// =====================================================================
-/* Subroutine */ int dlarf_(char *side, int *m, int *n, double *v, int *incv,
-	double *tau, double *c__, int *ldc, double *work)
-{
-    // Table of constant values
-    double c_b4 = 1.;
-    double c_b5 = 0.;
-    int c__1 = 1;
-
-    // System generated locals
-    int c_dim1, c_offset;
-    double d__1;
-
-    // Local variables
-    int i__;
-    int applyleft;
-    extern /* Subroutine */ int dger_(int *, int *, double *, double *, int *,
-	     double *, int *, double *, int *);
-    extern int lsame_(char *, char *);
-    extern /* Subroutine */ int dgemv_(char *, int *, int *, double *, double
-	    *, int *, double *, int *, double *, double *, int *);
-    int lastc, lastv;
-    extern int iladlc_(int *, int *, double *, int *), iladlr_(int *, int *,
-	    double *, int *);
-
-    //
-    // -- LAPACK auxiliary routine (version 3.7.0) --
-    // -- LAPACK is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    December 2016
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. Parameters ..
-    //    ..
-    //    .. Local Scalars ..
-    //    ..
-    //    .. External Subroutines ..
-    //    ..
-    //    .. External Functions ..
-    //    ..
-    //    .. Executable Statements ..
-    //
-    // Parameter adjustments
-    --v;
-    c_dim1 = *ldc;
-    c_offset = 1 + c_dim1;
-    c__ -= c_offset;
-    --work;
-
-    // Function Body
-    applyleft = lsame_(side, "L");
-    lastv = 0;
-    lastc = 0;
-    if (*tau != 0.) {
-	//    Set up variables for scanning V.  LASTV begins pointing to the end
-	//    of V.
-	if (applyleft) {
-	    lastv = *m;
-	} else {
-	    lastv = *n;
-	}
-	if (*incv > 0) {
-	    i__ = (lastv - 1) * *incv + 1;
-	} else {
-	    i__ = 1;
-	}
-	//    Look for the last non-zero row in V.
-	while(lastv > 0 && v[i__] == 0.) {
-	    --lastv;
-	    i__ -= *incv;
-	}
-	if (applyleft) {
-	    //    Scan for the last non-zero column in C(1:lastv,:).
-	    lastc = iladlc_(&lastv, n, &c__[c_offset], ldc);
-	} else {
-	    //    Scan for the last non-zero row in C(:,1:lastv).
-	    lastc = iladlr_(m, &lastv, &c__[c_offset], ldc);
-	}
-    }
-    //    Note that lastc.eq.0 renders the BLAS operations null; no special
-    //    case is needed at this level.
-    if (applyleft) {
-	//
-	//       Form  H * C
-	//
-	if (lastv > 0) {
-	    //
-	    //          w(1:lastc,1) := C(1:lastv,1:lastc)**T * v(1:lastv,1)
-	    //
-	    dgemv_("Transpose", &lastv, &lastc, &c_b4, &c__[c_offset], ldc, &
-		    v[1], incv, &c_b5, &work[1], &c__1);
-	    //
-	    //          C(1:lastv,1:lastc) := C(...) - v(1:lastv,1) * w(1:lastc,1)**T
-	    //
-	    d__1 = -(*tau);
-	    dger_(&lastv, &lastc, &d__1, &v[1], incv, &work[1], &c__1, &c__[
-		    c_offset], ldc);
-	}
-    } else {
-	//
-	//       Form  C * H
-	//
-	if (lastv > 0) {
-	    //
-	    //          w(1:lastc,1) := C(1:lastc,1:lastv) * v(1:lastv,1)
-	    //
-	    dgemv_("No transpose", &lastc, &lastv, &c_b4, &c__[c_offset], ldc,
-		     &v[1], incv, &c_b5, &work[1], &c__1);
-	    //
-	    //          C(1:lastc,1:lastv) := C(...) - w(1:lastc,1) * v(1:lastv,1)**T
-	    //
-	    d__1 = -(*tau);
-	    dger_(&lastc, &lastv, &d__1, &work[1], &c__1, &v[1], incv, &c__[
-		    c_offset], ldc);
-	}
-    }
-    return 0;
-    //
-    //    End of DLARF
-    //
-} // dlarf_
-
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-//> \brief \b ILADLC scans a matrix for its last non-zero column.
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-//> \htmlonly
-//> Download ILADLC + dependencies
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/iladlc.f">
-//> [TGZ]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/iladlc.f">
-//> [ZIP]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/iladlc.f">
-//> [TXT]</a>
-//> \endhtmlonly
-//
-// Definition:
-// ===========
-//
-//      INTEGER FUNCTION ILADLC( M, N, A, LDA )
-//
-//      .. Scalar Arguments ..
-//      INTEGER            M, N, LDA
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION   A( LDA, * )
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> ILADLC scans A for its last non-zero column.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] M
-//> \verbatim
-//>          M is INTEGER
-//>          The number of rows of the matrix A.
-//> \endverbatim
-//>
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>          The number of columns of the matrix A.
-//> \endverbatim
-//>
-//> \param[in] A
-//> \verbatim
-//>          A is DOUBLE PRECISION array, dimension (LDA,N)
-//>          The m by n matrix A.
-//> \endverbatim
-//>
-//> \param[in] LDA
-//> \verbatim
-//>          LDA is INTEGER
-//>          The leading dimension of the array A. LDA >= max(1,M).
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date December 2016
-//
-//> \ingroup OTHERauxiliary
-//
-// =====================================================================
-int iladlc_(int *m, int *n, double *a, int *lda)
-{
-    // System generated locals
-    int a_dim1, a_offset, ret_val, i__1;
-
-    // Local variables
-    int i__;
-
-    //
-    // -- LAPACK auxiliary routine (version 3.7.0) --
-    // -- LAPACK is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    December 2016
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. Parameters ..
-    //    ..
-    //    .. Local Scalars ..
-    //    ..
-    //    .. Executable Statements ..
-    //
-    //    Quick test for the common case where one corner is non-zero.
-    // Parameter adjustments
-    a_dim1 = *lda;
-    a_offset = 1 + a_dim1;
-    a -= a_offset;
-
-    // Function Body
-    if (*n == 0) {
-	ret_val = *n;
-    } else if (a[*n * a_dim1 + 1] != 0. || a[*m + *n * a_dim1] != 0.) {
-	ret_val = *n;
-    } else {
-	//    Now scan each column from the end, returning with the first non-zero.
-	for (ret_val = *n; ret_val >= 1; --ret_val) {
-	    i__1 = *m;
-	    for (i__ = 1; i__ <= i__1; ++i__) {
-		if (a[i__ + ret_val * a_dim1] != 0.) {
-		    return ret_val;
-		}
-	    }
-	}
-    }
-    return ret_val;
-} // iladlc_
-
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-//> \brief \b ILADLR scans a matrix for its last non-zero row.
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-//> \htmlonly
-//> Download ILADLR + dependencies
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/iladlr.f">
-//> [TGZ]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/iladlr.f">
-//> [ZIP]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/iladlr.f">
-//> [TXT]</a>
-//> \endhtmlonly
-//
-// Definition:
-// ===========
-//
-//      INTEGER FUNCTION ILADLR( M, N, A, LDA )
-//
-//      .. Scalar Arguments ..
-//      INTEGER            M, N, LDA
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION   A( LDA, * )
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> ILADLR scans A for its last non-zero row.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] M
-//> \verbatim
-//>          M is INTEGER
-//>          The number of rows of the matrix A.
-//> \endverbatim
-//>
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>          The number of columns of the matrix A.
-//> \endverbatim
-//>
-//> \param[in] A
-//> \verbatim
-//>          A is DOUBLE PRECISION array, dimension (LDA,N)
-//>          The m by n matrix A.
-//> \endverbatim
-//>
-//> \param[in] LDA
-//> \verbatim
-//>          LDA is INTEGER
-//>          The leading dimension of the array A. LDA >= max(1,M).
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date December 2016
-//
-//> \ingroup OTHERauxiliary
-//
-// =====================================================================
-int iladlr_(int *m, int *n, double *a, int *lda)
-{
-    // System generated locals
-    int a_dim1, a_offset, ret_val, i__1;
-
-    // Local variables
-    int i__, j;
-
-    //
-    // -- LAPACK auxiliary routine (version 3.7.0) --
-    // -- LAPACK is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    December 2016
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. Parameters ..
-    //    ..
-    //    .. Local Scalars ..
-    //    ..
-    //    .. Executable Statements ..
-    //
-    //    Quick test for the common case where one corner is non-zero.
-    // Parameter adjustments
-    a_dim1 = *lda;
-    a_offset = 1 + a_dim1;
-    a -= a_offset;
-
-    // Function Body
-    if (*m == 0) {
-	ret_val = *m;
-    } else if (a[*m + a_dim1] != 0. || a[*m + *n * a_dim1] != 0.) {
-	ret_val = *m;
-    } else {
-	//    Scan up each column tracking the last zero row seen.
-	ret_val = 0;
-	i__1 = *n;
-	for (j = 1; j <= i__1; ++j) {
-	    i__ = *m;
-	    while(a[max(i__,1) + j * a_dim1] == 0. && i__ >= 1) {
-		--i__;
-	    }
-	    ret_val = max(ret_val,i__);
-	}
-    }
-    return ret_val;
-} // iladlr_
-
@@ -1,824 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b DLARFB applies a block reflector or its transpose to a general rectangular matrix.
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-//> \htmlonly
-//> Download DLARFB + dependencies
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dlarfb.f">
-//> [TGZ]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dlarfb.f">
-//> [ZIP]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dlarfb.f">
-//> [TXT]</a>
-//> \endhtmlonly
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE DLARFB( SIDE, TRANS, DIRECT, STOREV, M, N, K, V, LDV,
-//                         T, LDT, C, LDC, WORK, LDWORK )
-//
-//      .. Scalar Arguments ..
-//      CHARACTER          DIRECT, SIDE, STOREV, TRANS
-//      INTEGER            K, LDC, LDT, LDV, LDWORK, M, N
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION   C( LDC, * ), T( LDT, * ), V( LDV, * ),
-//     $                   WORK( LDWORK, * )
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> DLARFB applies a real block reflector H or its transpose H**T to a
-//> real m by n matrix C, from either the left or the right.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] SIDE
-//> \verbatim
-//>          SIDE is CHARACTER*1
-//>          = 'L': apply H or H**T from the Left
-//>          = 'R': apply H or H**T from the Right
-//> \endverbatim
-//>
-//> \param[in] TRANS
-//> \verbatim
-//>          TRANS is CHARACTER*1
-//>          = 'N': apply H (No transpose)
-//>          = 'T': apply H**T (Transpose)
-//> \endverbatim
-//>
-//> \param[in] DIRECT
-//> \verbatim
-//>          DIRECT is CHARACTER*1
-//>          Indicates how H is formed from a product of elementary
-//>          reflectors
-//>          = 'F': H = H(1) H(2) . . . H(k) (Forward)
-//>          = 'B': H = H(k) . . . H(2) H(1) (Backward)
-//> \endverbatim
-//>
-//> \param[in] STOREV
-//> \verbatim
-//>          STOREV is CHARACTER*1
-//>          Indicates how the vectors which define the elementary
-//>          reflectors are stored:
-//>          = 'C': Columnwise
-//>          = 'R': Rowwise
-//> \endverbatim
-//>
-//> \param[in] M
-//> \verbatim
-//>          M is INTEGER
-//>          The number of rows of the matrix C.
-//> \endverbatim
-//>
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>          The number of columns of the matrix C.
-//> \endverbatim
-//>
-//> \param[in] K
-//> \verbatim
-//>          K is INTEGER
-//>          The order of the matrix T (= the number of elementary
-//>          reflectors whose product defines the block reflector).
-//>          If SIDE = 'L', M >= K >= 0;
-//>          if SIDE = 'R', N >= K >= 0.
-//> \endverbatim
-//>
-//> \param[in] V
-//> \verbatim
-//>          V is DOUBLE PRECISION array, dimension
-//>                                (LDV,K) if STOREV = 'C'
-//>                                (LDV,M) if STOREV = 'R' and SIDE = 'L'
-//>                                (LDV,N) if STOREV = 'R' and SIDE = 'R'
-//>          The matrix V. See Further Details.
-//> \endverbatim
-//>
-//> \param[in] LDV
-//> \verbatim
-//>          LDV is INTEGER
-//>          The leading dimension of the array V.
-//>          If STOREV = 'C' and SIDE = 'L', LDV >= max(1,M);
-//>          if STOREV = 'C' and SIDE = 'R', LDV >= max(1,N);
-//>          if STOREV = 'R', LDV >= K.
-//> \endverbatim
-//>
-//> \param[in] T
-//> \verbatim
-//>          T is DOUBLE PRECISION array, dimension (LDT,K)
-//>          The triangular k by k matrix T in the representation of the
-//>          block reflector.
-//> \endverbatim
-//>
-//> \param[in] LDT
-//> \verbatim
-//>          LDT is INTEGER
-//>          The leading dimension of the array T. LDT >= K.
-//> \endverbatim
-//>
-//> \param[in,out] C
-//> \verbatim
-//>          C is DOUBLE PRECISION array, dimension (LDC,N)
-//>          On entry, the m by n matrix C.
-//>          On exit, C is overwritten by H*C or H**T*C or C*H or C*H**T.
-//> \endverbatim
-//>
-//> \param[in] LDC
-//> \verbatim
-//>          LDC is INTEGER
-//>          The leading dimension of the array C. LDC >= max(1,M).
-//> \endverbatim
-//>
-//> \param[out] WORK
-//> \verbatim
-//>          WORK is DOUBLE PRECISION array, dimension (LDWORK,K)
-//> \endverbatim
-//>
-//> \param[in] LDWORK
-//> \verbatim
-//>          LDWORK is INTEGER
-//>          The leading dimension of the array WORK.
-//>          If SIDE = 'L', LDWORK >= max(1,N);
-//>          if SIDE = 'R', LDWORK >= max(1,M).
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date June 2013
-//
-//> \ingroup doubleOTHERauxiliary
-//
-//> \par Further Details:
-// =====================
-//>
-//> \verbatim
-//>
-//>  The shape of the matrix V and the storage of the vectors which define
-//>  the H(i) is best illustrated by the following example with n = 5 and
-//>  k = 3. The elements equal to 1 are not stored; the corresponding
-//>  array elements are modified but restored on exit. The rest of the
-//>  array is not used.
-//>
-//>  DIRECT = 'F' and STOREV = 'C':         DIRECT = 'F' and STOREV = 'R':
-//>
-//>               V = (  1       )                 V = (  1 v1 v1 v1 v1 )
-//>                   ( v1  1    )                     (     1 v2 v2 v2 )
-//>                   ( v1 v2  1 )                     (        1 v3 v3 )
-//>                   ( v1 v2 v3 )
-//>                   ( v1 v2 v3 )
-//>
-//>  DIRECT = 'B' and STOREV = 'C':         DIRECT = 'B' and STOREV = 'R':
-//>
-//>               V = ( v1 v2 v3 )                 V = ( v1 v1  1       )
-//>                   ( v1 v2 v3 )                     ( v2 v2 v2  1    )
-//>                   (  1 v2 v3 )                     ( v3 v3 v3 v3  1 )
-//>                   (     1 v3 )
-//>                   (        1 )
-//> \endverbatim
-//>
-// =====================================================================
-/* Subroutine */ int dlarfb_(char *side, char *trans, char *direct, char *
-	storev, int *m, int *n, int *k, double *v, int *ldv, double *t, int *
-	ldt, double *c__, int *ldc, double *work, int *ldwork)
-{
-    // Table of constant values
-    int c__1 = 1;
-    double c_b14 = 1.;
-    double c_b25 = -1.;
-
-    // System generated locals
-    int c_dim1, c_offset, t_dim1, t_offset, v_dim1, v_offset, work_dim1,
-	    work_offset, i__1, i__2;
-
-    // Local variables
-    int i__, j;
-    extern /* Subroutine */ int dgemm_(char *, char *, int *, int *, int *,
-	    double *, double *, int *, double *, int *, double *, double *,
-	    int *);
-    extern int lsame_(char *, char *);
-    extern /* Subroutine */ int dcopy_(int *, double *, int *, double *, int *
-	    ), dtrmm_(char *, char *, char *, char *, int *, int *, double *,
-	    double *, int *, double *, int *);
-    char transt[1+1]={'\0'};
-
-    //
-    // -- LAPACK auxiliary routine (version 3.7.0) --
-    // -- LAPACK is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    June 2013
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. Parameters ..
-    //    ..
-    //    .. Local Scalars ..
-    //    ..
-    //    .. External Functions ..
-    //    ..
-    //    .. External Subroutines ..
-    //    ..
-    //    .. Executable Statements ..
-    //
-    //    Quick return if possible
-    //
-    // Parameter adjustments
-    v_dim1 = *ldv;
-    v_offset = 1 + v_dim1;
-    v -= v_offset;
-    t_dim1 = *ldt;
-    t_offset = 1 + t_dim1;
-    t -= t_offset;
-    c_dim1 = *ldc;
-    c_offset = 1 + c_dim1;
-    c__ -= c_offset;
-    work_dim1 = *ldwork;
-    work_offset = 1 + work_dim1;
-    work -= work_offset;
-
-    // Function Body
-    if (*m <= 0 || *n <= 0) {
-	return 0;
-    }
-    if (lsame_(trans, "N")) {
-	*(unsigned char *)transt = 'T';
-    } else {
-	*(unsigned char *)transt = 'N';
-    }
-    if (lsame_(storev, "C")) {
-	if (lsame_(direct, "F")) {
-	    //
-	    //          Let  V =  ( V1 )    (first K rows)
-	    //                    ( V2 )
-	    //          where  V1  is unit lower triangular.
-	    //
-	    if (lsame_(side, "L")) {
-		//
-		//             Form  H * C  or  H**T * C  where  C = ( C1 )
-		//                                                   ( C2 )
-		//
-		//             W := C**T * V  =  (C1**T * V1 + C2**T * V2)  (stored in WORK)
-		//
-		//             W := C1**T
-		//
-		i__1 = *k;
-		for (j = 1; j <= i__1; ++j) {
-		    dcopy_(n, &c__[j + c_dim1], ldc, &work[j * work_dim1 + 1],
-			     &c__1);
-// L10:
-		}
-		//
-		//             W := W * V1
-		//
-		dtrmm_("Right", "Lower", "No transpose", "Unit", n, k, &c_b14,
-			 &v[v_offset], ldv, &work[work_offset], ldwork);
-		if (*m > *k) {
-		    //
-		    //                W := W + C2**T * V2
-		    //
-		    i__1 = *m - *k;
-		    dgemm_("Transpose", "No transpose", n, k, &i__1, &c_b14, &
-			    c__[*k + 1 + c_dim1], ldc, &v[*k + 1 + v_dim1],
-			    ldv, &c_b14, &work[work_offset], ldwork);
-		}
-		//
-		//             W := W * T**T  or  W * T
-		//
-		dtrmm_("Right", "Upper", transt, "Non-unit", n, k, &c_b14, &t[
-			t_offset], ldt, &work[work_offset], ldwork);
-		//
-		//             C := C - V * W**T
-		//
-		if (*m > *k) {
-		    //
-		    //                C2 := C2 - V2 * W**T
-		    //
-		    i__1 = *m - *k;
-		    dgemm_("No transpose", "Transpose", &i__1, n, k, &c_b25, &
-			    v[*k + 1 + v_dim1], ldv, &work[work_offset],
-			    ldwork, &c_b14, &c__[*k + 1 + c_dim1], ldc);
-		}
-		//
-		//             W := W * V1**T
-		//
-		dtrmm_("Right", "Lower", "Transpose", "Unit", n, k, &c_b14, &
-			v[v_offset], ldv, &work[work_offset], ldwork);
-		//
-		//             C1 := C1 - W**T
-		//
-		i__1 = *k;
-		for (j = 1; j <= i__1; ++j) {
-		    i__2 = *n;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			c__[j + i__ * c_dim1] -= work[i__ + j * work_dim1];
-// L20:
-		    }
-// L30:
-		}
-	    } else if (lsame_(side, "R")) {
-		//
-		//             Form  C * H  or  C * H**T  where  C = ( C1  C2 )
-		//
-		//             W := C * V  =  (C1*V1 + C2*V2)  (stored in WORK)
-		//
-		//             W := C1
-		//
-		i__1 = *k;
-		for (j = 1; j <= i__1; ++j) {
-		    dcopy_(m, &c__[j * c_dim1 + 1], &c__1, &work[j *
-			    work_dim1 + 1], &c__1);
-// L40:
-		}
-		//
-		//             W := W * V1
-		//
-		dtrmm_("Right", "Lower", "No transpose", "Unit", m, k, &c_b14,
-			 &v[v_offset], ldv, &work[work_offset], ldwork);
-		if (*n > *k) {
-		    //
-		    //                W := W + C2 * V2
-		    //
-		    i__1 = *n - *k;
-		    dgemm_("No transpose", "No transpose", m, k, &i__1, &
-			    c_b14, &c__[(*k + 1) * c_dim1 + 1], ldc, &v[*k +
-			    1 + v_dim1], ldv, &c_b14, &work[work_offset],
-			    ldwork);
-		}
-		//
-		//             W := W * T  or  W * T**T
-		//
-		dtrmm_("Right", "Upper", trans, "Non-unit", m, k, &c_b14, &t[
-			t_offset], ldt, &work[work_offset], ldwork);
-		//
-		//             C := C - W * V**T
-		//
-		if (*n > *k) {
-		    //
-		    //                C2 := C2 - W * V2**T
-		    //
-		    i__1 = *n - *k;
-		    dgemm_("No transpose", "Transpose", m, &i__1, k, &c_b25, &
-			    work[work_offset], ldwork, &v[*k + 1 + v_dim1],
-			    ldv, &c_b14, &c__[(*k + 1) * c_dim1 + 1], ldc);
-		}
-		//
-		//             W := W * V1**T
-		//
-		dtrmm_("Right", "Lower", "Transpose", "Unit", m, k, &c_b14, &
-			v[v_offset], ldv, &work[work_offset], ldwork);
-		//
-		//             C1 := C1 - W
-		//
-		i__1 = *k;
-		for (j = 1; j <= i__1; ++j) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			c__[i__ + j * c_dim1] -= work[i__ + j * work_dim1];
-// L50:
-		    }
-// L60:
-		}
-	    }
-	} else {
-	    //
-	    //          Let  V =  ( V1 )
-	    //                    ( V2 )    (last K rows)
-	    //          where  V2  is unit upper triangular.
-	    //
-	    if (lsame_(side, "L")) {
-		//
-		//             Form  H * C  or  H**T * C  where  C = ( C1 )
-		//                                                   ( C2 )
-		//
-		//             W := C**T * V  =  (C1**T * V1 + C2**T * V2)  (stored in WORK)
-		//
-		//             W := C2**T
-		//
-		i__1 = *k;
-		for (j = 1; j <= i__1; ++j) {
-		    dcopy_(n, &c__[*m - *k + j + c_dim1], ldc, &work[j *
-			    work_dim1 + 1], &c__1);
-// L70:
-		}
-		//
-		//             W := W * V2
-		//
-		dtrmm_("Right", "Upper", "No transpose", "Unit", n, k, &c_b14,
-			 &v[*m - *k + 1 + v_dim1], ldv, &work[work_offset],
-			ldwork);
-		if (*m > *k) {
-		    //
-		    //                W := W + C1**T * V1
-		    //
-		    i__1 = *m - *k;
-		    dgemm_("Transpose", "No transpose", n, k, &i__1, &c_b14, &
-			    c__[c_offset], ldc, &v[v_offset], ldv, &c_b14, &
-			    work[work_offset], ldwork);
-		}
-		//
-		//             W := W * T**T  or  W * T
-		//
-		dtrmm_("Right", "Lower", transt, "Non-unit", n, k, &c_b14, &t[
-			t_offset], ldt, &work[work_offset], ldwork);
-		//
-		//             C := C - V * W**T
-		//
-		if (*m > *k) {
-		    //
-		    //                C1 := C1 - V1 * W**T
-		    //
-		    i__1 = *m - *k;
-		    dgemm_("No transpose", "Transpose", &i__1, n, k, &c_b25, &
-			    v[v_offset], ldv, &work[work_offset], ldwork, &
-			    c_b14, &c__[c_offset], ldc);
-		}
-		//
-		//             W := W * V2**T
-		//
-		dtrmm_("Right", "Upper", "Transpose", "Unit", n, k, &c_b14, &
-			v[*m - *k + 1 + v_dim1], ldv, &work[work_offset],
-			ldwork);
-		//
-		//             C2 := C2 - W**T
-		//
-		i__1 = *k;
-		for (j = 1; j <= i__1; ++j) {
-		    i__2 = *n;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			c__[*m - *k + j + i__ * c_dim1] -= work[i__ + j *
-				work_dim1];
-// L80:
-		    }
-// L90:
-		}
-	    } else if (lsame_(side, "R")) {
-		//
-		//             Form  C * H  or  C * H**T  where  C = ( C1  C2 )
-		//
-		//             W := C * V  =  (C1*V1 + C2*V2)  (stored in WORK)
-		//
-		//             W := C2
-		//
-		i__1 = *k;
-		for (j = 1; j <= i__1; ++j) {
-		    dcopy_(m, &c__[(*n - *k + j) * c_dim1 + 1], &c__1, &work[
-			    j * work_dim1 + 1], &c__1);
-// L100:
-		}
-		//
-		//             W := W * V2
-		//
-		dtrmm_("Right", "Upper", "No transpose", "Unit", m, k, &c_b14,
-			 &v[*n - *k + 1 + v_dim1], ldv, &work[work_offset],
-			ldwork);
-		if (*n > *k) {
-		    //
-		    //                W := W + C1 * V1
-		    //
-		    i__1 = *n - *k;
-		    dgemm_("No transpose", "No transpose", m, k, &i__1, &
-			    c_b14, &c__[c_offset], ldc, &v[v_offset], ldv, &
-			    c_b14, &work[work_offset], ldwork);
-		}
-		//
-		//             W := W * T  or  W * T**T
-		//
-		dtrmm_("Right", "Lower", trans, "Non-unit", m, k, &c_b14, &t[
-			t_offset], ldt, &work[work_offset], ldwork);
-		//
-		//             C := C - W * V**T
-		//
-		if (*n > *k) {
-		    //
-		    //                C1 := C1 - W * V1**T
-		    //
-		    i__1 = *n - *k;
-		    dgemm_("No transpose", "Transpose", m, &i__1, k, &c_b25, &
-			    work[work_offset], ldwork, &v[v_offset], ldv, &
-			    c_b14, &c__[c_offset], ldc);
-		}
-		//
-		//             W := W * V2**T
-		//
-		dtrmm_("Right", "Upper", "Transpose", "Unit", m, k, &c_b14, &
-			v[*n - *k + 1 + v_dim1], ldv, &work[work_offset],
-			ldwork);
-		//
-		//             C2 := C2 - W
-		//
-		i__1 = *k;
-		for (j = 1; j <= i__1; ++j) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			c__[i__ + (*n - *k + j) * c_dim1] -= work[i__ + j *
-				work_dim1];
-// L110:
-		    }
-// L120:
-		}
-	    }
-	}
-    } else if (lsame_(storev, "R")) {
-	if (lsame_(direct, "F")) {
-	    //
-	    //          Let  V =  ( V1  V2 )    (V1: first K columns)
-	    //          where  V1  is unit upper triangular.
-	    //
-	    if (lsame_(side, "L")) {
-		//
-		//             Form  H * C  or  H**T * C  where  C = ( C1 )
-		//                                                   ( C2 )
-		//
-		//             W := C**T * V**T  =  (C1**T * V1**T + C2**T * V2**T) (stored in WORK)
-		//
-		//             W := C1**T
-		//
-		i__1 = *k;
-		for (j = 1; j <= i__1; ++j) {
-		    dcopy_(n, &c__[j + c_dim1], ldc, &work[j * work_dim1 + 1],
-			     &c__1);
-// L130:
-		}
-		//
-		//             W := W * V1**T
-		//
-		dtrmm_("Right", "Upper", "Transpose", "Unit", n, k, &c_b14, &
-			v[v_offset], ldv, &work[work_offset], ldwork);
-		if (*m > *k) {
-		    //
-		    //                W := W + C2**T * V2**T
-		    //
-		    i__1 = *m - *k;
-		    dgemm_("Transpose", "Transpose", n, k, &i__1, &c_b14, &
-			    c__[*k + 1 + c_dim1], ldc, &v[(*k + 1) * v_dim1 +
-			    1], ldv, &c_b14, &work[work_offset], ldwork);
-		}
-		//
-		//             W := W * T**T  or  W * T
-		//
-		dtrmm_("Right", "Upper", transt, "Non-unit", n, k, &c_b14, &t[
-			t_offset], ldt, &work[work_offset], ldwork);
-		//
-		//             C := C - V**T * W**T
-		//
-		if (*m > *k) {
-		    //
-		    //                C2 := C2 - V2**T * W**T
-		    //
-		    i__1 = *m - *k;
-		    dgemm_("Transpose", "Transpose", &i__1, n, k, &c_b25, &v[(
-			    *k + 1) * v_dim1 + 1], ldv, &work[work_offset],
-			    ldwork, &c_b14, &c__[*k + 1 + c_dim1], ldc);
-		}
-		//
-		//             W := W * V1
-		//
-		dtrmm_("Right", "Upper", "No transpose", "Unit", n, k, &c_b14,
-			 &v[v_offset], ldv, &work[work_offset], ldwork);
-		//
-		//             C1 := C1 - W**T
-		//
-		i__1 = *k;
-		for (j = 1; j <= i__1; ++j) {
-		    i__2 = *n;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			c__[j + i__ * c_dim1] -= work[i__ + j * work_dim1];
-// L140:
-		    }
-// L150:
-		}
-	    } else if (lsame_(side, "R")) {
-		//
-		//             Form  C * H  or  C * H**T  where  C = ( C1  C2 )
-		//
-		//             W := C * V**T  =  (C1*V1**T + C2*V2**T)  (stored in WORK)
-		//
-		//             W := C1
-		//
-		i__1 = *k;
-		for (j = 1; j <= i__1; ++j) {
-		    dcopy_(m, &c__[j * c_dim1 + 1], &c__1, &work[j *
-			    work_dim1 + 1], &c__1);
-// L160:
-		}
-		//
-		//             W := W * V1**T
-		//
-		dtrmm_("Right", "Upper", "Transpose", "Unit", m, k, &c_b14, &
-			v[v_offset], ldv, &work[work_offset], ldwork);
-		if (*n > *k) {
-		    //
-		    //                W := W + C2 * V2**T
-		    //
-		    i__1 = *n - *k;
-		    dgemm_("No transpose", "Transpose", m, k, &i__1, &c_b14, &
-			    c__[(*k + 1) * c_dim1 + 1], ldc, &v[(*k + 1) *
-			    v_dim1 + 1], ldv, &c_b14, &work[work_offset],
-			    ldwork);
-		}
-		//
-		//             W := W * T  or  W * T**T
-		//
-		dtrmm_("Right", "Upper", trans, "Non-unit", m, k, &c_b14, &t[
-			t_offset], ldt, &work[work_offset], ldwork);
-		//
-		//             C := C - W * V
-		//
-		if (*n > *k) {
-		    //
-		    //                C2 := C2 - W * V2
-		    //
-		    i__1 = *n - *k;
-		    dgemm_("No transpose", "No transpose", m, &i__1, k, &
-			    c_b25, &work[work_offset], ldwork, &v[(*k + 1) *
-			    v_dim1 + 1], ldv, &c_b14, &c__[(*k + 1) * c_dim1
-			    + 1], ldc);
-		}
-		//
-		//             W := W * V1
-		//
-		dtrmm_("Right", "Upper", "No transpose", "Unit", m, k, &c_b14,
-			 &v[v_offset], ldv, &work[work_offset], ldwork);
-		//
-		//             C1 := C1 - W
-		//
-		i__1 = *k;
-		for (j = 1; j <= i__1; ++j) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			c__[i__ + j * c_dim1] -= work[i__ + j * work_dim1];
-// L170:
-		    }
-// L180:
-		}
-	    }
-	} else {
-	    //
-	    //          Let  V =  ( V1  V2 )    (V2: last K columns)
-	    //          where  V2  is unit lower triangular.
-	    //
-	    if (lsame_(side, "L")) {
-		//
-		//             Form  H * C  or  H**T * C  where  C = ( C1 )
-		//                                                   ( C2 )
-		//
-		//             W := C**T * V**T  =  (C1**T * V1**T + C2**T * V2**T) (stored in WORK)
-		//
-		//             W := C2**T
-		//
-		i__1 = *k;
-		for (j = 1; j <= i__1; ++j) {
-		    dcopy_(n, &c__[*m - *k + j + c_dim1], ldc, &work[j *
-			    work_dim1 + 1], &c__1);
-// L190:
-		}
-		//
-		//             W := W * V2**T
-		//
-		dtrmm_("Right", "Lower", "Transpose", "Unit", n, k, &c_b14, &
-			v[(*m - *k + 1) * v_dim1 + 1], ldv, &work[work_offset]
-			, ldwork);
-		if (*m > *k) {
-		    //
-		    //                W := W + C1**T * V1**T
-		    //
-		    i__1 = *m - *k;
-		    dgemm_("Transpose", "Transpose", n, k, &i__1, &c_b14, &
-			    c__[c_offset], ldc, &v[v_offset], ldv, &c_b14, &
-			    work[work_offset], ldwork);
-		}
-		//
-		//             W := W * T**T  or  W * T
-		//
-		dtrmm_("Right", "Lower", transt, "Non-unit", n, k, &c_b14, &t[
-			t_offset], ldt, &work[work_offset], ldwork);
-		//
-		//             C := C - V**T * W**T
-		//
-		if (*m > *k) {
-		    //
-		    //                C1 := C1 - V1**T * W**T
-		    //
-		    i__1 = *m - *k;
-		    dgemm_("Transpose", "Transpose", &i__1, n, k, &c_b25, &v[
-			    v_offset], ldv, &work[work_offset], ldwork, &
-			    c_b14, &c__[c_offset], ldc);
-		}
-		//
-		//             W := W * V2
-		//
-		dtrmm_("Right", "Lower", "No transpose", "Unit", n, k, &c_b14,
-			 &v[(*m - *k + 1) * v_dim1 + 1], ldv, &work[
-			work_offset], ldwork);
-		//
-		//             C2 := C2 - W**T
-		//
-		i__1 = *k;
-		for (j = 1; j <= i__1; ++j) {
-		    i__2 = *n;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			c__[*m - *k + j + i__ * c_dim1] -= work[i__ + j *
-				work_dim1];
-// L200:
-		    }
-// L210:
-		}
-	    } else if (lsame_(side, "R")) {
-		//
-		//             Form  C * H  or  C * H'  where  C = ( C1  C2 )
-		//
-		//             W := C * V**T  =  (C1*V1**T + C2*V2**T)  (stored in WORK)
-		//
-		//             W := C2
-		//
-		i__1 = *k;
-		for (j = 1; j <= i__1; ++j) {
-		    dcopy_(m, &c__[(*n - *k + j) * c_dim1 + 1], &c__1, &work[
-			    j * work_dim1 + 1], &c__1);
-// L220:
-		}
-		//
-		//             W := W * V2**T
-		//
-		dtrmm_("Right", "Lower", "Transpose", "Unit", m, k, &c_b14, &
-			v[(*n - *k + 1) * v_dim1 + 1], ldv, &work[work_offset]
-			, ldwork);
-		if (*n > *k) {
-		    //
-		    //                W := W + C1 * V1**T
-		    //
-		    i__1 = *n - *k;
-		    dgemm_("No transpose", "Transpose", m, k, &i__1, &c_b14, &
-			    c__[c_offset], ldc, &v[v_offset], ldv, &c_b14, &
-			    work[work_offset], ldwork);
-		}
-		//
-		//             W := W * T  or  W * T**T
-		//
-		dtrmm_("Right", "Lower", trans, "Non-unit", m, k, &c_b14, &t[
-			t_offset], ldt, &work[work_offset], ldwork);
-		//
-		//             C := C - W * V
-		//
-		if (*n > *k) {
-		    //
-		    //                C1 := C1 - W * V1
-		    //
-		    i__1 = *n - *k;
-		    dgemm_("No transpose", "No transpose", m, &i__1, k, &
-			    c_b25, &work[work_offset], ldwork, &v[v_offset],
-			    ldv, &c_b14, &c__[c_offset], ldc);
-		}
-		//
-		//             W := W * V2
-		//
-		dtrmm_("Right", "Lower", "No transpose", "Unit", m, k, &c_b14,
-			 &v[(*n - *k + 1) * v_dim1 + 1], ldv, &work[
-			work_offset], ldwork);
-		//
-		//             C1 := C1 - W
-		//
-		i__1 = *k;
-		for (j = 1; j <= i__1; ++j) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			c__[i__ + (*n - *k + j) * c_dim1] -= work[i__ + j *
-				work_dim1];
-// L230:
-		    }
-// L240:
-		}
-	    }
-	}
-    }
-    return 0;
-    //
-    //    End of DLARFB
-    //
-} // dlarfb_
-
@@ -1,216 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b DLARFG generates an elementary reflector (Householder matrix).
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-//> \htmlonly
-//> Download DLARFG + dependencies
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dlarfg.f">
-//> [TGZ]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dlarfg.f">
-//> [ZIP]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dlarfg.f">
-//> [TXT]</a>
-//> \endhtmlonly
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE DLARFG( N, ALPHA, X, INCX, TAU )
-//
-//      .. Scalar Arguments ..
-//      INTEGER            INCX, N
-//      DOUBLE PRECISION   ALPHA, TAU
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION   X( * )
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> DLARFG generates a real elementary reflector H of order n, such
-//> that
-//>
-//>       H * ( alpha ) = ( beta ),   H**T * H = I.
-//>           (   x   )   (   0  )
-//>
-//> where alpha and beta are scalars, and x is an (n-1)-element real
-//> vector. H is represented in the form
-//>
-//>       H = I - tau * ( 1 ) * ( 1 v**T ) ,
-//>                     ( v )
-//>
-//> where tau is a real scalar and v is a real (n-1)-element
-//> vector.
-//>
-//> If the elements of x are all zero, then tau = 0 and H is taken to be
-//> the unit matrix.
-//>
-//> Otherwise  1 <= tau <= 2.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>          The order of the elementary reflector.
-//> \endverbatim
-//>
-//> \param[in,out] ALPHA
-//> \verbatim
-//>          ALPHA is DOUBLE PRECISION
-//>          On entry, the value alpha.
-//>          On exit, it is overwritten with the value beta.
-//> \endverbatim
-//>
-//> \param[in,out] X
-//> \verbatim
-//>          X is DOUBLE PRECISION array, dimension
-//>                         (1+(N-2)*abs(INCX))
-//>          On entry, the vector x.
-//>          On exit, it is overwritten with the vector v.
-//> \endverbatim
-//>
-//> \param[in] INCX
-//> \verbatim
-//>          INCX is INTEGER
-//>          The increment between elements of X. INCX > 0.
-//> \endverbatim
-//>
-//> \param[out] TAU
-//> \verbatim
-//>          TAU is DOUBLE PRECISION
-//>          The value tau.
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date November 2017
-//
-//> \ingroup doubleOTHERauxiliary
-//
-// =====================================================================
-/* Subroutine */ int dlarfg_(int *n, double *alpha, double *x, int *incx,
-	double *tau)
-{
-    // System generated locals
-    int i__1;
-    double d__1;
-
-    // Local variables
-    int j, knt;
-    double beta;
-    extern double dnrm2_(int *, double *, int *);
-    extern /* Subroutine */ int dscal_(int *, double *, double *, int *);
-    double xnorm;
-    extern double dlapy2_(double *, double *), dlamch_(char *);
-    double safmin, rsafmn;
-
-    //
-    // -- LAPACK auxiliary routine (version 3.8.0) --
-    // -- LAPACK is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    November 2017
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. Parameters ..
-    //    ..
-    //    .. Local Scalars ..
-    //    ..
-    //    .. External Functions ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    //    .. External Subroutines ..
-    //    ..
-    //    .. Executable Statements ..
-    //
-    // Parameter adjustments
-    --x;
-
-    // Function Body
-    if (*n <= 1) {
-	*tau = 0.;
-	return 0;
-    }
-    i__1 = *n - 1;
-    xnorm = dnrm2_(&i__1, &x[1], incx);
-    if (xnorm == 0.) {
-	//
-	//       H  =  I
-	//
-	*tau = 0.;
-    } else {
-	//
-	//       general case
-	//
-	d__1 = dlapy2_(alpha, &xnorm);
-	beta = -d_sign(&d__1, alpha);
-	safmin = dlamch_("S") / dlamch_("E");
-	knt = 0;
-	if (abs(beta) < safmin) {
-	    //
-	    //          XNORM, BETA may be inaccurate; scale X and recompute them
-	    //
-	    rsafmn = 1. / safmin;
-L10:
-	    ++knt;
-	    i__1 = *n - 1;
-	    dscal_(&i__1, &rsafmn, &x[1], incx);
-	    beta *= rsafmn;
-	    *alpha *= rsafmn;
-	    if (abs(beta) < safmin && knt < 20) {
-		goto L10;
-	    }
-	    //
-	    //          New BETA is at most 1, at least SAFMIN
-	    //
-	    i__1 = *n - 1;
-	    xnorm = dnrm2_(&i__1, &x[1], incx);
-	    d__1 = dlapy2_(alpha, &xnorm);
-	    beta = -d_sign(&d__1, alpha);
-	}
-	*tau = (beta - *alpha) / beta;
-	i__1 = *n - 1;
-	d__1 = 1. / (*alpha - beta);
-	dscal_(&i__1, &d__1, &x[1], incx);
-	//
-	//       If ALPHA is subnormal, it may lose relative accuracy
-	//
-	i__1 = knt;
-	for (j = 1; j <= i__1; ++j) {
-	    beta *= safmin;
-// L20:
-	}
-	*alpha = beta;
-    }
-    return 0;
-    //
-    //    End of DLARFG
-    //
-} // dlarfg_
-
@@ -1,389 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b DLARFT forms the triangular factor T of a block reflector H = I - vtvH
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-//> \htmlonly
-//> Download DLARFT + dependencies
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dlarft.f">
-//> [TGZ]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dlarft.f">
-//> [ZIP]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dlarft.f">
-//> [TXT]</a>
-//> \endhtmlonly
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE DLARFT( DIRECT, STOREV, N, K, V, LDV, TAU, T, LDT )
-//
-//      .. Scalar Arguments ..
-//      CHARACTER          DIRECT, STOREV
-//      INTEGER            K, LDT, LDV, N
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION   T( LDT, * ), TAU( * ), V( LDV, * )
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> DLARFT forms the triangular factor T of a real block reflector H
-//> of order n, which is defined as a product of k elementary reflectors.
-//>
-//> If DIRECT = 'F', H = H(1) H(2) . . . H(k) and T is upper triangular;
-//>
-//> If DIRECT = 'B', H = H(k) . . . H(2) H(1) and T is lower triangular.
-//>
-//> If STOREV = 'C', the vector which defines the elementary reflector
-//> H(i) is stored in the i-th column of the array V, and
-//>
-//>    H  =  I - V * T * V**T
-//>
-//> If STOREV = 'R', the vector which defines the elementary reflector
-//> H(i) is stored in the i-th row of the array V, and
-//>
-//>    H  =  I - V**T * T * V
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] DIRECT
-//> \verbatim
-//>          DIRECT is CHARACTER*1
-//>          Specifies the order in which the elementary reflectors are
-//>          multiplied to form the block reflector:
-//>          = 'F': H = H(1) H(2) . . . H(k) (Forward)
-//>          = 'B': H = H(k) . . . H(2) H(1) (Backward)
-//> \endverbatim
-//>
-//> \param[in] STOREV
-//> \verbatim
-//>          STOREV is CHARACTER*1
-//>          Specifies how the vectors which define the elementary
-//>          reflectors are stored (see also Further Details):
-//>          = 'C': columnwise
-//>          = 'R': rowwise
-//> \endverbatim
-//>
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>          The order of the block reflector H. N >= 0.
-//> \endverbatim
-//>
-//> \param[in] K
-//> \verbatim
-//>          K is INTEGER
-//>          The order of the triangular factor T (= the number of
-//>          elementary reflectors). K >= 1.
-//> \endverbatim
-//>
-//> \param[in] V
-//> \verbatim
-//>          V is DOUBLE PRECISION array, dimension
-//>                               (LDV,K) if STOREV = 'C'
-//>                               (LDV,N) if STOREV = 'R'
-//>          The matrix V. See further details.
-//> \endverbatim
-//>
-//> \param[in] LDV
-//> \verbatim
-//>          LDV is INTEGER
-//>          The leading dimension of the array V.
-//>          If STOREV = 'C', LDV >= max(1,N); if STOREV = 'R', LDV >= K.
-//> \endverbatim
-//>
-//> \param[in] TAU
-//> \verbatim
-//>          TAU is DOUBLE PRECISION array, dimension (K)
-//>          TAU(i) must contain the scalar factor of the elementary
-//>          reflector H(i).
-//> \endverbatim
-//>
-//> \param[out] T
-//> \verbatim
-//>          T is DOUBLE PRECISION array, dimension (LDT,K)
-//>          The k by k triangular factor T of the block reflector.
-//>          If DIRECT = 'F', T is upper triangular; if DIRECT = 'B', T is
-//>          lower triangular. The rest of the array is not used.
-//> \endverbatim
-//>
-//> \param[in] LDT
-//> \verbatim
-//>          LDT is INTEGER
-//>          The leading dimension of the array T. LDT >= K.
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date December 2016
-//
-//> \ingroup doubleOTHERauxiliary
-//
-//> \par Further Details:
-// =====================
-//>
-//> \verbatim
-//>
-//>  The shape of the matrix V and the storage of the vectors which define
-//>  the H(i) is best illustrated by the following example with n = 5 and
-//>  k = 3. The elements equal to 1 are not stored.
-//>
-//>  DIRECT = 'F' and STOREV = 'C':         DIRECT = 'F' and STOREV = 'R':
-//>
-//>               V = (  1       )                 V = (  1 v1 v1 v1 v1 )
-//>                   ( v1  1    )                     (     1 v2 v2 v2 )
-//>                   ( v1 v2  1 )                     (        1 v3 v3 )
-//>                   ( v1 v2 v3 )
-//>                   ( v1 v2 v3 )
-//>
-//>  DIRECT = 'B' and STOREV = 'C':         DIRECT = 'B' and STOREV = 'R':
-//>
-//>               V = ( v1 v2 v3 )                 V = ( v1 v1  1       )
-//>                   ( v1 v2 v3 )                     ( v2 v2 v2  1    )
-//>                   (  1 v2 v3 )                     ( v3 v3 v3 v3  1 )
-//>                   (     1 v3 )
-//>                   (        1 )
-//> \endverbatim
-//>
-// =====================================================================
-/* Subroutine */ int dlarft_(char *direct, char *storev, int *n, int *k,
-	double *v, int *ldv, double *tau, double *t, int *ldt)
-{
-    // Table of constant values
-    int c__1 = 1;
-    double c_b7 = 1.;
-
-    // System generated locals
-    int t_dim1, t_offset, v_dim1, v_offset, i__1, i__2, i__3;
-    double d__1;
-
-    // Local variables
-    int i__, j, prevlastv;
-    extern int lsame_(char *, char *);
-    extern /* Subroutine */ int dgemv_(char *, int *, int *, double *, double
-	    *, int *, double *, int *, double *, double *, int *);
-    int lastv;
-    extern /* Subroutine */ int dtrmv_(char *, char *, char *, int *, double *
-	    , int *, double *, int *);
-
-    //
-    // -- LAPACK auxiliary routine (version 3.7.0) --
-    // -- LAPACK is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    December 2016
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. Parameters ..
-    //    ..
-    //    .. Local Scalars ..
-    //    ..
-    //    .. External Subroutines ..
-    //    ..
-    //    .. External Functions ..
-    //    ..
-    //    .. Executable Statements ..
-    //
-    //    Quick return if possible
-    //
-    // Parameter adjustments
-    v_dim1 = *ldv;
-    v_offset = 1 + v_dim1;
-    v -= v_offset;
-    --tau;
-    t_dim1 = *ldt;
-    t_offset = 1 + t_dim1;
-    t -= t_offset;
-
-    // Function Body
-    if (*n == 0) {
-	return 0;
-    }
-    if (lsame_(direct, "F")) {
-	prevlastv = *n;
-	i__1 = *k;
-	for (i__ = 1; i__ <= i__1; ++i__) {
-	    prevlastv = max(i__,prevlastv);
-	    if (tau[i__] == 0.) {
-		//
-		//             H(i)  =  I
-		//
-		i__2 = i__;
-		for (j = 1; j <= i__2; ++j) {
-		    t[j + i__ * t_dim1] = 0.;
-		}
-	    } else {
-		//
-		//             general case
-		//
-		if (lsame_(storev, "C")) {
-		    //                Skip any trailing zeros.
-		    i__2 = i__ + 1;
-		    for (lastv = *n; lastv >= i__2; --lastv) {
-			if (v[lastv + i__ * v_dim1] != 0.) {
-			    break;
-			}
-		    }
-		    i__2 = i__ - 1;
-		    for (j = 1; j <= i__2; ++j) {
-			t[j + i__ * t_dim1] = -tau[i__] * v[i__ + j * v_dim1];
-		    }
-		    j = min(lastv,prevlastv);
-		    //
-		    //                T(1:i-1,i) := - tau(i) * V(i:j,1:i-1)**T * V(i:j,i)
-		    //
-		    i__2 = j - i__;
-		    i__3 = i__ - 1;
-		    d__1 = -tau[i__];
-		    dgemv_("Transpose", &i__2, &i__3, &d__1, &v[i__ + 1 +
-			    v_dim1], ldv, &v[i__ + 1 + i__ * v_dim1], &c__1, &
-			    c_b7, &t[i__ * t_dim1 + 1], &c__1);
-		} else {
-		    //                Skip any trailing zeros.
-		    i__2 = i__ + 1;
-		    for (lastv = *n; lastv >= i__2; --lastv) {
-			if (v[i__ + lastv * v_dim1] != 0.) {
-			    break;
-			}
-		    }
-		    i__2 = i__ - 1;
-		    for (j = 1; j <= i__2; ++j) {
-			t[j + i__ * t_dim1] = -tau[i__] * v[j + i__ * v_dim1];
-		    }
-		    j = min(lastv,prevlastv);
-		    //
-		    //                T(1:i-1,i) := - tau(i) * V(1:i-1,i:j) * V(i,i:j)**T
-		    //
-		    i__2 = i__ - 1;
-		    i__3 = j - i__;
-		    d__1 = -tau[i__];
-		    dgemv_("No transpose", &i__2, &i__3, &d__1, &v[(i__ + 1) *
-			     v_dim1 + 1], ldv, &v[i__ + (i__ + 1) * v_dim1],
-			    ldv, &c_b7, &t[i__ * t_dim1 + 1], &c__1);
-		}
-		//
-		//             T(1:i-1,i) := T(1:i-1,1:i-1) * T(1:i-1,i)
-		//
-		i__2 = i__ - 1;
-		dtrmv_("Upper", "No transpose", "Non-unit", &i__2, &t[
-			t_offset], ldt, &t[i__ * t_dim1 + 1], &c__1);
-		t[i__ + i__ * t_dim1] = tau[i__];
-		if (i__ > 1) {
-		    prevlastv = max(prevlastv,lastv);
-		} else {
-		    prevlastv = lastv;
-		}
-	    }
-	}
-    } else {
-	prevlastv = 1;
-	for (i__ = *k; i__ >= 1; --i__) {
-	    if (tau[i__] == 0.) {
-		//
-		//             H(i)  =  I
-		//
-		i__1 = *k;
-		for (j = i__; j <= i__1; ++j) {
-		    t[j + i__ * t_dim1] = 0.;
-		}
-	    } else {
-		//
-		//             general case
-		//
-		if (i__ < *k) {
-		    if (lsame_(storev, "C")) {
-			//                   Skip any leading zeros.
-			i__1 = i__ - 1;
-			for (lastv = 1; lastv <= i__1; ++lastv) {
-			    if (v[lastv + i__ * v_dim1] != 0.) {
-				break;
-			    }
-			}
-			i__1 = *k;
-			for (j = i__ + 1; j <= i__1; ++j) {
-			    t[j + i__ * t_dim1] = -tau[i__] * v[*n - *k + i__
-				    + j * v_dim1];
-			}
-			j = max(lastv,prevlastv);
-			//
-			//                   T(i+1:k,i) = -tau(i) * V(j:n-k+i,i+1:k)**T * V(j:n-k+i,i)
-			//
-			i__1 = *n - *k + i__ - j;
-			i__2 = *k - i__;
-			d__1 = -tau[i__];
-			dgemv_("Transpose", &i__1, &i__2, &d__1, &v[j + (i__
-				+ 1) * v_dim1], ldv, &v[j + i__ * v_dim1], &
-				c__1, &c_b7, &t[i__ + 1 + i__ * t_dim1], &
-				c__1);
-		    } else {
-			//                   Skip any leading zeros.
-			i__1 = i__ - 1;
-			for (lastv = 1; lastv <= i__1; ++lastv) {
-			    if (v[i__ + lastv * v_dim1] != 0.) {
-				break;
-			    }
-			}
-			i__1 = *k;
-			for (j = i__ + 1; j <= i__1; ++j) {
-			    t[j + i__ * t_dim1] = -tau[i__] * v[j + (*n - *k
-				    + i__) * v_dim1];
-			}
-			j = max(lastv,prevlastv);
-			//
-			//                   T(i+1:k,i) = -tau(i) * V(i+1:k,j:n-k+i) * V(i,j:n-k+i)**T
-			//
-			i__1 = *k - i__;
-			i__2 = *n - *k + i__ - j;
-			d__1 = -tau[i__];
-			dgemv_("No transpose", &i__1, &i__2, &d__1, &v[i__ +
-				1 + j * v_dim1], ldv, &v[i__ + j * v_dim1],
-				ldv, &c_b7, &t[i__ + 1 + i__ * t_dim1], &c__1)
-				;
-		    }
-		    //
-		    //                T(i+1:k,i) := T(i+1:k,i+1:k) * T(i+1:k,i)
-		    //
-		    i__1 = *k - i__;
-		    dtrmv_("Lower", "No transpose", "Non-unit", &i__1, &t[i__
-			    + 1 + (i__ + 1) * t_dim1], ldt, &t[i__ + 1 + i__ *
-			     t_dim1], &c__1);
-		    if (i__ > 1) {
-			prevlastv = min(prevlastv,lastv);
-		    } else {
-			prevlastv = lastv;
-		    }
-		}
-		t[i__ + i__ * t_dim1] = tau[i__];
-	    }
-	}
-    }
-    return 0;
-    //
-    //    End of DLARFT
-    //
-} // dlarft_
-
@@ -1,236 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b DLARTG generates a plane rotation with real cosine and real sine.
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-//> \htmlonly
-//> Download DLARTG + dependencies
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dlartg.f">
-//> [TGZ]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dlartg.f">
-//> [ZIP]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dlartg.f">
-//> [TXT]</a>
-//> \endhtmlonly
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE DLARTG( F, G, CS, SN, R )
-//
-//      .. Scalar Arguments ..
-//      DOUBLE PRECISION   CS, F, G, R, SN
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> DLARTG generate a plane rotation so that
-//>
-//>    [  CS  SN  ]  .  [ F ]  =  [ R ]   where CS**2 + SN**2 = 1.
-//>    [ -SN  CS  ]     [ G ]     [ 0 ]
-//>
-//> This is a slower, more accurate version of the BLAS1 routine DROTG,
-//> with the following other differences:
-//>    F and G are unchanged on return.
-//>    If G=0, then CS=1 and SN=0.
-//>    If F=0 and (G .ne. 0), then CS=0 and SN=1 without doing any
-//>       floating point operations (saves work in DBDSQR when
-//>       there are zeros on the diagonal).
-//>
-//> If F exceeds G in magnitude, CS will be positive.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] F
-//> \verbatim
-//>          F is DOUBLE PRECISION
-//>          The first component of vector to be rotated.
-//> \endverbatim
-//>
-//> \param[in] G
-//> \verbatim
-//>          G is DOUBLE PRECISION
-//>          The second component of vector to be rotated.
-//> \endverbatim
-//>
-//> \param[out] CS
-//> \verbatim
-//>          CS is DOUBLE PRECISION
-//>          The cosine of the rotation.
-//> \endverbatim
-//>
-//> \param[out] SN
-//> \verbatim
-//>          SN is DOUBLE PRECISION
-//>          The sine of the rotation.
-//> \endverbatim
-//>
-//> \param[out] R
-//> \verbatim
-//>          R is DOUBLE PRECISION
-//>          The nonzero component of the rotated vector.
-//>
-//>  This version has a few statements commented out for thread safety
-//>  (machine parameters are computed on each entry). 10 feb 03, SJH.
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date December 2016
-//
-//> \ingroup OTHERauxiliary
-//
-// =====================================================================
-/* Subroutine */ int dlartg_(double *f, double *g, double *cs, double *sn,
-	double *r__)
-{
-    // System generated locals
-    int i__1;
-    double d__1, d__2;
-
-    // Local variables
-    int i__;
-    double f1, g1, eps, scale;
-    int count;
-    double safmn2, safmx2;
-    extern double dlamch_(char *);
-    double safmin;
-
-    //
-    // -- LAPACK auxiliary routine (version 3.7.0) --
-    // -- LAPACK is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    December 2016
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. Parameters ..
-    //    ..
-    //    .. Local Scalars ..
-    //    LOGICAL            FIRST
-    //    ..
-    //    .. External Functions ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    //    .. Save statement ..
-    //    SAVE               FIRST, SAFMX2, SAFMIN, SAFMN2
-    //    ..
-    //    .. Data statements ..
-    //    DATA               FIRST / .TRUE. /
-    //    ..
-    //    .. Executable Statements ..
-    //
-    //    IF( FIRST ) THEN
-    safmin = dlamch_("S");
-    eps = dlamch_("E");
-    d__1 = dlamch_("B");
-    i__1 = (int) (log(safmin / eps) / log(dlamch_("B")) / 2.);
-    safmn2 = pow_di(&d__1, &i__1);
-    safmx2 = 1. / safmn2;
-    //       FIRST = .FALSE.
-    //    END IF
-    if (*g == 0.) {
-	*cs = 1.;
-	*sn = 0.;
-	*r__ = *f;
-    } else if (*f == 0.) {
-	*cs = 0.;
-	*sn = 1.;
-	*r__ = *g;
-    } else {
-	f1 = *f;
-	g1 = *g;
-	// Computing MAX
-	d__1 = abs(f1), d__2 = abs(g1);
-	scale = max(d__1,d__2);
-	if (scale >= safmx2) {
-	    count = 0;
-L10:
-	    ++count;
-	    f1 *= safmn2;
-	    g1 *= safmn2;
-	    // Computing MAX
-	    d__1 = abs(f1), d__2 = abs(g1);
-	    scale = max(d__1,d__2);
-	    if (scale >= safmx2) {
-		goto L10;
-	    }
-	    // Computing 2nd power
-	    d__1 = f1;
-	    // Computing 2nd power
-	    d__2 = g1;
-	    *r__ = sqrt(d__1 * d__1 + d__2 * d__2);
-	    *cs = f1 / *r__;
-	    *sn = g1 / *r__;
-	    i__1 = count;
-	    for (i__ = 1; i__ <= i__1; ++i__) {
-		*r__ *= safmx2;
-// L20:
-	    }
-	} else if (scale <= safmn2) {
-	    count = 0;
-L30:
-	    ++count;
-	    f1 *= safmx2;
-	    g1 *= safmx2;
-	    // Computing MAX
-	    d__1 = abs(f1), d__2 = abs(g1);
-	    scale = max(d__1,d__2);
-	    if (scale <= safmn2) {
-		goto L30;
-	    }
-	    // Computing 2nd power
-	    d__1 = f1;
-	    // Computing 2nd power
-	    d__2 = g1;
-	    *r__ = sqrt(d__1 * d__1 + d__2 * d__2);
-	    *cs = f1 / *r__;
-	    *sn = g1 / *r__;
-	    i__1 = count;
-	    for (i__ = 1; i__ <= i__1; ++i__) {
-		*r__ *= safmn2;
-// L40:
-	    }
-	} else {
-	    // Computing 2nd power
-	    d__1 = f1;
-	    // Computing 2nd power
-	    d__2 = g1;
-	    *r__ = sqrt(d__1 * d__1 + d__2 * d__2);
-	    *cs = f1 / *r__;
-	    *sn = g1 / *r__;
-	}
-	if (abs(*f) > abs(*g) && *cs < 0.) {
-	    *cs = -(*cs);
-	    *sn = -(*sn);
-	    *r__ = -(*r__);
-	}
-    }
-    return 0;
-    //
-    //    End of DLARTG
-    //
-} // dlartg_
-
@@ -1,413 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b DLASCL multiplies a general rectangular matrix by a real scalar defined as cto/cfrom.
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-//> \htmlonly
-//> Download DLASCL + dependencies
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dlascl.f">
-//> [TGZ]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dlascl.f">
-//> [ZIP]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dlascl.f">
-//> [TXT]</a>
-//> \endhtmlonly
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE DLASCL( TYPE, KL, KU, CFROM, CTO, M, N, A, LDA, INFO )
-//
-//      .. Scalar Arguments ..
-//      CHARACTER          TYPE
-//      INTEGER            INFO, KL, KU, LDA, M, N
-//      DOUBLE PRECISION   CFROM, CTO
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION   A( LDA, * )
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> DLASCL multiplies the M by N real matrix A by the real scalar
-//> CTO/CFROM.  This is done without over/underflow as long as the final
-//> result CTO*A(I,J)/CFROM does not over/underflow. TYPE specifies that
-//> A may be full, upper triangular, lower triangular, upper Hessenberg,
-//> or banded.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] TYPE
-//> \verbatim
-//>          TYPE is CHARACTER*1
-//>          TYPE indices the storage type of the input matrix.
-//>          = 'G':  A is a full matrix.
-//>          = 'L':  A is a lower triangular matrix.
-//>          = 'U':  A is an upper triangular matrix.
-//>          = 'H':  A is an upper Hessenberg matrix.
-//>          = 'B':  A is a symmetric band matrix with lower bandwidth KL
-//>                  and upper bandwidth KU and with the only the lower
-//>                  half stored.
-//>          = 'Q':  A is a symmetric band matrix with lower bandwidth KL
-//>                  and upper bandwidth KU and with the only the upper
-//>                  half stored.
-//>          = 'Z':  A is a band matrix with lower bandwidth KL and upper
-//>                  bandwidth KU. See DGBTRF for storage details.
-//> \endverbatim
-//>
-//> \param[in] KL
-//> \verbatim
-//>          KL is INTEGER
-//>          The lower bandwidth of A.  Referenced only if TYPE = 'B',
-//>          'Q' or 'Z'.
-//> \endverbatim
-//>
-//> \param[in] KU
-//> \verbatim
-//>          KU is INTEGER
-//>          The upper bandwidth of A.  Referenced only if TYPE = 'B',
-//>          'Q' or 'Z'.
-//> \endverbatim
-//>
-//> \param[in] CFROM
-//> \verbatim
-//>          CFROM is DOUBLE PRECISION
-//> \endverbatim
-//>
-//> \param[in] CTO
-//> \verbatim
-//>          CTO is DOUBLE PRECISION
-//>
-//>          The matrix A is multiplied by CTO/CFROM. A(I,J) is computed
-//>          without over/underflow if the final result CTO*A(I,J)/CFROM
-//>          can be represented without over/underflow.  CFROM must be
-//>          nonzero.
-//> \endverbatim
-//>
-//> \param[in] M
-//> \verbatim
-//>          M is INTEGER
-//>          The number of rows of the matrix A.  M >= 0.
-//> \endverbatim
-//>
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>          The number of columns of the matrix A.  N >= 0.
-//> \endverbatim
-//>
-//> \param[in,out] A
-//> \verbatim
-//>          A is DOUBLE PRECISION array, dimension (LDA,N)
-//>          The matrix to be multiplied by CTO/CFROM.  See TYPE for the
-//>          storage type.
-//> \endverbatim
-//>
-//> \param[in] LDA
-//> \verbatim
-//>          LDA is INTEGER
-//>          The leading dimension of the array A.
-//>          If TYPE = 'G', 'L', 'U', 'H', LDA >= max(1,M);
-//>             TYPE = 'B', LDA >= KL+1;
-//>             TYPE = 'Q', LDA >= KU+1;
-//>             TYPE = 'Z', LDA >= 2*KL+KU+1.
-//> \endverbatim
-//>
-//> \param[out] INFO
-//> \verbatim
-//>          INFO is INTEGER
-//>          0  - successful exit
-//>          <0 - if INFO = -i, the i-th argument had an illegal value.
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date June 2016
-//
-//> \ingroup OTHERauxiliary
-//
-// =====================================================================
-/* Subroutine */ int dlascl_(char *type__, int *kl, int *ku, double *cfrom,
-	double *cto, int *m, int *n, double *a, int *lda, int *info)
-{
-    // System generated locals
-    int a_dim1, a_offset, i__1, i__2, i__3, i__4, i__5;
-
-    // Local variables
-    int i__, j, k1, k2, k3, k4;
-    double mul, cto1;
-    int done;
-    double ctoc;
-    extern int lsame_(char *, char *);
-    int itype;
-    double cfrom1;
-    extern double dlamch_(char *);
-    double cfromc;
-    extern int disnan_(double *);
-    extern /* Subroutine */ int xerbla_(char *, int *);
-    double bignum, smlnum;
-
-    //
-    // -- LAPACK auxiliary routine (version 3.7.0) --
-    // -- LAPACK is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    June 2016
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. Parameters ..
-    //    ..
-    //    .. Local Scalars ..
-    //    ..
-    //    .. External Functions ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    //    .. External Subroutines ..
-    //    ..
-    //    .. Executable Statements ..
-    //
-    //    Test the input arguments
-    //
-    // Parameter adjustments
-    a_dim1 = *lda;
-    a_offset = 1 + a_dim1;
-    a -= a_offset;
-
-    // Function Body
-    *info = 0;
-    if (lsame_(type__, "G")) {
-	itype = 0;
-    } else if (lsame_(type__, "L")) {
-	itype = 1;
-    } else if (lsame_(type__, "U")) {
-	itype = 2;
-    } else if (lsame_(type__, "H")) {
-	itype = 3;
-    } else if (lsame_(type__, "B")) {
-	itype = 4;
-    } else if (lsame_(type__, "Q")) {
-	itype = 5;
-    } else if (lsame_(type__, "Z")) {
-	itype = 6;
-    } else {
-	itype = -1;
-    }
-    if (itype == -1) {
-	*info = -1;
-    } else if (*cfrom == 0. || disnan_(cfrom)) {
-	*info = -4;
-    } else if (disnan_(cto)) {
-	*info = -5;
-    } else if (*m < 0) {
-	*info = -6;
-    } else if (*n < 0 || itype == 4 && *n != *m || itype == 5 && *n != *m) {
-	*info = -7;
-    } else if (itype <= 3 && *lda < max(1,*m)) {
-	*info = -9;
-    } else if (itype >= 4) {
-	// Computing MAX
-	i__1 = *m - 1;
-	if (*kl < 0 || *kl > max(i__1,0)) {
-	    *info = -2;
-	} else /* if(complicated condition) */ {
-	    // Computing MAX
-	    i__1 = *n - 1;
-	    if (*ku < 0 || *ku > max(i__1,0) || (itype == 4 || itype == 5) &&
-		    *kl != *ku) {
-		*info = -3;
-	    } else if (itype == 4 && *lda < *kl + 1 || itype == 5 && *lda < *
-		    ku + 1 || itype == 6 && *lda < (*kl << 1) + *ku + 1) {
-		*info = -9;
-	    }
-	}
-    }
-    if (*info != 0) {
-	i__1 = -(*info);
-	xerbla_("DLASCL", &i__1);
-	return 0;
-    }
-    //
-    //    Quick return if possible
-    //
-    if (*n == 0 || *m == 0) {
-	return 0;
-    }
-    //
-    //    Get machine parameters
-    //
-    smlnum = dlamch_("S");
-    bignum = 1. / smlnum;
-    cfromc = *cfrom;
-    ctoc = *cto;
-L10:
-    cfrom1 = cfromc * smlnum;
-    if (cfrom1 == cfromc) {
-	//       CFROMC is an inf.  Multiply by a correctly signed zero for
-	//       finite CTOC, or a NaN if CTOC is infinite.
-	mul = ctoc / cfromc;
-	done = TRUE_;
-	cto1 = ctoc;
-    } else {
-	cto1 = ctoc / bignum;
-	if (cto1 == ctoc) {
-	    //          CTOC is either 0 or an inf.  In both cases, CTOC itself
-	    //          serves as the correct multiplication factor.
-	    mul = ctoc;
-	    done = TRUE_;
-	    cfromc = 1.;
-	} else if (abs(cfrom1) > abs(ctoc) && ctoc != 0.) {
-	    mul = smlnum;
-	    done = FALSE_;
-	    cfromc = cfrom1;
-	} else if (abs(cto1) > abs(cfromc)) {
-	    mul = bignum;
-	    done = FALSE_;
-	    ctoc = cto1;
-	} else {
-	    mul = ctoc / cfromc;
-	    done = TRUE_;
-	}
-    }
-    if (itype == 0) {
-	//
-	//       Full matrix
-	//
-	i__1 = *n;
-	for (j = 1; j <= i__1; ++j) {
-	    i__2 = *m;
-	    for (i__ = 1; i__ <= i__2; ++i__) {
-		a[i__ + j * a_dim1] *= mul;
-// L20:
-	    }
-// L30:
-	}
-    } else if (itype == 1) {
-	//
-	//       Lower triangular matrix
-	//
-	i__1 = *n;
-	for (j = 1; j <= i__1; ++j) {
-	    i__2 = *m;
-	    for (i__ = j; i__ <= i__2; ++i__) {
-		a[i__ + j * a_dim1] *= mul;
-// L40:
-	    }
-// L50:
-	}
-    } else if (itype == 2) {
-	//
-	//       Upper triangular matrix
-	//
-	i__1 = *n;
-	for (j = 1; j <= i__1; ++j) {
-	    i__2 = min(j,*m);
-	    for (i__ = 1; i__ <= i__2; ++i__) {
-		a[i__ + j * a_dim1] *= mul;
-// L60:
-	    }
-// L70:
-	}
-    } else if (itype == 3) {
-	//
-	//       Upper Hessenberg matrix
-	//
-	i__1 = *n;
-	for (j = 1; j <= i__1; ++j) {
-	    // Computing MIN
-	    i__3 = j + 1;
-	    i__2 = min(i__3,*m);
-	    for (i__ = 1; i__ <= i__2; ++i__) {
-		a[i__ + j * a_dim1] *= mul;
-// L80:
-	    }
-// L90:
-	}
-    } else if (itype == 4) {
-	//
-	//       Lower half of a symmetric band matrix
-	//
-	k3 = *kl + 1;
-	k4 = *n + 1;
-	i__1 = *n;
-	for (j = 1; j <= i__1; ++j) {
-	    // Computing MIN
-	    i__3 = k3, i__4 = k4 - j;
-	    i__2 = min(i__3,i__4);
-	    for (i__ = 1; i__ <= i__2; ++i__) {
-		a[i__ + j * a_dim1] *= mul;
-// L100:
-	    }
-// L110:
-	}
-    } else if (itype == 5) {
-	//
-	//       Upper half of a symmetric band matrix
-	//
-	k1 = *ku + 2;
-	k3 = *ku + 1;
-	i__1 = *n;
-	for (j = 1; j <= i__1; ++j) {
-	    // Computing MAX
-	    i__2 = k1 - j;
-	    i__3 = k3;
-	    for (i__ = max(i__2,1); i__ <= i__3; ++i__) {
-		a[i__ + j * a_dim1] *= mul;
-// L120:
-	    }
-// L130:
-	}
-    } else if (itype == 6) {
-	//
-	//       Band matrix
-	//
-	k1 = *kl + *ku + 2;
-	k2 = *kl + 1;
-	k3 = (*kl << 1) + *ku + 1;
-	k4 = *kl + *ku + 1 + *m;
-	i__1 = *n;
-	for (j = 1; j <= i__1; ++j) {
-	    // Computing MAX
-	    i__3 = k1 - j;
-	    // Computing MIN
-	    i__4 = k3, i__5 = k4 - j;
-	    i__2 = min(i__4,i__5);
-	    for (i__ = max(i__3,k2); i__ <= i__2; ++i__) {
-		a[i__ + j * a_dim1] *= mul;
-// L140:
-	    }
-// L150:
-	}
-    }
-    if (! done) {
-	goto L10;
-    }
-    return 0;
-    //
-    //    End of DLASCL
-    //
-} // dlascl_
-
@@ -1,209 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b DLASET initializes the off-diagonal elements and the diagonal elements of a matrix to given values.
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-//> \htmlonly
-//> Download DLASET + dependencies
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dlaset.f">
-//> [TGZ]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dlaset.f">
-//> [ZIP]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dlaset.f">
-//> [TXT]</a>
-//> \endhtmlonly
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE DLASET( UPLO, M, N, ALPHA, BETA, A, LDA )
-//
-//      .. Scalar Arguments ..
-//      CHARACTER          UPLO
-//      INTEGER            LDA, M, N
-//      DOUBLE PRECISION   ALPHA, BETA
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION   A( LDA, * )
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> DLASET initializes an m-by-n matrix A to BETA on the diagonal and
-//> ALPHA on the offdiagonals.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] UPLO
-//> \verbatim
-//>          UPLO is CHARACTER*1
-//>          Specifies the part of the matrix A to be set.
-//>          = 'U':      Upper triangular part is set; the strictly lower
-//>                      triangular part of A is not changed.
-//>          = 'L':      Lower triangular part is set; the strictly upper
-//>                      triangular part of A is not changed.
-//>          Otherwise:  All of the matrix A is set.
-//> \endverbatim
-//>
-//> \param[in] M
-//> \verbatim
-//>          M is INTEGER
-//>          The number of rows of the matrix A.  M >= 0.
-//> \endverbatim
-//>
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>          The number of columns of the matrix A.  N >= 0.
-//> \endverbatim
-//>
-//> \param[in] ALPHA
-//> \verbatim
-//>          ALPHA is DOUBLE PRECISION
-//>          The constant to which the offdiagonal elements are to be set.
-//> \endverbatim
-//>
-//> \param[in] BETA
-//> \verbatim
-//>          BETA is DOUBLE PRECISION
-//>          The constant to which the diagonal elements are to be set.
-//> \endverbatim
-//>
-//> \param[out] A
-//> \verbatim
-//>          A is DOUBLE PRECISION array, dimension (LDA,N)
-//>          On exit, the leading m-by-n submatrix of A is set as follows:
-//>
-//>          if UPLO = 'U', A(i,j) = ALPHA, 1<=i<=j-1, 1<=j<=n,
-//>          if UPLO = 'L', A(i,j) = ALPHA, j+1<=i<=m, 1<=j<=n,
-//>          otherwise,     A(i,j) = ALPHA, 1<=i<=m, 1<=j<=n, i.ne.j,
-//>
-//>          and, for all UPLO, A(i,i) = BETA, 1<=i<=min(m,n).
-//> \endverbatim
-//>
-//> \param[in] LDA
-//> \verbatim
-//>          LDA is INTEGER
-//>          The leading dimension of the array A.  LDA >= max(1,M).
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date December 2016
-//
-//> \ingroup OTHERauxiliary
-//
-// =====================================================================
-/* Subroutine */ int dlaset_(char *uplo, int *m, int *n, double *alpha,
-	double *beta, double *a, int *lda)
-{
-    // System generated locals
-    int a_dim1, a_offset, i__1, i__2, i__3;
-
-    // Local variables
-    int i__, j;
-    extern int lsame_(char *, char *);
-
-    //
-    // -- LAPACK auxiliary routine (version 3.7.0) --
-    // -- LAPACK is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    December 2016
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    //=====================================================================
-    //
-    //    .. Local Scalars ..
-    //    ..
-    //    .. External Functions ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    //    .. Executable Statements ..
-    //
-    // Parameter adjustments
-    a_dim1 = *lda;
-    a_offset = 1 + a_dim1;
-    a -= a_offset;
-
-    // Function Body
-    if (lsame_(uplo, "U")) {
-	//
-	//       Set the strictly upper triangular or trapezoidal part of the
-	//       array to ALPHA.
-	//
-	i__1 = *n;
-	for (j = 2; j <= i__1; ++j) {
-	    // Computing MIN
-	    i__3 = j - 1;
-	    i__2 = min(i__3,*m);
-	    for (i__ = 1; i__ <= i__2; ++i__) {
-		a[i__ + j * a_dim1] = *alpha;
-// L10:
-	    }
-// L20:
-	}
-    } else if (lsame_(uplo, "L")) {
-	//
-	//       Set the strictly lower triangular or trapezoidal part of the
-	//       array to ALPHA.
-	//
-	i__1 = min(*m,*n);
-	for (j = 1; j <= i__1; ++j) {
-	    i__2 = *m;
-	    for (i__ = j + 1; i__ <= i__2; ++i__) {
-		a[i__ + j * a_dim1] = *alpha;
-// L30:
-	    }
-// L40:
-	}
-    } else {
-	//
-	//       Set the leading m-by-n submatrix to ALPHA.
-	//
-	i__1 = *n;
-	for (j = 1; j <= i__1; ++j) {
-	    i__2 = *m;
-	    for (i__ = 1; i__ <= i__2; ++i__) {
-		a[i__ + j * a_dim1] = *alpha;
-// L50:
-	    }
-// L60:
-	}
-    }
-    //
-    //    Set the first min(M,N) diagonal elements to BETA.
-    //
-    i__1 = min(*m,*n);
-    for (i__ = 1; i__ <= i__1; ++i__) {
-	a[i__ + i__ * a_dim1] = *beta;
-// L70:
-    }
-    return 0;
-    //
-    //    End of DLASET
-    //
-} // dlaset_
-
@@ -1,172 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b DLASSQ updates a sum of squares represented in scaled form.
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-//> \htmlonly
-//> Download DLASSQ + dependencies
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dlassq.f">
-//> [TGZ]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dlassq.f">
-//> [ZIP]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dlassq.f">
-//> [TXT]</a>
-//> \endhtmlonly
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE DLASSQ( N, X, INCX, SCALE, SUMSQ )
-//
-//      .. Scalar Arguments ..
-//      INTEGER            INCX, N
-//      DOUBLE PRECISION   SCALE, SUMSQ
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION   X( * )
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> DLASSQ  returns the values  scl  and  smsq  such that
-//>
-//>    ( scl**2 )*smsq = x( 1 )**2 +...+ x( n )**2 + ( scale**2 )*sumsq,
-//>
-//> where  x( i ) = X( 1 + ( i - 1 )*INCX ). The value of  sumsq  is
-//> assumed to be non-negative and  scl  returns the value
-//>
-//>    scl = max( scale, abs( x( i ) ) ).
-//>
-//> scale and sumsq must be supplied in SCALE and SUMSQ and
-//> scl and smsq are overwritten on SCALE and SUMSQ respectively.
-//>
-//> The routine makes only one pass through the vector x.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>          The number of elements to be used from the vector X.
-//> \endverbatim
-//>
-//> \param[in] X
-//> \verbatim
-//>          X is DOUBLE PRECISION array, dimension (1+(N-1)*INCX)
-//>          The vector for which a scaled sum of squares is computed.
-//>             x( i )  = X( 1 + ( i - 1 )*INCX ), 1 <= i <= n.
-//> \endverbatim
-//>
-//> \param[in] INCX
-//> \verbatim
-//>          INCX is INTEGER
-//>          The increment between successive values of the vector X.
-//>          INCX > 0.
-//> \endverbatim
-//>
-//> \param[in,out] SCALE
-//> \verbatim
-//>          SCALE is DOUBLE PRECISION
-//>          On entry, the value  scale  in the equation above.
-//>          On exit, SCALE is overwritten with  scl , the scaling factor
-//>          for the sum of squares.
-//> \endverbatim
-//>
-//> \param[in,out] SUMSQ
-//> \verbatim
-//>          SUMSQ is DOUBLE PRECISION
-//>          On entry, the value  sumsq  in the equation above.
-//>          On exit, SUMSQ is overwritten with  smsq , the basic sum of
-//>          squares from which  scl  has been factored out.
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date December 2016
-//
-//> \ingroup OTHERauxiliary
-//
-// =====================================================================
-/* Subroutine */ int dlassq_(int *n, double *x, int *incx, double *scale,
-	double *sumsq)
-{
-    // System generated locals
-    int i__1, i__2;
-    double d__1;
-
-    // Local variables
-    int ix;
-    double absxi;
-    extern int disnan_(double *);
-
-    //
-    // -- LAPACK auxiliary routine (version 3.7.0) --
-    // -- LAPACK is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    December 2016
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    //=====================================================================
-    //
-    //    .. Parameters ..
-    //    ..
-    //    .. Local Scalars ..
-    //    ..
-    //    .. External Functions ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    //    .. Executable Statements ..
-    //
-    // Parameter adjustments
-    --x;
-
-    // Function Body
-    if (*n > 0) {
-	i__1 = (*n - 1) * *incx + 1;
-	i__2 = *incx;
-	for (ix = 1; i__2 < 0 ? ix >= i__1 : ix <= i__1; ix += i__2) {
-	    absxi = (d__1 = x[ix], abs(d__1));
-	    if (absxi > 0. || disnan_(&absxi)) {
-		if (*scale < absxi) {
-		    // Computing 2nd power
-		    d__1 = *scale / absxi;
-		    *sumsq = *sumsq * (d__1 * d__1) + 1;
-		    *scale = absxi;
-		} else {
-		    // Computing 2nd power
-		    d__1 = absxi / *scale;
-		    *sumsq += d__1 * d__1;
-		}
-	    }
-// L10:
-	}
-    }
-    return 0;
-    //
-    //    End of DLASSQ
-    //
-} // dlassq_
-
@@ -1,149 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b DNRM2
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-// Definition:
-// ===========
-//
-//      DOUBLE PRECISION FUNCTION DNRM2(N,X,INCX)
-//
-//      .. Scalar Arguments ..
-//      INTEGER INCX,N
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION X(*)
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> DNRM2 returns the euclidean norm of a vector via the function
-//> name, so that
-//>
-//>    DNRM2 := sqrt( x'*x )
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>         number of elements in input vector(s)
-//> \endverbatim
-//>
-//> \param[in] X
-//> \verbatim
-//>          X is DOUBLE PRECISION array, dimension ( 1 + ( N - 1 )*abs( INCX ) )
-//> \endverbatim
-//>
-//> \param[in] INCX
-//> \verbatim
-//>          INCX is INTEGER
-//>         storage spacing between elements of DX
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date November 2017
-//
-//> \ingroup double_blas_level1
-//
-//> \par Further Details:
-// =====================
-//>
-//> \verbatim
-//>
-//>  -- This version written on 25-October-1982.
-//>     Modified on 14-October-1993 to inline the call to DLASSQ.
-//>     Sven Hammarling, Nag Ltd.
-//> \endverbatim
-//>
-// =====================================================================
-double dnrm2_(int *n, double *x, int *incx)
-{
-    // System generated locals
-    int i__1, i__2;
-    double ret_val, d__1;
-
-    // Local variables
-    int ix;
-    double ssq, norm, scale, absxi;
-
-    //
-    // -- Reference BLAS level1 routine (version 3.8.0) --
-    // -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    November 2017
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. Parameters ..
-    //    ..
-    //    .. Local Scalars ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    // Parameter adjustments
-    --x;
-
-    // Function Body
-    if (*n < 1 || *incx < 1) {
-	norm = 0.;
-    } else if (*n == 1) {
-	norm = abs(x[1]);
-    } else {
-	scale = 0.;
-	ssq = 1.;
-	//       The following loop is equivalent to this call to the LAPACK
-	//       auxiliary routine:
-	//       CALL DLASSQ( N, X, INCX, SCALE, SSQ )
-	//
-	i__1 = (*n - 1) * *incx + 1;
-	i__2 = *incx;
-	for (ix = 1; i__2 < 0 ? ix >= i__1 : ix <= i__1; ix += i__2) {
-	    if (x[ix] != 0.) {
-		absxi = (d__1 = x[ix], abs(d__1));
-		if (scale < absxi) {
-		    // Computing 2nd power
-		    d__1 = scale / absxi;
-		    ssq = ssq * (d__1 * d__1) + 1.;
-		    scale = absxi;
-		} else {
-		    // Computing 2nd power
-		    d__1 = absxi / scale;
-		    ssq += d__1 * d__1;
-		}
-	    }
-// L10:
-	}
-	norm = scale * sqrt(ssq);
-    }
-    ret_val = norm;
-    return ret_val;
-    //
-    //    End of DNRM2.
-    //
-} // dnrm2_
-
@@ -1,571 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b DORG2R generates all or part of the orthogonal matrix Q from a QR factorization determined by sgeqrf (unblocked algorithm).
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-//> \htmlonly
-//> Download DORG2R + dependencies
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dorg2r.f">
-//> [TGZ]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dorg2r.f">
-//> [ZIP]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dorg2r.f">
-//> [TXT]</a>
-//> \endhtmlonly
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE DORG2R( M, N, K, A, LDA, TAU, WORK, INFO )
-//
-//      .. Scalar Arguments ..
-//      INTEGER            INFO, K, LDA, M, N
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION   A( LDA, * ), TAU( * ), WORK( * )
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> DORG2R generates an m by n real matrix Q with orthonormal columns,
-//> which is defined as the first n columns of a product of k elementary
-//> reflectors of order m
-//>
-//>       Q  =  H(1) H(2) . . . H(k)
-//>
-//> as returned by DGEQRF.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] M
-//> \verbatim
-//>          M is INTEGER
-//>          The number of rows of the matrix Q. M >= 0.
-//> \endverbatim
-//>
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>          The number of columns of the matrix Q. M >= N >= 0.
-//> \endverbatim
-//>
-//> \param[in] K
-//> \verbatim
-//>          K is INTEGER
-//>          The number of elementary reflectors whose product defines the
-//>          matrix Q. N >= K >= 0.
-//> \endverbatim
-//>
-//> \param[in,out] A
-//> \verbatim
-//>          A is DOUBLE PRECISION array, dimension (LDA,N)
-//>          On entry, the i-th column must contain the vector which
-//>          defines the elementary reflector H(i), for i = 1,2,...,k, as
-//>          returned by DGEQRF in the first k columns of its array
-//>          argument A.
-//>          On exit, the m-by-n matrix Q.
-//> \endverbatim
-//>
-//> \param[in] LDA
-//> \verbatim
-//>          LDA is INTEGER
-//>          The first dimension of the array A. LDA >= max(1,M).
-//> \endverbatim
-//>
-//> \param[in] TAU
-//> \verbatim
-//>          TAU is DOUBLE PRECISION array, dimension (K)
-//>          TAU(i) must contain the scalar factor of the elementary
-//>          reflector H(i), as returned by DGEQRF.
-//> \endverbatim
-//>
-//> \param[out] WORK
-//> \verbatim
-//>          WORK is DOUBLE PRECISION array, dimension (N)
-//> \endverbatim
-//>
-//> \param[out] INFO
-//> \verbatim
-//>          INFO is INTEGER
-//>          = 0: successful exit
-//>          < 0: if INFO = -i, the i-th argument has an illegal value
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date December 2016
-//
-//> \ingroup doubleOTHERcomputational
-//
-// =====================================================================
-/* Subroutine */ int dorg2r_(int *m, int *n, int *k, double *a, int *lda,
-	double *tau, double *work, int *info)
-{
-    // Table of constant values
-    int c__1 = 1;
-
-    // System generated locals
-    int a_dim1, a_offset, i__1, i__2;
-    double d__1;
-
-    // Local variables
-    int i__, j, l;
-    extern /* Subroutine */ int dscal_(int *, double *, double *, int *),
-	    dlarf_(char *, int *, int *, double *, int *, double *, double *,
-	    int *, double *), xerbla_(char *, int *);
-
-    //
-    // -- LAPACK computational routine (version 3.7.0) --
-    // -- LAPACK is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    December 2016
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. Parameters ..
-    //    ..
-    //    .. Local Scalars ..
-    //    ..
-    //    .. External Subroutines ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    //    .. Executable Statements ..
-    //
-    //    Test the input arguments
-    //
-    // Parameter adjustments
-    a_dim1 = *lda;
-    a_offset = 1 + a_dim1;
-    a -= a_offset;
-    --tau;
-    --work;
-
-    // Function Body
-    *info = 0;
-    if (*m < 0) {
-	*info = -1;
-    } else if (*n < 0 || *n > *m) {
-	*info = -2;
-    } else if (*k < 0 || *k > *n) {
-	*info = -3;
-    } else if (*lda < max(1,*m)) {
-	*info = -5;
-    }
-    if (*info != 0) {
-	i__1 = -(*info);
-	xerbla_("DORG2R", &i__1);
-	return 0;
-    }
-    //
-    //    Quick return if possible
-    //
-    if (*n <= 0) {
-	return 0;
-    }
-    //
-    //    Initialise columns k+1:n to columns of the unit matrix
-    //
-    i__1 = *n;
-    for (j = *k + 1; j <= i__1; ++j) {
-	i__2 = *m;
-	for (l = 1; l <= i__2; ++l) {
-	    a[l + j * a_dim1] = 0.;
-// L10:
-	}
-	a[j + j * a_dim1] = 1.;
-// L20:
-    }
-    for (i__ = *k; i__ >= 1; --i__) {
-	//
-	//       Apply H(i) to A(i:m,i:n) from the left
-	//
-	if (i__ < *n) {
-	    a[i__ + i__ * a_dim1] = 1.;
-	    i__1 = *m - i__ + 1;
-	    i__2 = *n - i__;
-	    dlarf_("Left", &i__1, &i__2, &a[i__ + i__ * a_dim1], &c__1, &tau[
-		    i__], &a[i__ + (i__ + 1) * a_dim1], lda, &work[1]);
-	}
-	if (i__ < *m) {
-	    i__1 = *m - i__;
-	    d__1 = -tau[i__];
-	    dscal_(&i__1, &d__1, &a[i__ + 1 + i__ * a_dim1], &c__1);
-	}
-	a[i__ + i__ * a_dim1] = 1. - tau[i__];
-	//
-	//       Set A(1:i-1,i) to zero
-	//
-	i__1 = i__ - 1;
-	for (l = 1; l <= i__1; ++l) {
-	    a[l + i__ * a_dim1] = 0.;
-// L30:
-	}
-// L40:
-    }
-    return 0;
-    //
-    //    End of DORG2R
-    //
-} // dorg2r_
-
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-//> \brief \b DORGQR
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-//> \htmlonly
-//> Download DORGQR + dependencies
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dorgqr.f">
-//> [TGZ]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dorgqr.f">
-//> [ZIP]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dorgqr.f">
-//> [TXT]</a>
-//> \endhtmlonly
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE DORGQR( M, N, K, A, LDA, TAU, WORK, LWORK, INFO )
-//
-//      .. Scalar Arguments ..
-//      INTEGER            INFO, K, LDA, LWORK, M, N
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION   A( LDA, * ), TAU( * ), WORK( * )
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> DORGQR generates an M-by-N real matrix Q with orthonormal columns,
-//> which is defined as the first N columns of a product of K elementary
-//> reflectors of order M
-//>
-//>       Q  =  H(1) H(2) . . . H(k)
-//>
-//> as returned by DGEQRF.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] M
-//> \verbatim
-//>          M is INTEGER
-//>          The number of rows of the matrix Q. M >= 0.
-//> \endverbatim
-//>
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>          The number of columns of the matrix Q. M >= N >= 0.
-//> \endverbatim
-//>
-//> \param[in] K
-//> \verbatim
-//>          K is INTEGER
-//>          The number of elementary reflectors whose product defines the
-//>          matrix Q. N >= K >= 0.
-//> \endverbatim
-//>
-//> \param[in,out] A
-//> \verbatim
-//>          A is DOUBLE PRECISION array, dimension (LDA,N)
-//>          On entry, the i-th column must contain the vector which
-//>          defines the elementary reflector H(i), for i = 1,2,...,k, as
-//>          returned by DGEQRF in the first k columns of its array
-//>          argument A.
-//>          On exit, the M-by-N matrix Q.
-//> \endverbatim
-//>
-//> \param[in] LDA
-//> \verbatim
-//>          LDA is INTEGER
-//>          The first dimension of the array A. LDA >= max(1,M).
-//> \endverbatim
-//>
-//> \param[in] TAU
-//> \verbatim
-//>          TAU is DOUBLE PRECISION array, dimension (K)
-//>          TAU(i) must contain the scalar factor of the elementary
-//>          reflector H(i), as returned by DGEQRF.
-//> \endverbatim
-//>
-//> \param[out] WORK
-//> \verbatim
-//>          WORK is DOUBLE PRECISION array, dimension (MAX(1,LWORK))
-//>          On exit, if INFO = 0, WORK(1) returns the optimal LWORK.
-//> \endverbatim
-//>
-//> \param[in] LWORK
-//> \verbatim
-//>          LWORK is INTEGER
-//>          The dimension of the array WORK. LWORK >= max(1,N).
-//>          For optimum performance LWORK >= N*NB, where NB is the
-//>          optimal blocksize.
-//>
-//>          If LWORK = -1, then a workspace query is assumed; the routine
-//>          only calculates the optimal size of the WORK array, returns
-//>          this value as the first entry of the WORK array, and no error
-//>          message related to LWORK is issued by XERBLA.
-//> \endverbatim
-//>
-//> \param[out] INFO
-//> \verbatim
-//>          INFO is INTEGER
-//>          = 0:  successful exit
-//>          < 0:  if INFO = -i, the i-th argument has an illegal value
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date December 2016
-//
-//> \ingroup doubleOTHERcomputational
-//
-// =====================================================================
-/* Subroutine */ int dorgqr_(int *m, int *n, int *k, double *a, int *lda,
-	double *tau, double *work, int *lwork, int *info)
-{
-    // Table of constant values
-    int c__1 = 1;
-    int c_n1 = -1;
-    int c__3 = 3;
-    int c__2 = 2;
-
-    // System generated locals
-    int a_dim1, a_offset, i__1, i__2, i__3;
-
-    // Local variables
-    int i__, j, l, ib, nb, ki, kk, nx, iws, nbmin, iinfo;
-    extern /* Subroutine */ int dorg2r_(int *, int *, int *, double *, int *,
-	    double *, double *, int *), dlarfb_(char *, char *, char *, char *
-	    , int *, int *, int *, double *, int *, double *, int *, double *,
-	     int *, double *, int *), dlarft_(char *, char *, int *, int *,
-	    double *, int *, double *, double *, int *), xerbla_(char *, int *
-	    );
-    extern int ilaenv_(int *, char *, char *, int *, int *, int *, int *);
-    int ldwork, lwkopt;
-    int lquery;
-
-    //
-    // -- LAPACK computational routine (version 3.7.0) --
-    // -- LAPACK is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    December 2016
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. Parameters ..
-    //    ..
-    //    .. Local Scalars ..
-    //    ..
-    //    .. External Subroutines ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    //    .. External Functions ..
-    //    ..
-    //    .. Executable Statements ..
-    //
-    //    Test the input arguments
-    //
-    // Parameter adjustments
-    a_dim1 = *lda;
-    a_offset = 1 + a_dim1;
-    a -= a_offset;
-    --tau;
-    --work;
-
-    // Function Body
-    *info = 0;
-    nb = ilaenv_(&c__1, "DORGQR", " ", m, n, k, &c_n1);
-    lwkopt = max(1,*n) * nb;
-    work[1] = (double) lwkopt;
-    lquery = *lwork == -1;
-    if (*m < 0) {
-	*info = -1;
-    } else if (*n < 0 || *n > *m) {
-	*info = -2;
-    } else if (*k < 0 || *k > *n) {
-	*info = -3;
-    } else if (*lda < max(1,*m)) {
-	*info = -5;
-    } else if (*lwork < max(1,*n) && ! lquery) {
-	*info = -8;
-    }
-    if (*info != 0) {
-	i__1 = -(*info);
-	xerbla_("DORGQR", &i__1);
-	return 0;
-    } else if (lquery) {
-	return 0;
-    }
-    //
-    //    Quick return if possible
-    //
-    if (*n <= 0) {
-	work[1] = 1.;
-	return 0;
-    }
-    nbmin = 2;
-    nx = 0;
-    iws = *n;
-    if (nb > 1 && nb < *k) {
-	//
-	//       Determine when to cross over from blocked to unblocked code.
-	//
-	// Computing MAX
-	i__1 = 0, i__2 = ilaenv_(&c__3, "DORGQR", " ", m, n, k, &c_n1);
-	nx = max(i__1,i__2);
-	if (nx < *k) {
-	    //
-	    //          Determine if workspace is large enough for blocked code.
-	    //
-	    ldwork = *n;
-	    iws = ldwork * nb;
-	    if (*lwork < iws) {
-		//
-		//             Not enough workspace to use optimal NB:  reduce NB and
-		//             determine the minimum value of NB.
-		//
-		nb = *lwork / ldwork;
-		// Computing MAX
-		i__1 = 2, i__2 = ilaenv_(&c__2, "DORGQR", " ", m, n, k, &c_n1)
-			;
-		nbmin = max(i__1,i__2);
-	    }
-	}
-    }
-    if (nb >= nbmin && nb < *k && nx < *k) {
-	//
-	//       Use blocked code after the last block.
-	//       The first kk columns are handled by the block method.
-	//
-	ki = (*k - nx - 1) / nb * nb;
-	// Computing MIN
-	i__1 = *k, i__2 = ki + nb;
-	kk = min(i__1,i__2);
-	//
-	//       Set A(1:kk,kk+1:n) to zero.
-	//
-	i__1 = *n;
-	for (j = kk + 1; j <= i__1; ++j) {
-	    i__2 = kk;
-	    for (i__ = 1; i__ <= i__2; ++i__) {
-		a[i__ + j * a_dim1] = 0.;
-// L10:
-	    }
-// L20:
-	}
-    } else {
-	kk = 0;
-    }
-    //
-    //    Use unblocked code for the last or only block.
-    //
-    if (kk < *n) {
-	i__1 = *m - kk;
-	i__2 = *n - kk;
-	i__3 = *k - kk;
-	dorg2r_(&i__1, &i__2, &i__3, &a[kk + 1 + (kk + 1) * a_dim1], lda, &
-		tau[kk + 1], &work[1], &iinfo);
-    }
-    if (kk > 0) {
-	//
-	//       Use blocked code
-	//
-	i__1 = -nb;
-	for (i__ = ki + 1; i__1 < 0 ? i__ >= 1 : i__ <= 1; i__ += i__1) {
-	    // Computing MIN
-	    i__2 = nb, i__3 = *k - i__ + 1;
-	    ib = min(i__2,i__3);
-	    if (i__ + ib <= *n) {
-		//
-		//             Form the triangular factor of the block reflector
-		//             H = H(i) H(i+1) . . . H(i+ib-1)
-		//
-		i__2 = *m - i__ + 1;
-		dlarft_("Forward", "Columnwise", &i__2, &ib, &a[i__ + i__ *
-			a_dim1], lda, &tau[i__], &work[1], &ldwork);
-		//
-		//             Apply H to A(i:m,i+ib:n) from the left
-		//
-		i__2 = *m - i__ + 1;
-		i__3 = *n - i__ - ib + 1;
-		dlarfb_("Left", "No transpose", "Forward", "Columnwise", &
-			i__2, &i__3, &ib, &a[i__ + i__ * a_dim1], lda, &work[
-			1], &ldwork, &a[i__ + (i__ + ib) * a_dim1], lda, &
-			work[ib + 1], &ldwork);
-	    }
-	    //
-	    //          Apply H to rows i:m of current block
-	    //
-	    i__2 = *m - i__ + 1;
-	    dorg2r_(&i__2, &ib, &ib, &a[i__ + i__ * a_dim1], lda, &tau[i__], &
-		    work[1], &iinfo);
-	    //
-	    //          Set rows 1:i-1 of current block to zero
-	    //
-	    i__2 = i__ + ib - 1;
-	    for (j = i__; j <= i__2; ++j) {
-		i__3 = i__ - 1;
-		for (l = 1; l <= i__3; ++l) {
-		    a[l + j * a_dim1] = 0.;
-// L30:
-		}
-// L40:
-	    }
-// L50:
-	}
-    }
-    work[1] = (double) iws;
-    return 0;
-    //
-    //    End of DORGQR
-    //
-} // dorgqr_
-
@@ -1,684 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b DORM2R multiplies a general matrix by the orthogonal matrix from a QR factorization determined by sgeqrf (unblocked algorithm).
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-//> \htmlonly
-//> Download DORM2R + dependencies
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dorm2r.f">
-//> [TGZ]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dorm2r.f">
-//> [ZIP]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dorm2r.f">
-//> [TXT]</a>
-//> \endhtmlonly
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE DORM2R( SIDE, TRANS, M, N, K, A, LDA, TAU, C, LDC,
-//                         WORK, INFO )
-//
-//      .. Scalar Arguments ..
-//      CHARACTER          SIDE, TRANS
-//      INTEGER            INFO, K, LDA, LDC, M, N
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION   A( LDA, * ), C( LDC, * ), TAU( * ), WORK( * )
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> DORM2R overwrites the general real m by n matrix C with
-//>
-//>       Q * C  if SIDE = 'L' and TRANS = 'N', or
-//>
-//>       Q**T* C  if SIDE = 'L' and TRANS = 'T', or
-//>
-//>       C * Q  if SIDE = 'R' and TRANS = 'N', or
-//>
-//>       C * Q**T if SIDE = 'R' and TRANS = 'T',
-//>
-//> where Q is a real orthogonal matrix defined as the product of k
-//> elementary reflectors
-//>
-//>       Q = H(1) H(2) . . . H(k)
-//>
-//> as returned by DGEQRF. Q is of order m if SIDE = 'L' and of order n
-//> if SIDE = 'R'.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] SIDE
-//> \verbatim
-//>          SIDE is CHARACTER*1
-//>          = 'L': apply Q or Q**T from the Left
-//>          = 'R': apply Q or Q**T from the Right
-//> \endverbatim
-//>
-//> \param[in] TRANS
-//> \verbatim
-//>          TRANS is CHARACTER*1
-//>          = 'N': apply Q  (No transpose)
-//>          = 'T': apply Q**T (Transpose)
-//> \endverbatim
-//>
-//> \param[in] M
-//> \verbatim
-//>          M is INTEGER
-//>          The number of rows of the matrix C. M >= 0.
-//> \endverbatim
-//>
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>          The number of columns of the matrix C. N >= 0.
-//> \endverbatim
-//>
-//> \param[in] K
-//> \verbatim
-//>          K is INTEGER
-//>          The number of elementary reflectors whose product defines
-//>          the matrix Q.
-//>          If SIDE = 'L', M >= K >= 0;
-//>          if SIDE = 'R', N >= K >= 0.
-//> \endverbatim
-//>
-//> \param[in] A
-//> \verbatim
-//>          A is DOUBLE PRECISION array, dimension (LDA,K)
-//>          The i-th column must contain the vector which defines the
-//>          elementary reflector H(i), for i = 1,2,...,k, as returned by
-//>          DGEQRF in the first k columns of its array argument A.
-//>          A is modified by the routine but restored on exit.
-//> \endverbatim
-//>
-//> \param[in] LDA
-//> \verbatim
-//>          LDA is INTEGER
-//>          The leading dimension of the array A.
-//>          If SIDE = 'L', LDA >= max(1,M);
-//>          if SIDE = 'R', LDA >= max(1,N).
-//> \endverbatim
-//>
-//> \param[in] TAU
-//> \verbatim
-//>          TAU is DOUBLE PRECISION array, dimension (K)
-//>          TAU(i) must contain the scalar factor of the elementary
-//>          reflector H(i), as returned by DGEQRF.
-//> \endverbatim
-//>
-//> \param[in,out] C
-//> \verbatim
-//>          C is DOUBLE PRECISION array, dimension (LDC,N)
-//>          On entry, the m by n matrix C.
-//>          On exit, C is overwritten by Q*C or Q**T*C or C*Q**T or C*Q.
-//> \endverbatim
-//>
-//> \param[in] LDC
-//> \verbatim
-//>          LDC is INTEGER
-//>          The leading dimension of the array C. LDC >= max(1,M).
-//> \endverbatim
-//>
-//> \param[out] WORK
-//> \verbatim
-//>          WORK is DOUBLE PRECISION array, dimension
-//>                                   (N) if SIDE = 'L',
-//>                                   (M) if SIDE = 'R'
-//> \endverbatim
-//>
-//> \param[out] INFO
-//> \verbatim
-//>          INFO is INTEGER
-//>          = 0: successful exit
-//>          < 0: if INFO = -i, the i-th argument had an illegal value
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date December 2016
-//
-//> \ingroup doubleOTHERcomputational
-//
-// =====================================================================
-/* Subroutine */ int dorm2r_(char *side, char *trans, int *m, int *n, int *k,
-	double *a, int *lda, double *tau, double *c__, int *ldc, double *work,
-	 int *info)
-{
-    // Table of constant values
-    int c__1 = 1;
-
-    // System generated locals
-    int a_dim1, a_offset, c_dim1, c_offset, i__1, i__2;
-
-    // Local variables
-    int i__, i1, i2, i3, ic, jc, mi, ni, nq;
-    double aii;
-    int left;
-    extern /* Subroutine */ int dlarf_(char *, int *, int *, double *, int *,
-	    double *, double *, int *, double *);
-    extern int lsame_(char *, char *);
-    extern /* Subroutine */ int xerbla_(char *, int *);
-    int notran;
-
-    //
-    // -- LAPACK computational routine (version 3.7.0) --
-    // -- LAPACK is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    December 2016
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. Parameters ..
-    //    ..
-    //    .. Local Scalars ..
-    //    ..
-    //    .. External Functions ..
-    //    ..
-    //    .. External Subroutines ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    //    .. Executable Statements ..
-    //
-    //    Test the input arguments
-    //
-    // Parameter adjustments
-    a_dim1 = *lda;
-    a_offset = 1 + a_dim1;
-    a -= a_offset;
-    --tau;
-    c_dim1 = *ldc;
-    c_offset = 1 + c_dim1;
-    c__ -= c_offset;
-    --work;
-
-    // Function Body
-    *info = 0;
-    left = lsame_(side, "L");
-    notran = lsame_(trans, "N");
-    //
-    //    NQ is the order of Q
-    //
-    if (left) {
-	nq = *m;
-    } else {
-	nq = *n;
-    }
-    if (! left && ! lsame_(side, "R")) {
-	*info = -1;
-    } else if (! notran && ! lsame_(trans, "T")) {
-	*info = -2;
-    } else if (*m < 0) {
-	*info = -3;
-    } else if (*n < 0) {
-	*info = -4;
-    } else if (*k < 0 || *k > nq) {
-	*info = -5;
-    } else if (*lda < max(1,nq)) {
-	*info = -7;
-    } else if (*ldc < max(1,*m)) {
-	*info = -10;
-    }
-    if (*info != 0) {
-	i__1 = -(*info);
-	xerbla_("DORM2R", &i__1);
-	return 0;
-    }
-    //
-    //    Quick return if possible
-    //
-    if (*m == 0 || *n == 0 || *k == 0) {
-	return 0;
-    }
-    if (left && ! notran || ! left && notran) {
-	i1 = 1;
-	i2 = *k;
-	i3 = 1;
-    } else {
-	i1 = *k;
-	i2 = 1;
-	i3 = -1;
-    }
-    if (left) {
-	ni = *n;
-	jc = 1;
-    } else {
-	mi = *m;
-	ic = 1;
-    }
-    i__1 = i2;
-    i__2 = i3;
-    for (i__ = i1; i__2 < 0 ? i__ >= i__1 : i__ <= i__1; i__ += i__2) {
-	if (left) {
-	    //
-	    //          H(i) is applied to C(i:m,1:n)
-	    //
-	    mi = *m - i__ + 1;
-	    ic = i__;
-	} else {
-	    //
-	    //          H(i) is applied to C(1:m,i:n)
-	    //
-	    ni = *n - i__ + 1;
-	    jc = i__;
-	}
-	//
-	//       Apply H(i)
-	//
-	aii = a[i__ + i__ * a_dim1];
-	a[i__ + i__ * a_dim1] = 1.;
-	dlarf_(side, &mi, &ni, &a[i__ + i__ * a_dim1], &c__1, &tau[i__], &c__[
-		ic + jc * c_dim1], ldc, &work[1]);
-	a[i__ + i__ * a_dim1] = aii;
-// L10:
-    }
-    return 0;
-    //
-    //    End of DORM2R
-    //
-} // dorm2r_
-
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-//> \brief \b DORMQR
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-//> \htmlonly
-//> Download DORMQR + dependencies
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dormqr.f">
-//> [TGZ]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dormqr.f">
-//> [ZIP]</a>
-//> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dormqr.f">
-//> [TXT]</a>
-//> \endhtmlonly
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE DORMQR( SIDE, TRANS, M, N, K, A, LDA, TAU, C, LDC,
-//                         WORK, LWORK, INFO )
-//
-//      .. Scalar Arguments ..
-//      CHARACTER          SIDE, TRANS
-//      INTEGER            INFO, K, LDA, LDC, LWORK, M, N
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION   A( LDA, * ), C( LDC, * ), TAU( * ), WORK( * )
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> DORMQR overwrites the general real M-by-N matrix C with
-//>
-//>                 SIDE = 'L'     SIDE = 'R'
-//> TRANS = 'N':      Q * C          C * Q
-//> TRANS = 'T':      Q**T * C       C * Q**T
-//>
-//> where Q is a real orthogonal matrix defined as the product of k
-//> elementary reflectors
-//>
-//>       Q = H(1) H(2) . . . H(k)
-//>
-//> as returned by DGEQRF. Q is of order M if SIDE = 'L' and of order N
-//> if SIDE = 'R'.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] SIDE
-//> \verbatim
-//>          SIDE is CHARACTER*1
-//>          = 'L': apply Q or Q**T from the Left;
-//>          = 'R': apply Q or Q**T from the Right.
-//> \endverbatim
-//>
-//> \param[in] TRANS
-//> \verbatim
-//>          TRANS is CHARACTER*1
-//>          = 'N':  No transpose, apply Q;
-//>          = 'T':  Transpose, apply Q**T.
-//> \endverbatim
-//>
-//> \param[in] M
-//> \verbatim
-//>          M is INTEGER
-//>          The number of rows of the matrix C. M >= 0.
-//> \endverbatim
-//>
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>          The number of columns of the matrix C. N >= 0.
-//> \endverbatim
-//>
-//> \param[in] K
-//> \verbatim
-//>          K is INTEGER
-//>          The number of elementary reflectors whose product defines
-//>          the matrix Q.
-//>          If SIDE = 'L', M >= K >= 0;
-//>          if SIDE = 'R', N >= K >= 0.
-//> \endverbatim
-//>
-//> \param[in] A
-//> \verbatim
-//>          A is DOUBLE PRECISION array, dimension (LDA,K)
-//>          The i-th column must contain the vector which defines the
-//>          elementary reflector H(i), for i = 1,2,...,k, as returned by
-//>          DGEQRF in the first k columns of its array argument A.
-//> \endverbatim
-//>
-//> \param[in] LDA
-//> \verbatim
-//>          LDA is INTEGER
-//>          The leading dimension of the array A.
-//>          If SIDE = 'L', LDA >= max(1,M);
-//>          if SIDE = 'R', LDA >= max(1,N).
-//> \endverbatim
-//>
-//> \param[in] TAU
-//> \verbatim
-//>          TAU is DOUBLE PRECISION array, dimension (K)
-//>          TAU(i) must contain the scalar factor of the elementary
-//>          reflector H(i), as returned by DGEQRF.
-//> \endverbatim
-//>
-//> \param[in,out] C
-//> \verbatim
-//>          C is DOUBLE PRECISION array, dimension (LDC,N)
-//>          On entry, the M-by-N matrix C.
-//>          On exit, C is overwritten by Q*C or Q**T*C or C*Q**T or C*Q.
-//> \endverbatim
-//>
-//> \param[in] LDC
-//> \verbatim
-//>          LDC is INTEGER
-//>          The leading dimension of the array C. LDC >= max(1,M).
-//> \endverbatim
-//>
-//> \param[out] WORK
-//> \verbatim
-//>          WORK is DOUBLE PRECISION array, dimension (MAX(1,LWORK))
-//>          On exit, if INFO = 0, WORK(1) returns the optimal LWORK.
-//> \endverbatim
-//>
-//> \param[in] LWORK
-//> \verbatim
-//>          LWORK is INTEGER
-//>          The dimension of the array WORK.
-//>          If SIDE = 'L', LWORK >= max(1,N);
-//>          if SIDE = 'R', LWORK >= max(1,M).
-//>          For good performance, LWORK should generally be larger.
-//>
-//>          If LWORK = -1, then a workspace query is assumed; the routine
-//>          only calculates the optimal size of the WORK array, returns
-//>          this value as the first entry of the WORK array, and no error
-//>          message related to LWORK is issued by XERBLA.
-//> \endverbatim
-//>
-//> \param[out] INFO
-//> \verbatim
-//>          INFO is INTEGER
-//>          = 0:  successful exit
-//>          < 0:  if INFO = -i, the i-th argument had an illegal value
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date December 2016
-//
-//> \ingroup doubleOTHERcomputational
-//
-// =====================================================================
-/* Subroutine */ int dormqr_(char *side, char *trans, int *m, int *n, int *k,
-	double *a, int *lda, double *tau, double *c__, int *ldc, double *work,
-	 int *lwork, int *info)
-{
-    // Table of constant values
-    int c__1 = 1;
-    int c_n1 = -1;
-    int c__2 = 2;
-    int c__65 = 65;
-
-    // System generated locals
-    address a__1[2];
-    int a_dim1, a_offset, c_dim1, c_offset, i__1, i__2, i__3[2], i__4, i__5;
-    char ch__1[2+1]={'\0'};
-
-    // Local variables
-    int i__, i1, i2, i3, ib, ic, jc, nb, mi, ni, nq, nw, iwt;
-    int left;
-    extern int lsame_(char *, char *);
-    int nbmin, iinfo;
-    extern /* Subroutine */ int dorm2r_(char *, char *, int *, int *, int *,
-	    double *, int *, double *, double *, int *, double *, int *),
-	    dlarfb_(char *, char *, char *, char *, int *, int *, int *,
-	    double *, int *, double *, int *, double *, int *, double *, int *
-	    ), dlarft_(char *, char *, int *, int *, double *, int *, double *
-	    , double *, int *), xerbla_(char *, int *);
-    extern int ilaenv_(int *, char *, char *, int *, int *, int *, int *);
-    int notran;
-    int ldwork, lwkopt;
-    int lquery;
-
-    //
-    // -- LAPACK computational routine (version 3.7.0) --
-    // -- LAPACK is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    December 2016
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. Parameters ..
-    //    ..
-    //    .. Local Scalars ..
-    //    ..
-    //    .. External Functions ..
-    //    ..
-    //    .. External Subroutines ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    //    .. Executable Statements ..
-    //
-    //    Test the input arguments
-    //
-    // Parameter adjustments
-    a_dim1 = *lda;
-    a_offset = 1 + a_dim1;
-    a -= a_offset;
-    --tau;
-    c_dim1 = *ldc;
-    c_offset = 1 + c_dim1;
-    c__ -= c_offset;
-    --work;
-
-    // Function Body
-    *info = 0;
-    left = lsame_(side, "L");
-    notran = lsame_(trans, "N");
-    lquery = *lwork == -1;
-    //
-    //    NQ is the order of Q and NW is the minimum dimension of WORK
-    //
-    if (left) {
-	nq = *m;
-	nw = *n;
-    } else {
-	nq = *n;
-	nw = *m;
-    }
-    if (! left && ! lsame_(side, "R")) {
-	*info = -1;
-    } else if (! notran && ! lsame_(trans, "T")) {
-	*info = -2;
-    } else if (*m < 0) {
-	*info = -3;
-    } else if (*n < 0) {
-	*info = -4;
-    } else if (*k < 0 || *k > nq) {
-	*info = -5;
-    } else if (*lda < max(1,nq)) {
-	*info = -7;
-    } else if (*ldc < max(1,*m)) {
-	*info = -10;
-    } else if (*lwork < max(1,nw) && ! lquery) {
-	*info = -12;
-    }
-    if (*info == 0) {
-	//
-	//       Compute the workspace requirements
-	//
-	// Computing MIN
-	// Writing concatenation
-	i__3[0] = 1, a__1[0] = side;
-	i__3[1] = 1, a__1[1] = trans;
-	s_cat(ch__1, a__1, i__3, &c__2);
-	i__1 = 64, i__2 = ilaenv_(&c__1, "DORMQR", ch__1, m, n, k, &c_n1);
-	nb = min(i__1,i__2);
-	lwkopt = max(1,nw) * nb + 4160;
-	work[1] = (double) lwkopt;
-    }
-    if (*info != 0) {
-	i__1 = -(*info);
-	xerbla_("DORMQR", &i__1);
-	return 0;
-    } else if (lquery) {
-	return 0;
-    }
-    //
-    //    Quick return if possible
-    //
-    if (*m == 0 || *n == 0 || *k == 0) {
-	work[1] = 1.;
-	return 0;
-    }
-    nbmin = 2;
-    ldwork = nw;
-    if (nb > 1 && nb < *k) {
-	if (*lwork < nw * nb + 4160) {
-	    nb = (*lwork - 4160) / ldwork;
-	    // Computing MAX
-	    // Writing concatenation
-	    i__3[0] = 1, a__1[0] = side;
-	    i__3[1] = 1, a__1[1] = trans;
-	    s_cat(ch__1, a__1, i__3, &c__2);
-	    i__1 = 2, i__2 = ilaenv_(&c__2, "DORMQR", ch__1, m, n, k, &c_n1);
-	    nbmin = max(i__1,i__2);
-	}
-    }
-    if (nb < nbmin || nb >= *k) {
-	//
-	//       Use unblocked code
-	//
-	dorm2r_(side, trans, m, n, k, &a[a_offset], lda, &tau[1], &c__[
-		c_offset], ldc, &work[1], &iinfo);
-    } else {
-	//
-	//       Use blocked code
-	//
-	iwt = nw * nb + 1;
-	if (left && ! notran || ! left && notran) {
-	    i1 = 1;
-	    i2 = *k;
-	    i3 = nb;
-	} else {
-	    i1 = (*k - 1) / nb * nb + 1;
-	    i2 = 1;
-	    i3 = -nb;
-	}
-	if (left) {
-	    ni = *n;
-	    jc = 1;
-	} else {
-	    mi = *m;
-	    ic = 1;
-	}
-	i__1 = i2;
-	i__2 = i3;
-	for (i__ = i1; i__2 < 0 ? i__ >= i__1 : i__ <= i__1; i__ += i__2) {
-	    // Computing MIN
-	    i__4 = nb, i__5 = *k - i__ + 1;
-	    ib = min(i__4,i__5);
-	    //
-	    //          Form the triangular factor of the block reflector
-	    //          H = H(i) H(i+1) . . . H(i+ib-1)
-	    //
-	    i__4 = nq - i__ + 1;
-	    dlarft_("Forward", "Columnwise", &i__4, &ib, &a[i__ + i__ *
-		    a_dim1], lda, &tau[i__], &work[iwt], &c__65);
-	    if (left) {
-		//
-		//             H or H**T is applied to C(i:m,1:n)
-		//
-		mi = *m - i__ + 1;
-		ic = i__;
-	    } else {
-		//
-		//             H or H**T is applied to C(1:m,i:n)
-		//
-		ni = *n - i__ + 1;
-		jc = i__;
-	    }
-	    //
-	    //          Apply H or H**T
-	    //
-	    dlarfb_(side, trans, "Forward", "Columnwise", &mi, &ni, &ib, &a[
-		    i__ + i__ * a_dim1], lda, &work[iwt], &c__65, &c__[ic +
-		    jc * c_dim1], ldc, &work[1], &ldwork);
-// L10:
-	}
-    }
-    work[1] = (double) lwkopt;
-    return 0;
-    //
-    //    End of DORMQR
-    //
-} // dormqr_
-
@@ -1,164 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b DROT
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE DROT(N,DX,INCX,DY,INCY,C,S)
-//
-//      .. Scalar Arguments ..
-//      DOUBLE PRECISION C,S
-//      INTEGER INCX,INCY,N
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION DX(*),DY(*)
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//>    DROT applies a plane rotation.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>         number of elements in input vector(s)
-//> \endverbatim
-//>
-//> \param[in,out] DX
-//> \verbatim
-//>          DX is DOUBLE PRECISION array, dimension ( 1 + ( N - 1 )*abs( INCX ) )
-//> \endverbatim
-//>
-//> \param[in] INCX
-//> \verbatim
-//>          INCX is INTEGER
-//>         storage spacing between elements of DX
-//> \endverbatim
-//>
-//> \param[in,out] DY
-//> \verbatim
-//>          DY is DOUBLE PRECISION array, dimension ( 1 + ( N - 1 )*abs( INCY ) )
-//> \endverbatim
-//>
-//> \param[in] INCY
-//> \verbatim
-//>          INCY is INTEGER
-//>         storage spacing between elements of DY
-//> \endverbatim
-//>
-//> \param[in] C
-//> \verbatim
-//>          C is DOUBLE PRECISION
-//> \endverbatim
-//>
-//> \param[in] S
-//> \verbatim
-//>          S is DOUBLE PRECISION
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date November 2017
-//
-//> \ingroup double_blas_level1
-//
-//> \par Further Details:
-// =====================
-//>
-//> \verbatim
-//>
-//>     jack dongarra, linpack, 3/11/78.
-//>     modified 12/3/93, array(1) declarations changed to array(*)
-//> \endverbatim
-//>
-// =====================================================================
-/* Subroutine */ int drot_(int *n, double *dx, int *incx, double *dy, int *
-	incy, double *c__, double *s)
-{
-    // System generated locals
-    int i__1;
-
-    // Local variables
-    int i__, ix, iy;
-    double dtemp;
-
-    //
-    // -- Reference BLAS level1 routine (version 3.8.0) --
-    // -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    November 2017
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. Local Scalars ..
-    //    ..
-    // Parameter adjustments
-    --dy;
-    --dx;
-
-    // Function Body
-    if (*n <= 0) {
-	return 0;
-    }
-    if (*incx == 1 && *incy == 1) {
-	//
-	//      code for both increments equal to 1
-	//
-	i__1 = *n;
-	for (i__ = 1; i__ <= i__1; ++i__) {
-	    dtemp = *c__ * dx[i__] + *s * dy[i__];
-	    dy[i__] = *c__ * dy[i__] - *s * dx[i__];
-	    dx[i__] = dtemp;
-	}
-    } else {
-	//
-	//      code for unequal increments or equal increments not equal
-	//        to 1
-	//
-	ix = 1;
-	iy = 1;
-	if (*incx < 0) {
-	    ix = (-(*n) + 1) * *incx + 1;
-	}
-	if (*incy < 0) {
-	    iy = (-(*n) + 1) * *incy + 1;
-	}
-	i__1 = *n;
-	for (i__ = 1; i__ <= i__1; ++i__) {
-	    dtemp = *c__ * dx[ix] + *s * dy[iy];
-	    dy[iy] = *c__ * dy[iy] - *s * dx[ix];
-	    dx[ix] = dtemp;
-	    ix += *incx;
-	    iy += *incy;
-	}
-    }
-    return 0;
-} // drot_
-
@@ -1,155 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b DSCAL
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE DSCAL(N,DA,DX,INCX)
-//
-//      .. Scalar Arguments ..
-//      DOUBLE PRECISION DA
-//      INTEGER INCX,N
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION DX(*)
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//>    DSCAL scales a vector by a constant.
-//>    uses unrolled loops for increment equal to 1.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>         number of elements in input vector(s)
-//> \endverbatim
-//>
-//> \param[in] DA
-//> \verbatim
-//>          DA is DOUBLE PRECISION
-//>           On entry, DA specifies the scalar alpha.
-//> \endverbatim
-//>
-//> \param[in,out] DX
-//> \verbatim
-//>          DX is DOUBLE PRECISION array, dimension ( 1 + ( N - 1 )*abs( INCX ) )
-//> \endverbatim
-//>
-//> \param[in] INCX
-//> \verbatim
-//>          INCX is INTEGER
-//>         storage spacing between elements of DX
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date November 2017
-//
-//> \ingroup double_blas_level1
-//
-//> \par Further Details:
-// =====================
-//>
-//> \verbatim
-//>
-//>     jack dongarra, linpack, 3/11/78.
-//>     modified 3/93 to return if incx .le. 0.
-//>     modified 12/3/93, array(1) declarations changed to array(*)
-//> \endverbatim
-//>
-// =====================================================================
-/* Subroutine */ int dscal_(int *n, double *da, double *dx, int *incx)
-{
-    // System generated locals
-    int i__1, i__2;
-
-    // Local variables
-    int i__, m, mp1, nincx;
-
-    //
-    // -- Reference BLAS level1 routine (version 3.8.0) --
-    // -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    November 2017
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. Local Scalars ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    // Parameter adjustments
-    --dx;
-
-    // Function Body
-    if (*n <= 0 || *incx <= 0) {
-	return 0;
-    }
-    if (*incx == 1) {
-	//
-	//       code for increment equal to 1
-	//
-	//
-	//       clean-up loop
-	//
-	m = *n % 5;
-	if (m != 0) {
-	    i__1 = m;
-	    for (i__ = 1; i__ <= i__1; ++i__) {
-		dx[i__] = *da * dx[i__];
-	    }
-	    if (*n < 5) {
-		return 0;
-	    }
-	}
-	mp1 = m + 1;
-	i__1 = *n;
-	for (i__ = mp1; i__ <= i__1; i__ += 5) {
-	    dx[i__] = *da * dx[i__];
-	    dx[i__ + 1] = *da * dx[i__ + 1];
-	    dx[i__ + 2] = *da * dx[i__ + 2];
-	    dx[i__ + 3] = *da * dx[i__ + 3];
-	    dx[i__ + 4] = *da * dx[i__ + 4];
-	}
-    } else {
-	//
-	//       code for increment not equal to 1
-	//
-	nincx = *n * *incx;
-	i__1 = nincx;
-	i__2 = *incx;
-	for (i__ = 1; i__2 < 0 ? i__ >= i__1 : i__ <= i__1; i__ += i__2) {
-	    dx[i__] = *da * dx[i__];
-	}
-    }
-    return 0;
-} // dscal_
-
@@ -1,178 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b DSWAP
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE DSWAP(N,DX,INCX,DY,INCY)
-//
-//      .. Scalar Arguments ..
-//      INTEGER INCX,INCY,N
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION DX(*),DY(*)
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//>    DSWAP interchanges two vectors.
-//>    uses unrolled loops for increments equal to 1.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>         number of elements in input vector(s)
-//> \endverbatim
-//>
-//> \param[in,out] DX
-//> \verbatim
-//>          DX is DOUBLE PRECISION array, dimension ( 1 + ( N - 1 )*abs( INCX ) )
-//> \endverbatim
-//>
-//> \param[in] INCX
-//> \verbatim
-//>          INCX is INTEGER
-//>         storage spacing between elements of DX
-//> \endverbatim
-//>
-//> \param[in,out] DY
-//> \verbatim
-//>          DY is DOUBLE PRECISION array, dimension ( 1 + ( N - 1 )*abs( INCY ) )
-//> \endverbatim
-//>
-//> \param[in] INCY
-//> \verbatim
-//>          INCY is INTEGER
-//>         storage spacing between elements of DY
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date November 2017
-//
-//> \ingroup double_blas_level1
-//
-//> \par Further Details:
-// =====================
-//>
-//> \verbatim
-//>
-//>     jack dongarra, linpack, 3/11/78.
-//>     modified 12/3/93, array(1) declarations changed to array(*)
-//> \endverbatim
-//>
-// =====================================================================
-/* Subroutine */ int dswap_(int *n, double *dx, int *incx, double *dy, int *
-	incy)
-{
-    // System generated locals
-    int i__1;
-
-    // Local variables
-    int i__, m, ix, iy, mp1;
-    double dtemp;
-
-    //
-    // -- Reference BLAS level1 routine (version 3.8.0) --
-    // -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    November 2017
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. Local Scalars ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    // Parameter adjustments
-    --dy;
-    --dx;
-
-    // Function Body
-    if (*n <= 0) {
-	return 0;
-    }
-    if (*incx == 1 && *incy == 1) {
-	//
-	//      code for both increments equal to 1
-	//
-	//
-	//      clean-up loop
-	//
-	m = *n % 3;
-	if (m != 0) {
-	    i__1 = m;
-	    for (i__ = 1; i__ <= i__1; ++i__) {
-		dtemp = dx[i__];
-		dx[i__] = dy[i__];
-		dy[i__] = dtemp;
-	    }
-	    if (*n < 3) {
-		return 0;
-	    }
-	}
-	mp1 = m + 1;
-	i__1 = *n;
-	for (i__ = mp1; i__ <= i__1; i__ += 3) {
-	    dtemp = dx[i__];
-	    dx[i__] = dy[i__];
-	    dy[i__] = dtemp;
-	    dtemp = dx[i__ + 1];
-	    dx[i__ + 1] = dy[i__ + 1];
-	    dy[i__ + 1] = dtemp;
-	    dtemp = dx[i__ + 2];
-	    dx[i__ + 2] = dy[i__ + 2];
-	    dy[i__ + 2] = dtemp;
-	}
-    } else {
-	//
-	//      code for unequal increments or equal increments not equal
-	//        to 1
-	//
-	ix = 1;
-	iy = 1;
-	if (*incx < 0) {
-	    ix = (-(*n) + 1) * *incx + 1;
-	}
-	if (*incy < 0) {
-	    iy = (-(*n) + 1) * *incy + 1;
-	}
-	i__1 = *n;
-	for (i__ = 1; i__ <= i__1; ++i__) {
-	    dtemp = dx[ix];
-	    dx[ix] = dy[iy];
-	    dy[iy] = dtemp;
-	    ix += *incx;
-	    iy += *incy;
-	}
-    }
-    return 0;
-} // dswap_
-
@@ -1,509 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b DTRMM
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE DTRMM(SIDE,UPLO,TRANSA,DIAG,M,N,ALPHA,A,LDA,B,LDB)
-//
-//      .. Scalar Arguments ..
-//      DOUBLE PRECISION ALPHA
-//      INTEGER LDA,LDB,M,N
-//      CHARACTER DIAG,SIDE,TRANSA,UPLO
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION A(LDA,*),B(LDB,*)
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> DTRMM  performs one of the matrix-matrix operations
-//>
-//>    B := alpha*op( A )*B,   or   B := alpha*B*op( A ),
-//>
-//> where  alpha  is a scalar,  B  is an m by n matrix,  A  is a unit, or
-//> non-unit,  upper or lower triangular matrix  and  op( A )  is one  of
-//>
-//>    op( A ) = A   or   op( A ) = A**T.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] SIDE
-//> \verbatim
-//>          SIDE is CHARACTER*1
-//>           On entry,  SIDE specifies whether  op( A ) multiplies B from
-//>           the left or right as follows:
-//>
-//>              SIDE = 'L' or 'l'   B := alpha*op( A )*B.
-//>
-//>              SIDE = 'R' or 'r'   B := alpha*B*op( A ).
-//> \endverbatim
-//>
-//> \param[in] UPLO
-//> \verbatim
-//>          UPLO is CHARACTER*1
-//>           On entry, UPLO specifies whether the matrix A is an upper or
-//>           lower triangular matrix as follows:
-//>
-//>              UPLO = 'U' or 'u'   A is an upper triangular matrix.
-//>
-//>              UPLO = 'L' or 'l'   A is a lower triangular matrix.
-//> \endverbatim
-//>
-//> \param[in] TRANSA
-//> \verbatim
-//>          TRANSA is CHARACTER*1
-//>           On entry, TRANSA specifies the form of op( A ) to be used in
-//>           the matrix multiplication as follows:
-//>
-//>              TRANSA = 'N' or 'n'   op( A ) = A.
-//>
-//>              TRANSA = 'T' or 't'   op( A ) = A**T.
-//>
-//>              TRANSA = 'C' or 'c'   op( A ) = A**T.
-//> \endverbatim
-//>
-//> \param[in] DIAG
-//> \verbatim
-//>          DIAG is CHARACTER*1
-//>           On entry, DIAG specifies whether or not A is unit triangular
-//>           as follows:
-//>
-//>              DIAG = 'U' or 'u'   A is assumed to be unit triangular.
-//>
-//>              DIAG = 'N' or 'n'   A is not assumed to be unit
-//>                                  triangular.
-//> \endverbatim
-//>
-//> \param[in] M
-//> \verbatim
-//>          M is INTEGER
-//>           On entry, M specifies the number of rows of B. M must be at
-//>           least zero.
-//> \endverbatim
-//>
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>           On entry, N specifies the number of columns of B.  N must be
-//>           at least zero.
-//> \endverbatim
-//>
-//> \param[in] ALPHA
-//> \verbatim
-//>          ALPHA is DOUBLE PRECISION.
-//>           On entry,  ALPHA specifies the scalar  alpha. When  alpha is
-//>           zero then  A is not referenced and  B need not be set before
-//>           entry.
-//> \endverbatim
-//>
-//> \param[in] A
-//> \verbatim
-//>           A is DOUBLE PRECISION array, dimension ( LDA, k ), where k is m
-//>           when  SIDE = 'L' or 'l'  and is  n  when  SIDE = 'R' or 'r'.
-//>           Before entry  with  UPLO = 'U' or 'u',  the  leading  k by k
-//>           upper triangular part of the array  A must contain the upper
-//>           triangular matrix  and the strictly lower triangular part of
-//>           A is not referenced.
-//>           Before entry  with  UPLO = 'L' or 'l',  the  leading  k by k
-//>           lower triangular part of the array  A must contain the lower
-//>           triangular matrix  and the strictly upper triangular part of
-//>           A is not referenced.
-//>           Note that when  DIAG = 'U' or 'u',  the diagonal elements of
-//>           A  are not referenced either,  but are assumed to be  unity.
-//> \endverbatim
-//>
-//> \param[in] LDA
-//> \verbatim
-//>          LDA is INTEGER
-//>           On entry, LDA specifies the first dimension of A as declared
-//>           in the calling (sub) program.  When  SIDE = 'L' or 'l'  then
-//>           LDA  must be at least  max( 1, m ),  when  SIDE = 'R' or 'r'
-//>           then LDA must be at least max( 1, n ).
-//> \endverbatim
-//>
-//> \param[in,out] B
-//> \verbatim
-//>          B is DOUBLE PRECISION array, dimension ( LDB, N )
-//>           Before entry,  the leading  m by n part of the array  B must
-//>           contain the matrix  B,  and  on exit  is overwritten  by the
-//>           transformed matrix.
-//> \endverbatim
-//>
-//> \param[in] LDB
-//> \verbatim
-//>          LDB is INTEGER
-//>           On entry, LDB specifies the first dimension of B as declared
-//>           in  the  calling  (sub)  program.   LDB  must  be  at  least
-//>           max( 1, m ).
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date December 2016
-//
-//> \ingroup double_blas_level3
-//
-//> \par Further Details:
-// =====================
-//>
-//> \verbatim
-//>
-//>  Level 3 Blas routine.
-//>
-//>  -- Written on 8-February-1989.
-//>     Jack Dongarra, Argonne National Laboratory.
-//>     Iain Duff, AERE Harwell.
-//>     Jeremy Du Croz, Numerical Algorithms Group Ltd.
-//>     Sven Hammarling, Numerical Algorithms Group Ltd.
-//> \endverbatim
-//>
-// =====================================================================
-/* Subroutine */ int dtrmm_(char *side, char *uplo, char *transa, char *diag,
-	int *m, int *n, double *alpha, double *a, int *lda, double *b, int *
-	ldb)
-{
-    // System generated locals
-    int a_dim1, a_offset, b_dim1, b_offset, i__1, i__2, i__3;
-
-    // Local variables
-    int i__, j, k, info;
-    double temp;
-    int lside;
-    extern int lsame_(char *, char *);
-    int nrowa;
-    int upper;
-    extern /* Subroutine */ int xerbla_(char *, int *);
-    int nounit;
-
-    //
-    // -- Reference BLAS level3 routine (version 3.7.0) --
-    // -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    December 2016
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. External Functions ..
-    //    ..
-    //    .. External Subroutines ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    //    .. Local Scalars ..
-    //    ..
-    //    .. Parameters ..
-    //    ..
-    //
-    //    Test the input parameters.
-    //
-    // Parameter adjustments
-    a_dim1 = *lda;
-    a_offset = 1 + a_dim1;
-    a -= a_offset;
-    b_dim1 = *ldb;
-    b_offset = 1 + b_dim1;
-    b -= b_offset;
-
-    // Function Body
-    lside = lsame_(side, "L");
-    if (lside) {
-	nrowa = *m;
-    } else {
-	nrowa = *n;
-    }
-    nounit = lsame_(diag, "N");
-    upper = lsame_(uplo, "U");
-    info = 0;
-    if (! lside && ! lsame_(side, "R")) {
-	info = 1;
-    } else if (! upper && ! lsame_(uplo, "L")) {
-	info = 2;
-    } else if (! lsame_(transa, "N") && ! lsame_(transa, "T") && ! lsame_(
-	    transa, "C")) {
-	info = 3;
-    } else if (! lsame_(diag, "U") && ! lsame_(diag, "N")) {
-	info = 4;
-    } else if (*m < 0) {
-	info = 5;
-    } else if (*n < 0) {
-	info = 6;
-    } else if (*lda < max(1,nrowa)) {
-	info = 9;
-    } else if (*ldb < max(1,*m)) {
-	info = 11;
-    }
-    if (info != 0) {
-	xerbla_("DTRMM ", &info);
-	return 0;
-    }
-    //
-    //    Quick return if possible.
-    //
-    if (*m == 0 || *n == 0) {
-	return 0;
-    }
-    //
-    //    And when  alpha.eq.zero.
-    //
-    if (*alpha == 0.) {
-	i__1 = *n;
-	for (j = 1; j <= i__1; ++j) {
-	    i__2 = *m;
-	    for (i__ = 1; i__ <= i__2; ++i__) {
-		b[i__ + j * b_dim1] = 0.;
-// L10:
-	    }
-// L20:
-	}
-	return 0;
-    }
-    //
-    //    Start the operations.
-    //
-    if (lside) {
-	if (lsame_(transa, "N")) {
-	    //
-	    //          Form  B := alpha*A*B.
-	    //
-	    if (upper) {
-		i__1 = *n;
-		for (j = 1; j <= i__1; ++j) {
-		    i__2 = *m;
-		    for (k = 1; k <= i__2; ++k) {
-			if (b[k + j * b_dim1] != 0.) {
-			    temp = *alpha * b[k + j * b_dim1];
-			    i__3 = k - 1;
-			    for (i__ = 1; i__ <= i__3; ++i__) {
-				b[i__ + j * b_dim1] += temp * a[i__ + k *
-					a_dim1];
-// L30:
-			    }
-			    if (nounit) {
-				temp *= a[k + k * a_dim1];
-			    }
-			    b[k + j * b_dim1] = temp;
-			}
-// L40:
-		    }
-// L50:
-		}
-	    } else {
-		i__1 = *n;
-		for (j = 1; j <= i__1; ++j) {
-		    for (k = *m; k >= 1; --k) {
-			if (b[k + j * b_dim1] != 0.) {
-			    temp = *alpha * b[k + j * b_dim1];
-			    b[k + j * b_dim1] = temp;
-			    if (nounit) {
-				b[k + j * b_dim1] *= a[k + k * a_dim1];
-			    }
-			    i__2 = *m;
-			    for (i__ = k + 1; i__ <= i__2; ++i__) {
-				b[i__ + j * b_dim1] += temp * a[i__ + k *
-					a_dim1];
-// L60:
-			    }
-			}
-// L70:
-		    }
-// L80:
-		}
-	    }
-	} else {
-	    //
-	    //          Form  B := alpha*A**T*B.
-	    //
-	    if (upper) {
-		i__1 = *n;
-		for (j = 1; j <= i__1; ++j) {
-		    for (i__ = *m; i__ >= 1; --i__) {
-			temp = b[i__ + j * b_dim1];
-			if (nounit) {
-			    temp *= a[i__ + i__ * a_dim1];
-			}
-			i__2 = i__ - 1;
-			for (k = 1; k <= i__2; ++k) {
-			    temp += a[k + i__ * a_dim1] * b[k + j * b_dim1];
-// L90:
-			}
-			b[i__ + j * b_dim1] = *alpha * temp;
-// L100:
-		    }
-// L110:
-		}
-	    } else {
-		i__1 = *n;
-		for (j = 1; j <= i__1; ++j) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			temp = b[i__ + j * b_dim1];
-			if (nounit) {
-			    temp *= a[i__ + i__ * a_dim1];
-			}
-			i__3 = *m;
-			for (k = i__ + 1; k <= i__3; ++k) {
-			    temp += a[k + i__ * a_dim1] * b[k + j * b_dim1];
-// L120:
-			}
-			b[i__ + j * b_dim1] = *alpha * temp;
-// L130:
-		    }
-// L140:
-		}
-	    }
-	}
-    } else {
-	if (lsame_(transa, "N")) {
-	    //
-	    //          Form  B := alpha*B*A.
-	    //
-	    if (upper) {
-		for (j = *n; j >= 1; --j) {
-		    temp = *alpha;
-		    if (nounit) {
-			temp *= a[j + j * a_dim1];
-		    }
-		    i__1 = *m;
-		    for (i__ = 1; i__ <= i__1; ++i__) {
-			b[i__ + j * b_dim1] = temp * b[i__ + j * b_dim1];
-// L150:
-		    }
-		    i__1 = j - 1;
-		    for (k = 1; k <= i__1; ++k) {
-			if (a[k + j * a_dim1] != 0.) {
-			    temp = *alpha * a[k + j * a_dim1];
-			    i__2 = *m;
-			    for (i__ = 1; i__ <= i__2; ++i__) {
-				b[i__ + j * b_dim1] += temp * b[i__ + k *
-					b_dim1];
-// L160:
-			    }
-			}
-// L170:
-		    }
-// L180:
-		}
-	    } else {
-		i__1 = *n;
-		for (j = 1; j <= i__1; ++j) {
-		    temp = *alpha;
-		    if (nounit) {
-			temp *= a[j + j * a_dim1];
-		    }
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			b[i__ + j * b_dim1] = temp * b[i__ + j * b_dim1];
-// L190:
-		    }
-		    i__2 = *n;
-		    for (k = j + 1; k <= i__2; ++k) {
-			if (a[k + j * a_dim1] != 0.) {
-			    temp = *alpha * a[k + j * a_dim1];
-			    i__3 = *m;
-			    for (i__ = 1; i__ <= i__3; ++i__) {
-				b[i__ + j * b_dim1] += temp * b[i__ + k *
-					b_dim1];
-// L200:
-			    }
-			}
-// L210:
-		    }
-// L220:
-		}
-	    }
-	} else {
-	    //
-	    //          Form  B := alpha*B*A**T.
-	    //
-	    if (upper) {
-		i__1 = *n;
-		for (k = 1; k <= i__1; ++k) {
-		    i__2 = k - 1;
-		    for (j = 1; j <= i__2; ++j) {
-			if (a[j + k * a_dim1] != 0.) {
-			    temp = *alpha * a[j + k * a_dim1];
-			    i__3 = *m;
-			    for (i__ = 1; i__ <= i__3; ++i__) {
-				b[i__ + j * b_dim1] += temp * b[i__ + k *
-					b_dim1];
-// L230:
-			    }
-			}
-// L240:
-		    }
-		    temp = *alpha;
-		    if (nounit) {
-			temp *= a[k + k * a_dim1];
-		    }
-		    if (temp != 1.) {
-			i__2 = *m;
-			for (i__ = 1; i__ <= i__2; ++i__) {
-			    b[i__ + k * b_dim1] = temp * b[i__ + k * b_dim1];
-// L250:
-			}
-		    }
-// L260:
-		}
-	    } else {
-		for (k = *n; k >= 1; --k) {
-		    i__1 = *n;
-		    for (j = k + 1; j <= i__1; ++j) {
-			if (a[j + k * a_dim1] != 0.) {
-			    temp = *alpha * a[j + k * a_dim1];
-			    i__2 = *m;
-			    for (i__ = 1; i__ <= i__2; ++i__) {
-				b[i__ + j * b_dim1] += temp * b[i__ + k *
-					b_dim1];
-// L270:
-			    }
-			}
-// L280:
-		    }
-		    temp = *alpha;
-		    if (nounit) {
-			temp *= a[k + k * a_dim1];
-		    }
-		    if (temp != 1.) {
-			i__1 = *m;
-			for (i__ = 1; i__ <= i__1; ++i__) {
-			    b[i__ + k * b_dim1] = temp * b[i__ + k * b_dim1];
-// L290:
-			}
-		    }
-// L300:
-		}
-	    }
-	}
-    }
-    return 0;
-    //
-    //    End of DTRMM .
-    //
-} // dtrmm_
-
@@ -1,396 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b DTRMV
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE DTRMV(UPLO,TRANS,DIAG,N,A,LDA,X,INCX)
-//
-//      .. Scalar Arguments ..
-//      INTEGER INCX,LDA,N
-//      CHARACTER DIAG,TRANS,UPLO
-//      ..
-//      .. Array Arguments ..
-//      DOUBLE PRECISION A(LDA,*),X(*)
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> DTRMV  performs one of the matrix-vector operations
-//>
-//>    x := A*x,   or   x := A**T*x,
-//>
-//> where x is an n element vector and  A is an n by n unit, or non-unit,
-//> upper or lower triangular matrix.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] UPLO
-//> \verbatim
-//>          UPLO is CHARACTER*1
-//>           On entry, UPLO specifies whether the matrix is an upper or
-//>           lower triangular matrix as follows:
-//>
-//>              UPLO = 'U' or 'u'   A is an upper triangular matrix.
-//>
-//>              UPLO = 'L' or 'l'   A is a lower triangular matrix.
-//> \endverbatim
-//>
-//> \param[in] TRANS
-//> \verbatim
-//>          TRANS is CHARACTER*1
-//>           On entry, TRANS specifies the operation to be performed as
-//>           follows:
-//>
-//>              TRANS = 'N' or 'n'   x := A*x.
-//>
-//>              TRANS = 'T' or 't'   x := A**T*x.
-//>
-//>              TRANS = 'C' or 'c'   x := A**T*x.
-//> \endverbatim
-//>
-//> \param[in] DIAG
-//> \verbatim
-//>          DIAG is CHARACTER*1
-//>           On entry, DIAG specifies whether or not A is unit
-//>           triangular as follows:
-//>
-//>              DIAG = 'U' or 'u'   A is assumed to be unit triangular.
-//>
-//>              DIAG = 'N' or 'n'   A is not assumed to be unit
-//>                                  triangular.
-//> \endverbatim
-//>
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>           On entry, N specifies the order of the matrix A.
-//>           N must be at least zero.
-//> \endverbatim
-//>
-//> \param[in] A
-//> \verbatim
-//>          A is DOUBLE PRECISION array, dimension ( LDA, N )
-//>           Before entry with  UPLO = 'U' or 'u', the leading n by n
-//>           upper triangular part of the array A must contain the upper
-//>           triangular matrix and the strictly lower triangular part of
-//>           A is not referenced.
-//>           Before entry with UPLO = 'L' or 'l', the leading n by n
-//>           lower triangular part of the array A must contain the lower
-//>           triangular matrix and the strictly upper triangular part of
-//>           A is not referenced.
-//>           Note that when  DIAG = 'U' or 'u', the diagonal elements of
-//>           A are not referenced either, but are assumed to be unity.
-//> \endverbatim
-//>
-//> \param[in] LDA
-//> \verbatim
-//>          LDA is INTEGER
-//>           On entry, LDA specifies the first dimension of A as declared
-//>           in the calling (sub) program. LDA must be at least
-//>           max( 1, n ).
-//> \endverbatim
-//>
-//> \param[in,out] X
-//> \verbatim
-//>          X is DOUBLE PRECISION array, dimension at least
-//>           ( 1 + ( n - 1 )*abs( INCX ) ).
-//>           Before entry, the incremented array X must contain the n
-//>           element vector x. On exit, X is overwritten with the
-//>           transformed vector x.
-//> \endverbatim
-//>
-//> \param[in] INCX
-//> \verbatim
-//>          INCX is INTEGER
-//>           On entry, INCX specifies the increment for the elements of
-//>           X. INCX must not be zero.
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date December 2016
-//
-//> \ingroup double_blas_level2
-//
-//> \par Further Details:
-// =====================
-//>
-//> \verbatim
-//>
-//>  Level 2 Blas routine.
-//>  The vector and matrix arguments are not referenced when N = 0, or M = 0
-//>
-//>  -- Written on 22-October-1986.
-//>     Jack Dongarra, Argonne National Lab.
-//>     Jeremy Du Croz, Nag Central Office.
-//>     Sven Hammarling, Nag Central Office.
-//>     Richard Hanson, Sandia National Labs.
-//> \endverbatim
-//>
-// =====================================================================
-/* Subroutine */ int dtrmv_(char *uplo, char *trans, char *diag, int *n,
-	double *a, int *lda, double *x, int *incx)
-{
-    // System generated locals
-    int a_dim1, a_offset, i__1, i__2;
-
-    // Local variables
-    int i__, j, ix, jx, kx, info;
-    double temp;
-    extern int lsame_(char *, char *);
-    extern /* Subroutine */ int xerbla_(char *, int *);
-    int nounit;
-
-    //
-    // -- Reference BLAS level2 routine (version 3.7.0) --
-    // -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    December 2016
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. Parameters ..
-    //    ..
-    //    .. Local Scalars ..
-    //    ..
-    //    .. External Functions ..
-    //    ..
-    //    .. External Subroutines ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    //
-    //    Test the input parameters.
-    //
-    // Parameter adjustments
-    a_dim1 = *lda;
-    a_offset = 1 + a_dim1;
-    a -= a_offset;
-    --x;
-
-    // Function Body
-    info = 0;
-    if (! lsame_(uplo, "U") && ! lsame_(uplo, "L")) {
-	info = 1;
-    } else if (! lsame_(trans, "N") && ! lsame_(trans, "T") && ! lsame_(trans,
-	     "C")) {
-	info = 2;
-    } else if (! lsame_(diag, "U") && ! lsame_(diag, "N")) {
-	info = 3;
-    } else if (*n < 0) {
-	info = 4;
-    } else if (*lda < max(1,*n)) {
-	info = 6;
-    } else if (*incx == 0) {
-	info = 8;
-    }
-    if (info != 0) {
-	xerbla_("DTRMV ", &info);
-	return 0;
-    }
-    //
-    //    Quick return if possible.
-    //
-    if (*n == 0) {
-	return 0;
-    }
-    nounit = lsame_(diag, "N");
-    //
-    //    Set up the start point in X if the increment is not unity. This
-    //    will be  ( N - 1 )*INCX  too small for descending loops.
-    //
-    if (*incx <= 0) {
-	kx = 1 - (*n - 1) * *incx;
-    } else if (*incx != 1) {
-	kx = 1;
-    }
-    //
-    //    Start the operations. In this version the elements of A are
-    //    accessed sequentially with one pass through A.
-    //
-    if (lsame_(trans, "N")) {
-	//
-	//       Form  x := A*x.
-	//
-	if (lsame_(uplo, "U")) {
-	    if (*incx == 1) {
-		i__1 = *n;
-		for (j = 1; j <= i__1; ++j) {
-		    if (x[j] != 0.) {
-			temp = x[j];
-			i__2 = j - 1;
-			for (i__ = 1; i__ <= i__2; ++i__) {
-			    x[i__] += temp * a[i__ + j * a_dim1];
-// L10:
-			}
-			if (nounit) {
-			    x[j] *= a[j + j * a_dim1];
-			}
-		    }
-// L20:
-		}
-	    } else {
-		jx = kx;
-		i__1 = *n;
-		for (j = 1; j <= i__1; ++j) {
-		    if (x[jx] != 0.) {
-			temp = x[jx];
-			ix = kx;
-			i__2 = j - 1;
-			for (i__ = 1; i__ <= i__2; ++i__) {
-			    x[ix] += temp * a[i__ + j * a_dim1];
-			    ix += *incx;
-// L30:
-			}
-			if (nounit) {
-			    x[jx] *= a[j + j * a_dim1];
-			}
-		    }
-		    jx += *incx;
-// L40:
-		}
-	    }
-	} else {
-	    if (*incx == 1) {
-		for (j = *n; j >= 1; --j) {
-		    if (x[j] != 0.) {
-			temp = x[j];
-			i__1 = j + 1;
-			for (i__ = *n; i__ >= i__1; --i__) {
-			    x[i__] += temp * a[i__ + j * a_dim1];
-// L50:
-			}
-			if (nounit) {
-			    x[j] *= a[j + j * a_dim1];
-			}
-		    }
-// L60:
-		}
-	    } else {
-		kx += (*n - 1) * *incx;
-		jx = kx;
-		for (j = *n; j >= 1; --j) {
-		    if (x[jx] != 0.) {
-			temp = x[jx];
-			ix = kx;
-			i__1 = j + 1;
-			for (i__ = *n; i__ >= i__1; --i__) {
-			    x[ix] += temp * a[i__ + j * a_dim1];
-			    ix -= *incx;
-// L70:
-			}
-			if (nounit) {
-			    x[jx] *= a[j + j * a_dim1];
-			}
-		    }
-		    jx -= *incx;
-// L80:
-		}
-	    }
-	}
-    } else {
-	//
-	//       Form  x := A**T*x.
-	//
-	if (lsame_(uplo, "U")) {
-	    if (*incx == 1) {
-		for (j = *n; j >= 1; --j) {
-		    temp = x[j];
-		    if (nounit) {
-			temp *= a[j + j * a_dim1];
-		    }
-		    for (i__ = j - 1; i__ >= 1; --i__) {
-			temp += a[i__ + j * a_dim1] * x[i__];
-// L90:
-		    }
-		    x[j] = temp;
-// L100:
-		}
-	    } else {
-		jx = kx + (*n - 1) * *incx;
-		for (j = *n; j >= 1; --j) {
-		    temp = x[jx];
-		    ix = jx;
-		    if (nounit) {
-			temp *= a[j + j * a_dim1];
-		    }
-		    for (i__ = j - 1; i__ >= 1; --i__) {
-			ix -= *incx;
-			temp += a[i__ + j * a_dim1] * x[ix];
-// L110:
-		    }
-		    x[jx] = temp;
-		    jx -= *incx;
-// L120:
-		}
-	    }
-	} else {
-	    if (*incx == 1) {
-		i__1 = *n;
-		for (j = 1; j <= i__1; ++j) {
-		    temp = x[j];
-		    if (nounit) {
-			temp *= a[j + j * a_dim1];
-		    }
-		    i__2 = *n;
-		    for (i__ = j + 1; i__ <= i__2; ++i__) {
-			temp += a[i__ + j * a_dim1] * x[i__];
-// L130:
-		    }
-		    x[j] = temp;
-// L140:
-		}
-	    } else {
-		jx = kx;
-		i__1 = *n;
-		for (j = 1; j <= i__1; ++j) {
-		    temp = x[jx];
-		    ix = jx;
-		    if (nounit) {
-			temp *= a[j + j * a_dim1];
-		    }
-		    i__2 = *n;
-		    for (i__ = j + 1; i__ <= i__2; ++i__) {
-			ix += *incx;
-			temp += a[i__ + j * a_dim1] * x[ix];
-// L150:
-		    }
-		    x[jx] = temp;
-		    jx += *incx;
-// L160:
-		}
-	    }
-	}
-    }
-    return 0;
-    //
-    //    End of DTRMV .
-    //
-} // dtrmv_
-
@@ -1,444 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b SGEMM
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE SGEMM(TRANSA,TRANSB,M,N,K,ALPHA,A,LDA,B,LDB,BETA,C,LDC)
-//
-//      .. Scalar Arguments ..
-//      REAL ALPHA,BETA
-//      INTEGER K,LDA,LDB,LDC,M,N
-//      CHARACTER TRANSA,TRANSB
-//      ..
-//      .. Array Arguments ..
-//      REAL A(LDA,*),B(LDB,*),C(LDC,*)
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> SGEMM  performs one of the matrix-matrix operations
-//>
-//>    C := alpha*op( A )*op( B ) + beta*C,
-//>
-//> where  op( X ) is one of
-//>
-//>    op( X ) = X   or   op( X ) = X**T,
-//>
-//> alpha and beta are scalars, and A, B and C are matrices, with op( A )
-//> an m by k matrix,  op( B )  a  k by n matrix and  C an m by n matrix.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] TRANSA
-//> \verbatim
-//>          TRANSA is CHARACTER*1
-//>           On entry, TRANSA specifies the form of op( A ) to be used in
-//>           the matrix multiplication as follows:
-//>
-//>              TRANSA = 'N' or 'n',  op( A ) = A.
-//>
-//>              TRANSA = 'T' or 't',  op( A ) = A**T.
-//>
-//>              TRANSA = 'C' or 'c',  op( A ) = A**T.
-//> \endverbatim
-//>
-//> \param[in] TRANSB
-//> \verbatim
-//>          TRANSB is CHARACTER*1
-//>           On entry, TRANSB specifies the form of op( B ) to be used in
-//>           the matrix multiplication as follows:
-//>
-//>              TRANSB = 'N' or 'n',  op( B ) = B.
-//>
-//>              TRANSB = 'T' or 't',  op( B ) = B**T.
-//>
-//>              TRANSB = 'C' or 'c',  op( B ) = B**T.
-//> \endverbatim
-//>
-//> \param[in] M
-//> \verbatim
-//>          M is INTEGER
-//>           On entry,  M  specifies  the number  of rows  of the  matrix
-//>           op( A )  and of the  matrix  C.  M  must  be at least  zero.
-//> \endverbatim
-//>
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>           On entry,  N  specifies the number  of columns of the matrix
-//>           op( B ) and the number of columns of the matrix C. N must be
-//>           at least zero.
-//> \endverbatim
-//>
-//> \param[in] K
-//> \verbatim
-//>          K is INTEGER
-//>           On entry,  K  specifies  the number of columns of the matrix
-//>           op( A ) and the number of rows of the matrix op( B ). K must
-//>           be at least  zero.
-//> \endverbatim
-//>
-//> \param[in] ALPHA
-//> \verbatim
-//>          ALPHA is REAL
-//>           On entry, ALPHA specifies the scalar alpha.
-//> \endverbatim
-//>
-//> \param[in] A
-//> \verbatim
-//>          A is REAL array, dimension ( LDA, ka ), where ka is
-//>           k  when  TRANSA = 'N' or 'n',  and is  m  otherwise.
-//>           Before entry with  TRANSA = 'N' or 'n',  the leading  m by k
-//>           part of the array  A  must contain the matrix  A,  otherwise
-//>           the leading  k by m  part of the array  A  must contain  the
-//>           matrix A.
-//> \endverbatim
-//>
-//> \param[in] LDA
-//> \verbatim
-//>          LDA is INTEGER
-//>           On entry, LDA specifies the first dimension of A as declared
-//>           in the calling (sub) program. When  TRANSA = 'N' or 'n' then
-//>           LDA must be at least  max( 1, m ), otherwise  LDA must be at
-//>           least  max( 1, k ).
-//> \endverbatim
-//>
-//> \param[in] B
-//> \verbatim
-//>          B is REAL array, dimension ( LDB, kb ), where kb is
-//>           n  when  TRANSB = 'N' or 'n',  and is  k  otherwise.
-//>           Before entry with  TRANSB = 'N' or 'n',  the leading  k by n
-//>           part of the array  B  must contain the matrix  B,  otherwise
-//>           the leading  n by k  part of the array  B  must contain  the
-//>           matrix B.
-//> \endverbatim
-//>
-//> \param[in] LDB
-//> \verbatim
-//>          LDB is INTEGER
-//>           On entry, LDB specifies the first dimension of B as declared
-//>           in the calling (sub) program. When  TRANSB = 'N' or 'n' then
-//>           LDB must be at least  max( 1, k ), otherwise  LDB must be at
-//>           least  max( 1, n ).
-//> \endverbatim
-//>
-//> \param[in] BETA
-//> \verbatim
-//>          BETA is REAL
-//>           On entry,  BETA  specifies the scalar  beta.  When  BETA  is
-//>           supplied as zero then C need not be set on input.
-//> \endverbatim
-//>
-//> \param[in,out] C
-//> \verbatim
-//>          C is REAL array, dimension ( LDC, N )
-//>           Before entry, the leading  m by n  part of the array  C must
-//>           contain the matrix  C,  except when  beta  is zero, in which
-//>           case C need not be set on entry.
-//>           On exit, the array  C  is overwritten by the  m by n  matrix
-//>           ( alpha*op( A )*op( B ) + beta*C ).
-//> \endverbatim
-//>
-//> \param[in] LDC
-//> \verbatim
-//>          LDC is INTEGER
-//>           On entry, LDC specifies the first dimension of C as declared
-//>           in  the  calling  (sub)  program.   LDC  must  be  at  least
-//>           max( 1, m ).
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date December 2016
-//
-//> \ingroup single_blas_level3
-//
-//> \par Further Details:
-// =====================
-//>
-//> \verbatim
-//>
-//>  Level 3 Blas routine.
-//>
-//>  -- Written on 8-February-1989.
-//>     Jack Dongarra, Argonne National Laboratory.
-//>     Iain Duff, AERE Harwell.
-//>     Jeremy Du Croz, Numerical Algorithms Group Ltd.
-//>     Sven Hammarling, Numerical Algorithms Group Ltd.
-//> \endverbatim
-//>
-// =====================================================================
-/* Subroutine */ int sgemm_(char *transa, char *transb, int *m, int *n, int *
-	k, float *alpha, float *a, int *lda, float *b, int *ldb, float *beta,
-	float *c__, int *ldc)
-{
-    // System generated locals
-    int a_dim1, a_offset, b_dim1, b_offset, c_dim1, c_offset, i__1, i__2,
-	    i__3;
-
-    // Local variables
-    int i__, j, l, info;
-    int nota, notb;
-    float temp;
-    int ncola;
-    extern int lsame_(char *, char *);
-    int nrowa, nrowb;
-    extern /* Subroutine */ int xerbla_(char *, int *);
-
-    //
-    // -- Reference BLAS level3 routine (version 3.7.0) --
-    // -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    December 2016
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. External Functions ..
-    //    ..
-    //    .. External Subroutines ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    //    .. Local Scalars ..
-    //    ..
-    //    .. Parameters ..
-    //    ..
-    //
-    //    Set  NOTA  and  NOTB  as  true if  A  and  B  respectively are not
-    //    transposed and set  NROWA, NCOLA and  NROWB  as the number of rows
-    //    and  columns of  A  and the  number of  rows  of  B  respectively.
-    //
-    // Parameter adjustments
-    a_dim1 = *lda;
-    a_offset = 1 + a_dim1;
-    a -= a_offset;
-    b_dim1 = *ldb;
-    b_offset = 1 + b_dim1;
-    b -= b_offset;
-    c_dim1 = *ldc;
-    c_offset = 1 + c_dim1;
-    c__ -= c_offset;
-
-    // Function Body
-    nota = lsame_(transa, "N");
-    notb = lsame_(transb, "N");
-    if (nota) {
-	nrowa = *m;
-	ncola = *k;
-    } else {
-	nrowa = *k;
-	ncola = *m;
-    }
-    if (notb) {
-	nrowb = *k;
-    } else {
-	nrowb = *n;
-    }
-    //
-    //    Test the input parameters.
-    //
-    info = 0;
-    if (! nota && ! lsame_(transa, "C") && ! lsame_(transa, "T")) {
-	info = 1;
-    } else if (! notb && ! lsame_(transb, "C") && ! lsame_(transb, "T")) {
-	info = 2;
-    } else if (*m < 0) {
-	info = 3;
-    } else if (*n < 0) {
-	info = 4;
-    } else if (*k < 0) {
-	info = 5;
-    } else if (*lda < max(1,nrowa)) {
-	info = 8;
-    } else if (*ldb < max(1,nrowb)) {
-	info = 10;
-    } else if (*ldc < max(1,*m)) {
-	info = 13;
-    }
-    if (info != 0) {
-	xerbla_("SGEMM ", &info);
-	return 0;
-    }
-    //
-    //    Quick return if possible.
-    //
-    if (*m == 0 || *n == 0 || (*alpha == 0.f || *k == 0) && *beta == 1.f) {
-	return 0;
-    }
-    //
-    //    And if  alpha.eq.zero.
-    //
-    if (*alpha == 0.f) {
-	if (*beta == 0.f) {
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    c__[i__ + j * c_dim1] = 0.f;
-// L10:
-		}
-// L20:
-	    }
-	} else {
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    c__[i__ + j * c_dim1] = *beta * c__[i__ + j * c_dim1];
-// L30:
-		}
-// L40:
-	    }
-	}
-	return 0;
-    }
-    //
-    //    Start the operations.
-    //
-    if (notb) {
-	if (nota) {
-	    //
-	    //          Form  C := alpha*A*B + beta*C.
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		if (*beta == 0.f) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			c__[i__ + j * c_dim1] = 0.f;
-// L50:
-		    }
-		} else if (*beta != 1.f) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			c__[i__ + j * c_dim1] = *beta * c__[i__ + j * c_dim1];
-// L60:
-		    }
-		}
-		i__2 = *k;
-		for (l = 1; l <= i__2; ++l) {
-		    temp = *alpha * b[l + j * b_dim1];
-		    i__3 = *m;
-		    for (i__ = 1; i__ <= i__3; ++i__) {
-			c__[i__ + j * c_dim1] += temp * a[i__ + l * a_dim1];
-// L70:
-		    }
-// L80:
-		}
-// L90:
-	    }
-	} else {
-	    //
-	    //          Form  C := alpha*A**T*B + beta*C
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    temp = 0.f;
-		    i__3 = *k;
-		    for (l = 1; l <= i__3; ++l) {
-			temp += a[l + i__ * a_dim1] * b[l + j * b_dim1];
-// L100:
-		    }
-		    if (*beta == 0.f) {
-			c__[i__ + j * c_dim1] = *alpha * temp;
-		    } else {
-			c__[i__ + j * c_dim1] = *alpha * temp + *beta * c__[
-				i__ + j * c_dim1];
-		    }
-// L110:
-		}
-// L120:
-	    }
-	}
-    } else {
-	if (nota) {
-	    //
-	    //          Form  C := alpha*A*B**T + beta*C
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		if (*beta == 0.f) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			c__[i__ + j * c_dim1] = 0.f;
-// L130:
-		    }
-		} else if (*beta != 1.f) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			c__[i__ + j * c_dim1] = *beta * c__[i__ + j * c_dim1];
-// L140:
-		    }
-		}
-		i__2 = *k;
-		for (l = 1; l <= i__2; ++l) {
-		    temp = *alpha * b[j + l * b_dim1];
-		    i__3 = *m;
-		    for (i__ = 1; i__ <= i__3; ++i__) {
-			c__[i__ + j * c_dim1] += temp * a[i__ + l * a_dim1];
-// L150:
-		    }
-// L160:
-		}
-// L170:
-	    }
-	} else {
-	    //
-	    //          Form  C := alpha*A**T*B**T + beta*C
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    temp = 0.f;
-		    i__3 = *k;
-		    for (l = 1; l <= i__3; ++l) {
-			temp += a[l + i__ * a_dim1] * b[j + l * b_dim1];
-// L180:
-		    }
-		    if (*beta == 0.f) {
-			c__[i__ + j * c_dim1] = *alpha * temp;
-		    } else {
-			c__[i__ + j * c_dim1] = *alpha * temp + *beta * c__[
-				i__ + j * c_dim1];
-		    }
-// L190:
-		}
-// L200:
-	    }
-	}
-    }
-    return 0;
-    //
-    //    End of SGEMM .
-    //
-} // sgemm_
-
@@ -1,752 +0,0 @@
-/*  -- translated by f2c (version 20201020 (for_lapack)). -- */
-
-#include "f2c.h"
-
-//> \brief \b ZGEMM
-//
-// =========== DOCUMENTATION ===========
-//
-// Online html documentation available at
-//           http://www.netlib.org/lapack/explore-html/
-//
-// Definition:
-// ===========
-//
-//      SUBROUTINE ZGEMM(TRANSA,TRANSB,M,N,K,ALPHA,A,LDA,B,LDB,BETA,C,LDC)
-//
-//      .. Scalar Arguments ..
-//      COMPLEX*16 ALPHA,BETA
-//      INTEGER K,LDA,LDB,LDC,M,N
-//      CHARACTER TRANSA,TRANSB
-//      ..
-//      .. Array Arguments ..
-//      COMPLEX*16 A(LDA,*),B(LDB,*),C(LDC,*)
-//      ..
-//
-//
-//> \par Purpose:
-// =============
-//>
-//> \verbatim
-//>
-//> ZGEMM  performs one of the matrix-matrix operations
-//>
-//>    C := alpha*op( A )*op( B ) + beta*C,
-//>
-//> where  op( X ) is one of
-//>
-//>    op( X ) = X   or   op( X ) = X**T   or   op( X ) = X**H,
-//>
-//> alpha and beta are scalars, and A, B and C are matrices, with op( A )
-//> an m by k matrix,  op( B )  a  k by n matrix and  C an m by n matrix.
-//> \endverbatim
-//
-// Arguments:
-// ==========
-//
-//> \param[in] TRANSA
-//> \verbatim
-//>          TRANSA is CHARACTER*1
-//>           On entry, TRANSA specifies the form of op( A ) to be used in
-//>           the matrix multiplication as follows:
-//>
-//>              TRANSA = 'N' or 'n',  op( A ) = A.
-//>
-//>              TRANSA = 'T' or 't',  op( A ) = A**T.
-//>
-//>              TRANSA = 'C' or 'c',  op( A ) = A**H.
-//> \endverbatim
-//>
-//> \param[in] TRANSB
-//> \verbatim
-//>          TRANSB is CHARACTER*1
-//>           On entry, TRANSB specifies the form of op( B ) to be used in
-//>           the matrix multiplication as follows:
-//>
-//>              TRANSB = 'N' or 'n',  op( B ) = B.
-//>
-//>              TRANSB = 'T' or 't',  op( B ) = B**T.
-//>
-//>              TRANSB = 'C' or 'c',  op( B ) = B**H.
-//> \endverbatim
-//>
-//> \param[in] M
-//> \verbatim
-//>          M is INTEGER
-//>           On entry,  M  specifies  the number  of rows  of the  matrix
-//>           op( A )  and of the  matrix  C.  M  must  be at least  zero.
-//> \endverbatim
-//>
-//> \param[in] N
-//> \verbatim
-//>          N is INTEGER
-//>           On entry,  N  specifies the number  of columns of the matrix
-//>           op( B ) and the number of columns of the matrix C. N must be
-//>           at least zero.
-//> \endverbatim
-//>
-//> \param[in] K
-//> \verbatim
-//>          K is INTEGER
-//>           On entry,  K  specifies  the number of columns of the matrix
-//>           op( A ) and the number of rows of the matrix op( B ). K must
-//>           be at least  zero.
-//> \endverbatim
-//>
-//> \param[in] ALPHA
-//> \verbatim
-//>          ALPHA is COMPLEX*16
-//>           On entry, ALPHA specifies the scalar alpha.
-//> \endverbatim
-//>
-//> \param[in] A
-//> \verbatim
-//>          A is COMPLEX*16 array, dimension ( LDA, ka ), where ka is
-//>           k  when  TRANSA = 'N' or 'n',  and is  m  otherwise.
-//>           Before entry with  TRANSA = 'N' or 'n',  the leading  m by k
-//>           part of the array  A  must contain the matrix  A,  otherwise
-//>           the leading  k by m  part of the array  A  must contain  the
-//>           matrix A.
-//> \endverbatim
-//>
-//> \param[in] LDA
-//> \verbatim
-//>          LDA is INTEGER
-//>           On entry, LDA specifies the first dimension of A as declared
-//>           in the calling (sub) program. When  TRANSA = 'N' or 'n' then
-//>           LDA must be at least  max( 1, m ), otherwise  LDA must be at
-//>           least  max( 1, k ).
-//> \endverbatim
-//>
-//> \param[in] B
-//> \verbatim
-//>          B is COMPLEX*16 array, dimension ( LDB, kb ), where kb is
-//>           n  when  TRANSB = 'N' or 'n',  and is  k  otherwise.
-//>           Before entry with  TRANSB = 'N' or 'n',  the leading  k by n
-//>           part of the array  B  must contain the matrix  B,  otherwise
-//>           the leading  n by k  part of the array  B  must contain  the
-//>           matrix B.
-//> \endverbatim
-//>
-//> \param[in] LDB
-//> \verbatim
-//>          LDB is INTEGER
-//>           On entry, LDB specifies the first dimension of B as declared
-//>           in the calling (sub) program. When  TRANSB = 'N' or 'n' then
-//>           LDB must be at least  max( 1, k ), otherwise  LDB must be at
-//>           least  max( 1, n ).
-//> \endverbatim
-//>
-//> \param[in] BETA
-//> \verbatim
-//>          BETA is COMPLEX*16
-//>           On entry,  BETA  specifies the scalar  beta.  When  BETA  is
-//>           supplied as zero then C need not be set on input.
-//> \endverbatim
-//>
-//> \param[in,out] C
-//> \verbatim
-//>          C is COMPLEX*16 array, dimension ( LDC, N )
-//>           Before entry, the leading  m by n  part of the array  C must
-//>           contain the matrix  C,  except when  beta  is zero, in which
-//>           case C need not be set on entry.
-//>           On exit, the array  C  is overwritten by the  m by n  matrix
-//>           ( alpha*op( A )*op( B ) + beta*C ).
-//> \endverbatim
-//>
-//> \param[in] LDC
-//> \verbatim
-//>          LDC is INTEGER
-//>           On entry, LDC specifies the first dimension of C as declared
-//>           in  the  calling  (sub)  program.   LDC  must  be  at  least
-//>           max( 1, m ).
-//> \endverbatim
-//
-// Authors:
-// ========
-//
-//> \author Univ. of Tennessee
-//> \author Univ. of California Berkeley
-//> \author Univ. of Colorado Denver
-//> \author NAG Ltd.
-//
-//> \date December 2016
-//
-//> \ingroup complex16_blas_level3
-//
-//> \par Further Details:
-// =====================
-//>
-//> \verbatim
-//>
-//>  Level 3 Blas routine.
-//>
-//>  -- Written on 8-February-1989.
-//>     Jack Dongarra, Argonne National Laboratory.
-//>     Iain Duff, AERE Harwell.
-//>     Jeremy Du Croz, Numerical Algorithms Group Ltd.
-//>     Sven Hammarling, Numerical Algorithms Group Ltd.
-//> \endverbatim
-//>
-// =====================================================================
-/* Subroutine */ int zgemm_(char *transa, char *transb, int *m, int *n, int *
-	k, doublecomplex *alpha, doublecomplex *a, int *lda, doublecomplex *b,
-	 int *ldb, doublecomplex *beta, doublecomplex *c__, int *ldc)
-{
-    // Table of constant values
-    doublecomplex c_b1 = {1.,0.};
-    doublecomplex c_b2 = {0.,0.};
-
-    // System generated locals
-    int a_dim1, a_offset, b_dim1, b_offset, c_dim1, c_offset, i__1, i__2,
-	    i__3, i__4, i__5, i__6;
-    doublecomplex z__1, z__2, z__3, z__4;
-
-    // Local variables
-    int i__, j, l, info;
-    int nota, notb;
-    doublecomplex temp;
-    int conja, conjb;
-    int ncola;
-    extern int lsame_(char *, char *);
-    int nrowa, nrowb;
-    extern /* Subroutine */ int xerbla_(char *, int *);
-
-    //
-    // -- Reference BLAS level3 routine (version 3.7.0) --
-    // -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
-    // -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-    //    December 2016
-    //
-    //    .. Scalar Arguments ..
-    //    ..
-    //    .. Array Arguments ..
-    //    ..
-    //
-    // =====================================================================
-    //
-    //    .. External Functions ..
-    //    ..
-    //    .. External Subroutines ..
-    //    ..
-    //    .. Intrinsic Functions ..
-    //    ..
-    //    .. Local Scalars ..
-    //    ..
-    //    .. Parameters ..
-    //    ..
-    //
-    //    Set  NOTA  and  NOTB  as  true if  A  and  B  respectively are not
-    //    conjugated or transposed, set  CONJA and CONJB  as true if  A  and
-    //    B  respectively are to be  transposed but  not conjugated  and set
-    //    NROWA, NCOLA and  NROWB  as the number of rows and  columns  of  A
-    //    and the number of rows of  B  respectively.
-    //
-    // Parameter adjustments
-    a_dim1 = *lda;
-    a_offset = 1 + a_dim1;
-    a -= a_offset;
-    b_dim1 = *ldb;
-    b_offset = 1 + b_dim1;
-    b -= b_offset;
-    c_dim1 = *ldc;
-    c_offset = 1 + c_dim1;
-    c__ -= c_offset;
-
-    // Function Body
-    nota = lsame_(transa, "N");
-    notb = lsame_(transb, "N");
-    conja = lsame_(transa, "C");
-    conjb = lsame_(transb, "C");
-    if (nota) {
-	nrowa = *m;
-	ncola = *k;
-    } else {
-	nrowa = *k;
-	ncola = *m;
-    }
-    if (notb) {
-	nrowb = *k;
-    } else {
-	nrowb = *n;
-    }
-    //
-    //    Test the input parameters.
-    //
-    info = 0;
-    if (! nota && ! conja && ! lsame_(transa, "T")) {
-	info = 1;
-    } else if (! notb && ! conjb && ! lsame_(transb, "T")) {
-	info = 2;
-    } else if (*m < 0) {
-	info = 3;
-    } else if (*n < 0) {
-	info = 4;
-    } else if (*k < 0) {
-	info = 5;
-    } else if (*lda < max(1,nrowa)) {
-	info = 8;
-    } else if (*ldb < max(1,nrowb)) {
-	info = 10;
-    } else if (*ldc < max(1,*m)) {
-	info = 13;
-    }
-    if (info != 0) {
-	xerbla_("ZGEMM ", &info);
-	return 0;
-    }
-    //
-    //    Quick return if possible.
-    //
-    if (*m == 0 || *n == 0 || (alpha->r == 0. && alpha->i == 0. || *k == 0) &&
-	     (beta->r == 1. && beta->i == 0.)) {
-	return 0;
-    }
-    //
-    //    And when  alpha.eq.zero.
-    //
-    if (alpha->r == 0. && alpha->i == 0.) {
-	if (beta->r == 0. && beta->i == 0.) {
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    i__3 = i__ + j * c_dim1;
-		    c__[i__3].r = 0., c__[i__3].i = 0.;
-// L10:
-		}
-// L20:
-	    }
-	} else {
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    i__3 = i__ + j * c_dim1;
-		    i__4 = i__ + j * c_dim1;
-		    z__1.r = beta->r * c__[i__4].r - beta->i * c__[i__4].i,
-			    z__1.i = beta->r * c__[i__4].i + beta->i * c__[
-			    i__4].r;
-		    c__[i__3].r = z__1.r, c__[i__3].i = z__1.i;
-// L30:
-		}
-// L40:
-	    }
-	}
-	return 0;
-    }
-    //
-    //    Start the operations.
-    //
-    if (notb) {
-	if (nota) {
-	    //
-	    //          Form  C := alpha*A*B + beta*C.
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		if (beta->r == 0. && beta->i == 0.) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			i__3 = i__ + j * c_dim1;
-			c__[i__3].r = 0., c__[i__3].i = 0.;
-// L50:
-		    }
-		} else if (beta->r != 1. || beta->i != 0.) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			i__3 = i__ + j * c_dim1;
-			i__4 = i__ + j * c_dim1;
-			z__1.r = beta->r * c__[i__4].r - beta->i * c__[i__4]
-				.i, z__1.i = beta->r * c__[i__4].i + beta->i *
-				 c__[i__4].r;
-			c__[i__3].r = z__1.r, c__[i__3].i = z__1.i;
-// L60:
-		    }
-		}
-		i__2 = *k;
-		for (l = 1; l <= i__2; ++l) {
-		    i__3 = l + j * b_dim1;
-		    z__1.r = alpha->r * b[i__3].r - alpha->i * b[i__3].i,
-			    z__1.i = alpha->r * b[i__3].i + alpha->i * b[i__3]
-			    .r;
-		    temp.r = z__1.r, temp.i = z__1.i;
-		    i__3 = *m;
-		    for (i__ = 1; i__ <= i__3; ++i__) {
-			i__4 = i__ + j * c_dim1;
-			i__5 = i__ + j * c_dim1;
-			i__6 = i__ + l * a_dim1;
-			z__2.r = temp.r * a[i__6].r - temp.i * a[i__6].i,
-				z__2.i = temp.r * a[i__6].i + temp.i * a[i__6]
-				.r;
-			z__1.r = c__[i__5].r + z__2.r, z__1.i = c__[i__5].i +
-				z__2.i;
-			c__[i__4].r = z__1.r, c__[i__4].i = z__1.i;
-// L70:
-		    }
-// L80:
-		}
-// L90:
-	    }
-	} else if (conja) {
-	    //
-	    //          Form  C := alpha*A**H*B + beta*C.
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    temp.r = 0., temp.i = 0.;
-		    i__3 = *k;
-		    for (l = 1; l <= i__3; ++l) {
-			d_cnjg(&z__3, &a[l + i__ * a_dim1]);
-			i__4 = l + j * b_dim1;
-			z__2.r = z__3.r * b[i__4].r - z__3.i * b[i__4].i,
-				z__2.i = z__3.r * b[i__4].i + z__3.i * b[i__4]
-				.r;
-			z__1.r = temp.r + z__2.r, z__1.i = temp.i + z__2.i;
-			temp.r = z__1.r, temp.i = z__1.i;
-// L100:
-		    }
-		    if (beta->r == 0. && beta->i == 0.) {
-			i__3 = i__ + j * c_dim1;
-			z__1.r = alpha->r * temp.r - alpha->i * temp.i,
-				z__1.i = alpha->r * temp.i + alpha->i *
-				temp.r;
-			c__[i__3].r = z__1.r, c__[i__3].i = z__1.i;
-		    } else {
-			i__3 = i__ + j * c_dim1;
-			z__2.r = alpha->r * temp.r - alpha->i * temp.i,
-				z__2.i = alpha->r * temp.i + alpha->i *
-				temp.r;
-			i__4 = i__ + j * c_dim1;
-			z__3.r = beta->r * c__[i__4].r - beta->i * c__[i__4]
-				.i, z__3.i = beta->r * c__[i__4].i + beta->i *
-				 c__[i__4].r;
-			z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i;
-			c__[i__3].r = z__1.r, c__[i__3].i = z__1.i;
-		    }
-// L110:
-		}
-// L120:
-	    }
-	} else {
-	    //
-	    //          Form  C := alpha*A**T*B + beta*C
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    temp.r = 0., temp.i = 0.;
-		    i__3 = *k;
-		    for (l = 1; l <= i__3; ++l) {
-			i__4 = l + i__ * a_dim1;
-			i__5 = l + j * b_dim1;
-			z__2.r = a[i__4].r * b[i__5].r - a[i__4].i * b[i__5]
-				.i, z__2.i = a[i__4].r * b[i__5].i + a[i__4]
-				.i * b[i__5].r;
-			z__1.r = temp.r + z__2.r, z__1.i = temp.i + z__2.i;
-			temp.r = z__1.r, temp.i = z__1.i;
-// L130:
-		    }
-		    if (beta->r == 0. && beta->i == 0.) {
-			i__3 = i__ + j * c_dim1;
-			z__1.r = alpha->r * temp.r - alpha->i * temp.i,
-				z__1.i = alpha->r * temp.i + alpha->i *
-				temp.r;
-			c__[i__3].r = z__1.r, c__[i__3].i = z__1.i;
-		    } else {
-			i__3 = i__ + j * c_dim1;
-			z__2.r = alpha->r * temp.r - alpha->i * temp.i,
-				z__2.i = alpha->r * temp.i + alpha->i *
-				temp.r;
-			i__4 = i__ + j * c_dim1;
-			z__3.r = beta->r * c__[i__4].r - beta->i * c__[i__4]
-				.i, z__3.i = beta->r * c__[i__4].i + beta->i *
-				 c__[i__4].r;
-			z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i;
-			c__[i__3].r = z__1.r, c__[i__3].i = z__1.i;
-		    }
-// L140:
-		}
-// L150:
-	    }
-	}
-    } else if (nota) {
-	if (conjb) {
-	    //
-	    //          Form  C := alpha*A*B**H + beta*C.
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		if (beta->r == 0. && beta->i == 0.) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			i__3 = i__ + j * c_dim1;
-			c__[i__3].r = 0., c__[i__3].i = 0.;
-// L160:
-		    }
-		} else if (beta->r != 1. || beta->i != 0.) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			i__3 = i__ + j * c_dim1;
-			i__4 = i__ + j * c_dim1;
-			z__1.r = beta->r * c__[i__4].r - beta->i * c__[i__4]
-				.i, z__1.i = beta->r * c__[i__4].i + beta->i *
-				 c__[i__4].r;
-			c__[i__3].r = z__1.r, c__[i__3].i = z__1.i;
-// L170:
-		    }
-		}
-		i__2 = *k;
-		for (l = 1; l <= i__2; ++l) {
-		    d_cnjg(&z__2, &b[j + l * b_dim1]);
-		    z__1.r = alpha->r * z__2.r - alpha->i * z__2.i, z__1.i =
-			    alpha->r * z__2.i + alpha->i * z__2.r;
-		    temp.r = z__1.r, temp.i = z__1.i;
-		    i__3 = *m;
-		    for (i__ = 1; i__ <= i__3; ++i__) {
-			i__4 = i__ + j * c_dim1;
-			i__5 = i__ + j * c_dim1;
-			i__6 = i__ + l * a_dim1;
-			z__2.r = temp.r * a[i__6].r - temp.i * a[i__6].i,
-				z__2.i = temp.r * a[i__6].i + temp.i * a[i__6]
-				.r;
-			z__1.r = c__[i__5].r + z__2.r, z__1.i = c__[i__5].i +
-				z__2.i;
-			c__[i__4].r = z__1.r, c__[i__4].i = z__1.i;
-// L180:
-		    }
-// L190:
-		}
-// L200:
-	    }
-	} else {
-	    //
-	    //          Form  C := alpha*A*B**T + beta*C
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		if (beta->r == 0. && beta->i == 0.) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			i__3 = i__ + j * c_dim1;
-			c__[i__3].r = 0., c__[i__3].i = 0.;
-// L210:
-		    }
-		} else if (beta->r != 1. || beta->i != 0.) {
-		    i__2 = *m;
-		    for (i__ = 1; i__ <= i__2; ++i__) {
-			i__3 = i__ + j * c_dim1;
-			i__4 = i__ + j * c_dim1;
-			z__1.r = beta->r * c__[i__4].r - beta->i * c__[i__4]
-				.i, z__1.i = beta->r * c__[i__4].i + beta->i *
-				 c__[i__4].r;
-			c__[i__3].r = z__1.r, c__[i__3].i = z__1.i;
-// L220:
-		    }
-		}
-		i__2 = *k;
-		for (l = 1; l <= i__2; ++l) {
-		    i__3 = j + l * b_dim1;
-		    z__1.r = alpha->r * b[i__3].r - alpha->i * b[i__3].i,
-			    z__1.i = alpha->r * b[i__3].i + alpha->i * b[i__3]
-			    .r;
-		    temp.r = z__1.r, temp.i = z__1.i;
-		    i__3 = *m;
-		    for (i__ = 1; i__ <= i__3; ++i__) {
-			i__4 = i__ + j * c_dim1;
-			i__5 = i__ + j * c_dim1;
-			i__6 = i__ + l * a_dim1;
-			z__2.r = temp.r * a[i__6].r - temp.i * a[i__6].i,
-				z__2.i = temp.r * a[i__6].i + temp.i * a[i__6]
-				.r;
-			z__1.r = c__[i__5].r + z__2.r, z__1.i = c__[i__5].i +
-				z__2.i;
-			c__[i__4].r = z__1.r, c__[i__4].i = z__1.i;
-// L230:
-		    }
-// L240:
-		}
-// L250:
-	    }
-	}
-    } else if (conja) {
-	if (conjb) {
-	    //
-	    //          Form  C := alpha*A**H*B**H + beta*C.
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    temp.r = 0., temp.i = 0.;
-		    i__3 = *k;
-		    for (l = 1; l <= i__3; ++l) {
-			d_cnjg(&z__3, &a[l + i__ * a_dim1]);
-			d_cnjg(&z__4, &b[j + l * b_dim1]);
-			z__2.r = z__3.r * z__4.r - z__3.i * z__4.i, z__2.i =
-				z__3.r * z__4.i + z__3.i * z__4.r;
-			z__1.r = temp.r + z__2.r, z__1.i = temp.i + z__2.i;
-			temp.r = z__1.r, temp.i = z__1.i;
-// L260:
-		    }
-		    if (beta->r == 0. && beta->i == 0.) {
-			i__3 = i__ + j * c_dim1;
-			z__1.r = alpha->r * temp.r - alpha->i * temp.i,
-				z__1.i = alpha->r * temp.i + alpha->i *
-				temp.r;
-			c__[i__3].r = z__1.r, c__[i__3].i = z__1.i;
-		    } else {
-			i__3 = i__ + j * c_dim1;
-			z__2.r = alpha->r * temp.r - alpha->i * temp.i,
-				z__2.i = alpha->r * temp.i + alpha->i *
-				temp.r;
-			i__4 = i__ + j * c_dim1;
-			z__3.r = beta->r * c__[i__4].r - beta->i * c__[i__4]
-				.i, z__3.i = beta->r * c__[i__4].i + beta->i *
-				 c__[i__4].r;
-			z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i;
-			c__[i__3].r = z__1.r, c__[i__3].i = z__1.i;
-		    }
-// L270:
-		}
-// L280:
-	    }
-	} else {
-	    //
-	    //          Form  C := alpha*A**H*B**T + beta*C
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    temp.r = 0., temp.i = 0.;
-		    i__3 = *k;
-		    for (l = 1; l <= i__3; ++l) {
-			d_cnjg(&z__3, &a[l + i__ * a_dim1]);
-			i__4 = j + l * b_dim1;
-			z__2.r = z__3.r * b[i__4].r - z__3.i * b[i__4].i,
-				z__2.i = z__3.r * b[i__4].i + z__3.i * b[i__4]
-				.r;
-			z__1.r = temp.r + z__2.r, z__1.i = temp.i + z__2.i;
-			temp.r = z__1.r, temp.i = z__1.i;
-// L290:
-		    }
-		    if (beta->r == 0. && beta->i == 0.) {
-			i__3 = i__ + j * c_dim1;
-			z__1.r = alpha->r * temp.r - alpha->i * temp.i,
-				z__1.i = alpha->r * temp.i + alpha->i *
-				temp.r;
-			c__[i__3].r = z__1.r, c__[i__3].i = z__1.i;
-		    } else {
-			i__3 = i__ + j * c_dim1;
-			z__2.r = alpha->r * temp.r - alpha->i * temp.i,
-				z__2.i = alpha->r * temp.i + alpha->i *
-				temp.r;
-			i__4 = i__ + j * c_dim1;
-			z__3.r = beta->r * c__[i__4].r - beta->i * c__[i__4]
-				.i, z__3.i = beta->r * c__[i__4].i + beta->i *
-				 c__[i__4].r;
-			z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i;
-			c__[i__3].r = z__1.r, c__[i__3].i = z__1.i;
-		    }
-// L300:
-		}
-// L310:
-	    }
-	}
-    } else {
-	if (conjb) {
-	    //
-	    //          Form  C := alpha*A**T*B**H + beta*C
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    temp.r = 0., temp.i = 0.;
-		    i__3 = *k;
-		    for (l = 1; l <= i__3; ++l) {
-			i__4 = l + i__ * a_dim1;
-			d_cnjg(&z__3, &b[j + l * b_dim1]);
-			z__2.r = a[i__4].r * z__3.r - a[i__4].i * z__3.i,
-				z__2.i = a[i__4].r * z__3.i + a[i__4].i *
-				z__3.r;
-			z__1.r = temp.r + z__2.r, z__1.i = temp.i + z__2.i;
-			temp.r = z__1.r, temp.i = z__1.i;
-// L320:
-		    }
-		    if (beta->r == 0. && beta->i == 0.) {
-			i__3 = i__ + j * c_dim1;
-			z__1.r = alpha->r * temp.r - alpha->i * temp.i,
-				z__1.i = alpha->r * temp.i + alpha->i *
-				temp.r;
-			c__[i__3].r = z__1.r, c__[i__3].i = z__1.i;
-		    } else {
-			i__3 = i__ + j * c_dim1;
-			z__2.r = alpha->r * temp.r - alpha->i * temp.i,
-				z__2.i = alpha->r * temp.i + alpha->i *
-				temp.r;
-			i__4 = i__ + j * c_dim1;
-			z__3.r = beta->r * c__[i__4].r - beta->i * c__[i__4]
-				.i, z__3.i = beta->r * c__[i__4].i + beta->i *
-				 c__[i__4].r;
-			z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i;
-			c__[i__3].r = z__1.r, c__[i__3].i = z__1.i;
-		    }
-// L330:
-		}
-// L340:
-	    }
-	} else {
-	    //
-	    //          Form  C := alpha*A**T*B**T + beta*C
-	    //
-	    i__1 = *n;
-	    for (j = 1; j <= i__1; ++j) {
-		i__2 = *m;
-		for (i__ = 1; i__ <= i__2; ++i__) {
-		    temp.r = 0., temp.i = 0.;
-		    i__3 = *k;
-		    for (l = 1; l <= i__3; ++l) {
-			i__4 = l + i__ * a_dim1;
-			i__5 = j + l * b_dim1;
-			z__2.r = a[i__4].r * b[i__5].r - a[i__4].i * b[i__5]
-				.i, z__2.i = a[i__4].r * b[i__5].i + a[i__4]
-				.i * b[i__5].r;
-			z__1.r = temp.r + z__2.r, z__1.i = temp.i + z__2.i;
-			temp.r = z__1.r, temp.i = z__1.i;
-// L350:
-		    }
-		    if (beta->r == 0. && beta->i == 0.) {
-			i__3 = i__ + j * c_dim1;
-			z__1.r = alpha->r * temp.r - alpha->i * temp.i,
-				z__1.i = alpha->r * temp.i + alpha->i *
-				temp.r;
-			c__[i__3].r = z__1.r, c__[i__3].i = z__1.i;
-		    } else {
-			i__3 = i__ + j * c_dim1;
-			z__2.r = alpha->r * temp.r - alpha->i * temp.i,
-				z__2.i = alpha->r * temp.i + alpha->i *
-				temp.r;
-			i__4 = i__ + j * c_dim1;
-			z__3.r = beta->r * c__[i__4].r - beta->i * c__[i__4]
-				.i, z__3.i = beta->r * c__[i__4].i + beta->i *
-				 c__[i__4].r;
-			z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i;
-			c__[i__3].r = z__1.r, c__[i__3].i = z__1.i;
-		    }
-// L360:
-		}
-// L370:
-	    }
-	}
-    }
-    return 0;
-    //
-    //    End of ZGEMM .
-    //
-} // zgemm_
-
@@ -1,8 +1,8 @@
-# Binaries branch name: ffmpeg/5.x_20260602
-# Binaries were created for OpenCV: a0a660fcb1e58a295e6caa6aee64ed4d369b0181
-ocv_update(FFMPEG_BINARIES_COMMIT "06dc20cad65dc7fcf784f70c95d46750520889a7")
-ocv_update(FFMPEG_FILE_HASH_BIN32 "9cef7a78b6f7ec8cf1a3935c058cfac5")
-ocv_update(FFMPEG_FILE_HASH_BIN64 "a821a1135251859655090c795af05789")
+# Binaries branch name: ffmpeg/4.x_20260715
+# Binaries were created for OpenCV: 6b640b424c516d27217700392a7ed6a362e040a5
+ocv_update(FFMPEG_BINARIES_COMMIT "bd9418020a5c342be979c56d6e6434261959d3af")
+ocv_update(FFMPEG_FILE_HASH_BIN32 "31968b434799d3dd56969b15aac3efa8")
+ocv_update(FFMPEG_FILE_HASH_BIN64 "84757ed0f16ddedce99227529d3165e9")
 ocv_update(FFMPEG_FILE_HASH_CMAKE "e09efc33312d1173be8a9446f3b088fe")

 function(download_win_ffmpeg script_var)
@@ -2,7 +2,7 @@ function(download_ippicv root_var)
  set(${root_var} "" PARENT_SCOPE)

  # Commit SHA in the opencv_3rdparty repo
-  set(IPPICV_COMMIT "406d398c436d0465c8e53dd432d9ecd9301d5f4a")
+  set(IPPICV_COMMIT "8338862a733cb3980d8b51d8e14917fe0e695f71")
  # Define actual ICV versions
  if(APPLE)
    set(IPPICV_COMMIT "0cc4aa06bf2bef4b05d237c69a5a96b9cd0cb85a")
@@ -14,8 +14,8 @@ function(download_ippicv root_var)
    set(OPENCV_ICV_PLATFORM "linux")
    set(OPENCV_ICV_PACKAGE_SUBDIR "ippicv_lnx")
    if(X86_64)
-      set(OPENCV_ICV_NAME "ippicv_2026.0.0_lnx_intel64_20260327_general.tgz")
-      set(OPENCV_ICV_HASH "9a3ee0c5c3c02102faa422d60bfd1f4a")
+      set(OPENCV_ICV_NAME "ippicv_2026.0.0_lnx_intel64_20260630_general.tgz")
+      set(OPENCV_ICV_HASH "a77e60db544e07a126ae98f5ffe83be1")
    else()
      if(ANDROID)
        set(IPPICV_COMMIT "c7c6d527dde5fee7cb914ee9e4e20f7436aab3a1")
@@ -31,8 +31,8 @@ function(download_ippicv root_var)
    set(OPENCV_ICV_PLATFORM "windows")
    set(OPENCV_ICV_PACKAGE_SUBDIR "ippicv_win")
    if(X86_64)
-      set(OPENCV_ICV_NAME "ippicv_2026.0.0_win_intel64_20260327_general.zip")
-      set(OPENCV_ICV_HASH "73bc67cd5e4c8da706fa88fe84630231")
+      set(OPENCV_ICV_NAME "ippicv_2026.0.0_win_intel64_20260630_general.zip")
+      set(OPENCV_ICV_HASH "d81c8b7d40da2867df82f0077a40afa1")
    else()
      set(IPPICV_COMMIT "7f55c0c26be418d494615afca15218566775c725")
      set(OPENCV_ICV_NAME "ippicv_2021.12.0_win_ia32_20240425_general.zip")
@@ -136,9 +136,6 @@ endif()

 set(JPEG_LIB_VERSION 70)

-# OpenCV
-set(JPEG_LIB_VERSION "${VERSION}-${JPEG_LIB_VERSION}" PARENT_SCOPE)
-
 set(THREAD_LOCAL "")  # WITH_TURBOJPEG is not used

 add_definitions(-DNO_GETENV -DNO_PUTENV)
@@ -0,0 +1,48 @@
+# ----------------------------------------------------------------------------
+#  CMake file for libjpeg. See root CMakeLists.txt
+#
+# ----------------------------------------------------------------------------
+project(${JPEG_LIBRARY})
+
+ocv_include_directories(${CMAKE_CURRENT_SOURCE_DIR})
+
+file(GLOB lib_srcs *.c)
+file(GLOB lib_hdrs *.h)
+
+if(ANDROID OR IOS OR APPLE)
+  ocv_list_filterout(lib_srcs jmemansi.c)
+else()
+  ocv_list_filterout(lib_srcs jmemnobs.c)
+endif()
+
+# ----------------------------------------------------------------------------------
+#         Define the library target:
+# ----------------------------------------------------------------------------------
+
+add_library(${JPEG_LIBRARY} STATIC ${OPENCV_3RDPARTY_EXCLUDE_FROM_ALL} ${lib_srcs} ${lib_hdrs})
+
+if(CV_GCC OR CV_CLANG)
+  set_source_files_properties(jcdctmgr.c PROPERTIES COMPILE_FLAGS "-O1")
+endif()
+
+ocv_warnings_disable(CMAKE_C_FLAGS -Wcast-align -Wshadow -Wunused -Wshift-negative-value -Wimplicit-fallthrough)
+ocv_warnings_disable(CMAKE_C_FLAGS -Wunused-parameter) # clang
+ocv_warnings_disable(CMAKE_C_FLAGS /wd4013 /wd4244 /wd4267) # vs2005
+
+set_target_properties(${JPEG_LIBRARY}
+  PROPERTIES OUTPUT_NAME ${JPEG_LIBRARY}
+  DEBUG_POSTFIX "${OPENCV_DEBUG_POSTFIX}"
+  COMPILE_PDB_NAME ${JPEG_LIBRARY}
+  COMPILE_PDB_NAME_DEBUG "${JPEG_LIBRARY}${OPENCV_DEBUG_POSTFIX}"
+  ARCHIVE_OUTPUT_DIRECTORY ${3P_LIBRARY_OUTPUT_PATH}
+  )
+
+if(ENABLE_SOLUTION_FOLDERS)
+  set_target_properties(${JPEG_LIBRARY} PROPERTIES FOLDER "3rdparty")
+endif()
+
+if(NOT BUILD_SHARED_LIBS)
+  ocv_install_target(${JPEG_LIBRARY} EXPORT OpenCVModules ARCHIVE DESTINATION ${OPENCV_3P_LIB_INSTALL_PATH} COMPONENT dev OPTIONAL)
+endif()
+
+ocv_install_3rdparty_licenses(libjpeg README)
@@ -0,0 +1,371 @@
+The Independent JPEG Group's JPEG software
+==========================================
+
+README for release 9d of 12-Jan-2020
+====================================
+
+This distribution contains the ninth public release of the Independent JPEG
+Group's free JPEG software.  You are welcome to redistribute this software and
+to use it for any purpose, subject to the conditions under LEGAL ISSUES, below.
+
+This software is the work of Tom Lane, Guido Vollbeding, Philip Gladstone,
+Bill Allombert, Jim Boucher, Lee Crocker, Bob Friesenhahn, Ben Jackson,
+John Korejwa, Julian Minguillon, Luis Ortiz, George Phillips, Davide Rossi,
+Ge' Weijers, and other members of the Independent JPEG Group.
+
+IJG is not affiliated with the ISO/IEC JTC1/SC29/WG1 standards committee
+(previously known as JPEG, together with ITU-T SG16).
+
+
+DOCUMENTATION ROADMAP
+=====================
+
+This file contains the following sections:
+
+OVERVIEW            General description of JPEG and the IJG software.
+LEGAL ISSUES        Copyright, lack of warranty, terms of distribution.
+REFERENCES          Where to learn more about JPEG.
+ARCHIVE LOCATIONS   Where to find newer versions of this software.
+ACKNOWLEDGMENTS     Special thanks.
+FILE FORMAT WARS    Software *not* to get.
+TO DO               Plans for future IJG releases.
+
+Other documentation files in the distribution are:
+
+User documentation:
+  install.txt       How to configure and install the IJG software.
+  usage.txt         Usage instructions for cjpeg, djpeg, jpegtran,
+                    rdjpgcom, and wrjpgcom.
+  *.1               Unix-style man pages for programs (same info as usage.txt).
+  wizard.txt        Advanced usage instructions for JPEG wizards only.
+  change.log        Version-to-version change highlights.
+Programmer and internal documentation:
+  libjpeg.txt       How to use the JPEG library in your own programs.
+  example.c         Sample code for calling the JPEG library.
+  structure.txt     Overview of the JPEG library's internal structure.
+  filelist.txt      Road map of IJG files.
+  coderules.txt     Coding style rules --- please read if you contribute code.
+
+Please read at least the files install.txt and usage.txt.  Some information
+can also be found in the JPEG FAQ (Frequently Asked Questions) article.  See
+ARCHIVE LOCATIONS below to find out where to obtain the FAQ article.
+
+If you want to understand how the JPEG code works, we suggest reading one or
+more of the REFERENCES, then looking at the documentation files (in roughly
+the order listed) before diving into the code.
+
+
+OVERVIEW
+========
+
+This package contains C software to implement JPEG image encoding, decoding,
+and transcoding.  JPEG (pronounced "jay-peg") is a standardized compression
+method for full-color and grayscale images.
+
+This software implements JPEG baseline, extended-sequential, and progressive
+compression processes.  Provision is made for supporting all variants of these
+processes, although some uncommon parameter settings aren't implemented yet.
+We have made no provision for supporting the hierarchical or lossless
+processes defined in the standard.
+
+We provide a set of library routines for reading and writing JPEG image files,
+plus two sample applications "cjpeg" and "djpeg", which use the library to
+perform conversion between JPEG and some other popular image file formats.
+The library is intended to be reused in other applications.
+
+In order to support file conversion and viewing software, we have included
+considerable functionality beyond the bare JPEG coding/decoding capability;
+for example, the color quantization modules are not strictly part of JPEG
+decoding, but they are essential for output to colormapped file formats or
+colormapped displays.  These extra functions can be compiled out of the
+library if not required for a particular application.
+
+We have also included "jpegtran", a utility for lossless transcoding between
+different JPEG processes, and "rdjpgcom" and "wrjpgcom", two simple
+applications for inserting and extracting textual comments in JFIF files.
+
+The emphasis in designing this software has been on achieving portability and
+flexibility, while also making it fast enough to be useful.  In particular,
+the software is not intended to be read as a tutorial on JPEG.  (See the
+REFERENCES section for introductory material.)  Rather, it is intended to
+be reliable, portable, industrial-strength code.  We do not claim to have
+achieved that goal in every aspect of the software, but we strive for it.
+
+We welcome the use of this software as a component of commercial products.
+No royalty is required, but we do ask for an acknowledgement in product
+documentation, as described under LEGAL ISSUES.
+
+
+LEGAL ISSUES
+============
+
+In plain English:
+
+1. We don't promise that this software works.  (But if you find any bugs,
+   please let us know!)
+2. You can use this software for whatever you want.  You don't have to pay us.
+3. You may not pretend that you wrote this software.  If you use it in a
+   program, you must acknowledge somewhere in your documentation that
+   you've used the IJG code.
+
+In legalese:
+
+The authors make NO WARRANTY or representation, either express or implied,
+with respect to this software, its quality, accuracy, merchantability, or
+fitness for a particular purpose.  This software is provided "AS IS", and you,
+its user, assume the entire risk as to its quality and accuracy.
+
+This software is copyright (C) 1991-2020, Thomas G. Lane, Guido Vollbeding.
+All Rights Reserved except as specified below.
+
+Permission is hereby granted to use, copy, modify, and distribute this
+software (or portions thereof) for any purpose, without fee, subject to these
+conditions:
+(1) If any part of the source code for this software is distributed, then this
+README file must be included, with this copyright and no-warranty notice
+unaltered; and any additions, deletions, or changes to the original files
+must be clearly indicated in accompanying documentation.
+(2) If only executable code is distributed, then the accompanying
+documentation must state that "this software is based in part on the work of
+the Independent JPEG Group".
+(3) Permission for use of this software is granted only if the user accepts
+full responsibility for any undesirable consequences; the authors accept
+NO LIABILITY for damages of any kind.
+
+These conditions apply to any software derived from or based on the IJG code,
+not just to the unmodified library.  If you use our work, you ought to
+acknowledge us.
+
+Permission is NOT granted for the use of any IJG author's name or company name
+in advertising or publicity relating to this software or products derived from
+it.  This software may be referred to only as "the Independent JPEG Group's
+software".
+
+We specifically permit and encourage the use of this software as the basis of
+commercial products, provided that all warranty or liability claims are
+assumed by the product vendor.
+
+
+The Unix configuration script "configure" was produced with GNU Autoconf.
+It is copyright by the Free Software Foundation but is freely distributable.
+The same holds for its supporting scripts (config.guess, config.sub,
+ltmain.sh).  Another support script, install-sh, is copyright by X Consortium
+but is also freely distributable.
+
+
+REFERENCES
+==========
+
+We recommend reading one or more of these references before trying to
+understand the innards of the JPEG software.
+
+The best short technical introduction to the JPEG compression algorithm is
+	Wallace, Gregory K.  "The JPEG Still Picture Compression Standard",
+	Communications of the ACM, April 1991 (vol. 34 no. 4), pp. 30-44.
+(Adjacent articles in that issue discuss MPEG motion picture compression,
+applications of JPEG, and related topics.)  If you don't have the CACM issue
+handy, a PDF file containing a revised version of Wallace's article is
+available at http://www.ijg.org/files/Wallace.JPEG.pdf.  The file (actually
+a preprint for an article that appeared in IEEE Trans. Consumer Electronics)
+omits the sample images that appeared in CACM, but it includes corrections
+and some added material.  Note: the Wallace article is copyright ACM and IEEE,
+and it may not be used for commercial purposes.
+
+A somewhat less technical, more leisurely introduction to JPEG can be found in
+"The Data Compression Book" by Mark Nelson and Jean-loup Gailly, published by
+M&T Books (New York), 2nd ed. 1996, ISBN 1-55851-434-1.  This book provides
+good explanations and example C code for a multitude of compression methods
+including JPEG.  It is an excellent source if you are comfortable reading C
+code but don't know much about data compression in general.  The book's JPEG
+sample code is far from industrial-strength, but when you are ready to look
+at a full implementation, you've got one here...
+
+The best currently available description of JPEG is the textbook "JPEG Still
+Image Data Compression Standard" by William B. Pennebaker and Joan L.
+Mitchell, published by Van Nostrand Reinhold, 1993, ISBN 0-442-01272-1.
+Price US$59.95, 638 pp.  The book includes the complete text of the ISO JPEG
+standards (DIS 10918-1 and draft DIS 10918-2).
+Although this is by far the most detailed and comprehensive exposition of
+JPEG publicly available, we point out that it is still missing an explanation
+of the most essential properties and algorithms of the underlying DCT
+technology.
+If you think that you know about DCT-based JPEG after reading this book,
+then you are in delusion.  The real fundamentals and corresponding potential
+of DCT-based JPEG are not publicly known so far, and that is the reason for
+all the mistaken developments taking place in the image coding domain.
+
+The original JPEG standard is divided into two parts, Part 1 being the actual
+specification, while Part 2 covers compliance testing methods.  Part 1 is
+titled "Digital Compression and Coding of Continuous-tone Still Images,
+Part 1: Requirements and guidelines" and has document numbers ISO/IEC IS
+10918-1, ITU-T T.81.  Part 2 is titled "Digital Compression and Coding of
+Continuous-tone Still Images, Part 2: Compliance testing" and has document
+numbers ISO/IEC IS 10918-2, ITU-T T.83.
+IJG JPEG 8 introduced an implementation of the JPEG SmartScale extension
+which is specified in two documents:  A contributed document at ITU and ISO
+with title "ITU-T JPEG-Plus Proposal for Extending ITU-T T.81 for Advanced
+Image Coding", April 2006, Geneva, Switzerland.  The latest version of this
+document is Revision 3.  And a contributed document ISO/IEC JTC1/SC29/WG1 N
+5799 with title "Evolution of JPEG", June/July 2011, Berlin, Germany.
+IJG JPEG 9 introduces a reversible color transform for improved lossless
+compression which is described in a contributed document ISO/IEC JTC1/SC29/
+WG1 N 6080 with title "JPEG 9 Lossless Coding", June/July 2012, Paris,
+France.
+
+The JPEG standard does not specify all details of an interchangeable file
+format.  For the omitted details we follow the "JFIF" conventions, version 2.
+JFIF version 1 has been adopted as Recommendation ITU-T T.871 (05/2011) :
+Information technology - Digital compression and coding of continuous-tone
+still images: JPEG File Interchange Format (JFIF).  It is available as a
+free download in PDF file format from http://www.itu.int/rec/T-REC-T.871.
+A PDF file of the older JFIF document is available at
+http://www.w3.org/Graphics/JPEG/jfif3.pdf.
+
+The TIFF 6.0 file format specification can be obtained by FTP from
+ftp://ftp.sgi.com/graphics/tiff/TIFF6.ps.gz.  The JPEG incorporation scheme
+found in the TIFF 6.0 spec of 3-June-92 has a number of serious problems.
+IJG does not recommend use of the TIFF 6.0 design (TIFF Compression tag 6).
+Instead, we recommend the JPEG design proposed by TIFF Technical Note #2
+(Compression tag 7).  Copies of this Note can be obtained from
+http://www.ijg.org/files/.  It is expected that the next revision
+of the TIFF spec will replace the 6.0 JPEG design with the Note's design.
+Although IJG's own code does not support TIFF/JPEG, the free libtiff library
+uses our library to implement TIFF/JPEG per the Note.
+
+
+ARCHIVE LOCATIONS
+=================
+
+The "official" archive site for this software is www.ijg.org.
+The most recent released version can always be found there in
+directory "files".  This particular version will be archived as
+http://www.ijg.org/files/jpegsrc.v9d.tar.gz, and in Windows-compatible
+"zip" archive format as http://www.ijg.org/files/jpegsr9d.zip.
+
+The JPEG FAQ (Frequently Asked Questions) article is a source of some
+general information about JPEG.
+It is available on the World Wide Web at http://www.faqs.org/faqs/jpeg-faq/
+and other news.answers archive sites, including the official news.answers
+archive at rtfm.mit.edu: ftp://rtfm.mit.edu/pub/usenet/news.answers/jpeg-faq/.
+If you don't have Web or FTP access, send e-mail to mail-server@rtfm.mit.edu
+with body
+	send usenet/news.answers/jpeg-faq/part1
+	send usenet/news.answers/jpeg-faq/part2
+
+
+ACKNOWLEDGMENTS
+===============
+
+Thank to Juergen Bruder for providing me with a copy of the common DCT
+algorithm article, only to find out that I had come to the same result
+in a more direct and comprehensible way with a more generative approach.
+
+Thank to Istvan Sebestyen and Joan L. Mitchell for inviting me to the
+ITU JPEG (Study Group 16) meeting in Geneva, Switzerland.
+
+Thank to Thomas Wiegand and Gary Sullivan for inviting me to the
+Joint Video Team (MPEG & ITU) meeting in Geneva, Switzerland.
+
+Thank to Thomas Richter and Daniel Lee for inviting me to the
+ISO/IEC JTC1/SC29/WG1 (previously known as JPEG, together with ITU-T SG16)
+meeting in Berlin, Germany.
+
+Thank to John Korejwa and Massimo Ballerini for inviting me to
+fruitful consultations in Boston, MA and Milan, Italy.
+
+Thank to Hendrik Elstner, Roland Fassauer, Simone Zuck, Guenther
+Maier-Gerber, Walter Stoeber, Fred Schmitz, and Norbert Braunagel
+for corresponding business development.
+
+Thank to Nico Zschach and Dirk Stelling of the technical support team
+at the Digital Images company in Halle for providing me with extra
+equipment for configuration tests.
+
+Thank to Richard F. Lyon (then of Foveon Inc.) for fruitful
+communication about JPEG configuration in Sigma Photo Pro software.
+
+Thank to Andrew Finkenstadt for hosting the ijg.org site.
+
+Thank to Thomas G. Lane for the original design and development of
+this singular software package.
+
+Thank to Lars Goehler, Andreas Heinecke, Sebastian Fuss, Yvonne Roebert,
+Andrej Werner, and Ulf-Dietrich Braumann for support and public relations.
+
+
+FILE FORMAT WARS
+================
+
+The ISO/IEC JTC1/SC29/WG1 standards committee (previously known as JPEG,
+together with ITU-T SG16) currently promotes different formats containing
+the name "JPEG" which is misleading because these formats are incompatible
+with original DCT-based JPEG and are based on faulty technologies.
+IJG therefore does not and will not support such momentary mistakes
+(see REFERENCES).
+There exist also distributions under the name "OpenJPEG" promoting such
+kind of formats which is misleading because they don't support original
+JPEG images.
+We have no sympathy for the promotion of inferior formats.  Indeed, one of
+the original reasons for developing this free software was to help force
+convergence on common, interoperable format standards for JPEG files.
+Don't use an incompatible file format!
+(In any case, our decoder will remain capable of reading existing JPEG
+image files indefinitely.)
+
+The ISO committee pretends to be "responsible for the popular JPEG" in their
+public reports which is not true because they don't respond to actual
+requirements for the maintenance of the original JPEG specification.
+Furthermore, the ISO committee pretends to "ensure interoperability" with
+their standards which is not true because their "standards" support only
+application-specific and proprietary use cases and contain mathematically
+incorrect code.
+
+There are currently different distributions in circulation containing the
+name "libjpeg" which is misleading because they don't have the features and
+are incompatible with formats supported by actual IJG libjpeg distributions.
+One of those fakes is released by members of the ISO committee and just uses
+the name of libjpeg for misdirection of people, similar to the abuse of the
+name JPEG as described above, while having nothing in common with actual IJG
+libjpeg distributions and containing mathematically incorrect code.
+The other one claims to be a "derivative" or "fork" of the original libjpeg,
+but violates the license conditions as described under LEGAL ISSUES above
+and violates basic C programming properties.
+We have no sympathy for the release of misleading, incorrect and illegal
+distributions derived from obsolete code bases.
+Don't use an obsolete code base!
+
+According to the UCC (Uniform Commercial Code) law, IJG has the lawful and
+legal right to foreclose on certain standardization bodies and other
+institutions or corporations that knowingly perform substantial and
+systematic deceptive acts and practices, fraud, theft, and damaging of the
+value of the people of this planet without their knowing, willing and
+intentional consent.
+The titles, ownership, and rights of these institutions and all their assets
+are now duly secured and held in trust for the free people of this planet.
+People of the planet, on every country, may have a financial interest in
+the assets of these former principals, agents, and beneficiaries of the
+foreclosed institutions and corporations.
+IJG asserts what is: that each man, woman, and child has unalienable value
+and rights granted and deposited in them by the Creator and not any one of
+the people is subordinate to any artificial principality, corporate fiction
+or the special interest of another without their appropriate knowing,
+willing and intentional consent made by contract or accommodation agreement.
+IJG expresses that which already was.
+The people have already determined and demanded that public administration
+entities, national governments, and their supporting judicial systems must
+be fully transparent, accountable, and liable.
+IJG has secured the value for all concerned free people of the planet.
+
+A partial list of foreclosed institutions and corporations ("Hall of Shame")
+is currently prepared and will be published later.
+
+
+TO DO
+=====
+
+Version 9 is the second release of a new generation JPEG standard
+to overcome the limitations of the original JPEG specification,
+and is the first true source reference JPEG codec.
+More features are being prepared for coming releases...
+
+Please send bug reports, offers of help, etc. to jpeg-info@jpegclub.org.
@@ -0,0 +1,502 @@
+CHANGE LOG for Independent JPEG Group's JPEG software
+
+
+Version 9d  12-Jan-2020
+-----------------------
+
+Optimize the optimal Huffman code table generation to produce
+slightly smaller files.  Thank to John Korejwa for suggestion.
+Note: Requires rebuild of testimgp.jpg.
+
+Decoding Huffman: Use default tables if tables are not defined.
+Thank to Simone Azzalin for report (Motion JPEG),
+and to Martin Strunz for hint.
+
+Add sanity check in optimal Huffman code table generation.
+Thank to Adam Farley for suggestion.
+
+rdtarga.c: use read_byte(), with EOF check, instead of getc()
+in read_*_pixel().
+Thank to Chijin Zhou for cjpeg potential vulnerability report.
+
+jmemnobs.c: respect the max_memory_to_use setting in
+jpeg_mem_available() computation.  Thank to Sheng Shu and
+Dongdong She for djpeg potential vulnerability report.
+
+jdarith.c, jdhuff.c: avoid left shift of negative value
+compiler warning in decode_mcu_AC_refine().
+Thank to Indu Bhagat for suggestion.
+
+Add x64 (64-bit) platform support, avoid compiler warnings.
+Thank to Jonathan Potter, Feiyun Wang, and Sheng Shu for suggestion.
+
+Adjust libjpeg version specification for pkg-config file.
+Thank to Chen Chen for suggestion.
+
+Restore GIF read and write support from libjpeg version 6a.
+Thank to Wolfgang Werner (W.W.) Heinz for suggestion.
+
+Improve consistency in raw (downsampled) image data processing mode.
+Thank to Zhongyuan Zhou for hint.
+
+Avoid out of bounds array read (AC derived table pointers)
+in start pass in jdhuff.c.  Thank to Peng Li for report.
+
+Improve code sanity (jdhuff.c).
+Thank to Reza Mirzazade farkhani for reports.
+
+Add jpegtran -drop option; add options to the crop extension and wipe
+to fill the extra area with content from the source image region,
+instead of gray out.
+
+
+Version 9c  14-Jan-2018
+-----------------------
+
+jpegtran: add an option to the -wipe switch to fill the region
+with the average of adjacent blocks, instead of gray out.
+Thank to Caitlyn Feddock and Maddie Ziegler for inspiration.
+
+Make range extension bits adjustable (in jpegint.h).
+Thank to Robin Watts for suggestion.
+
+Provide macros for fflush() and ferror() in jinclude.h in order
+to facilitate adaption by applications using an own FILE class.
+Thank to Gerhard Huber for suggestion.
+
+Add libjpeg pkg-config file.  Thank to Mark Lavi, Vincent Torri,
+Patrick McMunn, and Huw Davies for suggestion.
+
+Add sanity checks in cjpeg image reader modules.
+Thank to Bingchang, Liu for reports.
+
+
+Version 9b  17-Jan-2016
+-----------------------
+
+Improvements and optimizations in DCT and color calculations.
+Normalize range limit array composition and access pattern.
+Thank to Sia Furler and Maddie Ziegler for inspiration.
+
+Use merged upsample with scaled DCT sizes larger than 8.
+Thank to Taylor Hatala for inspiration.
+
+Check for excessive comment lengths in argument parsing in wrjpgcom.c.
+Thank to Julian Cohen for hint.
+
+Add makefile.b32 for use with Borland C++ 32-bit (bcc32).
+Thank to Joe Slater for contribution.
+
+Document 'f' specifier for jpegtran -crop specification.
+Thank to Michele Martone for suggestion.
+
+Use defined value from header instead of hardwired number in rdswitch.c.
+Thank to Robert Sprowson for hint.
+
+
+Version 9a  19-Jan-2014
+-----------------------
+
+Add support for wide gamut color spaces (JFIF version 2).
+Improve clarity and accuracy in color conversion modules.
+Note: Requires rebuild of test images.
+
+Extend the bit depth support to all values from 8 to 12
+(BITS_IN_JSAMPLE configuration option in jmorecfg.h).
+jpegtran now supports N bits sample data precision with all N from 8 to 12
+in a single instance.  Thank to Roland Fassauer for inspiration.
+
+Try to resolve issues with new boolean type definition.
+Thank also to v4hn for suggestion.
+
+Enable option to use default Huffman tables for lossless compression
+(for hardware solution), and in this case improve lossless RGB compression
+with reversible color transform.  Thank to Benny Alexandar for hint.
+
+Extend the entropy decoding structure, so that extraneous bytes between
+compressed scan data and following marker can be reported correctly.
+Thank to Nigel Tao for hint.
+
+Add jpegtran -wipe option and extension for -crop.
+Thank to Andrew Senior, David Clunie, and Josef Schmid for suggestion.
+
+
+Version 9  13-Jan-2013
+----------------------
+
+Add cjpeg -rgb1 option to create an RGB JPEG file, and insert
+a simple reversible color transform into the processing which
+significantly improves the compression.
+The recommended command for lossless coding of RGB images is now
+cjpeg -rgb1 -block 1 -arithmetic.
+As said, this option improves the compression significantly, but
+the files are not compatible with JPEG decoders prior to IJG v9
+due to the included color transform.
+The used color transform and marker signaling is compatible with
+other JPEG standards (e.g., JPEG-LS part 2).
+
+Remove the automatic de-ANSI-fication support (Automake 1.12).
+Thank also to Nitin A Kamble for suggestion.
+
+Add remark for jpeg_mem_dest() in jdatadst.c.
+Thank to Elie-Gregoire Khoury for the hint.
+
+Support files with invalid component identifiers (created
+by Adobe PDF).  Thank to Robin Watts for the suggestion.
+
+Adapt full buffer case in jcmainct.c for use with scaled DCT.
+Thank to Sergii Biloshytskyi for the suggestion.
+
+Add type identifier for declaration of noreturn functions.
+Thank to Brett L. Moore for the suggestion.
+
+Correct argument type in format string, avoid compiler warnings.
+Thank to Vincent Torri for hint.
+
+Add missing #include directives in configuration checks, avoid
+configuration errors.  Thank to John Spencer for the hint.
+
+
+Version 8d  15-Jan-2012
+-----------------------
+
+Add cjpeg -rgb option to create RGB JPEG files.
+Using this switch suppresses the conversion from RGB
+colorspace input to the default YCbCr JPEG colorspace.
+This feature allows true lossless JPEG coding of RGB color images.
+The recommended command for this purpose is currently
+cjpeg -rgb -block 1 -arithmetic.
+SmartScale capable decoder (introduced with IJG JPEG 8) required.
+Thank to Michael Koch for the initial suggestion.
+
+Add option to disable the region adjustment in the transupp crop code.
+Thank to Jeffrey Friedl for the suggestion.
+
+Thank to Richard Jones and Edd Dawson for various minor corrections.
+
+Thank to Akim Demaille for configure.ac cleanup.
+
+
+Version 8c  16-Jan-2011
+-----------------------
+
+Add option to compression library and cjpeg (-block N) to use
+different DCT block size.
+All N from 1 to 16 are possible.  Default is 8 (baseline format).
+Larger values produce higher compression,
+smaller values produce higher quality.
+SmartScale capable decoder (introduced with IJG JPEG 8) required.
+
+
+Version 8b  16-May-2010
+-----------------------
+
+Repair problem in new memory source manager with corrupt JPEG data.
+Thank to Ted Campbell and Samuel Chun for the report.
+
+Repair problem in Makefile.am test target.
+Thank to anonymous user for the report.
+
+Support MinGW installation with automatic configure.
+Thank to Volker Grabsch for the suggestion.
+
+
+Version 8a  28-Feb-2010
+-----------------------
+
+Writing tables-only datastreams via jpeg_write_tables works again.
+
+Support 32-bit BMPs (RGB image with Alpha channel) for read in cjpeg.
+Thank to Brett Blackham for the suggestion.
+
+Improve accuracy in floating point IDCT calculation.
+Thank to Robert Hooke for the hint.
+
+
+Version 8  10-Jan-2010
+----------------------
+
+jpegtran now supports the same -scale option as djpeg for "lossless" resize.
+An implementation of the JPEG SmartScale extension is required for this
+feature.  A (draft) specification of the JPEG SmartScale extension is
+available as a contributed document at ITU and ISO.  Revision 2 or later
+of the document is required (latest document version is Revision 3).
+The SmartScale extension will enable more features beside lossless resize
+in future implementations, as described in the document (new compression
+options).
+
+Add sanity check in BMP reader module to avoid cjpeg crash for empty input
+image (thank to Isaev Ildar of ISP RAS, Moscow, RU for reporting this error).
+
+Add data source and destination managers for read from and write to
+memory buffers.  New API functions jpeg_mem_src and jpeg_mem_dest.
+Thank to Roberto Boni from Italy for the suggestion.
+
+
+Version 7  27-Jun-2009
+----------------------
+
+New scaled DCTs implemented.
+djpeg now supports scalings N/8 with all N from 1 to 16.
+cjpeg now supports scalings 8/N with all N from 1 to 16.
+Scaled DCTs with size larger than 8 are now also used for resolving the
+common 2x2 chroma subsampling case without additional spatial resampling.
+Separate spatial resampling for those kind of files is now only necessary
+for N>8 scaling cases.
+Furthermore, separate scaled DCT functions are provided for direct resolving
+of the common asymmetric subsampling cases (2x1 and 1x2) without additional
+spatial resampling.
+
+cjpeg -quality option has been extended for support of separate quality
+settings for luminance and chrominance (or in general, for every provided
+quantization table slot).
+New API function jpeg_default_qtables() and q_scale_factor array in library.
+
+Added -nosmooth option to cjpeg, complementary to djpeg.
+New variable "do_fancy_downsampling" in library, complement to fancy
+upsampling.  Fancy upsampling now uses direct DCT scaling with sizes
+larger than 8.  The old method is not reversible and has been removed.
+
+Support arithmetic entropy encoding and decoding.
+Added files jaricom.c, jcarith.c, jdarith.c.
+
+Straighten the file structure:
+Removed files jidctred.c, jcphuff.c, jchuff.h, jdphuff.c, jdhuff.h.
+
+jpegtran has a new "lossless" cropping feature.
+
+Implement -perfect option in jpegtran, new API function
+jtransform_perfect_transform() in transupp. (DP 204_perfect.dpatch)
+
+Better error messages for jpegtran fopen failure.
+(DP 203_jpegtran_errmsg.dpatch)
+
+Fix byte order issue with 16bit PPM/PGM files in rdppm.c/wrppm.c:
+according to Netpbm, the de facto standard implementation of the PNM formats,
+the most significant byte is first. (DP 203_rdppm.dpatch)
+
+Add -raw option to rdjpgcom not to mangle the output.
+(DP 205_rdjpgcom_raw.dpatch)
+
+Make rdjpgcom locale aware. (DP 201_rdjpgcom_locale.dpatch)
+
+Add extern "C" to jpeglib.h.
+This avoids the need to put extern "C" { ... } around #include "jpeglib.h"
+in your C++ application.  Defining the symbol DONT_USE_EXTERN_C in the
+configuration prevents this. (DP 202_jpeglib.h_c++.dpatch)
+
+
+Version 6b  27-Mar-1998
+-----------------------
+
+jpegtran has new features for lossless image transformations (rotation
+and flipping) as well as "lossless" reduction to grayscale.
+
+jpegtran now copies comments by default; it has a -copy switch to enable
+copying all APPn blocks as well, or to suppress comments.  (Formerly it
+always suppressed comments and APPn blocks.)  jpegtran now also preserves
+JFIF version and resolution information.
+
+New decompressor library feature: COM and APPn markers found in the input
+file can be saved in memory for later use by the application.  (Before,
+you had to code this up yourself with a custom marker processor.)
+
+There is an unused field "void * client_data" now in compress and decompress
+parameter structs; this may be useful in some applications.
+
+JFIF version number information is now saved by the decoder and accepted by
+the encoder.  jpegtran uses this to copy the source file's version number,
+to ensure "jpegtran -copy all" won't create bogus files that contain JFXX
+extensions but claim to be version 1.01.  Applications that generate their
+own JFXX extension markers also (finally) have a supported way to cause the
+encoder to emit JFIF version number 1.02.
+
+djpeg's trace mode reports JFIF 1.02 thumbnail images as such, rather
+than as unknown APP0 markers.
+
+In -verbose mode, djpeg and rdjpgcom will try to print the contents of
+APP12 markers as text.  Some digital cameras store useful text information
+in APP12 markers.
+
+Handling of truncated data streams is more robust: blocks beyond the one in
+which the error occurs will be output as uniform gray, or left unchanged
+if decoding a progressive JPEG.  The appearance no longer depends on the
+Huffman tables being used.
+
+Huffman tables are checked for validity much more carefully than before.
+
+To avoid the Unisys LZW patent, djpeg's GIF output capability has been
+changed to produce "uncompressed GIFs", and cjpeg's GIF input capability
+has been removed altogether.  We're not happy about it either, but there
+seems to be no good alternative.
+
+The configure script now supports building libjpeg as a shared library
+on many flavors of Unix (all the ones that GNU libtool knows how to
+build shared libraries for).  Use "./configure --enable-shared" to
+try this out.
+
+New jconfig file and makefiles for Microsoft Visual C++ and Developer Studio.
+Also, a jconfig file and a build script for Metrowerks CodeWarrior
+on Apple Macintosh.  makefile.dj has been updated for DJGPP v2, and there
+are miscellaneous other minor improvements in the makefiles.
+
+jmemmac.c now knows how to create temporary files following Mac System 7
+conventions.
+
+djpeg's -map switch is now able to read raw-format PPM files reliably.
+
+cjpeg -progressive -restart no longer generates any unnecessary DRI markers.
+
+Multiple calls to jpeg_simple_progression for a single JPEG object
+no longer leak memory.
+
+
+Version 6a  7-Feb-96
+--------------------
+
+Library initialization sequence modified to detect version mismatches
+and struct field packing mismatches between library and calling application.
+This change requires applications to be recompiled, but does not require
+any application source code change.
+
+All routine declarations changed to the style "GLOBAL(type) name ...",
+that is, GLOBAL, LOCAL, METHODDEF, EXTERN are now macros taking the
+routine's return type as an argument.  This makes it possible to add
+Microsoft-style linkage keywords to all the routines by changing just
+these macros.  Note that any application code that was using these macros
+will have to be changed.
+
+DCT coefficient quantization tables are now stored in normal array order
+rather than zigzag order.  Application code that calls jpeg_add_quant_table,
+or otherwise manipulates quantization tables directly, will need to be
+changed.  If you need to make such code work with either older or newer
+versions of the library, a test like "#if JPEG_LIB_VERSION >= 61" is
+recommended.
+
+djpeg's trace capability now dumps DQT tables in natural order, not zigzag
+order.  This allows the trace output to be made into a "-qtables" file
+more easily.
+
+New system-dependent memory manager module for use on Apple Macintosh.
+
+Fix bug in cjpeg's -smooth option: last one or two scanlines would be
+duplicates of the prior line unless the image height mod 16 was 1 or 2.
+
+Repair minor problems in VMS, BCC, MC6 makefiles.
+
+New configure script based on latest GNU Autoconf.
+
+Correct the list of include files needed by MetroWerks C for ccommand().
+
+Numerous small documentation updates.
+
+
+Version 6  2-Aug-95
+-------------------
+
+Progressive JPEG support: library can read and write full progressive JPEG
+files.  A "buffered image" mode supports incremental decoding for on-the-fly
+display of progressive images.  Simply recompiling an existing IJG-v5-based
+decoder with v6 should allow it to read progressive files, though of course
+without any special progressive display.
+
+New "jpegtran" application performs lossless transcoding between different
+JPEG formats; primarily, it can be used to convert baseline to progressive
+JPEG and vice versa.  In support of jpegtran, the library now allows lossless
+reading and writing of JPEG files as DCT coefficient arrays.  This ability
+may be of use in other applications.
+
+Notes for programmers:
+* We changed jpeg_start_decompress() to be able to suspend; this makes all
+decoding modes available to suspending-input applications.  However,
+existing applications that use suspending input will need to be changed
+to check the return value from jpeg_start_decompress().  You don't need to
+do anything if you don't use a suspending data source.
+* We changed the interface to the virtual array routines: access_virt_array
+routines now take a count of the number of rows to access this time.  The
+last parameter to request_virt_array routines is now interpreted as the
+maximum number of rows that may be accessed at once, but not necessarily
+the height of every access.
+
+
+Version 5b  15-Mar-95
+---------------------
+
+Correct bugs with grayscale images having v_samp_factor > 1.
+
+jpeg_write_raw_data() now supports output suspension.
+
+Correct bugs in "configure" script for case of compiling in
+a directory other than the one containing the source files.
+
+Repair bug in jquant1.c: sometimes didn't use as many colors as it could.
+
+Borland C makefile and jconfig file work under either MS-DOS or OS/2.
+
+Miscellaneous improvements to documentation.
+
+
+Version 5a  7-Dec-94
+--------------------
+
+Changed color conversion roundoff behavior so that grayscale values are
+represented exactly.  (This causes test image files to change.)
+
+Make ordered dither use 16x16 instead of 4x4 pattern for a small quality
+improvement.
+
+New configure script based on latest GNU Autoconf.
+Fix configure script to handle CFLAGS correctly.
+Rename *.auto files to *.cfg, so that configure script still works if
+file names have been truncated for DOS.
+
+Fix bug in rdbmp.c: didn't allow for extra data between header and image.
+
+Modify rdppm.c/wrppm.c to handle 2-byte raw PPM/PGM formats for 12-bit data.
+
+Fix several bugs in rdrle.c.
+
+NEED_SHORT_EXTERNAL_NAMES option was broken.
+
+Revise jerror.h/jerror.c for more flexibility in message table.
+
+Repair oversight in jmemname.c NO_MKTEMP case: file could be there
+but unreadable.
+
+
+Version 5  24-Sep-94
+--------------------
+
+Version 5 represents a nearly complete redesign and rewrite of the IJG
+software.  Major user-visible changes include:
+  * Automatic configuration simplifies installation for most Unix systems.
+  * A range of speed vs. image quality tradeoffs are supported.
+    This includes resizing of an image during decompression: scaling down
+    by a factor of 1/2, 1/4, or 1/8 is handled very efficiently.
+  * New programs rdjpgcom and wrjpgcom allow insertion and extraction
+    of text comments in a JPEG file.
+
+The application programmer's interface to the library has changed completely.
+Notable improvements include:
+  * We have eliminated the use of callback routines for handling the
+    uncompressed image data.  The application now sees the library as a
+    set of routines that it calls to read or write image data on a
+    scanline-by-scanline basis.
+  * The application image data is represented in a conventional interleaved-
+    pixel format, rather than as a separate array for each color channel.
+    This can save a copying step in many programs.
+  * The handling of compressed data has been cleaned up: the application can
+    supply routines to source or sink the compressed data.  It is possible to
+    suspend processing on source/sink buffer overrun, although this is not
+    supported in all operating modes.
+  * All static state has been eliminated from the library, so that multiple
+    instances of compression or decompression can be active concurrently.
+  * JPEG abbreviated datastream formats are supported, ie, quantization and
+    Huffman tables can be stored separately from the image data.
+  * And not only that, but the documentation of the library has improved
+    considerably!
+
+
+The last widely used release before the version 5 rewrite was version 4A of
+18-Feb-93.  Change logs before that point have been discarded, since they
+are not of much interest after the rewrite.
@@ -0,0 +1,153 @@
+/*
+ * jaricom.c
+ *
+ * Developed 1997-2011 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains probability estimation tables for common use in
+ * arithmetic entropy encoding and decoding routines.
+ *
+ * This data represents Table D.3 in the JPEG spec (D.2 in the draft),
+ * ISO/IEC IS 10918-1 and CCITT Recommendation ITU-T T.81, and Table 24
+ * in the JBIG spec, ISO/IEC IS 11544 and CCITT Recommendation ITU-T T.82.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+/* The following #define specifies the packing of the four components
+ * into the compact INT32 representation.
+ * Note that this formula must match the actual arithmetic encoder
+ * and decoder implementation.  The implementation has to be changed
+ * if this formula is changed.
+ * The current organization is leaned on Markus Kuhn's JBIG
+ * implementation (jbig_tab.c).
+ */
+
+#define V(i,a,b,c,d) (((INT32)a << 16) | ((INT32)c << 8) | ((INT32)d << 7) | b)
+
+const INT32 jpeg_aritab[113+1] = {
+/*
+ * Index, Qe_Value, Next_Index_LPS, Next_Index_MPS, Switch_MPS
+ */
+  V(   0, 0x5a1d,   1,   1, 1 ),
+  V(   1, 0x2586,  14,   2, 0 ),
+  V(   2, 0x1114,  16,   3, 0 ),
+  V(   3, 0x080b,  18,   4, 0 ),
+  V(   4, 0x03d8,  20,   5, 0 ),
+  V(   5, 0x01da,  23,   6, 0 ),
+  V(   6, 0x00e5,  25,   7, 0 ),
+  V(   7, 0x006f,  28,   8, 0 ),
+  V(   8, 0x0036,  30,   9, 0 ),
+  V(   9, 0x001a,  33,  10, 0 ),
+  V(  10, 0x000d,  35,  11, 0 ),
+  V(  11, 0x0006,   9,  12, 0 ),
+  V(  12, 0x0003,  10,  13, 0 ),
+  V(  13, 0x0001,  12,  13, 0 ),
+  V(  14, 0x5a7f,  15,  15, 1 ),
+  V(  15, 0x3f25,  36,  16, 0 ),
+  V(  16, 0x2cf2,  38,  17, 0 ),
+  V(  17, 0x207c,  39,  18, 0 ),
+  V(  18, 0x17b9,  40,  19, 0 ),
+  V(  19, 0x1182,  42,  20, 0 ),
+  V(  20, 0x0cef,  43,  21, 0 ),
+  V(  21, 0x09a1,  45,  22, 0 ),
+  V(  22, 0x072f,  46,  23, 0 ),
+  V(  23, 0x055c,  48,  24, 0 ),
+  V(  24, 0x0406,  49,  25, 0 ),
+  V(  25, 0x0303,  51,  26, 0 ),
+  V(  26, 0x0240,  52,  27, 0 ),
+  V(  27, 0x01b1,  54,  28, 0 ),
+  V(  28, 0x0144,  56,  29, 0 ),
+  V(  29, 0x00f5,  57,  30, 0 ),
+  V(  30, 0x00b7,  59,  31, 0 ),
+  V(  31, 0x008a,  60,  32, 0 ),
+  V(  32, 0x0068,  62,  33, 0 ),
+  V(  33, 0x004e,  63,  34, 0 ),
+  V(  34, 0x003b,  32,  35, 0 ),
+  V(  35, 0x002c,  33,   9, 0 ),
+  V(  36, 0x5ae1,  37,  37, 1 ),
+  V(  37, 0x484c,  64,  38, 0 ),
+  V(  38, 0x3a0d,  65,  39, 0 ),
+  V(  39, 0x2ef1,  67,  40, 0 ),
+  V(  40, 0x261f,  68,  41, 0 ),
+  V(  41, 0x1f33,  69,  42, 0 ),
+  V(  42, 0x19a8,  70,  43, 0 ),
+  V(  43, 0x1518,  72,  44, 0 ),
+  V(  44, 0x1177,  73,  45, 0 ),
+  V(  45, 0x0e74,  74,  46, 0 ),
+  V(  46, 0x0bfb,  75,  47, 0 ),
+  V(  47, 0x09f8,  77,  48, 0 ),
+  V(  48, 0x0861,  78,  49, 0 ),
+  V(  49, 0x0706,  79,  50, 0 ),
+  V(  50, 0x05cd,  48,  51, 0 ),
+  V(  51, 0x04de,  50,  52, 0 ),
+  V(  52, 0x040f,  50,  53, 0 ),
+  V(  53, 0x0363,  51,  54, 0 ),
+  V(  54, 0x02d4,  52,  55, 0 ),
+  V(  55, 0x025c,  53,  56, 0 ),
+  V(  56, 0x01f8,  54,  57, 0 ),
+  V(  57, 0x01a4,  55,  58, 0 ),
+  V(  58, 0x0160,  56,  59, 0 ),
+  V(  59, 0x0125,  57,  60, 0 ),
+  V(  60, 0x00f6,  58,  61, 0 ),
+  V(  61, 0x00cb,  59,  62, 0 ),
+  V(  62, 0x00ab,  61,  63, 0 ),
+  V(  63, 0x008f,  61,  32, 0 ),
+  V(  64, 0x5b12,  65,  65, 1 ),
+  V(  65, 0x4d04,  80,  66, 0 ),
+  V(  66, 0x412c,  81,  67, 0 ),
+  V(  67, 0x37d8,  82,  68, 0 ),
+  V(  68, 0x2fe8,  83,  69, 0 ),
+  V(  69, 0x293c,  84,  70, 0 ),
+  V(  70, 0x2379,  86,  71, 0 ),
+  V(  71, 0x1edf,  87,  72, 0 ),
+  V(  72, 0x1aa9,  87,  73, 0 ),
+  V(  73, 0x174e,  72,  74, 0 ),
+  V(  74, 0x1424,  72,  75, 0 ),
+  V(  75, 0x119c,  74,  76, 0 ),
+  V(  76, 0x0f6b,  74,  77, 0 ),
+  V(  77, 0x0d51,  75,  78, 0 ),
+  V(  78, 0x0bb6,  77,  79, 0 ),
+  V(  79, 0x0a40,  77,  48, 0 ),
+  V(  80, 0x5832,  80,  81, 1 ),
+  V(  81, 0x4d1c,  88,  82, 0 ),
+  V(  82, 0x438e,  89,  83, 0 ),
+  V(  83, 0x3bdd,  90,  84, 0 ),
+  V(  84, 0x34ee,  91,  85, 0 ),
+  V(  85, 0x2eae,  92,  86, 0 ),
+  V(  86, 0x299a,  93,  87, 0 ),
+  V(  87, 0x2516,  86,  71, 0 ),
+  V(  88, 0x5570,  88,  89, 1 ),
+  V(  89, 0x4ca9,  95,  90, 0 ),
+  V(  90, 0x44d9,  96,  91, 0 ),
+  V(  91, 0x3e22,  97,  92, 0 ),
+  V(  92, 0x3824,  99,  93, 0 ),
+  V(  93, 0x32b4,  99,  94, 0 ),
+  V(  94, 0x2e17,  93,  86, 0 ),
+  V(  95, 0x56a8,  95,  96, 1 ),
+  V(  96, 0x4f46, 101,  97, 0 ),
+  V(  97, 0x47e5, 102,  98, 0 ),
+  V(  98, 0x41cf, 103,  99, 0 ),
+  V(  99, 0x3c3d, 104, 100, 0 ),
+  V( 100, 0x375e,  99,  93, 0 ),
+  V( 101, 0x5231, 105, 102, 0 ),
+  V( 102, 0x4c0f, 106, 103, 0 ),
+  V( 103, 0x4639, 107, 104, 0 ),
+  V( 104, 0x415e, 103,  99, 0 ),
+  V( 105, 0x5627, 105, 106, 1 ),
+  V( 106, 0x50e7, 108, 107, 0 ),
+  V( 107, 0x4b85, 109, 103, 0 ),
+  V( 108, 0x5597, 110, 109, 0 ),
+  V( 109, 0x504f, 111, 107, 0 ),
+  V( 110, 0x5a10, 110, 111, 1 ),
+  V( 111, 0x5522, 112, 109, 0 ),
+  V( 112, 0x59eb, 112, 111, 1 ),
+/*
+ * This last entry is used for fixed probability estimate of 0.5
+ * as suggested in Section 10.3 Table 5 of ITU-T Rec. T.851.
+ */
+  V( 113, 0x5a1d, 113, 113, 0 )
+};
@@ -0,0 +1,288 @@
+/*
+ * jcapimin.c
+ *
+ * Copyright (C) 1994-1998, Thomas G. Lane.
+ * Modified 2003-2010 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains application interface code for the compression half
+ * of the JPEG library.  These are the "minimum" API routines that may be
+ * needed in either the normal full-compression case or the transcoding-only
+ * case.
+ *
+ * Most of the routines intended to be called directly by an application
+ * are in this file or in jcapistd.c.  But also see jcparam.c for
+ * parameter-setup helper routines, jcomapi.c for routines shared by
+ * compression and decompression, and jctrans.c for the transcoding case.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/*
+ * Initialization of a JPEG compression object.
+ * The error manager must already be set up (in case memory manager fails).
+ */
+
+GLOBAL(void)
+jpeg_CreateCompress (j_compress_ptr cinfo, int version, size_t structsize)
+{
+  int i;
+
+  /* Guard against version mismatches between library and caller. */
+  cinfo->mem = NULL;		/* so jpeg_destroy knows mem mgr not called */
+  if (version != JPEG_LIB_VERSION)
+    ERREXIT2(cinfo, JERR_BAD_LIB_VERSION, JPEG_LIB_VERSION, version);
+  if (structsize != SIZEOF(struct jpeg_compress_struct))
+    ERREXIT2(cinfo, JERR_BAD_STRUCT_SIZE, 
+	     (int) SIZEOF(struct jpeg_compress_struct), (int) structsize);
+
+  /* For debugging purposes, we zero the whole master structure.
+   * But the application has already set the err pointer, and may have set
+   * client_data, so we have to save and restore those fields.
+   * Note: if application hasn't set client_data, tools like Purify may
+   * complain here.
+   */
+  {
+    struct jpeg_error_mgr * err = cinfo->err;
+    void * client_data = cinfo->client_data; /* ignore Purify complaint here */
+    MEMZERO(cinfo, SIZEOF(struct jpeg_compress_struct));
+    cinfo->err = err;
+    cinfo->client_data = client_data;
+  }
+  cinfo->is_decompressor = FALSE;
+
+  /* Initialize a memory manager instance for this object */
+  jinit_memory_mgr((j_common_ptr) cinfo);
+
+  /* Zero out pointers to permanent structures. */
+  cinfo->progress = NULL;
+  cinfo->dest = NULL;
+
+  cinfo->comp_info = NULL;
+
+  for (i = 0; i < NUM_QUANT_TBLS; i++) {
+    cinfo->quant_tbl_ptrs[i] = NULL;
+    cinfo->q_scale_factor[i] = 100;
+  }
+
+  for (i = 0; i < NUM_HUFF_TBLS; i++) {
+    cinfo->dc_huff_tbl_ptrs[i] = NULL;
+    cinfo->ac_huff_tbl_ptrs[i] = NULL;
+  }
+
+  /* Must do it here for emit_dqt in case jpeg_write_tables is used */
+  cinfo->block_size = DCTSIZE;
+  cinfo->natural_order = jpeg_natural_order;
+  cinfo->lim_Se = DCTSIZE2-1;
+
+  cinfo->script_space = NULL;
+
+  cinfo->input_gamma = 1.0;	/* in case application forgets */
+
+  /* OK, I'm ready */
+  cinfo->global_state = CSTATE_START;
+}
+
+
+/*
+ * Destruction of a JPEG compression object
+ */
+
+GLOBAL(void)
+jpeg_destroy_compress (j_compress_ptr cinfo)
+{
+  jpeg_destroy((j_common_ptr) cinfo); /* use common routine */
+}
+
+
+/*
+ * Abort processing of a JPEG compression operation,
+ * but don't destroy the object itself.
+ */
+
+GLOBAL(void)
+jpeg_abort_compress (j_compress_ptr cinfo)
+{
+  jpeg_abort((j_common_ptr) cinfo); /* use common routine */
+}
+
+
+/*
+ * Forcibly suppress or un-suppress all quantization and Huffman tables.
+ * Marks all currently defined tables as already written (if suppress)
+ * or not written (if !suppress).  This will control whether they get emitted
+ * by a subsequent jpeg_start_compress call.
+ *
+ * This routine is exported for use by applications that want to produce
+ * abbreviated JPEG datastreams.  It logically belongs in jcparam.c, but
+ * since it is called by jpeg_start_compress, we put it here --- otherwise
+ * jcparam.o would be linked whether the application used it or not.
+ */
+
+GLOBAL(void)
+jpeg_suppress_tables (j_compress_ptr cinfo, boolean suppress)
+{
+  int i;
+  JQUANT_TBL * qtbl;
+  JHUFF_TBL * htbl;
+
+  for (i = 0; i < NUM_QUANT_TBLS; i++) {
+    if ((qtbl = cinfo->quant_tbl_ptrs[i]) != NULL)
+      qtbl->sent_table = suppress;
+  }
+
+  for (i = 0; i < NUM_HUFF_TBLS; i++) {
+    if ((htbl = cinfo->dc_huff_tbl_ptrs[i]) != NULL)
+      htbl->sent_table = suppress;
+    if ((htbl = cinfo->ac_huff_tbl_ptrs[i]) != NULL)
+      htbl->sent_table = suppress;
+  }
+}
+
+
+/*
+ * Finish JPEG compression.
+ *
+ * If a multipass operating mode was selected, this may do a great deal of
+ * work including most of the actual output.
+ */
+
+GLOBAL(void)
+jpeg_finish_compress (j_compress_ptr cinfo)
+{
+  JDIMENSION iMCU_row;
+
+  if (cinfo->global_state == CSTATE_SCANNING ||
+      cinfo->global_state == CSTATE_RAW_OK) {
+    /* Terminate first pass */
+    if (cinfo->next_scanline < cinfo->image_height)
+      ERREXIT(cinfo, JERR_TOO_LITTLE_DATA);
+    (*cinfo->master->finish_pass) (cinfo);
+  } else if (cinfo->global_state != CSTATE_WRCOEFS)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+  /* Perform any remaining passes */
+  while (! cinfo->master->is_last_pass) {
+    (*cinfo->master->prepare_for_pass) (cinfo);
+    for (iMCU_row = 0; iMCU_row < cinfo->total_iMCU_rows; iMCU_row++) {
+      if (cinfo->progress != NULL) {
+	cinfo->progress->pass_counter = (long) iMCU_row;
+	cinfo->progress->pass_limit = (long) cinfo->total_iMCU_rows;
+	(*cinfo->progress->progress_monitor) ((j_common_ptr) cinfo);
+      }
+      /* We bypass the main controller and invoke coef controller directly;
+       * all work is being done from the coefficient buffer.
+       */
+      if (! (*cinfo->coef->compress_data) (cinfo, (JSAMPIMAGE) NULL))
+	ERREXIT(cinfo, JERR_CANT_SUSPEND);
+    }
+    (*cinfo->master->finish_pass) (cinfo);
+  }
+  /* Write EOI, do final cleanup */
+  (*cinfo->marker->write_file_trailer) (cinfo);
+  (*cinfo->dest->term_destination) (cinfo);
+  /* We can use jpeg_abort to release memory and reset global_state */
+  jpeg_abort((j_common_ptr) cinfo);
+}
+
+
+/*
+ * Write a special marker.
+ * This is only recommended for writing COM or APPn markers.
+ * Must be called after jpeg_start_compress() and before
+ * first call to jpeg_write_scanlines() or jpeg_write_raw_data().
+ */
+
+GLOBAL(void)
+jpeg_write_marker (j_compress_ptr cinfo, int marker,
+		   const JOCTET *dataptr, unsigned int datalen)
+{
+  JMETHOD(void, write_marker_byte, (j_compress_ptr info, int val));
+
+  if (cinfo->next_scanline != 0 ||
+      (cinfo->global_state != CSTATE_SCANNING &&
+       cinfo->global_state != CSTATE_RAW_OK &&
+       cinfo->global_state != CSTATE_WRCOEFS))
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+
+  (*cinfo->marker->write_marker_header) (cinfo, marker, datalen);
+  write_marker_byte = cinfo->marker->write_marker_byte;	/* copy for speed */
+  while (datalen--) {
+    (*write_marker_byte) (cinfo, *dataptr);
+    dataptr++;
+  }
+}
+
+/* Same, but piecemeal. */
+
+GLOBAL(void)
+jpeg_write_m_header (j_compress_ptr cinfo, int marker, unsigned int datalen)
+{
+  if (cinfo->next_scanline != 0 ||
+      (cinfo->global_state != CSTATE_SCANNING &&
+       cinfo->global_state != CSTATE_RAW_OK &&
+       cinfo->global_state != CSTATE_WRCOEFS))
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+
+  (*cinfo->marker->write_marker_header) (cinfo, marker, datalen);
+}
+
+GLOBAL(void)
+jpeg_write_m_byte (j_compress_ptr cinfo, int val)
+{
+  (*cinfo->marker->write_marker_byte) (cinfo, val);
+}
+
+
+/*
+ * Alternate compression function: just write an abbreviated table file.
+ * Before calling this, all parameters and a data destination must be set up.
+ *
+ * To produce a pair of files containing abbreviated tables and abbreviated
+ * image data, one would proceed as follows:
+ *
+ *		initialize JPEG object
+ *		set JPEG parameters
+ *		set destination to table file
+ *		jpeg_write_tables(cinfo);
+ *		set destination to image file
+ *		jpeg_start_compress(cinfo, FALSE);
+ *		write data...
+ *		jpeg_finish_compress(cinfo);
+ *
+ * jpeg_write_tables has the side effect of marking all tables written
+ * (same as jpeg_suppress_tables(..., TRUE)).  Thus a subsequent start_compress
+ * will not re-emit the tables unless it is passed write_all_tables=TRUE.
+ */
+
+GLOBAL(void)
+jpeg_write_tables (j_compress_ptr cinfo)
+{
+  if (cinfo->global_state != CSTATE_START)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+
+  /* (Re)initialize error mgr and destination modules */
+  (*cinfo->err->reset_error_mgr) ((j_common_ptr) cinfo);
+  (*cinfo->dest->init_destination) (cinfo);
+  /* Initialize the marker writer ... bit of a crock to do it here. */
+  jinit_marker_writer(cinfo);
+  /* Write them tables! */
+  (*cinfo->marker->write_tables_only) (cinfo);
+  /* And clean up. */
+  (*cinfo->dest->term_destination) (cinfo);
+  /*
+   * In library releases up through v6a, we called jpeg_abort() here to free
+   * any working memory allocated by the destination manager and marker
+   * writer.  Some applications had a problem with that: they allocated space
+   * of their own from the library memory manager, and didn't want it to go
+   * away during write_tables.  So now we do nothing.  This will cause a
+   * memory leak if an app calls write_tables repeatedly without doing a full
+   * compression cycle or otherwise resetting the JPEG object.  However, that
+   * seems less bad than unexpectedly freeing memory in the normal case.
+   * An app that prefers the old behavior can call jpeg_abort for itself after
+   * each call to jpeg_write_tables().
+   */
+}
@@ -0,0 +1,162 @@
+/*
+ * jcapistd.c
+ *
+ * Copyright (C) 1994-1996, Thomas G. Lane.
+ * Modified 2013 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains application interface code for the compression half
+ * of the JPEG library.  These are the "standard" API routines that are
+ * used in the normal full-compression case.  They are not used by a
+ * transcoding-only application.  Note that if an application links in
+ * jpeg_start_compress, it will end up linking in the entire compressor.
+ * We thus must separate this file from jcapimin.c to avoid linking the
+ * whole compression library into a transcoder.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/*
+ * Compression initialization.
+ * Before calling this, all parameters and a data destination must be set up.
+ *
+ * We require a write_all_tables parameter as a failsafe check when writing
+ * multiple datastreams from the same compression object.  Since prior runs
+ * will have left all the tables marked sent_table=TRUE, a subsequent run
+ * would emit an abbreviated stream (no tables) by default.  This may be what
+ * is wanted, but for safety's sake it should not be the default behavior:
+ * programmers should have to make a deliberate choice to emit abbreviated
+ * images.  Therefore the documentation and examples should encourage people
+ * to pass write_all_tables=TRUE; then it will take active thought to do the
+ * wrong thing.
+ */
+
+GLOBAL(void)
+jpeg_start_compress (j_compress_ptr cinfo, boolean write_all_tables)
+{
+  if (cinfo->global_state != CSTATE_START)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+
+  if (write_all_tables)
+    jpeg_suppress_tables(cinfo, FALSE);	/* mark all tables to be written */
+
+  /* (Re)initialize error mgr and destination modules */
+  (*cinfo->err->reset_error_mgr) ((j_common_ptr) cinfo);
+  (*cinfo->dest->init_destination) (cinfo);
+  /* Perform master selection of active modules */
+  jinit_compress_master(cinfo);
+  /* Set up for the first pass */
+  (*cinfo->master->prepare_for_pass) (cinfo);
+  /* Ready for application to drive first pass through jpeg_write_scanlines
+   * or jpeg_write_raw_data.
+   */
+  cinfo->next_scanline = 0;
+  cinfo->global_state = (cinfo->raw_data_in ? CSTATE_RAW_OK : CSTATE_SCANNING);
+}
+
+
+/*
+ * Write some scanlines of data to the JPEG compressor.
+ *
+ * The return value will be the number of lines actually written.
+ * This should be less than the supplied num_lines only in case that
+ * the data destination module has requested suspension of the compressor,
+ * or if more than image_height scanlines are passed in.
+ *
+ * Note: we warn about excess calls to jpeg_write_scanlines() since
+ * this likely signals an application programmer error.  However,
+ * excess scanlines passed in the last valid call are *silently* ignored,
+ * so that the application need not adjust num_lines for end-of-image
+ * when using a multiple-scanline buffer.
+ */
+
+GLOBAL(JDIMENSION)
+jpeg_write_scanlines (j_compress_ptr cinfo, JSAMPARRAY scanlines,
+		      JDIMENSION num_lines)
+{
+  JDIMENSION row_ctr, rows_left;
+
+  if (cinfo->global_state != CSTATE_SCANNING)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+  if (cinfo->next_scanline >= cinfo->image_height)
+    WARNMS(cinfo, JWRN_TOO_MUCH_DATA);
+
+  /* Call progress monitor hook if present */
+  if (cinfo->progress != NULL) {
+    cinfo->progress->pass_counter = (long) cinfo->next_scanline;
+    cinfo->progress->pass_limit = (long) cinfo->image_height;
+    (*cinfo->progress->progress_monitor) ((j_common_ptr) cinfo);
+  }
+
+  /* Give master control module another chance if this is first call to
+   * jpeg_write_scanlines.  This lets output of the frame/scan headers be
+   * delayed so that application can write COM, etc, markers between
+   * jpeg_start_compress and jpeg_write_scanlines.
+   */
+  if (cinfo->master->call_pass_startup)
+    (*cinfo->master->pass_startup) (cinfo);
+
+  /* Ignore any extra scanlines at bottom of image. */
+  rows_left = cinfo->image_height - cinfo->next_scanline;
+  if (num_lines > rows_left)
+    num_lines = rows_left;
+
+  row_ctr = 0;
+  (*cinfo->main->process_data) (cinfo, scanlines, &row_ctr, num_lines);
+  cinfo->next_scanline += row_ctr;
+  return row_ctr;
+}
+
+
+/*
+ * Alternate entry point to write raw data.
+ * Processes exactly one iMCU row per call, unless suspended.
+ */
+
+GLOBAL(JDIMENSION)
+jpeg_write_raw_data (j_compress_ptr cinfo, JSAMPIMAGE data,
+		     JDIMENSION num_lines)
+{
+  JDIMENSION lines_per_iMCU_row;
+
+  if (cinfo->global_state != CSTATE_RAW_OK)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+  if (cinfo->next_scanline >= cinfo->image_height) {
+    WARNMS(cinfo, JWRN_TOO_MUCH_DATA);
+    return 0;
+  }
+
+  /* Call progress monitor hook if present */
+  if (cinfo->progress != NULL) {
+    cinfo->progress->pass_counter = (long) cinfo->next_scanline;
+    cinfo->progress->pass_limit = (long) cinfo->image_height;
+    (*cinfo->progress->progress_monitor) ((j_common_ptr) cinfo);
+  }
+
+  /* Give master control module another chance if this is first call to
+   * jpeg_write_raw_data.  This lets output of the frame/scan headers be
+   * delayed so that application can write COM, etc, markers between
+   * jpeg_start_compress and jpeg_write_raw_data.
+   */
+  if (cinfo->master->call_pass_startup)
+    (*cinfo->master->pass_startup) (cinfo);
+
+  /* Verify that at least one iMCU row has been passed. */
+  lines_per_iMCU_row = cinfo->max_v_samp_factor * cinfo->min_DCT_v_scaled_size;
+  if (num_lines < lines_per_iMCU_row)
+    ERREXIT(cinfo, JERR_BUFFER_SIZE);
+
+  /* Directly compress the row. */
+  if (! (*cinfo->coef->compress_data) (cinfo, data)) {
+    /* If compressor did not consume the whole row, suspend processing. */
+    return 0;
+  }
+
+  /* OK, we processed one iMCU row. */
+  cinfo->next_scanline += lines_per_iMCU_row;
+  return lines_per_iMCU_row;
+}
@@ -0,0 +1,945 @@
+/*
+ * jcarith.c
+ *
+ * Developed 1997-2020 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains portable arithmetic entropy encoding routines for JPEG
+ * (implementing the ISO/IEC IS 10918-1 and CCITT Recommendation ITU-T T.81).
+ *
+ * Both sequential and progressive modes are supported in this single module.
+ *
+ * Suspension is not currently supported in this module.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/* Expanded entropy encoder object for arithmetic encoding. */
+
+typedef struct {
+  struct jpeg_entropy_encoder pub; /* public fields */
+
+  INT32 c; /* C register, base of coding interval, layout as in sec. D.1.3 */
+  INT32 a;               /* A register, normalized size of coding interval */
+  INT32 sc;        /* counter for stacked 0xFF values which might overflow */
+  INT32 zc;          /* counter for pending 0x00 output values which might *
+                          * be discarded at the end ("Pacman" termination) */
+  int ct;  /* bit shift counter, determines when next byte will be written */
+  int buffer;                /* buffer for most recent output byte != 0xFF */
+
+  int last_dc_val[MAX_COMPS_IN_SCAN]; /* last DC coef for each component */
+  int dc_context[MAX_COMPS_IN_SCAN]; /* context index for DC conditioning */
+
+  unsigned int restarts_to_go;	/* MCUs left in this restart interval */
+  int next_restart_num;		/* next restart number to write (0-7) */
+
+  /* Pointers to statistics areas (these workspaces have image lifespan) */
+  unsigned char * dc_stats[NUM_ARITH_TBLS];
+  unsigned char * ac_stats[NUM_ARITH_TBLS];
+
+  /* Statistics bin for coding with fixed probability 0.5 */
+  unsigned char fixed_bin[4];
+} arith_entropy_encoder;
+
+typedef arith_entropy_encoder * arith_entropy_ptr;
+
+/* The following two definitions specify the allocation chunk size
+ * for the statistics area.
+ * According to sections F.1.4.4.1.3 and F.1.4.4.2, we need at least
+ * 49 statistics bins for DC, and 245 statistics bins for AC coding.
+ *
+ * We use a compact representation with 1 byte per statistics bin,
+ * thus the numbers directly represent byte sizes.
+ * This 1 byte per statistics bin contains the meaning of the MPS
+ * (more probable symbol) in the highest bit (mask 0x80), and the
+ * index into the probability estimation state machine table
+ * in the lower bits (mask 0x7F).
+ */
+
+#define DC_STAT_BINS 64
+#define AC_STAT_BINS 256
+
+/* NOTE: Uncomment the following #define if you want to use the
+ * given formula for calculating the AC conditioning parameter Kx
+ * for spectral selection progressive coding in section G.1.3.2
+ * of the spec (Kx = Kmin + SRL (8 + Se - Kmin) 4).
+ * Although the spec and P&M authors claim that this "has proven
+ * to give good results for 8 bit precision samples", I'm not
+ * convinced yet that this is really beneficial.
+ * Early tests gave only very marginal compression enhancements
+ * (a few - around 5 or so - bytes even for very large files),
+ * which would turn out rather negative if we'd suppress the
+ * DAC (Define Arithmetic Conditioning) marker segments for
+ * the default parameters in the future.
+ * Note that currently the marker writing module emits 12-byte
+ * DAC segments for a full-component scan in a color image.
+ * This is not worth worrying about IMHO. However, since the
+ * spec defines the default values to be used if the tables
+ * are omitted (unlike Huffman tables, which are required
+ * anyway), one might optimize this behaviour in the future,
+ * and then it would be disadvantageous to use custom tables if
+ * they don't provide sufficient gain to exceed the DAC size.
+ *
+ * On the other hand, I'd consider it as a reasonable result
+ * that the conditioning has no significant influence on the
+ * compression performance. This means that the basic
+ * statistical model is already rather stable.
+ *
+ * Thus, at the moment, we use the default conditioning values
+ * anyway, and do not use the custom formula.
+ *
+#define CALCULATE_SPECTRAL_CONDITIONING
+ */
+
+/* IRIGHT_SHIFT is like RIGHT_SHIFT, but works on int rather than INT32.
+ * We assume that int right shift is unsigned if INT32 right shift is,
+ * which should be safe.
+ */
+
+#ifdef RIGHT_SHIFT_IS_UNSIGNED
+#define ISHIFT_TEMPS	int ishift_temp;
+#define IRIGHT_SHIFT(x,shft)  \
+	((ishift_temp = (x)) < 0 ? \
+	 (ishift_temp >> (shft)) | ((~0) << (16-(shft))) : \
+	 (ishift_temp >> (shft)))
+#else
+#define ISHIFT_TEMPS
+#define IRIGHT_SHIFT(x,shft)	((x) >> (shft))
+#endif
+
+
+LOCAL(void)
+emit_byte (int val, j_compress_ptr cinfo)
+/* Write next output byte; we do not support suspension in this module. */
+{
+  struct jpeg_destination_mgr * dest = cinfo->dest;
+
+  *dest->next_output_byte++ = (JOCTET) val;
+  if (--dest->free_in_buffer == 0)
+    if (! (*dest->empty_output_buffer) (cinfo))
+      ERREXIT(cinfo, JERR_CANT_SUSPEND);
+}
+
+
+/*
+ * Finish up at the end of an arithmetic-compressed scan.
+ */
+
+METHODDEF(void)
+finish_pass (j_compress_ptr cinfo)
+{
+  arith_entropy_ptr e = (arith_entropy_ptr) cinfo->entropy;
+  INT32 temp;
+
+  /* Section D.1.8: Termination of encoding */
+
+  /* Find the e->c in the coding interval with the largest
+   * number of trailing zero bits */
+  if ((temp = (e->a - 1 + e->c) & 0xFFFF0000L) < e->c)
+    e->c = temp + 0x8000L;
+  else
+    e->c = temp;
+  /* Send remaining bytes to output */
+  e->c <<= e->ct;
+  if (e->c & 0xF8000000L) {
+    /* One final overflow has to be handled */
+    if (e->buffer >= 0) {
+      if (e->zc)
+	do emit_byte(0x00, cinfo);
+	while (--e->zc);
+      emit_byte(e->buffer + 1, cinfo);
+      if (e->buffer + 1 == 0xFF)
+	emit_byte(0x00, cinfo);
+    }
+    e->zc += e->sc;  /* carry-over converts stacked 0xFF bytes to 0x00 */
+    e->sc = 0;
+  } else {
+    if (e->buffer == 0)
+      ++e->zc;
+    else if (e->buffer >= 0) {
+      if (e->zc)
+	do emit_byte(0x00, cinfo);
+	while (--e->zc);
+      emit_byte(e->buffer, cinfo);
+    }
+    if (e->sc) {
+      if (e->zc)
+	do emit_byte(0x00, cinfo);
+	while (--e->zc);
+      do {
+	emit_byte(0xFF, cinfo);
+	emit_byte(0x00, cinfo);
+      } while (--e->sc);
+    }
+  }
+  /* Output final bytes only if they are not 0x00 */
+  if (e->c & 0x7FFF800L) {
+    if (e->zc)  /* output final pending zero bytes */
+      do emit_byte(0x00, cinfo);
+      while (--e->zc);
+    emit_byte((int) ((e->c >> 19) & 0xFF), cinfo);
+    if (((e->c >> 19) & 0xFF) == 0xFF)
+      emit_byte(0x00, cinfo);
+    if (e->c & 0x7F800L) {
+      emit_byte((int) ((e->c >> 11) & 0xFF), cinfo);
+      if (((e->c >> 11) & 0xFF) == 0xFF)
+	emit_byte(0x00, cinfo);
+    }
+  }
+}
+
+
+/*
+ * The core arithmetic encoding routine (common in JPEG and JBIG).
+ * This needs to go as fast as possible.
+ * Machine-dependent optimization facilities
+ * are not utilized in this portable implementation.
+ * However, this code should be fairly efficient and
+ * may be a good base for further optimizations anyway.
+ *
+ * Parameter 'val' to be encoded may be 0 or 1 (binary decision).
+ *
+ * Note: I've added full "Pacman" termination support to the
+ * byte output routines, which is equivalent to the optional
+ * Discard_final_zeros procedure (Figure D.15) in the spec.
+ * Thus, we always produce the shortest possible output
+ * stream compliant to the spec (no trailing zero bytes,
+ * except for FF stuffing).
+ *
+ * I've also introduced a new scheme for accessing
+ * the probability estimation state machine table,
+ * derived from Markus Kuhn's JBIG implementation.
+ */
+
+LOCAL(void)
+arith_encode (j_compress_ptr cinfo, unsigned char *st, int val) 
+{
+  register arith_entropy_ptr e = (arith_entropy_ptr) cinfo->entropy;
+  register unsigned char nl, nm;
+  register INT32 qe, temp;
+  register int sv;
+
+  /* Fetch values from our compact representation of Table D.3(D.2):
+   * Qe values and probability estimation state machine
+   */
+  sv = *st;
+  qe = jpeg_aritab[sv & 0x7F];	/* => Qe_Value */
+  nl = qe & 0xFF; qe >>= 8;	/* Next_Index_LPS + Switch_MPS */
+  nm = qe & 0xFF; qe >>= 8;	/* Next_Index_MPS */
+
+  /* Encode & estimation procedures per sections D.1.4 & D.1.5 */
+  e->a -= qe;
+  if (val != (sv >> 7)) {
+    /* Encode the less probable symbol */
+    if (e->a >= qe) {
+      /* If the interval size (qe) for the less probable symbol (LPS)
+       * is larger than the interval size for the MPS, then exchange
+       * the two symbols for coding efficiency, otherwise code the LPS
+       * as usual: */
+      e->c += e->a;
+      e->a = qe;
+    }
+    *st = (sv & 0x80) ^ nl;	/* Estimate_after_LPS */
+  } else {
+    /* Encode the more probable symbol */
+    if (e->a >= 0x8000L)
+      return;  /* A >= 0x8000 -> ready, no renormalization required */
+    if (e->a < qe) {
+      /* If the interval size (qe) for the less probable symbol (LPS)
+       * is larger than the interval size for the MPS, then exchange
+       * the two symbols for coding efficiency: */
+      e->c += e->a;
+      e->a = qe;
+    }
+    *st = (sv & 0x80) ^ nm;	/* Estimate_after_MPS */
+  }
+
+  /* Renormalization & data output per section D.1.6 */
+  do {
+    e->a <<= 1;
+    e->c <<= 1;
+    if (--e->ct == 0) {
+      /* Another byte is ready for output */
+      temp = e->c >> 19;
+      if (temp > 0xFF) {
+	/* Handle overflow over all stacked 0xFF bytes */
+	if (e->buffer >= 0) {
+	  if (e->zc)
+	    do emit_byte(0x00, cinfo);
+	    while (--e->zc);
+	  emit_byte(e->buffer + 1, cinfo);
+	  if (e->buffer + 1 == 0xFF)
+	    emit_byte(0x00, cinfo);
+	}
+	e->zc += e->sc;  /* carry-over converts stacked 0xFF bytes to 0x00 */
+	e->sc = 0;
+	/* Note: The 3 spacer bits in the C register guarantee
+	 * that the new buffer byte can't be 0xFF here
+	 * (see page 160 in the P&M JPEG book). */
+	/* New output byte, might overflow later */
+	e->buffer = (int) (temp & 0xFF);
+      } else if (temp == 0xFF) {
+	++e->sc;  /* stack 0xFF byte (which might overflow later) */
+      } else {
+	/* Output all stacked 0xFF bytes, they will not overflow any more */
+	if (e->buffer == 0)
+	  ++e->zc;
+	else if (e->buffer >= 0) {
+	  if (e->zc)
+	    do emit_byte(0x00, cinfo);
+	    while (--e->zc);
+	  emit_byte(e->buffer, cinfo);
+	}
+	if (e->sc) {
+	  if (e->zc)
+	    do emit_byte(0x00, cinfo);
+	    while (--e->zc);
+	  do {
+	    emit_byte(0xFF, cinfo);
+	    emit_byte(0x00, cinfo);
+	  } while (--e->sc);
+	}
+	/* New output byte (can still overflow) */
+	e->buffer = (int) (temp & 0xFF);
+      }
+      e->c &= 0x7FFFFL;
+      e->ct += 8;
+    }
+  } while (e->a < 0x8000L);
+}
+
+
+/*
+ * Emit a restart marker & resynchronize predictions.
+ */
+
+LOCAL(void)
+emit_restart (j_compress_ptr cinfo, int restart_num)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  int ci;
+  jpeg_component_info * compptr;
+
+  finish_pass(cinfo);
+
+  emit_byte(0xFF, cinfo);
+  emit_byte(JPEG_RST0 + restart_num, cinfo);
+
+  /* Re-initialize statistics areas */
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+    compptr = cinfo->cur_comp_info[ci];
+    /* DC needs no table for refinement scan */
+    if (cinfo->Ss == 0 && cinfo->Ah == 0) {
+      MEMZERO(entropy->dc_stats[compptr->dc_tbl_no], DC_STAT_BINS);
+      /* Reset DC predictions to 0 */
+      entropy->last_dc_val[ci] = 0;
+      entropy->dc_context[ci] = 0;
+    }
+    /* AC needs no table when not present */
+    if (cinfo->Se) {
+      MEMZERO(entropy->ac_stats[compptr->ac_tbl_no], AC_STAT_BINS);
+    }
+  }
+
+  /* Reset arithmetic encoding variables */
+  entropy->c = 0;
+  entropy->a = 0x10000L;
+  entropy->sc = 0;
+  entropy->zc = 0;
+  entropy->ct = 11;
+  entropy->buffer = -1;  /* empty */
+}
+
+
+/*
+ * MCU encoding for DC initial scan (either spectral selection,
+ * or first pass of successive approximation).
+ */
+
+METHODDEF(boolean)
+encode_mcu_DC_first (j_compress_ptr cinfo, JBLOCKARRAY MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  unsigned char *st;
+  int blkn, ci, tbl;
+  int v, v2, m;
+  ISHIFT_TEMPS
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      emit_restart(cinfo, entropy->next_restart_num);
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  /* Encode the MCU data blocks */
+  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+    ci = cinfo->MCU_membership[blkn];
+    tbl = cinfo->cur_comp_info[ci]->dc_tbl_no;
+
+    /* Compute the DC value after the required point transform by Al.
+     * This is simply an arithmetic right shift.
+     */
+    m = IRIGHT_SHIFT((int) (MCU_data[blkn][0][0]), cinfo->Al);
+
+    /* Sections F.1.4.1 & F.1.4.4.1: Encoding of DC coefficients */
+
+    /* Table F.4: Point to statistics bin S0 for DC coefficient coding */
+    st = entropy->dc_stats[tbl] + entropy->dc_context[ci];
+
+    /* Figure F.4: Encode_DC_DIFF */
+    if ((v = m - entropy->last_dc_val[ci]) == 0) {
+      arith_encode(cinfo, st, 0);
+      entropy->dc_context[ci] = 0;	/* zero diff category */
+    } else {
+      entropy->last_dc_val[ci] = m;
+      arith_encode(cinfo, st, 1);
+      /* Figure F.6: Encoding nonzero value v */
+      /* Figure F.7: Encoding the sign of v */
+      if (v > 0) {
+	arith_encode(cinfo, st + 1, 0);	/* Table F.4: SS = S0 + 1 */
+	st += 2;			/* Table F.4: SP = S0 + 2 */
+	entropy->dc_context[ci] = 4;	/* small positive diff category */
+      } else {
+	v = -v;
+	arith_encode(cinfo, st + 1, 1);	/* Table F.4: SS = S0 + 1 */
+	st += 3;			/* Table F.4: SN = S0 + 3 */
+	entropy->dc_context[ci] = 8;	/* small negative diff category */
+      }
+      /* Figure F.8: Encoding the magnitude category of v */
+      m = 0;
+      if (v -= 1) {
+	arith_encode(cinfo, st, 1);
+	m = 1;
+	v2 = v;
+	st = entropy->dc_stats[tbl] + 20; /* Table F.4: X1 = 20 */
+	while (v2 >>= 1) {
+	  arith_encode(cinfo, st, 1);
+	  m <<= 1;
+	  st += 1;
+	}
+      }
+      arith_encode(cinfo, st, 0);
+      /* Section F.1.4.4.1.2: Establish dc_context conditioning category */
+      if (m < (int) ((1L << cinfo->arith_dc_L[tbl]) >> 1))
+	entropy->dc_context[ci] = 0;	/* zero diff category */
+      else if (m > (int) ((1L << cinfo->arith_dc_U[tbl]) >> 1))
+	entropy->dc_context[ci] += 8;	/* large diff category */
+      /* Figure F.9: Encoding the magnitude bit pattern of v */
+      st += 14;
+      while (m >>= 1)
+	arith_encode(cinfo, st, (m & v) ? 1 : 0);
+    }
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * MCU encoding for AC initial scan (either spectral selection,
+ * or first pass of successive approximation).
+ */
+
+METHODDEF(boolean)
+encode_mcu_AC_first (j_compress_ptr cinfo, JBLOCKARRAY MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  const int * natural_order;
+  JBLOCKROW block;
+  unsigned char *st;
+  int tbl, k, ke;
+  int v, v2, m;
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      emit_restart(cinfo, entropy->next_restart_num);
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  natural_order = cinfo->natural_order;
+
+  /* Encode the MCU data block */
+  block = MCU_data[0];
+  tbl = cinfo->cur_comp_info[0]->ac_tbl_no;
+
+  /* Sections F.1.4.2 & F.1.4.4.2: Encoding of AC coefficients */
+
+  /* Establish EOB (end-of-block) index */
+  ke = cinfo->Se;
+  do {
+    /* We must apply the point transform by Al.  For AC coefficients this
+     * is an integer division with rounding towards 0.  To do this portably
+     * in C, we shift after obtaining the absolute value.
+     */
+    if ((v = (*block)[natural_order[ke]]) >= 0) {
+      if (v >>= cinfo->Al) break;
+    } else {
+      v = -v;
+      if (v >>= cinfo->Al) break;
+    }
+  } while (--ke);
+
+  /* Figure F.5: Encode_AC_Coefficients */
+  for (k = cinfo->Ss - 1; k < ke;) {
+    st = entropy->ac_stats[tbl] + 3 * k;
+    arith_encode(cinfo, st, 0);		/* EOB decision */
+    for (;;) {
+      if ((v = (*block)[natural_order[++k]]) >= 0) {
+	if (v >>= cinfo->Al) {
+	  arith_encode(cinfo, st + 1, 1);
+	  arith_encode(cinfo, entropy->fixed_bin, 0);
+	  break;
+	}
+      } else {
+	v = -v;
+	if (v >>= cinfo->Al) {
+	  arith_encode(cinfo, st + 1, 1);
+	  arith_encode(cinfo, entropy->fixed_bin, 1);
+	  break;
+	}
+      }
+      arith_encode(cinfo, st + 1, 0);
+      st += 3;
+    }
+    st += 2;
+    /* Figure F.8: Encoding the magnitude category of v */
+    m = 0;
+    if (v -= 1) {
+      arith_encode(cinfo, st, 1);
+      m = 1;
+      v2 = v;
+      if (v2 >>= 1) {
+	arith_encode(cinfo, st, 1);
+	m <<= 1;
+	st = entropy->ac_stats[tbl] +
+	     (k <= cinfo->arith_ac_K[tbl] ? 189 : 217);
+	while (v2 >>= 1) {
+	  arith_encode(cinfo, st, 1);
+	  m <<= 1;
+	  st += 1;
+	}
+      }
+    }
+    arith_encode(cinfo, st, 0);
+    /* Figure F.9: Encoding the magnitude bit pattern of v */
+    st += 14;
+    while (m >>= 1)
+      arith_encode(cinfo, st, (m & v) ? 1 : 0);
+  }
+  /* Encode EOB decision only if k < cinfo->Se */
+  if (k < cinfo->Se) {
+    st = entropy->ac_stats[tbl] + 3 * k;
+    arith_encode(cinfo, st, 1);
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * MCU encoding for DC successive approximation refinement scan.
+ * Note: we assume such scans can be multi-component,
+ * although the spec is not very clear on the point.
+ */
+
+METHODDEF(boolean)
+encode_mcu_DC_refine (j_compress_ptr cinfo, JBLOCKARRAY MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  unsigned char *st;
+  int Al, blkn;
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      emit_restart(cinfo, entropy->next_restart_num);
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  st = entropy->fixed_bin;	/* use fixed probability estimation */
+  Al = cinfo->Al;
+
+  /* Encode the MCU data blocks */
+  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+    /* We simply emit the Al'th bit of the DC coefficient value. */
+    arith_encode(cinfo, st, (MCU_data[blkn][0][0] >> Al) & 1);
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * MCU encoding for AC successive approximation refinement scan.
+ */
+
+METHODDEF(boolean)
+encode_mcu_AC_refine (j_compress_ptr cinfo, JBLOCKARRAY MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  const int * natural_order;
+  JBLOCKROW block;
+  unsigned char *st;
+  int tbl, k, ke, kex;
+  int v;
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      emit_restart(cinfo, entropy->next_restart_num);
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  natural_order = cinfo->natural_order;
+
+  /* Encode the MCU data block */
+  block = MCU_data[0];
+  tbl = cinfo->cur_comp_info[0]->ac_tbl_no;
+
+  /* Section G.1.3.3: Encoding of AC coefficients */
+
+  /* Establish EOB (end-of-block) index */
+  ke = cinfo->Se;
+  do {
+    /* We must apply the point transform by Al.  For AC coefficients this
+     * is an integer division with rounding towards 0.  To do this portably
+     * in C, we shift after obtaining the absolute value.
+     */
+    if ((v = (*block)[natural_order[ke]]) >= 0) {
+      if (v >>= cinfo->Al) break;
+    } else {
+      v = -v;
+      if (v >>= cinfo->Al) break;
+    }
+  } while (--ke);
+
+  /* Establish EOBx (previous stage end-of-block) index */
+  for (kex = ke; kex > 0; kex--)
+    if ((v = (*block)[natural_order[kex]]) >= 0) {
+      if (v >>= cinfo->Ah) break;
+    } else {
+      v = -v;
+      if (v >>= cinfo->Ah) break;
+    }
+
+  /* Figure G.10: Encode_AC_Coefficients_SA */
+  for (k = cinfo->Ss - 1; k < ke;) {
+    st = entropy->ac_stats[tbl] + 3 * k;
+    if (k >= kex)
+      arith_encode(cinfo, st, 0);	/* EOB decision */
+    for (;;) {
+      if ((v = (*block)[natural_order[++k]]) >= 0) {
+	if (v >>= cinfo->Al) {
+	  if (v >> 1)			/* previously nonzero coef */
+	    arith_encode(cinfo, st + 2, (v & 1));
+	  else {			/* newly nonzero coef */
+	    arith_encode(cinfo, st + 1, 1);
+	    arith_encode(cinfo, entropy->fixed_bin, 0);
+	  }
+	  break;
+	}
+      } else {
+	v = -v;
+	if (v >>= cinfo->Al) {
+	  if (v >> 1)			/* previously nonzero coef */
+	    arith_encode(cinfo, st + 2, (v & 1));
+	  else {			/* newly nonzero coef */
+	    arith_encode(cinfo, st + 1, 1);
+	    arith_encode(cinfo, entropy->fixed_bin, 1);
+	  }
+	  break;
+	}
+      }
+      arith_encode(cinfo, st + 1, 0);
+      st += 3;
+    }
+  }
+  /* Encode EOB decision only if k < cinfo->Se */
+  if (k < cinfo->Se) {
+    st = entropy->ac_stats[tbl] + 3 * k;
+    arith_encode(cinfo, st, 1);
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * Encode and output one MCU's worth of arithmetic-compressed coefficients.
+ */
+
+METHODDEF(boolean)
+encode_mcu (j_compress_ptr cinfo, JBLOCKARRAY MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  const int * natural_order;
+  JBLOCKROW block;
+  unsigned char *st;
+  int tbl, k, ke;
+  int v, v2, m;
+  int blkn, ci;
+  jpeg_component_info * compptr;
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      emit_restart(cinfo, entropy->next_restart_num);
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  natural_order = cinfo->natural_order;
+
+  /* Encode the MCU data blocks */
+  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+    block = MCU_data[blkn];
+    ci = cinfo->MCU_membership[blkn];
+    compptr = cinfo->cur_comp_info[ci];
+
+    /* Sections F.1.4.1 & F.1.4.4.1: Encoding of DC coefficients */
+
+    tbl = compptr->dc_tbl_no;
+
+    /* Table F.4: Point to statistics bin S0 for DC coefficient coding */
+    st = entropy->dc_stats[tbl] + entropy->dc_context[ci];
+
+    /* Figure F.4: Encode_DC_DIFF */
+    if ((v = (*block)[0] - entropy->last_dc_val[ci]) == 0) {
+      arith_encode(cinfo, st, 0);
+      entropy->dc_context[ci] = 0;	/* zero diff category */
+    } else {
+      entropy->last_dc_val[ci] = (*block)[0];
+      arith_encode(cinfo, st, 1);
+      /* Figure F.6: Encoding nonzero value v */
+      /* Figure F.7: Encoding the sign of v */
+      if (v > 0) {
+	arith_encode(cinfo, st + 1, 0);	/* Table F.4: SS = S0 + 1 */
+	st += 2;			/* Table F.4: SP = S0 + 2 */
+	entropy->dc_context[ci] = 4;	/* small positive diff category */
+      } else {
+	v = -v;
+	arith_encode(cinfo, st + 1, 1);	/* Table F.4: SS = S0 + 1 */
+	st += 3;			/* Table F.4: SN = S0 + 3 */
+	entropy->dc_context[ci] = 8;	/* small negative diff category */
+      }
+      /* Figure F.8: Encoding the magnitude category of v */
+      m = 0;
+      if (v -= 1) {
+	arith_encode(cinfo, st, 1);
+	m = 1;
+	v2 = v;
+	st = entropy->dc_stats[tbl] + 20; /* Table F.4: X1 = 20 */
+	while (v2 >>= 1) {
+	  arith_encode(cinfo, st, 1);
+	  m <<= 1;
+	  st += 1;
+	}
+      }
+      arith_encode(cinfo, st, 0);
+      /* Section F.1.4.4.1.2: Establish dc_context conditioning category */
+      if (m < (int) ((1L << cinfo->arith_dc_L[tbl]) >> 1))
+	entropy->dc_context[ci] = 0;	/* zero diff category */
+      else if (m > (int) ((1L << cinfo->arith_dc_U[tbl]) >> 1))
+	entropy->dc_context[ci] += 8;	/* large diff category */
+      /* Figure F.9: Encoding the magnitude bit pattern of v */
+      st += 14;
+      while (m >>= 1)
+	arith_encode(cinfo, st, (m & v) ? 1 : 0);
+    }
+
+    /* Sections F.1.4.2 & F.1.4.4.2: Encoding of AC coefficients */
+
+    if ((ke = cinfo->lim_Se) == 0) continue;
+    tbl = compptr->ac_tbl_no;
+
+    /* Establish EOB (end-of-block) index */
+    do {
+      if ((*block)[natural_order[ke]]) break;
+    } while (--ke);
+
+    /* Figure F.5: Encode_AC_Coefficients */
+    for (k = 0; k < ke;) {
+      st = entropy->ac_stats[tbl] + 3 * k;
+      arith_encode(cinfo, st, 0);	/* EOB decision */
+      while ((v = (*block)[natural_order[++k]]) == 0) {
+	arith_encode(cinfo, st + 1, 0);
+	st += 3;
+      }
+      arith_encode(cinfo, st + 1, 1);
+      /* Figure F.6: Encoding nonzero value v */
+      /* Figure F.7: Encoding the sign of v */
+      if (v > 0) {
+	arith_encode(cinfo, entropy->fixed_bin, 0);
+      } else {
+	v = -v;
+	arith_encode(cinfo, entropy->fixed_bin, 1);
+      }
+      st += 2;
+      /* Figure F.8: Encoding the magnitude category of v */
+      m = 0;
+      if (v -= 1) {
+	arith_encode(cinfo, st, 1);
+	m = 1;
+	v2 = v;
+	if (v2 >>= 1) {
+	  arith_encode(cinfo, st, 1);
+	  m <<= 1;
+	  st = entropy->ac_stats[tbl] +
+	       (k <= cinfo->arith_ac_K[tbl] ? 189 : 217);
+	  while (v2 >>= 1) {
+	    arith_encode(cinfo, st, 1);
+	    m <<= 1;
+	    st += 1;
+	  }
+	}
+      }
+      arith_encode(cinfo, st, 0);
+      /* Figure F.9: Encoding the magnitude bit pattern of v */
+      st += 14;
+      while (m >>= 1)
+	arith_encode(cinfo, st, (m & v) ? 1 : 0);
+    }
+    /* Encode EOB decision only if k < cinfo->lim_Se */
+    if (k < cinfo->lim_Se) {
+      st = entropy->ac_stats[tbl] + 3 * k;
+      arith_encode(cinfo, st, 1);
+    }
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * Initialize for an arithmetic-compressed scan.
+ */
+
+METHODDEF(void)
+start_pass (j_compress_ptr cinfo, boolean gather_statistics)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  int ci, tbl;
+  jpeg_component_info * compptr;
+
+  if (gather_statistics)
+    /* Make sure to avoid that in the master control logic!
+     * We are fully adaptive here and need no extra
+     * statistics gathering pass!
+     */
+    ERREXIT(cinfo, JERR_NOT_COMPILED);
+
+  /* We assume jcmaster.c already validated the progressive scan parameters. */
+
+  /* Select execution routines */
+  if (cinfo->progressive_mode) {
+    if (cinfo->Ah == 0) {
+      if (cinfo->Ss == 0)
+	entropy->pub.encode_mcu = encode_mcu_DC_first;
+      else
+	entropy->pub.encode_mcu = encode_mcu_AC_first;
+    } else {
+      if (cinfo->Ss == 0)
+	entropy->pub.encode_mcu = encode_mcu_DC_refine;
+      else
+	entropy->pub.encode_mcu = encode_mcu_AC_refine;
+    }
+  } else
+    entropy->pub.encode_mcu = encode_mcu;
+
+  /* Allocate & initialize requested statistics areas */
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+    compptr = cinfo->cur_comp_info[ci];
+    /* DC needs no table for refinement scan */
+    if (cinfo->Ss == 0 && cinfo->Ah == 0) {
+      tbl = compptr->dc_tbl_no;
+      if (tbl < 0 || tbl >= NUM_ARITH_TBLS)
+	ERREXIT1(cinfo, JERR_NO_ARITH_TABLE, tbl);
+      if (entropy->dc_stats[tbl] == NULL)
+	entropy->dc_stats[tbl] = (unsigned char *) (*cinfo->mem->alloc_small)
+	  ((j_common_ptr) cinfo, JPOOL_IMAGE, DC_STAT_BINS);
+      MEMZERO(entropy->dc_stats[tbl], DC_STAT_BINS);
+      /* Initialize DC predictions to 0 */
+      entropy->last_dc_val[ci] = 0;
+      entropy->dc_context[ci] = 0;
+    }
+    /* AC needs no table when not present */
+    if (cinfo->Se) {
+      tbl = compptr->ac_tbl_no;
+      if (tbl < 0 || tbl >= NUM_ARITH_TBLS)
+	ERREXIT1(cinfo, JERR_NO_ARITH_TABLE, tbl);
+      if (entropy->ac_stats[tbl] == NULL)
+	entropy->ac_stats[tbl] = (unsigned char *) (*cinfo->mem->alloc_small)
+	  ((j_common_ptr) cinfo, JPOOL_IMAGE, AC_STAT_BINS);
+      MEMZERO(entropy->ac_stats[tbl], AC_STAT_BINS);
+#ifdef CALCULATE_SPECTRAL_CONDITIONING
+      if (cinfo->progressive_mode)
+	/* Section G.1.3.2: Set appropriate arithmetic conditioning value Kx */
+	cinfo->arith_ac_K[tbl] = cinfo->Ss + ((8 + cinfo->Se - cinfo->Ss) >> 4);
+#endif
+    }
+  }
+
+  /* Initialize arithmetic encoding variables */
+  entropy->c = 0;
+  entropy->a = 0x10000L;
+  entropy->sc = 0;
+  entropy->zc = 0;
+  entropy->ct = 11;
+  entropy->buffer = -1;  /* empty */
+
+  /* Initialize restart stuff */
+  entropy->restarts_to_go = cinfo->restart_interval;
+  entropy->next_restart_num = 0;
+}
+
+
+/*
+ * Module initialization routine for arithmetic entropy encoding.
+ */
+
+GLOBAL(void)
+jinit_arith_encoder (j_compress_ptr cinfo)
+{
+  arith_entropy_ptr entropy;
+  int i;
+
+  entropy = (arith_entropy_ptr) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, SIZEOF(arith_entropy_encoder));
+  cinfo->entropy = &entropy->pub;
+  entropy->pub.start_pass = start_pass;
+  entropy->pub.finish_pass = finish_pass;
+
+  /* Mark tables unallocated */
+  for (i = 0; i < NUM_ARITH_TBLS; i++) {
+    entropy->dc_stats[i] = NULL;
+    entropy->ac_stats[i] = NULL;
+  }
+
+  /* Initialize index for fixed probability estimation */
+  entropy->fixed_bin[0] = 113;
+}
@@ -0,0 +1,456 @@
+/*
+ * jccoefct.c
+ *
+ * Copyright (C) 1994-1997, Thomas G. Lane.
+ * Modified 2003-2022 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains the coefficient buffer controller for compression.
+ * This controller is the top level of the JPEG compressor proper.
+ * The coefficient buffer lies between forward-DCT and entropy encoding steps.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/* We use a full-image coefficient buffer when doing Huffman optimization,
+ * and also for writing multiple-scan JPEG files.  In all cases, the DCT
+ * step is run during the first pass, and subsequent passes need only read
+ * the buffered coefficients.
+ */
+#ifdef ENTROPY_OPT_SUPPORTED
+#define FULL_COEF_BUFFER_SUPPORTED
+#else
+#ifdef C_MULTISCAN_FILES_SUPPORTED
+#define FULL_COEF_BUFFER_SUPPORTED
+#endif
+#endif
+
+
+/* Private buffer controller object */
+
+typedef struct {
+  struct jpeg_c_coef_controller pub; /* public fields */
+
+  JDIMENSION iMCU_row_num;	/* iMCU row # within image */
+  JDIMENSION MCU_ctr;		/* counts MCUs processed in current row */
+  int MCU_vert_offset;		/* counts MCU rows within iMCU row */
+  int MCU_rows_per_iMCU_row;	/* number of such rows needed */
+
+  /* For single-pass compression, it's sufficient to buffer just one MCU
+   * (although this may prove a bit slow in practice).
+   * We append a workspace of C_MAX_BLOCKS_IN_MCU coefficient blocks,
+   * and reuse it for each MCU constructed and sent.
+   * In multi-pass modes, this array points to the current MCU's blocks
+   * within the virtual arrays.
+   */
+  JBLOCKROW MCU_buffer[C_MAX_BLOCKS_IN_MCU];
+
+  /* In multi-pass modes, we need a virtual block array for each component. */
+  jvirt_barray_ptr whole_image[MAX_COMPONENTS];
+
+  /* Workspace for single-pass compression (omitted otherwise). */
+  JBLOCK blk_buffer[C_MAX_BLOCKS_IN_MCU];
+} my_coef_controller;
+
+typedef my_coef_controller * my_coef_ptr;
+
+
+/* Forward declarations */
+METHODDEF(boolean) compress_data
+    JPP((j_compress_ptr cinfo, JSAMPIMAGE input_buf));
+#ifdef FULL_COEF_BUFFER_SUPPORTED
+METHODDEF(boolean) compress_first_pass
+    JPP((j_compress_ptr cinfo, JSAMPIMAGE input_buf));
+METHODDEF(boolean) compress_output
+    JPP((j_compress_ptr cinfo, JSAMPIMAGE input_buf));
+#endif
+
+
+LOCAL(void)
+start_iMCU_row (j_compress_ptr cinfo)
+/* Reset within-iMCU-row counters for a new row */
+{
+  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
+
+  /* In an interleaved scan, an MCU row is the same as an iMCU row.
+   * In a noninterleaved scan, an iMCU row has v_samp_factor MCU rows.
+   * But at the bottom of the image, process only what's left.
+   */
+  if (cinfo->comps_in_scan > 1) {
+    coef->MCU_rows_per_iMCU_row = 1;
+  } else {
+    if (coef->iMCU_row_num < (cinfo->total_iMCU_rows-1))
+      coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->v_samp_factor;
+    else
+      coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->last_row_height;
+  }
+
+  coef->MCU_ctr = 0;
+  coef->MCU_vert_offset = 0;
+}
+
+
+/*
+ * Initialize for a processing pass.
+ */
+
+METHODDEF(void)
+start_pass_coef (j_compress_ptr cinfo, J_BUF_MODE pass_mode)
+{
+  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
+
+  coef->iMCU_row_num = 0;
+  start_iMCU_row(cinfo);
+
+  switch (pass_mode) {
+  case JBUF_PASS_THRU:
+    if (coef->whole_image[0] != NULL)
+      ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+    coef->pub.compress_data = compress_data;
+    break;
+#ifdef FULL_COEF_BUFFER_SUPPORTED
+  case JBUF_SAVE_AND_PASS:
+    if (coef->whole_image[0] == NULL)
+      ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+    coef->pub.compress_data = compress_first_pass;
+    break;
+  case JBUF_CRANK_DEST:
+    if (coef->whole_image[0] == NULL)
+      ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+    coef->pub.compress_data = compress_output;
+    break;
+#endif
+  default:
+    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+  }
+}
+
+
+/*
+ * Process some data in the single-pass case.
+ * We process the equivalent of one fully interleaved MCU row ("iMCU" row)
+ * per call, ie, v_samp_factor block rows for each component in the image.
+ * Returns TRUE if the iMCU row is completed, FALSE if suspended.
+ *
+ * NB: input_buf contains a plane for each component in image,
+ * which we index according to the component's SOF position.
+ */
+
+METHODDEF(boolean)
+compress_data (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
+{
+  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
+  JDIMENSION MCU_col_num;	/* index of current MCU within row */
+  JDIMENSION last_MCU_col = cinfo->MCUs_per_row - 1;
+  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
+  int ci, xindex, yindex, yoffset, blockcnt;
+  JBLOCKROW blkp;
+  JSAMPARRAY input_ptr;
+  JDIMENSION xpos;
+  jpeg_component_info *compptr;
+  forward_DCT_ptr forward_DCT;
+
+  /* Loop to write as much as one whole iMCU row */
+  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
+       yoffset++) {
+    for (MCU_col_num = coef->MCU_ctr; MCU_col_num <= last_MCU_col;
+	 MCU_col_num++) {
+      /* Determine where data comes from in input_buf and do the DCT thing.
+       * Each call on forward_DCT processes a horizontal row of DCT blocks as
+       * wide as an MCU.  Dummy blocks at the right or bottom edge are filled in
+       * specially.  The data in them does not matter for image reconstruction,
+       * so we fill them with values that will encode to the smallest amount of
+       * data, viz: all zeroes in the AC entries, DC entries equal to previous
+       * block's DC value.  (Thanks to Thomas Kinsman for this idea.)
+       */
+      blkp = coef->blk_buffer;	/* pointer to current DCT block within MCU */
+      for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+	compptr = cinfo->cur_comp_info[ci];
+	forward_DCT = cinfo->fdct->forward_DCT[compptr->component_index];
+	input_ptr = input_buf[compptr->component_index] +
+	  yoffset * compptr->DCT_v_scaled_size;
+	/* ypos == (yoffset + yindex) * compptr->DCT_v_scaled_size */
+	blockcnt = (MCU_col_num < last_MCU_col) ? compptr->MCU_width
+						: compptr->last_col_width;
+	xpos = MCU_col_num * compptr->MCU_sample_width;
+	for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
+	  if (coef->iMCU_row_num < last_iMCU_row ||
+	      yoffset + yindex < compptr->last_row_height) {
+	    (*forward_DCT) (cinfo, compptr, input_ptr, blkp,
+			    xpos, (JDIMENSION) blockcnt);
+	    input_ptr += compptr->DCT_v_scaled_size;
+	    blkp += blockcnt;
+	    /* Dummy blocks at right edge */
+	    if ((xindex = compptr->MCU_width - blockcnt) == 0)
+	      continue;
+	  } else {
+	    /* At bottom of image, need a whole row of dummy blocks */
+	    xindex = compptr->MCU_width;
+	  }
+	  /* Fill in any dummy blocks needed in this row */
+	  MEMZERO(blkp, xindex * SIZEOF(JBLOCK));
+	  do {
+	    blkp[0][0] = blkp[-1][0];
+	    blkp++;
+	  } while (--xindex);
+	}
+      }
+      /* Try to write the MCU.  In event of a suspension failure, we will
+       * re-DCT the MCU on restart (a bit inefficient, could be fixed...)
+       */
+      if (! (*cinfo->entropy->encode_mcu) (cinfo, coef->MCU_buffer)) {
+	/* Suspension forced; update state counters and exit */
+	coef->MCU_vert_offset = yoffset;
+	coef->MCU_ctr = MCU_col_num;
+	return FALSE;
+      }
+    }
+    /* Completed an MCU row, but perhaps not an iMCU row */
+    coef->MCU_ctr = 0;
+  }
+  /* Completed the iMCU row, advance counters for next one */
+  coef->iMCU_row_num++;
+  start_iMCU_row(cinfo);
+  return TRUE;
+}
+
+
+#ifdef FULL_COEF_BUFFER_SUPPORTED
+
+/*
+ * Process some data in the first pass of a multi-pass case.
+ * We process the equivalent of one fully interleaved MCU row ("iMCU" row)
+ * per call, ie, v_samp_factor block rows for each component in the image.
+ * This amount of data is read from the source buffer, DCT'd and quantized,
+ * and saved into the virtual arrays.  We also generate suitable dummy blocks
+ * as needed at the right and lower edges.  (The dummy blocks are constructed
+ * in the virtual arrays, which have been padded appropriately.)  This makes
+ * it possible for subsequent passes not to worry about real vs. dummy blocks.
+ *
+ * We must also emit the data to the entropy encoder.  This is conveniently
+ * done by calling compress_output() after we've loaded the current strip
+ * of the virtual arrays.
+ *
+ * NB: input_buf contains a plane for each component in image.  All
+ * components are DCT'd and loaded into the virtual arrays in this pass.
+ * However, it may be that only a subset of the components are emitted to
+ * the entropy encoder during this first pass; be careful about looking
+ * at the scan-dependent variables (MCU dimensions, etc).
+ */
+
+METHODDEF(boolean)
+compress_first_pass (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
+{
+  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
+  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
+  JDIMENSION blocks_across, MCUs_across, MCUindex;
+  int bi, ci, h_samp_factor, block_row, block_rows, ndummy;
+  JCOEF lastDC;
+  jpeg_component_info *compptr;
+  JBLOCKARRAY buffer;
+  JBLOCKROW thisblockrow, lastblockrow;
+  JSAMPARRAY input_ptr;
+  forward_DCT_ptr forward_DCT;
+
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    /* Align the virtual buffer for this component. */
+    buffer = (*cinfo->mem->access_virt_barray)
+      ((j_common_ptr) cinfo, coef->whole_image[ci],
+       coef->iMCU_row_num * compptr->v_samp_factor,
+       (JDIMENSION) compptr->v_samp_factor, TRUE);
+    /* Count non-dummy DCT block rows in this iMCU row. */
+    if (coef->iMCU_row_num < last_iMCU_row)
+      block_rows = compptr->v_samp_factor;
+    else {
+      /* NB: can't use last_row_height here, since may not be set! */
+      block_rows = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
+      if (block_rows == 0) block_rows = compptr->v_samp_factor;
+    }
+    blocks_across = compptr->width_in_blocks;
+    h_samp_factor = compptr->h_samp_factor;
+    /* Count number of dummy blocks to be added at the right margin. */
+    ndummy = (int) (blocks_across % h_samp_factor);
+    if (ndummy > 0)
+      ndummy = h_samp_factor - ndummy;
+    forward_DCT = cinfo->fdct->forward_DCT[ci];
+    input_ptr = input_buf[ci];
+    /* Perform DCT for all non-dummy blocks in this iMCU row.  Each call
+     * on forward_DCT processes a complete horizontal row of DCT blocks.
+     */
+    for (block_row = 0; block_row < block_rows; block_row++) {
+      thisblockrow = buffer[block_row];
+      (*forward_DCT) (cinfo, compptr, input_ptr, thisblockrow,
+		      (JDIMENSION) 0, blocks_across);
+      input_ptr += compptr->DCT_v_scaled_size;
+      if (ndummy > 0) {
+	/* Create dummy blocks at the right edge of the image. */
+	thisblockrow += blocks_across; /* => first dummy block */
+	FMEMZERO((void FAR *) thisblockrow, ndummy * SIZEOF(JBLOCK));
+	lastDC = thisblockrow[-1][0];
+	for (bi = 0; bi < ndummy; bi++) {
+	  thisblockrow[bi][0] = lastDC;
+	}
+      }
+    }
+    /* If at end of image, create dummy block rows as needed.
+     * The tricky part here is that within each MCU, we want the DC values
+     * of the dummy blocks to match the last real block's DC value.
+     * This squeezes a few more bytes out of the resulting file...
+     */
+    if (block_row < compptr->v_samp_factor) {
+      blocks_across += ndummy;	/* include lower right corner */
+      MCUs_across = blocks_across / h_samp_factor;
+      do {
+	thisblockrow = buffer[block_row];
+	lastblockrow = buffer[block_row-1];
+	FMEMZERO((void FAR *) thisblockrow,
+		 (size_t) blocks_across * SIZEOF(JBLOCK));
+	for (MCUindex = 0; MCUindex < MCUs_across; MCUindex++) {
+	  lastDC = lastblockrow[h_samp_factor-1][0];
+	  for (bi = 0; bi < h_samp_factor; bi++) {
+	    thisblockrow[bi][0] = lastDC;
+	  }
+	  thisblockrow += h_samp_factor; /* advance to next MCU in row */
+	  lastblockrow += h_samp_factor;
+	}
+      } while (++block_row < compptr->v_samp_factor);
+    }
+  }
+  /* NB: compress_output will increment iMCU_row_num if successful.
+   * A suspension return will result in redoing all the work above next time.
+   */
+
+  /* Emit data to the entropy encoder, sharing code with subsequent passes */
+  return compress_output(cinfo, input_buf);
+}
+
+
+/*
+ * Process some data in subsequent passes of a multi-pass case.
+ * We process the equivalent of one fully interleaved MCU row ("iMCU" row)
+ * per call, ie, v_samp_factor block rows for each component in the scan.
+ * The data is obtained from the virtual arrays and fed to the entropy coder.
+ * Returns TRUE if the iMCU row is completed, FALSE if suspended.
+ *
+ * NB: input_buf is ignored; it is likely to be a NULL pointer.
+ */
+
+METHODDEF(boolean)
+compress_output (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
+{
+  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
+  JDIMENSION MCU_col_num;	/* index of current MCU within row */
+  int ci, xindex, yindex, yoffset;
+  JDIMENSION start_col;
+  JBLOCKARRAY blkp;
+  JBLOCKARRAY buffer[MAX_COMPS_IN_SCAN];
+  JBLOCKROW buffer_ptr;
+  jpeg_component_info *compptr;
+
+  /* Align the virtual buffers for the components used in this scan.
+   * NB: during first pass, this is safe only because the buffers will
+   * already be aligned properly, so jmemmgr.c won't need to do any I/O.
+   */
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+    compptr = cinfo->cur_comp_info[ci];
+    buffer[ci] = (*cinfo->mem->access_virt_barray)
+      ((j_common_ptr) cinfo, coef->whole_image[compptr->component_index],
+       coef->iMCU_row_num * compptr->v_samp_factor,
+       (JDIMENSION) compptr->v_samp_factor, FALSE);
+  }
+
+  /* Loop to process one whole iMCU row */
+  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
+       yoffset++) {
+    for (MCU_col_num = coef->MCU_ctr; MCU_col_num < cinfo->MCUs_per_row;
+	 MCU_col_num++) {
+      /* Construct list of pointers to DCT blocks belonging to this MCU */
+      blkp = coef->MCU_buffer;	/* pointer to current DCT block within MCU */
+      for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+	compptr = cinfo->cur_comp_info[ci];
+	start_col = MCU_col_num * compptr->MCU_width;
+	for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
+	  buffer_ptr = buffer[ci][yoffset + yindex] + start_col;
+	  xindex = compptr->MCU_width;
+	  do {
+	    *blkp++ = buffer_ptr++;
+	  } while (--xindex);
+	}
+      }
+      /* Try to write the MCU. */
+      if (! (*cinfo->entropy->encode_mcu) (cinfo, coef->MCU_buffer)) {
+	/* Suspension forced; update state counters and exit */
+	coef->MCU_vert_offset = yoffset;
+	coef->MCU_ctr = MCU_col_num;
+	return FALSE;
+      }
+    }
+    /* Completed an MCU row, but perhaps not an iMCU row */
+    coef->MCU_ctr = 0;
+  }
+  /* Completed the iMCU row, advance counters for next one */
+  coef->iMCU_row_num++;
+  start_iMCU_row(cinfo);
+  return TRUE;
+}
+
+#endif /* FULL_COEF_BUFFER_SUPPORTED */
+
+
+/*
+ * Initialize coefficient buffer controller.
+ */
+
+GLOBAL(void)
+jinit_c_coef_controller (j_compress_ptr cinfo, boolean need_full_buffer)
+{
+  my_coef_ptr coef;
+
+  if (need_full_buffer) {
+#ifdef FULL_COEF_BUFFER_SUPPORTED
+    /* Allocate a full-image virtual array for each component, */
+    /* padded to a multiple of samp_factor DCT blocks in each direction. */
+    int ci;
+    jpeg_component_info *compptr;
+
+    coef = (my_coef_ptr) (*cinfo->mem->alloc_small)
+      ((j_common_ptr) cinfo, JPOOL_IMAGE,
+       SIZEOF(my_coef_controller) - SIZEOF(coef->blk_buffer));
+    for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+	 ci++, compptr++) {
+      coef->whole_image[ci] = (*cinfo->mem->request_virt_barray)
+	((j_common_ptr) cinfo, JPOOL_IMAGE, FALSE,
+	 (JDIMENSION) jround_up((long) compptr->width_in_blocks,
+				(long) compptr->h_samp_factor),
+	 (JDIMENSION) jround_up((long) compptr->height_in_blocks,
+				(long) compptr->v_samp_factor),
+	 (JDIMENSION) compptr->v_samp_factor);
+    }
+#else
+    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+#endif
+  } else {
+    /* We only need a single-MCU buffer. */
+    JBLOCKARRAY blkp;
+    JBLOCKROW buffer_ptr;
+    int bi;
+
+    coef = (my_coef_ptr) (*cinfo->mem->alloc_small)
+      ((j_common_ptr) cinfo, JPOOL_IMAGE, SIZEOF(my_coef_controller));
+    blkp = coef->MCU_buffer;
+    buffer_ptr = coef->blk_buffer;
+    bi = C_MAX_BLOCKS_IN_MCU;
+    do {
+      *blkp++ = buffer_ptr++;
+    } while (--bi);
+    coef->whole_image[0] = NULL; /* flag for no virtual arrays */
+  }
+
+  coef->pub.start_pass = start_pass_coef;
+  cinfo->coef = &coef->pub;
+}
@@ -0,0 +1,598 @@
+/*
+ * jccolor.c
+ *
+ * Copyright (C) 1991-1996, Thomas G. Lane.
+ * Modified 2011-2023 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains input colorspace conversion routines.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/* Private subobject */
+
+typedef struct {
+  struct jpeg_color_converter pub; /* public fields */
+
+  /* Private state for RGB->YCC conversion */
+  INT32 * rgb_ycc_tab;		/* => table for RGB to YCbCr conversion */
+} my_color_converter;
+
+typedef my_color_converter * my_cconvert_ptr;
+
+
+/**************** RGB -> YCbCr conversion: most common case **************/
+
+/*
+ * YCbCr is defined per Recommendation ITU-R BT.601-7 (03/2011),
+ * previously known as Recommendation CCIR 601-1, except that Cb and Cr
+ * are normalized to the range 0..MAXJSAMPLE rather than -0.5 .. 0.5.
+ * sRGB (standard RGB color space) is defined per IEC 61966-2-1:1999.
+ * sYCC (standard luma-chroma-chroma color space with extended gamut)
+ * is defined per IEC 61966-2-1:1999 Amendment A1:2003 Annex F.
+ * bg-sRGB and bg-sYCC (big gamut standard color spaces)
+ * are defined per IEC 61966-2-1:1999 Amendment A1:2003 Annex G.
+ * Note that the derived conversion coefficients given in some of these
+ * documents are imprecise.  The general conversion equations are
+ *	Y  = Kr * R + (1 - Kr - Kb) * G + Kb * B
+ *	Cb = (B - Y) / (1 - Kb) / K
+ *	Cr = (R - Y) / (1 - Kr) / K
+ * With Kr = 0.299 and Kb = 0.114 (derived according to SMPTE RP 177-1993
+ * from the 1953 FCC NTSC primaries and CIE Illuminant C), K = 2 for sYCC,
+ * the conversion equations to be implemented are therefore
+ *	Y  =  0.299 * R + 0.587 * G + 0.114 * B
+ *	Cb = -0.168735892 * R - 0.331264108 * G + 0.5 * B + CENTERJSAMPLE
+ *	Cr =  0.5 * R - 0.418687589 * G - 0.081312411 * B + CENTERJSAMPLE
+ * Note: older versions of the IJG code used a zero offset of MAXJSAMPLE/2,
+ * rather than CENTERJSAMPLE, for Cb and Cr.  This gave equal positive and
+ * negative swings for Cb/Cr, but meant that grayscale values (Cb=Cr=0)
+ * were not represented exactly.  Now we sacrifice exact representation of
+ * maximum red and maximum blue in order to get exact grayscales.
+ *
+ * To avoid floating-point arithmetic, we represent the fractional constants
+ * as integers scaled up by 2^16 (about 4 digits precision); we have to divide
+ * the products by 2^16, with appropriate rounding, to get the correct answer.
+ *
+ * For even more speed, we avoid doing any multiplications in the inner loop
+ * by precalculating the constants times R,G,B for all possible values.
+ * For 8-bit JSAMPLEs this is very reasonable (only 256 entries per table);
+ * for 9-bit to 12-bit samples it is still acceptable.  It's not very
+ * reasonable for 16-bit samples, but if you want lossless storage
+ * you shouldn't be changing colorspace anyway.
+ * The CENTERJSAMPLE offsets and the rounding fudge-factor of 0.5 are included
+ * in the tables to save adding them separately in the inner loop.
+ */
+
+#define SCALEBITS	16	/* speediest right-shift on some machines */
+#define CBCR_OFFSET	((INT32) CENTERJSAMPLE << SCALEBITS)
+#define ONE_HALF	((INT32) 1 << (SCALEBITS-1))
+#define FIX(x)		((INT32) ((x) * (1L<<SCALEBITS) + 0.5))
+
+/* We allocate one big table and divide it up into eight parts, instead of
+ * doing eight alloc_small requests.  This lets us use a single table base
+ * address, which can be held in a register in the inner loops on many
+ * machines (more than can hold all eight addresses, anyway).
+ */
+
+#define R_Y_OFF		0			/* offset to R => Y section */
+#define G_Y_OFF		(1*(MAXJSAMPLE+1))	/* offset to G => Y section */
+#define B_Y_OFF		(2*(MAXJSAMPLE+1))	/* etc. */
+#define R_CB_OFF	(3*(MAXJSAMPLE+1))
+#define G_CB_OFF	(4*(MAXJSAMPLE+1))
+#define B_CB_OFF	(5*(MAXJSAMPLE+1))
+#define R_CR_OFF	B_CB_OFF		/* B=>Cb, R=>Cr are the same */
+#define G_CR_OFF	(6*(MAXJSAMPLE+1))
+#define B_CR_OFF	(7*(MAXJSAMPLE+1))
+#define TABLE_SIZE	(8*(MAXJSAMPLE+1))
+
+
+/*
+ * Initialize for RGB->YCC colorspace conversion.
+ */
+
+METHODDEF(void)
+rgb_ycc_start (j_compress_ptr cinfo)
+{
+  my_cconvert_ptr cconvert = (my_cconvert_ptr) cinfo->cconvert;
+  INT32 * rgb_ycc_tab;
+  INT32 i;
+
+  /* Allocate and fill in the conversion tables. */
+  cconvert->rgb_ycc_tab = rgb_ycc_tab = (INT32 *)
+    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				TABLE_SIZE * SIZEOF(INT32));
+
+  for (i = 0; i <= MAXJSAMPLE; i++) {
+    rgb_ycc_tab[i+R_Y_OFF] = FIX(0.299) * i;
+    rgb_ycc_tab[i+G_Y_OFF] = FIX(0.587) * i;
+    rgb_ycc_tab[i+B_Y_OFF] = FIX(0.114) * i + ONE_HALF;
+    rgb_ycc_tab[i+R_CB_OFF] = (- FIX(0.168735892)) * i;
+    rgb_ycc_tab[i+G_CB_OFF] = (- FIX(0.331264108)) * i;
+    /* We use a rounding fudge-factor of 0.5-epsilon for Cb and Cr.
+     * This ensures that the maximum output will round to MAXJSAMPLE
+     * not MAXJSAMPLE+1, and thus that we don't have to range-limit.
+     */
+    rgb_ycc_tab[i+B_CB_OFF] = (i << (SCALEBITS-1)) + CBCR_OFFSET + ONE_HALF-1;
+/*  B=>Cb and R=>Cr tables are the same
+    rgb_ycc_tab[i+R_CR_OFF] = (i << (SCALEBITS-1)) + CBCR_OFFSET + ONE_HALF-1;
+*/
+    rgb_ycc_tab[i+G_CR_OFF] = (- FIX(0.418687589)) * i;
+    rgb_ycc_tab[i+B_CR_OFF] = (- FIX(0.081312411)) * i;
+  }
+}
+
+
+/*
+ * Convert some rows of samples to the JPEG colorspace.
+ *
+ * Note that we change from the application's interleaved-pixel format
+ * to our internal noninterleaved, one-plane-per-component format.  The
+ * input buffer is therefore three times as wide as the output buffer.
+ *
+ * A starting row offset is provided only for the output buffer.  The
+ * caller can easily adjust the passed input_buf value to accommodate
+ * any row offset required on that side.
+ */
+
+METHODDEF(void)
+rgb_ycc_convert (j_compress_ptr cinfo,
+		 JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+		 JDIMENSION output_row, int num_rows)
+{
+  my_cconvert_ptr cconvert = (my_cconvert_ptr) cinfo->cconvert;
+  register int r, g, b;
+  register INT32 * ctab = cconvert->rgb_ycc_tab;
+  register JSAMPROW inptr;
+  register JSAMPROW outptr0, outptr1, outptr2;
+  register JDIMENSION col;
+  JDIMENSION num_cols = cinfo->image_width;
+
+  while (--num_rows >= 0) {
+    inptr = *input_buf++;
+    outptr0 = output_buf[0][output_row];
+    outptr1 = output_buf[1][output_row];
+    outptr2 = output_buf[2][output_row];
+    output_row++;
+    for (col = 0; col < num_cols; col++) {
+      r = GETJSAMPLE(inptr[RGB_RED]);
+      g = GETJSAMPLE(inptr[RGB_GREEN]);
+      b = GETJSAMPLE(inptr[RGB_BLUE]);
+      inptr += RGB_PIXELSIZE;
+      /* If the inputs are 0..MAXJSAMPLE, the outputs of these equations
+       * must be too; we do not need an explicit range-limiting operation.
+       * Hence the value being shifted is never negative, and we don't
+       * need the general RIGHT_SHIFT macro.
+       */
+      /* Y */
+      outptr0[col] = (JSAMPLE)
+		((ctab[r+R_Y_OFF] + ctab[g+G_Y_OFF] + ctab[b+B_Y_OFF])
+		 >> SCALEBITS);
+      /* Cb */
+      outptr1[col] = (JSAMPLE)
+		((ctab[r+R_CB_OFF] + ctab[g+G_CB_OFF] + ctab[b+B_CB_OFF])
+		 >> SCALEBITS);
+      /* Cr */
+      outptr2[col] = (JSAMPLE)
+		((ctab[r+R_CR_OFF] + ctab[g+G_CR_OFF] + ctab[b+B_CR_OFF])
+		 >> SCALEBITS);
+    }
+  }
+}
+
+
+/**************** Cases other than RGB -> YCbCr **************/
+
+
+/*
+ * Convert some rows of samples to the JPEG colorspace.
+ * This version handles RGB->grayscale conversion,
+ * which is the same as the RGB->Y portion of RGB->YCbCr.
+ * We assume rgb_ycc_start has been called (we only use the Y tables).
+ */
+
+METHODDEF(void)
+rgb_gray_convert (j_compress_ptr cinfo,
+		  JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+		  JDIMENSION output_row, int num_rows)
+{
+  my_cconvert_ptr cconvert = (my_cconvert_ptr) cinfo->cconvert;
+  register INT32 y;
+  register INT32 * ctab = cconvert->rgb_ycc_tab;
+  register JSAMPROW inptr;
+  register JSAMPROW outptr;
+  register JDIMENSION col;
+  JDIMENSION num_cols = cinfo->image_width;
+
+  while (--num_rows >= 0) {
+    inptr = *input_buf++;
+    outptr = output_buf[0][output_row++];
+    for (col = 0; col < num_cols; col++) {
+      y  = ctab[R_Y_OFF + GETJSAMPLE(inptr[RGB_RED])];
+      y += ctab[G_Y_OFF + GETJSAMPLE(inptr[RGB_GREEN])];
+      y += ctab[B_Y_OFF + GETJSAMPLE(inptr[RGB_BLUE])];
+      inptr += RGB_PIXELSIZE;
+      outptr[col] = (JSAMPLE) (y >> SCALEBITS);
+    }
+  }
+}
+
+
+/*
+ * Convert some rows of samples to the JPEG colorspace.
+ * This version handles Adobe-style CMYK->YCCK conversion,
+ * where we convert R=1-C, G=1-M, and B=1-Y to YCbCr using the
+ * same conversion as above, while passing K (black) unchanged.
+ * We assume rgb_ycc_start has been called.
+ */
+
+METHODDEF(void)
+cmyk_ycck_convert (j_compress_ptr cinfo,
+		   JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+		   JDIMENSION output_row, int num_rows)
+{
+  my_cconvert_ptr cconvert = (my_cconvert_ptr) cinfo->cconvert;
+  register int r, g, b;
+  register INT32 * ctab = cconvert->rgb_ycc_tab;
+  register JSAMPROW inptr;
+  register JSAMPROW outptr0, outptr1, outptr2, outptr3;
+  register JDIMENSION col;
+  JDIMENSION num_cols = cinfo->image_width;
+
+  while (--num_rows >= 0) {
+    inptr = *input_buf++;
+    outptr0 = output_buf[0][output_row];
+    outptr1 = output_buf[1][output_row];
+    outptr2 = output_buf[2][output_row];
+    outptr3 = output_buf[3][output_row];
+    output_row++;
+    for (col = 0; col < num_cols; col++) {
+      r = MAXJSAMPLE - GETJSAMPLE(inptr[0]);
+      g = MAXJSAMPLE - GETJSAMPLE(inptr[1]);
+      b = MAXJSAMPLE - GETJSAMPLE(inptr[2]);
+      /* K passes through as-is */
+      outptr3[col] = inptr[3];	/* don't need GETJSAMPLE here */
+      inptr += 4;
+      /* If the inputs are 0..MAXJSAMPLE, the outputs of these equations
+       * must be too; we do not need an explicit range-limiting operation.
+       * Hence the value being shifted is never negative, and we don't
+       * need the general RIGHT_SHIFT macro.
+       */
+      /* Y */
+      outptr0[col] = (JSAMPLE)
+		((ctab[r+R_Y_OFF] + ctab[g+G_Y_OFF] + ctab[b+B_Y_OFF])
+		 >> SCALEBITS);
+      /* Cb */
+      outptr1[col] = (JSAMPLE)
+		((ctab[r+R_CB_OFF] + ctab[g+G_CB_OFF] + ctab[b+B_CB_OFF])
+		 >> SCALEBITS);
+      /* Cr */
+      outptr2[col] = (JSAMPLE)
+		((ctab[r+R_CR_OFF] + ctab[g+G_CR_OFF] + ctab[b+B_CR_OFF])
+		 >> SCALEBITS);
+    }
+  }
+}
+
+
+/*
+ * Convert some rows of samples to the JPEG colorspace.
+ * [R,G,B] to [R-G,G,B-G] conversion with modulo calculation
+ * (forward reversible color transform).
+ * This can be seen as an adaption of the general RGB->YCbCr
+ * conversion equation with Kr = Kb = 0, while replacing the
+ * normalization by modulo calculation.
+ */
+
+METHODDEF(void)
+rgb_rgb1_convert (j_compress_ptr cinfo,
+		  JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+		  JDIMENSION output_row, int num_rows)
+{
+  register int r, g, b;
+  register JSAMPROW inptr;
+  register JSAMPROW outptr0, outptr1, outptr2;
+  register JDIMENSION col;
+  JDIMENSION num_cols = cinfo->image_width;
+
+  while (--num_rows >= 0) {
+    inptr = *input_buf++;
+    outptr0 = output_buf[0][output_row];
+    outptr1 = output_buf[1][output_row];
+    outptr2 = output_buf[2][output_row];
+    output_row++;
+    for (col = 0; col < num_cols; col++) {
+      r = GETJSAMPLE(inptr[RGB_RED]);
+      g = GETJSAMPLE(inptr[RGB_GREEN]);
+      b = GETJSAMPLE(inptr[RGB_BLUE]);
+      inptr += RGB_PIXELSIZE;
+      /* Assume that MAXJSAMPLE+1 is a power of 2, so that the MOD
+       * (modulo) operator is equivalent to the bitmask operator AND.
+       */
+      outptr0[col] = (JSAMPLE) ((r - g + CENTERJSAMPLE) & MAXJSAMPLE);
+      outptr1[col] = (JSAMPLE) g;
+      outptr2[col] = (JSAMPLE) ((b - g + CENTERJSAMPLE) & MAXJSAMPLE);
+    }
+  }
+}
+
+
+/*
+ * Convert some rows of samples to the JPEG colorspace.
+ * This version handles grayscale output with no conversion.
+ * The source can be either plain grayscale or YCC (since Y == gray).
+ */
+
+METHODDEF(void)
+grayscale_convert (j_compress_ptr cinfo,
+		   JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+		   JDIMENSION output_row, int num_rows)
+{
+  register JSAMPROW inptr;
+  register JSAMPROW outptr;
+  register JDIMENSION count;
+  register int instride = cinfo->input_components;
+  JDIMENSION num_cols = cinfo->image_width;
+
+  while (--num_rows >= 0) {
+    inptr = *input_buf++;
+    outptr = output_buf[0][output_row++];
+    for (count = num_cols; count > 0; count--) {
+      *outptr++ = *inptr;	/* don't need GETJSAMPLE() here */
+      inptr += instride;
+    }
+  }
+}
+
+
+/*
+ * Convert some rows of samples to the JPEG colorspace.
+ * No colorspace conversion, but change from interleaved
+ * to separate-planes representation.
+ */
+
+METHODDEF(void)
+rgb_convert (j_compress_ptr cinfo,
+	     JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+	     JDIMENSION output_row, int num_rows)
+{
+  register JSAMPROW inptr;
+  register JSAMPROW outptr0, outptr1, outptr2;
+  register JDIMENSION col;
+  JDIMENSION num_cols = cinfo->image_width;
+
+  while (--num_rows >= 0) {
+    inptr = *input_buf++;
+    outptr0 = output_buf[0][output_row];
+    outptr1 = output_buf[1][output_row];
+    outptr2 = output_buf[2][output_row];
+    output_row++;
+    for (col = 0; col < num_cols; col++) {
+      /* We can dispense with GETJSAMPLE() here */
+      outptr0[col] = inptr[RGB_RED];
+      outptr1[col] = inptr[RGB_GREEN];
+      outptr2[col] = inptr[RGB_BLUE];
+      inptr += RGB_PIXELSIZE;
+    }
+  }
+}
+
+
+/*
+ * Convert some rows of samples to the JPEG colorspace.
+ * This version handles multi-component colorspaces without conversion.
+ * We assume input_components == num_components.
+ */
+
+METHODDEF(void)
+null_convert (j_compress_ptr cinfo,
+	      JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+	      JDIMENSION output_row, int num_rows)
+{
+  register JSAMPROW inptr;
+  register JSAMPROW outptr;
+  register JDIMENSION count;
+  register int num_comps = cinfo->num_components;
+  JDIMENSION num_cols = cinfo->image_width;
+  int ci;
+
+  while (--num_rows >= 0) {
+    /* It seems fastest to make a separate pass for each component. */
+    for (ci = 0; ci < num_comps; ci++) {
+      inptr = input_buf[0] + ci;
+      outptr = output_buf[ci][output_row];
+      for (count = num_cols; count > 0; count--) {
+	*outptr++ = *inptr;	/* don't need GETJSAMPLE() here */
+	inptr += num_comps;
+      }
+    }
+    input_buf++;
+    output_row++;
+  }
+}
+
+
+/*
+ * Empty method for start_pass.
+ */
+
+METHODDEF(void)
+null_method (j_compress_ptr cinfo)
+{
+  /* no work needed */
+}
+
+
+/*
+ * Module initialization routine for input colorspace conversion.
+ */
+
+GLOBAL(void)
+jinit_color_converter (j_compress_ptr cinfo)
+{
+  my_cconvert_ptr cconvert;
+
+  cconvert = (my_cconvert_ptr) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, SIZEOF(my_color_converter));
+  cinfo->cconvert = &cconvert->pub;
+  /* set start_pass to null method until we find out differently */
+  cconvert->pub.start_pass = null_method;
+
+  /* Make sure input_components agrees with in_color_space */
+  switch (cinfo->in_color_space) {
+  case JCS_GRAYSCALE:
+    if (cinfo->input_components != 1)
+      ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE);
+    break;
+
+  case JCS_RGB:
+  case JCS_BG_RGB:
+#if RGB_PIXELSIZE != 3
+    if (cinfo->input_components != RGB_PIXELSIZE)
+      ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE);
+    break;
+#endif /* else share code with YCbCr */
+
+  case JCS_YCbCr:
+  case JCS_BG_YCC:
+    if (cinfo->input_components != 3)
+      ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE);
+    break;
+
+  case JCS_CMYK:
+  case JCS_YCCK:
+    if (cinfo->input_components != 4)
+      ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE);
+    break;
+
+  default:			/* JCS_UNKNOWN can be anything */
+    if (cinfo->input_components < 1)
+      ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE);
+  }
+
+  /* Support color transform only for RGB colorspaces */
+  if (cinfo->color_transform &&
+      cinfo->jpeg_color_space != JCS_RGB &&
+      cinfo->jpeg_color_space != JCS_BG_RGB)
+    ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+
+  /* Check num_components, set conversion method based on requested space */
+  switch (cinfo->jpeg_color_space) {
+  case JCS_GRAYSCALE:
+    if (cinfo->num_components != 1)
+      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
+    switch (cinfo->in_color_space) {
+    case JCS_GRAYSCALE:
+    case JCS_YCbCr:
+    case JCS_BG_YCC:
+      cconvert->pub.color_convert = grayscale_convert;
+      break;
+    case JCS_RGB:
+      cconvert->pub.start_pass = rgb_ycc_start;
+      cconvert->pub.color_convert = rgb_gray_convert;
+      break;
+    default:
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    }
+    break;
+
+  case JCS_RGB:
+  case JCS_BG_RGB:
+    if (cinfo->num_components != 3)
+      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
+    if (cinfo->in_color_space != cinfo->jpeg_color_space)
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    switch (cinfo->color_transform) {
+    case JCT_NONE:
+      cconvert->pub.color_convert = rgb_convert;
+      break;
+    case JCT_SUBTRACT_GREEN:
+      cconvert->pub.color_convert = rgb_rgb1_convert;
+      break;
+    default:
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    }
+    break;
+
+  case JCS_YCbCr:
+    if (cinfo->num_components != 3)
+      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
+    switch (cinfo->in_color_space) {
+    case JCS_RGB:
+      cconvert->pub.start_pass = rgb_ycc_start;
+      cconvert->pub.color_convert = rgb_ycc_convert;
+      break;
+    case JCS_YCbCr:
+      cconvert->pub.color_convert = null_convert;
+      break;
+    default:
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    }
+    break;
+
+  case JCS_BG_YCC:
+    if (cinfo->num_components != 3)
+      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
+    switch (cinfo->in_color_space) {
+    case JCS_RGB:
+      /* For conversion from normal RGB input to BG_YCC representation,
+       * the Cb/Cr values are first computed as usual, and then
+       * quantized further after DCT processing by a factor of
+       * 2 in reference to the nominal quantization factor.
+       */
+      /* need quantization scale by factor of 2 after DCT */
+      cinfo->comp_info[1].component_needed = TRUE;
+      cinfo->comp_info[2].component_needed = TRUE;
+      /* compute normal YCC first */
+      cconvert->pub.start_pass = rgb_ycc_start;
+      cconvert->pub.color_convert = rgb_ycc_convert;
+      break;
+    case JCS_YCbCr:
+      /* need quantization scale by factor of 2 after DCT */
+      cinfo->comp_info[1].component_needed = TRUE;
+      cinfo->comp_info[2].component_needed = TRUE;
+      /*FALLTHROUGH*/
+    case JCS_BG_YCC:
+      /* Pass through for BG_YCC input */
+      cconvert->pub.color_convert = null_convert;
+      break;
+    default:
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    }
+    break;
+
+  case JCS_CMYK:
+    if (cinfo->num_components != 4)
+      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
+    if (cinfo->in_color_space != JCS_CMYK)
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    cconvert->pub.color_convert = null_convert;
+    break;
+
+  case JCS_YCCK:
+    if (cinfo->num_components != 4)
+      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
+    switch (cinfo->in_color_space) {
+    case JCS_CMYK:
+      cconvert->pub.start_pass = rgb_ycc_start;
+      cconvert->pub.color_convert = cmyk_ycck_convert;
+      break;
+    case JCS_YCCK:
+      cconvert->pub.color_convert = null_convert;
+      break;
+    default:
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    }
+    break;
+
+  default:			/* allow null conversion of JCS_UNKNOWN */
+    if (cinfo->jpeg_color_space != cinfo->in_color_space ||
+	cinfo->num_components != cinfo->input_components)
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    cconvert->pub.color_convert = null_convert;
+  }
+}
@@ -0,0 +1,466 @@
+/*
+ * jcdctmgr.c
+ *
+ * Copyright (C) 1994-1996, Thomas G. Lane.
+ * Modified 2003-2020 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains the forward-DCT management logic.
+ * This code selects a particular DCT implementation to be used,
+ * and it performs related housekeeping chores including coefficient
+ * quantization.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jdct.h"		/* Private declarations for DCT subsystem */
+
+
+/* Private subobject for this module */
+
+typedef struct {
+  struct jpeg_forward_dct pub;	/* public fields */
+
+  /* Pointer to the DCT routine actually in use */
+  forward_DCT_method_ptr do_dct[MAX_COMPONENTS];
+
+#ifdef DCT_FLOAT_SUPPORTED
+  /* Same as above for the floating-point case. */
+  float_DCT_method_ptr do_float_dct[MAX_COMPONENTS];
+#endif
+} my_fdct_controller;
+
+typedef my_fdct_controller * my_fdct_ptr;
+
+
+/* The allocated post-DCT divisor tables -- big enough for any
+ * supported variant and not identical to the quant table entries,
+ * because of scaling (especially for an unnormalized DCT) --
+ * are pointed to by dct_table in the per-component comp_info
+ * structures.  Each table is given in normal array order.
+ */
+
+typedef union {
+  DCTELEM int_array[DCTSIZE2];
+#ifdef DCT_FLOAT_SUPPORTED
+  FAST_FLOAT float_array[DCTSIZE2];
+#endif
+} divisor_table;
+
+
+/* The current scaled-DCT routines require ISLOW-style divisor tables,
+ * so be sure to compile that code if either ISLOW or SCALING is requested.
+ */
+#ifdef DCT_ISLOW_SUPPORTED
+#define PROVIDE_ISLOW_TABLES
+#else
+#ifdef DCT_SCALING_SUPPORTED
+#define PROVIDE_ISLOW_TABLES
+#endif
+#endif
+
+
+/*
+ * Perform forward DCT on one or more blocks of a component.
+ *
+ * The input samples are taken from the sample_data[] array starting at
+ * position start_col, and moving to the right for any additional blocks.
+ * The quantized coefficients are returned in coef_blocks[].
+ */
+
+METHODDEF(void)
+forward_DCT (j_compress_ptr cinfo, jpeg_component_info * compptr,
+	     JSAMPARRAY sample_data, JBLOCKROW coef_blocks,
+	     JDIMENSION start_col, JDIMENSION num_blocks)
+/* This version is used for integer DCT implementations. */
+{
+  /* This routine is heavily used, so it's worth coding it tightly. */
+  my_fdct_ptr fdct = (my_fdct_ptr) cinfo->fdct;
+  forward_DCT_method_ptr do_dct = fdct->do_dct[compptr->component_index];
+  DCTELEM * divisors = (DCTELEM *) compptr->dct_table;
+  DCTELEM workspace[DCTSIZE2];	/* work area for FDCT subroutine */
+  JDIMENSION bi;
+
+  for (bi = 0; bi < num_blocks; bi++, start_col += compptr->DCT_h_scaled_size) {
+    /* Perform the DCT */
+    (*do_dct) (workspace, sample_data, start_col);
+
+    /* Quantize/descale the coefficients, and store into coef_blocks[] */
+    { register DCTELEM temp, qval;
+      register int i;
+      register JCOEFPTR output_ptr = coef_blocks[bi];
+
+      for (i = 0; i < DCTSIZE2; i++) {
+	qval = divisors[i];
+	temp = workspace[i];
+	/* Divide the coefficient value by qval, ensuring proper rounding.
+	 * Since C does not specify the direction of rounding for negative
+	 * quotients, we have to force the dividend positive for portability.
+	 *
+	 * In most files, at least half of the output values will be zero
+	 * (at default quantization settings, more like three-quarters...)
+	 * so we should ensure that this case is fast.  On many machines,
+	 * a comparison is enough cheaper than a divide to make a special test
+	 * a win.  Since both inputs will be nonnegative, we need only test
+	 * for a < b to discover whether a/b is 0.
+	 * If your machine's division is fast enough, define FAST_DIVIDE.
+	 */
+#ifdef FAST_DIVIDE
+#define DIVIDE_BY(a,b)	a /= b
+#else
+#define DIVIDE_BY(a,b)	if (a >= b) a /= b; else a = 0
+#endif
+	if (temp < 0) {
+	  temp = -temp;
+	  temp += qval>>1;	/* for rounding */
+	  DIVIDE_BY(temp, qval);
+	  temp = -temp;
+	} else {
+	  temp += qval>>1;	/* for rounding */
+	  DIVIDE_BY(temp, qval);
+	}
+	output_ptr[i] = (JCOEF) temp;
+      }
+    }
+  }
+}
+
+
+#ifdef DCT_FLOAT_SUPPORTED
+
+METHODDEF(void)
+forward_DCT_float (j_compress_ptr cinfo, jpeg_component_info * compptr,
+		   JSAMPARRAY sample_data, JBLOCKROW coef_blocks,
+		   JDIMENSION start_col, JDIMENSION num_blocks)
+/* This version is used for floating-point DCT implementations. */
+{
+  /* This routine is heavily used, so it's worth coding it tightly. */
+  my_fdct_ptr fdct = (my_fdct_ptr) cinfo->fdct;
+  float_DCT_method_ptr do_dct = fdct->do_float_dct[compptr->component_index];
+  FAST_FLOAT * divisors = (FAST_FLOAT *) compptr->dct_table;
+  FAST_FLOAT workspace[DCTSIZE2]; /* work area for FDCT subroutine */
+  JDIMENSION bi;
+
+  for (bi = 0; bi < num_blocks; bi++, start_col += compptr->DCT_h_scaled_size) {
+    /* Perform the DCT */
+    (*do_dct) (workspace, sample_data, start_col);
+
+    /* Quantize/descale the coefficients, and store into coef_blocks[] */
+    { register FAST_FLOAT temp;
+      register int i;
+      register JCOEFPTR output_ptr = coef_blocks[bi];
+
+      for (i = 0; i < DCTSIZE2; i++) {
+	/* Apply the quantization and scaling factor */
+	temp = workspace[i] * divisors[i];
+	/* Round to nearest integer.
+	 * Since C does not specify the direction of rounding for negative
+	 * quotients, we have to force the dividend positive for portability.
+	 * The maximum coefficient size is +-16K (for 12-bit data), so this
+	 * code should work for either 16-bit or 32-bit ints.
+	 */
+	output_ptr[i] = (JCOEF) ((int) (temp + (FAST_FLOAT) 16384.5) - 16384);
+      }
+    }
+  }
+}
+
+#endif /* DCT_FLOAT_SUPPORTED */
+
+
+/*
+ * Initialize for a processing pass.
+ * Verify that all referenced Q-tables are present, and set up
+ * the divisor table for each one.
+ * In the current implementation, DCT of all components is done during
+ * the first pass, even if only some components will be output in the
+ * first scan.  Hence all components should be examined here.
+ */
+
+METHODDEF(void)
+start_pass_fdctmgr (j_compress_ptr cinfo)
+{
+  my_fdct_ptr fdct = (my_fdct_ptr) cinfo->fdct;
+  int ci, qtblno, i;
+  jpeg_component_info *compptr;
+  int method = 0;
+  JQUANT_TBL * qtbl;
+  DCTELEM * dtbl;
+
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    /* Select the proper DCT routine for this component's scaling */
+    switch ((compptr->DCT_h_scaled_size << 8) + compptr->DCT_v_scaled_size) {
+#ifdef DCT_SCALING_SUPPORTED
+    case ((1 << 8) + 1):
+      fdct->do_dct[ci] = jpeg_fdct_1x1;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((2 << 8) + 2):
+      fdct->do_dct[ci] = jpeg_fdct_2x2;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((3 << 8) + 3):
+      fdct->do_dct[ci] = jpeg_fdct_3x3;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((4 << 8) + 4):
+      fdct->do_dct[ci] = jpeg_fdct_4x4;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((5 << 8) + 5):
+      fdct->do_dct[ci] = jpeg_fdct_5x5;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((6 << 8) + 6):
+      fdct->do_dct[ci] = jpeg_fdct_6x6;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((7 << 8) + 7):
+      fdct->do_dct[ci] = jpeg_fdct_7x7;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((9 << 8) + 9):
+      fdct->do_dct[ci] = jpeg_fdct_9x9;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((10 << 8) + 10):
+      fdct->do_dct[ci] = jpeg_fdct_10x10;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((11 << 8) + 11):
+      fdct->do_dct[ci] = jpeg_fdct_11x11;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((12 << 8) + 12):
+      fdct->do_dct[ci] = jpeg_fdct_12x12;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((13 << 8) + 13):
+      fdct->do_dct[ci] = jpeg_fdct_13x13;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((14 << 8) + 14):
+      fdct->do_dct[ci] = jpeg_fdct_14x14;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((15 << 8) + 15):
+      fdct->do_dct[ci] = jpeg_fdct_15x15;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((16 << 8) + 16):
+      fdct->do_dct[ci] = jpeg_fdct_16x16;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((16 << 8) + 8):
+      fdct->do_dct[ci] = jpeg_fdct_16x8;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((14 << 8) + 7):
+      fdct->do_dct[ci] = jpeg_fdct_14x7;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((12 << 8) + 6):
+      fdct->do_dct[ci] = jpeg_fdct_12x6;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((10 << 8) + 5):
+      fdct->do_dct[ci] = jpeg_fdct_10x5;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((8 << 8) + 4):
+      fdct->do_dct[ci] = jpeg_fdct_8x4;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((6 << 8) + 3):
+      fdct->do_dct[ci] = jpeg_fdct_6x3;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((4 << 8) + 2):
+      fdct->do_dct[ci] = jpeg_fdct_4x2;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((2 << 8) + 1):
+      fdct->do_dct[ci] = jpeg_fdct_2x1;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((8 << 8) + 16):
+      fdct->do_dct[ci] = jpeg_fdct_8x16;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((7 << 8) + 14):
+      fdct->do_dct[ci] = jpeg_fdct_7x14;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((6 << 8) + 12):
+      fdct->do_dct[ci] = jpeg_fdct_6x12;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((5 << 8) + 10):
+      fdct->do_dct[ci] = jpeg_fdct_5x10;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((4 << 8) + 8):
+      fdct->do_dct[ci] = jpeg_fdct_4x8;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((3 << 8) + 6):
+      fdct->do_dct[ci] = jpeg_fdct_3x6;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((2 << 8) + 4):
+      fdct->do_dct[ci] = jpeg_fdct_2x4;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((1 << 8) + 2):
+      fdct->do_dct[ci] = jpeg_fdct_1x2;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+#endif
+    case ((DCTSIZE << 8) + DCTSIZE):
+      switch (cinfo->dct_method) {
+#ifdef DCT_ISLOW_SUPPORTED
+      case JDCT_ISLOW:
+	fdct->do_dct[ci] = jpeg_fdct_islow;
+	method = JDCT_ISLOW;
+	break;
+#endif
+#ifdef DCT_IFAST_SUPPORTED
+      case JDCT_IFAST:
+	fdct->do_dct[ci] = jpeg_fdct_ifast;
+	method = JDCT_IFAST;
+	break;
+#endif
+#ifdef DCT_FLOAT_SUPPORTED
+      case JDCT_FLOAT:
+	fdct->do_float_dct[ci] = jpeg_fdct_float;
+	method = JDCT_FLOAT;
+	break;
+#endif
+      default:
+	ERREXIT(cinfo, JERR_NOT_COMPILED);
+      }
+      break;
+    default:
+      ERREXIT2(cinfo, JERR_BAD_DCTSIZE,
+	       compptr->DCT_h_scaled_size, compptr->DCT_v_scaled_size);
+    }
+    qtblno = compptr->quant_tbl_no;
+    /* Make sure specified quantization table is present */
+    if (qtblno < 0 || qtblno >= NUM_QUANT_TBLS ||
+	cinfo->quant_tbl_ptrs[qtblno] == NULL)
+      ERREXIT1(cinfo, JERR_NO_QUANT_TABLE, qtblno);
+    qtbl = cinfo->quant_tbl_ptrs[qtblno];
+    /* Create divisor table from quant table */
+    switch (method) {
+#ifdef PROVIDE_ISLOW_TABLES
+    case JDCT_ISLOW:
+      /* For LL&M IDCT method, divisors are equal to raw quantization
+       * coefficients multiplied by 8 (to counteract scaling).
+       */
+      dtbl = (DCTELEM *) compptr->dct_table;
+      for (i = 0; i < DCTSIZE2; i++) {
+	dtbl[i] =
+	  ((DCTELEM) qtbl->quantval[i]) << (compptr->component_needed ? 4 : 3);
+      }
+      fdct->pub.forward_DCT[ci] = forward_DCT;
+      break;
+#endif
+#ifdef DCT_IFAST_SUPPORTED
+    case JDCT_IFAST:
+      {
+	/* For AA&N IDCT method, divisors are equal to quantization
+	 * coefficients scaled by scalefactor[row]*scalefactor[col], where
+	 *   scalefactor[0] = 1
+	 *   scalefactor[k] = cos(k*PI/16) * sqrt(2)    for k=1..7
+	 * We apply a further scale factor of 8.
+	 */
+#define CONST_BITS 14
+	static const INT16 aanscales[DCTSIZE2] = {
+	  /* precomputed values scaled up by 14 bits */
+	  16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
+	  22725, 31521, 29692, 26722, 22725, 17855, 12299,  6270,
+	  21407, 29692, 27969, 25172, 21407, 16819, 11585,  5906,
+	  19266, 26722, 25172, 22654, 19266, 15137, 10426,  5315,
+	  16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
+	  12873, 17855, 16819, 15137, 12873, 10114,  6967,  3552,
+	   8867, 12299, 11585, 10426,  8867,  6967,  4799,  2446,
+	   4520,  6270,  5906,  5315,  4520,  3552,  2446,  1247
+	};
+	SHIFT_TEMPS
+
+	dtbl = (DCTELEM *) compptr->dct_table;
+	for (i = 0; i < DCTSIZE2; i++) {
+	  dtbl[i] = (DCTELEM)
+	    DESCALE(MULTIPLY16V16((INT32) qtbl->quantval[i],
+				  (INT32) aanscales[i]),
+		    compptr->component_needed ? CONST_BITS-4 : CONST_BITS-3);
+	}
+      }
+      fdct->pub.forward_DCT[ci] = forward_DCT;
+      break;
+#endif
+#ifdef DCT_FLOAT_SUPPORTED
+    case JDCT_FLOAT:
+      {
+	/* For float AA&N IDCT method, divisors are equal to quantization
+	 * coefficients scaled by scalefactor[row]*scalefactor[col], where
+	 *   scalefactor[0] = 1
+	 *   scalefactor[k] = cos(k*PI/16) * sqrt(2)    for k=1..7
+	 * We apply a further scale factor of 8.
+	 * What's actually stored is 1/divisor so that the inner loop can
+	 * use a multiplication rather than a division.
+	 */
+	FAST_FLOAT * fdtbl = (FAST_FLOAT *) compptr->dct_table;
+	int row, col;
+	static const double aanscalefactor[DCTSIZE] = {
+	  1.0, 1.387039845, 1.306562965, 1.175875602,
+	  1.0, 0.785694958, 0.541196100, 0.275899379
+	};
+
+	i = 0;
+	for (row = 0; row < DCTSIZE; row++) {
+	  for (col = 0; col < DCTSIZE; col++) {
+	    fdtbl[i] = (FAST_FLOAT)
+	      (1.0 / ((double) qtbl->quantval[i] *
+		      aanscalefactor[row] * aanscalefactor[col] *
+		      (compptr->component_needed ? 16.0 : 8.0)));
+	    i++;
+	  }
+	}
+      }
+      fdct->pub.forward_DCT[ci] = forward_DCT_float;
+      break;
+#endif
+    default:
+      ERREXIT(cinfo, JERR_NOT_COMPILED);
+    }
+  }
+}
+
+
+/*
+ * Initialize FDCT manager.
+ */
+
+GLOBAL(void)
+jinit_forward_dct (j_compress_ptr cinfo)
+{
+  my_fdct_ptr fdct;
+  int ci;
+  jpeg_component_info *compptr;
+
+  fdct = (my_fdct_ptr) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, SIZEOF(my_fdct_controller));
+  cinfo->fdct = &fdct->pub;
+  fdct->pub.start_pass = start_pass_fdctmgr;
+
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    /* Allocate a divisor table for each component */
+    compptr->dct_table = (*cinfo->mem->alloc_small)
+      ((j_common_ptr) cinfo, JPOOL_IMAGE, SIZEOF(divisor_table));
+  }
+}
@@ -0,0 +1,249 @@
+/*
+ * jcinit.c
+ *
+ * Copyright (C) 1991-1997, Thomas G. Lane.
+ * Modified 2003-2017 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains initialization logic for the JPEG compressor.
+ * This routine is in charge of selecting the modules to be executed and
+ * making an initialization call to each one.
+ *
+ * Logically, this code belongs in jcmaster.c.  It's split out because
+ * linking this routine implies linking the entire compression library.
+ * For a transcoding-only application, we want to be able to use jcmaster.c
+ * without linking in the whole library.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/*
+ * Compute JPEG image dimensions and related values.
+ * NOTE: this is exported for possible use by application.
+ * Hence it mustn't do anything that can't be done twice.
+ */
+
+GLOBAL(void)
+jpeg_calc_jpeg_dimensions (j_compress_ptr cinfo)
+/* Do computations that are needed before master selection phase */
+{
+  /* Sanity check on input image dimensions to prevent overflow in
+   * following calculations.
+   * We do check jpeg_width and jpeg_height in initial_setup in jcmaster.c,
+   * but image_width and image_height can come from arbitrary data,
+   * and we need some space for multiplication by block_size.
+   */
+  if (((long) cinfo->image_width >> 24) || ((long) cinfo->image_height >> 24))
+    ERREXIT1(cinfo, JERR_IMAGE_TOO_BIG, (unsigned int) JPEG_MAX_DIMENSION);
+
+#ifdef DCT_SCALING_SUPPORTED
+
+  /* Compute actual JPEG image dimensions and DCT scaling choices. */
+  if (cinfo->scale_num >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/1 scaling */
+    cinfo->jpeg_width = cinfo->image_width * cinfo->block_size;
+    cinfo->jpeg_height = cinfo->image_height * cinfo->block_size;
+    cinfo->min_DCT_h_scaled_size = 1;
+    cinfo->min_DCT_v_scaled_size = 1;
+  } else if (cinfo->scale_num * 2 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/2 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 2L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 2L);
+    cinfo->min_DCT_h_scaled_size = 2;
+    cinfo->min_DCT_v_scaled_size = 2;
+  } else if (cinfo->scale_num * 3 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/3 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 3L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 3L);
+    cinfo->min_DCT_h_scaled_size = 3;
+    cinfo->min_DCT_v_scaled_size = 3;
+  } else if (cinfo->scale_num * 4 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/4 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 4L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 4L);
+    cinfo->min_DCT_h_scaled_size = 4;
+    cinfo->min_DCT_v_scaled_size = 4;
+  } else if (cinfo->scale_num * 5 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/5 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 5L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 5L);
+    cinfo->min_DCT_h_scaled_size = 5;
+    cinfo->min_DCT_v_scaled_size = 5;
+  } else if (cinfo->scale_num * 6 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/6 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 6L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 6L);
+    cinfo->min_DCT_h_scaled_size = 6;
+    cinfo->min_DCT_v_scaled_size = 6;
+  } else if (cinfo->scale_num * 7 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/7 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 7L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 7L);
+    cinfo->min_DCT_h_scaled_size = 7;
+    cinfo->min_DCT_v_scaled_size = 7;
+  } else if (cinfo->scale_num * 8 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/8 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 8L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 8L);
+    cinfo->min_DCT_h_scaled_size = 8;
+    cinfo->min_DCT_v_scaled_size = 8;
+  } else if (cinfo->scale_num * 9 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/9 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 9L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 9L);
+    cinfo->min_DCT_h_scaled_size = 9;
+    cinfo->min_DCT_v_scaled_size = 9;
+  } else if (cinfo->scale_num * 10 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/10 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 10L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 10L);
+    cinfo->min_DCT_h_scaled_size = 10;
+    cinfo->min_DCT_v_scaled_size = 10;
+  } else if (cinfo->scale_num * 11 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/11 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 11L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 11L);
+    cinfo->min_DCT_h_scaled_size = 11;
+    cinfo->min_DCT_v_scaled_size = 11;
+  } else if (cinfo->scale_num * 12 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/12 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 12L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 12L);
+    cinfo->min_DCT_h_scaled_size = 12;
+    cinfo->min_DCT_v_scaled_size = 12;
+  } else if (cinfo->scale_num * 13 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/13 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 13L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 13L);
+    cinfo->min_DCT_h_scaled_size = 13;
+    cinfo->min_DCT_v_scaled_size = 13;
+  } else if (cinfo->scale_num * 14 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/14 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 14L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 14L);
+    cinfo->min_DCT_h_scaled_size = 14;
+    cinfo->min_DCT_v_scaled_size = 14;
+  } else if (cinfo->scale_num * 15 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/15 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 15L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 15L);
+    cinfo->min_DCT_h_scaled_size = 15;
+    cinfo->min_DCT_v_scaled_size = 15;
+  } else {
+    /* Provide block_size/16 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 16L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 16L);
+    cinfo->min_DCT_h_scaled_size = 16;
+    cinfo->min_DCT_v_scaled_size = 16;
+  }
+
+#else /* !DCT_SCALING_SUPPORTED */
+
+  /* Hardwire it to "no scaling" */
+  cinfo->jpeg_width = cinfo->image_width;
+  cinfo->jpeg_height = cinfo->image_height;
+  cinfo->min_DCT_h_scaled_size = DCTSIZE;
+  cinfo->min_DCT_v_scaled_size = DCTSIZE;
+
+#endif /* DCT_SCALING_SUPPORTED */
+}
+
+
+/*
+ * Master selection of compression modules.
+ * This is done once at the start of processing an image.  We determine
+ * which modules will be used and give them appropriate initialization calls.
+ */
+
+GLOBAL(void)
+jinit_compress_master (j_compress_ptr cinfo)
+{
+  long samplesperrow;
+  JDIMENSION jd_samplesperrow;
+
+  /* For now, precision must match compiled-in value... */
+  if (cinfo->data_precision != BITS_IN_JSAMPLE)
+    ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision);
+
+  /* Sanity check on input image dimensions */
+  if (cinfo->image_height <= 0 || cinfo->image_width <= 0 ||
+      cinfo->input_components <= 0)
+    ERREXIT(cinfo, JERR_EMPTY_IMAGE);
+
+  /* Width of an input scanline must be representable as JDIMENSION. */
+  samplesperrow = (long) cinfo->image_width * (long) cinfo->input_components;
+  jd_samplesperrow = (JDIMENSION) samplesperrow;
+  if ((long) jd_samplesperrow != samplesperrow)
+    ERREXIT(cinfo, JERR_WIDTH_OVERFLOW);
+
+  /* Compute JPEG image dimensions and related values. */
+  jpeg_calc_jpeg_dimensions(cinfo);
+
+  /* Initialize master control (includes parameter checking/processing) */
+  jinit_c_master_control(cinfo, FALSE /* full compression */);
+
+  /* Preprocessing */
+  if (! cinfo->raw_data_in) {
+    jinit_color_converter(cinfo);
+    jinit_downsampler(cinfo);
+    jinit_c_prep_controller(cinfo, FALSE /* never need full buffer here */);
+  }
+  /* Forward DCT */
+  jinit_forward_dct(cinfo);
+  /* Entropy encoding: either Huffman or arithmetic coding. */
+  if (cinfo->arith_code)
+    jinit_arith_encoder(cinfo);
+  else {
+    jinit_huff_encoder(cinfo);
+  }
+
+  /* Need a full-image coefficient buffer in any multi-pass mode. */
+  jinit_c_coef_controller(cinfo,
+		(boolean) (cinfo->num_scans > 1 || cinfo->optimize_coding));
+  jinit_c_main_controller(cinfo, FALSE /* never need full buffer here */);
+
+  jinit_marker_writer(cinfo);
+
+  /* We can now tell the memory manager to allocate virtual arrays. */
+  (*cinfo->mem->realize_virt_arrays) ((j_common_ptr) cinfo);
+
+  /* Write the datastream header (SOI) immediately.
+   * Frame and scan headers are postponed till later.
+   * This lets application insert special markers after the SOI.
+   */
+  (*cinfo->marker->write_file_header) (cinfo);
+}
@@ -0,0 +1,297 @@
+/*
+ * jcmainct.c
+ *
+ * Copyright (C) 1994-1996, Thomas G. Lane.
+ * Modified 2003-2012 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains the main buffer controller for compression.
+ * The main buffer lies between the pre-processor and the JPEG
+ * compressor proper; it holds downsampled data in the JPEG colorspace.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/* Note: currently, there is no operating mode in which a full-image buffer
+ * is needed at this step.  If there were, that mode could not be used with
+ * "raw data" input, since this module is bypassed in that case.  However,
+ * we've left the code here for possible use in special applications.
+ */
+#undef FULL_MAIN_BUFFER_SUPPORTED
+
+
+/* Private buffer controller object */
+
+typedef struct {
+  struct jpeg_c_main_controller pub; /* public fields */
+
+  JDIMENSION cur_iMCU_row;	/* number of current iMCU row */
+  JDIMENSION rowgroup_ctr;	/* counts row groups received in iMCU row */
+  boolean suspended;		/* remember if we suspended output */
+  J_BUF_MODE pass_mode;		/* current operating mode */
+
+  /* If using just a strip buffer, this points to the entire set of buffers
+   * (we allocate one for each component).  In the full-image case, this
+   * points to the currently accessible strips of the virtual arrays.
+   */
+  JSAMPARRAY buffer[MAX_COMPONENTS];
+
+#ifdef FULL_MAIN_BUFFER_SUPPORTED
+  /* If using full-image storage, this array holds pointers to virtual-array
+   * control blocks for each component.  Unused if not full-image storage.
+   */
+  jvirt_sarray_ptr whole_image[MAX_COMPONENTS];
+#endif
+} my_main_controller;
+
+typedef my_main_controller * my_main_ptr;
+
+
+/* Forward declarations */
+METHODDEF(void) process_data_simple_main
+	JPP((j_compress_ptr cinfo, JSAMPARRAY input_buf,
+	     JDIMENSION *in_row_ctr, JDIMENSION in_rows_avail));
+#ifdef FULL_MAIN_BUFFER_SUPPORTED
+METHODDEF(void) process_data_buffer_main
+	JPP((j_compress_ptr cinfo, JSAMPARRAY input_buf,
+	     JDIMENSION *in_row_ctr, JDIMENSION in_rows_avail));
+#endif
+
+
+/*
+ * Initialize for a processing pass.
+ */
+
+METHODDEF(void)
+start_pass_main (j_compress_ptr cinfo, J_BUF_MODE pass_mode)
+{
+  my_main_ptr mainp = (my_main_ptr) cinfo->main;
+
+  /* Do nothing in raw-data mode. */
+  if (cinfo->raw_data_in)
+    return;
+
+  mainp->cur_iMCU_row = 0;	/* initialize counters */
+  mainp->rowgroup_ctr = 0;
+  mainp->suspended = FALSE;
+  mainp->pass_mode = pass_mode;	/* save mode for use by process_data */
+
+  switch (pass_mode) {
+  case JBUF_PASS_THRU:
+#ifdef FULL_MAIN_BUFFER_SUPPORTED
+    if (mainp->whole_image[0] != NULL)
+      ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+#endif
+    mainp->pub.process_data = process_data_simple_main;
+    break;
+#ifdef FULL_MAIN_BUFFER_SUPPORTED
+  case JBUF_SAVE_SOURCE:
+  case JBUF_CRANK_DEST:
+  case JBUF_SAVE_AND_PASS:
+    if (mainp->whole_image[0] == NULL)
+      ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+    mainp->pub.process_data = process_data_buffer_main;
+    break;
+#endif
+  default:
+    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+    break;
+  }
+}
+
+
+/*
+ * Process some data.
+ * This routine handles the simple pass-through mode,
+ * where we have only a strip buffer.
+ */
+
+METHODDEF(void)
+process_data_simple_main (j_compress_ptr cinfo,
+			  JSAMPARRAY input_buf, JDIMENSION *in_row_ctr,
+			  JDIMENSION in_rows_avail)
+{
+  my_main_ptr mainp = (my_main_ptr) cinfo->main;
+
+  while (mainp->cur_iMCU_row < cinfo->total_iMCU_rows) {
+    /* Read input data if we haven't filled the main buffer yet */
+    if (mainp->rowgroup_ctr < (JDIMENSION) cinfo->min_DCT_v_scaled_size)
+      (*cinfo->prep->pre_process_data) (cinfo,
+					input_buf, in_row_ctr, in_rows_avail,
+					mainp->buffer, &mainp->rowgroup_ctr,
+					(JDIMENSION) cinfo->min_DCT_v_scaled_size);
+
+    /* If we don't have a full iMCU row buffered, return to application for
+     * more data.  Note that preprocessor will always pad to fill the iMCU row
+     * at the bottom of the image.
+     */
+    if (mainp->rowgroup_ctr != (JDIMENSION) cinfo->min_DCT_v_scaled_size)
+      return;
+
+    /* Send the completed row to the compressor */
+    if (! (*cinfo->coef->compress_data) (cinfo, mainp->buffer)) {
+      /* If compressor did not consume the whole row, then we must need to
+       * suspend processing and return to the application.  In this situation
+       * we pretend we didn't yet consume the last input row; otherwise, if
+       * it happened to be the last row of the image, the application would
+       * think we were done.
+       */
+      if (! mainp->suspended) {
+	(*in_row_ctr)--;
+	mainp->suspended = TRUE;
+      }
+      return;
+    }
+    /* We did finish the row.  Undo our little suspension hack if a previous
+     * call suspended; then mark the main buffer empty.
+     */
+    if (mainp->suspended) {
+      (*in_row_ctr)++;
+      mainp->suspended = FALSE;
+    }
+    mainp->rowgroup_ctr = 0;
+    mainp->cur_iMCU_row++;
+  }
+}
+
+
+#ifdef FULL_MAIN_BUFFER_SUPPORTED
+
+/*
+ * Process some data.
+ * This routine handles all of the modes that use a full-size buffer.
+ */
+
+METHODDEF(void)
+process_data_buffer_main (j_compress_ptr cinfo,
+			  JSAMPARRAY input_buf, JDIMENSION *in_row_ctr,
+			  JDIMENSION in_rows_avail)
+{
+  my_main_ptr mainp = (my_main_ptr) cinfo->main;
+  int ci;
+  jpeg_component_info *compptr;
+  boolean writing = (mainp->pass_mode != JBUF_CRANK_DEST);
+
+  while (mainp->cur_iMCU_row < cinfo->total_iMCU_rows) {
+    /* Realign the virtual buffers if at the start of an iMCU row. */
+    if (mainp->rowgroup_ctr == 0) {
+      for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+	   ci++, compptr++) {
+	mainp->buffer[ci] = (*cinfo->mem->access_virt_sarray)
+	  ((j_common_ptr) cinfo, mainp->whole_image[ci], mainp->cur_iMCU_row *
+	   ((JDIMENSION) (compptr->v_samp_factor * cinfo->min_DCT_v_scaled_size)),
+	   (JDIMENSION) (compptr->v_samp_factor * cinfo->min_DCT_v_scaled_size),
+	   writing);
+      }
+      /* In a read pass, pretend we just read some source data. */
+      if (! writing) {
+	*in_row_ctr += (JDIMENSION)
+	  (cinfo->max_v_samp_factor * cinfo->min_DCT_v_scaled_size);
+	mainp->rowgroup_ctr = (JDIMENSION) cinfo->min_DCT_v_scaled_size;
+      }
+    }
+
+    /* If a write pass, read input data until the current iMCU row is full. */
+    /* Note: preprocessor will pad if necessary to fill the last iMCU row. */
+    if (writing) {
+      (*cinfo->prep->pre_process_data) (cinfo,
+					input_buf, in_row_ctr, in_rows_avail,
+					mainp->buffer, &mainp->rowgroup_ctr,
+					(JDIMENSION) cinfo->min_DCT_v_scaled_size);
+      /* Return to application if we need more data to fill the iMCU row. */
+      if (mainp->rowgroup_ctr < (JDIMENSION) cinfo->min_DCT_v_scaled_size)
+	return;
+    }
+
+    /* Emit data, unless this is a sink-only pass. */
+    if (mainp->pass_mode != JBUF_SAVE_SOURCE) {
+      if (! (*cinfo->coef->compress_data) (cinfo, mainp->buffer)) {
+	/* If compressor did not consume the whole row, then we must need to
+	 * suspend processing and return to the application.  In this situation
+	 * we pretend we didn't yet consume the last input row; otherwise, if
+	 * it happened to be the last row of the image, the application would
+	 * think we were done.
+	 */
+	if (! mainp->suspended) {
+	  (*in_row_ctr)--;
+	  mainp->suspended = TRUE;
+	}
+	return;
+      }
+      /* We did finish the row.  Undo our little suspension hack if a previous
+       * call suspended; then mark the main buffer empty.
+       */
+      if (mainp->suspended) {
+	(*in_row_ctr)++;
+	mainp->suspended = FALSE;
+      }
+    }
+
+    /* If get here, we are done with this iMCU row.  Mark buffer empty. */
+    mainp->rowgroup_ctr = 0;
+    mainp->cur_iMCU_row++;
+  }
+}
+
+#endif /* FULL_MAIN_BUFFER_SUPPORTED */
+
+
+/*
+ * Initialize main buffer controller.
+ */
+
+GLOBAL(void)
+jinit_c_main_controller (j_compress_ptr cinfo, boolean need_full_buffer)
+{
+  my_main_ptr mainp;
+  int ci;
+  jpeg_component_info *compptr;
+
+  mainp = (my_main_ptr)
+    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				SIZEOF(my_main_controller));
+  cinfo->main = &mainp->pub;
+  mainp->pub.start_pass = start_pass_main;
+
+  /* We don't need to create a buffer in raw-data mode. */
+  if (cinfo->raw_data_in)
+    return;
+
+  /* Create the buffer.  It holds downsampled data, so each component
+   * may be of a different size.
+   */
+  if (need_full_buffer) {
+#ifdef FULL_MAIN_BUFFER_SUPPORTED
+    /* Allocate a full-image virtual array for each component */
+    /* Note we pad the bottom to a multiple of the iMCU height */
+    for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+	 ci++, compptr++) {
+      mainp->whole_image[ci] = (*cinfo->mem->request_virt_sarray)
+	((j_common_ptr) cinfo, JPOOL_IMAGE, FALSE,
+	 compptr->width_in_blocks * ((JDIMENSION) compptr->DCT_h_scaled_size),
+	 ((JDIMENSION) jround_up((long) compptr->height_in_blocks,
+				 (long) compptr->v_samp_factor)) *
+	 ((JDIMENSION) cinfo->min_DCT_v_scaled_size),
+	 (JDIMENSION) (compptr->v_samp_factor * compptr->DCT_v_scaled_size));
+    }
+#else
+    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+#endif
+  } else {
+#ifdef FULL_MAIN_BUFFER_SUPPORTED
+    mainp->whole_image[0] = NULL; /* flag for no virtual arrays */
+#endif
+    /* Allocate a strip buffer for each component */
+    for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+	 ci++, compptr++) {
+      mainp->buffer[ci] = (*cinfo->mem->alloc_sarray)
+	((j_common_ptr) cinfo, JPOOL_IMAGE,
+	 compptr->width_in_blocks * ((JDIMENSION) compptr->DCT_h_scaled_size),
+	 (JDIMENSION) (compptr->v_samp_factor * compptr->DCT_v_scaled_size));
+    }
+  }
+}
@@ -0,0 +1,717 @@
+/*
+ * jcmarker.c
+ *
+ * Copyright (C) 1991-1998, Thomas G. Lane.
+ * Modified 2003-2019 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains routines to write JPEG datastream markers.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+typedef enum {			/* JPEG marker codes */
+  M_SOF0  = 0xc0,
+  M_SOF1  = 0xc1,
+  M_SOF2  = 0xc2,
+  M_SOF3  = 0xc3,
+
+  M_SOF5  = 0xc5,
+  M_SOF6  = 0xc6,
+  M_SOF7  = 0xc7,
+
+  M_JPG   = 0xc8,
+  M_SOF9  = 0xc9,
+  M_SOF10 = 0xca,
+  M_SOF11 = 0xcb,
+
+  M_SOF13 = 0xcd,
+  M_SOF14 = 0xce,
+  M_SOF15 = 0xcf,
+
+  M_DHT   = 0xc4,
+
+  M_DAC   = 0xcc,
+
+  M_RST0  = 0xd0,
+  M_RST1  = 0xd1,
+  M_RST2  = 0xd2,
+  M_RST3  = 0xd3,
+  M_RST4  = 0xd4,
+  M_RST5  = 0xd5,
+  M_RST6  = 0xd6,
+  M_RST7  = 0xd7,
+
+  M_SOI   = 0xd8,
+  M_EOI   = 0xd9,
+  M_SOS   = 0xda,
+  M_DQT   = 0xdb,
+  M_DNL   = 0xdc,
+  M_DRI   = 0xdd,
+  M_DHP   = 0xde,
+  M_EXP   = 0xdf,
+
+  M_APP0  = 0xe0,
+  M_APP1  = 0xe1,
+  M_APP2  = 0xe2,
+  M_APP3  = 0xe3,
+  M_APP4  = 0xe4,
+  M_APP5  = 0xe5,
+  M_APP6  = 0xe6,
+  M_APP7  = 0xe7,
+  M_APP8  = 0xe8,
+  M_APP9  = 0xe9,
+  M_APP10 = 0xea,
+  M_APP11 = 0xeb,
+  M_APP12 = 0xec,
+  M_APP13 = 0xed,
+  M_APP14 = 0xee,
+  M_APP15 = 0xef,
+
+  M_JPG0  = 0xf0,
+  M_JPG8  = 0xf8,
+  M_JPG13 = 0xfd,
+  M_COM   = 0xfe,
+
+  M_TEM   = 0x01,
+
+  M_ERROR = 0x100
+} JPEG_MARKER;
+
+
+/* Private state */
+
+typedef struct {
+  struct jpeg_marker_writer pub; /* public fields */
+
+  unsigned int last_restart_interval; /* last DRI value emitted; 0 after SOI */
+} my_marker_writer;
+
+typedef my_marker_writer * my_marker_ptr;
+
+
+/*
+ * Basic output routines.
+ *
+ * Note that we do not support suspension while writing a marker.
+ * Therefore, an application using suspension must ensure that there is
+ * enough buffer space for the initial markers (typ. 600-700 bytes) before
+ * calling jpeg_start_compress, and enough space to write the trailing EOI
+ * (a few bytes) before calling jpeg_finish_compress.  Multipass compression
+ * modes are not supported at all with suspension, so those two are the only
+ * points where markers will be written.
+ */
+
+LOCAL(void)
+emit_byte (j_compress_ptr cinfo, int val)
+/* Emit a byte */
+{
+  struct jpeg_destination_mgr * dest = cinfo->dest;
+
+  *(dest->next_output_byte)++ = (JOCTET) val;
+  if (--dest->free_in_buffer == 0) {
+    if (! (*dest->empty_output_buffer) (cinfo))
+      ERREXIT(cinfo, JERR_CANT_SUSPEND);
+  }
+}
+
+
+LOCAL(void)
+emit_marker (j_compress_ptr cinfo, JPEG_MARKER mark)
+/* Emit a marker code */
+{
+  emit_byte(cinfo, 0xFF);
+  emit_byte(cinfo, (int) mark);
+}
+
+
+LOCAL(void)
+emit_2bytes (j_compress_ptr cinfo, int value)
+/* Emit a 2-byte integer; these are always MSB first in JPEG files */
+{
+  emit_byte(cinfo, (value >> 8) & 0xFF);
+  emit_byte(cinfo, value & 0xFF);
+}
+
+
+/*
+ * Routines to write specific marker types.
+ */
+
+LOCAL(int)
+emit_dqt (j_compress_ptr cinfo, int index)
+/* Emit a DQT marker */
+/* Returns the precision used (0 = 8bits, 1 = 16bits) for baseline checking */
+{
+  JQUANT_TBL * qtbl = cinfo->quant_tbl_ptrs[index];
+  int prec;
+  int i;
+
+  if (qtbl == NULL)
+    ERREXIT1(cinfo, JERR_NO_QUANT_TABLE, index);
+
+  prec = 0;
+  for (i = 0; i <= cinfo->lim_Se; i++) {
+    if (qtbl->quantval[cinfo->natural_order[i]] > 255)
+      prec = 1;
+  }
+
+  if (! qtbl->sent_table) {
+    emit_marker(cinfo, M_DQT);
+
+    emit_2bytes(cinfo,
+      prec ? cinfo->lim_Se * 2 + 2 + 1 + 2 : cinfo->lim_Se + 1 + 1 + 2);
+
+    emit_byte(cinfo, index + (prec<<4));
+
+    for (i = 0; i <= cinfo->lim_Se; i++) {
+      /* The table entries must be emitted in zigzag order. */
+      unsigned int qval = qtbl->quantval[cinfo->natural_order[i]];
+      if (prec)
+	emit_byte(cinfo, (int) (qval >> 8));
+      emit_byte(cinfo, (int) (qval & 0xFF));
+    }
+
+    qtbl->sent_table = TRUE;
+  }
+
+  return prec;
+}
+
+
+LOCAL(void)
+emit_dht (j_compress_ptr cinfo, int index, boolean is_ac)
+/* Emit a DHT marker */
+{
+  JHUFF_TBL * htbl;
+  int length, i;
+  
+  if (is_ac) {
+    htbl = cinfo->ac_huff_tbl_ptrs[index];
+    index += 0x10;		/* output index has AC bit set */
+  } else {
+    htbl = cinfo->dc_huff_tbl_ptrs[index];
+  }
+
+  if (htbl == NULL)
+    ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, index);
+  
+  if (! htbl->sent_table) {
+    emit_marker(cinfo, M_DHT);
+    
+    length = 0;
+    for (i = 1; i <= 16; i++)
+      length += htbl->bits[i];
+    
+    emit_2bytes(cinfo, length + 2 + 1 + 16);
+    emit_byte(cinfo, index);
+    
+    for (i = 1; i <= 16; i++)
+      emit_byte(cinfo, htbl->bits[i]);
+    
+    for (i = 0; i < length; i++)
+      emit_byte(cinfo, htbl->huffval[i]);
+    
+    htbl->sent_table = TRUE;
+  }
+}
+
+
+LOCAL(void)
+emit_dac (j_compress_ptr cinfo)
+/* Emit a DAC marker */
+/* Since the useful info is so small, we want to emit all the tables in */
+/* one DAC marker.  Therefore this routine does its own scan of the table. */
+{
+#ifdef C_ARITH_CODING_SUPPORTED
+  char dc_in_use[NUM_ARITH_TBLS];
+  char ac_in_use[NUM_ARITH_TBLS];
+  int length, i;
+  jpeg_component_info *compptr;
+
+  for (i = 0; i < NUM_ARITH_TBLS; i++)
+    dc_in_use[i] = ac_in_use[i] = 0;
+
+  for (i = 0; i < cinfo->comps_in_scan; i++) {
+    compptr = cinfo->cur_comp_info[i];
+    /* DC needs no table for refinement scan */
+    if (cinfo->Ss == 0 && cinfo->Ah == 0)
+      dc_in_use[compptr->dc_tbl_no] = 1;
+    /* AC needs no table when not present */
+    if (cinfo->Se)
+      ac_in_use[compptr->ac_tbl_no] = 1;
+  }
+
+  length = 0;
+  for (i = 0; i < NUM_ARITH_TBLS; i++)
+    length += dc_in_use[i] + ac_in_use[i];
+
+  if (length) {
+    emit_marker(cinfo, M_DAC);
+
+    emit_2bytes(cinfo, length*2 + 2);
+
+    for (i = 0; i < NUM_ARITH_TBLS; i++) {
+      if (dc_in_use[i]) {
+	emit_byte(cinfo, i);
+	emit_byte(cinfo, cinfo->arith_dc_L[i] + (cinfo->arith_dc_U[i]<<4));
+      }
+      if (ac_in_use[i]) {
+	emit_byte(cinfo, i + 0x10);
+	emit_byte(cinfo, cinfo->arith_ac_K[i]);
+      }
+    }
+  }
+#endif /* C_ARITH_CODING_SUPPORTED */
+}
+
+
+LOCAL(void)
+emit_dri (j_compress_ptr cinfo)
+/* Emit a DRI marker */
+{
+  emit_marker(cinfo, M_DRI);
+  
+  emit_2bytes(cinfo, 4);	/* fixed length */
+
+  emit_2bytes(cinfo, (int) cinfo->restart_interval);
+}
+
+
+LOCAL(void)
+emit_lse_ict (j_compress_ptr cinfo)
+/* Emit an LSE inverse color transform specification marker */
+{
+  /* Support only 1 transform */
+  if (cinfo->color_transform != JCT_SUBTRACT_GREEN ||
+      cinfo->num_components < 3)
+    ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+
+  emit_marker(cinfo, M_JPG8);
+  
+  emit_2bytes(cinfo, 24);	/* fixed length */
+
+  emit_byte(cinfo, 0x0D);	/* ID inverse transform specification */
+  emit_2bytes(cinfo, MAXJSAMPLE);	/* MAXTRANS */
+  emit_byte(cinfo, 3);		/* Nt=3 */
+  emit_byte(cinfo, cinfo->comp_info[1].component_id);
+  emit_byte(cinfo, cinfo->comp_info[0].component_id);
+  emit_byte(cinfo, cinfo->comp_info[2].component_id);
+  emit_byte(cinfo, 0x80);	/* F1: CENTER1=1, NORM1=0 */
+  emit_2bytes(cinfo, 0);	/* A(1,1)=0 */
+  emit_2bytes(cinfo, 0);	/* A(1,2)=0 */
+  emit_byte(cinfo, 0);		/* F2: CENTER2=0, NORM2=0 */
+  emit_2bytes(cinfo, 1);	/* A(2,1)=1 */
+  emit_2bytes(cinfo, 0);	/* A(2,2)=0 */
+  emit_byte(cinfo, 0);		/* F3: CENTER3=0, NORM3=0 */
+  emit_2bytes(cinfo, 1);	/* A(3,1)=1 */
+  emit_2bytes(cinfo, 0);	/* A(3,2)=0 */
+}
+
+
+LOCAL(void)
+emit_sof (j_compress_ptr cinfo, JPEG_MARKER code)
+/* Emit a SOF marker */
+{
+  int ci;
+  jpeg_component_info *compptr;
+  
+  emit_marker(cinfo, code);
+  
+  emit_2bytes(cinfo, 3 * cinfo->num_components + 2 + 5 + 1); /* length */
+
+  /* Make sure image isn't bigger than SOF field can handle */
+  if ((long) cinfo->jpeg_height > 65535L ||
+      (long) cinfo->jpeg_width > 65535L)
+    ERREXIT1(cinfo, JERR_IMAGE_TOO_BIG, (unsigned int) 65535);
+
+  emit_byte(cinfo, cinfo->data_precision);
+  emit_2bytes(cinfo, (int) cinfo->jpeg_height);
+  emit_2bytes(cinfo, (int) cinfo->jpeg_width);
+
+  emit_byte(cinfo, cinfo->num_components);
+
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    emit_byte(cinfo, compptr->component_id);
+    emit_byte(cinfo, (compptr->h_samp_factor << 4) + compptr->v_samp_factor);
+    emit_byte(cinfo, compptr->quant_tbl_no);
+  }
+}
+
+
+LOCAL(void)
+emit_sos (j_compress_ptr cinfo)
+/* Emit a SOS marker */
+{
+  int i, td, ta;
+  jpeg_component_info *compptr;
+  
+  emit_marker(cinfo, M_SOS);
+  
+  emit_2bytes(cinfo, 2 * cinfo->comps_in_scan + 2 + 1 + 3); /* length */
+  
+  emit_byte(cinfo, cinfo->comps_in_scan);
+  
+  for (i = 0; i < cinfo->comps_in_scan; i++) {
+    compptr = cinfo->cur_comp_info[i];
+    emit_byte(cinfo, compptr->component_id);
+
+    /* We emit 0 for unused field(s); this is recommended by the P&M text
+     * but does not seem to be specified in the standard.
+     */
+
+    /* DC needs no table for refinement scan */
+    td = cinfo->Ss == 0 && cinfo->Ah == 0 ? compptr->dc_tbl_no : 0;
+    /* AC needs no table when not present */
+    ta = cinfo->Se ? compptr->ac_tbl_no : 0;
+
+    emit_byte(cinfo, (td << 4) + ta);
+  }
+
+  emit_byte(cinfo, cinfo->Ss);
+  emit_byte(cinfo, cinfo->Se);
+  emit_byte(cinfo, (cinfo->Ah << 4) + cinfo->Al);
+}
+
+
+LOCAL(void)
+emit_pseudo_sos (j_compress_ptr cinfo)
+/* Emit a pseudo SOS marker */
+{
+  emit_marker(cinfo, M_SOS);
+  
+  emit_2bytes(cinfo, 2 + 1 + 3); /* length */
+  
+  emit_byte(cinfo, 0); /* Ns */
+
+  emit_byte(cinfo, 0); /* Ss */
+  emit_byte(cinfo, cinfo->block_size * cinfo->block_size - 1); /* Se */
+  emit_byte(cinfo, 0); /* Ah/Al */
+}
+
+
+LOCAL(void)
+emit_jfif_app0 (j_compress_ptr cinfo)
+/* Emit a JFIF-compliant APP0 marker */
+{
+  /*
+   * Length of APP0 block	(2 bytes)
+   * Block ID			(4 bytes - ASCII "JFIF")
+   * Zero byte			(1 byte to terminate the ID string)
+   * Version Major, Minor	(2 bytes - major first)
+   * Units			(1 byte - 0x00 = none, 0x01 = inch, 0x02 = cm)
+   * Xdpu			(2 bytes - dots per unit horizontal)
+   * Ydpu			(2 bytes - dots per unit vertical)
+   * Thumbnail X size		(1 byte)
+   * Thumbnail Y size		(1 byte)
+   */
+  
+  emit_marker(cinfo, M_APP0);
+  
+  emit_2bytes(cinfo, 2 + 4 + 1 + 2 + 1 + 2 + 2 + 1 + 1); /* length */
+
+  emit_byte(cinfo, 0x4A);	/* Identifier: ASCII "JFIF" */
+  emit_byte(cinfo, 0x46);
+  emit_byte(cinfo, 0x49);
+  emit_byte(cinfo, 0x46);
+  emit_byte(cinfo, 0);
+  emit_byte(cinfo, cinfo->JFIF_major_version); /* Version fields */
+  emit_byte(cinfo, cinfo->JFIF_minor_version);
+  emit_byte(cinfo, cinfo->density_unit); /* Pixel size information */
+  emit_2bytes(cinfo, (int) cinfo->X_density);
+  emit_2bytes(cinfo, (int) cinfo->Y_density);
+  emit_byte(cinfo, 0);		/* No thumbnail image */
+  emit_byte(cinfo, 0);
+}
+
+
+LOCAL(void)
+emit_adobe_app14 (j_compress_ptr cinfo)
+/* Emit an Adobe APP14 marker */
+{
+  /*
+   * Length of APP14 block	(2 bytes)
+   * Block ID			(5 bytes - ASCII "Adobe")
+   * Version Number		(2 bytes - currently 100)
+   * Flags0			(2 bytes - currently 0)
+   * Flags1			(2 bytes - currently 0)
+   * Color transform		(1 byte)
+   *
+   * Although Adobe TN 5116 mentions Version = 101, all the Adobe files
+   * now in circulation seem to use Version = 100, so that's what we write.
+   *
+   * We write the color transform byte as 1 if the JPEG color space is
+   * YCbCr, 2 if it's YCCK, 0 otherwise.  Adobe's definition has to do with
+   * whether the encoder performed a transformation, which is pretty useless.
+   */
+  
+  emit_marker(cinfo, M_APP14);
+  
+  emit_2bytes(cinfo, 2 + 5 + 2 + 2 + 2 + 1); /* length */
+
+  emit_byte(cinfo, 0x41);	/* Identifier: ASCII "Adobe" */
+  emit_byte(cinfo, 0x64);
+  emit_byte(cinfo, 0x6F);
+  emit_byte(cinfo, 0x62);
+  emit_byte(cinfo, 0x65);
+  emit_2bytes(cinfo, 100);	/* Version */
+  emit_2bytes(cinfo, 0);	/* Flags0 */
+  emit_2bytes(cinfo, 0);	/* Flags1 */
+  switch (cinfo->jpeg_color_space) {
+  case JCS_YCbCr:
+    emit_byte(cinfo, 1);	/* Color transform = 1 */
+    break;
+  case JCS_YCCK:
+    emit_byte(cinfo, 2);	/* Color transform = 2 */
+    break;
+  default:
+    emit_byte(cinfo, 0);	/* Color transform = 0 */
+  }
+}
+
+
+/*
+ * These routines allow writing an arbitrary marker with parameters.
+ * The only intended use is to emit COM or APPn markers after calling
+ * write_file_header and before calling write_frame_header.
+ * Other uses are not guaranteed to produce desirable results.
+ * Counting the parameter bytes properly is the caller's responsibility.
+ */
+
+METHODDEF(void)
+write_marker_header (j_compress_ptr cinfo, int marker, unsigned int datalen)
+/* Emit an arbitrary marker header */
+{
+  if (datalen > (unsigned int) 65533)		/* safety check */
+    ERREXIT(cinfo, JERR_BAD_LENGTH);
+
+  emit_marker(cinfo, (JPEG_MARKER) marker);
+
+  emit_2bytes(cinfo, (int) (datalen + 2));	/* total length */
+}
+
+METHODDEF(void)
+write_marker_byte (j_compress_ptr cinfo, int val)
+/* Emit one byte of marker parameters following write_marker_header */
+{
+  emit_byte(cinfo, val);
+}
+
+
+/*
+ * Write datastream header.
+ * This consists of an SOI and optional APPn markers.
+ * We recommend use of the JFIF marker, but not the Adobe marker,
+ * when using YCbCr or grayscale data.  The JFIF marker is also used
+ * for other standard JPEG colorspaces.  The Adobe marker is helpful
+ * to distinguish RGB, CMYK, and YCCK colorspaces.
+ * Note that an application can write additional header markers after
+ * jpeg_start_compress returns.
+ */
+
+METHODDEF(void)
+write_file_header (j_compress_ptr cinfo)
+{
+  my_marker_ptr marker = (my_marker_ptr) cinfo->marker;
+
+  emit_marker(cinfo, M_SOI);	/* first the SOI */
+
+  /* SOI is defined to reset restart interval to 0 */
+  marker->last_restart_interval = 0;
+
+  if (cinfo->write_JFIF_header)	/* next an optional JFIF APP0 */
+    emit_jfif_app0(cinfo);
+  if (cinfo->write_Adobe_marker) /* next an optional Adobe APP14 */
+    emit_adobe_app14(cinfo);
+}
+
+
+/*
+ * Write frame header.
+ * This consists of DQT and SOFn markers,
+ * a conditional LSE marker and a conditional pseudo SOS marker.
+ * Note that we do not emit the SOF until we have emitted the DQT(s).
+ * This avoids compatibility problems with incorrect implementations that
+ * try to error-check the quant table numbers as soon as they see the SOF.
+ */
+
+METHODDEF(void)
+write_frame_header (j_compress_ptr cinfo)
+{
+  int ci, prec;
+  boolean is_baseline;
+  jpeg_component_info *compptr;
+  
+  /* Emit DQT for each quantization table.
+   * Note that emit_dqt() suppresses any duplicate tables.
+   */
+  prec = 0;
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    prec += emit_dqt(cinfo, compptr->quant_tbl_no);
+  }
+  /* now prec is nonzero iff there are any 16-bit quant tables. */
+
+  /* Check for a non-baseline specification.
+   * Note we assume that Huffman table numbers won't be changed later.
+   */
+  if (cinfo->arith_code || cinfo->progressive_mode ||
+      cinfo->data_precision != 8 || cinfo->block_size != DCTSIZE) {
+    is_baseline = FALSE;
+  } else {
+    is_baseline = TRUE;
+    for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+	 ci++, compptr++) {
+      if (compptr->dc_tbl_no > 1 || compptr->ac_tbl_no > 1)
+	is_baseline = FALSE;
+    }
+    if (prec && is_baseline) {
+      is_baseline = FALSE;
+      /* If it's baseline except for quantizer size, warn the user */
+      TRACEMS(cinfo, 0, JTRC_16BIT_TABLES);
+    }
+  }
+
+  /* Emit the proper SOF marker */
+  if (cinfo->arith_code) {
+    if (cinfo->progressive_mode)
+      emit_sof(cinfo, M_SOF10); /* SOF code for progressive arithmetic */
+    else
+      emit_sof(cinfo, M_SOF9);  /* SOF code for sequential arithmetic */
+  } else {
+    if (cinfo->progressive_mode)
+      emit_sof(cinfo, M_SOF2);	/* SOF code for progressive Huffman */
+    else if (is_baseline)
+      emit_sof(cinfo, M_SOF0);	/* SOF code for baseline implementation */
+    else
+      emit_sof(cinfo, M_SOF1);	/* SOF code for non-baseline Huffman file */
+  }
+
+  /* Check to emit LSE inverse color transform specification marker */
+  if (cinfo->color_transform)
+    emit_lse_ict(cinfo);
+
+  /* Check to emit pseudo SOS marker */
+  if (cinfo->progressive_mode && cinfo->block_size != DCTSIZE)
+    emit_pseudo_sos(cinfo);
+}
+
+
+/*
+ * Write scan header.
+ * This consists of DHT or DAC markers, optional DRI, and SOS.
+ * Compressed data will be written following the SOS.
+ */
+
+METHODDEF(void)
+write_scan_header (j_compress_ptr cinfo)
+{
+  my_marker_ptr marker = (my_marker_ptr) cinfo->marker;
+  int i;
+  jpeg_component_info *compptr;
+
+  if (cinfo->arith_code) {
+    /* Emit arith conditioning info.  We may have some duplication
+     * if the file has multiple scans, but it's so small it's hardly
+     * worth worrying about.
+     */
+    emit_dac(cinfo);
+  } else {
+    /* Emit Huffman tables.
+     * Note that emit_dht() suppresses any duplicate tables.
+     */
+    for (i = 0; i < cinfo->comps_in_scan; i++) {
+      compptr = cinfo->cur_comp_info[i];
+      /* DC needs no table for refinement scan */
+      if (cinfo->Ss == 0 && cinfo->Ah == 0)
+	emit_dht(cinfo, compptr->dc_tbl_no, FALSE);
+      /* AC needs no table when not present */
+      if (cinfo->Se)
+	emit_dht(cinfo, compptr->ac_tbl_no, TRUE);
+    }
+  }
+
+  /* Emit DRI if required --- note that DRI value could change for each scan.
+   * We avoid wasting space with unnecessary DRIs, however.
+   */
+  if (cinfo->restart_interval != marker->last_restart_interval) {
+    emit_dri(cinfo);
+    marker->last_restart_interval = cinfo->restart_interval;
+  }
+
+  emit_sos(cinfo);
+}
+
+
+/*
+ * Write datastream trailer.
+ */
+
+METHODDEF(void)
+write_file_trailer (j_compress_ptr cinfo)
+{
+  emit_marker(cinfo, M_EOI);
+}
+
+
+/*
+ * Write an abbreviated table-specification datastream.
+ * This consists of SOI, DQT and DHT tables, and EOI.
+ * Any table that is defined and not marked sent_table = TRUE will be
+ * emitted.  Note that all tables will be marked sent_table = TRUE at exit.
+ */
+
+METHODDEF(void)
+write_tables_only (j_compress_ptr cinfo)
+{
+  int i;
+
+  emit_marker(cinfo, M_SOI);
+
+  for (i = 0; i < NUM_QUANT_TBLS; i++) {
+    if (cinfo->quant_tbl_ptrs[i] != NULL)
+      (void) emit_dqt(cinfo, i);
+  }
+
+  if (! cinfo->arith_code) {
+    for (i = 0; i < NUM_HUFF_TBLS; i++) {
+      if (cinfo->dc_huff_tbl_ptrs[i] != NULL)
+	emit_dht(cinfo, i, FALSE);
+      if (cinfo->ac_huff_tbl_ptrs[i] != NULL)
+	emit_dht(cinfo, i, TRUE);
+    }
+  }
+
+  emit_marker(cinfo, M_EOI);
+}
+
+
+/*
+ * Initialize the marker writer module.
+ */
+
+GLOBAL(void)
+jinit_marker_writer (j_compress_ptr cinfo)
+{
+  my_marker_ptr marker;
+
+  /* Create the subobject */
+  marker = (my_marker_ptr) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, SIZEOF(my_marker_writer));
+  cinfo->marker = &marker->pub;
+  /* Initialize method pointers */
+  marker->pub.write_file_header = write_file_header;
+  marker->pub.write_frame_header = write_frame_header;
+  marker->pub.write_scan_header = write_scan_header;
+  marker->pub.write_file_trailer = write_file_trailer;
+  marker->pub.write_tables_only = write_tables_only;
+  marker->pub.write_marker_header = write_marker_header;
+  marker->pub.write_marker_byte = write_marker_byte;
+  /* Initialize private state */
+  marker->last_restart_interval = 0;
+}
@@ -0,0 +1,675 @@
+/*
+ * jcmaster.c
+ *
+ * Copyright (C) 1991-1997, Thomas G. Lane.
+ * Modified 2003-2020 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains master control logic for the JPEG compressor.
+ * These routines are concerned with parameter validation, initial setup,
+ * and inter-pass control (determining the number of passes and the work 
+ * to be done in each pass).
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/* Private state */
+
+typedef enum {
+	main_pass,		/* input data, also do first output step */
+	huff_opt_pass,		/* Huffman code optimization pass */
+	output_pass		/* data output pass */
+} c_pass_type;
+
+typedef struct {
+  struct jpeg_comp_master pub;	/* public fields */
+
+  c_pass_type pass_type;	/* the type of the current pass */
+
+  int pass_number;		/* # of passes completed */
+  int total_passes;		/* total # of passes needed */
+
+  int scan_number;		/* current index in scan_info[] */
+} my_comp_master;
+
+typedef my_comp_master * my_master_ptr;
+
+
+/*
+ * Support routines that do various essential calculations.
+ */
+
+LOCAL(void)
+initial_setup (j_compress_ptr cinfo)
+/* Do computations that are needed before master selection phase */
+{
+  int ci, ssize;
+  jpeg_component_info *compptr;
+
+  /* Sanity check on block_size */
+  if (cinfo->block_size < 1 || cinfo->block_size > 16)
+    ERREXIT2(cinfo, JERR_BAD_DCTSIZE, cinfo->block_size, cinfo->block_size);
+
+  /* Derive natural_order from block_size */
+  switch (cinfo->block_size) {
+  case 2: cinfo->natural_order = jpeg_natural_order2; break;
+  case 3: cinfo->natural_order = jpeg_natural_order3; break;
+  case 4: cinfo->natural_order = jpeg_natural_order4; break;
+  case 5: cinfo->natural_order = jpeg_natural_order5; break;
+  case 6: cinfo->natural_order = jpeg_natural_order6; break;
+  case 7: cinfo->natural_order = jpeg_natural_order7; break;
+  default: cinfo->natural_order = jpeg_natural_order;
+  }
+
+  /* Derive lim_Se from block_size */
+  cinfo->lim_Se = cinfo->block_size < DCTSIZE ?
+    cinfo->block_size * cinfo->block_size - 1 : DCTSIZE2-1;
+
+  /* Sanity check on image dimensions */
+  if (cinfo->jpeg_height <= 0 || cinfo->jpeg_width <= 0 ||
+      cinfo->num_components <= 0)
+    ERREXIT(cinfo, JERR_EMPTY_IMAGE);
+
+  /* Make sure image isn't bigger than I can handle */
+  if ((long) cinfo->jpeg_height > (long) JPEG_MAX_DIMENSION ||
+      (long) cinfo->jpeg_width > (long) JPEG_MAX_DIMENSION)
+    ERREXIT1(cinfo, JERR_IMAGE_TOO_BIG, (unsigned int) JPEG_MAX_DIMENSION);
+
+  /* Only 8 to 12 bits data precision are supported for DCT based JPEG */
+  if (cinfo->data_precision < 8 || cinfo->data_precision > 12)
+    ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision);
+
+  /* Check that number of components won't exceed internal array sizes */
+  if (cinfo->num_components > MAX_COMPONENTS)
+    ERREXIT2(cinfo, JERR_COMPONENT_COUNT, cinfo->num_components,
+	     MAX_COMPONENTS);
+
+  /* Compute maximum sampling factors; check factor validity */
+  cinfo->max_h_samp_factor = 1;
+  cinfo->max_v_samp_factor = 1;
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    if (compptr->h_samp_factor<=0 || compptr->h_samp_factor>MAX_SAMP_FACTOR ||
+	compptr->v_samp_factor<=0 || compptr->v_samp_factor>MAX_SAMP_FACTOR)
+      ERREXIT(cinfo, JERR_BAD_SAMPLING);
+    cinfo->max_h_samp_factor = MAX(cinfo->max_h_samp_factor,
+				   compptr->h_samp_factor);
+    cinfo->max_v_samp_factor = MAX(cinfo->max_v_samp_factor,
+				   compptr->v_samp_factor);
+  }
+
+  /* Compute dimensions of components */
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    /* Fill in the correct component_index value; don't rely on application */
+    compptr->component_index = ci;
+    /* In selecting the actual DCT scaling for each component, we try to
+     * scale down the chroma components via DCT scaling rather than downsampling.
+     * This saves time if the downsampler gets to use 1:1 scaling.
+     * Note this code adapts subsampling ratios which are powers of 2.
+     */
+    ssize = 1;
+#ifdef DCT_SCALING_SUPPORTED
+    if (! cinfo->raw_data_in)
+      while (cinfo->min_DCT_h_scaled_size * ssize <=
+	     (cinfo->do_fancy_downsampling ? DCTSIZE : DCTSIZE / 2) &&
+	     (cinfo->max_h_samp_factor % (compptr->h_samp_factor * ssize * 2)) ==
+	     0) {
+	ssize = ssize * 2;
+      }
+#endif
+    compptr->DCT_h_scaled_size = cinfo->min_DCT_h_scaled_size * ssize;
+    ssize = 1;
+#ifdef DCT_SCALING_SUPPORTED
+    if (! cinfo->raw_data_in)
+      while (cinfo->min_DCT_v_scaled_size * ssize <=
+	     (cinfo->do_fancy_downsampling ? DCTSIZE : DCTSIZE / 2) &&
+	     (cinfo->max_v_samp_factor % (compptr->v_samp_factor * ssize * 2)) ==
+	     0) {
+	ssize = ssize * 2;
+      }
+#endif
+    compptr->DCT_v_scaled_size = cinfo->min_DCT_v_scaled_size * ssize;
+
+    /* We don't support DCT ratios larger than 2. */
+    if (compptr->DCT_h_scaled_size > compptr->DCT_v_scaled_size * 2)
+	compptr->DCT_h_scaled_size = compptr->DCT_v_scaled_size * 2;
+    else if (compptr->DCT_v_scaled_size > compptr->DCT_h_scaled_size * 2)
+	compptr->DCT_v_scaled_size = compptr->DCT_h_scaled_size * 2;
+
+    /* Size in DCT blocks */
+    compptr->width_in_blocks = (JDIMENSION)
+      jdiv_round_up((long) cinfo->jpeg_width * (long) compptr->h_samp_factor,
+		    (long) (cinfo->max_h_samp_factor * cinfo->block_size));
+    compptr->height_in_blocks = (JDIMENSION)
+      jdiv_round_up((long) cinfo->jpeg_height * (long) compptr->v_samp_factor,
+		    (long) (cinfo->max_v_samp_factor * cinfo->block_size));
+    /* Size in samples */
+    compptr->downsampled_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->jpeg_width *
+		    (long) (compptr->h_samp_factor * compptr->DCT_h_scaled_size),
+		    (long) (cinfo->max_h_samp_factor * cinfo->block_size));
+    compptr->downsampled_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->jpeg_height *
+		    (long) (compptr->v_samp_factor * compptr->DCT_v_scaled_size),
+		    (long) (cinfo->max_v_samp_factor * cinfo->block_size));
+    /* Don't need quantization scale after DCT,
+     * until color conversion says otherwise.
+     */
+    compptr->component_needed = FALSE;
+  }
+
+  /* Compute number of fully interleaved MCU rows (number of times that
+   * main controller will call coefficient controller).
+   */
+  cinfo->total_iMCU_rows = (JDIMENSION)
+    jdiv_round_up((long) cinfo->jpeg_height,
+		  (long) (cinfo->max_v_samp_factor * cinfo->block_size));
+}
+
+
+#ifdef C_MULTISCAN_FILES_SUPPORTED
+
+LOCAL(void)
+validate_script (j_compress_ptr cinfo)
+/* Verify that the scan script in cinfo->scan_info[] is valid; also
+ * determine whether it uses progressive JPEG, and set cinfo->progressive_mode.
+ */
+{
+  const jpeg_scan_info * scanptr;
+  int scanno, ncomps, ci, coefi, thisi;
+  int Ss, Se, Ah, Al;
+  boolean component_sent[MAX_COMPONENTS];
+#ifdef C_PROGRESSIVE_SUPPORTED
+  int * last_bitpos_ptr;
+  int last_bitpos[MAX_COMPONENTS][DCTSIZE2];
+  /* -1 until that coefficient has been seen; then last Al for it */
+#endif
+
+  if (cinfo->num_scans <= 0)
+    ERREXIT1(cinfo, JERR_BAD_SCAN_SCRIPT, 0);
+
+  /* For sequential JPEG, all scans must have Ss=0, Se=DCTSIZE2-1;
+   * for progressive JPEG, no scan can have this.
+   */
+  scanptr = cinfo->scan_info;
+  if (scanptr->Ss != 0 || scanptr->Se != DCTSIZE2-1) {
+#ifdef C_PROGRESSIVE_SUPPORTED
+    cinfo->progressive_mode = TRUE;
+    last_bitpos_ptr = & last_bitpos[0][0];
+    for (ci = 0; ci < cinfo->num_components; ci++) 
+      for (coefi = 0; coefi < DCTSIZE2; coefi++)
+	*last_bitpos_ptr++ = -1;
+#else
+    ERREXIT(cinfo, JERR_NOT_COMPILED);
+#endif
+  } else {
+    cinfo->progressive_mode = FALSE;
+    for (ci = 0; ci < cinfo->num_components; ci++) 
+      component_sent[ci] = FALSE;
+  }
+
+  for (scanno = 1; scanno <= cinfo->num_scans; scanptr++, scanno++) {
+    /* Validate component indexes */
+    ncomps = scanptr->comps_in_scan;
+    if (ncomps <= 0 || ncomps > MAX_COMPS_IN_SCAN)
+      ERREXIT2(cinfo, JERR_COMPONENT_COUNT, ncomps, MAX_COMPS_IN_SCAN);
+    for (ci = 0; ci < ncomps; ci++) {
+      thisi = scanptr->component_index[ci];
+      if (thisi < 0 || thisi >= cinfo->num_components)
+	ERREXIT1(cinfo, JERR_BAD_SCAN_SCRIPT, scanno);
+      /* Components must appear in SOF order within each scan */
+      if (ci > 0 && thisi <= scanptr->component_index[ci-1])
+	ERREXIT1(cinfo, JERR_BAD_SCAN_SCRIPT, scanno);
+    }
+    /* Validate progression parameters */
+    Ss = scanptr->Ss;
+    Se = scanptr->Se;
+    Ah = scanptr->Ah;
+    Al = scanptr->Al;
+    if (cinfo->progressive_mode) {
+#ifdef C_PROGRESSIVE_SUPPORTED
+      /* The JPEG spec simply gives the ranges 0..13 for Ah and Al, but that
+       * seems wrong: the upper bound ought to depend on data precision.
+       * Perhaps they really meant 0..N+1 for N-bit precision.
+       * Here we allow 0..10 for 8-bit data; Al larger than 10 results in
+       * out-of-range reconstructed DC values during the first DC scan,
+       * which might cause problems for some decoders.
+       */
+      if (Ss < 0 || Ss >= DCTSIZE2 || Se < Ss || Se >= DCTSIZE2 ||
+	  Ah < 0 || Ah > (cinfo->data_precision > 8 ? 13 : 10) ||
+	  Al < 0 || Al > (cinfo->data_precision > 8 ? 13 : 10))
+	ERREXIT1(cinfo, JERR_BAD_PROG_SCRIPT, scanno);
+      if (Ss == 0) {
+	if (Se != 0)		/* DC and AC together not OK */
+	  ERREXIT1(cinfo, JERR_BAD_PROG_SCRIPT, scanno);
+      } else {
+	if (ncomps != 1)	/* AC scans must be for only one component */
+	  ERREXIT1(cinfo, JERR_BAD_PROG_SCRIPT, scanno);
+      }
+      for (ci = 0; ci < ncomps; ci++) {
+	last_bitpos_ptr = & last_bitpos[scanptr->component_index[ci]][0];
+	if (Ss != 0 && last_bitpos_ptr[0] < 0) /* AC without prior DC scan */
+	  ERREXIT1(cinfo, JERR_BAD_PROG_SCRIPT, scanno);
+	for (coefi = Ss; coefi <= Se; coefi++) {
+	  if (last_bitpos_ptr[coefi] < 0) {
+	    /* first scan of this coefficient */
+	    if (Ah != 0)
+	      ERREXIT1(cinfo, JERR_BAD_PROG_SCRIPT, scanno);
+	  } else {
+	    /* not first scan */
+	    if (Ah != last_bitpos_ptr[coefi] || Al != Ah-1)
+	      ERREXIT1(cinfo, JERR_BAD_PROG_SCRIPT, scanno);
+	  }
+	  last_bitpos_ptr[coefi] = Al;
+	}
+      }
+#endif
+    } else {
+      /* For sequential JPEG, all progression parameters must be these: */
+      if (Ss != 0 || Se != DCTSIZE2-1 || Ah != 0 || Al != 0)
+	ERREXIT1(cinfo, JERR_BAD_PROG_SCRIPT, scanno);
+      /* Make sure components are not sent twice */
+      for (ci = 0; ci < ncomps; ci++) {
+	thisi = scanptr->component_index[ci];
+	if (component_sent[thisi])
+	  ERREXIT1(cinfo, JERR_BAD_SCAN_SCRIPT, scanno);
+	component_sent[thisi] = TRUE;
+      }
+    }
+  }
+
+  /* Now verify that everything got sent. */
+  if (cinfo->progressive_mode) {
+#ifdef C_PROGRESSIVE_SUPPORTED
+    /* For progressive mode, we only check that at least some DC data
+     * got sent for each component; the spec does not require that all bits
+     * of all coefficients be transmitted.  Would it be wiser to enforce
+     * transmission of all coefficient bits??
+     */
+    for (ci = 0; ci < cinfo->num_components; ci++) {
+      if (last_bitpos[ci][0] < 0)
+	ERREXIT(cinfo, JERR_MISSING_DATA);
+    }
+#endif
+  } else {
+    for (ci = 0; ci < cinfo->num_components; ci++) {
+      if (! component_sent[ci])
+	ERREXIT(cinfo, JERR_MISSING_DATA);
+    }
+  }
+}
+
+
+LOCAL(void)
+reduce_script (j_compress_ptr cinfo)
+/* Adapt scan script for use with reduced block size;
+ * assume that script has been validated before.
+ */
+{
+  jpeg_scan_info * scanptr;
+  int idxout, idxin;
+
+  /* Circumvent const declaration for this function */
+  scanptr = (jpeg_scan_info *) cinfo->scan_info;
+  idxout = 0;
+
+  for (idxin = 0; idxin < cinfo->num_scans; idxin++) {
+    /* After skipping, idxout becomes smaller than idxin */
+    if (idxin != idxout)
+      /* Copy rest of data;
+       * note we stay in given chunk of allocated memory.
+       */
+      scanptr[idxout] = scanptr[idxin];
+    if (scanptr[idxout].Ss > cinfo->lim_Se)
+      /* Entire scan out of range - skip this entry */
+      continue;
+    if (scanptr[idxout].Se > cinfo->lim_Se)
+      /* Limit scan to end of block */
+      scanptr[idxout].Se = cinfo->lim_Se;
+    idxout++;
+  }
+
+  cinfo->num_scans = idxout;
+}
+
+#endif /* C_MULTISCAN_FILES_SUPPORTED */
+
+
+LOCAL(void)
+select_scan_parameters (j_compress_ptr cinfo)
+/* Set up the scan parameters for the current scan */
+{
+  int ci;
+
+#ifdef C_MULTISCAN_FILES_SUPPORTED
+  if (cinfo->scan_info != NULL) {
+    /* Prepare for current scan --- the script is already validated */
+    my_master_ptr master = (my_master_ptr) cinfo->master;
+    const jpeg_scan_info * scanptr = cinfo->scan_info + master->scan_number;
+
+    cinfo->comps_in_scan = scanptr->comps_in_scan;
+    for (ci = 0; ci < scanptr->comps_in_scan; ci++) {
+      cinfo->cur_comp_info[ci] =
+	&cinfo->comp_info[scanptr->component_index[ci]];
+    }
+    if (cinfo->progressive_mode) {
+      cinfo->Ss = scanptr->Ss;
+      cinfo->Se = scanptr->Se;
+      cinfo->Ah = scanptr->Ah;
+      cinfo->Al = scanptr->Al;
+      return;
+    }
+  }
+  else
+#endif
+  {
+    /* Prepare for single sequential-JPEG scan containing all components */
+    if (cinfo->num_components > MAX_COMPS_IN_SCAN)
+      ERREXIT2(cinfo, JERR_COMPONENT_COUNT, cinfo->num_components,
+	       MAX_COMPS_IN_SCAN);
+    cinfo->comps_in_scan = cinfo->num_components;
+    for (ci = 0; ci < cinfo->num_components; ci++) {
+      cinfo->cur_comp_info[ci] = &cinfo->comp_info[ci];
+    }
+  }
+  cinfo->Ss = 0;
+  cinfo->Se = cinfo->block_size * cinfo->block_size - 1;
+  cinfo->Ah = 0;
+  cinfo->Al = 0;
+}
+
+
+LOCAL(void)
+per_scan_setup (j_compress_ptr cinfo)
+/* Do computations that are needed before processing a JPEG scan */
+/* cinfo->comps_in_scan and cinfo->cur_comp_info[] are already set */
+{
+  int ci, mcublks, tmp;
+  jpeg_component_info *compptr;
+
+  if (cinfo->comps_in_scan == 1) {
+
+    /* Noninterleaved (single-component) scan */
+    compptr = cinfo->cur_comp_info[0];
+
+    /* Overall image size in MCUs */
+    cinfo->MCUs_per_row = compptr->width_in_blocks;
+    cinfo->MCU_rows_in_scan = compptr->height_in_blocks;
+
+    /* For noninterleaved scan, always one block per MCU */
+    compptr->MCU_width = 1;
+    compptr->MCU_height = 1;
+    compptr->MCU_blocks = 1;
+    compptr->MCU_sample_width = compptr->DCT_h_scaled_size;
+    compptr->last_col_width = 1;
+    /* For noninterleaved scans, it is convenient to define last_row_height
+     * as the number of block rows present in the last iMCU row.
+     */
+    tmp = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
+    if (tmp == 0) tmp = compptr->v_samp_factor;
+    compptr->last_row_height = tmp;
+
+    /* Prepare array describing MCU composition */
+    cinfo->blocks_in_MCU = 1;
+    cinfo->MCU_membership[0] = 0;
+
+  } else {
+
+    /* Interleaved (multi-component) scan */
+    if (cinfo->comps_in_scan <= 0 || cinfo->comps_in_scan > MAX_COMPS_IN_SCAN)
+      ERREXIT2(cinfo, JERR_COMPONENT_COUNT, cinfo->comps_in_scan,
+	       MAX_COMPS_IN_SCAN);
+
+    /* Overall image size in MCUs */
+    cinfo->MCUs_per_row = (JDIMENSION)
+      jdiv_round_up((long) cinfo->jpeg_width,
+		    (long) (cinfo->max_h_samp_factor * cinfo->block_size));
+    cinfo->MCU_rows_in_scan = cinfo->total_iMCU_rows;
+
+    cinfo->blocks_in_MCU = 0;
+
+    for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+      compptr = cinfo->cur_comp_info[ci];
+      /* Sampling factors give # of blocks of component in each MCU */
+      compptr->MCU_width = compptr->h_samp_factor;
+      compptr->MCU_height = compptr->v_samp_factor;
+      compptr->MCU_blocks = compptr->MCU_width * compptr->MCU_height;
+      compptr->MCU_sample_width = compptr->MCU_width * compptr->DCT_h_scaled_size;
+      /* Figure number of non-dummy blocks in last MCU column & row */
+      tmp = (int) (compptr->width_in_blocks % compptr->MCU_width);
+      if (tmp == 0) tmp = compptr->MCU_width;
+      compptr->last_col_width = tmp;
+      tmp = (int) (compptr->height_in_blocks % compptr->MCU_height);
+      if (tmp == 0) tmp = compptr->MCU_height;
+      compptr->last_row_height = tmp;
+      /* Prepare array describing MCU composition */
+      mcublks = compptr->MCU_blocks;
+      if (cinfo->blocks_in_MCU + mcublks > C_MAX_BLOCKS_IN_MCU)
+	ERREXIT(cinfo, JERR_BAD_MCU_SIZE);
+      while (mcublks-- > 0) {
+	cinfo->MCU_membership[cinfo->blocks_in_MCU++] = ci;
+      }
+    }
+
+  }
+
+  /* Convert restart specified in rows to actual MCU count. */
+  /* Note that count must fit in 16 bits, so we provide limiting. */
+  if (cinfo->restart_in_rows > 0) {
+    long nominal = (long) cinfo->restart_in_rows * (long) cinfo->MCUs_per_row;
+    cinfo->restart_interval = (unsigned int) MIN(nominal, 65535L);
+  }
+}
+
+
+/*
+ * Per-pass setup.
+ * This is called at the beginning of each pass.  We determine which modules
+ * will be active during this pass and give them appropriate start_pass calls.
+ * We also set is_last_pass to indicate whether any more passes will be
+ * required.
+ */
+
+METHODDEF(void)
+prepare_for_pass (j_compress_ptr cinfo)
+{
+  my_master_ptr master = (my_master_ptr) cinfo->master;
+
+  switch (master->pass_type) {
+  case main_pass:
+    /* Initial pass: will collect input data, and do either Huffman
+     * optimization or data output for the first scan.
+     */
+    select_scan_parameters(cinfo);
+    per_scan_setup(cinfo);
+    if (! cinfo->raw_data_in) {
+      (*cinfo->cconvert->start_pass) (cinfo);
+      (*cinfo->downsample->start_pass) (cinfo);
+      (*cinfo->prep->start_pass) (cinfo, JBUF_PASS_THRU);
+    }
+    (*cinfo->fdct->start_pass) (cinfo);
+    (*cinfo->entropy->start_pass) (cinfo, cinfo->optimize_coding);
+    (*cinfo->coef->start_pass) (cinfo,
+				(master->total_passes > 1 ?
+				 JBUF_SAVE_AND_PASS : JBUF_PASS_THRU));
+    (*cinfo->main->start_pass) (cinfo, JBUF_PASS_THRU);
+    if (cinfo->optimize_coding) {
+      /* No immediate data output; postpone writing frame/scan headers */
+      master->pub.call_pass_startup = FALSE;
+    } else {
+      /* Will write frame/scan headers at first jpeg_write_scanlines call */
+      master->pub.call_pass_startup = TRUE;
+    }
+    break;
+#ifdef ENTROPY_OPT_SUPPORTED
+  case huff_opt_pass:
+    /* Do Huffman optimization for a scan after the first one. */
+    select_scan_parameters(cinfo);
+    per_scan_setup(cinfo);
+    if (cinfo->Ss != 0 || cinfo->Ah == 0) {
+      (*cinfo->entropy->start_pass) (cinfo, TRUE);
+      (*cinfo->coef->start_pass) (cinfo, JBUF_CRANK_DEST);
+      master->pub.call_pass_startup = FALSE;
+      break;
+    }
+    /* Special case: Huffman DC refinement scans need no Huffman table
+     * and therefore we can skip the optimization pass for them.
+     */
+    master->pass_type = output_pass;
+    master->pass_number++;
+    /*FALLTHROUGH*/
+#endif
+  case output_pass:
+    /* Do a data-output pass. */
+    /* We need not repeat per-scan setup if prior optimization pass did it. */
+    if (! cinfo->optimize_coding) {
+      select_scan_parameters(cinfo);
+      per_scan_setup(cinfo);
+    }
+    (*cinfo->entropy->start_pass) (cinfo, FALSE);
+    (*cinfo->coef->start_pass) (cinfo, JBUF_CRANK_DEST);
+    /* We emit frame/scan headers now */
+    if (master->scan_number == 0)
+      (*cinfo->marker->write_frame_header) (cinfo);
+    (*cinfo->marker->write_scan_header) (cinfo);
+    master->pub.call_pass_startup = FALSE;
+    break;
+  default:
+    ERREXIT(cinfo, JERR_NOT_COMPILED);
+  }
+
+  master->pub.is_last_pass = (master->pass_number == master->total_passes-1);
+
+  /* Set up progress monitor's pass info if present */
+  if (cinfo->progress != NULL) {
+    cinfo->progress->completed_passes = master->pass_number;
+    cinfo->progress->total_passes = master->total_passes;
+  }
+}
+
+
+/*
+ * Special start-of-pass hook.
+ * This is called by jpeg_write_scanlines if call_pass_startup is TRUE.
+ * In single-pass processing, we need this hook because we don't want to
+ * write frame/scan headers during jpeg_start_compress; we want to let the
+ * application write COM markers etc. between jpeg_start_compress and the
+ * jpeg_write_scanlines loop.
+ * In multi-pass processing, this routine is not used.
+ */
+
+METHODDEF(void)
+pass_startup (j_compress_ptr cinfo)
+{
+  cinfo->master->call_pass_startup = FALSE; /* reset flag so call only once */
+
+  (*cinfo->marker->write_frame_header) (cinfo);
+  (*cinfo->marker->write_scan_header) (cinfo);
+}
+
+
+/*
+ * Finish up at end of pass.
+ */
+
+METHODDEF(void)
+finish_pass_master (j_compress_ptr cinfo)
+{
+  my_master_ptr master = (my_master_ptr) cinfo->master;
+
+  /* The entropy coder always needs an end-of-pass call,
+   * either to analyze statistics or to flush its output buffer.
+   */
+  (*cinfo->entropy->finish_pass) (cinfo);
+
+  /* Update state for next pass */
+  switch (master->pass_type) {
+  case main_pass:
+    /* next pass is either output of scan 0 (after optimization)
+     * or output of scan 1 (if no optimization).
+     */
+    master->pass_type = output_pass;
+    if (! cinfo->optimize_coding)
+      master->scan_number++;
+    break;
+  case huff_opt_pass:
+    /* next pass is always output of current scan */
+    master->pass_type = output_pass;
+    break;
+  case output_pass:
+    /* next pass is either optimization or output of next scan */
+    if (cinfo->optimize_coding)
+      master->pass_type = huff_opt_pass;
+    master->scan_number++;
+    break;
+  }
+
+  master->pass_number++;
+}
+
+
+/*
+ * Initialize master compression control.
+ */
+
+GLOBAL(void)
+jinit_c_master_control (j_compress_ptr cinfo, boolean transcode_only)
+{
+  my_master_ptr master;
+
+  master = (my_master_ptr) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, SIZEOF(my_comp_master));
+  cinfo->master = &master->pub;
+  master->pub.prepare_for_pass = prepare_for_pass;
+  master->pub.pass_startup = pass_startup;
+  master->pub.finish_pass = finish_pass_master;
+  master->pub.is_last_pass = FALSE;
+
+  /* Validate parameters, determine derived values */
+  initial_setup(cinfo);
+
+  if (cinfo->scan_info != NULL) {
+#ifdef C_MULTISCAN_FILES_SUPPORTED
+    validate_script(cinfo);
+    if (cinfo->block_size < DCTSIZE)
+      reduce_script(cinfo);
+#else
+    ERREXIT(cinfo, JERR_NOT_COMPILED);
+#endif
+  } else {
+    cinfo->progressive_mode = FALSE;
+    cinfo->num_scans = 1;
+  }
+
+  if (cinfo->optimize_coding)
+    cinfo->arith_code = FALSE; /* disable arithmetic coding */
+  else if (! cinfo->arith_code &&
+	   (cinfo->progressive_mode ||
+	    (cinfo->block_size > 1 && cinfo->block_size < DCTSIZE)))
+    /* TEMPORARY HACK ??? */
+    /* assume default tables no good for progressive or reduced AC mode */
+    cinfo->optimize_coding = TRUE; /* force Huffman optimization */
+
+  /* Initialize my private state */
+  if (transcode_only) {
+    /* no main pass in transcoding */
+    if (cinfo->optimize_coding)
+      master->pass_type = huff_opt_pass;
+    else
+      master->pass_type = output_pass;
+  } else {
+    /* for normal compression, first pass is always this type: */
+    master->pass_type = main_pass;
+  }
+  master->scan_number = 0;
+  master->pass_number = 0;
+  if (cinfo->optimize_coding)
+    master->total_passes = cinfo->num_scans * 2;
+  else
+    master->total_passes = cinfo->num_scans;
+}
@@ -0,0 +1,244 @@
+/*
+ * jcomapi.c
+ *
+ * Copyright (C) 1994-1997, Thomas G. Lane.
+ * Modified 2019 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains application interface routines that are used for both
+ * compression and decompression.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/*
+ * Abort processing of a JPEG compression or decompression operation,
+ * but don't destroy the object itself.
+ *
+ * For this, we merely clean up all the nonpermanent memory pools.
+ * Note that temp files (virtual arrays) are not allowed to belong to
+ * the permanent pool, so we will be able to close all temp files here.
+ * Closing a data source or destination, if necessary, is the application's
+ * responsibility.
+ */
+
+GLOBAL(void)
+jpeg_abort (j_common_ptr cinfo)
+{
+  int pool;
+
+  /* Do nothing if called on a not-initialized or destroyed JPEG object. */
+  if (cinfo->mem == NULL)
+    return;
+
+  /* Releasing pools in reverse order might help avoid fragmentation
+   * with some (brain-damaged) malloc libraries.
+   */
+  for (pool = JPOOL_NUMPOOLS-1; pool > JPOOL_PERMANENT; pool--) {
+    (*cinfo->mem->free_pool) (cinfo, pool);
+  }
+
+  /* Reset overall state for possible reuse of object */
+  if (cinfo->is_decompressor) {
+    cinfo->global_state = DSTATE_START;
+    /* Try to keep application from accessing now-deleted marker list.
+     * A bit kludgy to do it here, but this is the most central place.
+     */
+    ((j_decompress_ptr) cinfo)->marker_list = NULL;
+  } else {
+    cinfo->global_state = CSTATE_START;
+  }
+}
+
+
+/*
+ * Destruction of a JPEG object.
+ *
+ * Everything gets deallocated except the master jpeg_compress_struct itself
+ * and the error manager struct.  Both of these are supplied by the application
+ * and must be freed, if necessary, by the application.  (Often they are on
+ * the stack and so don't need to be freed anyway.)
+ * Closing a data source or destination, if necessary, is the application's
+ * responsibility.
+ */
+
+GLOBAL(void)
+jpeg_destroy (j_common_ptr cinfo)
+{
+  /* We need only tell the memory manager to release everything. */
+  /* NB: mem pointer is NULL if memory mgr failed to initialize. */
+  if (cinfo->mem != NULL)
+    (*cinfo->mem->self_destruct) (cinfo);
+  cinfo->mem = NULL;		/* be safe if jpeg_destroy is called twice */
+  cinfo->global_state = 0;	/* mark it destroyed */
+}
+
+
+/*
+ * Convenience routines for allocating quantization and Huffman tables.
+ * (Would jutils.c be a more reasonable place to put these?)
+ */
+
+GLOBAL(JQUANT_TBL *)
+jpeg_alloc_quant_table (j_common_ptr cinfo)
+{
+  JQUANT_TBL *tbl;
+
+  tbl = (JQUANT_TBL *)
+    (*cinfo->mem->alloc_small) (cinfo, JPOOL_PERMANENT, SIZEOF(JQUANT_TBL));
+  tbl->sent_table = FALSE;	/* make sure this is false in any new table */
+  return tbl;
+}
+
+
+GLOBAL(JHUFF_TBL *)
+jpeg_alloc_huff_table (j_common_ptr cinfo)
+{
+  JHUFF_TBL *tbl;
+
+  tbl = (JHUFF_TBL *)
+    (*cinfo->mem->alloc_small) (cinfo, JPOOL_PERMANENT, SIZEOF(JHUFF_TBL));
+  tbl->sent_table = FALSE;	/* make sure this is false in any new table */
+  return tbl;
+}
+
+
+/*
+ * Set up the standard Huffman tables (cf. JPEG standard section K.3).
+ * IMPORTANT: these are only valid for 8-bit data precision!
+ * (Would jutils.c be a more reasonable place to put this?)
+ */
+
+GLOBAL(JHUFF_TBL *)
+jpeg_std_huff_table (j_common_ptr cinfo, boolean isDC, int tblno)
+{
+  JHUFF_TBL **htblptr, *htbl;
+  const UINT8 *bits, *val;
+  int nsymbols, len;
+
+  static const UINT8 bits_dc_luminance[17] =
+    { /* 0-base */ 0, 0, 1, 5, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0 };
+  static const UINT8 val_dc_luminance[] =
+    { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 };
+
+  static const UINT8 bits_dc_chrominance[17] =
+    { /* 0-base */ 0, 0, 3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0 };
+  static const UINT8 val_dc_chrominance[] =
+    { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 };
+
+  static const UINT8 bits_ac_luminance[17] =
+    { /* 0-base */ 0, 0, 2, 1, 3, 3, 2, 4, 3, 5, 5, 4, 4, 0, 0, 1, 0x7d };
+  static const UINT8 val_ac_luminance[] =
+    { 0x01, 0x02, 0x03, 0x00, 0x04, 0x11, 0x05, 0x12,
+      0x21, 0x31, 0x41, 0x06, 0x13, 0x51, 0x61, 0x07,
+      0x22, 0x71, 0x14, 0x32, 0x81, 0x91, 0xa1, 0x08,
+      0x23, 0x42, 0xb1, 0xc1, 0x15, 0x52, 0xd1, 0xf0,
+      0x24, 0x33, 0x62, 0x72, 0x82, 0x09, 0x0a, 0x16,
+      0x17, 0x18, 0x19, 0x1a, 0x25, 0x26, 0x27, 0x28,
+      0x29, 0x2a, 0x34, 0x35, 0x36, 0x37, 0x38, 0x39,
+      0x3a, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48, 0x49,
+      0x4a, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58, 0x59,
+      0x5a, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68, 0x69,
+      0x6a, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78, 0x79,
+      0x7a, 0x83, 0x84, 0x85, 0x86, 0x87, 0x88, 0x89,
+      0x8a, 0x92, 0x93, 0x94, 0x95, 0x96, 0x97, 0x98,
+      0x99, 0x9a, 0xa2, 0xa3, 0xa4, 0xa5, 0xa6, 0xa7,
+      0xa8, 0xa9, 0xaa, 0xb2, 0xb3, 0xb4, 0xb5, 0xb6,
+      0xb7, 0xb8, 0xb9, 0xba, 0xc2, 0xc3, 0xc4, 0xc5,
+      0xc6, 0xc7, 0xc8, 0xc9, 0xca, 0xd2, 0xd3, 0xd4,
+      0xd5, 0xd6, 0xd7, 0xd8, 0xd9, 0xda, 0xe1, 0xe2,
+      0xe3, 0xe4, 0xe5, 0xe6, 0xe7, 0xe8, 0xe9, 0xea,
+      0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8,
+      0xf9, 0xfa };
+
+  static const UINT8 bits_ac_chrominance[17] =
+    { /* 0-base */ 0, 0, 2, 1, 2, 4, 4, 3, 4, 7, 5, 4, 4, 0, 1, 2, 0x77 };
+  static const UINT8 val_ac_chrominance[] =
+    { 0x00, 0x01, 0x02, 0x03, 0x11, 0x04, 0x05, 0x21,
+      0x31, 0x06, 0x12, 0x41, 0x51, 0x07, 0x61, 0x71,
+      0x13, 0x22, 0x32, 0x81, 0x08, 0x14, 0x42, 0x91,
+      0xa1, 0xb1, 0xc1, 0x09, 0x23, 0x33, 0x52, 0xf0,
+      0x15, 0x62, 0x72, 0xd1, 0x0a, 0x16, 0x24, 0x34,
+      0xe1, 0x25, 0xf1, 0x17, 0x18, 0x19, 0x1a, 0x26,
+      0x27, 0x28, 0x29, 0x2a, 0x35, 0x36, 0x37, 0x38,
+      0x39, 0x3a, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48,
+      0x49, 0x4a, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58,
+      0x59, 0x5a, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68,
+      0x69, 0x6a, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78,
+      0x79, 0x7a, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87,
+      0x88, 0x89, 0x8a, 0x92, 0x93, 0x94, 0x95, 0x96,
+      0x97, 0x98, 0x99, 0x9a, 0xa2, 0xa3, 0xa4, 0xa5,
+      0xa6, 0xa7, 0xa8, 0xa9, 0xaa, 0xb2, 0xb3, 0xb4,
+      0xb5, 0xb6, 0xb7, 0xb8, 0xb9, 0xba, 0xc2, 0xc3,
+      0xc4, 0xc5, 0xc6, 0xc7, 0xc8, 0xc9, 0xca, 0xd2,
+      0xd3, 0xd4, 0xd5, 0xd6, 0xd7, 0xd8, 0xd9, 0xda,
+      0xe2, 0xe3, 0xe4, 0xe5, 0xe6, 0xe7, 0xe8, 0xe9,
+      0xea, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8,
+      0xf9, 0xfa };
+
+  if (cinfo->is_decompressor) {
+    if (isDC)
+      htblptr = ((j_decompress_ptr) cinfo)->dc_huff_tbl_ptrs;
+    else
+      htblptr = ((j_decompress_ptr) cinfo)->ac_huff_tbl_ptrs;
+  } else {
+    if (isDC)
+      htblptr = ((j_compress_ptr) cinfo)->dc_huff_tbl_ptrs;
+    else
+      htblptr = ((j_compress_ptr) cinfo)->ac_huff_tbl_ptrs;
+  }
+
+  switch (tblno) {
+  case 0:
+    if (isDC) {
+      bits = bits_dc_luminance;
+      val = val_dc_luminance;
+    } else {
+      bits = bits_ac_luminance;
+      val = val_ac_luminance;
+    }
+    break;
+  case 1:
+    if (isDC) {
+      bits = bits_dc_chrominance;
+      val = val_dc_chrominance;
+    } else {
+      bits = bits_ac_chrominance;
+      val = val_ac_chrominance;
+    }
+    break;
+  default:
+    ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tblno);
+    return NULL; /* avoid compiler warnings for uninitialized variables */
+  }
+
+  if (htblptr[tblno] == NULL)
+    htblptr[tblno] = jpeg_alloc_huff_table(cinfo);
+
+  htbl = htblptr[tblno];
+
+  /* Copy the number-of-symbols-of-each-code-length counts */
+  MEMCOPY(htbl->bits, bits, SIZEOF(htbl->bits));
+
+  /* Validate the counts.  We do this here mainly so we can copy the right
+   * number of symbols from the val[] array, without risking marching off
+   * the end of memory.  jxhuff.c will do a more thorough test later.
+   */
+  nsymbols = 0;
+  for (len = 1; len <= 16; len++)
+    nsymbols += bits[len];
+  if (nsymbols > 256)
+    ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);
+
+  if (nsymbols > 0)
+    MEMCOPY(htbl->huffval, val, nsymbols * SIZEOF(UINT8));
+
+  /* Initialize sent_table FALSE so table will be written to JPEG file. */
+  htbl->sent_table = FALSE;
+
+  return htbl;
+}
@@ -0,0 +1,64 @@
+/* jconfig.vc --- jconfig.h for Microsoft Visual C++ on Windows 9x or NT.
+ * This file also works for Borland/Embarcadero C++ for Win32 or Win64
+ * (CLI: bcc32, bcc32c, bcc32x, bcc64; GUI IDE: C++Builder/RAD Studio).
+ * See jconfig.txt for explanations.
+ */
+
+#define HAVE_PROTOTYPES
+#define HAVE_UNSIGNED_CHAR
+#define HAVE_UNSIGNED_SHORT
+/* #define void char */
+/* #define const */
+#undef CHAR_IS_UNSIGNED
+#define HAVE_STDDEF_H
+#define HAVE_STDLIB_H
+#undef NEED_BSD_STRINGS
+#undef NEED_SYS_TYPES_H
+#undef NEED_FAR_POINTERS	/* we presume a 32-bit flat memory model */
+#undef NEED_SHORT_EXTERNAL_NAMES
+#undef INCOMPLETE_TYPES_BROKEN
+
+/* Define "boolean" as unsigned char, not enum, per Windows custom */
+#ifndef __RPCNDR_H__		/* don't conflict if rpcndr.h already read */
+typedef unsigned char boolean;
+#endif
+#ifndef FALSE			/* in case these macros already exist */
+#define FALSE	0		/* values of boolean */
+#endif
+#ifndef TRUE
+#define TRUE	1
+#endif
+#define HAVE_BOOLEAN		/* prevent jmorecfg.h from redefining it */
+
+/* Define custom RGB color order, prevent jmorecfg.h from redefinition */
+#undef JPEG_HAVE_RGB_CUSTOM
+/* Use Windows custom BGR color order defined in jmorecfg.h */
+#undef JPEG_USE_RGB_CUSTOM
+
+/* Define custom file I/O functions, prevent jinclude.h from redefinition */
+#undef JPEG_HAVE_FILE_IO_CUSTOM
+/* Use Delphi custom file I/O functions defined in jinclude.h */
+#undef JPEG_USE_FILE_IO_CUSTOM
+
+
+#ifdef JPEG_INTERNALS
+
+#undef RIGHT_SHIFT_IS_UNSIGNED
+
+#endif /* JPEG_INTERNALS */
+
+#ifdef JPEG_CJPEG_DJPEG
+
+#define BMP_SUPPORTED		/* BMP image file format */
+#define GIF_SUPPORTED		/* GIF image file format */
+#define PPM_SUPPORTED		/* PBMPLUS PPM/PGM image file format */
+#undef RLE_SUPPORTED		/* Utah RLE image file format */
+#define TARGA_SUPPORTED		/* Targa image file format */
+
+#define TWO_FILE_COMMANDLINE	/* optional */
+#define USE_SETMODE	/* Microsoft/Borland/Embarcadero have setmode() */
+#undef NEED_SIGNAL_CATCHER
+#undef DONT_USE_B_MODE
+#undef PROGRESS_REPORT		/* optional */
+
+#endif /* JPEG_CJPEG_DJPEG */
@@ -0,0 +1,591 @@
+/*
+ * jcparam.c
+ *
+ * Copyright (C) 1991-1998, Thomas G. Lane.
+ * Modified 2003-2022 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains optional default-setting code for the JPEG compressor.
+ * Applications do not have to use this file, but those that don't use it
+ * must know a lot more about the innards of the JPEG code.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/*
+ * Quantization table setup routines
+ */
+
+GLOBAL(void)
+jpeg_add_quant_table (j_compress_ptr cinfo, int which_tbl,
+		      const unsigned int *basic_table,
+		      int scale_factor, boolean force_baseline)
+/* Define a quantization table equal to the basic_table times
+ * a scale factor (given as a percentage).
+ * If force_baseline is TRUE, the computed quantization table entries
+ * are limited to 1..255 for JPEG baseline compatibility.
+ */
+{
+  JQUANT_TBL ** qtblptr;
+  int i;
+  long temp;
+
+  /* Safety check to ensure start_compress not called yet. */
+  if (cinfo->global_state != CSTATE_START)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+
+  if (which_tbl < 0 || which_tbl >= NUM_QUANT_TBLS)
+    ERREXIT1(cinfo, JERR_DQT_INDEX, which_tbl);
+
+  qtblptr = & cinfo->quant_tbl_ptrs[which_tbl];
+
+  if (*qtblptr == NULL)
+    *qtblptr = jpeg_alloc_quant_table((j_common_ptr) cinfo);
+
+  for (i = 0; i < DCTSIZE2; i++) {
+    temp = ((long) basic_table[i] * scale_factor + 50L) / 100L;
+    /* limit the values to the valid range */
+    if (temp <= 0L) temp = 1L;
+    if (temp > 32767L) temp = 32767L; /* max quantizer needed for 12 bits */
+    if (force_baseline && temp > 255L)
+      temp = 255L;		/* limit to baseline range if requested */
+    (*qtblptr)->quantval[i] = (UINT16) temp;
+  }
+
+  /* Initialize sent_table FALSE so table will be written to JPEG file. */
+  (*qtblptr)->sent_table = FALSE;
+}
+
+
+/* These are the sample quantization tables given in JPEG spec section K.1.
+ * NOTE: chrominance DC value is changed from 17 to 16 for lossless support.
+ * The spec says that the values given produce "good" quality,
+ * and when divided by 2, "very good" quality.
+ */
+static const unsigned int std_luminance_quant_tbl[DCTSIZE2] = {
+  16,  11,  10,  16,  24,  40,  51,  61,
+  12,  12,  14,  19,  26,  58,  60,  55,
+  14,  13,  16,  24,  40,  57,  69,  56,
+  14,  17,  22,  29,  51,  87,  80,  62,
+  18,  22,  37,  56,  68, 109, 103,  77,
+  24,  35,  55,  64,  81, 104, 113,  92,
+  49,  64,  78,  87, 103, 121, 120, 101,
+  72,  92,  95,  98, 112, 100, 103,  99
+};
+static const unsigned int std_chrominance_quant_tbl[DCTSIZE2] = {
+  16,  18,  24,  47,  99,  99,  99,  99,
+  18,  21,  26,  66,  99,  99,  99,  99,
+  24,  26,  56,  99,  99,  99,  99,  99,
+  47,  66,  99,  99,  99,  99,  99,  99,
+  99,  99,  99,  99,  99,  99,  99,  99,
+  99,  99,  99,  99,  99,  99,  99,  99,
+  99,  99,  99,  99,  99,  99,  99,  99,
+  99,  99,  99,  99,  99,  99,  99,  99
+};
+
+
+GLOBAL(void)
+jpeg_default_qtables (j_compress_ptr cinfo, boolean force_baseline)
+/* Set or change the 'quality' (quantization) setting, using default tables
+ * and straight percentage-scaling quality scales.
+ * This entry point allows different scalings for luminance and chrominance.
+ */
+{
+  /* Set up two quantization tables using the specified scaling */
+  jpeg_add_quant_table(cinfo, 0, std_luminance_quant_tbl,
+		       cinfo->q_scale_factor[0], force_baseline);
+  jpeg_add_quant_table(cinfo, 1, std_chrominance_quant_tbl,
+		       cinfo->q_scale_factor[1], force_baseline);
+}
+
+
+GLOBAL(void)
+jpeg_set_linear_quality (j_compress_ptr cinfo, int scale_factor,
+			 boolean force_baseline)
+/* Set or change the 'quality' (quantization) setting, using default tables
+ * and a straight percentage-scaling quality scale.  In most cases it's better
+ * to use jpeg_set_quality (below); this entry point is provided for
+ * applications that insist on a linear percentage scaling.
+ */
+{
+  /* Set up two quantization tables using the specified scaling */
+  jpeg_add_quant_table(cinfo, 0, std_luminance_quant_tbl,
+		       scale_factor, force_baseline);
+  jpeg_add_quant_table(cinfo, 1, std_chrominance_quant_tbl,
+		       scale_factor, force_baseline);
+}
+
+
+GLOBAL(int)
+jpeg_quality_scaling (int quality)
+/* Convert a user-specified quality rating to a percentage scaling factor
+ * for an underlying quantization table, using our recommended scaling curve.
+ * The input 'quality' factor should be 0 (terrible) to 100 (very good).
+ */
+{
+  /* Safety limit on quality factor.  Convert 0 to 1 to avoid zero divide. */
+  if (quality <= 0) quality = 1;
+  if (quality > 100) quality = 100;
+
+  /* The basic table is used as-is (scaling 100) for a quality of 50.
+   * Qualities 50..100 are converted to scaling percentage 200 - 2*Q;
+   * note that at Q=100 the scaling is 0, which will cause jpeg_add_quant_table
+   * to make all the table entries 1 (hence, minimum quantization loss).
+   * Qualities 1..50 are converted to scaling percentage 5000/Q.
+   */
+  if (quality < 50)
+    quality = 5000 / quality;
+  else
+    quality = 200 - quality*2;
+
+  return quality;
+}
+
+
+GLOBAL(void)
+jpeg_set_quality (j_compress_ptr cinfo, int quality, boolean force_baseline)
+/* Set or change the 'quality' (quantization) setting, using default tables.
+ * This is the standard quality-adjusting entry point for typical user
+ * interfaces; only those who want detailed control over quantization tables
+ * would use the preceding routines directly.
+ */
+{
+  /* Convert user 0-100 rating to percentage scaling */
+  quality = jpeg_quality_scaling(quality);
+
+  /* Set up standard quality tables */
+  jpeg_set_linear_quality(cinfo, quality, force_baseline);
+}
+
+
+/*
+ * Reset standard Huffman tables
+ */
+
+LOCAL(void)
+std_huff_tables (j_compress_ptr cinfo)
+{
+  if (cinfo->dc_huff_tbl_ptrs[0] != NULL)
+    (void) jpeg_std_huff_table((j_common_ptr) cinfo, TRUE, 0);
+
+  if (cinfo->ac_huff_tbl_ptrs[0] != NULL)
+    (void) jpeg_std_huff_table((j_common_ptr) cinfo, FALSE, 0);
+
+  if (cinfo->dc_huff_tbl_ptrs[1] != NULL)
+    (void) jpeg_std_huff_table((j_common_ptr) cinfo, TRUE, 1);
+
+  if (cinfo->ac_huff_tbl_ptrs[1] != NULL)
+    (void) jpeg_std_huff_table((j_common_ptr) cinfo, FALSE, 1);
+}
+
+
+/*
+ * Default parameter setup for compression.
+ *
+ * Applications that don't choose to use this routine must do their
+ * own setup of all these parameters.  Alternately, you can call this
+ * to establish defaults and then alter parameters selectively.  This
+ * is the recommended approach since, if we add any new parameters,
+ * your code will still work (they'll be set to reasonable defaults).
+ */
+
+GLOBAL(void)
+jpeg_set_defaults (j_compress_ptr cinfo)
+{
+  int i;
+
+  /* Safety check to ensure start_compress not called yet. */
+  if (cinfo->global_state != CSTATE_START)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+
+  /* Allocate comp_info array large enough for maximum component count.
+   * Array is made permanent in case application wants to compress
+   * multiple images at same param settings.
+   */
+  if (cinfo->comp_info == NULL)
+    cinfo->comp_info = (jpeg_component_info *)
+      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_PERMANENT,
+				  MAX_COMPONENTS * SIZEOF(jpeg_component_info));
+
+  /* Initialize everything not dependent on the color space */
+
+  cinfo->scale_num = 1;		/* 1:1 scaling */
+  cinfo->scale_denom = 1;
+  cinfo->data_precision = BITS_IN_JSAMPLE;
+  /* Set up two quantization tables using default quality of 75 */
+  jpeg_set_quality(cinfo, 75, TRUE);
+  /* Reset standard Huffman tables */
+  std_huff_tables(cinfo);
+
+  /* Initialize default arithmetic coding conditioning */
+  for (i = 0; i < NUM_ARITH_TBLS; i++) {
+    cinfo->arith_dc_L[i] = 0;
+    cinfo->arith_dc_U[i] = 1;
+    cinfo->arith_ac_K[i] = 5;
+  }
+
+  /* Default is no multiple-scan output */
+  cinfo->scan_info = NULL;
+  cinfo->num_scans = 0;
+
+  /* Expect normal source image, not raw downsampled data */
+  cinfo->raw_data_in = FALSE;
+
+  /* The standard Huffman tables are only valid for 8-bit data precision.
+   * If the precision is higher, use arithmetic coding.
+   * (Alternatively, using Huffman coding would be possible with forcing
+   * optimization on so that usable tables will be computed, or by
+   * supplying default tables that are valid for the desired precision.)
+   * Otherwise, use Huffman coding by default.
+   */
+  cinfo->arith_code = cinfo->data_precision > 8 ? TRUE : FALSE;
+
+  /* By default, don't do extra passes to optimize entropy coding */
+  cinfo->optimize_coding = FALSE;
+
+  /* By default, use the simpler non-cosited sampling alignment */
+  cinfo->CCIR601_sampling = FALSE;
+
+  /* By default, apply fancy downsampling */
+  cinfo->do_fancy_downsampling = TRUE;
+
+  /* No input smoothing */
+  cinfo->smoothing_factor = 0;
+
+  /* DCT algorithm preference */
+  cinfo->dct_method = JDCT_DEFAULT;
+
+  /* No restart markers */
+  cinfo->restart_interval = 0;
+  cinfo->restart_in_rows = 0;
+
+  /* Fill in default JFIF marker parameters.  Note that whether the marker
+   * will actually be written is determined by jpeg_set_colorspace.
+   *
+   * By default, the library emits JFIF version code 1.01.
+   * An application that wants to emit JFIF 1.02 extension markers should set
+   * JFIF_minor_version to 2.  We could probably get away with just defaulting
+   * to 1.02, but there may still be some decoders in use that will complain
+   * about that; saying 1.01 should minimize compatibility problems.
+   *
+   * For wide gamut colorspaces (BG_RGB and BG_YCC), the major version will be
+   * overridden by jpeg_set_colorspace and set to 2.
+   */
+  cinfo->JFIF_major_version = 1; /* Default JFIF version = 1.01 */
+  cinfo->JFIF_minor_version = 1;
+  cinfo->density_unit = 0;	/* Pixel size is unknown by default */
+  cinfo->X_density = 1;		/* Pixel aspect ratio is square by default */
+  cinfo->Y_density = 1;
+
+  /* No color transform */
+  cinfo->color_transform = JCT_NONE;
+
+  /* Choose JPEG colorspace based on input space, set defaults accordingly */
+
+  jpeg_default_colorspace(cinfo);
+}
+
+
+/*
+ * Select an appropriate JPEG colorspace for in_color_space.
+ */
+
+GLOBAL(void)
+jpeg_default_colorspace (j_compress_ptr cinfo)
+{
+  switch (cinfo->in_color_space) {
+  case JCS_UNKNOWN:
+    jpeg_set_colorspace(cinfo, JCS_UNKNOWN);
+    break;
+  case JCS_GRAYSCALE:
+    jpeg_set_colorspace(cinfo, JCS_GRAYSCALE);
+    break;
+  case JCS_RGB:
+    jpeg_set_colorspace(cinfo, JCS_YCbCr);
+    break;
+  case JCS_YCbCr:
+    jpeg_set_colorspace(cinfo, JCS_YCbCr);
+    break;
+  case JCS_CMYK:
+    jpeg_set_colorspace(cinfo, JCS_CMYK); /* By default, no translation */
+    break;
+  case JCS_YCCK:
+    jpeg_set_colorspace(cinfo, JCS_YCCK);
+    break;
+  case JCS_BG_RGB:
+    /* No translation for now -- conversion to BG_YCC not yet supportet */
+    jpeg_set_colorspace(cinfo, JCS_BG_RGB);
+    break;
+  case JCS_BG_YCC:
+    jpeg_set_colorspace(cinfo, JCS_BG_YCC);
+    break;
+  default:
+    ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE);
+  }
+}
+
+
+/*
+ * Set the JPEG colorspace, and choose colorspace-dependent default values.
+ */
+
+GLOBAL(void)
+jpeg_set_colorspace (j_compress_ptr cinfo, J_COLOR_SPACE colorspace)
+{
+  jpeg_component_info * compptr;
+  int ci;
+
+#define SET_COMP(index,id,hsamp,vsamp,quant,dctbl,actbl)  \
+  (compptr = &cinfo->comp_info[index], \
+   compptr->component_id = (id), \
+   compptr->h_samp_factor = (hsamp), \
+   compptr->v_samp_factor = (vsamp), \
+   compptr->quant_tbl_no = (quant), \
+   compptr->dc_tbl_no = (dctbl), \
+   compptr->ac_tbl_no = (actbl) )
+
+  /* Safety check to ensure start_compress not called yet. */
+  if (cinfo->global_state != CSTATE_START)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+
+  /* For all colorspaces, we use Q and Huff tables 0 for luminance components,
+   * tables 1 for chrominance components.
+   */
+
+  cinfo->jpeg_color_space = colorspace;
+
+  cinfo->write_JFIF_header = FALSE; /* No marker for non-JFIF colorspaces */
+  cinfo->write_Adobe_marker = FALSE; /* write no Adobe marker by default */
+
+  switch (colorspace) {
+  case JCS_UNKNOWN:
+    cinfo->num_components = cinfo->input_components;
+    if (cinfo->num_components < 1 || cinfo->num_components > MAX_COMPONENTS)
+      ERREXIT2(cinfo, JERR_COMPONENT_COUNT, cinfo->num_components,
+	       MAX_COMPONENTS);
+    for (ci = 0; ci < cinfo->num_components; ci++) {
+      SET_COMP(ci, ci, 1,1, 0, 0,0);
+    }
+    break;
+  case JCS_GRAYSCALE:
+    cinfo->write_JFIF_header = TRUE; /* Write a JFIF marker */
+    cinfo->num_components = 1;
+    /* JFIF specifies component ID 1 */
+    SET_COMP(0, 0x01, 1,1, 0, 0,0);
+    break;
+  case JCS_RGB:
+    cinfo->write_Adobe_marker = TRUE; /* write Adobe marker to flag RGB */
+    cinfo->num_components = 3;
+    SET_COMP(0, 0x52 /* 'R' */, 1,1,
+		cinfo->color_transform == JCT_SUBTRACT_GREEN ? 1 : 0,
+		cinfo->color_transform == JCT_SUBTRACT_GREEN ? 1 : 0,
+		cinfo->color_transform == JCT_SUBTRACT_GREEN ? 1 : 0);
+    SET_COMP(1, 0x47 /* 'G' */, 1,1, 0, 0,0);
+    SET_COMP(2, 0x42 /* 'B' */, 1,1,
+		cinfo->color_transform == JCT_SUBTRACT_GREEN ? 1 : 0,
+		cinfo->color_transform == JCT_SUBTRACT_GREEN ? 1 : 0,
+		cinfo->color_transform == JCT_SUBTRACT_GREEN ? 1 : 0);
+    break;
+  case JCS_YCbCr:
+    cinfo->write_JFIF_header = TRUE; /* Write a JFIF marker */
+    cinfo->num_components = 3;
+    /* JFIF specifies component IDs 1,2,3 */
+    /* We default to 2x2 subsamples of chrominance */
+    SET_COMP(0, 0x01, 2,2, 0, 0,0);
+    SET_COMP(1, 0x02, 1,1, 1, 1,1);
+    SET_COMP(2, 0x03, 1,1, 1, 1,1);
+    break;
+  case JCS_CMYK:
+    cinfo->write_Adobe_marker = TRUE; /* write Adobe marker to flag CMYK */
+    cinfo->num_components = 4;
+    SET_COMP(0, 0x43 /* 'C' */, 1,1, 0, 0,0);
+    SET_COMP(1, 0x4D /* 'M' */, 1,1, 0, 0,0);
+    SET_COMP(2, 0x59 /* 'Y' */, 1,1, 0, 0,0);
+    SET_COMP(3, 0x4B /* 'K' */, 1,1, 0, 0,0);
+    break;
+  case JCS_YCCK:
+    cinfo->write_Adobe_marker = TRUE; /* write Adobe marker to flag YCCK */
+    cinfo->num_components = 4;
+    SET_COMP(0, 0x01, 2,2, 0, 0,0);
+    SET_COMP(1, 0x02, 1,1, 1, 1,1);
+    SET_COMP(2, 0x03, 1,1, 1, 1,1);
+    SET_COMP(3, 0x04, 2,2, 0, 0,0);
+    break;
+  case JCS_BG_RGB:
+    cinfo->write_JFIF_header = TRUE; /* Write a JFIF marker */
+    cinfo->JFIF_major_version = 2;   /* Set JFIF major version = 2 */
+    cinfo->num_components = 3;
+    /* Add offset 0x20 to the normal R/G/B component IDs */
+    SET_COMP(0, 0x72 /* 'r' */, 1,1,
+		cinfo->color_transform == JCT_SUBTRACT_GREEN ? 1 : 0,
+		cinfo->color_transform == JCT_SUBTRACT_GREEN ? 1 : 0,
+		cinfo->color_transform == JCT_SUBTRACT_GREEN ? 1 : 0);
+    SET_COMP(1, 0x67 /* 'g' */, 1,1, 0, 0,0);
+    SET_COMP(2, 0x62 /* 'b' */, 1,1,
+		cinfo->color_transform == JCT_SUBTRACT_GREEN ? 1 : 0,
+		cinfo->color_transform == JCT_SUBTRACT_GREEN ? 1 : 0,
+		cinfo->color_transform == JCT_SUBTRACT_GREEN ? 1 : 0);
+    break;
+  case JCS_BG_YCC:
+    cinfo->write_JFIF_header = TRUE; /* Write a JFIF marker */
+    cinfo->JFIF_major_version = 2;   /* Set JFIF major version = 2 */
+    cinfo->num_components = 3;
+    /* Add offset 0x20 to the normal Cb/Cr component IDs */
+    /* We default to 2x2 subsamples of chrominance */
+    SET_COMP(0, 0x01, 2,2, 0, 0,0);
+    SET_COMP(1, 0x22, 1,1, 1, 1,1);
+    SET_COMP(2, 0x23, 1,1, 1, 1,1);
+    break;
+  default:
+    ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
+  }
+}
+
+
+#ifdef C_PROGRESSIVE_SUPPORTED
+
+LOCAL(jpeg_scan_info *)
+fill_a_scan (jpeg_scan_info * scanptr, int ci,
+	     int Ss, int Se, int Ah, int Al)
+/* Support routine: generate one scan for specified component */
+{
+  scanptr->comps_in_scan = 1;
+  scanptr->component_index[0] = ci;
+  scanptr->Ss = Ss;
+  scanptr->Se = Se;
+  scanptr->Ah = Ah;
+  scanptr->Al = Al;
+  scanptr++;
+  return scanptr;
+}
+
+LOCAL(jpeg_scan_info *)
+fill_scans (jpeg_scan_info * scanptr, int ncomps,
+	    int Ss, int Se, int Ah, int Al)
+/* Support routine: generate one scan for each component */
+{
+  int ci;
+
+  for (ci = 0; ci < ncomps; ci++) {
+    scanptr->comps_in_scan = 1;
+    scanptr->component_index[0] = ci;
+    scanptr->Ss = Ss;
+    scanptr->Se = Se;
+    scanptr->Ah = Ah;
+    scanptr->Al = Al;
+    scanptr++;
+  }
+  return scanptr;
+}
+
+LOCAL(jpeg_scan_info *)
+fill_dc_scans (jpeg_scan_info * scanptr, int ncomps, int Ah, int Al)
+/* Support routine: generate interleaved DC scan if possible, else N scans */
+{
+  int ci;
+
+  if (ncomps <= MAX_COMPS_IN_SCAN) {
+    /* Single interleaved DC scan */
+    scanptr->comps_in_scan = ncomps;
+    for (ci = 0; ci < ncomps; ci++)
+      scanptr->component_index[ci] = ci;
+    scanptr->Ss = scanptr->Se = 0;
+    scanptr->Ah = Ah;
+    scanptr->Al = Al;
+    scanptr++;
+  } else {
+    /* Noninterleaved DC scan for each component */
+    scanptr = fill_scans(scanptr, ncomps, 0, 0, Ah, Al);
+  }
+  return scanptr;
+}
+
+
+/*
+ * Create a recommended progressive-JPEG script.
+ * cinfo->num_components and cinfo->jpeg_color_space must be correct.
+ */
+
+GLOBAL(void)
+jpeg_simple_progression (j_compress_ptr cinfo)
+{
+  int ncomps = cinfo->num_components;
+  int nscans;
+  jpeg_scan_info * scanptr;
+
+  /* Safety check to ensure start_compress not called yet. */
+  if (cinfo->global_state != CSTATE_START)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+
+  /* Figure space needed for script.  Calculation must match code below! */
+  if (ncomps == 3 &&
+      (cinfo->jpeg_color_space == JCS_YCbCr ||
+       cinfo->jpeg_color_space == JCS_BG_YCC)) {
+    /* Custom script for YCC color images. */
+    nscans = 10;
+  } else {
+    /* All-purpose script for other color spaces. */
+    if (ncomps > MAX_COMPS_IN_SCAN)
+      nscans = 6 * ncomps;	/* 2 DC + 4 AC scans per component */
+    else
+      nscans = 2 + 4 * ncomps;	/* 2 DC scans; 4 AC scans per component */
+  }
+
+  /* Allocate space for script.
+   * We need to put it in the permanent pool in case the application performs
+   * multiple compressions without changing the settings.  To avoid a memory
+   * leak if jpeg_simple_progression is called repeatedly for the same JPEG
+   * object, we try to re-use previously allocated space, and we allocate
+   * enough space to handle YCC even if initially asked for grayscale.
+   */
+  if (cinfo->script_space == NULL || cinfo->script_space_size < nscans) {
+    cinfo->script_space_size = MAX(nscans, 10);
+    cinfo->script_space = (jpeg_scan_info *)
+      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_PERMANENT,
+			cinfo->script_space_size * SIZEOF(jpeg_scan_info));
+  }
+  scanptr = cinfo->script_space;
+  cinfo->scan_info = scanptr;
+  cinfo->num_scans = nscans;
+
+  if (ncomps == 3 &&
+      (cinfo->jpeg_color_space == JCS_YCbCr ||
+       cinfo->jpeg_color_space == JCS_BG_YCC)) {
+    /* Custom script for YCC color images. */
+    /* Initial DC scan */
+    scanptr = fill_dc_scans(scanptr, ncomps, 0, 1);
+    /* Initial AC scan: get some luma data out in a hurry */
+    scanptr = fill_a_scan(scanptr, 0, 1, 5, 0, 2);
+    /* Chroma data is too small to be worth expending many scans on */
+    scanptr = fill_a_scan(scanptr, 2, 1, 63, 0, 1);
+    scanptr = fill_a_scan(scanptr, 1, 1, 63, 0, 1);
+    /* Complete spectral selection for luma AC */
+    scanptr = fill_a_scan(scanptr, 0, 6, 63, 0, 2);
+    /* Refine next bit of luma AC */
+    scanptr = fill_a_scan(scanptr, 0, 1, 63, 2, 1);
+    /* Finish DC successive approximation */
+    scanptr = fill_dc_scans(scanptr, ncomps, 1, 0);
+    /* Finish AC successive approximation */
+    scanptr = fill_a_scan(scanptr, 2, 1, 63, 1, 0);
+    scanptr = fill_a_scan(scanptr, 1, 1, 63, 1, 0);
+    /* Luma bottom bit comes last since it's usually largest scan */
+    scanptr = fill_a_scan(scanptr, 0, 1, 63, 1, 0);
+  } else {
+    /* All-purpose script for other color spaces. */
+    /* Successive approximation first pass */
+    scanptr = fill_dc_scans(scanptr, ncomps, 0, 1);
+    scanptr = fill_scans(scanptr, ncomps, 1, 5, 0, 2);
+    scanptr = fill_scans(scanptr, ncomps, 6, 63, 0, 2);
+    /* Successive approximation second pass */
+    scanptr = fill_scans(scanptr, ncomps, 1, 63, 2, 1);
+    /* Successive approximation final pass */
+    scanptr = fill_dc_scans(scanptr, ncomps, 1, 0);
+    scanptr = fill_scans(scanptr, ncomps, 1, 63, 1, 0);
+  }
+}
+
+#endif /* C_PROGRESSIVE_SUPPORTED */
@@ -0,0 +1,358 @@
+/*
+ * jcprepct.c
+ *
+ * Copyright (C) 1994-1996, Thomas G. Lane.
+ * Modified 2003-2020 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains the compression preprocessing controller.
+ * This controller manages the color conversion, downsampling,
+ * and edge expansion steps.
+ *
+ * Most of the complexity here is associated with buffering input rows
+ * as required by the downsampler.  See the comments at the head of
+ * jcsample.c for the downsampler's needs.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/* At present, jcsample.c can request context rows only for smoothing.
+ * In the future, we might also need context rows for CCIR601 sampling
+ * or other more-complex downsampling procedures.  The code to support
+ * context rows should be compiled only if needed.
+ */
+#ifdef INPUT_SMOOTHING_SUPPORTED
+#define CONTEXT_ROWS_SUPPORTED
+#endif
+
+
+/*
+ * For the simple (no-context-row) case, we just need to buffer one
+ * row group's worth of pixels for the downsampling step.  At the bottom of
+ * the image, we pad to a full row group by replicating the last pixel row.
+ * The downsampler's last output row is then replicated if needed to pad
+ * out to a full iMCU row.
+ *
+ * When providing context rows, we must buffer three row groups' worth of
+ * pixels.  Three row groups are physically allocated, but the row pointer
+ * arrays are made five row groups high, with the extra pointers above and
+ * below "wrapping around" to point to the last and first real row groups.
+ * This allows the downsampler to access the proper context rows.
+ * At the top and bottom of the image, we create dummy context rows by
+ * copying the first or last real pixel row.  This copying could be avoided
+ * by pointer hacking as is done in jdmainct.c, but it doesn't seem worth the
+ * trouble on the compression side.
+ */
+
+
+/* Private buffer controller object */
+
+typedef struct {
+  struct jpeg_c_prep_controller pub; /* public fields */
+
+  /* Downsampling input buffer.  This buffer holds color-converted data
+   * until we have enough to do a downsample step.
+   */
+  JSAMPARRAY color_buf[MAX_COMPONENTS];
+
+  JDIMENSION rows_to_go;	/* counts rows remaining in source image */
+  int next_buf_row;		/* index of next row to store in color_buf */
+
+#ifdef CONTEXT_ROWS_SUPPORTED	/* only needed for context case */
+  int this_row_group;		/* starting row index of group to process */
+  int next_buf_stop;		/* downsample when we reach this index */
+#endif
+} my_prep_controller;
+
+typedef my_prep_controller * my_prep_ptr;
+
+
+/*
+ * Initialize for a processing pass.
+ */
+
+METHODDEF(void)
+start_pass_prep (j_compress_ptr cinfo, J_BUF_MODE pass_mode)
+{
+  my_prep_ptr prep = (my_prep_ptr) cinfo->prep;
+
+  if (pass_mode != JBUF_PASS_THRU)
+    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+
+  /* Initialize total-height counter for detecting bottom of image */
+  prep->rows_to_go = cinfo->image_height;
+  /* Mark the conversion buffer empty */
+  prep->next_buf_row = 0;
+#ifdef CONTEXT_ROWS_SUPPORTED
+  /* Preset additional state variables for context mode.
+   * These aren't used in non-context mode, so we needn't test which mode.
+   */
+  prep->this_row_group = 0;
+  /* Set next_buf_stop to stop after two row groups have been read in. */
+  prep->next_buf_stop = 2 * cinfo->max_v_samp_factor;
+#endif
+}
+
+
+/*
+ * Expand an image vertically from height input_rows to height output_rows,
+ * by duplicating the bottom row.
+ */
+
+LOCAL(void)
+expand_bottom_edge (JSAMPARRAY image_data, JDIMENSION num_cols,
+		    int input_rows, int output_rows)
+{
+  register int row;
+
+  for (row = input_rows; row < output_rows; row++) {
+    jcopy_sample_rows(image_data + input_rows - 1,
+		      image_data + row,
+		      1, num_cols);
+  }
+}
+
+
+/*
+ * Process some data in the simple no-context case.
+ *
+ * Preprocessor output data is counted in "row groups".  A row group
+ * is defined to be v_samp_factor sample rows of each component.
+ * Downsampling will produce this much data from each max_v_samp_factor
+ * input rows.
+ */
+
+METHODDEF(void)
+pre_process_data (j_compress_ptr cinfo,
+		  JSAMPARRAY input_buf, JDIMENSION *in_row_ctr,
+		  JDIMENSION in_rows_avail,
+		  JSAMPIMAGE output_buf, JDIMENSION *out_row_group_ctr,
+		  JDIMENSION out_row_groups_avail)
+{
+  my_prep_ptr prep = (my_prep_ptr) cinfo->prep;
+  int numrows, ci;
+  JDIMENSION inrows;
+  jpeg_component_info * compptr;
+
+  while (*in_row_ctr < in_rows_avail &&
+	 *out_row_group_ctr < out_row_groups_avail) {
+    /* Do color conversion to fill the conversion buffer. */
+    inrows = in_rows_avail - *in_row_ctr;
+    numrows = cinfo->max_v_samp_factor - prep->next_buf_row;
+    numrows = (int) MIN((JDIMENSION) numrows, inrows);
+    (*cinfo->cconvert->color_convert) (cinfo, input_buf + *in_row_ctr,
+				       prep->color_buf,
+				       (JDIMENSION) prep->next_buf_row,
+				       numrows);
+    *in_row_ctr += numrows;
+    prep->next_buf_row += numrows;
+    prep->rows_to_go -= numrows;
+    /* If at bottom of image, pad to fill the conversion buffer. */
+    if (prep->rows_to_go == 0 &&
+	prep->next_buf_row < cinfo->max_v_samp_factor) {
+      for (ci = 0; ci < cinfo->num_components; ci++) {
+	expand_bottom_edge(prep->color_buf[ci], cinfo->image_width,
+			   prep->next_buf_row, cinfo->max_v_samp_factor);
+      }
+      prep->next_buf_row = cinfo->max_v_samp_factor;
+    }
+    /* If we've filled the conversion buffer, empty it. */
+    if (prep->next_buf_row == cinfo->max_v_samp_factor) {
+      (*cinfo->downsample->downsample) (cinfo,
+					prep->color_buf, (JDIMENSION) 0,
+					output_buf, *out_row_group_ctr);
+      prep->next_buf_row = 0;
+      (*out_row_group_ctr)++;
+    }
+    /* If at bottom of image, pad the output to a full iMCU height.
+     * Note we assume the caller is providing a one-iMCU-height output buffer!
+     */
+    if (prep->rows_to_go == 0 &&
+	*out_row_group_ctr < out_row_groups_avail) {
+      for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+	   ci++, compptr++) {
+	numrows = (compptr->v_samp_factor * compptr->DCT_v_scaled_size) /
+		  cinfo->min_DCT_v_scaled_size;
+	expand_bottom_edge(output_buf[ci],
+			   compptr->width_in_blocks * compptr->DCT_h_scaled_size,
+			   (int) (*out_row_group_ctr * numrows),
+			   (int) (out_row_groups_avail * numrows));
+      }
+      *out_row_group_ctr = out_row_groups_avail;
+      break;			/* can exit outer loop without test */
+    }
+  }
+}
+
+
+#ifdef CONTEXT_ROWS_SUPPORTED
+
+/*
+ * Process some data in the context case.
+ */
+
+METHODDEF(void)
+pre_process_context (j_compress_ptr cinfo,
+		     JSAMPARRAY input_buf, JDIMENSION *in_row_ctr,
+		     JDIMENSION in_rows_avail,
+		     JSAMPIMAGE output_buf, JDIMENSION *out_row_group_ctr,
+		     JDIMENSION out_row_groups_avail)
+{
+  my_prep_ptr prep = (my_prep_ptr) cinfo->prep;
+  int numrows, ci;
+  int buf_height = cinfo->max_v_samp_factor * 3;
+  JDIMENSION inrows;
+
+  while (*out_row_group_ctr < out_row_groups_avail) {
+    if (*in_row_ctr < in_rows_avail) {
+      /* Do color conversion to fill the conversion buffer. */
+      inrows = in_rows_avail - *in_row_ctr;
+      numrows = prep->next_buf_stop - prep->next_buf_row;
+      numrows = (int) MIN((JDIMENSION) numrows, inrows);
+      (*cinfo->cconvert->color_convert) (cinfo, input_buf + *in_row_ctr,
+					 prep->color_buf,
+					 (JDIMENSION) prep->next_buf_row,
+					 numrows);
+      /* Pad at top of image, if first time through */
+      if (prep->rows_to_go == cinfo->image_height) {
+	for (ci = 0; ci < cinfo->num_components; ci++) {
+	  int row;
+	  for (row = 1; row <= cinfo->max_v_samp_factor; row++) {
+	    jcopy_sample_rows(prep->color_buf[ci],
+			      prep->color_buf[ci] - row,
+			      1, cinfo->image_width);
+	  }
+	}
+      }
+      *in_row_ctr += numrows;
+      prep->next_buf_row += numrows;
+      prep->rows_to_go -= numrows;
+    } else {
+      /* Return for more data, unless we are at the bottom of the image. */
+      if (prep->rows_to_go != 0)
+	break;
+      /* When at bottom of image, pad to fill the conversion buffer. */
+      if (prep->next_buf_row < prep->next_buf_stop) {
+	for (ci = 0; ci < cinfo->num_components; ci++) {
+	  expand_bottom_edge(prep->color_buf[ci], cinfo->image_width,
+			     prep->next_buf_row, prep->next_buf_stop);
+	}
+	prep->next_buf_row = prep->next_buf_stop;
+      }
+    }
+    /* If we've gotten enough data, downsample a row group. */
+    if (prep->next_buf_row == prep->next_buf_stop) {
+      (*cinfo->downsample->downsample) (cinfo,
+					prep->color_buf,
+					(JDIMENSION) prep->this_row_group,
+					output_buf, *out_row_group_ctr);
+      (*out_row_group_ctr)++;
+      /* Advance pointers with wraparound as necessary. */
+      prep->this_row_group += cinfo->max_v_samp_factor;
+      if (prep->this_row_group >= buf_height)
+	prep->this_row_group = 0;
+      if (prep->next_buf_row >= buf_height)
+	prep->next_buf_row = 0;
+      prep->next_buf_stop = prep->next_buf_row + cinfo->max_v_samp_factor;
+    }
+  }
+}
+
+
+/*
+ * Create the wrapped-around downsampling input buffer needed for context mode.
+ */
+
+LOCAL(void)
+create_context_buffer (j_compress_ptr cinfo)
+{
+  my_prep_ptr prep = (my_prep_ptr) cinfo->prep;
+  int rgroup_height = cinfo->max_v_samp_factor;
+  int ci, i;
+  jpeg_component_info * compptr;
+  JSAMPARRAY true_buffer, fake_buffer;
+
+  /* Grab enough space for fake row pointers for all the components;
+   * we need five row groups' worth of pointers for each component.
+   */
+  fake_buffer = (JSAMPARRAY) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE,
+     (cinfo->num_components * 5 * rgroup_height) * SIZEOF(JSAMPROW));
+
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    /* Allocate the actual buffer space (3 row groups) for this component.
+     * We make the buffer wide enough to allow the downsampler to edge-expand
+     * horizontally within the buffer, if it so chooses.
+     */
+    true_buffer = (*cinfo->mem->alloc_sarray)
+      ((j_common_ptr) cinfo, JPOOL_IMAGE,
+       (JDIMENSION) (((long) compptr->width_in_blocks *
+		      cinfo->min_DCT_h_scaled_size *
+		      cinfo->max_h_samp_factor) / compptr->h_samp_factor),
+       (JDIMENSION) (3 * rgroup_height));
+    /* Copy true buffer row pointers into the middle of the fake row array */
+    MEMCOPY(fake_buffer + rgroup_height, true_buffer,
+	    3 * rgroup_height * SIZEOF(JSAMPROW));
+    /* Fill in the above and below wraparound pointers */
+    for (i = 0; i < rgroup_height; i++) {
+      fake_buffer[i] = true_buffer[2 * rgroup_height + i];
+      fake_buffer[4 * rgroup_height + i] = true_buffer[i];
+    }
+    prep->color_buf[ci] = fake_buffer + rgroup_height;
+    fake_buffer += 5 * rgroup_height; /* point to space for next component */
+  }
+}
+
+#endif /* CONTEXT_ROWS_SUPPORTED */
+
+
+/*
+ * Initialize preprocessing controller.
+ */
+
+GLOBAL(void)
+jinit_c_prep_controller (j_compress_ptr cinfo, boolean need_full_buffer)
+{
+  my_prep_ptr prep;
+  int ci;
+  jpeg_component_info * compptr;
+
+  if (need_full_buffer)		/* safety check */
+    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+
+  prep = (my_prep_ptr) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, SIZEOF(my_prep_controller));
+  cinfo->prep = &prep->pub;
+  prep->pub.start_pass = start_pass_prep;
+
+  /* Allocate the color conversion buffer.
+   * We make the buffer wide enough to allow the downsampler to edge-expand
+   * horizontally within the buffer, if it so chooses.
+   */
+  if (cinfo->downsample->need_context_rows) {
+    /* Set up to provide context rows */
+#ifdef CONTEXT_ROWS_SUPPORTED
+    prep->pub.pre_process_data = pre_process_context;
+    create_context_buffer(cinfo);
+#else
+    ERREXIT(cinfo, JERR_NOT_COMPILED);
+#endif
+  } else {
+    /* No context, just make it tall enough for one row group */
+    prep->pub.pre_process_data = pre_process_data;
+    for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+	 ci++, compptr++) {
+      prep->color_buf[ci] = (*cinfo->mem->alloc_sarray)
+	((j_common_ptr) cinfo, JPOOL_IMAGE,
+	 (JDIMENSION) (((long) compptr->width_in_blocks *
+			cinfo->min_DCT_h_scaled_size *
+			cinfo->max_h_samp_factor) / compptr->h_samp_factor),
+	 (JDIMENSION) cinfo->max_v_samp_factor);
+    }
+  }
+}
@@ -0,0 +1,545 @@
+/*
+ * jcsample.c
+ *
+ * Copyright (C) 1991-1996, Thomas G. Lane.
+ * Modified 2003-2020 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains downsampling routines.
+ *
+ * Downsampling input data is counted in "row groups".  A row group
+ * is defined to be max_v_samp_factor pixel rows of each component,
+ * from which the downsampler produces v_samp_factor sample rows.
+ * A single row group is processed in each call to the downsampler module.
+ *
+ * The downsampler is responsible for edge-expansion of its output data
+ * to fill an integral number of DCT blocks horizontally.  The source buffer
+ * may be modified if it is helpful for this purpose (the source buffer is
+ * allocated wide enough to correspond to the desired output width).
+ * The caller (the prep controller) is responsible for vertical padding.
+ *
+ * The downsampler may request "context rows" by setting need_context_rows
+ * during startup.  In this case, the input arrays will contain at least
+ * one row group's worth of pixels above and below the passed-in data;
+ * the caller will create dummy rows at image top and bottom by replicating
+ * the first or last real pixel row.
+ *
+ * An excellent reference for image resampling is
+ *   Digital Image Warping, George Wolberg, 1990.
+ *   Pub. by IEEE Computer Society Press, Los Alamitos, CA. ISBN 0-8186-8944-7.
+ *
+ * The downsampling algorithm used here is a simple average of the source
+ * pixels covered by the output pixel.  The hi-falutin sampling literature
+ * refers to this as a "box filter".  In general the characteristics of a box
+ * filter are not very good, but for the specific cases we normally use (1:1
+ * and 2:1 ratios) the box is equivalent to a "triangle filter" which is not
+ * nearly so bad.  If you intend to use other sampling ratios, you'd be well
+ * advised to improve this code.
+ *
+ * A simple input-smoothing capability is provided.  This is mainly intended
+ * for cleaning up color-dithered GIF input files (if you find it inadequate,
+ * we suggest using an external filtering program such as pnmconvol).  When
+ * enabled, each input pixel P is replaced by a weighted sum of itself and its
+ * eight neighbors.  P's weight is 1-8*SF and each neighbor's weight is SF,
+ * where SF = (smoothing_factor / 1024).
+ * Currently, smoothing is only supported for 2h2v sampling factors.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/* Pointer to routine to downsample a single component */
+typedef JMETHOD(void, downsample1_ptr,
+		(j_compress_ptr cinfo, jpeg_component_info * compptr,
+		 JSAMPARRAY input_data, JSAMPARRAY output_data));
+
+/* Private subobject */
+
+typedef struct {
+  struct jpeg_downsampler pub;	/* public fields */
+
+  /* Downsampling method pointers, one per component */
+  downsample1_ptr methods[MAX_COMPONENTS];
+
+  /* Height of an output row group for each component. */
+  int rowgroup_height[MAX_COMPONENTS];
+
+  /* These arrays save pixel expansion factors so that int_downsample need not
+   * recompute them each time.  They are unused for other downsampling methods.
+   */
+  UINT8 h_expand[MAX_COMPONENTS];
+  UINT8 v_expand[MAX_COMPONENTS];
+} my_downsampler;
+
+typedef my_downsampler * my_downsample_ptr;
+
+
+/*
+ * Initialize for a downsampling pass.
+ */
+
+METHODDEF(void)
+start_pass_downsample (j_compress_ptr cinfo)
+{
+  /* no work for now */
+}
+
+
+/*
+ * Expand a component horizontally from width input_cols to width output_cols,
+ * by duplicating the rightmost samples.
+ */
+
+LOCAL(void)
+expand_right_edge (JSAMPARRAY image_data, int num_rows,
+		   JDIMENSION input_cols, JDIMENSION output_cols)
+{
+  register JSAMPROW ptr;
+  register JSAMPLE pixval;
+  register int count;
+  int row;
+  int numcols = (int) (output_cols - input_cols);
+
+  if (numcols > 0) {
+    for (row = 0; row < num_rows; row++) {
+      ptr = image_data[row] + input_cols;
+      pixval = ptr[-1];		/* don't need GETJSAMPLE() here */
+      for (count = numcols; count > 0; count--)
+	*ptr++ = pixval;
+    }
+  }
+}
+
+
+/*
+ * Do downsampling for a whole row group (all components).
+ *
+ * In this version we simply downsample each component independently.
+ */
+
+METHODDEF(void)
+sep_downsample (j_compress_ptr cinfo,
+		JSAMPIMAGE input_buf, JDIMENSION in_row_index,
+		JSAMPIMAGE output_buf, JDIMENSION out_row_group_index)
+{
+  my_downsample_ptr downsample = (my_downsample_ptr) cinfo->downsample;
+  int ci;
+  jpeg_component_info * compptr;
+  JSAMPARRAY in_ptr, out_ptr;
+
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    in_ptr = input_buf[ci] + in_row_index;
+    out_ptr = output_buf[ci] +
+	      (out_row_group_index * downsample->rowgroup_height[ci]);
+    (*downsample->methods[ci]) (cinfo, compptr, in_ptr, out_ptr);
+  }
+}
+
+
+/*
+ * Downsample pixel values of a single component.
+ * One row group is processed per call.
+ * This version handles arbitrary integral sampling ratios, without smoothing.
+ * Note that this version is not actually used for customary sampling ratios.
+ */
+
+METHODDEF(void)
+int_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
+		JSAMPARRAY input_data, JSAMPARRAY output_data)
+{
+  my_downsample_ptr downsample = (my_downsample_ptr) cinfo->downsample;
+  int inrow, outrow, h_expand, v_expand, numpix, numpix2, h, v;
+  JDIMENSION outcol, outcol_h;	/* outcol_h == outcol*h_expand */
+  JDIMENSION output_cols = compptr->width_in_blocks * compptr->DCT_h_scaled_size;
+  JSAMPROW inptr, outptr;
+  INT32 outvalue;
+
+  h_expand = downsample->h_expand[compptr->component_index];
+  v_expand = downsample->v_expand[compptr->component_index];
+  numpix = h_expand * v_expand;
+  numpix2 = numpix/2;
+
+  /* Expand input data enough to let all the output samples be generated
+   * by the standard loop.  Special-casing padded output would be more
+   * efficient.
+   */
+  expand_right_edge(input_data, cinfo->max_v_samp_factor,
+		    cinfo->image_width, output_cols * h_expand);
+
+  inrow = outrow = 0;
+  while (inrow < cinfo->max_v_samp_factor) {
+    outptr = output_data[outrow];
+    for (outcol = 0, outcol_h = 0; outcol < output_cols;
+	 outcol++, outcol_h += h_expand) {
+      outvalue = 0;
+      for (v = 0; v < v_expand; v++) {
+	inptr = input_data[inrow+v] + outcol_h;
+	for (h = 0; h < h_expand; h++) {
+	  outvalue += (INT32) GETJSAMPLE(*inptr++);
+	}
+      }
+      *outptr++ = (JSAMPLE) ((outvalue + numpix2) / numpix);
+    }
+    inrow += v_expand;
+    outrow++;
+  }
+}
+
+
+/*
+ * Downsample pixel values of a single component.
+ * This version handles the special case of a full-size component,
+ * without smoothing.
+ */
+
+METHODDEF(void)
+fullsize_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
+		     JSAMPARRAY input_data, JSAMPARRAY output_data)
+{
+  /* Copy the data */
+  jcopy_sample_rows(input_data, output_data,
+		    cinfo->max_v_samp_factor, cinfo->image_width);
+  /* Edge-expand */
+  expand_right_edge(output_data, cinfo->max_v_samp_factor, cinfo->image_width,
+		    compptr->width_in_blocks * compptr->DCT_h_scaled_size);
+}
+
+
+/*
+ * Downsample pixel values of a single component.
+ * This version handles the common case of 2:1 horizontal and 1:1 vertical,
+ * without smoothing.
+ *
+ * A note about the "bias" calculations: when rounding fractional values to
+ * integer, we do not want to always round 0.5 up to the next integer.
+ * If we did that, we'd introduce a noticeable bias towards larger values.
+ * Instead, this code is arranged so that 0.5 will be rounded up or down at
+ * alternate pixel locations (a simple ordered dither pattern).
+ */
+
+METHODDEF(void)
+h2v1_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
+		 JSAMPARRAY input_data, JSAMPARRAY output_data)
+{
+  int inrow;
+  JDIMENSION outcol;
+  JDIMENSION output_cols = compptr->width_in_blocks * compptr->DCT_h_scaled_size;
+  register JSAMPROW inptr, outptr;
+  register int bias;
+
+  /* Expand input data enough to let all the output samples be generated
+   * by the standard loop.  Special-casing padded output would be more
+   * efficient.
+   */
+  expand_right_edge(input_data, cinfo->max_v_samp_factor,
+		    cinfo->image_width, output_cols * 2);
+
+  for (inrow = 0; inrow < cinfo->max_v_samp_factor; inrow++) {
+    outptr = output_data[inrow];
+    inptr = input_data[inrow];
+    bias = 0;			/* bias = 0,1,0,1,... for successive samples */
+    for (outcol = 0; outcol < output_cols; outcol++) {
+      *outptr++ = (JSAMPLE) ((GETJSAMPLE(*inptr) + GETJSAMPLE(inptr[1])
+			      + bias) >> 1);
+      bias ^= 1;		/* 0=>1, 1=>0 */
+      inptr += 2;
+    }
+  }
+}
+
+
+/*
+ * Downsample pixel values of a single component.
+ * This version handles the standard case of 2:1 horizontal and 2:1 vertical,
+ * without smoothing.
+ */
+
+METHODDEF(void)
+h2v2_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
+		 JSAMPARRAY input_data, JSAMPARRAY output_data)
+{
+  int inrow, outrow;
+  JDIMENSION outcol;
+  JDIMENSION output_cols = compptr->width_in_blocks * compptr->DCT_h_scaled_size;
+  register JSAMPROW inptr0, inptr1, outptr;
+  register int bias;
+
+  /* Expand input data enough to let all the output samples be generated
+   * by the standard loop.  Special-casing padded output would be more
+   * efficient.
+   */
+  expand_right_edge(input_data, cinfo->max_v_samp_factor,
+		    cinfo->image_width, output_cols * 2);
+
+  inrow = outrow = 0;
+  while (inrow < cinfo->max_v_samp_factor) {
+    outptr = output_data[outrow];
+    inptr0 = input_data[inrow];
+    inptr1 = input_data[inrow+1];
+    bias = 1;			/* bias = 1,2,1,2,... for successive samples */
+    for (outcol = 0; outcol < output_cols; outcol++) {
+      *outptr++ = (JSAMPLE) ((GETJSAMPLE(*inptr0) + GETJSAMPLE(inptr0[1]) +
+			      GETJSAMPLE(*inptr1) + GETJSAMPLE(inptr1[1])
+			      + bias) >> 2);
+      bias ^= 3;		/* 1=>2, 2=>1 */
+      inptr0 += 2; inptr1 += 2;
+    }
+    inrow += 2;
+    outrow++;
+  }
+}
+
+
+#ifdef INPUT_SMOOTHING_SUPPORTED
+
+/*
+ * Downsample pixel values of a single component.
+ * This version handles the standard case of 2:1 horizontal and 2:1 vertical,
+ * with smoothing.  One row of context is required.
+ */
+
+METHODDEF(void)
+h2v2_smooth_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
+			JSAMPARRAY input_data, JSAMPARRAY output_data)
+{
+  int inrow, outrow;
+  JDIMENSION colctr;
+  JDIMENSION output_cols = compptr->width_in_blocks * compptr->DCT_h_scaled_size;
+  register JSAMPROW inptr0, inptr1, above_ptr, below_ptr, outptr;
+  INT32 membersum, neighsum, memberscale, neighscale;
+
+  /* Expand input data enough to let all the output samples be generated
+   * by the standard loop.  Special-casing padded output would be more
+   * efficient.
+   */
+  expand_right_edge(input_data - 1, cinfo->max_v_samp_factor + 2,
+		    cinfo->image_width, output_cols * 2);
+
+  /* We don't bother to form the individual "smoothed" input pixel values;
+   * we can directly compute the output which is the average of the four
+   * smoothed values.  Each of the four member pixels contributes a fraction
+   * (1-8*SF) to its own smoothed image and a fraction SF to each of the three
+   * other smoothed pixels, therefore a total fraction (1-5*SF)/4 to the final
+   * output.  The four corner-adjacent neighbor pixels contribute a fraction
+   * SF to just one smoothed pixel, or SF/4 to the final output; while the
+   * eight edge-adjacent neighbors contribute SF to each of two smoothed
+   * pixels, or SF/2 overall.  In order to use integer arithmetic, these
+   * factors are scaled by 2^16 = 65536.
+   * Also recall that SF = smoothing_factor / 1024.
+   */
+
+  memberscale = 16384 - cinfo->smoothing_factor * 80; /* scaled (1-5*SF)/4 */
+  neighscale = cinfo->smoothing_factor * 16; /* scaled SF/4 */
+
+  inrow = outrow = 0;
+  while (inrow < cinfo->max_v_samp_factor) {
+    outptr = output_data[outrow];
+    inptr0 = input_data[inrow];
+    inptr1 = input_data[inrow+1];
+    above_ptr = input_data[inrow-1];
+    below_ptr = input_data[inrow+2];
+
+    /* Special case for first column: pretend column -1 is same as column 0 */
+    membersum = GETJSAMPLE(*inptr0) + GETJSAMPLE(inptr0[1]) +
+		GETJSAMPLE(*inptr1) + GETJSAMPLE(inptr1[1]);
+    neighsum = GETJSAMPLE(*above_ptr) + GETJSAMPLE(above_ptr[1]) +
+	       GETJSAMPLE(*below_ptr) + GETJSAMPLE(below_ptr[1]) +
+	       GETJSAMPLE(*inptr0) + GETJSAMPLE(inptr0[2]) +
+	       GETJSAMPLE(*inptr1) + GETJSAMPLE(inptr1[2]);
+    neighsum += neighsum;
+    neighsum += GETJSAMPLE(*above_ptr) + GETJSAMPLE(above_ptr[2]) +
+		GETJSAMPLE(*below_ptr) + GETJSAMPLE(below_ptr[2]);
+    membersum = membersum * memberscale + neighsum * neighscale;
+    *outptr++ = (JSAMPLE) ((membersum + 32768) >> 16);
+    inptr0 += 2; inptr1 += 2; above_ptr += 2; below_ptr += 2;
+
+    for (colctr = output_cols - 2; colctr > 0; colctr--) {
+      /* sum of pixels directly mapped to this output element */
+      membersum = GETJSAMPLE(*inptr0) + GETJSAMPLE(inptr0[1]) +
+		  GETJSAMPLE(*inptr1) + GETJSAMPLE(inptr1[1]);
+      /* sum of edge-neighbor pixels */
+      neighsum = GETJSAMPLE(*above_ptr) + GETJSAMPLE(above_ptr[1]) +
+		 GETJSAMPLE(*below_ptr) + GETJSAMPLE(below_ptr[1]) +
+		 GETJSAMPLE(inptr0[-1]) + GETJSAMPLE(inptr0[2]) +
+		 GETJSAMPLE(inptr1[-1]) + GETJSAMPLE(inptr1[2]);
+      /* The edge-neighbors count twice as much as corner-neighbors */
+      neighsum += neighsum;
+      /* Add in the corner-neighbors */
+      neighsum += GETJSAMPLE(above_ptr[-1]) + GETJSAMPLE(above_ptr[2]) +
+		  GETJSAMPLE(below_ptr[-1]) + GETJSAMPLE(below_ptr[2]);
+      /* form final output scaled up by 2^16 */
+      membersum = membersum * memberscale + neighsum * neighscale;
+      /* round, descale and output it */
+      *outptr++ = (JSAMPLE) ((membersum + 32768) >> 16);
+      inptr0 += 2; inptr1 += 2; above_ptr += 2; below_ptr += 2;
+    }
+
+    /* Special case for last column */
+    membersum = GETJSAMPLE(*inptr0) + GETJSAMPLE(inptr0[1]) +
+		GETJSAMPLE(*inptr1) + GETJSAMPLE(inptr1[1]);
+    neighsum = GETJSAMPLE(*above_ptr) + GETJSAMPLE(above_ptr[1]) +
+	       GETJSAMPLE(*below_ptr) + GETJSAMPLE(below_ptr[1]) +
+	       GETJSAMPLE(inptr0[-1]) + GETJSAMPLE(inptr0[1]) +
+	       GETJSAMPLE(inptr1[-1]) + GETJSAMPLE(inptr1[1]);
+    neighsum += neighsum;
+    neighsum += GETJSAMPLE(above_ptr[-1]) + GETJSAMPLE(above_ptr[1]) +
+		GETJSAMPLE(below_ptr[-1]) + GETJSAMPLE(below_ptr[1]);
+    membersum = membersum * memberscale + neighsum * neighscale;
+    *outptr = (JSAMPLE) ((membersum + 32768) >> 16);
+
+    inrow += 2;
+    outrow++;
+  }
+}
+
+
+/*
+ * Downsample pixel values of a single component.
+ * This version handles the special case of a full-size component,
+ * with smoothing.  One row of context is required.
+ */
+
+METHODDEF(void)
+fullsize_smooth_downsample (j_compress_ptr cinfo, jpeg_component_info *compptr,
+			    JSAMPARRAY input_data, JSAMPARRAY output_data)
+{
+  int inrow;
+  JDIMENSION colctr;
+  JDIMENSION output_cols = compptr->width_in_blocks * compptr->DCT_h_scaled_size;
+  register JSAMPROW inptr, above_ptr, below_ptr, outptr;
+  INT32 membersum, neighsum, memberscale, neighscale;
+  int colsum, lastcolsum, nextcolsum;
+
+  /* Expand input data enough to let all the output samples be generated
+   * by the standard loop.  Special-casing padded output would be more
+   * efficient.
+   */
+  expand_right_edge(input_data - 1, cinfo->max_v_samp_factor + 2,
+		    cinfo->image_width, output_cols);
+
+  /* Each of the eight neighbor pixels contributes a fraction SF to the
+   * smoothed pixel, while the main pixel contributes (1-8*SF).  In order
+   * to use integer arithmetic, these factors are multiplied by 2^16 = 65536.
+   * Also recall that SF = smoothing_factor / 1024.
+   */
+
+  memberscale = 65536L - cinfo->smoothing_factor * 512L; /* scaled 1-8*SF */
+  neighscale = cinfo->smoothing_factor * 64; /* scaled SF */
+
+  for (inrow = 0; inrow < cinfo->max_v_samp_factor; inrow++) {
+    outptr = output_data[inrow];
+    inptr = input_data[inrow];
+    above_ptr = input_data[inrow-1];
+    below_ptr = input_data[inrow+1];
+
+    /* Special case for first column */
+    colsum = GETJSAMPLE(*above_ptr++) + GETJSAMPLE(*below_ptr++) +
+	     GETJSAMPLE(*inptr);
+    membersum = GETJSAMPLE(*inptr++);
+    nextcolsum = GETJSAMPLE(*above_ptr) + GETJSAMPLE(*below_ptr) +
+		 GETJSAMPLE(*inptr);
+    neighsum = colsum + (colsum - membersum) + nextcolsum;
+    membersum = membersum * memberscale + neighsum * neighscale;
+    *outptr++ = (JSAMPLE) ((membersum + 32768) >> 16);
+    lastcolsum = colsum; colsum = nextcolsum;
+
+    for (colctr = output_cols - 2; colctr > 0; colctr--) {
+      membersum = GETJSAMPLE(*inptr++);
+      above_ptr++; below_ptr++;
+      nextcolsum = GETJSAMPLE(*above_ptr) + GETJSAMPLE(*below_ptr) +
+		   GETJSAMPLE(*inptr);
+      neighsum = lastcolsum + (colsum - membersum) + nextcolsum;
+      membersum = membersum * memberscale + neighsum * neighscale;
+      *outptr++ = (JSAMPLE) ((membersum + 32768) >> 16);
+      lastcolsum = colsum; colsum = nextcolsum;
+    }
+
+    /* Special case for last column */
+    membersum = GETJSAMPLE(*inptr);
+    neighsum = lastcolsum + (colsum - membersum) + colsum;
+    membersum = membersum * memberscale + neighsum * neighscale;
+    *outptr = (JSAMPLE) ((membersum + 32768) >> 16);
+
+  }
+}
+
+#endif /* INPUT_SMOOTHING_SUPPORTED */
+
+
+/*
+ * Module initialization routine for downsampling.
+ * Note that we must select a routine for each component.
+ */
+
+GLOBAL(void)
+jinit_downsampler (j_compress_ptr cinfo)
+{
+  my_downsample_ptr downsample;
+  int ci;
+  jpeg_component_info * compptr;
+  boolean smoothok = TRUE;
+  int h_in_group, v_in_group, h_out_group, v_out_group;
+
+  downsample = (my_downsample_ptr) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, SIZEOF(my_downsampler));
+  cinfo->downsample = &downsample->pub;
+  downsample->pub.start_pass = start_pass_downsample;
+  downsample->pub.downsample = sep_downsample;
+  downsample->pub.need_context_rows = FALSE;
+
+  if (cinfo->CCIR601_sampling)
+    ERREXIT(cinfo, JERR_CCIR601_NOTIMPL);
+
+  /* Verify we can handle the sampling factors, and set up method pointers */
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    /* Compute size of an "output group" for DCT scaling.  This many samples
+     * are to be converted from max_h_samp_factor * max_v_samp_factor pixels.
+     */
+    h_out_group = (compptr->h_samp_factor * compptr->DCT_h_scaled_size) /
+		  cinfo->min_DCT_h_scaled_size;
+    v_out_group = (compptr->v_samp_factor * compptr->DCT_v_scaled_size) /
+		  cinfo->min_DCT_v_scaled_size;
+    h_in_group = cinfo->max_h_samp_factor;
+    v_in_group = cinfo->max_v_samp_factor;
+    downsample->rowgroup_height[ci] = v_out_group; /* save for use later */
+    if (h_in_group == h_out_group && v_in_group == v_out_group) {
+#ifdef INPUT_SMOOTHING_SUPPORTED
+      if (cinfo->smoothing_factor) {
+	downsample->methods[ci] = fullsize_smooth_downsample;
+	downsample->pub.need_context_rows = TRUE;
+      } else
+#endif
+	downsample->methods[ci] = fullsize_downsample;
+    } else if (h_in_group == h_out_group * 2 &&
+	       v_in_group == v_out_group) {
+      smoothok = FALSE;
+      downsample->methods[ci] = h2v1_downsample;
+    } else if (h_in_group == h_out_group * 2 &&
+	       v_in_group == v_out_group * 2) {
+#ifdef INPUT_SMOOTHING_SUPPORTED
+      if (cinfo->smoothing_factor) {
+	downsample->methods[ci] = h2v2_smooth_downsample;
+	downsample->pub.need_context_rows = TRUE;
+      } else
+#endif
+	downsample->methods[ci] = h2v2_downsample;
+    } else if ((h_in_group % h_out_group) == 0 &&
+	       (v_in_group % v_out_group) == 0) {
+      smoothok = FALSE;
+      downsample->methods[ci] = int_downsample;
+      downsample->h_expand[ci] = (UINT8) (h_in_group / h_out_group);
+      downsample->v_expand[ci] = (UINT8) (v_in_group / v_out_group);
+    } else
+      ERREXIT(cinfo, JERR_FRACT_SAMPLE_NOTIMPL);
+  }
+
+#ifdef INPUT_SMOOTHING_SUPPORTED
+  if (cinfo->smoothing_factor && !smoothok)
+    TRACEMS(cinfo, 0, JTRC_SMOOTH_NOTIMPL);
+#endif
+}
@@ -0,0 +1,399 @@
+/*
+ * jctrans.c
+ *
+ * Copyright (C) 1995-1998, Thomas G. Lane.
+ * Modified 2000-2020 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains library routines for transcoding compression,
+ * that is, writing raw DCT coefficient arrays to an output JPEG file.
+ * The routines in jcapimin.c will also be needed by a transcoder.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/* Forward declarations */
+LOCAL(void) transencode_master_selection
+	JPP((j_compress_ptr cinfo, jvirt_barray_ptr * coef_arrays));
+LOCAL(void) transencode_coef_controller
+	JPP((j_compress_ptr cinfo, jvirt_barray_ptr * coef_arrays));
+
+
+/*
+ * Compression initialization for writing raw-coefficient data.
+ * Before calling this, all parameters and a data destination must be set up.
+ * Call jpeg_finish_compress() to actually write the data.
+ *
+ * The number of passed virtual arrays must match cinfo->num_components.
+ * Note that the virtual arrays need not be filled or even realized at
+ * the time write_coefficients is called; indeed, if the virtual arrays
+ * were requested from this compression object's memory manager, they
+ * typically will be realized during this routine and filled afterwards.
+ */
+
+GLOBAL(void)
+jpeg_write_coefficients (j_compress_ptr cinfo, jvirt_barray_ptr * coef_arrays)
+{
+  if (cinfo->global_state != CSTATE_START)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+  /* Mark all tables to be written */
+  jpeg_suppress_tables(cinfo, FALSE);
+  /* (Re)initialize error mgr and destination modules */
+  (*cinfo->err->reset_error_mgr) ((j_common_ptr) cinfo);
+  (*cinfo->dest->init_destination) (cinfo);
+  /* Perform master selection of active modules */
+  transencode_master_selection(cinfo, coef_arrays);
+  /* Wait for jpeg_finish_compress() call */
+  cinfo->next_scanline = 0;	/* so jpeg_write_marker works */
+  cinfo->global_state = CSTATE_WRCOEFS;
+}
+
+
+/*
+ * Initialize the compression object with default parameters,
+ * then copy from the source object all parameters needed for lossless
+ * transcoding.  Parameters that can be varied without loss (such as
+ * scan script and Huffman optimization) are left in their default states.
+ */
+
+GLOBAL(void)
+jpeg_copy_critical_parameters (j_decompress_ptr srcinfo,
+			       j_compress_ptr dstinfo)
+{
+  JQUANT_TBL ** qtblptr;
+  jpeg_component_info *incomp, *outcomp;
+  JQUANT_TBL *c_quant, *slot_quant;
+  int tblno, ci, coefi;
+
+  /* Safety check to ensure start_compress not called yet. */
+  if (dstinfo->global_state != CSTATE_START)
+    ERREXIT1(dstinfo, JERR_BAD_STATE, dstinfo->global_state);
+  /* Copy fundamental image dimensions */
+  dstinfo->image_width = srcinfo->image_width;
+  dstinfo->image_height = srcinfo->image_height;
+  dstinfo->input_components = srcinfo->num_components;
+  dstinfo->in_color_space = srcinfo->jpeg_color_space;
+  dstinfo->jpeg_width = srcinfo->output_width;
+  dstinfo->jpeg_height = srcinfo->output_height;
+  dstinfo->min_DCT_h_scaled_size = srcinfo->min_DCT_h_scaled_size;
+  dstinfo->min_DCT_v_scaled_size = srcinfo->min_DCT_v_scaled_size;
+  /* Initialize all parameters to default values */
+  jpeg_set_defaults(dstinfo);
+  /* jpeg_set_defaults may choose wrong colorspace, eg YCbCr if input is RGB.
+   * Fix it to get the right header markers for the image colorspace.
+   * Note: Entropy table assignment in jpeg_set_colorspace
+   * depends on color_transform.
+   * Adaption is also required for setting the appropriate
+   * entropy coding mode dependent on image data precision.
+   */
+  dstinfo->color_transform = srcinfo->color_transform;
+  jpeg_set_colorspace(dstinfo, srcinfo->jpeg_color_space);
+  dstinfo->data_precision = srcinfo->data_precision;
+  dstinfo->arith_code = srcinfo->data_precision > 8 ? TRUE : FALSE;
+  dstinfo->CCIR601_sampling = srcinfo->CCIR601_sampling;
+  /* Copy the source's quantization tables. */
+  for (tblno = 0; tblno < NUM_QUANT_TBLS; tblno++) {
+    if (srcinfo->quant_tbl_ptrs[tblno] != NULL) {
+      qtblptr = & dstinfo->quant_tbl_ptrs[tblno];
+      if (*qtblptr == NULL)
+	*qtblptr = jpeg_alloc_quant_table((j_common_ptr) dstinfo);
+      MEMCOPY((*qtblptr)->quantval,
+	      srcinfo->quant_tbl_ptrs[tblno]->quantval,
+	      SIZEOF((*qtblptr)->quantval));
+      (*qtblptr)->sent_table = FALSE;
+    }
+  }
+  /* Copy the source's per-component info.
+   * Note we assume jpeg_set_defaults has allocated the dest comp_info array.
+   */
+  dstinfo->num_components = srcinfo->num_components;
+  if (dstinfo->num_components < 1 || dstinfo->num_components > MAX_COMPONENTS)
+    ERREXIT2(dstinfo, JERR_COMPONENT_COUNT, dstinfo->num_components,
+	     MAX_COMPONENTS);
+  for (ci = 0, incomp = srcinfo->comp_info, outcomp = dstinfo->comp_info;
+       ci < dstinfo->num_components; ci++, incomp++, outcomp++) {
+    outcomp->component_id = incomp->component_id;
+    outcomp->h_samp_factor = incomp->h_samp_factor;
+    outcomp->v_samp_factor = incomp->v_samp_factor;
+    outcomp->quant_tbl_no = incomp->quant_tbl_no;
+    /* Make sure saved quantization table for component matches the qtable
+     * slot.  If not, the input file re-used this qtable slot.
+     * IJG encoder currently cannot duplicate this.
+     */
+    tblno = outcomp->quant_tbl_no;
+    if (tblno < 0 || tblno >= NUM_QUANT_TBLS ||
+	srcinfo->quant_tbl_ptrs[tblno] == NULL)
+      ERREXIT1(dstinfo, JERR_NO_QUANT_TABLE, tblno);
+    slot_quant = srcinfo->quant_tbl_ptrs[tblno];
+    c_quant = incomp->quant_table;
+    if (c_quant != NULL) {
+      for (coefi = 0; coefi < DCTSIZE2; coefi++) {
+	if (c_quant->quantval[coefi] != slot_quant->quantval[coefi])
+	  ERREXIT1(dstinfo, JERR_MISMATCHED_QUANT_TABLE, tblno);
+      }
+    }
+    /* Note: we do not copy the source's entropy table assignments;
+     * instead we rely on jpeg_set_colorspace to have made a suitable choice.
+     */
+  }
+  /* Also copy JFIF version and resolution information, if available.
+   * Strictly speaking this isn't "critical" info, but it's nearly
+   * always appropriate to copy it if available.  In particular,
+   * if the application chooses to copy JFIF 1.02 extension markers from
+   * the source file, we need to copy the version to make sure we don't
+   * emit a file that has 1.02 extensions but a claimed version of 1.01.
+   */
+  if (srcinfo->saw_JFIF_marker) {
+    if (srcinfo->JFIF_major_version == 1 ||
+	srcinfo->JFIF_major_version == 2) {
+      dstinfo->JFIF_major_version = srcinfo->JFIF_major_version;
+      dstinfo->JFIF_minor_version = srcinfo->JFIF_minor_version;
+    }
+    dstinfo->density_unit = srcinfo->density_unit;
+    dstinfo->X_density = srcinfo->X_density;
+    dstinfo->Y_density = srcinfo->Y_density;
+  }
+}
+
+
+LOCAL(void)
+jpeg_calc_trans_dimensions (j_compress_ptr cinfo)
+/* Do computations that are needed before master selection phase */
+{
+  if (cinfo->min_DCT_h_scaled_size != cinfo->min_DCT_v_scaled_size)
+    ERREXIT2(cinfo, JERR_BAD_DCTSIZE,
+	     cinfo->min_DCT_h_scaled_size, cinfo->min_DCT_v_scaled_size);
+
+  cinfo->block_size = cinfo->min_DCT_h_scaled_size;
+}
+
+
+/*
+ * Master selection of compression modules for transcoding.
+ * This substitutes for jcinit.c's initialization of the full compressor.
+ */
+
+LOCAL(void)
+transencode_master_selection (j_compress_ptr cinfo,
+			      jvirt_barray_ptr * coef_arrays)
+{
+  /* Do computations that are needed before master selection phase */
+  jpeg_calc_trans_dimensions(cinfo);
+
+  /* Initialize master control (includes parameter checking/processing) */
+  jinit_c_master_control(cinfo, TRUE /* transcode only */);
+
+  /* Entropy encoding: either Huffman or arithmetic coding. */
+  if (cinfo->arith_code)
+    jinit_arith_encoder(cinfo);
+  else {
+    jinit_huff_encoder(cinfo);
+  }
+
+  /* We need a special coefficient buffer controller. */
+  transencode_coef_controller(cinfo, coef_arrays);
+
+  jinit_marker_writer(cinfo);
+
+  /* We can now tell the memory manager to allocate virtual arrays. */
+  (*cinfo->mem->realize_virt_arrays) ((j_common_ptr) cinfo);
+
+  /* Write the datastream header (SOI, JFIF) immediately.
+   * Frame and scan headers are postponed till later.
+   * This lets application insert special markers after the SOI.
+   */
+  (*cinfo->marker->write_file_header) (cinfo);
+}
+
+
+/*
+ * The rest of this file is a special implementation of the coefficient
+ * buffer controller.  This is similar to jccoefct.c, but it handles only
+ * output from presupplied virtual arrays.  Furthermore, we generate any
+ * dummy padding blocks on-the-fly rather than expecting them to be present
+ * in the arrays.
+ */
+
+/* Private buffer controller object */
+
+typedef struct {
+  struct jpeg_c_coef_controller pub; /* public fields */
+
+  JDIMENSION iMCU_row_num;	/* iMCU row # within image */
+  JDIMENSION MCU_ctr;		/* counts MCUs processed in current row */
+  int MCU_vert_offset;		/* counts MCU rows within iMCU row */
+  int MCU_rows_per_iMCU_row;	/* number of such rows needed */
+
+  /* Virtual block array for each component. */
+  jvirt_barray_ptr * whole_image;
+
+  /* Workspace for constructing dummy blocks at right/bottom edges. */
+  JBLOCK dummy_buffer[C_MAX_BLOCKS_IN_MCU];
+} my_coef_controller;
+
+typedef my_coef_controller * my_coef_ptr;
+
+
+LOCAL(void)
+start_iMCU_row (j_compress_ptr cinfo)
+/* Reset within-iMCU-row counters for a new row */
+{
+  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
+
+  /* In an interleaved scan, an MCU row is the same as an iMCU row.
+   * In a noninterleaved scan, an iMCU row has v_samp_factor MCU rows.
+   * But at the bottom of the image, process only what's left.
+   */
+  if (cinfo->comps_in_scan > 1) {
+    coef->MCU_rows_per_iMCU_row = 1;
+  } else {
+    if (coef->iMCU_row_num < (cinfo->total_iMCU_rows-1))
+      coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->v_samp_factor;
+    else
+      coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->last_row_height;
+  }
+
+  coef->MCU_ctr = 0;
+  coef->MCU_vert_offset = 0;
+}
+
+
+/*
+ * Initialize for a processing pass.
+ */
+
+METHODDEF(void)
+start_pass_coef (j_compress_ptr cinfo, J_BUF_MODE pass_mode)
+{
+  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
+
+  if (pass_mode != JBUF_CRANK_DEST)
+    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+
+  coef->iMCU_row_num = 0;
+  start_iMCU_row(cinfo);
+}
+
+
+/*
+ * Process some data.
+ * We process the equivalent of one fully interleaved MCU row ("iMCU" row)
+ * per call, ie, v_samp_factor block rows for each component in the scan.
+ * The data is obtained from the virtual arrays and fed to the entropy coder.
+ * Returns TRUE if the iMCU row is completed, FALSE if suspended.
+ *
+ * NB: input_buf is ignored; it is likely to be a NULL pointer.
+ */
+
+METHODDEF(boolean)
+compress_output (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
+{
+  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
+  JDIMENSION MCU_col_num;	/* index of current MCU within row */
+  JDIMENSION last_MCU_col = cinfo->MCUs_per_row - 1;
+  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
+  int blkn, ci, xindex, yindex, yoffset, blockcnt;
+  JDIMENSION start_col;
+  JBLOCKARRAY buffer[MAX_COMPS_IN_SCAN];
+  JBLOCKROW MCU_buffer[C_MAX_BLOCKS_IN_MCU];
+  JBLOCKROW buffer_ptr;
+  jpeg_component_info *compptr;
+
+  /* Align the virtual buffers for the components used in this scan. */
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+    compptr = cinfo->cur_comp_info[ci];
+    buffer[ci] = (*cinfo->mem->access_virt_barray)
+      ((j_common_ptr) cinfo, coef->whole_image[compptr->component_index],
+       coef->iMCU_row_num * compptr->v_samp_factor,
+       (JDIMENSION) compptr->v_samp_factor, FALSE);
+  }
+
+  /* Loop to process one whole iMCU row */
+  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
+       yoffset++) {
+    for (MCU_col_num = coef->MCU_ctr; MCU_col_num <= last_MCU_col;
+	 MCU_col_num++) {
+      /* Construct list of pointers to DCT blocks belonging to this MCU */
+      blkn = 0;			/* index of current DCT block within MCU */
+      for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+	compptr = cinfo->cur_comp_info[ci];
+	blockcnt = (MCU_col_num < last_MCU_col) ? compptr->MCU_width
+						: compptr->last_col_width;
+	start_col = MCU_col_num * compptr->MCU_width;
+	for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
+	  if (coef->iMCU_row_num < last_iMCU_row ||
+	      yoffset + yindex < compptr->last_row_height) {
+	    /* Fill in pointers to real blocks in this row */
+	    buffer_ptr = buffer[ci][yoffset + yindex] + start_col;
+	    xindex = blockcnt;
+	    do {
+	      MCU_buffer[blkn++] = buffer_ptr++;
+	    } while (--xindex);
+	    /* Dummy blocks at right edge */
+	    if ((xindex = compptr->MCU_width - blockcnt) == 0)
+	      continue;
+	  } else {
+	    /* At bottom of image, need a whole row of dummy blocks */
+	    xindex = compptr->MCU_width;
+	  }
+	  /* Fill in any dummy blocks needed in this row.
+	   * Dummy blocks are filled in the same way as in jccoefct.c:
+	   * all zeroes in the AC entries, DC entries equal to previous
+	   * block's DC value.  The init routine has already zeroed the
+	   * AC entries, so we need only set the DC entries correctly.
+	   */
+	  buffer_ptr = coef->dummy_buffer + blkn;
+	  do {
+	    buffer_ptr[0][0] = MCU_buffer[blkn-1][0][0];
+	    MCU_buffer[blkn++] = buffer_ptr++;
+	  } while (--xindex);
+	}
+      }
+      /* Try to write the MCU. */
+      if (! (*cinfo->entropy->encode_mcu) (cinfo, MCU_buffer)) {
+	/* Suspension forced; update state counters and exit */
+	coef->MCU_vert_offset = yoffset;
+	coef->MCU_ctr = MCU_col_num;
+	return FALSE;
+      }
+    }
+    /* Completed an MCU row, but perhaps not an iMCU row */
+    coef->MCU_ctr = 0;
+  }
+  /* Completed the iMCU row, advance counters for next one */
+  coef->iMCU_row_num++;
+  start_iMCU_row(cinfo);
+  return TRUE;
+}
+
+
+/*
+ * Initialize coefficient buffer controller.
+ *
+ * Each passed coefficient array must be the right size for that
+ * coefficient: width_in_blocks wide and height_in_blocks high,
+ * with unitheight at least v_samp_factor.
+ */
+
+LOCAL(void)
+transencode_coef_controller (j_compress_ptr cinfo,
+			     jvirt_barray_ptr * coef_arrays)
+{
+  my_coef_ptr coef;
+
+  coef = (my_coef_ptr) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, SIZEOF(my_coef_controller));
+  cinfo->coef = &coef->pub;
+  coef->pub.start_pass = start_pass_coef;
+  coef->pub.compress_data = compress_output;
+
+  /* Save pointer to virtual arrays */
+  coef->whole_image = coef_arrays;
+
+  /* Pre-zero space for dummy DCT blocks */
+  MEMZERO(coef->dummy_buffer, SIZEOF(coef->dummy_buffer));
+}
@@ -0,0 +1,412 @@
+/*
+ * jdapimin.c
+ *
+ * Copyright (C) 1994-1998, Thomas G. Lane.
+ * Modified 2009-2020 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains application interface code for the decompression half
+ * of the JPEG library.  These are the "minimum" API routines that may be
+ * needed in either the normal full-decompression case or the
+ * transcoding-only case.
+ *
+ * Most of the routines intended to be called directly by an application
+ * are in this file or in jdapistd.c.  But also see jcomapi.c for routines
+ * shared by compression and decompression, and jdtrans.c for the transcoding
+ * case.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/*
+ * Initialization of a JPEG decompression object.
+ * The error manager must already be set up (in case memory manager fails).
+ */
+
+GLOBAL(void)
+jpeg_CreateDecompress (j_decompress_ptr cinfo, int version, size_t structsize)
+{
+  int i;
+
+  /* Guard against version mismatches between library and caller. */
+  cinfo->mem = NULL;		/* so jpeg_destroy knows mem mgr not called */
+  if (version != JPEG_LIB_VERSION)
+    ERREXIT2(cinfo, JERR_BAD_LIB_VERSION, JPEG_LIB_VERSION, version);
+  if (structsize != SIZEOF(struct jpeg_decompress_struct))
+    ERREXIT2(cinfo, JERR_BAD_STRUCT_SIZE, 
+	     (int) SIZEOF(struct jpeg_decompress_struct), (int) structsize);
+
+  /* For debugging purposes, we zero the whole master structure.
+   * But the application has already set the err pointer, and may have set
+   * client_data, so we have to save and restore those fields.
+   * Note: if application hasn't set client_data, tools like Purify may
+   * complain here.
+   */
+  {
+    struct jpeg_error_mgr * err = cinfo->err;
+    void * client_data = cinfo->client_data; /* ignore Purify complaint here */
+    MEMZERO(cinfo, SIZEOF(struct jpeg_decompress_struct));
+    cinfo->err = err;
+    cinfo->client_data = client_data;
+  }
+  cinfo->is_decompressor = TRUE;
+
+  /* Initialize a memory manager instance for this object */
+  jinit_memory_mgr((j_common_ptr) cinfo);
+
+  /* Zero out pointers to permanent structures. */
+  cinfo->progress = NULL;
+  cinfo->src = NULL;
+
+  for (i = 0; i < NUM_QUANT_TBLS; i++)
+    cinfo->quant_tbl_ptrs[i] = NULL;
+
+  for (i = 0; i < NUM_HUFF_TBLS; i++) {
+    cinfo->dc_huff_tbl_ptrs[i] = NULL;
+    cinfo->ac_huff_tbl_ptrs[i] = NULL;
+  }
+
+  /* Initialize marker processor so application can override methods
+   * for COM, APPn markers before calling jpeg_read_header.
+   */
+  cinfo->marker_list = NULL;
+  jinit_marker_reader(cinfo);
+
+  /* And initialize the overall input controller. */
+  jinit_input_controller(cinfo);
+
+  /* OK, I'm ready */
+  cinfo->global_state = DSTATE_START;
+}
+
+
+/*
+ * Destruction of a JPEG decompression object
+ */
+
+GLOBAL(void)
+jpeg_destroy_decompress (j_decompress_ptr cinfo)
+{
+  jpeg_destroy((j_common_ptr) cinfo); /* use common routine */
+}
+
+
+/*
+ * Abort processing of a JPEG decompression operation,
+ * but don't destroy the object itself.
+ */
+
+GLOBAL(void)
+jpeg_abort_decompress (j_decompress_ptr cinfo)
+{
+  jpeg_abort((j_common_ptr) cinfo); /* use common routine */
+}
+
+
+/*
+ * Set default decompression parameters.
+ */
+
+LOCAL(void)
+default_decompress_parms (j_decompress_ptr cinfo)
+{
+  int cid0, cid1, cid2, cid3;
+
+  /* Guess the input colorspace, and set output colorspace accordingly. */
+  /* Note application may override our guesses. */
+  switch (cinfo->num_components) {
+  case 1:
+    cinfo->jpeg_color_space = JCS_GRAYSCALE;
+    cinfo->out_color_space = JCS_GRAYSCALE;
+    break;
+
+  case 3:
+    cid0 = cinfo->comp_info[0].component_id;
+    cid1 = cinfo->comp_info[1].component_id;
+    cid2 = cinfo->comp_info[2].component_id;
+
+    /* For robust detection of standard colorspaces
+     * regardless of the presence of special markers,
+     * check component IDs from SOF marker first.
+     */
+    if      (cid0 == 0x01 && cid1 == 0x02 && cid2 == 0x03)
+      cinfo->jpeg_color_space = JCS_YCbCr;
+    else if (cid0 == 0x01 && cid1 == 0x22 && cid2 == 0x23)
+      cinfo->jpeg_color_space = JCS_BG_YCC;
+    else if (cid0 == 0x52 && cid1 == 0x47 && cid2 == 0x42)
+      cinfo->jpeg_color_space = JCS_RGB;	/* ASCII 'R', 'G', 'B' */
+    else if (cid0 == 0x72 && cid1 == 0x67 && cid2 == 0x62)
+      cinfo->jpeg_color_space = JCS_BG_RGB;	/* ASCII 'r', 'g', 'b' */
+    else if (cinfo->saw_JFIF_marker)
+      cinfo->jpeg_color_space = JCS_YCbCr;	/* assume it's YCbCr */
+    else if (cinfo->saw_Adobe_marker) {
+      switch (cinfo->Adobe_transform) {
+      case 0:
+	cinfo->jpeg_color_space = JCS_RGB;
+	break;
+      case 1:
+	cinfo->jpeg_color_space = JCS_YCbCr;
+	break;
+      default:
+	WARNMS1(cinfo, JWRN_ADOBE_XFORM, cinfo->Adobe_transform);
+	cinfo->jpeg_color_space = JCS_YCbCr;	/* assume it's YCbCr */
+      }
+    } else {
+      TRACEMS3(cinfo, 1, JTRC_UNKNOWN_IDS, cid0, cid1, cid2);
+      cinfo->jpeg_color_space = JCS_YCbCr;	/* assume it's YCbCr */
+    }
+    /* Always guess RGB is proper output colorspace. */
+    cinfo->out_color_space = JCS_RGB;
+    break;
+
+  case 4:
+    cid0 = cinfo->comp_info[0].component_id;
+    cid1 = cinfo->comp_info[1].component_id;
+    cid2 = cinfo->comp_info[2].component_id;
+    cid3 = cinfo->comp_info[3].component_id;
+
+    /* For robust detection of standard colorspaces
+     * regardless of the presence of special markers,
+     * check component IDs from SOF marker first.
+     */
+    if      (cid0 == 0x01 && cid1 == 0x02 && cid2 == 0x03 && cid3 == 0x04)
+      cinfo->jpeg_color_space = JCS_YCCK;
+    else if (cid0 == 0x43 && cid1 == 0x4D && cid2 == 0x59 && cid3 == 0x4B)
+      cinfo->jpeg_color_space = JCS_CMYK;   /* ASCII 'C', 'M', 'Y', 'K' */
+    else if (cinfo->saw_Adobe_marker) {
+      switch (cinfo->Adobe_transform) {
+      case 0:
+	cinfo->jpeg_color_space = JCS_CMYK;
+	break;
+      case 2:
+	cinfo->jpeg_color_space = JCS_YCCK;
+	break;
+      default:
+	WARNMS1(cinfo, JWRN_ADOBE_XFORM, cinfo->Adobe_transform);
+	cinfo->jpeg_color_space = JCS_YCCK;	/* assume it's YCCK */
+      }
+    } else {
+      /* Unknown IDs and no special markers, assume straight CMYK. */
+      cinfo->jpeg_color_space = JCS_CMYK;
+    }
+    cinfo->out_color_space = JCS_CMYK;
+    break;
+
+  default:
+    cinfo->jpeg_color_space = JCS_UNKNOWN;
+    cinfo->out_color_space = JCS_UNKNOWN;
+  }
+
+  /* Set defaults for other decompression parameters. */
+  cinfo->scale_num = cinfo->block_size;		/* 1:1 scaling */
+  cinfo->scale_denom = cinfo->block_size;
+  cinfo->output_gamma = 1.0;
+  cinfo->buffered_image = FALSE;
+  cinfo->raw_data_out = FALSE;
+  cinfo->dct_method = JDCT_DEFAULT;
+  cinfo->do_fancy_upsampling = TRUE;
+  cinfo->do_block_smoothing = TRUE;
+  cinfo->quantize_colors = FALSE;
+  /* We set these in case application only sets quantize_colors. */
+  cinfo->dither_mode = JDITHER_FS;
+#ifdef QUANT_2PASS_SUPPORTED
+  cinfo->two_pass_quantize = TRUE;
+#else
+  cinfo->two_pass_quantize = FALSE;
+#endif
+  cinfo->desired_number_of_colors = 256;
+  cinfo->colormap = NULL;
+  /* Initialize for no mode change in buffered-image mode. */
+  cinfo->enable_1pass_quant = FALSE;
+  cinfo->enable_external_quant = FALSE;
+  cinfo->enable_2pass_quant = FALSE;
+}
+
+
+/*
+ * Decompression startup: read start of JPEG datastream to see what's there.
+ * Need only initialize JPEG object and supply a data source before calling.
+ *
+ * This routine will read as far as the first SOS marker (ie, actual start of
+ * compressed data), and will save all tables and parameters in the JPEG
+ * object.  It will also initialize the decompression parameters to default
+ * values, and finally return JPEG_HEADER_OK.  On return, the application may
+ * adjust the decompression parameters and then call jpeg_start_decompress.
+ * (Or, if the application only wanted to determine the image parameters,
+ * the data need not be decompressed.  In that case, call jpeg_abort or
+ * jpeg_destroy to release any temporary space.)
+ * If an abbreviated (tables only) datastream is presented, the routine will
+ * return JPEG_HEADER_TABLES_ONLY upon reaching EOI.  The application may then
+ * re-use the JPEG object to read the abbreviated image datastream(s).
+ * It is unnecessary (but OK) to call jpeg_abort in this case.
+ * The JPEG_SUSPENDED return code only occurs if the data source module
+ * requests suspension of the decompressor.  In this case the application
+ * should load more source data and then re-call jpeg_read_header to resume
+ * processing.
+ * If a non-suspending data source is used and require_image is TRUE, then the
+ * return code need not be inspected since only JPEG_HEADER_OK is possible.
+ *
+ * This routine is now just a front end to jpeg_consume_input, with some
+ * extra error checking.
+ */
+
+GLOBAL(int)
+jpeg_read_header (j_decompress_ptr cinfo, boolean require_image)
+{
+  int retcode;
+
+  if (cinfo->global_state != DSTATE_START &&
+      cinfo->global_state != DSTATE_INHEADER)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+
+  retcode = jpeg_consume_input(cinfo);
+
+  switch (retcode) {
+  case JPEG_REACHED_SOS:
+    retcode = JPEG_HEADER_OK;
+    break;
+  case JPEG_REACHED_EOI:
+    if (require_image)		/* Complain if application wanted an image */
+      ERREXIT(cinfo, JERR_NO_IMAGE);
+    /* Reset to start state; it would be safer to require the application to
+     * call jpeg_abort, but we can't change it now for compatibility reasons.
+     * A side effect is to free any temporary memory (there shouldn't be any).
+     */
+    jpeg_abort((j_common_ptr) cinfo); /* sets state = DSTATE_START */
+    retcode = JPEG_HEADER_TABLES_ONLY;
+    break;
+  case JPEG_SUSPENDED:
+    /* no work */
+    break;
+  }
+
+  return retcode;
+}
+
+
+/*
+ * Consume data in advance of what the decompressor requires.
+ * This can be called at any time once the decompressor object has
+ * been created and a data source has been set up.
+ *
+ * This routine is essentially a state machine that handles a couple
+ * of critical state-transition actions, namely initial setup and
+ * transition from header scanning to ready-for-start_decompress.
+ * All the actual input is done via the input controller's consume_input
+ * method.
+ */
+
+GLOBAL(int)
+jpeg_consume_input (j_decompress_ptr cinfo)
+{
+  int retcode = JPEG_SUSPENDED;
+
+  /* NB: every possible DSTATE value should be listed in this switch */
+  switch (cinfo->global_state) {
+  case DSTATE_START:
+    /* Start-of-datastream actions: reset appropriate modules */
+    (*cinfo->inputctl->reset_input_controller) (cinfo);
+    /* Initialize application's data source module */
+    (*cinfo->src->init_source) (cinfo);
+    cinfo->global_state = DSTATE_INHEADER;
+    /*FALLTHROUGH*/
+  case DSTATE_INHEADER:
+    retcode = (*cinfo->inputctl->consume_input) (cinfo);
+    if (retcode == JPEG_REACHED_SOS) { /* Found SOS, prepare to decompress */
+      /* Set up default parameters based on header data */
+      default_decompress_parms(cinfo);
+      /* Set global state: ready for start_decompress */
+      cinfo->global_state = DSTATE_READY;
+    }
+    break;
+  case DSTATE_READY:
+    /* Can't advance past first SOS until start_decompress is called */
+    retcode = JPEG_REACHED_SOS;
+    break;
+  case DSTATE_PRELOAD:
+  case DSTATE_PRESCAN:
+  case DSTATE_SCANNING:
+  case DSTATE_RAW_OK:
+  case DSTATE_BUFIMAGE:
+  case DSTATE_BUFPOST:
+  case DSTATE_STOPPING:
+    retcode = (*cinfo->inputctl->consume_input) (cinfo);
+    break;
+  default:
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+  }
+  return retcode;
+}
+
+
+/*
+ * Have we finished reading the input file?
+ */
+
+GLOBAL(boolean)
+jpeg_input_complete (j_decompress_ptr cinfo)
+{
+  /* Check for valid jpeg object */
+  if (cinfo->global_state < DSTATE_START ||
+      cinfo->global_state > DSTATE_STOPPING)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+  return cinfo->inputctl->eoi_reached;
+}
+
+
+/*
+ * Is there more than one scan?
+ */
+
+GLOBAL(boolean)
+jpeg_has_multiple_scans (j_decompress_ptr cinfo)
+{
+  /* Only valid after jpeg_read_header completes */
+  if (cinfo->global_state < DSTATE_READY ||
+      cinfo->global_state > DSTATE_STOPPING)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+  return cinfo->inputctl->has_multiple_scans;
+}
+
+
+/*
+ * Finish JPEG decompression.
+ *
+ * This will normally just verify the file trailer and release temp storage.
+ *
+ * Returns FALSE if suspended.  The return value need be inspected only if
+ * a suspending data source is used.
+ */
+
+GLOBAL(boolean)
+jpeg_finish_decompress (j_decompress_ptr cinfo)
+{
+  if ((cinfo->global_state == DSTATE_SCANNING ||
+       cinfo->global_state == DSTATE_RAW_OK) && ! cinfo->buffered_image) {
+    /* Terminate final pass of non-buffered mode */
+    if (cinfo->output_scanline < cinfo->output_height)
+      ERREXIT(cinfo, JERR_TOO_LITTLE_DATA);
+    (*cinfo->master->finish_output_pass) (cinfo);
+    cinfo->global_state = DSTATE_STOPPING;
+  } else if (cinfo->global_state == DSTATE_BUFIMAGE) {
+    /* Finishing after a buffered-image operation */
+    cinfo->global_state = DSTATE_STOPPING;
+  } else if (cinfo->global_state != DSTATE_STOPPING) {
+    /* STOPPING = repeat call after a suspension, anything else is error */
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+  }
+  /* Read until EOI */
+  while (! cinfo->inputctl->eoi_reached) {
+    if ((*cinfo->inputctl->consume_input) (cinfo) == JPEG_SUSPENDED)
+      return FALSE;		/* Suspend, come back later */
+  }
+  /* Do final cleanup */
+  (*cinfo->src->term_source) (cinfo);
+  /* We can use jpeg_abort to release memory and reset global_state */
+  jpeg_abort((j_common_ptr) cinfo);
+  return TRUE;
+}
@@ -0,0 +1,276 @@
+/*
+ * jdapistd.c
+ *
+ * Copyright (C) 1994-1996, Thomas G. Lane.
+ * Modified 2002-2013 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains application interface code for the decompression half
+ * of the JPEG library.  These are the "standard" API routines that are
+ * used in the normal full-decompression case.  They are not used by a
+ * transcoding-only application.  Note that if an application links in
+ * jpeg_start_decompress, it will end up linking in the entire decompressor.
+ * We thus must separate this file from jdapimin.c to avoid linking the
+ * whole decompression library into a transcoder.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/* Forward declarations */
+LOCAL(boolean) output_pass_setup JPP((j_decompress_ptr cinfo));
+
+
+/*
+ * Decompression initialization.
+ * jpeg_read_header must be completed before calling this.
+ *
+ * If a multipass operating mode was selected, this will do all but the
+ * last pass, and thus may take a great deal of time.
+ *
+ * Returns FALSE if suspended.  The return value need be inspected only if
+ * a suspending data source is used.
+ */
+
+GLOBAL(boolean)
+jpeg_start_decompress (j_decompress_ptr cinfo)
+{
+  if (cinfo->global_state == DSTATE_READY) {
+    /* First call: initialize master control, select active modules */
+    jinit_master_decompress(cinfo);
+    if (cinfo->buffered_image) {
+      /* No more work here; expecting jpeg_start_output next */
+      cinfo->global_state = DSTATE_BUFIMAGE;
+      return TRUE;
+    }
+    cinfo->global_state = DSTATE_PRELOAD;
+  }
+  if (cinfo->global_state == DSTATE_PRELOAD) {
+    /* If file has multiple scans, absorb them all into the coef buffer */
+    if (cinfo->inputctl->has_multiple_scans) {
+#ifdef D_MULTISCAN_FILES_SUPPORTED
+      for (;;) {
+	int retcode;
+	/* Call progress monitor hook if present */
+	if (cinfo->progress != NULL)
+	  (*cinfo->progress->progress_monitor) ((j_common_ptr) cinfo);
+	/* Absorb some more input */
+	retcode = (*cinfo->inputctl->consume_input) (cinfo);
+	if (retcode == JPEG_SUSPENDED)
+	  return FALSE;
+	if (retcode == JPEG_REACHED_EOI)
+	  break;
+	/* Advance progress counter if appropriate */
+	if (cinfo->progress != NULL &&
+	    (retcode == JPEG_ROW_COMPLETED || retcode == JPEG_REACHED_SOS)) {
+	  if (++cinfo->progress->pass_counter >= cinfo->progress->pass_limit) {
+	    /* jdmaster underestimated number of scans; ratchet up one scan */
+	    cinfo->progress->pass_limit += (long) cinfo->total_iMCU_rows;
+	  }
+	}
+      }
+#else
+      ERREXIT(cinfo, JERR_NOT_COMPILED);
+#endif /* D_MULTISCAN_FILES_SUPPORTED */
+    }
+    cinfo->output_scan_number = cinfo->input_scan_number;
+  } else if (cinfo->global_state != DSTATE_PRESCAN)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+  /* Perform any dummy output passes, and set up for the final pass */
+  return output_pass_setup(cinfo);
+}
+
+
+/*
+ * Set up for an output pass, and perform any dummy pass(es) needed.
+ * Common subroutine for jpeg_start_decompress and jpeg_start_output.
+ * Entry: global_state = DSTATE_PRESCAN only if previously suspended.
+ * Exit: If done, returns TRUE and sets global_state for proper output mode.
+ *       If suspended, returns FALSE and sets global_state = DSTATE_PRESCAN.
+ */
+
+LOCAL(boolean)
+output_pass_setup (j_decompress_ptr cinfo)
+{
+  if (cinfo->global_state != DSTATE_PRESCAN) {
+    /* First call: do pass setup */
+    (*cinfo->master->prepare_for_output_pass) (cinfo);
+    cinfo->output_scanline = 0;
+    cinfo->global_state = DSTATE_PRESCAN;
+  }
+  /* Loop over any required dummy passes */
+  while (cinfo->master->is_dummy_pass) {
+#ifdef QUANT_2PASS_SUPPORTED
+    /* Crank through the dummy pass */
+    while (cinfo->output_scanline < cinfo->output_height) {
+      JDIMENSION last_scanline;
+      /* Call progress monitor hook if present */
+      if (cinfo->progress != NULL) {
+	cinfo->progress->pass_counter = (long) cinfo->output_scanline;
+	cinfo->progress->pass_limit = (long) cinfo->output_height;
+	(*cinfo->progress->progress_monitor) ((j_common_ptr) cinfo);
+      }
+      /* Process some data */
+      last_scanline = cinfo->output_scanline;
+      (*cinfo->main->process_data) (cinfo, (JSAMPARRAY) NULL,
+				    &cinfo->output_scanline, (JDIMENSION) 0);
+      if (cinfo->output_scanline == last_scanline)
+	return FALSE;		/* No progress made, must suspend */
+    }
+    /* Finish up dummy pass, and set up for another one */
+    (*cinfo->master->finish_output_pass) (cinfo);
+    (*cinfo->master->prepare_for_output_pass) (cinfo);
+    cinfo->output_scanline = 0;
+#else
+    ERREXIT(cinfo, JERR_NOT_COMPILED);
+#endif /* QUANT_2PASS_SUPPORTED */
+  }
+  /* Ready for application to drive output pass through
+   * jpeg_read_scanlines or jpeg_read_raw_data.
+   */
+  cinfo->global_state = cinfo->raw_data_out ? DSTATE_RAW_OK : DSTATE_SCANNING;
+  return TRUE;
+}
+
+
+/*
+ * Read some scanlines of data from the JPEG decompressor.
+ *
+ * The return value will be the number of lines actually read.
+ * This may be less than the number requested in several cases,
+ * including bottom of image, data source suspension, and operating
+ * modes that emit multiple scanlines at a time.
+ *
+ * Note: we warn about excess calls to jpeg_read_scanlines() since
+ * this likely signals an application programmer error.  However,
+ * an oversize buffer (max_lines > scanlines remaining) is not an error.
+ */
+
+GLOBAL(JDIMENSION)
+jpeg_read_scanlines (j_decompress_ptr cinfo, JSAMPARRAY scanlines,
+		     JDIMENSION max_lines)
+{
+  JDIMENSION row_ctr;
+
+  if (cinfo->global_state != DSTATE_SCANNING)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+  if (cinfo->output_scanline >= cinfo->output_height) {
+    WARNMS(cinfo, JWRN_TOO_MUCH_DATA);
+    return 0;
+  }
+
+  /* Call progress monitor hook if present */
+  if (cinfo->progress != NULL) {
+    cinfo->progress->pass_counter = (long) cinfo->output_scanline;
+    cinfo->progress->pass_limit = (long) cinfo->output_height;
+    (*cinfo->progress->progress_monitor) ((j_common_ptr) cinfo);
+  }
+
+  /* Process some data */
+  row_ctr = 0;
+  (*cinfo->main->process_data) (cinfo, scanlines, &row_ctr, max_lines);
+  cinfo->output_scanline += row_ctr;
+  return row_ctr;
+}
+
+
+/*
+ * Alternate entry point to read raw data.
+ * Processes exactly one iMCU row per call, unless suspended.
+ */
+
+GLOBAL(JDIMENSION)
+jpeg_read_raw_data (j_decompress_ptr cinfo, JSAMPIMAGE data,
+		    JDIMENSION max_lines)
+{
+  JDIMENSION lines_per_iMCU_row;
+
+  if (cinfo->global_state != DSTATE_RAW_OK)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+  if (cinfo->output_scanline >= cinfo->output_height) {
+    WARNMS(cinfo, JWRN_TOO_MUCH_DATA);
+    return 0;
+  }
+
+  /* Call progress monitor hook if present */
+  if (cinfo->progress != NULL) {
+    cinfo->progress->pass_counter = (long) cinfo->output_scanline;
+    cinfo->progress->pass_limit = (long) cinfo->output_height;
+    (*cinfo->progress->progress_monitor) ((j_common_ptr) cinfo);
+  }
+
+  /* Verify that at least one iMCU row can be returned. */
+  lines_per_iMCU_row = cinfo->max_v_samp_factor * cinfo->min_DCT_v_scaled_size;
+  if (max_lines < lines_per_iMCU_row)
+    ERREXIT(cinfo, JERR_BUFFER_SIZE);
+
+  /* Decompress directly into user's buffer. */
+  if (! (*cinfo->coef->decompress_data) (cinfo, data))
+    return 0;			/* suspension forced, can do nothing more */
+
+  /* OK, we processed one iMCU row. */
+  cinfo->output_scanline += lines_per_iMCU_row;
+  return lines_per_iMCU_row;
+}
+
+
+/* Additional entry points for buffered-image mode. */
+
+#ifdef D_MULTISCAN_FILES_SUPPORTED
+
+/*
+ * Initialize for an output pass in buffered-image mode.
+ */
+
+GLOBAL(boolean)
+jpeg_start_output (j_decompress_ptr cinfo, int scan_number)
+{
+  if (cinfo->global_state != DSTATE_BUFIMAGE &&
+      cinfo->global_state != DSTATE_PRESCAN)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+  /* Limit scan number to valid range */
+  if (scan_number <= 0)
+    scan_number = 1;
+  if (cinfo->inputctl->eoi_reached &&
+      scan_number > cinfo->input_scan_number)
+    scan_number = cinfo->input_scan_number;
+  cinfo->output_scan_number = scan_number;
+  /* Perform any dummy output passes, and set up for the real pass */
+  return output_pass_setup(cinfo);
+}
+
+
+/*
+ * Finish up after an output pass in buffered-image mode.
+ *
+ * Returns FALSE if suspended.  The return value need be inspected only if
+ * a suspending data source is used.
+ */
+
+GLOBAL(boolean)
+jpeg_finish_output (j_decompress_ptr cinfo)
+{
+  if ((cinfo->global_state == DSTATE_SCANNING ||
+       cinfo->global_state == DSTATE_RAW_OK) && cinfo->buffered_image) {
+    /* Terminate this pass. */
+    /* We do not require the whole pass to have been completed. */
+    (*cinfo->master->finish_output_pass) (cinfo);
+    cinfo->global_state = DSTATE_BUFPOST;
+  } else if (cinfo->global_state != DSTATE_BUFPOST) {
+    /* BUFPOST = repeat call after a suspension, anything else is error */
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+  }
+  /* Read markers looking for SOS or EOI */
+  while (cinfo->input_scan_number <= cinfo->output_scan_number &&
+	 ! cinfo->inputctl->eoi_reached) {
+    if ((*cinfo->inputctl->consume_input) (cinfo) == JPEG_SUSPENDED)
+      return FALSE;		/* Suspend, come back later */
+  }
+  cinfo->global_state = DSTATE_BUFIMAGE;
+  return TRUE;
+}
+
+#endif /* D_MULTISCAN_FILES_SUPPORTED */
@@ -0,0 +1,796 @@
+/*
+ * jdarith.c
+ *
+ * Developed 1997-2020 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains portable arithmetic entropy decoding routines for JPEG
+ * (implementing the ISO/IEC IS 10918-1 and CCITT Recommendation ITU-T T.81).
+ *
+ * Both sequential and progressive modes are supported in this single module.
+ *
+ * Suspension is not currently supported in this module.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/* Expanded entropy decoder object for arithmetic decoding. */
+
+typedef struct {
+  struct jpeg_entropy_decoder pub; /* public fields */
+
+  INT32 c;       /* C register, base of coding interval + input bit buffer */
+  INT32 a;               /* A register, normalized size of coding interval */
+  int ct;     /* bit shift counter, # of bits left in bit buffer part of C */
+                                                         /* init: ct = -16 */
+                                                         /* run: ct = 0..7 */
+                                                         /* error: ct = -1 */
+  int last_dc_val[MAX_COMPS_IN_SCAN]; /* last DC coef for each component */
+  int dc_context[MAX_COMPS_IN_SCAN]; /* context index for DC conditioning */
+
+  unsigned int restarts_to_go;	/* MCUs left in this restart interval */
+
+  /* Pointers to statistics areas (these workspaces have image lifespan) */
+  unsigned char * dc_stats[NUM_ARITH_TBLS];
+  unsigned char * ac_stats[NUM_ARITH_TBLS];
+
+  /* Statistics bin for coding with fixed probability 0.5 */
+  unsigned char fixed_bin[4];
+} arith_entropy_decoder;
+
+typedef arith_entropy_decoder * arith_entropy_ptr;
+
+/* The following two definitions specify the allocation chunk size
+ * for the statistics area.
+ * According to sections F.1.4.4.1.3 and F.1.4.4.2, we need at least
+ * 49 statistics bins for DC, and 245 statistics bins for AC coding.
+ *
+ * We use a compact representation with 1 byte per statistics bin,
+ * thus the numbers directly represent byte sizes.
+ * This 1 byte per statistics bin contains the meaning of the MPS
+ * (more probable symbol) in the highest bit (mask 0x80), and the
+ * index into the probability estimation state machine table
+ * in the lower bits (mask 0x7F).
+ */
+
+#define DC_STAT_BINS 64
+#define AC_STAT_BINS 256
+
+
+LOCAL(int)
+get_byte (j_decompress_ptr cinfo)
+/* Read next input byte; we do not support suspension in this module. */
+{
+  struct jpeg_source_mgr * src = cinfo->src;
+
+  if (src->bytes_in_buffer == 0)
+    if (! (*src->fill_input_buffer) (cinfo))
+      ERREXIT(cinfo, JERR_CANT_SUSPEND);
+  src->bytes_in_buffer--;
+  return GETJOCTET(*src->next_input_byte++);
+}
+
+
+/*
+ * The core arithmetic decoding routine (common in JPEG and JBIG).
+ * This needs to go as fast as possible.
+ * Machine-dependent optimization facilities
+ * are not utilized in this portable implementation.
+ * However, this code should be fairly efficient and
+ * may be a good base for further optimizations anyway.
+ *
+ * Return value is 0 or 1 (binary decision).
+ *
+ * Note: I've changed the handling of the code base & bit
+ * buffer register C compared to other implementations
+ * based on the standards layout & procedures.
+ * While it also contains both the actual base of the
+ * coding interval (16 bits) and the next-bits buffer,
+ * the cut-point between these two parts is floating
+ * (instead of fixed) with the bit shift counter CT.
+ * Thus, we also need only one (variable instead of
+ * fixed size) shift for the LPS/MPS decision, and
+ * we can do away with any renormalization update
+ * of C (except for new data insertion, of course).
+ *
+ * I've also introduced a new scheme for accessing
+ * the probability estimation state machine table,
+ * derived from Markus Kuhn's JBIG implementation.
+ */
+
+LOCAL(int)
+arith_decode (j_decompress_ptr cinfo, unsigned char *st)
+{
+  register arith_entropy_ptr e = (arith_entropy_ptr) cinfo->entropy;
+  register unsigned char nl, nm;
+  register INT32 qe, temp;
+  register int sv, data;
+
+  /* Renormalization & data input per section D.2.6 */
+  while (e->a < 0x8000L) {
+    if (--e->ct < 0) {
+      /* Need to fetch next data byte */
+      if (cinfo->unread_marker)
+	data = 0;		/* stuff zero data */
+      else {
+	data = get_byte(cinfo);	/* read next input byte */
+	if (data == 0xFF) {	/* zero stuff or marker code */
+	  do data = get_byte(cinfo);
+	  while (data == 0xFF);	/* swallow extra 0xFF bytes */
+	  if (data == 0)
+	    data = 0xFF;	/* discard stuffed zero byte */
+	  else {
+	    /* Note: Different from the Huffman decoder, hitting
+	     * a marker while processing the compressed data
+	     * segment is legal in arithmetic coding.
+	     * The convention is to supply zero data
+	     * then until decoding is complete.
+	     */
+	    cinfo->unread_marker = data;
+	    data = 0;
+	  }
+	}
+      }
+      e->c = (e->c << 8) | data; /* insert data into C register */
+      if ((e->ct += 8) < 0)	 /* update bit shift counter */
+	/* Need more initial bytes */
+	if (++e->ct == 0)
+	  /* Got 2 initial bytes -> re-init A and exit loop */
+	  e->a = 0x8000L; /* => e->a = 0x10000L after loop exit */
+    }
+    e->a <<= 1;
+  }
+
+  /* Fetch values from our compact representation of Table D.3(D.2):
+   * Qe values and probability estimation state machine
+   */
+  sv = *st;
+  qe = jpeg_aritab[sv & 0x7F];	/* => Qe_Value */
+  nl = qe & 0xFF; qe >>= 8;	/* Next_Index_LPS + Switch_MPS */
+  nm = qe & 0xFF; qe >>= 8;	/* Next_Index_MPS */
+
+  /* Decode & estimation procedures per sections D.2.4 & D.2.5 */
+  temp = e->a - qe;
+  e->a = temp;
+  temp <<= e->ct;
+  if (e->c >= temp) {
+    e->c -= temp;
+    /* Conditional LPS (less probable symbol) exchange */
+    if (e->a < qe) {
+      e->a = qe;
+      *st = (sv & 0x80) ^ nm;	/* Estimate_after_MPS */
+    } else {
+      e->a = qe;
+      *st = (sv & 0x80) ^ nl;	/* Estimate_after_LPS */
+      sv ^= 0x80;		/* Exchange LPS/MPS */
+    }
+  } else if (e->a < 0x8000L) {
+    /* Conditional MPS (more probable symbol) exchange */
+    if (e->a < qe) {
+      *st = (sv & 0x80) ^ nl;	/* Estimate_after_LPS */
+      sv ^= 0x80;		/* Exchange LPS/MPS */
+    } else {
+      *st = (sv & 0x80) ^ nm;	/* Estimate_after_MPS */
+    }
+  }
+
+  return sv >> 7;
+}
+
+
+/*
+ * Check for a restart marker & resynchronize decoder.
+ */
+
+LOCAL(void)
+process_restart (j_decompress_ptr cinfo)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  int ci;
+  jpeg_component_info * compptr;
+
+  /* Advance past the RSTn marker */
+  if (! (*cinfo->marker->read_restart_marker) (cinfo))
+    ERREXIT(cinfo, JERR_CANT_SUSPEND);
+
+  /* Re-initialize statistics areas */
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+    compptr = cinfo->cur_comp_info[ci];
+    if (! cinfo->progressive_mode || (cinfo->Ss == 0 && cinfo->Ah == 0)) {
+      MEMZERO(entropy->dc_stats[compptr->dc_tbl_no], DC_STAT_BINS);
+      /* Reset DC predictions to 0 */
+      entropy->last_dc_val[ci] = 0;
+      entropy->dc_context[ci] = 0;
+    }
+    if ((! cinfo->progressive_mode && cinfo->lim_Se) ||
+	(cinfo->progressive_mode && cinfo->Ss)) {
+      MEMZERO(entropy->ac_stats[compptr->ac_tbl_no], AC_STAT_BINS);
+    }
+  }
+
+  /* Reset arithmetic decoding variables */
+  entropy->c = 0;
+  entropy->a = 0;
+  entropy->ct = -16;	/* force reading 2 initial bytes to fill C */
+
+  /* Reset restart counter */
+  entropy->restarts_to_go = cinfo->restart_interval;
+}
+
+
+/*
+ * Arithmetic MCU decoding.
+ * Each of these routines decodes and returns one MCU's worth of
+ * arithmetic-compressed coefficients.
+ * The coefficients are reordered from zigzag order into natural array order,
+ * but are not dequantized.
+ *
+ * The i'th block of the MCU is stored into the block pointed to by
+ * MCU_data[i].  WE ASSUME THIS AREA IS INITIALLY ZEROED BY THE CALLER.
+ */
+
+/*
+ * MCU decoding for DC initial scan (either spectral selection,
+ * or first pass of successive approximation).
+ */
+
+METHODDEF(boolean)
+decode_mcu_DC_first (j_decompress_ptr cinfo, JBLOCKARRAY MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  JBLOCKROW block;
+  unsigned char *st;
+  int blkn, ci, tbl, sign;
+  int v, m;
+
+  /* Process restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0)
+      process_restart(cinfo);
+    entropy->restarts_to_go--;
+  }
+
+  if (entropy->ct == -1) return TRUE;	/* if error do nothing */
+
+  /* Outer loop handles each block in the MCU */
+
+  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+    block = MCU_data[blkn];
+    ci = cinfo->MCU_membership[blkn];
+    tbl = cinfo->cur_comp_info[ci]->dc_tbl_no;
+
+    /* Sections F.2.4.1 & F.1.4.4.1: Decoding of DC coefficients */
+
+    /* Table F.4: Point to statistics bin S0 for DC coefficient coding */
+    st = entropy->dc_stats[tbl] + entropy->dc_context[ci];
+
+    /* Figure F.19: Decode_DC_DIFF */
+    if (arith_decode(cinfo, st) == 0)
+      entropy->dc_context[ci] = 0;
+    else {
+      /* Figure F.21: Decoding nonzero value v */
+      /* Figure F.22: Decoding the sign of v */
+      sign = arith_decode(cinfo, st + 1);
+      st += 2; st += sign;
+      /* Figure F.23: Decoding the magnitude category of v */
+      if ((m = arith_decode(cinfo, st)) != 0) {
+	st = entropy->dc_stats[tbl] + 20;	/* Table F.4: X1 = 20 */
+	while (arith_decode(cinfo, st)) {
+	  if ((m <<= 1) == (int) 0x8000U) {
+	    WARNMS(cinfo, JWRN_ARITH_BAD_CODE);
+	    entropy->ct = -1;			/* magnitude overflow */
+	    return TRUE;
+	  }
+	  st += 1;
+	}
+      }
+      /* Section F.1.4.4.1.2: Establish dc_context conditioning category */
+      if (m < (int) ((1L << cinfo->arith_dc_L[tbl]) >> 1))
+	entropy->dc_context[ci] = 0;		   /* zero diff category */
+      else if (m > (int) ((1L << cinfo->arith_dc_U[tbl]) >> 1))
+	entropy->dc_context[ci] = 12 + (sign * 4); /* large diff category */
+      else
+	entropy->dc_context[ci] = 4 + (sign * 4);  /* small diff category */
+      v = m;
+      /* Figure F.24: Decoding the magnitude bit pattern of v */
+      st += 14;
+      while (m >>= 1)
+	if (arith_decode(cinfo, st)) v |= m;
+      v += 1; if (sign) v = -v;
+      entropy->last_dc_val[ci] += v;
+    }
+
+    /* Scale and output the DC coefficient (assumes jpeg_natural_order[0]=0) */
+    (*block)[0] = (JCOEF) (entropy->last_dc_val[ci] << cinfo->Al);
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * MCU decoding for AC initial scan (either spectral selection,
+ * or first pass of successive approximation).
+ */
+
+METHODDEF(boolean)
+decode_mcu_AC_first (j_decompress_ptr cinfo, JBLOCKARRAY MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  JBLOCKROW block;
+  unsigned char *st;
+  int tbl, sign, k;
+  int v, m;
+  const int * natural_order;
+
+  /* Process restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0)
+      process_restart(cinfo);
+    entropy->restarts_to_go--;
+  }
+
+  if (entropy->ct == -1) return TRUE;	/* if error do nothing */
+
+  natural_order = cinfo->natural_order;
+
+  /* There is always only one block per MCU */
+  block = MCU_data[0];
+  tbl = cinfo->cur_comp_info[0]->ac_tbl_no;
+
+  /* Sections F.2.4.2 & F.1.4.4.2: Decoding of AC coefficients */
+
+  /* Figure F.20: Decode_AC_coefficients */
+  k = cinfo->Ss - 1;
+  do {
+    st = entropy->ac_stats[tbl] + 3 * k;
+    if (arith_decode(cinfo, st)) break;		/* EOB flag */
+    for (;;) {
+      k++;
+      if (arith_decode(cinfo, st + 1)) break;
+      st += 3;
+      if (k >= cinfo->Se) {
+	WARNMS(cinfo, JWRN_ARITH_BAD_CODE);
+	entropy->ct = -1;			/* spectral overflow */
+	return TRUE;
+      }
+    }
+    /* Figure F.21: Decoding nonzero value v */
+    /* Figure F.22: Decoding the sign of v */
+    sign = arith_decode(cinfo, entropy->fixed_bin);
+    st += 2;
+    /* Figure F.23: Decoding the magnitude category of v */
+    if ((m = arith_decode(cinfo, st)) != 0) {
+      if (arith_decode(cinfo, st)) {
+	m <<= 1;
+	st = entropy->ac_stats[tbl] +
+	     (k <= cinfo->arith_ac_K[tbl] ? 189 : 217);
+	while (arith_decode(cinfo, st)) {
+	  if ((m <<= 1) == (int) 0x8000U) {
+	    WARNMS(cinfo, JWRN_ARITH_BAD_CODE);
+	    entropy->ct = -1;			/* magnitude overflow */
+	    return TRUE;
+	  }
+	  st += 1;
+	}
+      }
+    }
+    v = m;
+    /* Figure F.24: Decoding the magnitude bit pattern of v */
+    st += 14;
+    while (m >>= 1)
+      if (arith_decode(cinfo, st)) v |= m;
+    v += 1; if (sign) v = -v;
+    /* Scale and output coefficient in natural (dezigzagged) order */
+    (*block)[natural_order[k]] = (JCOEF) (v << cinfo->Al);
+  } while (k < cinfo->Se);
+
+  return TRUE;
+}
+
+
+/*
+ * MCU decoding for DC successive approximation refinement scan.
+ * Note: we assume such scans can be multi-component,
+ * although the spec is not very clear on the point.
+ */
+
+METHODDEF(boolean)
+decode_mcu_DC_refine (j_decompress_ptr cinfo, JBLOCKARRAY MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  unsigned char *st;
+  JCOEF p1;
+  int blkn;
+
+  /* Process restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0)
+      process_restart(cinfo);
+    entropy->restarts_to_go--;
+  }
+
+  st = entropy->fixed_bin;	/* use fixed probability estimation */
+  p1 = 1 << cinfo->Al;		/* 1 in the bit position being coded */
+
+  /* Outer loop handles each block in the MCU */
+
+  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+    /* Encoded data is simply the next bit of the two's-complement DC value */
+    if (arith_decode(cinfo, st))
+      MCU_data[blkn][0][0] |= p1;
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * MCU decoding for AC successive approximation refinement scan.
+ */
+
+METHODDEF(boolean)
+decode_mcu_AC_refine (j_decompress_ptr cinfo, JBLOCKARRAY MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  JBLOCKROW block;
+  JCOEFPTR thiscoef;
+  unsigned char *st;
+  int tbl, k, kex;
+  JCOEF p1, m1;
+  const int * natural_order;
+
+  /* Process restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0)
+      process_restart(cinfo);
+    entropy->restarts_to_go--;
+  }
+
+  if (entropy->ct == -1) return TRUE;	/* if error do nothing */
+
+  natural_order = cinfo->natural_order;
+
+  /* There is always only one block per MCU */
+  block = MCU_data[0];
+  tbl = cinfo->cur_comp_info[0]->ac_tbl_no;
+
+  p1 = 1 << cinfo->Al;		/* 1 in the bit position being coded */
+  m1 = -p1;			/* -1 in the bit position being coded */
+
+  /* Establish EOBx (previous stage end-of-block) index */
+  kex = cinfo->Se;
+  do {
+    if ((*block)[natural_order[kex]]) break;
+  } while (--kex);
+
+  k = cinfo->Ss - 1;
+  do {
+    st = entropy->ac_stats[tbl] + 3 * k;
+    if (k >= kex)
+      if (arith_decode(cinfo, st)) break;	/* EOB flag */
+    for (;;) {
+      thiscoef = *block + natural_order[++k];
+      if (*thiscoef) {				/* previously nonzero coef */
+	if (arith_decode(cinfo, st + 2)) {
+	  if (*thiscoef < 0)
+	    *thiscoef += m1;
+	  else
+	    *thiscoef += p1;
+	}
+	break;
+      }
+      if (arith_decode(cinfo, st + 1)) {	/* newly nonzero coef */
+	if (arith_decode(cinfo, entropy->fixed_bin))
+	  *thiscoef = m1;
+	else
+	  *thiscoef = p1;
+	break;
+      }
+      st += 3;
+      if (k >= cinfo->Se) {
+	WARNMS(cinfo, JWRN_ARITH_BAD_CODE);
+	entropy->ct = -1;			/* spectral overflow */
+	return TRUE;
+      }
+    }
+  } while (k < cinfo->Se);
+
+  return TRUE;
+}
+
+
+/*
+ * Decode one MCU's worth of arithmetic-compressed coefficients.
+ */
+
+METHODDEF(boolean)
+decode_mcu (j_decompress_ptr cinfo, JBLOCKARRAY MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  jpeg_component_info * compptr;
+  JBLOCKROW block;
+  unsigned char *st;
+  int blkn, ci, tbl, sign, k;
+  int v, m;
+  const int * natural_order;
+
+  /* Process restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0)
+      process_restart(cinfo);
+    entropy->restarts_to_go--;
+  }
+
+  if (entropy->ct == -1) return TRUE;	/* if error do nothing */
+
+  natural_order = cinfo->natural_order;
+
+  /* Outer loop handles each block in the MCU */
+
+  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+    block = MCU_data[blkn];
+    ci = cinfo->MCU_membership[blkn];
+    compptr = cinfo->cur_comp_info[ci];
+
+    /* Sections F.2.4.1 & F.1.4.4.1: Decoding of DC coefficients */
+
+    tbl = compptr->dc_tbl_no;
+
+    /* Table F.4: Point to statistics bin S0 for DC coefficient coding */
+    st = entropy->dc_stats[tbl] + entropy->dc_context[ci];
+
+    /* Figure F.19: Decode_DC_DIFF */
+    if (arith_decode(cinfo, st) == 0)
+      entropy->dc_context[ci] = 0;
+    else {
+      /* Figure F.21: Decoding nonzero value v */
+      /* Figure F.22: Decoding the sign of v */
+      sign = arith_decode(cinfo, st + 1);
+      st += 2; st += sign;
+      /* Figure F.23: Decoding the magnitude category of v */
+      if ((m = arith_decode(cinfo, st)) != 0) {
+	st = entropy->dc_stats[tbl] + 20;	/* Table F.4: X1 = 20 */
+	while (arith_decode(cinfo, st)) {
+	  if ((m <<= 1) == (int) 0x8000U) {
+	    WARNMS(cinfo, JWRN_ARITH_BAD_CODE);
+	    entropy->ct = -1;			/* magnitude overflow */
+	    return TRUE;
+	  }
+	  st += 1;
+	}
+      }
+      /* Section F.1.4.4.1.2: Establish dc_context conditioning category */
+      if (m < (int) ((1L << cinfo->arith_dc_L[tbl]) >> 1))
+	entropy->dc_context[ci] = 0;		   /* zero diff category */
+      else if (m > (int) ((1L << cinfo->arith_dc_U[tbl]) >> 1))
+	entropy->dc_context[ci] = 12 + (sign * 4); /* large diff category */
+      else
+	entropy->dc_context[ci] = 4 + (sign * 4);  /* small diff category */
+      v = m;
+      /* Figure F.24: Decoding the magnitude bit pattern of v */
+      st += 14;
+      while (m >>= 1)
+	if (arith_decode(cinfo, st)) v |= m;
+      v += 1; if (sign) v = -v;
+      entropy->last_dc_val[ci] += v;
+    }
+
+    (*block)[0] = (JCOEF) entropy->last_dc_val[ci];
+
+    /* Sections F.2.4.2 & F.1.4.4.2: Decoding of AC coefficients */
+
+    if (cinfo->lim_Se == 0) continue;
+    tbl = compptr->ac_tbl_no;
+    k = 0;
+
+    /* Figure F.20: Decode_AC_coefficients */
+    do {
+      st = entropy->ac_stats[tbl] + 3 * k;
+      if (arith_decode(cinfo, st)) break;	/* EOB flag */
+      for (;;) {
+	k++;
+	if (arith_decode(cinfo, st + 1)) break;
+	st += 3;
+	if (k >= cinfo->lim_Se) {
+	  WARNMS(cinfo, JWRN_ARITH_BAD_CODE);
+	  entropy->ct = -1;			/* spectral overflow */
+	  return TRUE;
+	}
+      }
+      /* Figure F.21: Decoding nonzero value v */
+      /* Figure F.22: Decoding the sign of v */
+      sign = arith_decode(cinfo, entropy->fixed_bin);
+      st += 2;
+      /* Figure F.23: Decoding the magnitude category of v */
+      if ((m = arith_decode(cinfo, st)) != 0) {
+	if (arith_decode(cinfo, st)) {
+	  m <<= 1;
+	  st = entropy->ac_stats[tbl] +
+	       (k <= cinfo->arith_ac_K[tbl] ? 189 : 217);
+	  while (arith_decode(cinfo, st)) {
+	    if ((m <<= 1) == (int) 0x8000U) {
+	      WARNMS(cinfo, JWRN_ARITH_BAD_CODE);
+	      entropy->ct = -1;			/* magnitude overflow */
+	      return TRUE;
+	    }
+	    st += 1;
+	  }
+	}
+      }
+      v = m;
+      /* Figure F.24: Decoding the magnitude bit pattern of v */
+      st += 14;
+      while (m >>= 1)
+	if (arith_decode(cinfo, st)) v |= m;
+      v += 1; if (sign) v = -v;
+      (*block)[natural_order[k]] = (JCOEF) v;
+    } while (k < cinfo->lim_Se);
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * Initialize for an arithmetic-compressed scan.
+ */
+
+METHODDEF(void)
+start_pass (j_decompress_ptr cinfo)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  int ci, tbl;
+  jpeg_component_info * compptr;
+
+  if (cinfo->progressive_mode) {
+    /* Validate progressive scan parameters */
+    if (cinfo->Ss == 0) {
+      if (cinfo->Se != 0)
+	goto bad;
+    } else {
+      /* need not check Ss/Se < 0 since they came from unsigned bytes */
+      if (cinfo->Se < cinfo->Ss || cinfo->Se > cinfo->lim_Se)
+	goto bad;
+      /* AC scans may have only one component */
+      if (cinfo->comps_in_scan != 1)
+	goto bad;
+    }
+    if (cinfo->Ah != 0) {
+      /* Successive approximation refinement scan: must have Al = Ah-1. */
+      if (cinfo->Ah-1 != cinfo->Al)
+	goto bad;
+    }
+    if (cinfo->Al > 13) {	/* need not check for < 0 */
+      bad:
+      ERREXIT4(cinfo, JERR_BAD_PROGRESSION,
+	       cinfo->Ss, cinfo->Se, cinfo->Ah, cinfo->Al);
+    }
+    /* Update progression status, and verify that scan order is legal.
+     * Note that inter-scan inconsistencies are treated as warnings
+     * not fatal errors ... not clear if this is right way to behave.
+     */
+    for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+      int coefi, cindex = cinfo->cur_comp_info[ci]->component_index;
+      int *coef_bit_ptr = & cinfo->coef_bits[cindex][0];
+      if (cinfo->Ss && coef_bit_ptr[0] < 0) /* AC without prior DC scan */
+	WARNMS2(cinfo, JWRN_BOGUS_PROGRESSION, cindex, 0);
+      for (coefi = cinfo->Ss; coefi <= cinfo->Se; coefi++) {
+	int expected = (coef_bit_ptr[coefi] < 0) ? 0 : coef_bit_ptr[coefi];
+	if (cinfo->Ah != expected)
+	  WARNMS2(cinfo, JWRN_BOGUS_PROGRESSION, cindex, coefi);
+	coef_bit_ptr[coefi] = cinfo->Al;
+      }
+    }
+    /* Select MCU decoding routine */
+    if (cinfo->Ah == 0) {
+      if (cinfo->Ss == 0)
+	entropy->pub.decode_mcu = decode_mcu_DC_first;
+      else
+	entropy->pub.decode_mcu = decode_mcu_AC_first;
+    } else {
+      if (cinfo->Ss == 0)
+	entropy->pub.decode_mcu = decode_mcu_DC_refine;
+      else
+	entropy->pub.decode_mcu = decode_mcu_AC_refine;
+    }
+  } else {
+    /* Check that the scan parameters Ss, Se, Ah/Al are OK for sequential JPEG.
+     * This ought to be an error condition, but we make it a warning.
+     */
+    if (cinfo->Ss != 0 || cinfo->Ah != 0 || cinfo->Al != 0 ||
+	(cinfo->Se < DCTSIZE2 && cinfo->Se != cinfo->lim_Se))
+      WARNMS(cinfo, JWRN_NOT_SEQUENTIAL);
+    /* Select MCU decoding routine */
+    entropy->pub.decode_mcu = decode_mcu;
+  }
+
+  /* Allocate & initialize requested statistics areas */
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+    compptr = cinfo->cur_comp_info[ci];
+    if (! cinfo->progressive_mode || (cinfo->Ss == 0 && cinfo->Ah == 0)) {
+      tbl = compptr->dc_tbl_no;
+      if (tbl < 0 || tbl >= NUM_ARITH_TBLS)
+	ERREXIT1(cinfo, JERR_NO_ARITH_TABLE, tbl);
+      if (entropy->dc_stats[tbl] == NULL)
+	entropy->dc_stats[tbl] = (unsigned char *) (*cinfo->mem->alloc_small)
+	  ((j_common_ptr) cinfo, JPOOL_IMAGE, DC_STAT_BINS);
+      MEMZERO(entropy->dc_stats[tbl], DC_STAT_BINS);
+      /* Initialize DC predictions to 0 */
+      entropy->last_dc_val[ci] = 0;
+      entropy->dc_context[ci] = 0;
+    }
+    if ((! cinfo->progressive_mode && cinfo->lim_Se) ||
+	(cinfo->progressive_mode && cinfo->Ss)) {
+      tbl = compptr->ac_tbl_no;
+      if (tbl < 0 || tbl >= NUM_ARITH_TBLS)
+	ERREXIT1(cinfo, JERR_NO_ARITH_TABLE, tbl);
+      if (entropy->ac_stats[tbl] == NULL)
+	entropy->ac_stats[tbl] = (unsigned char *) (*cinfo->mem->alloc_small)
+	  ((j_common_ptr) cinfo, JPOOL_IMAGE, AC_STAT_BINS);
+      MEMZERO(entropy->ac_stats[tbl], AC_STAT_BINS);
+    }
+  }
+
+  /* Initialize arithmetic decoding variables */
+  entropy->c = 0;
+  entropy->a = 0;
+  entropy->ct = -16;	/* force reading 2 initial bytes to fill C */
+
+  /* Initialize restart counter */
+  entropy->restarts_to_go = cinfo->restart_interval;
+}
+
+
+/*
+ * Finish up at the end of an arithmetic-compressed scan.
+ */
+
+METHODDEF(void)
+finish_pass (j_decompress_ptr cinfo)
+{
+  /* no work necessary here */
+}
+
+
+/*
+ * Module initialization routine for arithmetic entropy decoding.
+ */
+
+GLOBAL(void)
+jinit_arith_decoder (j_decompress_ptr cinfo)
+{
+  arith_entropy_ptr entropy;
+  int i;
+
+  entropy = (arith_entropy_ptr) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, SIZEOF(arith_entropy_decoder));
+  cinfo->entropy = &entropy->pub;
+  entropy->pub.start_pass = start_pass;
+  entropy->pub.finish_pass = finish_pass;
+
+  /* Mark tables unallocated */
+  for (i = 0; i < NUM_ARITH_TBLS; i++) {
+    entropy->dc_stats[i] = NULL;
+    entropy->ac_stats[i] = NULL;
+  }
+
+  /* Initialize index for fixed probability estimation */
+  entropy->fixed_bin[0] = 113;
+
+  if (cinfo->progressive_mode) {
+    /* Create progression status table */
+    int *coef_bit_ptr, ci;
+    cinfo->coef_bits = (int (*)[DCTSIZE2]) (*cinfo->mem->alloc_small)
+      ((j_common_ptr) cinfo, JPOOL_IMAGE,
+       cinfo->num_components * DCTSIZE2 * SIZEOF(int));
+    coef_bit_ptr = & cinfo->coef_bits[0][0];
+    for (ci = 0; ci < cinfo->num_components; ci++) 
+      for (i = 0; i < DCTSIZE2; i++)
+	*coef_bit_ptr++ = -1;
+  }
+}
@@ -0,0 +1,263 @@
+/*
+ * jdatadst.c
+ *
+ * Copyright (C) 1994-1996, Thomas G. Lane.
+ * Modified 2009-2022 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains compression data destination routines for the case of
+ * emitting JPEG data to memory or to a file (or any stdio stream).
+ * While these routines are sufficient for most applications,
+ * some will want to use a different destination manager.
+ * IMPORTANT: we assume that fwrite() will correctly transcribe an array of
+ * JOCTETs into 8-bit-wide elements on external storage.  If char is wider
+ * than 8 bits on your machine, you may need to do some tweaking.
+ */
+
+/* this is not a core library module, so it doesn't define JPEG_INTERNALS */
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jerror.h"
+
+#ifndef HAVE_STDLIB_H		/* <stdlib.h> should declare malloc(),free() */
+extern void * malloc JPP((size_t size));
+extern void free JPP((void *ptr));
+#endif
+
+
+/* Expanded data destination object for stdio output */
+
+#define OUTPUT_BUF_SIZE  4096	/* choose an efficiently fwrite'able size */
+
+typedef struct {
+  struct jpeg_destination_mgr pub; /* public fields */
+
+  FILE * outfile;		/* target stream */
+  JOCTET buffer[OUTPUT_BUF_SIZE]; /* output buffer */
+} my_destination_mgr;
+
+typedef my_destination_mgr * my_dest_ptr;
+
+
+/* Expanded data destination object for memory output */
+
+typedef struct {
+  struct jpeg_destination_mgr pub; /* public fields */
+
+  unsigned char ** outbuffer;	/* target buffer */
+  size_t * outsize;
+  unsigned char * newbuffer;	/* newly allocated buffer */
+  JOCTET * buffer;		/* start of buffer */
+  size_t bufsize;
+} my_mem_destination_mgr;
+
+typedef my_mem_destination_mgr * my_mem_dest_ptr;
+
+
+/*
+ * Initialize destination --- called by jpeg_start_compress
+ * before any data is actually written.
+ */
+
+METHODDEF(void)
+init_destination (j_compress_ptr cinfo)
+{
+  my_dest_ptr dest = (my_dest_ptr) cinfo->dest;
+
+  dest->pub.next_output_byte = dest->buffer;
+  dest->pub.free_in_buffer = OUTPUT_BUF_SIZE;
+}
+
+METHODDEF(void)
+init_mem_destination (j_compress_ptr cinfo)
+{
+  /* no work necessary here */
+}
+
+
+/*
+ * Empty the output buffer --- called whenever buffer fills up.
+ *
+ * In typical applications, this should write the entire output buffer
+ * (ignoring the current state of next_output_byte & free_in_buffer),
+ * reset the pointer & count to the start of the buffer, and return TRUE
+ * indicating that the buffer has been dumped.
+ *
+ * In applications that need to be able to suspend compression due to output
+ * overrun, a FALSE return indicates that the buffer cannot be emptied now.
+ * In this situation, the compressor will return to its caller (possibly with
+ * an indication that it has not accepted all the supplied scanlines).  The
+ * application should resume compression after it has made more room in the
+ * output buffer.  Note that there are substantial restrictions on the use of
+ * suspension --- see the documentation.
+ *
+ * When suspending, the compressor will back up to a convenient restart point
+ * (typically the start of the current MCU). next_output_byte & free_in_buffer
+ * indicate where the restart point will be if the current call returns FALSE.
+ * Data beyond this point will be regenerated after resumption, so do not
+ * write it out when emptying the buffer externally.
+ */
+
+METHODDEF(boolean)
+empty_output_buffer (j_compress_ptr cinfo)
+{
+  my_dest_ptr dest = (my_dest_ptr) cinfo->dest;
+
+  if (JFWRITE(dest->outfile, dest->buffer, OUTPUT_BUF_SIZE) !=
+      (size_t) OUTPUT_BUF_SIZE)
+    ERREXIT(cinfo, JERR_FILE_WRITE);
+
+  dest->pub.next_output_byte = dest->buffer;
+  dest->pub.free_in_buffer = OUTPUT_BUF_SIZE;
+
+  return TRUE;
+}
+
+METHODDEF(boolean)
+empty_mem_output_buffer (j_compress_ptr cinfo)
+{
+  size_t nextsize;
+  JOCTET * nextbuffer;
+  my_mem_dest_ptr dest = (my_mem_dest_ptr) cinfo->dest;
+
+  /* Try to allocate new buffer with double size */
+  nextsize = dest->bufsize * 2;
+  nextbuffer = (JOCTET *) malloc(nextsize);
+
+  if (nextbuffer == NULL)
+    ERREXIT1(cinfo, JERR_OUT_OF_MEMORY, 11);
+
+  MEMCOPY(nextbuffer, dest->buffer, dest->bufsize);
+
+  if (dest->newbuffer != NULL)
+    free(dest->newbuffer);
+
+  dest->newbuffer = nextbuffer;
+
+  dest->pub.next_output_byte = nextbuffer + dest->bufsize;
+  dest->pub.free_in_buffer = dest->bufsize;
+
+  dest->buffer = nextbuffer;
+  dest->bufsize = nextsize;
+
+  return TRUE;
+}
+
+
+/*
+ * Terminate destination --- called by jpeg_finish_compress
+ * after all data has been written.  Usually needs to flush buffer.
+ *
+ * NB: *not* called by jpeg_abort or jpeg_destroy; surrounding
+ * application must deal with any cleanup that should happen even
+ * for error exit.
+ */
+
+METHODDEF(void)
+term_destination (j_compress_ptr cinfo)
+{
+  my_dest_ptr dest = (my_dest_ptr) cinfo->dest;
+  size_t datacount = OUTPUT_BUF_SIZE - dest->pub.free_in_buffer;
+
+  /* Write any data remaining in the buffer */
+  if (datacount > 0) {
+    if (JFWRITE(dest->outfile, dest->buffer, datacount) != datacount)
+      ERREXIT(cinfo, JERR_FILE_WRITE);
+  }
+  JFFLUSH(dest->outfile);
+  /* Make sure we wrote the output file OK */
+  if (JFERROR(dest->outfile))
+    ERREXIT(cinfo, JERR_FILE_WRITE);
+}
+
+METHODDEF(void)
+term_mem_destination (j_compress_ptr cinfo)
+{
+  my_mem_dest_ptr dest = (my_mem_dest_ptr) cinfo->dest;
+
+  *dest->outbuffer = dest->buffer;
+  *dest->outsize = dest->bufsize - dest->pub.free_in_buffer;
+}
+
+
+/*
+ * Prepare for output to a stdio stream.
+ * The caller must have already opened the stream,
+ * and is responsible for closing it after finishing compression.
+ */
+
+GLOBAL(void)
+jpeg_stdio_dest (j_compress_ptr cinfo, FILE * outfile)
+{
+  my_dest_ptr dest;
+
+  /* The destination object is made permanent so that multiple JPEG images
+   * can be written to the same file without re-executing jpeg_stdio_dest.
+   * This makes it dangerous to use this manager and a different destination
+   * manager serially with the same JPEG object, because their private object
+   * sizes may be different.  Caveat programmer.
+   */
+  if (cinfo->dest == NULL) {	/* first time for this JPEG object? */
+    cinfo->dest = (struct jpeg_destination_mgr *) (*cinfo->mem->alloc_small)
+      ((j_common_ptr) cinfo, JPOOL_PERMANENT, SIZEOF(my_destination_mgr));
+  }
+
+  dest = (my_dest_ptr) cinfo->dest;
+  dest->pub.init_destination = init_destination;
+  dest->pub.empty_output_buffer = empty_output_buffer;
+  dest->pub.term_destination = term_destination;
+  dest->outfile = outfile;
+}
+
+
+/*
+ * Prepare for output to a memory buffer.
+ * The caller may supply an own initial buffer with appropriate size.
+ * Otherwise, or when the actual data output exceeds the given size,
+ * the library adapts the buffer size as necessary.
+ * The standard library functions malloc/free are used for allocating
+ * larger memory, so the buffer is available to the application after
+ * finishing compression, and then the application is responsible for
+ * freeing the requested memory.
+ * Note:  An initial buffer supplied by the caller is expected to be
+ * managed by the application.  The library does not free such buffer
+ * when allocating a larger buffer.
+ */
+
+GLOBAL(void)
+jpeg_mem_dest (j_compress_ptr cinfo,
+	       unsigned char ** outbuffer, size_t * outsize)
+{
+  my_mem_dest_ptr dest;
+
+  if (outbuffer == NULL || outsize == NULL)	/* sanity check */
+    ERREXIT(cinfo, JERR_BUFFER_SIZE);
+
+  /* The destination object is made permanent so that multiple JPEG images
+   * can be written to the same buffer without re-executing jpeg_mem_dest.
+   */
+  if (cinfo->dest == NULL) {	/* first time for this JPEG object? */
+    cinfo->dest = (struct jpeg_destination_mgr *) (*cinfo->mem->alloc_small)
+      ((j_common_ptr) cinfo, JPOOL_PERMANENT, SIZEOF(my_mem_destination_mgr));
+  }
+
+  dest = (my_mem_dest_ptr) cinfo->dest;
+  dest->pub.init_destination = init_mem_destination;
+  dest->pub.empty_output_buffer = empty_mem_output_buffer;
+  dest->pub.term_destination = term_mem_destination;
+  dest->outbuffer = outbuffer;
+  dest->outsize = outsize;
+  dest->newbuffer = NULL;
+
+  if (*outbuffer == NULL || *outsize == 0) {
+    /* Allocate initial buffer */
+    dest->newbuffer = *outbuffer = (unsigned char *) malloc(OUTPUT_BUF_SIZE);
+    if (dest->newbuffer == NULL)
+      ERREXIT1(cinfo, JERR_OUT_OF_MEMORY, 10);
+    *outsize = OUTPUT_BUF_SIZE;
+  }
+
+  dest->pub.next_output_byte = dest->buffer = *outbuffer;
+  dest->pub.free_in_buffer = dest->bufsize = *outsize;
+}
@@ -0,0 +1,271 @@
+/*
+ * jdatasrc.c
+ *
+ * Copyright (C) 1994-1996, Thomas G. Lane.
+ * Modified 2009-2022 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains decompression data source routines for the case of
+ * reading JPEG data from memory or from a file (or any stdio stream).
+ * While these routines are sufficient for most applications,
+ * some will want to use a different source manager.
+ * IMPORTANT: we assume that fread() will correctly transcribe an array of
+ * JOCTETs from 8-bit-wide elements on external storage.  If char is wider
+ * than 8 bits on your machine, you may need to do some tweaking.
+ */
+
+/* this is not a core library module, so it doesn't define JPEG_INTERNALS */
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jerror.h"
+
+
+/* Expanded data source object for stdio input */
+
+#define INPUT_BUF_SIZE  4096	/* choose an efficiently fread'able size */
+
+typedef struct {
+  struct jpeg_source_mgr pub;	/* public fields */
+
+  FILE * infile;		/* source stream */
+  JOCTET buffer[INPUT_BUF_SIZE]; /* input buffer */
+  boolean start_of_file;	/* have we gotten any data yet? */
+} my_source_mgr;
+
+typedef my_source_mgr * my_src_ptr;
+
+
+/*
+ * Initialize source --- called by jpeg_read_header
+ * before any data is actually read.
+ */
+
+METHODDEF(void)
+init_source (j_decompress_ptr cinfo)
+{
+  my_src_ptr src = (my_src_ptr) cinfo->src;
+
+  /* We reset the empty-input-file flag for each image,
+   * but we don't clear the input buffer.
+   * This is correct behavior for reading a series of images from one source.
+   */
+  src->start_of_file = TRUE;
+}
+
+METHODDEF(void)
+init_mem_source (j_decompress_ptr cinfo)
+{
+  /* no work necessary here */
+}
+
+
+/*
+ * Fill the input buffer --- called whenever buffer is emptied.
+ *
+ * In typical applications, this should read fresh data into the buffer
+ * (ignoring the current state of next_input_byte & bytes_in_buffer),
+ * reset the pointer & count to the start of the buffer, and return TRUE
+ * indicating that the buffer has been reloaded.  It is not necessary to
+ * fill the buffer entirely, only to obtain at least one more byte.
+ *
+ * There is no such thing as an EOF return.  If the end of the file has been
+ * reached, the routine has a choice of ERREXIT() or inserting fake data into
+ * the buffer.  In most cases, generating a warning message and inserting a
+ * fake EOI marker is the best course of action --- this will allow the
+ * decompressor to output however much of the image is there.  However,
+ * the resulting error message is misleading if the real problem is an empty
+ * input file, so we handle that case specially.
+ *
+ * In applications that need to be able to suspend compression due to input
+ * not being available yet, a FALSE return indicates that no more data can be
+ * obtained right now, but more may be forthcoming later.  In this situation,
+ * the decompressor will return to its caller (with an indication of the
+ * number of scanlines it has read, if any).  The application should resume
+ * decompression after it has loaded more data into the input buffer.  Note
+ * that there are substantial restrictions on the use of suspension --- see
+ * the documentation.
+ *
+ * When suspending, the decompressor will back up to a convenient restart point
+ * (typically the start of the current MCU). next_input_byte & bytes_in_buffer
+ * indicate where the restart point will be if the current call returns FALSE.
+ * Data beyond this point must be rescanned after resumption, so move it to
+ * the front of the buffer rather than discarding it.
+ */
+
+METHODDEF(boolean)
+fill_input_buffer (j_decompress_ptr cinfo)
+{
+  my_src_ptr src = (my_src_ptr) cinfo->src;
+  size_t nbytes;
+
+  nbytes = JFREAD(src->infile, src->buffer, INPUT_BUF_SIZE);
+
+  if (nbytes <= 0) {
+    if (src->start_of_file)	/* Treat empty input file as fatal error */
+      ERREXIT(cinfo, JERR_INPUT_EMPTY);
+    WARNMS(cinfo, JWRN_JPEG_EOF);
+    /* Insert a fake EOI marker */
+    src->buffer[0] = (JOCTET) 0xFF;
+    src->buffer[1] = (JOCTET) JPEG_EOI;
+    nbytes = 2;
+  }
+
+  src->pub.next_input_byte = src->buffer;
+  src->pub.bytes_in_buffer = nbytes;
+  src->start_of_file = FALSE;
+
+  return TRUE;
+}
+
+METHODDEF(boolean)
+fill_mem_input_buffer (j_decompress_ptr cinfo)
+{
+  static const JOCTET mybuffer[4] = {
+    (JOCTET) 0xFF, (JOCTET) JPEG_EOI, 0, 0
+  };
+
+  /* The whole JPEG data is expected to reside in the supplied memory
+   * buffer, so any request for more data beyond the given buffer size
+   * is treated as an error.
+   */
+  WARNMS(cinfo, JWRN_JPEG_EOF);
+
+  /* Insert a fake EOI marker */
+
+  cinfo->src->next_input_byte = mybuffer;
+  cinfo->src->bytes_in_buffer = 2;
+
+  return TRUE;
+}
+
+
+/*
+ * Skip data --- used to skip over a potentially large amount of
+ * uninteresting data (such as an APPn marker).
+ *
+ * Writers of suspendable-input applications must note that skip_input_data
+ * is not granted the right to give a suspension return.  If the skip extends
+ * beyond the data currently in the buffer, the buffer can be marked empty so
+ * that the next read will cause a fill_input_buffer call that can suspend.
+ * Arranging for additional bytes to be discarded before reloading the input
+ * buffer is the application writer's problem.
+ */
+
+METHODDEF(void)
+skip_input_data (j_decompress_ptr cinfo, long num_bytes)
+{
+  struct jpeg_source_mgr * src = cinfo->src;
+  size_t nbytes;
+
+  /* Just a dumb implementation for now.  Could use fseek() except
+   * it doesn't work on pipes.  Not clear that being smart is worth
+   * any trouble anyway --- large skips are infrequent.
+   */
+  if (num_bytes > 0) {
+    nbytes = (size_t) num_bytes;
+    while (nbytes > src->bytes_in_buffer) {
+      nbytes -= src->bytes_in_buffer;
+      (void) (*src->fill_input_buffer) (cinfo);
+      /* note we assume that fill_input_buffer will never return FALSE,
+       * so suspension need not be handled.
+       */
+    }
+    src->next_input_byte += nbytes;
+    src->bytes_in_buffer -= nbytes;
+  }
+}
+
+
+/*
+ * An additional method that can be provided by data source modules is the
+ * resync_to_restart method for error recovery in the presence of RST markers.
+ * For the moment, this source module just uses the default resync method
+ * provided by the JPEG library.  That method assumes that no backtracking
+ * is possible.
+ */
+
+
+/*
+ * Terminate source --- called by jpeg_finish_decompress
+ * after all data has been read.  Often a no-op.
+ *
+ * NB: *not* called by jpeg_abort or jpeg_destroy; surrounding
+ * application must deal with any cleanup that should happen even
+ * for error exit.
+ */
+
+METHODDEF(void)
+term_source (j_decompress_ptr cinfo)
+{
+  /* no work necessary here */
+}
+
+
+/*
+ * Prepare for input from a stdio stream.
+ * The caller must have already opened the stream,
+ * and is responsible for closing it after finishing decompression.
+ */
+
+GLOBAL(void)
+jpeg_stdio_src (j_decompress_ptr cinfo, FILE * infile)
+{
+  my_src_ptr src;
+
+  /* The source object including the input buffer is made permanent so that
+   * a series of JPEG images can be read from the same file by calling
+   * jpeg_stdio_src only before the first one.  (If we discarded the buffer
+   * at the end of one image, we'd likely lose the start of the next one.)
+   * This makes it unsafe to use this manager and a different source
+   * manager serially with the same JPEG object.  Caveat programmer.
+   */
+  if (cinfo->src == NULL) {	/* first time for this JPEG object? */
+    cinfo->src = (struct jpeg_source_mgr *) (*cinfo->mem->alloc_small)
+      ((j_common_ptr) cinfo, JPOOL_PERMANENT, SIZEOF(my_source_mgr));
+  }
+
+  src = (my_src_ptr) cinfo->src;
+  src->pub.init_source = init_source;
+  src->pub.fill_input_buffer = fill_input_buffer;
+  src->pub.skip_input_data = skip_input_data;
+  src->pub.resync_to_restart = jpeg_resync_to_restart; /* use default method */
+  src->pub.term_source = term_source;
+  src->infile = infile;
+  src->pub.bytes_in_buffer = 0; /* forces fill_input_buffer on first read */
+  src->pub.next_input_byte = NULL; /* until buffer loaded */
+}
+
+
+/*
+ * Prepare for input from a supplied memory buffer.
+ * The buffer must contain the whole JPEG data.
+ */
+
+GLOBAL(void)
+jpeg_mem_src (j_decompress_ptr cinfo,
+	      const unsigned char * inbuffer, size_t insize)
+{
+  struct jpeg_source_mgr * src;
+
+  if (inbuffer == NULL || insize == 0)	/* Treat empty input as fatal error */
+    ERREXIT(cinfo, JERR_INPUT_EMPTY);
+
+  /* The source object is made permanent so that a series of JPEG images
+   * can be read from the same buffer by calling jpeg_mem_src only before
+   * the first one.
+   */
+  if (cinfo->src == NULL) {	/* first time for this JPEG object? */
+    cinfo->src = (struct jpeg_source_mgr *) (*cinfo->mem->alloc_small)
+      ((j_common_ptr) cinfo, JPOOL_PERMANENT, SIZEOF(struct jpeg_source_mgr));
+  }
+
+  src = cinfo->src;
+  src->init_source = init_mem_source;
+  src->fill_input_buffer = fill_mem_input_buffer;
+  src->skip_input_data = skip_input_data;
+  src->resync_to_restart = jpeg_resync_to_restart; /* use default method */
+  src->term_source = term_source;
+  src->bytes_in_buffer = insize;
+  src->next_input_byte = (const JOCTET *) inbuffer;
+}
@@ -0,0 +1,744 @@
+/*
+ * jdcoefct.c
+ *
+ * Copyright (C) 1994-1997, Thomas G. Lane.
+ * Modified 2002-2020 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains the coefficient buffer controller for decompression.
+ * This controller is the top level of the JPEG decompressor proper.
+ * The coefficient buffer lies between entropy decoding and inverse-DCT steps.
+ *
+ * In buffered-image mode, this controller is the interface between
+ * input-oriented processing and output-oriented processing.
+ * Also, the input side (only) is used when reading a file for transcoding.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/* Block smoothing is only applicable for progressive JPEG, so: */
+#ifndef D_PROGRESSIVE_SUPPORTED
+#undef BLOCK_SMOOTHING_SUPPORTED
+#endif
+
+
+/* Private buffer controller object */
+
+typedef struct {
+  struct jpeg_d_coef_controller pub; /* public fields */
+
+  /* These variables keep track of the current location of the input side. */
+  /* cinfo->input_iMCU_row is also used for this. */
+  JDIMENSION MCU_ctr;		/* counts MCUs processed in current row */
+  int MCU_vert_offset;		/* counts MCU rows within iMCU row */
+  int MCU_rows_per_iMCU_row;	/* number of such rows needed */
+
+  /* The output side's location is represented by cinfo->output_iMCU_row. */
+
+  /* In single-pass modes, it's sufficient to buffer just one MCU.
+   * We append a workspace of D_MAX_BLOCKS_IN_MCU coefficient blocks,
+   * and let the entropy decoder write into that workspace each time.
+   * In multi-pass modes, this array points to the current MCU's blocks
+   * within the virtual arrays; it is used only by the input side.
+   */
+  JBLOCKROW MCU_buffer[D_MAX_BLOCKS_IN_MCU];
+
+#ifdef D_MULTISCAN_FILES_SUPPORTED
+  /* In multi-pass modes, we need a virtual block array for each component. */
+  jvirt_barray_ptr whole_image[MAX_COMPONENTS];
+#endif
+
+#ifdef BLOCK_SMOOTHING_SUPPORTED
+  /* When doing block smoothing, we latch coefficient Al values here */
+  int * coef_bits_latch;
+#define SAVED_COEFS  6		/* we save coef_bits[0..5] */
+#endif
+
+  /* Workspace for single-pass modes (omitted otherwise). */
+  JBLOCK blk_buffer[D_MAX_BLOCKS_IN_MCU];
+} my_coef_controller;
+
+typedef my_coef_controller * my_coef_ptr;
+
+
+/* Forward declarations */
+METHODDEF(int) decompress_onepass
+	JPP((j_decompress_ptr cinfo, JSAMPIMAGE output_buf));
+#ifdef D_MULTISCAN_FILES_SUPPORTED
+METHODDEF(int) decompress_data
+	JPP((j_decompress_ptr cinfo, JSAMPIMAGE output_buf));
+#endif
+#ifdef BLOCK_SMOOTHING_SUPPORTED
+LOCAL(boolean) smoothing_ok JPP((j_decompress_ptr cinfo));
+METHODDEF(int) decompress_smooth_data
+	JPP((j_decompress_ptr cinfo, JSAMPIMAGE output_buf));
+#endif
+
+
+LOCAL(void)
+start_iMCU_row (j_decompress_ptr cinfo)
+/* Reset within-iMCU-row counters for a new row (input side) */
+{
+  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
+
+  /* In an interleaved scan, an MCU row is the same as an iMCU row.
+   * In a noninterleaved scan, an iMCU row has v_samp_factor MCU rows.
+   * But at the bottom of the image, process only what's left.
+   */
+  if (cinfo->comps_in_scan > 1) {
+    coef->MCU_rows_per_iMCU_row = 1;
+  } else {
+    if (cinfo->input_iMCU_row < (cinfo->total_iMCU_rows-1))
+      coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->v_samp_factor;
+    else
+      coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->last_row_height;
+  }
+
+  coef->MCU_ctr = 0;
+  coef->MCU_vert_offset = 0;
+}
+
+
+/*
+ * Initialize for an input processing pass.
+ */
+
+METHODDEF(void)
+start_input_pass (j_decompress_ptr cinfo)
+{
+  cinfo->input_iMCU_row = 0;
+  start_iMCU_row(cinfo);
+}
+
+
+/*
+ * Initialize for an output processing pass.
+ */
+
+METHODDEF(void)
+start_output_pass (j_decompress_ptr cinfo)
+{
+#ifdef BLOCK_SMOOTHING_SUPPORTED
+  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
+
+  /* If multipass, check to see whether to use block smoothing on this pass */
+  if (coef->pub.coef_arrays != NULL) {
+    if (cinfo->do_block_smoothing && smoothing_ok(cinfo))
+      coef->pub.decompress_data = decompress_smooth_data;
+    else
+      coef->pub.decompress_data = decompress_data;
+  }
+#endif
+  cinfo->output_iMCU_row = 0;
+}
+
+
+/*
+ * Decompress and return some data in the single-pass case.
+ * Always attempts to emit one fully interleaved MCU row ("iMCU" row).
+ * Input and output must run in lockstep since we have only a one-MCU buffer.
+ * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED.
+ *
+ * NB: output_buf contains a plane for each component in image,
+ * which we index according to the component's SOF position.
+ */
+
+METHODDEF(int)
+decompress_onepass (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
+{
+  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
+  JDIMENSION MCU_col_num;	/* index of current MCU within row */
+  JDIMENSION last_MCU_col = cinfo->MCUs_per_row - 1;
+  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
+  int ci, xindex, yindex, yoffset, useful_width;
+  JBLOCKROW blkp;
+  JSAMPARRAY output_ptr;
+  JDIMENSION start_col, output_col;
+  jpeg_component_info *compptr;
+  inverse_DCT_method_ptr inverse_DCT;
+
+  /* Loop to process as much as one whole iMCU row */
+  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
+       yoffset++) {
+    for (MCU_col_num = coef->MCU_ctr; MCU_col_num <= last_MCU_col;
+	 MCU_col_num++) {
+      blkp = coef->blk_buffer;	/* pointer to current DCT block within MCU */
+      /* Try to fetch an MCU.  Entropy decoder expects buffer to be zeroed. */
+      if (cinfo->lim_Se)	/* can bypass in DC only case */
+	MEMZERO(blkp, cinfo->blocks_in_MCU * SIZEOF(JBLOCK));
+      if (! (*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) {
+	/* Suspension forced; update state counters and exit */
+	coef->MCU_vert_offset = yoffset;
+	coef->MCU_ctr = MCU_col_num;
+	return JPEG_SUSPENDED;
+      }
+      /* Determine where data should go in output_buf and do the IDCT thing.
+       * We skip dummy blocks at the right and bottom edges (but blkp gets
+       * incremented past them!).
+       */
+      for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+	compptr = cinfo->cur_comp_info[ci];
+	/* Don't bother to IDCT an uninteresting component. */
+	if (! compptr->component_needed) {
+	  blkp += compptr->MCU_blocks;
+	  continue;
+	}
+	inverse_DCT = cinfo->idct->inverse_DCT[compptr->component_index];
+	output_ptr = output_buf[compptr->component_index] +
+	  yoffset * compptr->DCT_v_scaled_size;
+	useful_width = (MCU_col_num < last_MCU_col) ? compptr->MCU_width
+						    : compptr->last_col_width;
+	start_col = MCU_col_num * compptr->MCU_sample_width;
+	for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
+	  if (cinfo->input_iMCU_row < last_iMCU_row ||
+	      yoffset + yindex < compptr->last_row_height) {
+	    output_col = start_col;
+	    for (xindex = 0; xindex < useful_width; xindex++) {
+	      (*inverse_DCT) (cinfo, compptr, (JCOEFPTR) (blkp + xindex),
+			      output_ptr, output_col);
+	      output_col += compptr->DCT_h_scaled_size;
+	    }
+	    output_ptr += compptr->DCT_v_scaled_size;
+	  }
+	  blkp += compptr->MCU_width;
+	}
+      }
+    }
+    /* Completed an MCU row, but perhaps not an iMCU row */
+    coef->MCU_ctr = 0;
+  }
+  /* Completed the iMCU row, advance counters for next one */
+  cinfo->output_iMCU_row++;
+  if (++(cinfo->input_iMCU_row) <= last_iMCU_row) {
+    start_iMCU_row(cinfo);
+    return JPEG_ROW_COMPLETED;
+  }
+  /* Completed the scan */
+  (*cinfo->inputctl->finish_input_pass) (cinfo);
+  return JPEG_SCAN_COMPLETED;
+}
+
+
+/*
+ * Dummy consume-input routine for single-pass operation.
+ */
+
+METHODDEF(int)
+dummy_consume_data (j_decompress_ptr cinfo)
+{
+  return JPEG_SUSPENDED;	/* Always indicate nothing was done */
+}
+
+
+#ifdef D_MULTISCAN_FILES_SUPPORTED
+
+/*
+ * Consume input data and store it in the full-image coefficient buffer.
+ * We read as much as one fully interleaved MCU row ("iMCU" row) per call,
+ * ie, v_samp_factor block rows for each component in the scan.
+ * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED.
+ */
+
+METHODDEF(int)
+consume_data (j_decompress_ptr cinfo)
+{
+  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
+  JDIMENSION MCU_col_num;	/* index of current MCU within row */
+  int ci, xindex, yindex, yoffset;
+  JDIMENSION start_col;
+  JBLOCKARRAY blkp;
+  JBLOCKARRAY buffer[MAX_COMPS_IN_SCAN];
+  JBLOCKROW buffer_ptr;
+  jpeg_component_info *compptr;
+
+  /* Align the virtual buffers for the components used in this scan. */
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+    compptr = cinfo->cur_comp_info[ci];
+    buffer[ci] = (*cinfo->mem->access_virt_barray)
+      ((j_common_ptr) cinfo, coef->whole_image[compptr->component_index],
+       cinfo->input_iMCU_row * compptr->v_samp_factor,
+       (JDIMENSION) compptr->v_samp_factor, TRUE);
+    /* Note: entropy decoder expects buffer to be zeroed,
+     * but this is handled automatically by the memory manager
+     * because we requested a pre-zeroed array.
+     */
+  }
+
+  /* Loop to process one whole iMCU row */
+  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
+       yoffset++) {
+    for (MCU_col_num = coef->MCU_ctr; MCU_col_num < cinfo->MCUs_per_row;
+	 MCU_col_num++) {
+      /* Construct list of pointers to DCT blocks belonging to this MCU */
+      blkp = coef->MCU_buffer;	/* pointer to current DCT block within MCU */
+      for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+	compptr = cinfo->cur_comp_info[ci];
+	start_col = MCU_col_num * compptr->MCU_width;
+	for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
+	  buffer_ptr = buffer[ci][yoffset + yindex] + start_col;
+	  xindex = compptr->MCU_width;
+	  do {
+	    *blkp++ = buffer_ptr++;
+	  } while (--xindex);
+	}
+      }
+      /* Try to fetch the MCU. */
+      if (! (*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) {
+	/* Suspension forced; update state counters and exit */
+	coef->MCU_vert_offset = yoffset;
+	coef->MCU_ctr = MCU_col_num;
+	return JPEG_SUSPENDED;
+      }
+    }
+    /* Completed an MCU row, but perhaps not an iMCU row */
+    coef->MCU_ctr = 0;
+  }
+  /* Completed the iMCU row, advance counters for next one */
+  if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) {
+    start_iMCU_row(cinfo);
+    return JPEG_ROW_COMPLETED;
+  }
+  /* Completed the scan */
+  (*cinfo->inputctl->finish_input_pass) (cinfo);
+  return JPEG_SCAN_COMPLETED;
+}
+
+
+/*
+ * Decompress and return some data in the multi-pass case.
+ * Always attempts to emit one fully interleaved MCU row ("iMCU" row).
+ * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED.
+ *
+ * NB: output_buf contains a plane for each component in image.
+ */
+
+METHODDEF(int)
+decompress_data (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
+{
+  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
+  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
+  JDIMENSION block_num;
+  int ci, block_row, block_rows;
+  JBLOCKARRAY buffer;
+  JBLOCKROW buffer_ptr;
+  JSAMPARRAY output_ptr;
+  JDIMENSION output_col;
+  jpeg_component_info *compptr;
+  inverse_DCT_method_ptr inverse_DCT;
+
+  /* Force some input to be done if we are getting ahead of the input. */
+  while (cinfo->input_scan_number < cinfo->output_scan_number ||
+	 (cinfo->input_scan_number == cinfo->output_scan_number &&
+	  cinfo->input_iMCU_row <= cinfo->output_iMCU_row)) {
+    if ((*cinfo->inputctl->consume_input)(cinfo) == JPEG_SUSPENDED)
+      return JPEG_SUSPENDED;
+  }
+
+  /* OK, output from the virtual arrays. */
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    /* Don't bother to IDCT an uninteresting component. */
+    if (! compptr->component_needed)
+      continue;
+    /* Align the virtual buffer for this component. */
+    buffer = (*cinfo->mem->access_virt_barray)
+      ((j_common_ptr) cinfo, coef->whole_image[ci],
+       cinfo->output_iMCU_row * compptr->v_samp_factor,
+       (JDIMENSION) compptr->v_samp_factor, FALSE);
+    /* Count non-dummy DCT block rows in this iMCU row. */
+    if (cinfo->output_iMCU_row < last_iMCU_row)
+      block_rows = compptr->v_samp_factor;
+    else {
+      /* NB: can't use last_row_height here; it is input-side-dependent! */
+      block_rows = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
+      if (block_rows == 0) block_rows = compptr->v_samp_factor;
+    }
+    inverse_DCT = cinfo->idct->inverse_DCT[ci];
+    output_ptr = output_buf[ci];
+    /* Loop over all DCT blocks to be processed. */
+    for (block_row = 0; block_row < block_rows; block_row++) {
+      buffer_ptr = buffer[block_row];
+      output_col = 0;
+      for (block_num = 0; block_num < compptr->width_in_blocks; block_num++) {
+	(*inverse_DCT) (cinfo, compptr, (JCOEFPTR) buffer_ptr,
+			output_ptr, output_col);
+	buffer_ptr++;
+	output_col += compptr->DCT_h_scaled_size;
+      }
+      output_ptr += compptr->DCT_v_scaled_size;
+    }
+  }
+
+  if (++(cinfo->output_iMCU_row) <= last_iMCU_row)
+    return JPEG_ROW_COMPLETED;
+  return JPEG_SCAN_COMPLETED;
+}
+
+#endif /* D_MULTISCAN_FILES_SUPPORTED */
+
+
+#ifdef BLOCK_SMOOTHING_SUPPORTED
+
+/*
+ * This code applies interblock smoothing as described by section K.8
+ * of the JPEG standard: the first 5 AC coefficients are estimated from
+ * the DC values of a DCT block and its 8 neighboring blocks.
+ * We apply smoothing only for progressive JPEG decoding, and only if
+ * the coefficients it can estimate are not yet known to full precision.
+ */
+
+/* Natural-order array positions of the first 5 zigzag-order coefficients */
+#define Q01_POS  1
+#define Q10_POS  8
+#define Q20_POS  16
+#define Q11_POS  9
+#define Q02_POS  2
+
+/*
+ * Determine whether block smoothing is applicable and safe.
+ * We also latch the current states of the coef_bits[] entries for the
+ * AC coefficients; otherwise, if the input side of the decompressor
+ * advances into a new scan, we might think the coefficients are known
+ * more accurately than they really are.
+ */
+
+LOCAL(boolean)
+smoothing_ok (j_decompress_ptr cinfo)
+{
+  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
+  boolean smoothing_useful = FALSE;
+  int ci, coefi;
+  jpeg_component_info *compptr;
+  JQUANT_TBL * qtable;
+  int * coef_bits;
+  int * coef_bits_latch;
+
+  if (! cinfo->progressive_mode || cinfo->coef_bits == NULL)
+    return FALSE;
+
+  /* Allocate latch area if not already done */
+  if (coef->coef_bits_latch == NULL)
+    coef->coef_bits_latch = (int *) (*cinfo->mem->alloc_small)
+      ((j_common_ptr) cinfo, JPOOL_IMAGE,
+       cinfo->num_components * (SAVED_COEFS * SIZEOF(int)));
+  coef_bits_latch = coef->coef_bits_latch;
+
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    /* All components' quantization values must already be latched. */
+    if ((qtable = compptr->quant_table) == NULL)
+      return FALSE;
+    /* Verify DC & first 5 AC quantizers are nonzero to avoid zero-divide. */
+    if (qtable->quantval[0] == 0 ||
+	qtable->quantval[Q01_POS] == 0 ||
+	qtable->quantval[Q10_POS] == 0 ||
+	qtable->quantval[Q20_POS] == 0 ||
+	qtable->quantval[Q11_POS] == 0 ||
+	qtable->quantval[Q02_POS] == 0)
+      return FALSE;
+    /* DC values must be at least partly known for all components. */
+    coef_bits = cinfo->coef_bits[ci];
+    if (coef_bits[0] < 0)
+      return FALSE;
+    /* Block smoothing is helpful if some AC coefficients remain inaccurate. */
+    for (coefi = 1; coefi <= 5; coefi++) {
+      coef_bits_latch[coefi] = coef_bits[coefi];
+      if (coef_bits[coefi] != 0)
+	smoothing_useful = TRUE;
+    }
+    coef_bits_latch += SAVED_COEFS;
+  }
+
+  return smoothing_useful;
+}
+
+
+/*
+ * Variant of decompress_data for use when doing block smoothing.
+ */
+
+METHODDEF(int)
+decompress_smooth_data (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
+{
+  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
+  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
+  JDIMENSION block_num, last_block_column;
+  int ci, block_row, block_rows, access_rows;
+  JBLOCKARRAY buffer;
+  JBLOCKROW buffer_ptr, prev_block_row, next_block_row;
+  JSAMPARRAY output_ptr;
+  JDIMENSION output_col;
+  jpeg_component_info *compptr;
+  inverse_DCT_method_ptr inverse_DCT;
+  boolean first_row, last_row;
+  JBLOCK workspace;
+  int *coef_bits;
+  JQUANT_TBL *quanttbl;
+  INT32 Q00,Q01,Q02,Q10,Q11,Q20, num;
+  int DC1,DC2,DC3,DC4,DC5,DC6,DC7,DC8,DC9;
+  int Al, pred;
+
+  /* Force some input to be done if we are getting ahead of the input. */
+  while (cinfo->input_scan_number <= cinfo->output_scan_number &&
+	 ! cinfo->inputctl->eoi_reached) {
+    if (cinfo->input_scan_number == cinfo->output_scan_number) {
+      /* If input is working on current scan, we ordinarily want it to
+       * have completed the current row.  But if input scan is DC,
+       * we want it to keep one row ahead so that next block row's DC
+       * values are up to date.
+       */
+      JDIMENSION delta = (cinfo->Ss == 0) ? 1 : 0;
+      if (cinfo->input_iMCU_row > cinfo->output_iMCU_row+delta)
+	break;
+    }
+    if ((*cinfo->inputctl->consume_input)(cinfo) == JPEG_SUSPENDED)
+      return JPEG_SUSPENDED;
+  }
+
+  /* OK, output from the virtual arrays. */
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    /* Don't bother to IDCT an uninteresting component. */
+    if (! compptr->component_needed)
+      continue;
+    /* Count non-dummy DCT block rows in this iMCU row. */
+    if (cinfo->output_iMCU_row < last_iMCU_row) {
+      block_rows = compptr->v_samp_factor;
+      access_rows = block_rows * 2; /* this and next iMCU row */
+      last_row = FALSE;
+    } else {
+      /* NB: can't use last_row_height here; it is input-side-dependent! */
+      block_rows = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
+      if (block_rows == 0) block_rows = compptr->v_samp_factor;
+      access_rows = block_rows; /* this iMCU row only */
+      last_row = TRUE;
+    }
+    /* Align the virtual buffer for this component. */
+    if (cinfo->output_iMCU_row > 0) {
+      access_rows += compptr->v_samp_factor; /* prior iMCU row too */
+      buffer = (*cinfo->mem->access_virt_barray)
+	((j_common_ptr) cinfo, coef->whole_image[ci],
+	 (cinfo->output_iMCU_row - 1) * compptr->v_samp_factor,
+	 (JDIMENSION) access_rows, FALSE);
+      buffer += compptr->v_samp_factor;	/* point to current iMCU row */
+      first_row = FALSE;
+    } else {
+      buffer = (*cinfo->mem->access_virt_barray)
+	((j_common_ptr) cinfo, coef->whole_image[ci],
+	 (JDIMENSION) 0, (JDIMENSION) access_rows, FALSE);
+      first_row = TRUE;
+    }
+    /* Fetch component-dependent info */
+    coef_bits = coef->coef_bits_latch + (ci * SAVED_COEFS);
+    quanttbl = compptr->quant_table;
+    Q00 = quanttbl->quantval[0];
+    Q01 = quanttbl->quantval[Q01_POS];
+    Q10 = quanttbl->quantval[Q10_POS];
+    Q20 = quanttbl->quantval[Q20_POS];
+    Q11 = quanttbl->quantval[Q11_POS];
+    Q02 = quanttbl->quantval[Q02_POS];
+    inverse_DCT = cinfo->idct->inverse_DCT[ci];
+    output_ptr = output_buf[ci];
+    /* Loop over all DCT blocks to be processed. */
+    for (block_row = 0; block_row < block_rows; block_row++) {
+      buffer_ptr = buffer[block_row];
+      if (first_row && block_row == 0)
+	prev_block_row = buffer_ptr;
+      else
+	prev_block_row = buffer[block_row-1];
+      if (last_row && block_row == block_rows-1)
+	next_block_row = buffer_ptr;
+      else
+	next_block_row = buffer[block_row+1];
+      /* We fetch the surrounding DC values using a sliding-register approach.
+       * Initialize all nine here so as to do the right thing on narrow pics.
+       */
+      DC1 = DC2 = DC3 = (int) prev_block_row[0][0];
+      DC4 = DC5 = DC6 = (int) buffer_ptr[0][0];
+      DC7 = DC8 = DC9 = (int) next_block_row[0][0];
+      output_col = 0;
+      last_block_column = compptr->width_in_blocks - 1;
+      for (block_num = 0; block_num <= last_block_column; block_num++) {
+	/* Fetch current DCT block into workspace so we can modify it. */
+	jcopy_block_row(buffer_ptr, (JBLOCKROW) workspace, (JDIMENSION) 1);
+	/* Update DC values */
+	if (block_num < last_block_column) {
+	  DC3 = (int) prev_block_row[1][0];
+	  DC6 = (int) buffer_ptr[1][0];
+	  DC9 = (int) next_block_row[1][0];
+	}
+	/* Compute coefficient estimates per K.8.
+	 * An estimate is applied only if coefficient is still zero,
+	 * and is not known to be fully accurate.
+	 */
+	/* AC01 */
+	if ((Al=coef_bits[1]) != 0 && workspace[1] == 0) {
+	  num = 36 * Q00 * (DC4 - DC6);
+	  if (num >= 0) {
+	    pred = (int) (((Q01<<7) + num) / (Q01<<8));
+	    if (Al > 0 && pred >= (1<<Al))
+	      pred = (1<<Al)-1;
+	  } else {
+	    pred = (int) (((Q01<<7) - num) / (Q01<<8));
+	    if (Al > 0 && pred >= (1<<Al))
+	      pred = (1<<Al)-1;
+	    pred = -pred;
+	  }
+	  workspace[1] = (JCOEF) pred;
+	}
+	/* AC10 */
+	if ((Al=coef_bits[2]) != 0 && workspace[8] == 0) {
+	  num = 36 * Q00 * (DC2 - DC8);
+	  if (num >= 0) {
+	    pred = (int) (((Q10<<7) + num) / (Q10<<8));
+	    if (Al > 0 && pred >= (1<<Al))
+	      pred = (1<<Al)-1;
+	  } else {
+	    pred = (int) (((Q10<<7) - num) / (Q10<<8));
+	    if (Al > 0 && pred >= (1<<Al))
+	      pred = (1<<Al)-1;
+	    pred = -pred;
+	  }
+	  workspace[8] = (JCOEF) pred;
+	}
+	/* AC20 */
+	if ((Al=coef_bits[3]) != 0 && workspace[16] == 0) {
+	  num = 9 * Q00 * (DC2 + DC8 - 2*DC5);
+	  if (num >= 0) {
+	    pred = (int) (((Q20<<7) + num) / (Q20<<8));
+	    if (Al > 0 && pred >= (1<<Al))
+	      pred = (1<<Al)-1;
+	  } else {
+	    pred = (int) (((Q20<<7) - num) / (Q20<<8));
+	    if (Al > 0 && pred >= (1<<Al))
+	      pred = (1<<Al)-1;
+	    pred = -pred;
+	  }
+	  workspace[16] = (JCOEF) pred;
+	}
+	/* AC11 */
+	if ((Al=coef_bits[4]) != 0 && workspace[9] == 0) {
+	  num = 5 * Q00 * (DC1 - DC3 - DC7 + DC9);
+	  if (num >= 0) {
+	    pred = (int) (((Q11<<7) + num) / (Q11<<8));
+	    if (Al > 0 && pred >= (1<<Al))
+	      pred = (1<<Al)-1;
+	  } else {
+	    pred = (int) (((Q11<<7) - num) / (Q11<<8));
+	    if (Al > 0 && pred >= (1<<Al))
+	      pred = (1<<Al)-1;
+	    pred = -pred;
+	  }
+	  workspace[9] = (JCOEF) pred;
+	}
+	/* AC02 */
+	if ((Al=coef_bits[5]) != 0 && workspace[2] == 0) {
+	  num = 9 * Q00 * (DC4 + DC6 - 2*DC5);
+	  if (num >= 0) {
+	    pred = (int) (((Q02<<7) + num) / (Q02<<8));
+	    if (Al > 0 && pred >= (1<<Al))
+	      pred = (1<<Al)-1;
+	  } else {
+	    pred = (int) (((Q02<<7) - num) / (Q02<<8));
+	    if (Al > 0 && pred >= (1<<Al))
+	      pred = (1<<Al)-1;
+	    pred = -pred;
+	  }
+	  workspace[2] = (JCOEF) pred;
+	}
+	/* OK, do the IDCT */
+	(*inverse_DCT) (cinfo, compptr, (JCOEFPTR) workspace,
+			output_ptr, output_col);
+	/* Advance for next column */
+	DC1 = DC2; DC2 = DC3;
+	DC4 = DC5; DC5 = DC6;
+	DC7 = DC8; DC8 = DC9;
+	buffer_ptr++, prev_block_row++, next_block_row++;
+	output_col += compptr->DCT_h_scaled_size;
+      }
+      output_ptr += compptr->DCT_v_scaled_size;
+    }
+  }
+
+  if (++(cinfo->output_iMCU_row) <= last_iMCU_row)
+    return JPEG_ROW_COMPLETED;
+  return JPEG_SCAN_COMPLETED;
+}
+
+#endif /* BLOCK_SMOOTHING_SUPPORTED */
+
+
+/*
+ * Initialize coefficient buffer controller.
+ */
+
+GLOBAL(void)
+jinit_d_coef_controller (j_decompress_ptr cinfo, boolean need_full_buffer)
+{
+  my_coef_ptr coef;
+
+  if (need_full_buffer) {
+#ifdef D_MULTISCAN_FILES_SUPPORTED
+    /* Allocate a full-image virtual array for each component, */
+    /* padded to a multiple of samp_factor DCT blocks in each direction. */
+    /* Note we ask for a pre-zeroed array. */
+    int ci, access_rows;
+    jpeg_component_info *compptr;
+
+    coef = (my_coef_ptr) (*cinfo->mem->alloc_small)
+      ((j_common_ptr) cinfo, JPOOL_IMAGE,
+       SIZEOF(my_coef_controller) - SIZEOF(coef->blk_buffer));
+    for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+	 ci++, compptr++) {
+      access_rows = compptr->v_samp_factor;
+#ifdef BLOCK_SMOOTHING_SUPPORTED
+      /* If block smoothing could be used, need a bigger window */
+      if (cinfo->progressive_mode)
+	access_rows *= 3;
+#endif
+      coef->whole_image[ci] = (*cinfo->mem->request_virt_barray)
+	((j_common_ptr) cinfo, JPOOL_IMAGE, TRUE,
+	 (JDIMENSION) jround_up((long) compptr->width_in_blocks,
+				(long) compptr->h_samp_factor),
+	 (JDIMENSION) jround_up((long) compptr->height_in_blocks,
+				(long) compptr->v_samp_factor),
+	 (JDIMENSION) access_rows);
+    }
+    coef->pub.consume_data = consume_data;
+    coef->pub.decompress_data = decompress_data;
+    coef->pub.coef_arrays = coef->whole_image; /* link to virtual arrays */
+#else
+    ERREXIT(cinfo, JERR_NOT_COMPILED);
+#endif
+  } else {
+    /* We only need a single-MCU buffer. */
+    JBLOCKARRAY blkp;
+    JBLOCKROW buffer_ptr;
+    int bi;
+
+    coef = (my_coef_ptr) (*cinfo->mem->alloc_small)
+      ((j_common_ptr) cinfo, JPOOL_IMAGE, SIZEOF(my_coef_controller));
+    buffer_ptr = coef->blk_buffer;
+    if (cinfo->lim_Se == 0)	/* DC only case: want to bypass later */
+      MEMZERO(buffer_ptr, SIZEOF(coef->blk_buffer));
+    blkp = coef->MCU_buffer;
+    bi = D_MAX_BLOCKS_IN_MCU;
+    do {
+      *blkp++ = buffer_ptr++;
+    } while (--bi);
+    coef->pub.consume_data = dummy_consume_data;
+    coef->pub.decompress_data = decompress_onepass;
+    coef->pub.coef_arrays = NULL; /* flag for no virtual arrays */
+  }
+
+  coef->pub.start_input_pass = start_input_pass;
+  coef->pub.start_output_pass = start_output_pass;
+#ifdef BLOCK_SMOOTHING_SUPPORTED
+  coef->coef_bits_latch = NULL;
+#endif
+  cinfo->coef = &coef->pub;
+}
@@ -0,0 +1,769 @@
+/*
+ * jdcolor.c
+ *
+ * Copyright (C) 1991-1997, Thomas G. Lane.
+ * Modified 2011-2023 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains output colorspace conversion routines.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+#if RANGE_BITS < 2
+  /* Deliberate syntax err */
+  Sorry, this code requires 2 or more range extension bits.
+#endif
+
+
+/* Private subobject */
+
+typedef struct {
+  struct jpeg_color_deconverter pub; /* public fields */
+
+  /* Private state for YCbCr->RGB and BG_YCC->RGB conversion */
+  int * Cr_r_tab;		/* => table for Cr to R conversion */
+  int * Cb_b_tab;		/* => table for Cb to B conversion */
+  INT32 * Cr_g_tab;		/* => table for Cr to G conversion */
+  INT32 * Cb_g_tab;		/* => table for Cb to G conversion */
+
+  /* Private state for RGB->Y conversion */
+  INT32 * R_y_tab;		/* => table for R to Y conversion */
+  INT32 * G_y_tab;		/* => table for G to Y conversion */
+  INT32 * B_y_tab;		/* => table for B to Y conversion */
+} my_color_deconverter;
+
+typedef my_color_deconverter * my_cconvert_ptr;
+
+
+/***************  YCbCr -> RGB conversion: most common case **************/
+/*************** BG_YCC -> RGB conversion: less common case **************/
+/***************    RGB -> Y   conversion: less common case **************/
+
+/*
+ * YCbCr is defined per Recommendation ITU-R BT.601-7 (03/2011),
+ * previously known as Recommendation CCIR 601-1, except that Cb and Cr
+ * are normalized to the range 0..MAXJSAMPLE rather than -0.5 .. 0.5.
+ * sRGB (standard RGB color space) is defined per IEC 61966-2-1:1999.
+ * sYCC (standard luma-chroma-chroma color space with extended gamut)
+ * is defined per IEC 61966-2-1:1999 Amendment A1:2003 Annex F.
+ * bg-sRGB and bg-sYCC (big gamut standard color spaces)
+ * are defined per IEC 61966-2-1:1999 Amendment A1:2003 Annex G.
+ * Note that the derived conversion coefficients given in some of these
+ * documents are imprecise.  The general conversion equations are
+ *
+ *	R = Y + K * (1 - Kr) * Cr
+ *	G = Y - K * (Kb * (1 - Kb) * Cb + Kr * (1 - Kr) * Cr) / (1 - Kr - Kb)
+ *	B = Y + K * (1 - Kb) * Cb
+ *
+ *	Y = Kr * R + (1 - Kr - Kb) * G + Kb * B
+ *
+ * With Kr = 0.299 and Kb = 0.114 (derived according to SMPTE RP 177-1993
+ * from the 1953 FCC NTSC primaries and CIE Illuminant C), K = 2 for sYCC,
+ * the conversion equations to be implemented are therefore
+ *
+ *	R = Y + 1.402 * Cr
+ *	G = Y - 0.344136286 * Cb - 0.714136286 * Cr
+ *	B = Y + 1.772 * Cb
+ *
+ *	Y = 0.299 * R + 0.587 * G + 0.114 * B
+ *
+ * where Cb and Cr represent the incoming values less CENTERJSAMPLE.
+ * For bg-sYCC, with K = 4, the equations are
+ *
+ *	R = Y + 2.804 * Cr
+ *	G = Y - 0.688272572 * Cb - 1.428272572 * Cr
+ *	B = Y + 3.544 * Cb
+ *
+ * To avoid floating-point arithmetic, we represent the fractional constants
+ * as integers scaled up by 2^16 (about 4 digits precision); we have to divide
+ * the products by 2^16, with appropriate rounding, to get the correct answer.
+ * Notice that Y, being an integral input, does not contribute any fraction
+ * so it need not participate in the rounding.
+ *
+ * For even more speed, we avoid doing any multiplications in the inner loop
+ * by precalculating the constants times Cb and Cr for all possible values.
+ * For 8-bit JSAMPLEs this is very reasonable (only 256 entries per table);
+ * for 9-bit to 12-bit samples it is still acceptable.  It's not very
+ * reasonable for 16-bit samples, but if you want lossless storage
+ * you shouldn't be changing colorspace anyway.
+ * The Cr=>R and Cb=>B values can be rounded to integers in advance;
+ * the values for the G calculation are left scaled up,
+ * since we must add them together before rounding.
+ */
+
+#define SCALEBITS	16	/* speediest right-shift on some machines */
+#define ONE_HALF	((INT32) 1 << (SCALEBITS-1))
+#define FIX(x)		((INT32) ((x) * (1L<<SCALEBITS) + 0.5))
+
+
+/*
+ * Initialize tables for YCbCr->RGB and BG_YCC->RGB colorspace conversion.
+ */
+
+LOCAL(void)
+build_ycc_rgb_table (j_decompress_ptr cinfo)
+/* Normal case, sYCC */
+{
+  my_cconvert_ptr cconvert = (my_cconvert_ptr) cinfo->cconvert;
+  int i;
+  INT32 x;
+  SHIFT_TEMPS
+
+  cconvert->Cr_r_tab = (int *) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, (MAXJSAMPLE+1) * SIZEOF(int));
+  cconvert->Cb_b_tab = (int *) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, (MAXJSAMPLE+1) * SIZEOF(int));
+  cconvert->Cr_g_tab = (INT32 *) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, (MAXJSAMPLE+1) * SIZEOF(INT32));
+  cconvert->Cb_g_tab = (INT32 *) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, (MAXJSAMPLE+1) * SIZEOF(INT32));
+
+  for (i = 0, x = -CENTERJSAMPLE; i <= MAXJSAMPLE; i++, x++) {
+    /* i is the actual input pixel value, in the range 0..MAXJSAMPLE */
+    /* The Cb or Cr value we are thinking of is x = i - CENTERJSAMPLE */
+    /* Cr=>R value is nearest int to 1.402 * x */
+    cconvert->Cr_r_tab[i] = (int) DESCALE(FIX(1.402) * x, SCALEBITS);
+    /* Cb=>B value is nearest int to 1.772 * x */
+    cconvert->Cb_b_tab[i] = (int) DESCALE(FIX(1.772) * x, SCALEBITS);
+    /* Cr=>G value is scaled-up -0.714136286 * x */
+    cconvert->Cr_g_tab[i] = (- FIX(0.714136286)) * x;
+    /* Cb=>G value is scaled-up -0.344136286 * x */
+    /* We also add in ONE_HALF so that need not do it in inner loop */
+    cconvert->Cb_g_tab[i] = (- FIX(0.344136286)) * x + ONE_HALF;
+  }
+}
+
+
+LOCAL(void)
+build_bg_ycc_rgb_table (j_decompress_ptr cinfo)
+/* Wide gamut case, bg-sYCC */
+{
+  my_cconvert_ptr cconvert = (my_cconvert_ptr) cinfo->cconvert;
+  int i;
+  INT32 x;
+  SHIFT_TEMPS
+
+  cconvert->Cr_r_tab = (int *) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, (MAXJSAMPLE+1) * SIZEOF(int));
+  cconvert->Cb_b_tab = (int *) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, (MAXJSAMPLE+1) * SIZEOF(int));
+  cconvert->Cr_g_tab = (INT32 *) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, (MAXJSAMPLE+1) * SIZEOF(INT32));
+  cconvert->Cb_g_tab = (INT32 *) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, (MAXJSAMPLE+1) * SIZEOF(INT32));
+
+  for (i = 0, x = -CENTERJSAMPLE; i <= MAXJSAMPLE; i++, x++) {
+    /* i is the actual input pixel value, in the range 0..MAXJSAMPLE */
+    /* The Cb or Cr value we are thinking of is x = i - CENTERJSAMPLE */
+    /* Cr=>R value is nearest int to 2.804 * x */
+    cconvert->Cr_r_tab[i] = (int) DESCALE(FIX(2.804) * x, SCALEBITS);
+    /* Cb=>B value is nearest int to 3.544 * x */
+    cconvert->Cb_b_tab[i] = (int) DESCALE(FIX(3.544) * x, SCALEBITS);
+    /* Cr=>G value is scaled-up -1.428272572 * x */
+    cconvert->Cr_g_tab[i] = (- FIX(1.428272572)) * x;
+    /* Cb=>G value is scaled-up -0.688272572 * x */
+    /* We also add in ONE_HALF so that need not do it in inner loop */
+    cconvert->Cb_g_tab[i] = (- FIX(0.688272572)) * x + ONE_HALF;
+  }
+}
+
+
+/*
+ * Convert some rows of samples to the output colorspace.
+ *
+ * Note that we change from noninterleaved, one-plane-per-component format
+ * to interleaved-pixel format.  The output buffer is therefore three times
+ * as wide as the input buffer.
+ *
+ * A starting row offset is provided only for the input buffer.  The caller
+ * can easily adjust the passed output_buf value to accommodate any row
+ * offset required on that side.
+ */
+
+METHODDEF(void)
+ycc_rgb_convert (j_decompress_ptr cinfo,
+		 JSAMPIMAGE input_buf, JDIMENSION input_row,
+		 JSAMPARRAY output_buf, int num_rows)
+{
+  my_cconvert_ptr cconvert = (my_cconvert_ptr) cinfo->cconvert;
+  register int y, cb, cr;
+  register JSAMPROW outptr;
+  register JSAMPROW inptr0, inptr1, inptr2;
+  register JDIMENSION col;
+  JDIMENSION num_cols = cinfo->output_width;
+  /* copy these pointers into registers if possible */
+  register JSAMPLE * range_limit = cinfo->sample_range_limit;
+  register int * Crrtab = cconvert->Cr_r_tab;
+  register int * Cbbtab = cconvert->Cb_b_tab;
+  register INT32 * Crgtab = cconvert->Cr_g_tab;
+  register INT32 * Cbgtab = cconvert->Cb_g_tab;
+  SHIFT_TEMPS
+
+  while (--num_rows >= 0) {
+    inptr0 = input_buf[0][input_row];
+    inptr1 = input_buf[1][input_row];
+    inptr2 = input_buf[2][input_row];
+    input_row++;
+    outptr = *output_buf++;
+    for (col = 0; col < num_cols; col++) {
+      y  = GETJSAMPLE(inptr0[col]);
+      cb = GETJSAMPLE(inptr1[col]);
+      cr = GETJSAMPLE(inptr2[col]);
+      /* Range-limiting is essential due to noise introduced by DCT losses,
+       * for extended gamut (sYCC) and wide gamut (bg-sYCC) encodings.
+       */
+      outptr[RGB_RED]   = range_limit[y + Crrtab[cr]];
+      outptr[RGB_GREEN] = range_limit[y +
+			      ((int) RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr],
+						 SCALEBITS))];
+      outptr[RGB_BLUE]  = range_limit[y + Cbbtab[cb]];
+      outptr += RGB_PIXELSIZE;
+    }
+  }
+}
+
+
+/**************** Cases other than YCC -> RGB ****************/
+
+
+/*
+ * Initialize for RGB->grayscale colorspace conversion.
+ */
+
+LOCAL(void)
+build_rgb_y_table (j_decompress_ptr cinfo)
+{
+  my_cconvert_ptr cconvert = (my_cconvert_ptr) cinfo->cconvert;
+  INT32 i;
+
+  cconvert->R_y_tab = (INT32 *) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, (MAXJSAMPLE+1) * SIZEOF(INT32));
+  cconvert->G_y_tab = (INT32 *) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, (MAXJSAMPLE+1) * SIZEOF(INT32));
+  cconvert->B_y_tab = (INT32 *) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, (MAXJSAMPLE+1) * SIZEOF(INT32));
+
+  for (i = 0; i <= MAXJSAMPLE; i++) {
+    cconvert->R_y_tab[i] = FIX(0.299) * i;
+    cconvert->G_y_tab[i] = FIX(0.587) * i;
+    cconvert->B_y_tab[i] = FIX(0.114) * i + ONE_HALF;
+  }
+}
+
+
+/*
+ * Convert RGB to grayscale.
+ */
+
+METHODDEF(void)
+rgb_gray_convert (j_decompress_ptr cinfo,
+		  JSAMPIMAGE input_buf, JDIMENSION input_row,
+		  JSAMPARRAY output_buf, int num_rows)
+{
+  my_cconvert_ptr cconvert = (my_cconvert_ptr) cinfo->cconvert;
+  register INT32 y;
+  register INT32 * Rytab = cconvert->R_y_tab;
+  register INT32 * Gytab = cconvert->G_y_tab;
+  register INT32 * Bytab = cconvert->B_y_tab;
+  register JSAMPROW outptr;
+  register JSAMPROW inptr0, inptr1, inptr2;
+  register JDIMENSION col;
+  JDIMENSION num_cols = cinfo->output_width;
+
+  while (--num_rows >= 0) {
+    inptr0 = input_buf[0][input_row];
+    inptr1 = input_buf[1][input_row];
+    inptr2 = input_buf[2][input_row];
+    input_row++;
+    outptr = *output_buf++;
+    for (col = 0; col < num_cols; col++) {
+      y  = Rytab[GETJSAMPLE(inptr0[col])];
+      y += Gytab[GETJSAMPLE(inptr1[col])];
+      y += Bytab[GETJSAMPLE(inptr2[col])];
+      outptr[col] = (JSAMPLE) (y >> SCALEBITS);
+    }
+  }
+}
+
+
+/*
+ * Convert some rows of samples to the output colorspace.
+ * [R-G,G,B-G] to [R,G,B] conversion with modulo calculation
+ * (inverse color transform).
+ * This can be seen as an adaption of the general YCbCr->RGB
+ * conversion equation with Kr = Kb = 0, while replacing the
+ * normalization by modulo calculation.
+ */
+
+METHODDEF(void)
+rgb1_rgb_convert (j_decompress_ptr cinfo,
+		  JSAMPIMAGE input_buf, JDIMENSION input_row,
+		  JSAMPARRAY output_buf, int num_rows)
+{
+  register int r, g, b;
+  register JSAMPROW outptr;
+  register JSAMPROW inptr0, inptr1, inptr2;
+  register JDIMENSION col;
+  JDIMENSION num_cols = cinfo->output_width;
+
+  while (--num_rows >= 0) {
+    inptr0 = input_buf[0][input_row];
+    inptr1 = input_buf[1][input_row];
+    inptr2 = input_buf[2][input_row];
+    input_row++;
+    outptr = *output_buf++;
+    for (col = 0; col < num_cols; col++) {
+      r = GETJSAMPLE(inptr0[col]);
+      g = GETJSAMPLE(inptr1[col]);
+      b = GETJSAMPLE(inptr2[col]);
+      /* Assume that MAXJSAMPLE+1 is a power of 2, so that the MOD
+       * (modulo) operator is equivalent to the bitmask operator AND.
+       */
+      outptr[RGB_RED]   = (JSAMPLE) ((r + g - CENTERJSAMPLE) & MAXJSAMPLE);
+      outptr[RGB_GREEN] = (JSAMPLE) g;
+      outptr[RGB_BLUE]  = (JSAMPLE) ((b + g - CENTERJSAMPLE) & MAXJSAMPLE);
+      outptr += RGB_PIXELSIZE;
+    }
+  }
+}
+
+
+/*
+ * [R-G,G,B-G] to grayscale conversion with modulo calculation
+ * (inverse color transform).
+ */
+
+METHODDEF(void)
+rgb1_gray_convert (j_decompress_ptr cinfo,
+		   JSAMPIMAGE input_buf, JDIMENSION input_row,
+		   JSAMPARRAY output_buf, int num_rows)
+{
+  my_cconvert_ptr cconvert = (my_cconvert_ptr) cinfo->cconvert;
+  register int r, g, b;
+  register INT32 y;
+  register INT32 * Rytab = cconvert->R_y_tab;
+  register INT32 * Gytab = cconvert->G_y_tab;
+  register INT32 * Bytab = cconvert->B_y_tab;
+  register JSAMPROW outptr;
+  register JSAMPROW inptr0, inptr1, inptr2;
+  register JDIMENSION col;
+  JDIMENSION num_cols = cinfo->output_width;
+
+  while (--num_rows >= 0) {
+    inptr0 = input_buf[0][input_row];
+    inptr1 = input_buf[1][input_row];
+    inptr2 = input_buf[2][input_row];
+    input_row++;
+    outptr = *output_buf++;
+    for (col = 0; col < num_cols; col++) {
+      r = GETJSAMPLE(inptr0[col]);
+      g = GETJSAMPLE(inptr1[col]);
+      b = GETJSAMPLE(inptr2[col]);
+      /* Assume that MAXJSAMPLE+1 is a power of 2, so that the MOD
+       * (modulo) operator is equivalent to the bitmask operator AND.
+       */
+      y  = Rytab[(r + g - CENTERJSAMPLE) & MAXJSAMPLE];
+      y += Gytab[g];
+      y += Bytab[(b + g - CENTERJSAMPLE) & MAXJSAMPLE];
+      outptr[col] = (JSAMPLE) (y >> SCALEBITS);
+    }
+  }
+}
+
+
+/*
+ * Convert some rows of samples to the output colorspace.
+ * No colorspace change, but conversion from separate-planes
+ * to interleaved representation.
+ */
+
+METHODDEF(void)
+rgb_convert (j_decompress_ptr cinfo,
+	     JSAMPIMAGE input_buf, JDIMENSION input_row,
+	     JSAMPARRAY output_buf, int num_rows)
+{
+  register JSAMPROW outptr;
+  register JSAMPROW inptr0, inptr1, inptr2;
+  register JDIMENSION col;
+  JDIMENSION num_cols = cinfo->output_width;
+
+  while (--num_rows >= 0) {
+    inptr0 = input_buf[0][input_row];
+    inptr1 = input_buf[1][input_row];
+    inptr2 = input_buf[2][input_row];
+    input_row++;
+    outptr = *output_buf++;
+    for (col = 0; col < num_cols; col++) {
+      /* We can dispense with GETJSAMPLE() here */
+      outptr[RGB_RED]   = inptr0[col];
+      outptr[RGB_GREEN] = inptr1[col];
+      outptr[RGB_BLUE]  = inptr2[col];
+      outptr += RGB_PIXELSIZE;
+    }
+  }
+}
+
+
+/*
+ * Color conversion for no colorspace change: just copy the data,
+ * converting from separate-planes to interleaved representation.
+ * Note: Omit uninteresting components in output buffer.
+ */
+
+METHODDEF(void)
+null_convert (j_decompress_ptr cinfo,
+	      JSAMPIMAGE input_buf, JDIMENSION input_row,
+	      JSAMPARRAY output_buf, int num_rows)
+{
+  register JSAMPROW outptr;
+  register JSAMPROW inptr;
+  register JDIMENSION count;
+  register int out_comps = cinfo->out_color_components;
+  JDIMENSION num_cols = cinfo->output_width;
+  JSAMPROW startptr;
+  int ci;
+  jpeg_component_info *compptr;
+
+  while (--num_rows >= 0) {
+    /* It seems fastest to make a separate pass for each component. */
+    startptr = *output_buf++;
+    for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+	 ci++, compptr++) {
+      if (! compptr->component_needed)
+	continue;		/* skip uninteresting component */
+      inptr = input_buf[ci][input_row];
+      outptr = startptr++;
+      for (count = num_cols; count > 0; count--) {
+	*outptr = *inptr++;	/* don't need GETJSAMPLE() here */
+	outptr += out_comps;
+      }
+    }
+    input_row++;
+  }
+}
+
+
+/*
+ * Color conversion for grayscale: just copy the data.
+ * This also works for YCC -> grayscale conversion, in which
+ * we just copy the Y (luminance) component and ignore chrominance.
+ */
+
+METHODDEF(void)
+grayscale_convert (j_decompress_ptr cinfo,
+		   JSAMPIMAGE input_buf, JDIMENSION input_row,
+		   JSAMPARRAY output_buf, int num_rows)
+{
+  jcopy_sample_rows(input_buf[0] + input_row, output_buf,
+		    num_rows, cinfo->output_width);
+}
+
+
+/*
+ * Convert grayscale to RGB: just duplicate the graylevel three times.
+ * This is provided to support applications that don't want to cope
+ * with grayscale as a separate case.
+ */
+
+METHODDEF(void)
+gray_rgb_convert (j_decompress_ptr cinfo,
+		  JSAMPIMAGE input_buf, JDIMENSION input_row,
+		  JSAMPARRAY output_buf, int num_rows)
+{
+  register JSAMPROW outptr;
+  register JSAMPROW inptr;
+  register JDIMENSION col;
+  JDIMENSION num_cols = cinfo->output_width;
+
+  while (--num_rows >= 0) {
+    inptr = input_buf[0][input_row++];
+    outptr = *output_buf++;
+    for (col = 0; col < num_cols; col++) {
+      /* We can dispense with GETJSAMPLE() here */
+      outptr[RGB_RED] = outptr[RGB_GREEN] = outptr[RGB_BLUE] = inptr[col];
+      outptr += RGB_PIXELSIZE;
+    }
+  }
+}
+
+
+/*
+ * Convert some rows of samples to the output colorspace.
+ * This version handles Adobe-style YCCK->CMYK conversion,
+ * where we convert YCbCr to R=1-C, G=1-M, and B=1-Y using the
+ * same conversion as above, while passing K (black) unchanged.
+ * We assume build_ycc_rgb_table has been called.
+ */
+
+METHODDEF(void)
+ycck_cmyk_convert (j_decompress_ptr cinfo,
+		   JSAMPIMAGE input_buf, JDIMENSION input_row,
+		   JSAMPARRAY output_buf, int num_rows)
+{
+  my_cconvert_ptr cconvert = (my_cconvert_ptr) cinfo->cconvert;
+  register int y, cb, cr;
+  register JSAMPROW outptr;
+  register JSAMPROW inptr0, inptr1, inptr2, inptr3;
+  register JDIMENSION col;
+  JDIMENSION num_cols = cinfo->output_width;
+  /* copy these pointers into registers if possible */
+  register JSAMPLE * range_limit = cinfo->sample_range_limit;
+  register int * Crrtab = cconvert->Cr_r_tab;
+  register int * Cbbtab = cconvert->Cb_b_tab;
+  register INT32 * Crgtab = cconvert->Cr_g_tab;
+  register INT32 * Cbgtab = cconvert->Cb_g_tab;
+  SHIFT_TEMPS
+
+  while (--num_rows >= 0) {
+    inptr0 = input_buf[0][input_row];
+    inptr1 = input_buf[1][input_row];
+    inptr2 = input_buf[2][input_row];
+    inptr3 = input_buf[3][input_row];
+    input_row++;
+    outptr = *output_buf++;
+    for (col = 0; col < num_cols; col++) {
+      y  = GETJSAMPLE(inptr0[col]);
+      cb = GETJSAMPLE(inptr1[col]);
+      cr = GETJSAMPLE(inptr2[col]);
+      /* Range-limiting is essential due to noise introduced by DCT losses,
+       * and for extended gamut encodings (sYCC).
+       */
+      outptr[0] = range_limit[MAXJSAMPLE - (y + Crrtab[cr])];	/* red */
+      outptr[1] = range_limit[MAXJSAMPLE - (y +			/* green */
+			      ((int) RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr],
+						 SCALEBITS)))];
+      outptr[2] = range_limit[MAXJSAMPLE - (y + Cbbtab[cb])];	/* blue */
+      /* K passes through unchanged */
+      outptr[3] = inptr3[col];	/* don't need GETJSAMPLE here */
+      outptr += 4;
+    }
+  }
+}
+
+
+/*
+ * Convert CMYK to YK part of YCCK for colorless output.
+ * We assume build_rgb_y_table has been called.
+ */
+
+METHODDEF(void)
+cmyk_yk_convert (j_decompress_ptr cinfo,
+		 JSAMPIMAGE input_buf, JDIMENSION input_row,
+		 JSAMPARRAY output_buf, int num_rows)
+{
+  my_cconvert_ptr cconvert = (my_cconvert_ptr) cinfo->cconvert;
+  register INT32 y;
+  register INT32 * Rytab = cconvert->R_y_tab;
+  register INT32 * Gytab = cconvert->G_y_tab;
+  register INT32 * Bytab = cconvert->B_y_tab;
+  register JSAMPROW outptr;
+  register JSAMPROW inptr0, inptr1, inptr2, inptr3;
+  register JDIMENSION col;
+  JDIMENSION num_cols = cinfo->output_width;
+
+  while (--num_rows >= 0) {
+    inptr0 = input_buf[0][input_row];
+    inptr1 = input_buf[1][input_row];
+    inptr2 = input_buf[2][input_row];
+    inptr3 = input_buf[3][input_row];
+    input_row++;
+    outptr = *output_buf++;
+    for (col = 0; col < num_cols; col++) {
+      y  = Rytab[MAXJSAMPLE - GETJSAMPLE(inptr0[col])];
+      y += Gytab[MAXJSAMPLE - GETJSAMPLE(inptr1[col])];
+      y += Bytab[MAXJSAMPLE - GETJSAMPLE(inptr2[col])];
+      outptr[0] = (JSAMPLE) (y >> SCALEBITS);
+      /* K passes through unchanged */
+      outptr[1] = inptr3[col];	/* don't need GETJSAMPLE here */
+      outptr += 2;
+    }
+  }
+}
+
+
+/*
+ * Empty method for start_pass.
+ */
+
+METHODDEF(void)
+start_pass_dcolor (j_decompress_ptr cinfo)
+{
+  /* no work needed */
+}
+
+
+/*
+ * Module initialization routine for output colorspace conversion.
+ */
+
+GLOBAL(void)
+jinit_color_deconverter (j_decompress_ptr cinfo)
+{
+  my_cconvert_ptr cconvert;
+  int ci, i;
+
+  cconvert = (my_cconvert_ptr) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, SIZEOF(my_color_deconverter));
+  cinfo->cconvert = &cconvert->pub;
+  cconvert->pub.start_pass = start_pass_dcolor;
+
+  /* Make sure num_components agrees with jpeg_color_space */
+  switch (cinfo->jpeg_color_space) {
+  case JCS_GRAYSCALE:
+    if (cinfo->num_components != 1)
+      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
+    break;
+
+  case JCS_RGB:
+  case JCS_YCbCr:
+  case JCS_BG_RGB:
+  case JCS_BG_YCC:
+    if (cinfo->num_components != 3)
+      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
+    break;
+
+  case JCS_CMYK:
+  case JCS_YCCK:
+    if (cinfo->num_components != 4)
+      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
+    break;
+
+  default:			/* JCS_UNKNOWN can be anything */
+    if (cinfo->num_components < 1)
+      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
+  }
+
+  /* Support color transform only for RGB colorspaces */
+  if (cinfo->color_transform &&
+      cinfo->jpeg_color_space != JCS_RGB &&
+      cinfo->jpeg_color_space != JCS_BG_RGB)
+    ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+
+  /* Set out_color_components and conversion method based on requested space.
+   * Also adjust the component_needed flags for any unused components,
+   * so that earlier pipeline stages can avoid useless computation.
+   */
+
+  switch (cinfo->out_color_space) {
+  case JCS_GRAYSCALE:
+    cinfo->out_color_components = 1;
+    switch (cinfo->jpeg_color_space) {
+    case JCS_GRAYSCALE:
+    case JCS_YCbCr:
+    case JCS_BG_YCC:
+      cconvert->pub.color_convert = grayscale_convert;
+      /* For color->grayscale conversion, only the Y (0) component is needed */
+      for (ci = 1; ci < cinfo->num_components; ci++)
+	cinfo->comp_info[ci].component_needed = FALSE;
+      break;
+    case JCS_RGB:
+      switch (cinfo->color_transform) {
+      case JCT_NONE:
+	cconvert->pub.color_convert = rgb_gray_convert;
+	break;
+      case JCT_SUBTRACT_GREEN:
+	cconvert->pub.color_convert = rgb1_gray_convert;
+	break;
+      default:
+	ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+      }
+      build_rgb_y_table(cinfo);
+      break;
+    default:
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    }
+    break;
+
+  case JCS_RGB:
+    cinfo->out_color_components = RGB_PIXELSIZE;
+    switch (cinfo->jpeg_color_space) {
+    case JCS_GRAYSCALE:
+      cconvert->pub.color_convert = gray_rgb_convert;
+      break;
+    case JCS_YCbCr:
+      cconvert->pub.color_convert = ycc_rgb_convert;
+      build_ycc_rgb_table(cinfo);
+      break;
+    case JCS_BG_YCC:
+      cconvert->pub.color_convert = ycc_rgb_convert;
+      build_bg_ycc_rgb_table(cinfo);
+      break;
+    case JCS_RGB:
+      switch (cinfo->color_transform) {
+      case JCT_NONE:
+	cconvert->pub.color_convert = rgb_convert;
+	break;
+      case JCT_SUBTRACT_GREEN:
+	cconvert->pub.color_convert = rgb1_rgb_convert;
+	break;
+      default:
+	ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+      }
+      break;
+    default:
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    }
+    break;
+
+  case JCS_BG_RGB:
+    if (cinfo->jpeg_color_space != JCS_BG_RGB)
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    cinfo->out_color_components = RGB_PIXELSIZE;
+    switch (cinfo->color_transform) {
+    case JCT_NONE:
+      cconvert->pub.color_convert = rgb_convert;
+      break;
+    case JCT_SUBTRACT_GREEN:
+      cconvert->pub.color_convert = rgb1_rgb_convert;
+      break;
+    default:
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    }
+    break;
+
+  case JCS_CMYK:
+    if (cinfo->jpeg_color_space != JCS_YCCK)
+      goto def_label;
+    cinfo->out_color_components = 4;
+    cconvert->pub.color_convert = ycck_cmyk_convert;
+    build_ycc_rgb_table(cinfo);
+    break;
+
+  case JCS_YCCK:
+    if (cinfo->jpeg_color_space != JCS_CMYK ||
+	/* Support only YK part of YCCK for colorless output */
+	! cinfo->comp_info[0].component_needed ||
+	  cinfo->comp_info[1].component_needed ||
+	  cinfo->comp_info[2].component_needed ||
+	! cinfo->comp_info[3].component_needed)
+      goto def_label;
+    cinfo->out_color_components = 2;
+    /* Need all components on input side */
+    cinfo->comp_info[1].component_needed = TRUE;
+    cinfo->comp_info[2].component_needed = TRUE;
+    cconvert->pub.color_convert = cmyk_yk_convert;
+    build_rgb_y_table(cinfo);
+    break;
+
+  default: def_label:	/* permit null conversion to same output space */
+    if (cinfo->out_color_space != cinfo->jpeg_color_space)
+      /* unsupported non-null conversion */
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    i = 0;
+    for (ci = 0; ci < cinfo->num_components; ci++)
+      if (cinfo->comp_info[ci].component_needed)
+	i++;		/* count output color components */
+    cinfo->out_color_components = i;
+    cconvert->pub.color_convert = null_convert;
+  }
+
+  if (cinfo->quantize_colors)
+    cinfo->output_components = 1; /* single colormapped output component */
+  else
+    cinfo->output_components = cinfo->out_color_components;
+}
@@ -0,0 +1,409 @@
+/*
+ * jdct.h
+ *
+ * Copyright (C) 1994-1996, Thomas G. Lane.
+ * Modified 2002-2023 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This include file contains common declarations for the forward and
+ * inverse DCT modules.  These declarations are private to the DCT managers
+ * (jcdctmgr.c, jddctmgr.c) and the individual DCT algorithms.
+ * The individual DCT algorithms are kept in separate files to ease 
+ * machine-dependent tuning (e.g., assembly coding).
+ */
+
+
+/*
+ * A forward DCT routine is given a pointer to an input sample array and
+ * a pointer to a work area of type DCTELEM[]; the DCT is to be performed
+ * in-place in that buffer.  Type DCTELEM is int for 8-bit samples, INT32
+ * for 12-bit samples.  (NOTE: Floating-point DCT implementations use an
+ * array of type FAST_FLOAT, instead.)
+ * The input data is to be fetched from the sample array starting at a
+ * specified column.  (Any row offset needed will be applied to the array
+ * pointer before it is passed to the FDCT code.)
+ * Note that the number of samples fetched by the FDCT routine is
+ * DCT_h_scaled_size * DCT_v_scaled_size.
+ * The DCT outputs are returned scaled up by a factor of 8; they therefore
+ * have a range of +-8K for 8-bit data, +-128K for 12-bit data.  This
+ * convention improves accuracy in integer implementations and saves some
+ * work in floating-point ones.
+ * Quantization of the output coefficients is done by jcdctmgr.c.
+ */
+
+#if BITS_IN_JSAMPLE == 8
+typedef int DCTELEM;		/* 16 or 32 bits is fine */
+#else
+typedef INT32 DCTELEM;		/* must have 32 bits */
+#endif
+
+typedef JMETHOD(void, forward_DCT_method_ptr, (DCTELEM * data,
+					       JSAMPARRAY sample_data,
+					       JDIMENSION start_col));
+typedef JMETHOD(void, float_DCT_method_ptr, (FAST_FLOAT * data,
+					     JSAMPARRAY sample_data,
+					     JDIMENSION start_col));
+
+
+/*
+ * An inverse DCT routine is given a pointer to the input JBLOCK and a pointer
+ * to an output sample array.  The routine must dequantize the input data as
+ * well as perform the IDCT; for dequantization, it uses the multiplier table
+ * pointed to by compptr->dct_table.  The output data is to be placed into the
+ * sample array starting at a specified column.  (Any row offset needed will
+ * be applied to the array pointer before it is passed to the IDCT code.)
+ * Note that the number of samples emitted by the IDCT routine is
+ * DCT_h_scaled_size * DCT_v_scaled_size.
+ */
+
+/* typedef inverse_DCT_method_ptr is declared in jpegint.h */
+
+/*
+ * Each IDCT routine has its own ideas about the best dct_table element type.
+ */
+
+typedef MULTIPLIER ISLOW_MULT_TYPE; /* short or int, whichever is faster */
+#if BITS_IN_JSAMPLE == 8
+typedef MULTIPLIER IFAST_MULT_TYPE; /* 16 bits is OK, use short if faster */
+#define IFAST_SCALE_BITS  2	/* fractional bits in scale factors */
+#else
+typedef INT32 IFAST_MULT_TYPE;	/* need 32 bits for scaled quantizers */
+#define IFAST_SCALE_BITS  13	/* fractional bits in scale factors */
+#endif
+typedef FAST_FLOAT FLOAT_MULT_TYPE; /* preferred floating type */
+
+
+/*
+ * Each IDCT routine is responsible for range-limiting its results and
+ * converting them to unsigned form (0..MAXJSAMPLE).  The raw outputs could
+ * be quite far out of range if the input data is corrupt, so a bulletproof
+ * range-limiting step is required.  We use a mask-and-table-lookup method
+ * to do the combined operations quickly, assuming that RANGE_CENTER
+ * (defined in jpegint.h) is a power of 2.  See the comments with
+ * prepare_range_limit_table (in jdmaster.c) for more info.
+ */
+
+#define RANGE_MASK  (RANGE_CENTER * 2 - 1)
+#define RANGE_SUBSET  (RANGE_CENTER - CENTERJSAMPLE)
+
+#define IDCT_range_limit(cinfo)  ((cinfo)->sample_range_limit - RANGE_SUBSET)
+
+
+/* Short forms of external names for systems with brain-damaged linkers. */
+
+#ifdef NEED_SHORT_EXTERNAL_NAMES
+#define jpeg_fdct_islow		jFDislow
+#define jpeg_fdct_ifast		jFDifast
+#define jpeg_fdct_float		jFDfloat
+#define jpeg_fdct_7x7		jFD7x7
+#define jpeg_fdct_6x6		jFD6x6
+#define jpeg_fdct_5x5		jFD5x5
+#define jpeg_fdct_4x4		jFD4x4
+#define jpeg_fdct_3x3		jFD3x3
+#define jpeg_fdct_2x2		jFD2x2
+#define jpeg_fdct_1x1		jFD1x1
+#define jpeg_fdct_9x9		jFD9x9
+#define jpeg_fdct_10x10		jFD10x10
+#define jpeg_fdct_11x11		jFD11x11
+#define jpeg_fdct_12x12		jFD12x12
+#define jpeg_fdct_13x13		jFD13x13
+#define jpeg_fdct_14x14		jFD14x14
+#define jpeg_fdct_15x15		jFD15x15
+#define jpeg_fdct_16x16		jFD16x16
+#define jpeg_fdct_16x8		jFD16x8
+#define jpeg_fdct_14x7		jFD14x7
+#define jpeg_fdct_12x6		jFD12x6
+#define jpeg_fdct_10x5		jFD10x5
+#define jpeg_fdct_8x4		jFD8x4
+#define jpeg_fdct_6x3		jFD6x3
+#define jpeg_fdct_4x2		jFD4x2
+#define jpeg_fdct_2x1		jFD2x1
+#define jpeg_fdct_8x16		jFD8x16
+#define jpeg_fdct_7x14		jFD7x14
+#define jpeg_fdct_6x12		jFD6x12
+#define jpeg_fdct_5x10		jFD5x10
+#define jpeg_fdct_4x8		jFD4x8
+#define jpeg_fdct_3x6		jFD3x6
+#define jpeg_fdct_2x4		jFD2x4
+#define jpeg_fdct_1x2		jFD1x2
+#define jpeg_idct_islow		jRDislow
+#define jpeg_idct_ifast		jRDifast
+#define jpeg_idct_float		jRDfloat
+#define jpeg_idct_7x7		jRD7x7
+#define jpeg_idct_6x6		jRD6x6
+#define jpeg_idct_5x5		jRD5x5
+#define jpeg_idct_4x4		jRD4x4
+#define jpeg_idct_3x3		jRD3x3
+#define jpeg_idct_2x2		jRD2x2
+#define jpeg_idct_1x1		jRD1x1
+#define jpeg_idct_9x9		jRD9x9
+#define jpeg_idct_10x10		jRD10x10
+#define jpeg_idct_11x11		jRD11x11
+#define jpeg_idct_12x12		jRD12x12
+#define jpeg_idct_13x13		jRD13x13
+#define jpeg_idct_14x14		jRD14x14
+#define jpeg_idct_15x15		jRD15x15
+#define jpeg_idct_16x16		jRD16x16
+#define jpeg_idct_16x8		jRD16x8
+#define jpeg_idct_14x7		jRD14x7
+#define jpeg_idct_12x6		jRD12x6
+#define jpeg_idct_10x5		jRD10x5
+#define jpeg_idct_8x4		jRD8x4
+#define jpeg_idct_6x3		jRD6x3
+#define jpeg_idct_4x2		jRD4x2
+#define jpeg_idct_2x1		jRD2x1
+#define jpeg_idct_8x16		jRD8x16
+#define jpeg_idct_7x14		jRD7x14
+#define jpeg_idct_6x12		jRD6x12
+#define jpeg_idct_5x10		jRD5x10
+#define jpeg_idct_4x8		jRD4x8
+#define jpeg_idct_3x6		jRD3x6
+#define jpeg_idct_2x4		jRD2x4
+#define jpeg_idct_1x2		jRD1x2
+#endif /* NEED_SHORT_EXTERNAL_NAMES */
+
+/* Extern declarations for the forward and inverse DCT routines. */
+
+EXTERN(void) jpeg_fdct_islow
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_ifast
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_float
+    JPP((FAST_FLOAT * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_7x7
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_6x6
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_5x5
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_4x4
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_3x3
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_2x2
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_1x1
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_9x9
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_10x10
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_11x11
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_12x12
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_13x13
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_14x14
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_15x15
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_16x16
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_16x8
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_14x7
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_12x6
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_10x5
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_8x4
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_6x3
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_4x2
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_2x1
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_8x16
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_7x14
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_6x12
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_5x10
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_4x8
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_3x6
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_2x4
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_1x2
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+
+EXTERN(void) jpeg_idct_islow
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_ifast
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_float
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_7x7
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_6x6
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_5x5
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_4x4
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_3x3
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_2x2
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_1x1
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_9x9
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_10x10
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_11x11
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_12x12
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_13x13
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_14x14
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_15x15
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_16x16
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_16x8
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_14x7
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_12x6
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_10x5
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_8x4
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_6x3
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_4x2
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_2x1
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_8x16
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_7x14
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_6x12
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_5x10
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_4x8
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_3x6
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_2x4
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_1x2
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+
+
+/*
+ * Macros for handling fixed-point arithmetic; these are used by many
+ * but not all of the DCT/IDCT modules.
+ *
+ * All values are expected to be of type INT32.
+ * Fractional constants are scaled left by CONST_BITS bits.
+ * CONST_BITS is defined within each module using these macros,
+ * and may differ from one module to the next.
+ */
+
+#define ONE	((INT32) 1)
+#define CONST_SCALE (ONE << CONST_BITS)
+
+/* Convert a positive real constant to an integer scaled by CONST_SCALE.
+ * Caution: some C compilers fail to reduce "FIX(constant)" at compile time,
+ * thus causing a lot of useless floating-point operations at run time.
+ */
+
+#define FIX(x)	((INT32) ((x) * CONST_SCALE + 0.5))
+
+/* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
+ * This macro is used only when the two inputs will actually be no more than
+ * 16 bits wide, so that a 16x16->32 bit multiply can be used instead of a
+ * full 32x32 multiply.  This provides a useful speedup on many machines.
+ * Unfortunately there is no way to specify a 16x16->32 multiply portably
+ * in C, but some C compilers will do the right thing if you provide the
+ * correct combination of casts.
+ */
+
+#ifdef SHORTxSHORT_32		/* may work if 'int' is 32 bits */
+#define MULTIPLY16C16(var,const)  (((INT16) (var)) * ((INT16) (const)))
+#endif
+#ifdef SHORTxLCONST_32		/* known to work with Microsoft C 6.0 */
+#define MULTIPLY16C16(var,const)  (((INT16) (var)) * ((INT32) (const)))
+#endif
+
+#ifndef MULTIPLY16C16		/* default definition */
+#define MULTIPLY16C16(var,const)  ((var) * (const))
+#endif
+
+/* Same except both inputs are variables. */
+
+#ifdef SHORTxSHORT_32		/* may work if 'int' is 32 bits */
+#define MULTIPLY16V16(var1,var2)  (((INT16) (var1)) * ((INT16) (var2)))
+#endif
+
+#ifndef MULTIPLY16V16		/* default definition */
+#define MULTIPLY16V16(var1,var2)  ((var1) * (var2))
+#endif
+
+/* Like RIGHT_SHIFT, but applies to a DCTELEM.
+ * We assume that int right shift is unsigned if INT32 right shift is.
+ */
+
+#ifdef RIGHT_SHIFT_IS_UNSIGNED
+#define ISHIFT_TEMPS	DCTELEM ishift_temp;
+#if BITS_IN_JSAMPLE == 8
+#define DCTELEMBITS  16		/* DCTELEM may be 16 or 32 bits */
+#else
+#define DCTELEMBITS  32		/* DCTELEM must be 32 bits */
+#endif
+#define IRIGHT_SHIFT(x,shft)  \
+    ((ishift_temp = (x)) < 0 ? \
+     (ishift_temp >> (shft)) | ((~((DCTELEM) 0)) << (DCTELEMBITS-(shft))) : \
+     (ishift_temp >> (shft)))
+#else
+#define ISHIFT_TEMPS
+#define IRIGHT_SHIFT(x,shft)	((x) >> (shft))
+#endif
@@ -0,0 +1,384 @@
+/*
+ * jddctmgr.c
+ *
+ * Copyright (C) 1994-1996, Thomas G. Lane.
+ * Modified 2002-2013 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains the inverse-DCT management logic.
+ * This code selects a particular IDCT implementation to be used,
+ * and it performs related housekeeping chores.  No code in this file
+ * is executed per IDCT step, only during output pass setup.
+ *
+ * Note that the IDCT routines are responsible for performing coefficient
+ * dequantization as well as the IDCT proper.  This module sets up the
+ * dequantization multiplier table needed by the IDCT routine.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jdct.h"		/* Private declarations for DCT subsystem */
+
+
+/*
+ * The decompressor input side (jdinput.c) saves away the appropriate
+ * quantization table for each component at the start of the first scan
+ * involving that component.  (This is necessary in order to correctly
+ * decode files that reuse Q-table slots.)
+ * When we are ready to make an output pass, the saved Q-table is converted
+ * to a multiplier table that will actually be used by the IDCT routine.
+ * The multiplier table contents are IDCT-method-dependent.  To support
+ * application changes in IDCT method between scans, we can remake the
+ * multiplier tables if necessary.
+ * In buffered-image mode, the first output pass may occur before any data
+ * has been seen for some components, and thus before their Q-tables have
+ * been saved away.  To handle this case, multiplier tables are preset
+ * to zeroes; the result of the IDCT will be a neutral gray level.
+ */
+
+
+/* Private subobject for this module */
+
+typedef struct {
+  struct jpeg_inverse_dct pub;	/* public fields */
+
+  /* This array contains the IDCT method code that each multiplier table
+   * is currently set up for, or -1 if it's not yet set up.
+   * The actual multiplier tables are pointed to by dct_table in the
+   * per-component comp_info structures.
+   */
+  int cur_method[MAX_COMPONENTS];
+} my_idct_controller;
+
+typedef my_idct_controller * my_idct_ptr;
+
+
+/* Allocated multiplier tables: big enough for any supported variant */
+
+typedef union {
+  ISLOW_MULT_TYPE islow_array[DCTSIZE2];
+#ifdef DCT_IFAST_SUPPORTED
+  IFAST_MULT_TYPE ifast_array[DCTSIZE2];
+#endif
+#ifdef DCT_FLOAT_SUPPORTED
+  FLOAT_MULT_TYPE float_array[DCTSIZE2];
+#endif
+} multiplier_table;
+
+
+/* The current scaled-IDCT routines require ISLOW-style multiplier tables,
+ * so be sure to compile that code if either ISLOW or SCALING is requested.
+ */
+#ifdef DCT_ISLOW_SUPPORTED
+#define PROVIDE_ISLOW_TABLES
+#else
+#ifdef IDCT_SCALING_SUPPORTED
+#define PROVIDE_ISLOW_TABLES
+#endif
+#endif
+
+
+/*
+ * Prepare for an output pass.
+ * Here we select the proper IDCT routine for each component and build
+ * a matching multiplier table.
+ */
+
+METHODDEF(void)
+start_pass (j_decompress_ptr cinfo)
+{
+  my_idct_ptr idct = (my_idct_ptr) cinfo->idct;
+  int ci, i;
+  jpeg_component_info *compptr;
+  int method = 0;
+  inverse_DCT_method_ptr method_ptr = NULL;
+  JQUANT_TBL * qtbl;
+
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    /* Select the proper IDCT routine for this component's scaling */
+    switch ((compptr->DCT_h_scaled_size << 8) + compptr->DCT_v_scaled_size) {
+#ifdef IDCT_SCALING_SUPPORTED
+    case ((1 << 8) + 1):
+      method_ptr = jpeg_idct_1x1;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((2 << 8) + 2):
+      method_ptr = jpeg_idct_2x2;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((3 << 8) + 3):
+      method_ptr = jpeg_idct_3x3;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((4 << 8) + 4):
+      method_ptr = jpeg_idct_4x4;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((5 << 8) + 5):
+      method_ptr = jpeg_idct_5x5;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((6 << 8) + 6):
+      method_ptr = jpeg_idct_6x6;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((7 << 8) + 7):
+      method_ptr = jpeg_idct_7x7;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((9 << 8) + 9):
+      method_ptr = jpeg_idct_9x9;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((10 << 8) + 10):
+      method_ptr = jpeg_idct_10x10;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((11 << 8) + 11):
+      method_ptr = jpeg_idct_11x11;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((12 << 8) + 12):
+      method_ptr = jpeg_idct_12x12;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((13 << 8) + 13):
+      method_ptr = jpeg_idct_13x13;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((14 << 8) + 14):
+      method_ptr = jpeg_idct_14x14;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((15 << 8) + 15):
+      method_ptr = jpeg_idct_15x15;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((16 << 8) + 16):
+      method_ptr = jpeg_idct_16x16;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((16 << 8) + 8):
+      method_ptr = jpeg_idct_16x8;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((14 << 8) + 7):
+      method_ptr = jpeg_idct_14x7;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((12 << 8) + 6):
+      method_ptr = jpeg_idct_12x6;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((10 << 8) + 5):
+      method_ptr = jpeg_idct_10x5;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((8 << 8) + 4):
+      method_ptr = jpeg_idct_8x4;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((6 << 8) + 3):
+      method_ptr = jpeg_idct_6x3;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((4 << 8) + 2):
+      method_ptr = jpeg_idct_4x2;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((2 << 8) + 1):
+      method_ptr = jpeg_idct_2x1;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((8 << 8) + 16):
+      method_ptr = jpeg_idct_8x16;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((7 << 8) + 14):
+      method_ptr = jpeg_idct_7x14;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((6 << 8) + 12):
+      method_ptr = jpeg_idct_6x12;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((5 << 8) + 10):
+      method_ptr = jpeg_idct_5x10;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((4 << 8) + 8):
+      method_ptr = jpeg_idct_4x8;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((3 << 8) + 6):
+      method_ptr = jpeg_idct_3x6;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((2 << 8) + 4):
+      method_ptr = jpeg_idct_2x4;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((1 << 8) + 2):
+      method_ptr = jpeg_idct_1x2;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+#endif
+    case ((DCTSIZE << 8) + DCTSIZE):
+      switch (cinfo->dct_method) {
+#ifdef DCT_ISLOW_SUPPORTED
+      case JDCT_ISLOW:
+	method_ptr = jpeg_idct_islow;
+	method = JDCT_ISLOW;
+	break;
+#endif
+#ifdef DCT_IFAST_SUPPORTED
+      case JDCT_IFAST:
+	method_ptr = jpeg_idct_ifast;
+	method = JDCT_IFAST;
+	break;
+#endif
+#ifdef DCT_FLOAT_SUPPORTED
+      case JDCT_FLOAT:
+	method_ptr = jpeg_idct_float;
+	method = JDCT_FLOAT;
+	break;
+#endif
+      default:
+	ERREXIT(cinfo, JERR_NOT_COMPILED);
+	break;
+      }
+      break;
+    default:
+      ERREXIT2(cinfo, JERR_BAD_DCTSIZE,
+	       compptr->DCT_h_scaled_size, compptr->DCT_v_scaled_size);
+      break;
+    }
+    idct->pub.inverse_DCT[ci] = method_ptr;
+    /* Create multiplier table from quant table.
+     * However, we can skip this if the component is uninteresting
+     * or if we already built the table.  Also, if no quant table
+     * has yet been saved for the component, we leave the
+     * multiplier table all-zero; we'll be reading zeroes from the
+     * coefficient controller's buffer anyway.
+     */
+    if (! compptr->component_needed || idct->cur_method[ci] == method)
+      continue;
+    qtbl = compptr->quant_table;
+    if (qtbl == NULL)		/* happens if no data yet for component */
+      continue;
+    idct->cur_method[ci] = method;
+    switch (method) {
+#ifdef PROVIDE_ISLOW_TABLES
+    case JDCT_ISLOW:
+      {
+	/* For LL&M IDCT method, multipliers are equal to raw quantization
+	 * coefficients, but are stored as ints to ensure access efficiency.
+	 */
+	ISLOW_MULT_TYPE * ismtbl = (ISLOW_MULT_TYPE *) compptr->dct_table;
+	for (i = 0; i < DCTSIZE2; i++) {
+	  ismtbl[i] = (ISLOW_MULT_TYPE) qtbl->quantval[i];
+	}
+      }
+      break;
+#endif
+#ifdef DCT_IFAST_SUPPORTED
+    case JDCT_IFAST:
+      {
+	/* For AA&N IDCT method, multipliers are equal to quantization
+	 * coefficients scaled by scalefactor[row]*scalefactor[col], where
+	 *   scalefactor[0] = 1
+	 *   scalefactor[k] = cos(k*PI/16) * sqrt(2)    for k=1..7
+	 * For integer operation, the multiplier table is to be scaled by
+	 * IFAST_SCALE_BITS.
+	 */
+	IFAST_MULT_TYPE * ifmtbl = (IFAST_MULT_TYPE *) compptr->dct_table;
+#define CONST_BITS 14
+	static const INT16 aanscales[DCTSIZE2] = {
+	  /* precomputed values scaled up by 14 bits */
+	  16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
+	  22725, 31521, 29692, 26722, 22725, 17855, 12299,  6270,
+	  21407, 29692, 27969, 25172, 21407, 16819, 11585,  5906,
+	  19266, 26722, 25172, 22654, 19266, 15137, 10426,  5315,
+	  16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
+	  12873, 17855, 16819, 15137, 12873, 10114,  6967,  3552,
+	   8867, 12299, 11585, 10426,  8867,  6967,  4799,  2446,
+	   4520,  6270,  5906,  5315,  4520,  3552,  2446,  1247
+	};
+	SHIFT_TEMPS
+
+	for (i = 0; i < DCTSIZE2; i++) {
+	  ifmtbl[i] = (IFAST_MULT_TYPE)
+	    DESCALE(MULTIPLY16V16((INT32) qtbl->quantval[i],
+				  (INT32) aanscales[i]),
+		    CONST_BITS-IFAST_SCALE_BITS);
+	}
+      }
+      break;
+#endif
+#ifdef DCT_FLOAT_SUPPORTED
+    case JDCT_FLOAT:
+      {
+	/* For float AA&N IDCT method, multipliers are equal to quantization
+	 * coefficients scaled by scalefactor[row]*scalefactor[col], where
+	 *   scalefactor[0] = 1
+	 *   scalefactor[k] = cos(k*PI/16) * sqrt(2)    for k=1..7
+	 * We apply a further scale factor of 1/8.
+	 */
+	FLOAT_MULT_TYPE * fmtbl = (FLOAT_MULT_TYPE *) compptr->dct_table;
+	int row, col;
+	static const double aanscalefactor[DCTSIZE] = {
+	  1.0, 1.387039845, 1.306562965, 1.175875602,
+	  1.0, 0.785694958, 0.541196100, 0.275899379
+	};
+
+	i = 0;
+	for (row = 0; row < DCTSIZE; row++) {
+	  for (col = 0; col < DCTSIZE; col++) {
+	    fmtbl[i] = (FLOAT_MULT_TYPE)
+	      ((double) qtbl->quantval[i] *
+	       aanscalefactor[row] * aanscalefactor[col] * 0.125);
+	    i++;
+	  }
+	}
+      }
+      break;
+#endif
+    default:
+      ERREXIT(cinfo, JERR_NOT_COMPILED);
+      break;
+    }
+  }
+}
+
+
+/*
+ * Initialize IDCT manager.
+ */
+
+GLOBAL(void)
+jinit_inverse_dct (j_decompress_ptr cinfo)
+{
+  my_idct_ptr idct;
+  int ci;
+  jpeg_component_info *compptr;
+
+  idct = (my_idct_ptr)
+    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				SIZEOF(my_idct_controller));
+  cinfo->idct = &idct->pub;
+  idct->pub.start_pass = start_pass;
+
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    /* Allocate and pre-zero a multiplier table for each component */
+    compptr->dct_table =
+      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				  SIZEOF(multiplier_table));
+    MEMZERO(compptr->dct_table, SIZEOF(multiplier_table));
+    /* Mark multiplier table not yet set up for any method */
+    idct->cur_method[ci] = -1;
+  }
+}
@@ -0,0 +1,657 @@
+/*
+ * jdinput.c
+ *
+ * Copyright (C) 1991-1997, Thomas G. Lane.
+ * Modified 2002-2020 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains input control logic for the JPEG decompressor.
+ * These routines are concerned with controlling the decompressor's input
+ * processing (marker reading and coefficient decoding).  The actual input
+ * reading is done in jdmarker.c, jdhuff.c, and jdarith.c.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/* Private state */
+
+typedef struct {
+  struct jpeg_input_controller pub; /* public fields */
+
+  int inheaders;		/* Nonzero until first SOS is reached */
+} my_input_controller;
+
+typedef my_input_controller * my_inputctl_ptr;
+
+
+/* Forward declarations */
+METHODDEF(int) consume_markers JPP((j_decompress_ptr cinfo));
+
+
+/*
+ * Routines to calculate various quantities related to the size of the image.
+ */
+
+
+/*
+ * Compute output image dimensions and related values.
+ * NOTE: this is exported for possible use by application.
+ * Hence it mustn't do anything that can't be done twice.
+ */
+
+GLOBAL(void)
+jpeg_core_output_dimensions (j_decompress_ptr cinfo)
+/* Do computations that are needed before master selection phase.
+ * This function is used for transcoding and full decompression.
+ */
+{
+#ifdef IDCT_SCALING_SUPPORTED
+  int ci;
+  jpeg_component_info *compptr;
+
+  /* Compute actual output image dimensions and DCT scaling choices. */
+  if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom) {
+    /* Provide 1/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 1;
+    cinfo->min_DCT_v_scaled_size = 1;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 2) {
+    /* Provide 2/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 2L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 2L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 2;
+    cinfo->min_DCT_v_scaled_size = 2;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 3) {
+    /* Provide 3/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 3L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 3L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 3;
+    cinfo->min_DCT_v_scaled_size = 3;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 4) {
+    /* Provide 4/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 4L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 4L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 4;
+    cinfo->min_DCT_v_scaled_size = 4;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 5) {
+    /* Provide 5/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 5L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 5L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 5;
+    cinfo->min_DCT_v_scaled_size = 5;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 6) {
+    /* Provide 6/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 6L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 6L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 6;
+    cinfo->min_DCT_v_scaled_size = 6;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 7) {
+    /* Provide 7/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 7L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 7L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 7;
+    cinfo->min_DCT_v_scaled_size = 7;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 8) {
+    /* Provide 8/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 8L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 8L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 8;
+    cinfo->min_DCT_v_scaled_size = 8;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 9) {
+    /* Provide 9/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 9L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 9L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 9;
+    cinfo->min_DCT_v_scaled_size = 9;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 10) {
+    /* Provide 10/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 10L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 10L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 10;
+    cinfo->min_DCT_v_scaled_size = 10;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 11) {
+    /* Provide 11/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 11L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 11L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 11;
+    cinfo->min_DCT_v_scaled_size = 11;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 12) {
+    /* Provide 12/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 12L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 12L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 12;
+    cinfo->min_DCT_v_scaled_size = 12;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 13) {
+    /* Provide 13/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 13L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 13L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 13;
+    cinfo->min_DCT_v_scaled_size = 13;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 14) {
+    /* Provide 14/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 14L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 14L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 14;
+    cinfo->min_DCT_v_scaled_size = 14;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 15) {
+    /* Provide 15/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 15L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 15L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 15;
+    cinfo->min_DCT_v_scaled_size = 15;
+  } else {
+    /* Provide 16/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 16L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 16L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 16;
+    cinfo->min_DCT_v_scaled_size = 16;
+  }
+
+  /* Recompute dimensions of components */
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    compptr->DCT_h_scaled_size = cinfo->min_DCT_h_scaled_size;
+    compptr->DCT_v_scaled_size = cinfo->min_DCT_v_scaled_size;
+  }
+
+#else /* !IDCT_SCALING_SUPPORTED */
+
+  /* Hardwire it to "no scaling" */
+  cinfo->output_width = cinfo->image_width;
+  cinfo->output_height = cinfo->image_height;
+  /* initial_setup has already initialized DCT_scaled_size,
+   * and has computed unscaled downsampled_width and downsampled_height.
+   */
+
+#endif /* IDCT_SCALING_SUPPORTED */
+}
+
+
+LOCAL(void)
+initial_setup (j_decompress_ptr cinfo)
+/* Called once, when first SOS marker is reached */
+{
+  int ci;
+  jpeg_component_info *compptr;
+
+  /* Make sure image isn't bigger than I can handle */
+  if ((long) cinfo->image_height > (long) JPEG_MAX_DIMENSION ||
+      (long) cinfo->image_width > (long) JPEG_MAX_DIMENSION)
+    ERREXIT1(cinfo, JERR_IMAGE_TOO_BIG, (unsigned int) JPEG_MAX_DIMENSION);
+
+  /* Only 8 to 12 bits data precision are supported for DCT based JPEG */
+  if (cinfo->data_precision < 8 || cinfo->data_precision > 12)
+    ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision);
+
+  /* Check that number of components won't exceed internal array sizes */
+  if (cinfo->num_components > MAX_COMPONENTS)
+    ERREXIT2(cinfo, JERR_COMPONENT_COUNT, cinfo->num_components,
+	     MAX_COMPONENTS);
+
+  /* Compute maximum sampling factors; check factor validity */
+  cinfo->max_h_samp_factor = 1;
+  cinfo->max_v_samp_factor = 1;
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    if (compptr->h_samp_factor<=0 || compptr->h_samp_factor>MAX_SAMP_FACTOR ||
+	compptr->v_samp_factor<=0 || compptr->v_samp_factor>MAX_SAMP_FACTOR)
+      ERREXIT(cinfo, JERR_BAD_SAMPLING);
+    cinfo->max_h_samp_factor = MAX(cinfo->max_h_samp_factor,
+				   compptr->h_samp_factor);
+    cinfo->max_v_samp_factor = MAX(cinfo->max_v_samp_factor,
+				   compptr->v_samp_factor);
+  }
+
+  /* Derive block_size, natural_order, and lim_Se */
+  if (cinfo->is_baseline || (cinfo->progressive_mode &&
+      cinfo->comps_in_scan)) { /* no pseudo SOS marker */
+    cinfo->block_size = DCTSIZE;
+    cinfo->natural_order = jpeg_natural_order;
+    cinfo->lim_Se = DCTSIZE2-1;
+  } else
+    switch (cinfo->Se) {
+    case (1*1-1):
+      cinfo->block_size = 1;
+      cinfo->natural_order = jpeg_natural_order; /* not needed */
+      cinfo->lim_Se = cinfo->Se;
+      break;
+    case (2*2-1):
+      cinfo->block_size = 2;
+      cinfo->natural_order = jpeg_natural_order2;
+      cinfo->lim_Se = cinfo->Se;
+      break;
+    case (3*3-1):
+      cinfo->block_size = 3;
+      cinfo->natural_order = jpeg_natural_order3;
+      cinfo->lim_Se = cinfo->Se;
+      break;
+    case (4*4-1):
+      cinfo->block_size = 4;
+      cinfo->natural_order = jpeg_natural_order4;
+      cinfo->lim_Se = cinfo->Se;
+      break;
+    case (5*5-1):
+      cinfo->block_size = 5;
+      cinfo->natural_order = jpeg_natural_order5;
+      cinfo->lim_Se = cinfo->Se;
+      break;
+    case (6*6-1):
+      cinfo->block_size = 6;
+      cinfo->natural_order = jpeg_natural_order6;
+      cinfo->lim_Se = cinfo->Se;
+      break;
+    case (7*7-1):
+      cinfo->block_size = 7;
+      cinfo->natural_order = jpeg_natural_order7;
+      cinfo->lim_Se = cinfo->Se;
+      break;
+    case (8*8-1):
+      cinfo->block_size = 8;
+      cinfo->natural_order = jpeg_natural_order;
+      cinfo->lim_Se = DCTSIZE2-1;
+      break;
+    case (9*9-1):
+      cinfo->block_size = 9;
+      cinfo->natural_order = jpeg_natural_order;
+      cinfo->lim_Se = DCTSIZE2-1;
+      break;
+    case (10*10-1):
+      cinfo->block_size = 10;
+      cinfo->natural_order = jpeg_natural_order;
+      cinfo->lim_Se = DCTSIZE2-1;
+      break;
+    case (11*11-1):
+      cinfo->block_size = 11;
+      cinfo->natural_order = jpeg_natural_order;
+      cinfo->lim_Se = DCTSIZE2-1;
+      break;
+    case (12*12-1):
+      cinfo->block_size = 12;
+      cinfo->natural_order = jpeg_natural_order;
+      cinfo->lim_Se = DCTSIZE2-1;
+      break;
+    case (13*13-1):
+      cinfo->block_size = 13;
+      cinfo->natural_order = jpeg_natural_order;
+      cinfo->lim_Se = DCTSIZE2-1;
+      break;
+    case (14*14-1):
+      cinfo->block_size = 14;
+      cinfo->natural_order = jpeg_natural_order;
+      cinfo->lim_Se = DCTSIZE2-1;
+      break;
+    case (15*15-1):
+      cinfo->block_size = 15;
+      cinfo->natural_order = jpeg_natural_order;
+      cinfo->lim_Se = DCTSIZE2-1;
+      break;
+    case (16*16-1):
+      cinfo->block_size = 16;
+      cinfo->natural_order = jpeg_natural_order;
+      cinfo->lim_Se = DCTSIZE2-1;
+      break;
+    default:
+      ERREXIT4(cinfo, JERR_BAD_PROGRESSION,
+	       cinfo->Ss, cinfo->Se, cinfo->Ah, cinfo->Al);
+    }
+
+  /* We initialize DCT_scaled_size and min_DCT_scaled_size to block_size.
+   * In the full decompressor,
+   * this will be overridden by jpeg_calc_output_dimensions in jdmaster.c;
+   * but in the transcoder,
+   * jpeg_calc_output_dimensions is not used, so we must do it here.
+   */
+  cinfo->min_DCT_h_scaled_size = cinfo->block_size;
+  cinfo->min_DCT_v_scaled_size = cinfo->block_size;
+
+  /* Compute dimensions of components */
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    compptr->DCT_h_scaled_size = cinfo->block_size;
+    compptr->DCT_v_scaled_size = cinfo->block_size;
+    /* Size in DCT blocks */
+    compptr->width_in_blocks = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * (long) compptr->h_samp_factor,
+		    (long) (cinfo->max_h_samp_factor * cinfo->block_size));
+    compptr->height_in_blocks = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * (long) compptr->v_samp_factor,
+		    (long) (cinfo->max_v_samp_factor * cinfo->block_size));
+    /* downsampled_width and downsampled_height will also be overridden by
+     * jdmaster.c if we are doing full decompression.  The transcoder library
+     * doesn't use these values, but the calling application might.
+     */
+    /* Size in samples */
+    compptr->downsampled_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * (long) compptr->h_samp_factor,
+		    (long) cinfo->max_h_samp_factor);
+    compptr->downsampled_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * (long) compptr->v_samp_factor,
+		    (long) cinfo->max_v_samp_factor);
+    /* Mark component needed, until color conversion says otherwise */
+    compptr->component_needed = TRUE;
+    /* Mark no quantization table yet saved for component */
+    compptr->quant_table = NULL;
+  }
+
+  /* Compute number of fully interleaved MCU rows. */
+  cinfo->total_iMCU_rows = (JDIMENSION)
+    jdiv_round_up((long) cinfo->image_height,
+	          (long) (cinfo->max_v_samp_factor * cinfo->block_size));
+
+  /* Decide whether file contains multiple scans */
+  if (cinfo->comps_in_scan < cinfo->num_components || cinfo->progressive_mode)
+    cinfo->inputctl->has_multiple_scans = TRUE;
+  else
+    cinfo->inputctl->has_multiple_scans = FALSE;
+}
+
+
+LOCAL(void)
+per_scan_setup (j_decompress_ptr cinfo)
+/* Do computations that are needed before processing a JPEG scan */
+/* cinfo->comps_in_scan and cinfo->cur_comp_info[] were set from SOS marker */
+{
+  int ci, mcublks, tmp;
+  jpeg_component_info *compptr;
+
+  if (cinfo->comps_in_scan == 1) {
+
+    /* Noninterleaved (single-component) scan */
+    compptr = cinfo->cur_comp_info[0];
+
+    /* Overall image size in MCUs */
+    cinfo->MCUs_per_row = compptr->width_in_blocks;
+    cinfo->MCU_rows_in_scan = compptr->height_in_blocks;
+
+    /* For noninterleaved scan, always one block per MCU */
+    compptr->MCU_width = 1;
+    compptr->MCU_height = 1;
+    compptr->MCU_blocks = 1;
+    compptr->MCU_sample_width = compptr->DCT_h_scaled_size;
+    compptr->last_col_width = 1;
+    /* For noninterleaved scans, it is convenient to define last_row_height
+     * as the number of block rows present in the last iMCU row.
+     */
+    tmp = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
+    if (tmp == 0) tmp = compptr->v_samp_factor;
+    compptr->last_row_height = tmp;
+
+    /* Prepare array describing MCU composition */
+    cinfo->blocks_in_MCU = 1;
+    cinfo->MCU_membership[0] = 0;
+
+  } else {
+
+    /* Interleaved (multi-component) scan */
+    if (cinfo->comps_in_scan <= 0 || cinfo->comps_in_scan > MAX_COMPS_IN_SCAN)
+      ERREXIT2(cinfo, JERR_COMPONENT_COUNT, cinfo->comps_in_scan,
+	       MAX_COMPS_IN_SCAN);
+
+    /* Overall image size in MCUs */
+    cinfo->MCUs_per_row = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width,
+		    (long) (cinfo->max_h_samp_factor * cinfo->block_size));
+    cinfo->MCU_rows_in_scan = cinfo->total_iMCU_rows;
+
+    cinfo->blocks_in_MCU = 0;
+
+    for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+      compptr = cinfo->cur_comp_info[ci];
+      /* Sampling factors give # of blocks of component in each MCU */
+      compptr->MCU_width = compptr->h_samp_factor;
+      compptr->MCU_height = compptr->v_samp_factor;
+      compptr->MCU_blocks = compptr->MCU_width * compptr->MCU_height;
+      compptr->MCU_sample_width = compptr->MCU_width * compptr->DCT_h_scaled_size;
+      /* Figure number of non-dummy blocks in last MCU column & row */
+      tmp = (int) (compptr->width_in_blocks % compptr->MCU_width);
+      if (tmp == 0) tmp = compptr->MCU_width;
+      compptr->last_col_width = tmp;
+      tmp = (int) (compptr->height_in_blocks % compptr->MCU_height);
+      if (tmp == 0) tmp = compptr->MCU_height;
+      compptr->last_row_height = tmp;
+      /* Prepare array describing MCU composition */
+      mcublks = compptr->MCU_blocks;
+      if (cinfo->blocks_in_MCU + mcublks > D_MAX_BLOCKS_IN_MCU)
+	ERREXIT(cinfo, JERR_BAD_MCU_SIZE);
+      while (mcublks-- > 0) {
+	cinfo->MCU_membership[cinfo->blocks_in_MCU++] = ci;
+      }
+    }
+
+  }
+}
+
+
+/*
+ * Save away a copy of the Q-table referenced by each component present
+ * in the current scan, unless already saved during a prior scan.
+ *
+ * In a multiple-scan JPEG file, the encoder could assign different components
+ * the same Q-table slot number, but change table definitions between scans
+ * so that each component uses a different Q-table.  (The IJG encoder is not
+ * currently capable of doing this, but other encoders might.)  Since we want
+ * to be able to dequantize all the components at the end of the file, this
+ * means that we have to save away the table actually used for each component.
+ * We do this by copying the table at the start of the first scan containing
+ * the component.
+ * The JPEG spec prohibits the encoder from changing the contents of a Q-table
+ * slot between scans of a component using that slot.  If the encoder does so
+ * anyway, this decoder will simply use the Q-table values that were current
+ * at the start of the first scan for the component.
+ *
+ * The decompressor output side looks only at the saved quant tables,
+ * not at the current Q-table slots.
+ */
+
+LOCAL(void)
+latch_quant_tables (j_decompress_ptr cinfo)
+{
+  int ci, qtblno;
+  jpeg_component_info *compptr;
+  JQUANT_TBL * qtbl;
+
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+    compptr = cinfo->cur_comp_info[ci];
+    /* No work if we already saved Q-table for this component */
+    if (compptr->quant_table != NULL)
+      continue;
+    /* Make sure specified quantization table is present */
+    qtblno = compptr->quant_tbl_no;
+    if (qtblno < 0 || qtblno >= NUM_QUANT_TBLS ||
+	cinfo->quant_tbl_ptrs[qtblno] == NULL)
+      ERREXIT1(cinfo, JERR_NO_QUANT_TABLE, qtblno);
+    /* OK, save away the quantization table */
+    qtbl = (JQUANT_TBL *) (*cinfo->mem->alloc_small)
+      ((j_common_ptr) cinfo, JPOOL_IMAGE, SIZEOF(JQUANT_TBL));
+    MEMCOPY(qtbl, cinfo->quant_tbl_ptrs[qtblno], SIZEOF(JQUANT_TBL));
+    compptr->quant_table = qtbl;
+  }
+}
+
+
+/*
+ * Initialize the input modules to read a scan of compressed data.
+ * The first call to this is done by jdmaster.c after initializing
+ * the entire decompressor (during jpeg_start_decompress).
+ * Subsequent calls come from consume_markers, below.
+ */
+
+METHODDEF(void)
+start_input_pass (j_decompress_ptr cinfo)
+{
+  per_scan_setup(cinfo);
+  latch_quant_tables(cinfo);
+  (*cinfo->entropy->start_pass) (cinfo);
+  (*cinfo->coef->start_input_pass) (cinfo);
+  cinfo->inputctl->consume_input = cinfo->coef->consume_data;
+}
+
+
+/*
+ * Finish up after inputting a compressed-data scan.
+ * This is called by the coefficient controller after it's read all
+ * the expected data of the scan.
+ */
+
+METHODDEF(void)
+finish_input_pass (j_decompress_ptr cinfo)
+{
+  (*cinfo->entropy->finish_pass) (cinfo);
+  cinfo->inputctl->consume_input = consume_markers;
+}
+
+
+/*
+ * Read JPEG markers before, between, or after compressed-data scans.
+ * Change state as necessary when a new scan is reached.
+ * Return value is JPEG_SUSPENDED, JPEG_REACHED_SOS, or JPEG_REACHED_EOI.
+ *
+ * The consume_input method pointer points either here or to the
+ * coefficient controller's consume_data routine, depending on whether
+ * we are reading a compressed data segment or inter-segment markers.
+ *
+ * Note: This function should NOT return a pseudo SOS marker (with zero
+ * component number) to the caller.  A pseudo marker received by
+ * read_markers is processed and then skipped for other markers.
+ */
+
+METHODDEF(int)
+consume_markers (j_decompress_ptr cinfo)
+{
+  my_inputctl_ptr inputctl = (my_inputctl_ptr) cinfo->inputctl;
+  int val;
+
+  if (inputctl->pub.eoi_reached) /* After hitting EOI, read no further */
+    return JPEG_REACHED_EOI;
+
+  for (;;) {			/* Loop to pass pseudo SOS marker */
+    val = (*cinfo->marker->read_markers) (cinfo);
+
+    switch (val) {
+    case JPEG_REACHED_SOS:	/* Found SOS */
+      if (inputctl->inheaders) { /* 1st SOS */
+	if (inputctl->inheaders == 1)
+	  initial_setup(cinfo);
+	if (cinfo->comps_in_scan == 0) { /* pseudo SOS marker */
+	  inputctl->inheaders = 2;
+	  break;
+	}
+	inputctl->inheaders = 0;
+	/* Note: start_input_pass must be called by jdmaster.c
+	 * before any more input can be consumed.  jdapimin.c is
+	 * responsible for enforcing this sequencing.
+	 */
+      } else {			/* 2nd or later SOS marker */
+	if (! inputctl->pub.has_multiple_scans)
+	  ERREXIT(cinfo, JERR_EOI_EXPECTED); /* Oops, I wasn't expecting this! */
+	if (cinfo->comps_in_scan == 0) /* unexpected pseudo SOS marker */
+	  break;
+	start_input_pass(cinfo);
+      }
+      return val;
+    case JPEG_REACHED_EOI:	/* Found EOI */
+      inputctl->pub.eoi_reached = TRUE;
+      if (inputctl->inheaders) { /* Tables-only datastream, apparently */
+	if (cinfo->marker->saw_SOF)
+	  ERREXIT(cinfo, JERR_SOF_NO_SOS);
+      } else {
+	/* Prevent infinite loop in coef ctlr's decompress_data routine
+	 * if user set output_scan_number larger than number of scans.
+	 */
+	if (cinfo->output_scan_number > cinfo->input_scan_number)
+	  cinfo->output_scan_number = cinfo->input_scan_number;
+      }
+      return val;
+    case JPEG_SUSPENDED:
+      return val;
+    default:
+      return val;
+    }
+  }
+}
+
+
+/*
+ * Reset state to begin a fresh datastream.
+ */
+
+METHODDEF(void)
+reset_input_controller (j_decompress_ptr cinfo)
+{
+  my_inputctl_ptr inputctl = (my_inputctl_ptr) cinfo->inputctl;
+
+  inputctl->pub.consume_input = consume_markers;
+  inputctl->pub.has_multiple_scans = FALSE; /* "unknown" would be better */
+  inputctl->pub.eoi_reached = FALSE;
+  inputctl->inheaders = 1;
+  /* Reset other modules */
+  (*cinfo->err->reset_error_mgr) ((j_common_ptr) cinfo);
+  (*cinfo->marker->reset_marker_reader) (cinfo);
+  /* Reset progression state -- would be cleaner if entropy decoder did this */
+  cinfo->coef_bits = NULL;
+}
+
+
+/*
+ * Initialize the input controller module.
+ * This is called only once, when the decompression object is created.
+ */
+
+GLOBAL(void)
+jinit_input_controller (j_decompress_ptr cinfo)
+{
+  my_inputctl_ptr inputctl;
+
+  /* Create subobject in permanent pool */
+  inputctl = (my_inputctl_ptr) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_PERMANENT, SIZEOF(my_input_controller));
+  cinfo->inputctl = &inputctl->pub;
+  /* Initialize method pointers */
+  inputctl->pub.consume_input = consume_markers;
+  inputctl->pub.reset_input_controller = reset_input_controller;
+  inputctl->pub.start_input_pass = start_input_pass;
+  inputctl->pub.finish_input_pass = finish_input_pass;
+  /* Initialize state: can't use reset_input_controller since we don't
+   * want to try to reset other modules yet.
+   */
+  inputctl->pub.has_multiple_scans = FALSE; /* "unknown" would be better */
+  inputctl->pub.eoi_reached = FALSE;
+  inputctl->inheaders = 1;
+}
@@ -0,0 +1,511 @@
+/*
+ * jdmainct.c
+ *
+ * Copyright (C) 1994-1996, Thomas G. Lane.
+ * Modified 2002-2020 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains the main buffer controller for decompression.
+ * The main buffer lies between the JPEG decompressor proper and the
+ * post-processor; it holds downsampled data in the JPEG colorspace.
+ *
+ * Note that this code is bypassed in raw-data mode, since the application
+ * supplies the equivalent of the main buffer in that case.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/*
+ * In the current system design, the main buffer need never be a full-image
+ * buffer; any full-height buffers will be found inside the coefficient or
+ * postprocessing controllers.  Nonetheless, the main controller is not
+ * trivial.  Its responsibility is to provide context rows for upsampling/
+ * rescaling, and doing this in an efficient fashion is a bit tricky.
+ *
+ * Postprocessor input data is counted in "row groups".  A row group is
+ * defined to be (v_samp_factor * DCT_v_scaled_size / min_DCT_v_scaled_size)
+ * sample rows of each component.  (We require DCT_scaled_size values to be
+ * chosen such that these numbers are integers.  In practice DCT_scaled_size
+ * values will likely be powers of two, so we actually have the stronger
+ * condition that DCT_scaled_size / min_DCT_scaled_size is an integer.)
+ * Upsampling will typically produce max_v_samp_factor pixel rows from each
+ * row group (times any additional scale factor that the upsampler is
+ * applying).
+ *
+ * The coefficient controller will deliver data to us one iMCU row at a time;
+ * each iMCU row contains v_samp_factor * DCT_v_scaled_size sample rows, or
+ * exactly min_DCT_v_scaled_size row groups.  (This amount of data corresponds
+ * to one row of MCUs when the image is fully interleaved.)  Note that the
+ * number of sample rows varies across components, but the number of row
+ * groups does not.  Some garbage sample rows may be included in the last iMCU
+ * row at the bottom of the image.
+ *
+ * Depending on the vertical scaling algorithm used, the upsampler may need
+ * access to the sample row(s) above and below its current input row group.
+ * The upsampler is required to set need_context_rows TRUE at global selection
+ * time if so.  When need_context_rows is FALSE, this controller can simply
+ * obtain one iMCU row at a time from the coefficient controller and dole it
+ * out as row groups to the postprocessor.
+ *
+ * When need_context_rows is TRUE, this controller guarantees that the buffer
+ * passed to postprocessing contains at least one row group's worth of samples
+ * above and below the row group(s) being processed.  Note that the context
+ * rows "above" the first passed row group appear at negative row offsets in
+ * the passed buffer.  At the top and bottom of the image, the required
+ * context rows are manufactured by duplicating the first or last real sample
+ * row; this avoids having special cases in the upsampling inner loops.
+ *
+ * The amount of context is fixed at one row group just because that's a
+ * convenient number for this controller to work with.  The existing
+ * upsamplers really only need one sample row of context.  An upsampler
+ * supporting arbitrary output rescaling might wish for more than one row
+ * group of context when shrinking the image; tough, we don't handle that.
+ * (This is justified by the assumption that downsizing will be handled mostly
+ * by adjusting the DCT_scaled_size values, so that the actual scale factor at
+ * the upsample step needn't be much less than one.)
+ *
+ * To provide the desired context, we have to retain the last two row groups
+ * of one iMCU row while reading in the next iMCU row.  (The last row group
+ * can't be processed until we have another row group for its below-context,
+ * and so we have to save the next-to-last group too for its above-context.)
+ * We could do this most simply by copying data around in our buffer, but
+ * that'd be very slow.  We can avoid copying any data by creating a rather
+ * strange pointer structure.  Here's how it works.  We allocate a workspace
+ * consisting of M+2 row groups (where M = min_DCT_v_scaled_size is the number
+ * of row groups per iMCU row).  We create two sets of redundant pointers to
+ * the workspace.  Labeling the physical row groups 0 to M+1, the synthesized
+ * pointer lists look like this:
+ *                   M+1                          M-1
+ * master pointer --> 0         master pointer --> 0
+ *                    1                            1
+ *                   ...                          ...
+ *                   M-3                          M-3
+ *                   M-2                           M
+ *                   M-1                          M+1
+ *                    M                           M-2
+ *                   M+1                          M-1
+ *                    0                            0
+ * We read alternate iMCU rows using each master pointer; thus the last two
+ * row groups of the previous iMCU row remain un-overwritten in the workspace.
+ * The pointer lists are set up so that the required context rows appear to
+ * be adjacent to the proper places when we pass the pointer lists to the
+ * upsampler.
+ *
+ * The above pictures describe the normal state of the pointer lists.
+ * At top and bottom of the image, we diddle the pointer lists to duplicate
+ * the first or last sample row as necessary (this is cheaper than copying
+ * sample rows around).
+ *
+ * This scheme breaks down if M < 2, ie, min_DCT_v_scaled_size is 1.  In that
+ * situation each iMCU row provides only one row group so the buffering logic
+ * must be different (eg, we must read two iMCU rows before we can emit the
+ * first row group).  For now, we simply do not support providing context
+ * rows when min_DCT_v_scaled_size is 1.  That combination seems unlikely to
+ * be worth providing --- if someone wants a 1/8th-size preview, they probably
+ * want it quick and dirty, so a context-free upsampler is sufficient.
+ */
+
+
+/* Private buffer controller object */
+
+typedef struct {
+  struct jpeg_d_main_controller pub; /* public fields */
+
+  /* Pointer to allocated workspace (M or M+2 row groups). */
+  JSAMPARRAY buffer[MAX_COMPONENTS];
+
+  JDIMENSION rowgroup_ctr;	/* counts row groups output to postprocessor */
+  JDIMENSION rowgroups_avail;	/* row groups available to postprocessor */
+
+  /* Remaining fields are only used in the context case. */
+
+  boolean buffer_full;		/* Have we gotten an iMCU row from decoder? */
+
+  /* These are the master pointers to the funny-order pointer lists. */
+  JSAMPIMAGE xbuffer[2];	/* pointers to weird pointer lists */
+
+  int whichptr;			/* indicates which pointer set is now in use */
+  int context_state;		/* process_data state machine status */
+  JDIMENSION iMCU_row_ctr;	/* counts iMCU rows to detect image top/bot */
+} my_main_controller;
+
+typedef my_main_controller * my_main_ptr;
+
+/* context_state values: */
+#define CTX_PREPARE_FOR_IMCU	0	/* need to prepare for MCU row */
+#define CTX_PROCESS_IMCU	1	/* feeding iMCU to postprocessor */
+#define CTX_POSTPONED_ROW	2	/* feeding postponed row group */
+
+
+/* Forward declarations */
+METHODDEF(void) process_data_simple_main
+	JPP((j_decompress_ptr cinfo, JSAMPARRAY output_buf,
+	     JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail));
+METHODDEF(void) process_data_context_main
+	JPP((j_decompress_ptr cinfo, JSAMPARRAY output_buf,
+	     JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail));
+#ifdef QUANT_2PASS_SUPPORTED
+METHODDEF(void) process_data_crank_post
+	JPP((j_decompress_ptr cinfo, JSAMPARRAY output_buf,
+	     JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail));
+#endif
+
+
+LOCAL(void)
+alloc_funny_pointers (j_decompress_ptr cinfo)
+/* Allocate space for the funny pointer lists.
+ * This is done only once, not once per pass.
+ */
+{
+  my_main_ptr mainp = (my_main_ptr) cinfo->main;
+  int ci, rgroup;
+  int M = cinfo->min_DCT_v_scaled_size;
+  jpeg_component_info *compptr;
+  JSAMPARRAY xbuf;
+
+  /* Get top-level space for component array pointers.
+   * We alloc both arrays with one call to save a few cycles.
+   */
+  mainp->xbuffer[0] = (JSAMPIMAGE) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE,
+     cinfo->num_components * 2 * SIZEOF(JSAMPARRAY));
+  mainp->xbuffer[1] = mainp->xbuffer[0] + cinfo->num_components;
+
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    if (! compptr->component_needed)
+      continue;			/* skip uninteresting component */
+    rgroup = (compptr->v_samp_factor * compptr->DCT_v_scaled_size) /
+      cinfo->min_DCT_v_scaled_size; /* height of a row group of component */
+    /* Get space for pointer lists --- M+4 row groups in each list.
+     * We alloc both pointer lists with one call to save a few cycles.
+     */
+    xbuf = (JSAMPARRAY) (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo,
+      JPOOL_IMAGE, 2 * (rgroup * (M + 4)) * SIZEOF(JSAMPROW));
+    xbuf += rgroup;		/* want one row group at negative offsets */
+    mainp->xbuffer[0][ci] = xbuf;
+    xbuf += rgroup * (M + 4);
+    mainp->xbuffer[1][ci] = xbuf;
+  }
+}
+
+
+LOCAL(void)
+make_funny_pointers (j_decompress_ptr cinfo)
+/* Create the funny pointer lists discussed in the comments above.
+ * The actual workspace is already allocated (in mainp->buffer),
+ * and the space for the pointer lists is allocated too.
+ * This routine just fills in the curiously ordered lists.
+ * This will be repeated at the beginning of each pass.
+ */
+{
+  my_main_ptr mainp = (my_main_ptr) cinfo->main;
+  int ci, i, rgroup;
+  int M = cinfo->min_DCT_v_scaled_size;
+  jpeg_component_info *compptr;
+  JSAMPARRAY buf, xbuf0, xbuf1;
+
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    if (! compptr->component_needed)
+      continue;			/* skip uninteresting component */
+    rgroup = (compptr->v_samp_factor * compptr->DCT_v_scaled_size) /
+      cinfo->min_DCT_v_scaled_size; /* height of a row group of component */
+    xbuf0 = mainp->xbuffer[0][ci];
+    xbuf1 = mainp->xbuffer[1][ci];
+    /* First copy the workspace pointers as-is */
+    buf = mainp->buffer[ci];
+    for (i = 0; i < rgroup * (M + 2); i++) {
+      xbuf0[i] = xbuf1[i] = buf[i];
+    }
+    /* In the second list, put the last four row groups in swapped order */
+    for (i = 0; i < rgroup * 2; i++) {
+      xbuf1[rgroup*(M-2) + i] = buf[rgroup*M + i];
+      xbuf1[rgroup*M + i] = buf[rgroup*(M-2) + i];
+    }
+    /* The wraparound pointers at top and bottom will be filled later
+     * (see set_wraparound_pointers, below).  Initially we want the "above"
+     * pointers to duplicate the first actual data line.  This only needs
+     * to happen in xbuffer[0].
+     */
+    for (i = 0; i < rgroup; i++) {
+      xbuf0[i - rgroup] = xbuf0[0];
+    }
+  }
+}
+
+
+LOCAL(void)
+set_wraparound_pointers (j_decompress_ptr cinfo)
+/* Set up the "wraparound" pointers at top and bottom of the pointer lists.
+ * This changes the pointer list state from top-of-image to the normal state.
+ */
+{
+  my_main_ptr mainp = (my_main_ptr) cinfo->main;
+  int ci, i, rgroup;
+  int M = cinfo->min_DCT_v_scaled_size;
+  jpeg_component_info *compptr;
+  JSAMPARRAY xbuf0, xbuf1;
+
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    if (! compptr->component_needed)
+      continue;			/* skip uninteresting component */
+    rgroup = (compptr->v_samp_factor * compptr->DCT_v_scaled_size) /
+      cinfo->min_DCT_v_scaled_size; /* height of a row group of component */
+    xbuf0 = mainp->xbuffer[0][ci];
+    xbuf1 = mainp->xbuffer[1][ci];
+    for (i = 0; i < rgroup; i++) {
+      xbuf0[i - rgroup] = xbuf0[rgroup*(M+1) + i];
+      xbuf1[i - rgroup] = xbuf1[rgroup*(M+1) + i];
+      xbuf0[rgroup*(M+2) + i] = xbuf0[i];
+      xbuf1[rgroup*(M+2) + i] = xbuf1[i];
+    }
+  }
+}
+
+
+LOCAL(void)
+set_bottom_pointers (j_decompress_ptr cinfo)
+/* Change the pointer lists to duplicate the last sample row at the bottom
+ * of the image.  whichptr indicates which xbuffer holds the final iMCU row.
+ * Also sets rowgroups_avail to indicate number of nondummy row groups in row.
+ */
+{
+  my_main_ptr mainp = (my_main_ptr) cinfo->main;
+  int ci, i, rgroup, iMCUheight, rows_left;
+  jpeg_component_info *compptr;
+  JSAMPARRAY xbuf;
+
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    if (! compptr->component_needed)
+      continue;			/* skip uninteresting component */
+    /* Count sample rows in one iMCU row and in one row group */
+    iMCUheight = compptr->v_samp_factor * compptr->DCT_v_scaled_size;
+    rgroup = iMCUheight / cinfo->min_DCT_v_scaled_size;
+    /* Count nondummy sample rows remaining for this component */
+    rows_left = (int) (compptr->downsampled_height % (JDIMENSION) iMCUheight);
+    if (rows_left == 0) rows_left = iMCUheight;
+    /* Count nondummy row groups.  Should get same answer for each component,
+     * so we need only do it once.
+     */
+    if (ci == 0) {
+      mainp->rowgroups_avail = (JDIMENSION) ((rows_left-1) / rgroup + 1);
+    }
+    /* Duplicate the last real sample row rgroup*2 times; this pads out the
+     * last partial rowgroup and ensures at least one full rowgroup of context.
+     */
+    xbuf = mainp->xbuffer[mainp->whichptr][ci];
+    for (i = 0; i < rgroup * 2; i++) {
+      xbuf[rows_left + i] = xbuf[rows_left-1];
+    }
+  }
+}
+
+
+/*
+ * Initialize for a processing pass.
+ */
+
+METHODDEF(void)
+start_pass_main (j_decompress_ptr cinfo, J_BUF_MODE pass_mode)
+{
+  my_main_ptr mainp = (my_main_ptr) cinfo->main;
+
+  switch (pass_mode) {
+  case JBUF_PASS_THRU:
+    if (cinfo->upsample->need_context_rows) {
+      mainp->pub.process_data = process_data_context_main;
+      make_funny_pointers(cinfo); /* Create the xbuffer[] lists */
+      mainp->whichptr = 0;	/* Read first iMCU row into xbuffer[0] */
+      mainp->context_state = CTX_PREPARE_FOR_IMCU;
+      mainp->iMCU_row_ctr = 0;
+      mainp->buffer_full = FALSE; /* Mark buffer empty */
+    } else {
+      /* Simple case with no context needed */
+      mainp->pub.process_data = process_data_simple_main;
+      mainp->rowgroup_ctr = mainp->rowgroups_avail; /* Mark buffer empty */
+    }
+    break;
+#ifdef QUANT_2PASS_SUPPORTED
+  case JBUF_CRANK_DEST:
+    /* For last pass of 2-pass quantization, just crank the postprocessor */
+    mainp->pub.process_data = process_data_crank_post;
+    break;
+#endif
+  default:
+    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+  }
+}
+
+
+/*
+ * Process some data.
+ * This handles the simple case where no context is required.
+ */
+
+METHODDEF(void)
+process_data_simple_main (j_decompress_ptr cinfo, JSAMPARRAY output_buf,
+			  JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail)
+{
+  my_main_ptr mainp = (my_main_ptr) cinfo->main;
+
+  /* Read input data if we haven't filled the main buffer yet */
+  if (mainp->rowgroup_ctr >= mainp->rowgroups_avail) {
+    if (! (*cinfo->coef->decompress_data) (cinfo, mainp->buffer))
+      return;			/* suspension forced, can do nothing more */
+    mainp->rowgroup_ctr = 0;	/* OK, we have an iMCU row to work with */
+  }
+
+  /* Note: at the bottom of the image, we may pass extra garbage row groups
+   * to the postprocessor.  The postprocessor has to check for bottom
+   * of image anyway (at row resolution), so no point in us doing it too.
+   */
+
+  /* Feed the postprocessor */
+  (*cinfo->post->post_process_data) (cinfo, mainp->buffer,
+			&mainp->rowgroup_ctr, mainp->rowgroups_avail,
+			output_buf, out_row_ctr, out_rows_avail);
+}
+
+
+/*
+ * Process some data.
+ * This handles the case where context rows must be provided.
+ */
+
+METHODDEF(void)
+process_data_context_main (j_decompress_ptr cinfo, JSAMPARRAY output_buf,
+			   JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail)
+{
+  my_main_ptr mainp = (my_main_ptr) cinfo->main;
+
+  /* Read input data if we haven't filled the main buffer yet */
+  if (! mainp->buffer_full) {
+    if (! (*cinfo->coef->decompress_data) (cinfo,
+					   mainp->xbuffer[mainp->whichptr]))
+      return;			/* suspension forced, can do nothing more */
+    mainp->buffer_full = TRUE;	/* OK, we have an iMCU row to work with */
+    mainp->iMCU_row_ctr++;	/* count rows received */
+  }
+
+  /* Postprocessor typically will not swallow all the input data it is handed
+   * in one call (due to filling the output buffer first).  Must be prepared
+   * to exit and restart.  This switch lets us keep track of how far we got.
+   * Note that each case falls through to the next on successful completion.
+   */
+  switch (mainp->context_state) {
+  case CTX_POSTPONED_ROW:
+    /* Call postprocessor using previously set pointers for postponed row */
+    (*cinfo->post->post_process_data) (cinfo, mainp->xbuffer[mainp->whichptr],
+			&mainp->rowgroup_ctr, mainp->rowgroups_avail,
+			output_buf, out_row_ctr, out_rows_avail);
+    if (mainp->rowgroup_ctr < mainp->rowgroups_avail)
+      return;			/* Need to suspend */
+    mainp->context_state = CTX_PREPARE_FOR_IMCU;
+    if (*out_row_ctr >= out_rows_avail)
+      return;			/* Postprocessor exactly filled output buf */
+    /*FALLTHROUGH*/
+  case CTX_PREPARE_FOR_IMCU:
+    /* Prepare to process first M-1 row groups of this iMCU row */
+    mainp->rowgroup_ctr = 0;
+    mainp->rowgroups_avail = (JDIMENSION) (cinfo->min_DCT_v_scaled_size - 1);
+    /* Check for bottom of image: if so, tweak pointers to "duplicate"
+     * the last sample row, and adjust rowgroups_avail to ignore padding rows.
+     */
+    if (mainp->iMCU_row_ctr == cinfo->total_iMCU_rows)
+      set_bottom_pointers(cinfo);
+    mainp->context_state = CTX_PROCESS_IMCU;
+    /*FALLTHROUGH*/
+  case CTX_PROCESS_IMCU:
+    /* Call postprocessor using previously set pointers */
+    (*cinfo->post->post_process_data) (cinfo, mainp->xbuffer[mainp->whichptr],
+			&mainp->rowgroup_ctr, mainp->rowgroups_avail,
+			output_buf, out_row_ctr, out_rows_avail);
+    if (mainp->rowgroup_ctr < mainp->rowgroups_avail)
+      return;			/* Need to suspend */
+    /* After the first iMCU, change wraparound pointers to normal state */
+    if (mainp->iMCU_row_ctr == 1)
+      set_wraparound_pointers(cinfo);
+    /* Prepare to load new iMCU row using other xbuffer list */
+    mainp->whichptr ^= 1;	/* 0=>1 or 1=>0 */
+    mainp->buffer_full = FALSE;
+    /* Still need to process last row group of this iMCU row, */
+    /* which is saved at index M+1 of the other xbuffer */
+    mainp->rowgroup_ctr = (JDIMENSION) (cinfo->min_DCT_v_scaled_size + 1);
+    mainp->rowgroups_avail = (JDIMENSION) (cinfo->min_DCT_v_scaled_size + 2);
+    mainp->context_state = CTX_POSTPONED_ROW;
+  }
+}
+
+
+/*
+ * Process some data.
+ * Final pass of two-pass quantization: just call the postprocessor.
+ * Source data will be the postprocessor controller's internal buffer.
+ */
+
+#ifdef QUANT_2PASS_SUPPORTED
+
+METHODDEF(void)
+process_data_crank_post (j_decompress_ptr cinfo, JSAMPARRAY output_buf,
+			 JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail)
+{
+  (*cinfo->post->post_process_data) (cinfo, (JSAMPIMAGE) NULL,
+			(JDIMENSION *) NULL, (JDIMENSION) 0,
+			output_buf, out_row_ctr, out_rows_avail);
+}
+
+#endif /* QUANT_2PASS_SUPPORTED */
+
+
+/*
+ * Initialize main buffer controller.
+ */
+
+GLOBAL(void)
+jinit_d_main_controller (j_decompress_ptr cinfo, boolean need_full_buffer)
+{
+  my_main_ptr mainp;
+  int ci, rgroup, ngroups;
+  jpeg_component_info *compptr;
+
+  mainp = (my_main_ptr) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, SIZEOF(my_main_controller));
+  cinfo->main = &mainp->pub;
+  mainp->pub.start_pass = start_pass_main;
+
+  if (need_full_buffer)		/* shouldn't happen */
+    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+
+  /* Allocate the workspace.
+   * ngroups is the number of row groups we need.
+   */
+  if (cinfo->upsample->need_context_rows) {
+    if (cinfo->min_DCT_v_scaled_size < 2) /* unsupported, see comments above */
+      ERREXIT(cinfo, JERR_NOTIMPL);
+    alloc_funny_pointers(cinfo); /* Alloc space for xbuffer[] lists */
+    ngroups = cinfo->min_DCT_v_scaled_size + 2;
+  } else {
+    /* There are always min_DCT_v_scaled_size row groups in an iMCU row. */
+    ngroups = cinfo->min_DCT_v_scaled_size;
+    mainp->rowgroups_avail = (JDIMENSION) ngroups;
+  }
+
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    if (! compptr->component_needed)
+      continue;			/* skip uninteresting component */
+    rgroup = (compptr->v_samp_factor * compptr->DCT_v_scaled_size) /
+      cinfo->min_DCT_v_scaled_size; /* height of a row group of component */
+    mainp->buffer[ci] = (*cinfo->mem->alloc_sarray)
+      ((j_common_ptr) cinfo, JPOOL_IMAGE,
+       compptr->width_in_blocks * ((JDIMENSION) compptr->DCT_h_scaled_size),
+       (JDIMENSION) (rgroup * ngroups));
+  }
+}
@@ -0,0 +1,532 @@
+/*
+ * jdmaster.c
+ *
+ * Copyright (C) 1991-1997, Thomas G. Lane.
+ * Modified 2002-2020 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains master control logic for the JPEG decompressor.
+ * These routines are concerned with selecting the modules to be executed
+ * and with determining the number of passes and the work to be done in each
+ * pass.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/* Private state */
+
+typedef struct {
+  struct jpeg_decomp_master pub; /* public fields */
+
+  int pass_number;		/* # of passes completed */
+
+  boolean using_merged_upsample; /* TRUE if using merged upsample/cconvert */
+
+  /* Saved references to initialized quantizer modules,
+   * in case we need to switch modes.
+   */
+  struct jpeg_color_quantizer * quantizer_1pass;
+  struct jpeg_color_quantizer * quantizer_2pass;
+} my_decomp_master;
+
+typedef my_decomp_master * my_master_ptr;
+
+
+/*
+ * Determine whether merged upsample/color conversion should be used.
+ * CRUCIAL: this must match the actual capabilities of jdmerge.c!
+ */
+
+LOCAL(boolean)
+use_merged_upsample (j_decompress_ptr cinfo)
+{
+#ifdef UPSAMPLE_MERGING_SUPPORTED
+  /* Merging is the equivalent of plain box-filter upsampling. */
+  /* The following condition is only needed if fancy shall select
+   * a different upsampling method.  In our current implementation
+   * fancy only affects the DCT scaling, thus we can use fancy
+   * upsampling and merged upsample simultaneously, in particular
+   * with scaled DCT sizes larger than the default DCTSIZE.
+   */
+#if 0
+  if (cinfo->do_fancy_upsampling)
+    return FALSE;
+#endif
+  if (cinfo->CCIR601_sampling)
+    return FALSE;
+  /* jdmerge.c only supports YCC=>RGB color conversion */
+  if ((cinfo->jpeg_color_space != JCS_YCbCr &&
+       cinfo->jpeg_color_space != JCS_BG_YCC) ||
+      cinfo->num_components != 3 ||
+      cinfo->out_color_space != JCS_RGB ||
+      cinfo->out_color_components != RGB_PIXELSIZE ||
+      cinfo->color_transform)
+    return FALSE;
+  /* and it only handles 2h1v or 2h2v sampling ratios */
+  if (cinfo->comp_info[0].h_samp_factor != 2 ||
+      cinfo->comp_info[1].h_samp_factor != 1 ||
+      cinfo->comp_info[2].h_samp_factor != 1 ||
+      cinfo->comp_info[0].v_samp_factor >  2 ||
+      cinfo->comp_info[1].v_samp_factor != 1 ||
+      cinfo->comp_info[2].v_samp_factor != 1)
+    return FALSE;
+  /* furthermore, it doesn't work if we've scaled the IDCTs differently */
+  if (cinfo->comp_info[0].DCT_h_scaled_size != cinfo->min_DCT_h_scaled_size ||
+      cinfo->comp_info[1].DCT_h_scaled_size != cinfo->min_DCT_h_scaled_size ||
+      cinfo->comp_info[2].DCT_h_scaled_size != cinfo->min_DCT_h_scaled_size ||
+      cinfo->comp_info[0].DCT_v_scaled_size != cinfo->min_DCT_v_scaled_size ||
+      cinfo->comp_info[1].DCT_v_scaled_size != cinfo->min_DCT_v_scaled_size ||
+      cinfo->comp_info[2].DCT_v_scaled_size != cinfo->min_DCT_v_scaled_size)
+    return FALSE;
+  /* ??? also need to test for upsample-time rescaling, when & if supported */
+  return TRUE;			/* by golly, it'll work... */
+#else
+  return FALSE;
+#endif
+}
+
+
+/*
+ * Compute output image dimensions and related values.
+ * NOTE: this is exported for possible use by application.
+ * Hence it mustn't do anything that can't be done twice.
+ * Also note that it may be called before the master module is initialized!
+ */
+
+GLOBAL(void)
+jpeg_calc_output_dimensions (j_decompress_ptr cinfo)
+/* Do computations that are needed before master selection phase.
+ * This function is used for full decompression.
+ */
+{
+  int ci, i;
+  jpeg_component_info *compptr;
+
+  /* Prevent application from calling me at wrong times */
+  if (cinfo->global_state != DSTATE_READY)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+
+  /* Compute core output image dimensions and DCT scaling choices. */
+  jpeg_core_output_dimensions(cinfo);
+
+#ifdef IDCT_SCALING_SUPPORTED
+
+  /* In selecting the actual DCT scaling for each component, we try to
+   * scale up the chroma components via IDCT scaling rather than upsampling.
+   * This saves time if the upsampler gets to use 1:1 scaling.
+   * Note this code adapts subsampling ratios which are powers of 2.
+   */
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    int ssize = 1;
+    if (! cinfo->raw_data_out)
+      while (cinfo->min_DCT_h_scaled_size * ssize <=
+	     (cinfo->do_fancy_upsampling ? DCTSIZE : DCTSIZE / 2) &&
+	     (cinfo->max_h_samp_factor % (compptr->h_samp_factor * ssize * 2)) ==
+	     0) {
+	ssize = ssize * 2;
+      }
+    compptr->DCT_h_scaled_size = cinfo->min_DCT_h_scaled_size * ssize;
+    ssize = 1;
+    if (! cinfo->raw_data_out)
+      while (cinfo->min_DCT_v_scaled_size * ssize <=
+	     (cinfo->do_fancy_upsampling ? DCTSIZE : DCTSIZE / 2) &&
+	     (cinfo->max_v_samp_factor % (compptr->v_samp_factor * ssize * 2)) ==
+	     0) {
+	ssize = ssize * 2;
+      }
+    compptr->DCT_v_scaled_size = cinfo->min_DCT_v_scaled_size * ssize;
+
+    /* We don't support IDCT ratios larger than 2. */
+    if (compptr->DCT_h_scaled_size > compptr->DCT_v_scaled_size * 2)
+	compptr->DCT_h_scaled_size = compptr->DCT_v_scaled_size * 2;
+    else if (compptr->DCT_v_scaled_size > compptr->DCT_h_scaled_size * 2)
+	compptr->DCT_v_scaled_size = compptr->DCT_h_scaled_size * 2;
+
+    /* Recompute downsampled dimensions of components;
+     * application needs to know these if using raw downsampled data.
+     */
+    /* Size in samples, after IDCT scaling */
+    compptr->downsampled_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width *
+		    (long) (compptr->h_samp_factor * compptr->DCT_h_scaled_size),
+		    (long) (cinfo->max_h_samp_factor * cinfo->block_size));
+    compptr->downsampled_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height *
+		    (long) (compptr->v_samp_factor * compptr->DCT_v_scaled_size),
+		    (long) (cinfo->max_v_samp_factor * cinfo->block_size));
+  }
+
+#endif /* IDCT_SCALING_SUPPORTED */
+
+  /* Report number of components in selected colorspace. */
+  /* This should correspond to the actual code in the color conversion module. */
+  switch (cinfo->out_color_space) {
+  case JCS_GRAYSCALE:
+    cinfo->out_color_components = 1;
+    break;
+  case JCS_RGB:
+  case JCS_BG_RGB:
+    cinfo->out_color_components = RGB_PIXELSIZE;
+    break;
+  default:	/* YCCK <=> CMYK conversion or same colorspace as in file */
+    i = 0;
+    for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+	 ci++, compptr++)
+      if (compptr->component_needed)
+	i++;	/* count output color components */
+    cinfo->out_color_components = i;
+  }
+  cinfo->output_components = (cinfo->quantize_colors ? 1 :
+			      cinfo->out_color_components);
+
+  /* See if upsampler will want to emit more than one row at a time */
+  if (use_merged_upsample(cinfo))
+    cinfo->rec_outbuf_height = cinfo->max_v_samp_factor;
+  else
+    cinfo->rec_outbuf_height = 1;
+}
+
+
+/*
+ * Several decompression processes need to range-limit values to the range
+ * 0..MAXJSAMPLE; the input value may fall somewhat outside this range
+ * due to noise introduced by quantization, roundoff error, etc.  These
+ * processes are inner loops and need to be as fast as possible.  On most
+ * machines, particularly CPUs with pipelines or instruction prefetch,
+ * a (subscript-check-less) C table lookup
+ *		x = sample_range_limit[x];
+ * is faster than explicit tests
+ *		if (x < 0)  x = 0;
+ *		else if (x > MAXJSAMPLE)  x = MAXJSAMPLE;
+ * These processes all use a common table prepared by the routine below.
+ *
+ * For most steps we can mathematically guarantee that the initial value
+ * of x is within 2*(MAXJSAMPLE+1) of the legal range, so a table running
+ * from -2*(MAXJSAMPLE+1) to 3*MAXJSAMPLE+2 is sufficient.  But for the
+ * initial limiting step (just after the IDCT), a wildly out-of-range value
+ * is possible if the input data is corrupt.  To avoid any chance of indexing
+ * off the end of memory and getting a bad-pointer trap, we perform the
+ * post-IDCT limiting thus:
+ *		x = (sample_range_limit - SUBSET)[(x + CENTER) & MASK];
+ * where MASK is 2 bits wider than legal sample data, ie 10 bits for 8-bit
+ * samples.  Under normal circumstances this is more than enough range and
+ * a correct output will be generated; with bogus input data the mask will
+ * cause wraparound, and we will safely generate a bogus-but-in-range output.
+ * For the post-IDCT step, we want to convert the data from signed to unsigned
+ * representation by adding CENTERJSAMPLE at the same time that we limit it.
+ * This is accomplished with SUBSET = CENTER - CENTERJSAMPLE.
+ *
+ * Note that the table is allocated in near data space on PCs; it's small
+ * enough and used often enough to justify this.
+ */
+
+LOCAL(void)
+prepare_range_limit_table (j_decompress_ptr cinfo)
+/* Allocate and fill in the sample_range_limit table */
+{
+  JSAMPLE * table;
+  int i;
+
+  table = (JSAMPLE *) (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo,
+    JPOOL_IMAGE, (RANGE_CENTER * 2 + MAXJSAMPLE + 1) * SIZEOF(JSAMPLE));
+  /* First segment of range limit table: limit[x] = 0 for x < 0 */
+  MEMZERO(table, RANGE_CENTER * SIZEOF(JSAMPLE));
+  table += RANGE_CENTER;	/* allow negative subscripts of table */
+  cinfo->sample_range_limit = table;
+  /* Main part of range limit table: limit[x] = x */
+  for (i = 0; i <= MAXJSAMPLE; i++)
+    table[i] = (JSAMPLE) i;
+  /* End of range limit table: limit[x] = MAXJSAMPLE for x > MAXJSAMPLE */
+  for (; i <=  MAXJSAMPLE + RANGE_CENTER; i++)
+    table[i] = MAXJSAMPLE;
+}
+
+
+/*
+ * Master selection of decompression modules.
+ * This is done once at jpeg_start_decompress time.  We determine
+ * which modules will be used and give them appropriate initialization calls.
+ * We also initialize the decompressor input side to begin consuming data.
+ *
+ * Since jpeg_read_header has finished, we know what is in the SOF
+ * and (first) SOS markers.  We also have all the application parameter
+ * settings.
+ */
+
+LOCAL(void)
+master_selection (j_decompress_ptr cinfo)
+{
+  my_master_ptr master = (my_master_ptr) cinfo->master;
+  boolean use_c_buffer;
+  long samplesperrow;
+  JDIMENSION jd_samplesperrow;
+
+  /* For now, precision must match compiled-in value... */
+  if (cinfo->data_precision != BITS_IN_JSAMPLE)
+    ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision);
+
+  /* Initialize dimensions and other stuff */
+  jpeg_calc_output_dimensions(cinfo);
+  prepare_range_limit_table(cinfo);
+
+  /* Sanity check on image dimensions */
+  if (cinfo->output_height <= 0 || cinfo->output_width <= 0 ||
+      cinfo->out_color_components <= 0)
+    ERREXIT(cinfo, JERR_EMPTY_IMAGE);
+
+  /* Width of an output scanline must be representable as JDIMENSION. */
+  samplesperrow = (long) cinfo->output_width * (long) cinfo->out_color_components;
+  jd_samplesperrow = (JDIMENSION) samplesperrow;
+  if ((long) jd_samplesperrow != samplesperrow)
+    ERREXIT(cinfo, JERR_WIDTH_OVERFLOW);
+
+  /* Initialize my private state */
+  master->pass_number = 0;
+  master->using_merged_upsample = use_merged_upsample(cinfo);
+
+  /* Color quantizer selection */
+  master->quantizer_1pass = NULL;
+  master->quantizer_2pass = NULL;
+  /* No mode changes if not using buffered-image mode. */
+  if (! cinfo->quantize_colors || ! cinfo->buffered_image) {
+    cinfo->enable_1pass_quant = FALSE;
+    cinfo->enable_external_quant = FALSE;
+    cinfo->enable_2pass_quant = FALSE;
+  }
+  if (cinfo->quantize_colors) {
+    if (cinfo->raw_data_out)
+      ERREXIT(cinfo, JERR_NOTIMPL);
+    /* 2-pass quantizer only works in 3-component color space. */
+    if (cinfo->out_color_components != 3) {
+      cinfo->enable_1pass_quant = TRUE;
+      cinfo->enable_external_quant = FALSE;
+      cinfo->enable_2pass_quant = FALSE;
+      cinfo->colormap = NULL;
+    } else if (cinfo->colormap != NULL) {
+      cinfo->enable_external_quant = TRUE;
+    } else if (cinfo->two_pass_quantize) {
+      cinfo->enable_2pass_quant = TRUE;
+    } else {
+      cinfo->enable_1pass_quant = TRUE;
+    }
+
+    if (cinfo->enable_1pass_quant) {
+#ifdef QUANT_1PASS_SUPPORTED
+      jinit_1pass_quantizer(cinfo);
+      master->quantizer_1pass = cinfo->cquantize;
+#else
+      ERREXIT(cinfo, JERR_NOT_COMPILED);
+#endif
+    }
+
+    /* We use the 2-pass code to map to external colormaps. */
+    if (cinfo->enable_2pass_quant || cinfo->enable_external_quant) {
+#ifdef QUANT_2PASS_SUPPORTED
+      jinit_2pass_quantizer(cinfo);
+      master->quantizer_2pass = cinfo->cquantize;
+#else
+      ERREXIT(cinfo, JERR_NOT_COMPILED);
+#endif
+    }
+    /* If both quantizers are initialized, the 2-pass one is left active;
+     * this is necessary for starting with quantization to an external map.
+     */
+  }
+
+  /* Post-processing: in particular, color conversion first */
+  if (! cinfo->raw_data_out) {
+    if (master->using_merged_upsample) {
+#ifdef UPSAMPLE_MERGING_SUPPORTED
+      jinit_merged_upsampler(cinfo); /* does color conversion too */
+#else
+      ERREXIT(cinfo, JERR_NOT_COMPILED);
+#endif
+    } else {
+      jinit_color_deconverter(cinfo);
+      jinit_upsampler(cinfo);
+    }
+    jinit_d_post_controller(cinfo, cinfo->enable_2pass_quant);
+  }
+  /* Inverse DCT */
+  jinit_inverse_dct(cinfo);
+  /* Entropy decoding: either Huffman or arithmetic coding. */
+  if (cinfo->arith_code)
+    jinit_arith_decoder(cinfo);
+  else {
+    jinit_huff_decoder(cinfo);
+  }
+
+  /* Initialize principal buffer controllers. */
+  use_c_buffer = cinfo->inputctl->has_multiple_scans || cinfo->buffered_image;
+  jinit_d_coef_controller(cinfo, use_c_buffer);
+
+  if (! cinfo->raw_data_out)
+    jinit_d_main_controller(cinfo, FALSE /* never need full buffer here */);
+
+  /* We can now tell the memory manager to allocate virtual arrays. */
+  (*cinfo->mem->realize_virt_arrays) ((j_common_ptr) cinfo);
+
+  /* Initialize input side of decompressor to consume first scan. */
+  (*cinfo->inputctl->start_input_pass) (cinfo);
+
+#ifdef D_MULTISCAN_FILES_SUPPORTED
+  /* If jpeg_start_decompress will read the whole file, initialize
+   * progress monitoring appropriately.  The input step is counted
+   * as one pass.
+   */
+  if (cinfo->progress != NULL && ! cinfo->buffered_image &&
+      cinfo->inputctl->has_multiple_scans) {
+    int nscans;
+    /* Estimate number of scans to set pass_limit. */
+    if (cinfo->progressive_mode) {
+      /* Arbitrarily estimate 2 interleaved DC scans + 3 AC scans/component. */
+      nscans = 2 + 3 * cinfo->num_components;
+    } else {
+      /* For a nonprogressive multiscan file, estimate 1 scan per component. */
+      nscans = cinfo->num_components;
+    }
+    cinfo->progress->pass_counter = 0L;
+    cinfo->progress->pass_limit = (long) cinfo->total_iMCU_rows * nscans;
+    cinfo->progress->completed_passes = 0;
+    cinfo->progress->total_passes = (cinfo->enable_2pass_quant ? 3 : 2);
+    /* Count the input pass as done */
+    master->pass_number++;
+  }
+#endif /* D_MULTISCAN_FILES_SUPPORTED */
+}
+
+
+/*
+ * Per-pass setup.
+ * This is called at the beginning of each output pass.  We determine which
+ * modules will be active during this pass and give them appropriate
+ * start_pass calls.  We also set is_dummy_pass to indicate whether this
+ * is a "real" output pass or a dummy pass for color quantization.
+ * (In the latter case, jdapistd.c will crank the pass to completion.)
+ */
+
+METHODDEF(void)
+prepare_for_output_pass (j_decompress_ptr cinfo)
+{
+  my_master_ptr master = (my_master_ptr) cinfo->master;
+
+  if (master->pub.is_dummy_pass) {
+#ifdef QUANT_2PASS_SUPPORTED
+    /* Final pass of 2-pass quantization */
+    master->pub.is_dummy_pass = FALSE;
+    (*cinfo->cquantize->start_pass) (cinfo, FALSE);
+    (*cinfo->post->start_pass) (cinfo, JBUF_CRANK_DEST);
+    (*cinfo->main->start_pass) (cinfo, JBUF_CRANK_DEST);
+#else
+    ERREXIT(cinfo, JERR_NOT_COMPILED);
+#endif /* QUANT_2PASS_SUPPORTED */
+  } else {
+    if (cinfo->quantize_colors && cinfo->colormap == NULL) {
+      /* Select new quantization method */
+      if (cinfo->two_pass_quantize && cinfo->enable_2pass_quant) {
+	cinfo->cquantize = master->quantizer_2pass;
+	master->pub.is_dummy_pass = TRUE;
+      } else if (cinfo->enable_1pass_quant) {
+	cinfo->cquantize = master->quantizer_1pass;
+      } else {
+	ERREXIT(cinfo, JERR_MODE_CHANGE);
+      }
+    }
+    (*cinfo->idct->start_pass) (cinfo);
+    (*cinfo->coef->start_output_pass) (cinfo);
+    if (! cinfo->raw_data_out) {
+      if (! master->using_merged_upsample)
+	(*cinfo->cconvert->start_pass) (cinfo);
+      (*cinfo->upsample->start_pass) (cinfo);
+      if (cinfo->quantize_colors)
+	(*cinfo->cquantize->start_pass) (cinfo, master->pub.is_dummy_pass);
+      (*cinfo->post->start_pass) (cinfo,
+	    (master->pub.is_dummy_pass ? JBUF_SAVE_AND_PASS : JBUF_PASS_THRU));
+      (*cinfo->main->start_pass) (cinfo, JBUF_PASS_THRU);
+    }
+  }
+
+  /* Set up progress monitor's pass info if present */
+  if (cinfo->progress != NULL) {
+    cinfo->progress->completed_passes = master->pass_number;
+    cinfo->progress->total_passes = master->pass_number +
+				    (master->pub.is_dummy_pass ? 2 : 1);
+    /* In buffered-image mode, we assume one more output pass if EOI not
+     * yet reached, but no more passes if EOI has been reached.
+     */
+    if (cinfo->buffered_image && ! cinfo->inputctl->eoi_reached) {
+      cinfo->progress->total_passes += (cinfo->enable_2pass_quant ? 2 : 1);
+    }
+  }
+}
+
+
+/*
+ * Finish up at end of an output pass.
+ */
+
+METHODDEF(void)
+finish_output_pass (j_decompress_ptr cinfo)
+{
+  my_master_ptr master = (my_master_ptr) cinfo->master;
+
+  if (cinfo->quantize_colors)
+    (*cinfo->cquantize->finish_pass) (cinfo);
+  master->pass_number++;
+}
+
+
+#ifdef D_MULTISCAN_FILES_SUPPORTED
+
+/*
+ * Switch to a new external colormap between output passes.
+ */
+
+GLOBAL(void)
+jpeg_new_colormap (j_decompress_ptr cinfo)
+{
+  my_master_ptr master = (my_master_ptr) cinfo->master;
+
+  /* Prevent application from calling me at wrong times */
+  if (cinfo->global_state != DSTATE_BUFIMAGE)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+
+  if (cinfo->quantize_colors && cinfo->enable_external_quant &&
+      cinfo->colormap != NULL) {
+    /* Select 2-pass quantizer for external colormap use */
+    cinfo->cquantize = master->quantizer_2pass;
+    /* Notify quantizer of colormap change */
+    (*cinfo->cquantize->new_color_map) (cinfo);
+    master->pub.is_dummy_pass = FALSE; /* just in case */
+  } else
+    ERREXIT(cinfo, JERR_MODE_CHANGE);
+}
+
+#endif /* D_MULTISCAN_FILES_SUPPORTED */
+
+
+/*
+ * Initialize master decompression control and select active modules.
+ * This is performed at the start of jpeg_start_decompress.
+ */
+
+GLOBAL(void)
+jinit_master_decompress (j_decompress_ptr cinfo)
+{
+  my_master_ptr master;
+
+  master = (my_master_ptr) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, SIZEOF(my_decomp_master));
+  cinfo->master = &master->pub;
+  master->pub.prepare_for_output_pass = prepare_for_output_pass;
+  master->pub.finish_output_pass = finish_output_pass;
+
+  master->pub.is_dummy_pass = FALSE;
+
+  master_selection(cinfo);
+}
@@ -0,0 +1,437 @@
+/*
+ * jdmerge.c
+ *
+ * Copyright (C) 1994-1996, Thomas G. Lane.
+ * Modified 2013-2022 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains code for merged upsampling/color conversion.
+ *
+ * This file combines functions from jdsample.c and jdcolor.c;
+ * read those files first to understand what's going on.
+ *
+ * When the chroma components are to be upsampled by simple replication
+ * (ie, box filtering), we can save some work in color conversion by
+ * calculating all the output pixels corresponding to a pair of chroma
+ * samples at one time.  In the conversion equations
+ *	R = Y           + K1 * Cr
+ *	G = Y + K2 * Cb + K3 * Cr
+ *	B = Y + K4 * Cb
+ * only the Y term varies among the group of pixels corresponding to a pair
+ * of chroma samples, so the rest of the terms can be calculated just once.
+ * At typical sampling ratios, this eliminates half or three-quarters
+ * of the multiplications needed for color conversion.
+ *
+ * This file currently provides implementations for the following cases:
+ *	YCC => RGB color conversion only (YCbCr or BG_YCC).
+ *	Sampling ratios of 2h1v or 2h2v.
+ *	No scaling needed at upsample time.
+ *	Corner-aligned (non-CCIR601) sampling alignment.
+ * Other special cases could be added, but in most applications these
+ * are the only common cases.  (For uncommon cases we fall back on
+ * the more general code in jdsample.c and jdcolor.c.)
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+#ifdef UPSAMPLE_MERGING_SUPPORTED
+
+
+#if RANGE_BITS < 2
+  /* Deliberate syntax err */
+  Sorry, this code requires 2 or more range extension bits.
+#endif
+
+
+/* Private subobject */
+
+typedef struct {
+  struct jpeg_upsampler pub;	/* public fields */
+
+  /* Pointer to routine to do actual upsampling/conversion of one row group */
+  JMETHOD(void, upmethod, (j_decompress_ptr cinfo,
+			   JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr,
+			   JSAMPARRAY output_buf));
+
+  /* Private state for YCC->RGB conversion */
+  int * Cr_r_tab;		/* => table for Cr to R conversion */
+  int * Cb_b_tab;		/* => table for Cb to B conversion */
+  INT32 * Cr_g_tab;		/* => table for Cr to G conversion */
+  INT32 * Cb_g_tab;		/* => table for Cb to G conversion */
+
+  /* For 2:1 vertical sampling, we produce two output rows at a time.
+   * We need a "spare" row buffer to hold the second output row if the
+   * application provides just a one-row buffer; we also use the spare
+   * to discard the dummy last row if the image height is odd.
+   */
+  JSAMPROW spare_row;
+  boolean spare_full;		/* T if spare buffer is occupied */
+
+  JDIMENSION out_row_width;	/* samples per output row */
+  JDIMENSION rows_to_go;	/* counts rows remaining in image */
+} my_upsampler;
+
+typedef my_upsampler * my_upsample_ptr;
+
+#define SCALEBITS	16	/* speediest right-shift on some machines */
+#define ONE_HALF	((INT32) 1 << (SCALEBITS-1))
+#define FIX(x)		((INT32) ((x) * (1L<<SCALEBITS) + 0.5))
+
+
+/*
+ * Initialize tables for YCbCr->RGB and BG_YCC->RGB colorspace conversion.
+ * This is taken directly from jdcolor.c; see that file for more info.
+ */
+
+LOCAL(void)
+build_ycc_rgb_table (j_decompress_ptr cinfo)
+/* Normal case, sYCC */
+{
+  my_upsample_ptr upsample = (my_upsample_ptr) cinfo->upsample;
+  int i;
+  INT32 x;
+  SHIFT_TEMPS
+
+  upsample->Cr_r_tab = (int *) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, (MAXJSAMPLE+1) * SIZEOF(int));
+  upsample->Cb_b_tab = (int *) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, (MAXJSAMPLE+1) * SIZEOF(int));
+  upsample->Cr_g_tab = (INT32 *) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, (MAXJSAMPLE+1) * SIZEOF(INT32));
+  upsample->Cb_g_tab = (INT32 *) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, (MAXJSAMPLE+1) * SIZEOF(INT32));
+
+  for (i = 0, x = -CENTERJSAMPLE; i <= MAXJSAMPLE; i++, x++) {
+    /* i is the actual input pixel value, in the range 0..MAXJSAMPLE */
+    /* The Cb or Cr value we are thinking of is x = i - CENTERJSAMPLE */
+    /* Cr=>R value is nearest int to 1.402 * x */
+    upsample->Cr_r_tab[i] = (int) DESCALE(FIX(1.402) * x, SCALEBITS);
+    /* Cb=>B value is nearest int to 1.772 * x */
+    upsample->Cb_b_tab[i] = (int) DESCALE(FIX(1.772) * x, SCALEBITS);
+    /* Cr=>G value is scaled-up -0.714136286 * x */
+    upsample->Cr_g_tab[i] = (- FIX(0.714136286)) * x;
+    /* Cb=>G value is scaled-up -0.344136286 * x */
+    /* We also add in ONE_HALF so that need not do it in inner loop */
+    upsample->Cb_g_tab[i] = (- FIX(0.344136286)) * x + ONE_HALF;
+  }
+}
+
+
+LOCAL(void)
+build_bg_ycc_rgb_table (j_decompress_ptr cinfo)
+/* Wide gamut case, bg-sYCC */
+{
+  my_upsample_ptr upsample = (my_upsample_ptr) cinfo->upsample;
+  int i;
+  INT32 x;
+  SHIFT_TEMPS
+
+  upsample->Cr_r_tab = (int *) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, (MAXJSAMPLE+1) * SIZEOF(int));
+  upsample->Cb_b_tab = (int *) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, (MAXJSAMPLE+1) * SIZEOF(int));
+  upsample->Cr_g_tab = (INT32 *) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, (MAXJSAMPLE+1) * SIZEOF(INT32));
+  upsample->Cb_g_tab = (INT32 *) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, (MAXJSAMPLE+1) * SIZEOF(INT32));
+
+  for (i = 0, x = -CENTERJSAMPLE; i <= MAXJSAMPLE; i++, x++) {
+    /* i is the actual input pixel value, in the range 0..MAXJSAMPLE */
+    /* The Cb or Cr value we are thinking of is x = i - CENTERJSAMPLE */
+    /* Cr=>R value is nearest int to 2.804 * x */
+    upsample->Cr_r_tab[i] = (int) DESCALE(FIX(2.804) * x, SCALEBITS);
+    /* Cb=>B value is nearest int to 3.544 * x */
+    upsample->Cb_b_tab[i] = (int) DESCALE(FIX(3.544) * x, SCALEBITS);
+    /* Cr=>G value is scaled-up -1.428272572 * x */
+    upsample->Cr_g_tab[i] = (- FIX(1.428272572)) * x;
+    /* Cb=>G value is scaled-up -0.688272572 * x */
+    /* We also add in ONE_HALF so that need not do it in inner loop */
+    upsample->Cb_g_tab[i] = (- FIX(0.688272572)) * x + ONE_HALF;
+  }
+}
+
+
+/*
+ * Initialize for an upsampling pass.
+ */
+
+METHODDEF(void)
+start_pass_merged_upsample (j_decompress_ptr cinfo)
+{
+  my_upsample_ptr upsample = (my_upsample_ptr) cinfo->upsample;
+
+  /* Mark the spare buffer empty */
+  upsample->spare_full = FALSE;
+  /* Initialize total-height counter for detecting bottom of image */
+  upsample->rows_to_go = cinfo->output_height;
+}
+
+
+/*
+ * Control routine to do upsampling (and color conversion).
+ *
+ * The control routine just handles the row buffering considerations.
+ */
+
+METHODDEF(void)
+merged_2v_upsample (j_decompress_ptr cinfo,
+		    JSAMPIMAGE input_buf, JDIMENSION *in_row_group_ctr,
+		    JDIMENSION in_row_groups_avail,
+		    JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
+		    JDIMENSION out_rows_avail)
+/* 2:1 vertical sampling case: may need a spare row. */
+{
+  my_upsample_ptr upsample = (my_upsample_ptr) cinfo->upsample;
+  JSAMPROW work_ptrs[2];
+  JDIMENSION num_rows;		/* number of rows returned to caller */
+
+  if (upsample->spare_full) {
+    /* If we have a spare row saved from a previous cycle, just return it. */
+    jcopy_sample_rows(& upsample->spare_row, output_buf + *out_row_ctr,
+		      1, upsample->out_row_width);
+    num_rows = 1;
+    upsample->spare_full = FALSE;
+  } else {
+    /* Figure number of rows to return to caller. */
+    num_rows = 2;
+    /* Not more than the distance to the end of the image. */
+    if (num_rows > upsample->rows_to_go)
+      num_rows = upsample->rows_to_go;
+    /* And not more than what the client can accept: */
+    out_rows_avail -= *out_row_ctr;
+    if (num_rows > out_rows_avail)
+      num_rows = out_rows_avail;
+    /* Create output pointer array for upsampler. */
+    work_ptrs[0] = output_buf[*out_row_ctr];
+    if (num_rows > 1) {
+      work_ptrs[1] = output_buf[*out_row_ctr + 1];
+    } else {
+      work_ptrs[1] = upsample->spare_row;
+      upsample->spare_full = TRUE;
+    }
+    /* Now do the upsampling. */
+    (*upsample->upmethod) (cinfo, input_buf, *in_row_group_ctr, work_ptrs);
+  }
+
+  /* Adjust counts */
+  *out_row_ctr += num_rows;
+  upsample->rows_to_go -= num_rows;
+  /* When the buffer is emptied, declare this input row group consumed */
+  if (! upsample->spare_full)
+    (*in_row_group_ctr)++;
+}
+
+
+METHODDEF(void)
+merged_1v_upsample (j_decompress_ptr cinfo,
+		    JSAMPIMAGE input_buf, JDIMENSION *in_row_group_ctr,
+		    JDIMENSION in_row_groups_avail,
+		    JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
+		    JDIMENSION out_rows_avail)
+/* 1:1 vertical sampling case: much easier, never need a spare row. */
+{
+  my_upsample_ptr upsample = (my_upsample_ptr) cinfo->upsample;
+
+  /* Just do the upsampling. */
+  (*upsample->upmethod) (cinfo, input_buf, *in_row_group_ctr,
+			 output_buf + *out_row_ctr);
+  /* Adjust counts */
+  (*out_row_ctr)++;
+  (*in_row_group_ctr)++;
+}
+
+
+/*
+ * These are the routines invoked by the control routines to do
+ * the actual upsampling/conversion.  One row group is processed per call.
+ *
+ * Note: since we may be writing directly into application-supplied buffers,
+ * we have to be honest about the output width; we can't assume the buffer
+ * has been rounded up to an even width.
+ */
+
+
+/*
+ * Upsample and color convert for the case of 2:1 horizontal and 1:1 vertical.
+ */
+
+METHODDEF(void)
+h2v1_merged_upsample (j_decompress_ptr cinfo,
+		      JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr,
+		      JSAMPARRAY output_buf)
+{
+  my_upsample_ptr upsample = (my_upsample_ptr) cinfo->upsample;
+  register int y, cred, cgreen, cblue;
+  int cb, cr;
+  register JSAMPROW outptr;
+  JSAMPROW inptr0, inptr1, inptr2;
+  JDIMENSION col;
+  /* copy these pointers into registers if possible */
+  register JSAMPLE * range_limit = cinfo->sample_range_limit;
+  int * Crrtab = upsample->Cr_r_tab;
+  int * Cbbtab = upsample->Cb_b_tab;
+  INT32 * Crgtab = upsample->Cr_g_tab;
+  INT32 * Cbgtab = upsample->Cb_g_tab;
+  SHIFT_TEMPS
+
+  inptr0 = input_buf[0][in_row_group_ctr];
+  inptr1 = input_buf[1][in_row_group_ctr];
+  inptr2 = input_buf[2][in_row_group_ctr];
+  outptr = output_buf[0];
+  /* Loop for each pair of output pixels */
+  for (col = cinfo->output_width >> 1; col > 0; col--) {
+    /* Do the chroma part of the calculation */
+    cb = GETJSAMPLE(*inptr1++);
+    cr = GETJSAMPLE(*inptr2++);
+    cgreen = (int) RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS);
+    cblue  = Cbbtab[cb];
+    cred   = Crrtab[cr];
+    /* Fetch 2 Y values and emit 2 pixels */
+    y  = GETJSAMPLE(*inptr0++);
+    outptr[RGB_RED]   = range_limit[y + cred];
+    outptr[RGB_GREEN] = range_limit[y + cgreen];
+    outptr[RGB_BLUE]  = range_limit[y + cblue];
+    outptr += RGB_PIXELSIZE;
+    y  = GETJSAMPLE(*inptr0++);
+    outptr[RGB_RED]   = range_limit[y + cred];
+    outptr[RGB_GREEN] = range_limit[y + cgreen];
+    outptr[RGB_BLUE]  = range_limit[y + cblue];
+    outptr += RGB_PIXELSIZE;
+  }
+  /* If image width is odd, do the last output column separately */
+  if (cinfo->output_width & 1) {
+    y  = GETJSAMPLE(*inptr0);
+    cb = GETJSAMPLE(*inptr1);
+    cr = GETJSAMPLE(*inptr2);
+    outptr[RGB_RED]   = range_limit[y + Crrtab[cr]];
+    outptr[RGB_GREEN] = range_limit[y +
+			      ((int) RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr],
+						 SCALEBITS))];
+    outptr[RGB_BLUE]  = range_limit[y + Cbbtab[cb]];
+  }
+}
+
+
+/*
+ * Upsample and color convert for the case of 2:1 horizontal and 2:1 vertical.
+ */
+
+METHODDEF(void)
+h2v2_merged_upsample (j_decompress_ptr cinfo,
+		      JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr,
+		      JSAMPARRAY output_buf)
+{
+  my_upsample_ptr upsample = (my_upsample_ptr) cinfo->upsample;
+  register int y, cred, cgreen, cblue;
+  int cb, cr;
+  register JSAMPROW outptr0, outptr1;
+  JSAMPROW inptr00, inptr01, inptr1, inptr2;
+  JDIMENSION col;
+  /* copy these pointers into registers if possible */
+  register JSAMPLE * range_limit = cinfo->sample_range_limit;
+  int * Crrtab = upsample->Cr_r_tab;
+  int * Cbbtab = upsample->Cb_b_tab;
+  INT32 * Crgtab = upsample->Cr_g_tab;
+  INT32 * Cbgtab = upsample->Cb_g_tab;
+  SHIFT_TEMPS
+
+  inptr00 = input_buf[0][in_row_group_ctr*2];
+  inptr01 = input_buf[0][in_row_group_ctr*2 + 1];
+  inptr1 = input_buf[1][in_row_group_ctr];
+  inptr2 = input_buf[2][in_row_group_ctr];
+  outptr0 = output_buf[0];
+  outptr1 = output_buf[1];
+  /* Loop for each group of output pixels */
+  for (col = cinfo->output_width >> 1; col > 0; col--) {
+    /* Do the chroma part of the calculation */
+    cb = GETJSAMPLE(*inptr1++);
+    cr = GETJSAMPLE(*inptr2++);
+    cgreen = (int) RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS);
+    cblue  = Cbbtab[cb];
+    cred   = Crrtab[cr];
+    /* Fetch 4 Y values and emit 4 pixels */
+    y  = GETJSAMPLE(*inptr00++);
+    outptr0[RGB_RED]   = range_limit[y + cred];
+    outptr0[RGB_GREEN] = range_limit[y + cgreen];
+    outptr0[RGB_BLUE]  = range_limit[y + cblue];
+    outptr0 += RGB_PIXELSIZE;
+    y  = GETJSAMPLE(*inptr00++);
+    outptr0[RGB_RED]   = range_limit[y + cred];
+    outptr0[RGB_GREEN] = range_limit[y + cgreen];
+    outptr0[RGB_BLUE]  = range_limit[y + cblue];
+    outptr0 += RGB_PIXELSIZE;
+    y  = GETJSAMPLE(*inptr01++);
+    outptr1[RGB_RED]   = range_limit[y + cred];
+    outptr1[RGB_GREEN] = range_limit[y + cgreen];
+    outptr1[RGB_BLUE]  = range_limit[y + cblue];
+    outptr1 += RGB_PIXELSIZE;
+    y  = GETJSAMPLE(*inptr01++);
+    outptr1[RGB_RED]   = range_limit[y + cred];
+    outptr1[RGB_GREEN] = range_limit[y + cgreen];
+    outptr1[RGB_BLUE]  = range_limit[y + cblue];
+    outptr1 += RGB_PIXELSIZE;
+  }
+  /* If image width is odd, do the last output column separately */
+  if (cinfo->output_width & 1) {
+    cb = GETJSAMPLE(*inptr1);
+    cr = GETJSAMPLE(*inptr2);
+    cgreen = (int) RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS);
+    cblue  = Cbbtab[cb];
+    cred   = Crrtab[cr];
+    y  = GETJSAMPLE(*inptr00);
+    outptr0[RGB_RED]   = range_limit[y + cred];
+    outptr0[RGB_GREEN] = range_limit[y + cgreen];
+    outptr0[RGB_BLUE]  = range_limit[y + cblue];
+    y  = GETJSAMPLE(*inptr01);
+    outptr1[RGB_RED]   = range_limit[y + cred];
+    outptr1[RGB_GREEN] = range_limit[y + cgreen];
+    outptr1[RGB_BLUE]  = range_limit[y + cblue];
+  }
+}
+
+
+/*
+ * Module initialization routine for merged upsampling/color conversion.
+ *
+ * NB: this is called under the conditions determined by use_merged_upsample()
+ * in jdmaster.c.  That routine MUST correspond to the actual capabilities
+ * of this module; no safety checks are made here.
+ */
+
+GLOBAL(void)
+jinit_merged_upsampler (j_decompress_ptr cinfo)
+{
+  my_upsample_ptr upsample;
+
+  upsample = (my_upsample_ptr) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, SIZEOF(my_upsampler));
+  cinfo->upsample = &upsample->pub;
+  upsample->pub.start_pass = start_pass_merged_upsample;
+  upsample->pub.need_context_rows = FALSE;
+
+  upsample->out_row_width = cinfo->output_width * cinfo->out_color_components;
+
+  if (cinfo->max_v_samp_factor == 2) {
+    upsample->pub.upsample = merged_2v_upsample;
+    upsample->upmethod = h2v2_merged_upsample;
+    /* Allocate a spare row buffer */
+    upsample->spare_row = (JSAMPROW) (*cinfo->mem->alloc_large)
+      ((j_common_ptr) cinfo, JPOOL_IMAGE,
+       (size_t) upsample->out_row_width * SIZEOF(JSAMPLE));
+  } else {
+    upsample->pub.upsample = merged_1v_upsample;
+    upsample->upmethod = h2v1_merged_upsample;
+    /* No spare row needed */
+    upsample->spare_row = NULL;
+  }
+
+  if (cinfo->jpeg_color_space == JCS_BG_YCC)
+    build_bg_ycc_rgb_table(cinfo);
+  else
+    build_ycc_rgb_table(cinfo);
+}
+
+#endif /* UPSAMPLE_MERGING_SUPPORTED */
@@ -0,0 +1,290 @@
+/*
+ * jdpostct.c
+ *
+ * Copyright (C) 1994-1996, Thomas G. Lane.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains the decompression postprocessing controller.
+ * This controller manages the upsampling, color conversion, and color
+ * quantization/reduction steps; specifically, it controls the buffering
+ * between upsample/color conversion and color quantization/reduction.
+ *
+ * If no color quantization/reduction is required, then this module has no
+ * work to do, and it just hands off to the upsample/color conversion code.
+ * An integrated upsample/convert/quantize process would replace this module
+ * entirely.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/* Private buffer controller object */
+
+typedef struct {
+  struct jpeg_d_post_controller pub; /* public fields */
+
+  /* Color quantization source buffer: this holds output data from
+   * the upsample/color conversion step to be passed to the quantizer.
+   * For two-pass color quantization, we need a full-image buffer;
+   * for one-pass operation, a strip buffer is sufficient.
+   */
+  jvirt_sarray_ptr whole_image;	/* virtual array, or NULL if one-pass */
+  JSAMPARRAY buffer;		/* strip buffer, or current strip of virtual */
+  JDIMENSION strip_height;	/* buffer size in rows */
+  /* for two-pass mode only: */
+  JDIMENSION starting_row;	/* row # of first row in current strip */
+  JDIMENSION next_row;		/* index of next row to fill/empty in strip */
+} my_post_controller;
+
+typedef my_post_controller * my_post_ptr;
+
+
+/* Forward declarations */
+METHODDEF(void) post_process_1pass
+	JPP((j_decompress_ptr cinfo,
+	     JSAMPIMAGE input_buf, JDIMENSION *in_row_group_ctr,
+	     JDIMENSION in_row_groups_avail,
+	     JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
+	     JDIMENSION out_rows_avail));
+#ifdef QUANT_2PASS_SUPPORTED
+METHODDEF(void) post_process_prepass
+	JPP((j_decompress_ptr cinfo,
+	     JSAMPIMAGE input_buf, JDIMENSION *in_row_group_ctr,
+	     JDIMENSION in_row_groups_avail,
+	     JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
+	     JDIMENSION out_rows_avail));
+METHODDEF(void) post_process_2pass
+	JPP((j_decompress_ptr cinfo,
+	     JSAMPIMAGE input_buf, JDIMENSION *in_row_group_ctr,
+	     JDIMENSION in_row_groups_avail,
+	     JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
+	     JDIMENSION out_rows_avail));
+#endif
+
+
+/*
+ * Initialize for a processing pass.
+ */
+
+METHODDEF(void)
+start_pass_dpost (j_decompress_ptr cinfo, J_BUF_MODE pass_mode)
+{
+  my_post_ptr post = (my_post_ptr) cinfo->post;
+
+  switch (pass_mode) {
+  case JBUF_PASS_THRU:
+    if (cinfo->quantize_colors) {
+      /* Single-pass processing with color quantization. */
+      post->pub.post_process_data = post_process_1pass;
+      /* We could be doing buffered-image output before starting a 2-pass
+       * color quantization; in that case, jinit_d_post_controller did not
+       * allocate a strip buffer.  Use the virtual-array buffer as workspace.
+       */
+      if (post->buffer == NULL) {
+	post->buffer = (*cinfo->mem->access_virt_sarray)
+	  ((j_common_ptr) cinfo, post->whole_image,
+	   (JDIMENSION) 0, post->strip_height, TRUE);
+      }
+    } else {
+      /* For single-pass processing without color quantization,
+       * I have no work to do; just call the upsampler directly.
+       */
+      post->pub.post_process_data = cinfo->upsample->upsample;
+    }
+    break;
+#ifdef QUANT_2PASS_SUPPORTED
+  case JBUF_SAVE_AND_PASS:
+    /* First pass of 2-pass quantization */
+    if (post->whole_image == NULL)
+      ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+    post->pub.post_process_data = post_process_prepass;
+    break;
+  case JBUF_CRANK_DEST:
+    /* Second pass of 2-pass quantization */
+    if (post->whole_image == NULL)
+      ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+    post->pub.post_process_data = post_process_2pass;
+    break;
+#endif /* QUANT_2PASS_SUPPORTED */
+  default:
+    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+    break;
+  }
+  post->starting_row = post->next_row = 0;
+}
+
+
+/*
+ * Process some data in the one-pass (strip buffer) case.
+ * This is used for color precision reduction as well as one-pass quantization.
+ */
+
+METHODDEF(void)
+post_process_1pass (j_decompress_ptr cinfo,
+		    JSAMPIMAGE input_buf, JDIMENSION *in_row_group_ctr,
+		    JDIMENSION in_row_groups_avail,
+		    JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
+		    JDIMENSION out_rows_avail)
+{
+  my_post_ptr post = (my_post_ptr) cinfo->post;
+  JDIMENSION num_rows, max_rows;
+
+  /* Fill the buffer, but not more than what we can dump out in one go. */
+  /* Note we rely on the upsampler to detect bottom of image. */
+  max_rows = out_rows_avail - *out_row_ctr;
+  if (max_rows > post->strip_height)
+    max_rows = post->strip_height;
+  num_rows = 0;
+  (*cinfo->upsample->upsample) (cinfo,
+		input_buf, in_row_group_ctr, in_row_groups_avail,
+		post->buffer, &num_rows, max_rows);
+  /* Quantize and emit data. */
+  (*cinfo->cquantize->color_quantize) (cinfo,
+		post->buffer, output_buf + *out_row_ctr, (int) num_rows);
+  *out_row_ctr += num_rows;
+}
+
+
+#ifdef QUANT_2PASS_SUPPORTED
+
+/*
+ * Process some data in the first pass of 2-pass quantization.
+ */
+
+METHODDEF(void)
+post_process_prepass (j_decompress_ptr cinfo,
+		      JSAMPIMAGE input_buf, JDIMENSION *in_row_group_ctr,
+		      JDIMENSION in_row_groups_avail,
+		      JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
+		      JDIMENSION out_rows_avail)
+{
+  my_post_ptr post = (my_post_ptr) cinfo->post;
+  JDIMENSION old_next_row, num_rows;
+
+  /* Reposition virtual buffer if at start of strip. */
+  if (post->next_row == 0) {
+    post->buffer = (*cinfo->mem->access_virt_sarray)
+	((j_common_ptr) cinfo, post->whole_image,
+	 post->starting_row, post->strip_height, TRUE);
+  }
+
+  /* Upsample some data (up to a strip height's worth). */
+  old_next_row = post->next_row;
+  (*cinfo->upsample->upsample) (cinfo,
+		input_buf, in_row_group_ctr, in_row_groups_avail,
+		post->buffer, &post->next_row, post->strip_height);
+
+  /* Allow quantizer to scan new data.  No data is emitted, */
+  /* but we advance out_row_ctr so outer loop can tell when we're done. */
+  if (post->next_row > old_next_row) {
+    num_rows = post->next_row - old_next_row;
+    (*cinfo->cquantize->color_quantize) (cinfo, post->buffer + old_next_row,
+					 (JSAMPARRAY) NULL, (int) num_rows);
+    *out_row_ctr += num_rows;
+  }
+
+  /* Advance if we filled the strip. */
+  if (post->next_row >= post->strip_height) {
+    post->starting_row += post->strip_height;
+    post->next_row = 0;
+  }
+}
+
+
+/*
+ * Process some data in the second pass of 2-pass quantization.
+ */
+
+METHODDEF(void)
+post_process_2pass (j_decompress_ptr cinfo,
+		    JSAMPIMAGE input_buf, JDIMENSION *in_row_group_ctr,
+		    JDIMENSION in_row_groups_avail,
+		    JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
+		    JDIMENSION out_rows_avail)
+{
+  my_post_ptr post = (my_post_ptr) cinfo->post;
+  JDIMENSION num_rows, max_rows;
+
+  /* Reposition virtual buffer if at start of strip. */
+  if (post->next_row == 0) {
+    post->buffer = (*cinfo->mem->access_virt_sarray)
+	((j_common_ptr) cinfo, post->whole_image,
+	 post->starting_row, post->strip_height, FALSE);
+  }
+
+  /* Determine number of rows to emit. */
+  num_rows = post->strip_height - post->next_row; /* available in strip */
+  max_rows = out_rows_avail - *out_row_ctr; /* available in output area */
+  if (num_rows > max_rows)
+    num_rows = max_rows;
+  /* We have to check bottom of image here, can't depend on upsampler. */
+  max_rows = cinfo->output_height - post->starting_row;
+  if (num_rows > max_rows)
+    num_rows = max_rows;
+
+  /* Quantize and emit data. */
+  (*cinfo->cquantize->color_quantize) (cinfo,
+		post->buffer + post->next_row, output_buf + *out_row_ctr,
+		(int) num_rows);
+  *out_row_ctr += num_rows;
+
+  /* Advance if we filled the strip. */
+  post->next_row += num_rows;
+  if (post->next_row >= post->strip_height) {
+    post->starting_row += post->strip_height;
+    post->next_row = 0;
+  }
+}
+
+#endif /* QUANT_2PASS_SUPPORTED */
+
+
+/*
+ * Initialize postprocessing controller.
+ */
+
+GLOBAL(void)
+jinit_d_post_controller (j_decompress_ptr cinfo, boolean need_full_buffer)
+{
+  my_post_ptr post;
+
+  post = (my_post_ptr)
+    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				SIZEOF(my_post_controller));
+  cinfo->post = (struct jpeg_d_post_controller *) post;
+  post->pub.start_pass = start_pass_dpost;
+  post->whole_image = NULL;	/* flag for no virtual arrays */
+  post->buffer = NULL;		/* flag for no strip buffer */
+
+  /* Create the quantization buffer, if needed */
+  if (cinfo->quantize_colors) {
+    /* The buffer strip height is max_v_samp_factor, which is typically
+     * an efficient number of rows for upsampling to return.
+     * (In the presence of output rescaling, we might want to be smarter?)
+     */
+    post->strip_height = (JDIMENSION) cinfo->max_v_samp_factor;
+    if (need_full_buffer) {
+      /* Two-pass color quantization: need full-image storage. */
+      /* We round up the number of rows to a multiple of the strip height. */
+#ifdef QUANT_2PASS_SUPPORTED
+      post->whole_image = (*cinfo->mem->request_virt_sarray)
+	((j_common_ptr) cinfo, JPOOL_IMAGE, FALSE,
+	 cinfo->output_width * cinfo->out_color_components,
+	 (JDIMENSION) jround_up((long) cinfo->output_height,
+				(long) post->strip_height),
+	 post->strip_height);
+#else
+      ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+#endif /* QUANT_2PASS_SUPPORTED */
+    } else {
+      /* One-pass color quantization: just make a strip buffer. */
+      post->buffer = (*cinfo->mem->alloc_sarray)
+	((j_common_ptr) cinfo, JPOOL_IMAGE,
+	 cinfo->output_width * cinfo->out_color_components,
+	 post->strip_height);
+    }
+  }
+}
@@ -0,0 +1,341 @@
+/*
+ * jdsample.c
+ *
+ * Copyright (C) 1991-1996, Thomas G. Lane.
+ * Modified 2002-2020 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains upsampling routines.
+ *
+ * Upsampling input data is counted in "row groups".  A row group
+ * is defined to be (v_samp_factor * DCT_v_scaled_size / min_DCT_v_scaled_size)
+ * sample rows of each component.  Upsampling will normally produce
+ * max_v_samp_factor pixel rows from each row group (but this could vary
+ * if the upsampler is applying a scale factor of its own).
+ *
+ * An excellent reference for image resampling is
+ *   Digital Image Warping, George Wolberg, 1990.
+ *   Pub. by IEEE Computer Society Press, Los Alamitos, CA. ISBN 0-8186-8944-7.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/* Pointer to routine to upsample a single component */
+typedef JMETHOD(void, upsample1_ptr,
+		(j_decompress_ptr cinfo, jpeg_component_info * compptr,
+		 JSAMPARRAY input_data, JSAMPIMAGE output_data_ptr));
+
+/* Private subobject */
+
+typedef struct {
+  struct jpeg_upsampler pub;	/* public fields */
+
+  /* Color conversion buffer.  When using separate upsampling and color
+   * conversion steps, this buffer holds one upsampled row group until it
+   * has been color converted and output.
+   * Note: we do not allocate any storage for component(s) which are full-size,
+   * ie do not need rescaling.  The corresponding entry of color_buf[] is
+   * simply set to point to the input data array, thereby avoiding copying.
+   */
+  JSAMPARRAY color_buf[MAX_COMPONENTS];
+
+  /* Per-component upsampling method pointers */
+  upsample1_ptr methods[MAX_COMPONENTS];
+
+  int next_row_out;		/* counts rows emitted from color_buf */
+  JDIMENSION rows_to_go;	/* counts rows remaining in image */
+
+  /* Height of an input row group for each component. */
+  int rowgroup_height[MAX_COMPONENTS];
+
+  /* These arrays save pixel expansion factors so that int_expand need not
+   * recompute them each time.  They are unused for other upsampling methods.
+   */
+  UINT8 h_expand[MAX_COMPONENTS];
+  UINT8 v_expand[MAX_COMPONENTS];
+} my_upsampler;
+
+typedef my_upsampler * my_upsample_ptr;
+
+
+/*
+ * Initialize for an upsampling pass.
+ */
+
+METHODDEF(void)
+start_pass_upsample (j_decompress_ptr cinfo)
+{
+  my_upsample_ptr upsample = (my_upsample_ptr) cinfo->upsample;
+
+  /* Mark the conversion buffer empty */
+  upsample->next_row_out = cinfo->max_v_samp_factor;
+  /* Initialize total-height counter for detecting bottom of image */
+  upsample->rows_to_go = cinfo->output_height;
+}
+
+
+/*
+ * Control routine to do upsampling (and color conversion).
+ *
+ * In this version we upsample each component independently.
+ * We upsample one row group into the conversion buffer, then apply
+ * color conversion a row at a time.
+ */
+
+METHODDEF(void)
+sep_upsample (j_decompress_ptr cinfo,
+	      JSAMPIMAGE input_buf, JDIMENSION *in_row_group_ctr,
+	      JDIMENSION in_row_groups_avail,
+	      JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
+	      JDIMENSION out_rows_avail)
+{
+  my_upsample_ptr upsample = (my_upsample_ptr) cinfo->upsample;
+  int ci;
+  jpeg_component_info * compptr;
+  JDIMENSION num_rows;
+
+  /* Fill the conversion buffer, if it's empty */
+  if (upsample->next_row_out >= cinfo->max_v_samp_factor) {
+    for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+	 ci++, compptr++) {
+      /* Don't bother to upsample an uninteresting component. */
+      if (! compptr->component_needed)
+	continue;
+      /* Invoke per-component upsample method.  Notice we pass a POINTER
+       * to color_buf[ci], so that fullsize_upsample can change it.
+       */
+      (*upsample->methods[ci]) (cinfo, compptr,
+	input_buf[ci] + (*in_row_group_ctr * upsample->rowgroup_height[ci]),
+	upsample->color_buf + ci);
+    }
+    upsample->next_row_out = 0;
+  }
+
+  /* Color-convert and emit rows */
+
+  /* How many we have in the buffer: */
+  num_rows = (JDIMENSION) (cinfo->max_v_samp_factor - upsample->next_row_out);
+  /* Not more than the distance to the end of the image.  Need this test
+   * in case the image height is not a multiple of max_v_samp_factor:
+   */
+  if (num_rows > upsample->rows_to_go) 
+    num_rows = upsample->rows_to_go;
+  /* And not more than what the client can accept: */
+  out_rows_avail -= *out_row_ctr;
+  if (num_rows > out_rows_avail)
+    num_rows = out_rows_avail;
+
+  (*cinfo->cconvert->color_convert) (cinfo, upsample->color_buf,
+				     (JDIMENSION) upsample->next_row_out,
+				     output_buf + *out_row_ctr,
+				     (int) num_rows);
+
+  /* Adjust counts */
+  *out_row_ctr += num_rows;
+  upsample->rows_to_go -= num_rows;
+  upsample->next_row_out += num_rows;
+  /* When the buffer is emptied, declare this input row group consumed */
+  if (upsample->next_row_out >= cinfo->max_v_samp_factor)
+    (*in_row_group_ctr)++;
+}
+
+
+/*
+ * These are the routines invoked by sep_upsample to upsample pixel values
+ * of a single component.  One row group is processed per call.
+ */
+
+
+/*
+ * For full-size components, we just make color_buf[ci] point at the
+ * input buffer, and thus avoid copying any data.  Note that this is
+ * safe only because sep_upsample doesn't declare the input row group
+ * "consumed" until we are done color converting and emitting it.
+ */
+
+METHODDEF(void)
+fullsize_upsample (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+		   JSAMPARRAY input_data, JSAMPIMAGE output_data_ptr)
+{
+  *output_data_ptr = input_data;
+}
+
+
+/*
+ * This version handles any integral sampling ratios.
+ * This is not used for typical JPEG files, so it need not be fast.
+ * Nor, for that matter, is it particularly accurate: the algorithm is
+ * simple replication of the input pixel onto the corresponding output
+ * pixels.  The hi-falutin sampling literature refers to this as a
+ * "box filter".  A box filter tends to introduce visible artifacts,
+ * so if you are actually going to use 3:1 or 4:1 sampling ratios
+ * you would be well advised to improve this code.
+ */
+
+METHODDEF(void)
+int_upsample (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	      JSAMPARRAY input_data, JSAMPIMAGE output_data_ptr)
+{
+  my_upsample_ptr upsample = (my_upsample_ptr) cinfo->upsample;
+  JSAMPARRAY output_data, output_end;
+  register JSAMPROW inptr, outptr;
+  register JSAMPLE invalue;
+  register int h;
+  JSAMPROW outend;
+  int h_expand, v_expand;
+
+  h_expand = upsample->h_expand[compptr->component_index];
+  v_expand = upsample->v_expand[compptr->component_index];
+
+  output_data = *output_data_ptr;
+  output_end = output_data + cinfo->max_v_samp_factor;
+  for (; output_data < output_end; output_data += v_expand) {
+    /* Generate one output row with proper horizontal expansion */
+    inptr = *input_data++;
+    outptr = *output_data;
+    outend = outptr + cinfo->output_width;
+    while (outptr < outend) {
+      invalue = *inptr++;	/* don't need GETJSAMPLE() here */
+      for (h = h_expand; h > 0; h--) {
+	*outptr++ = invalue;
+      }
+    }
+    /* Generate any additional output rows by duplicating the first one */
+    if (v_expand > 1) {
+      jcopy_sample_rows(output_data, output_data + 1,
+			v_expand - 1, cinfo->output_width);
+    }
+  }
+}
+
+
+/*
+ * Fast processing for the common case of 2:1 horizontal and 1:1 vertical.
+ * It's still a box filter.
+ */
+
+METHODDEF(void)
+h2v1_upsample (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	       JSAMPARRAY input_data, JSAMPIMAGE output_data_ptr)
+{
+  JSAMPARRAY output_data = *output_data_ptr;
+  register JSAMPROW inptr, outptr;
+  register JSAMPLE invalue;
+  JSAMPROW outend;
+  int outrow;
+
+  for (outrow = 0; outrow < cinfo->max_v_samp_factor; outrow++) {
+    inptr = input_data[outrow];
+    outptr = output_data[outrow];
+    outend = outptr + cinfo->output_width;
+    while (outptr < outend) {
+      invalue = *inptr++;	/* don't need GETJSAMPLE() here */
+      *outptr++ = invalue;
+      *outptr++ = invalue;
+    }
+  }
+}
+
+
+/*
+ * Fast processing for the common case of 2:1 horizontal and 2:1 vertical.
+ * It's still a box filter.
+ */
+
+METHODDEF(void)
+h2v2_upsample (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	       JSAMPARRAY input_data, JSAMPIMAGE output_data_ptr)
+{
+  JSAMPARRAY output_data, output_end;
+  register JSAMPROW inptr, outptr;
+  register JSAMPLE invalue;
+  JSAMPROW outend;
+
+  output_data = *output_data_ptr;
+  output_end = output_data + cinfo->max_v_samp_factor;
+  for (; output_data < output_end; output_data += 2) {
+    inptr = *input_data++;
+    outptr = *output_data;
+    outend = outptr + cinfo->output_width;
+    while (outptr < outend) {
+      invalue = *inptr++;	/* don't need GETJSAMPLE() here */
+      *outptr++ = invalue;
+      *outptr++ = invalue;
+    }
+    jcopy_sample_rows(output_data, output_data + 1,
+		      1, cinfo->output_width);
+  }
+}
+
+
+/*
+ * Module initialization routine for upsampling.
+ */
+
+GLOBAL(void)
+jinit_upsampler (j_decompress_ptr cinfo)
+{
+  my_upsample_ptr upsample;
+  int ci;
+  jpeg_component_info * compptr;
+  int h_in_group, v_in_group, h_out_group, v_out_group;
+
+  upsample = (my_upsample_ptr) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, SIZEOF(my_upsampler));
+  cinfo->upsample = &upsample->pub;
+  upsample->pub.start_pass = start_pass_upsample;
+  upsample->pub.upsample = sep_upsample;
+  upsample->pub.need_context_rows = FALSE; /* until we find out differently */
+
+  if (cinfo->CCIR601_sampling)	/* this isn't supported */
+    ERREXIT(cinfo, JERR_CCIR601_NOTIMPL);
+
+  /* Verify we can handle the sampling factors, select per-component methods,
+   * and create storage as needed.
+   */
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    /* Don't bother to upsample an uninteresting component. */
+    if (! compptr->component_needed)
+      continue;
+    /* Compute size of an "input group" after IDCT scaling.  This many samples
+     * are to be converted to max_h_samp_factor * max_v_samp_factor pixels.
+     */
+    h_in_group = (compptr->h_samp_factor * compptr->DCT_h_scaled_size) /
+		 cinfo->min_DCT_h_scaled_size;
+    v_in_group = (compptr->v_samp_factor * compptr->DCT_v_scaled_size) /
+		 cinfo->min_DCT_v_scaled_size;
+    h_out_group = cinfo->max_h_samp_factor;
+    v_out_group = cinfo->max_v_samp_factor;
+    upsample->rowgroup_height[ci] = v_in_group; /* save for use later */
+    if (h_in_group == h_out_group && v_in_group == v_out_group) {
+      /* Fullsize components can be processed without any work. */
+      upsample->methods[ci] = fullsize_upsample;
+      continue;		/* don't need to allocate buffer */
+    }
+    if (h_in_group * 2 == h_out_group && v_in_group == v_out_group) {
+      /* Special case for 2h1v upsampling */
+      upsample->methods[ci] = h2v1_upsample;
+    } else if (h_in_group * 2 == h_out_group &&
+	       v_in_group * 2 == v_out_group) {
+      /* Special case for 2h2v upsampling */
+      upsample->methods[ci] = h2v2_upsample;
+    } else if ((h_out_group % h_in_group) == 0 &&
+	       (v_out_group % v_in_group) == 0) {
+      /* Generic integral-factors upsampling method */
+      upsample->methods[ci] = int_upsample;
+      upsample->h_expand[ci] = (UINT8) (h_out_group / h_in_group);
+      upsample->v_expand[ci] = (UINT8) (v_out_group / v_in_group);
+    } else
+      ERREXIT(cinfo, JERR_FRACT_SAMPLE_NOTIMPL);
+    upsample->color_buf[ci] = (*cinfo->mem->alloc_sarray)
+      ((j_common_ptr) cinfo, JPOOL_IMAGE,
+       (JDIMENSION) jround_up((long) cinfo->output_width,
+			      (long) cinfo->max_h_samp_factor),
+       (JDIMENSION) cinfo->max_v_samp_factor);
+  }
+}
@@ -0,0 +1,140 @@
+/*
+ * jdtrans.c
+ *
+ * Copyright (C) 1995-1997, Thomas G. Lane.
+ * Modified 2000-2009 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains library routines for transcoding decompression,
+ * that is, reading raw DCT coefficient arrays from an input JPEG file.
+ * The routines in jdapimin.c will also be needed by a transcoder.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/* Forward declarations */
+LOCAL(void) transdecode_master_selection JPP((j_decompress_ptr cinfo));
+
+
+/*
+ * Read the coefficient arrays from a JPEG file.
+ * jpeg_read_header must be completed before calling this.
+ *
+ * The entire image is read into a set of virtual coefficient-block arrays,
+ * one per component.  The return value is a pointer to the array of
+ * virtual-array descriptors.  These can be manipulated directly via the
+ * JPEG memory manager, or handed off to jpeg_write_coefficients().
+ * To release the memory occupied by the virtual arrays, call
+ * jpeg_finish_decompress() when done with the data.
+ *
+ * An alternative usage is to simply obtain access to the coefficient arrays
+ * during a buffered-image-mode decompression operation.  This is allowed
+ * after any jpeg_finish_output() call.  The arrays can be accessed until
+ * jpeg_finish_decompress() is called.  (Note that any call to the library
+ * may reposition the arrays, so don't rely on access_virt_barray() results
+ * to stay valid across library calls.)
+ *
+ * Returns NULL if suspended.  This case need be checked only if
+ * a suspending data source is used.
+ */
+
+GLOBAL(jvirt_barray_ptr *)
+jpeg_read_coefficients (j_decompress_ptr cinfo)
+{
+  if (cinfo->global_state == DSTATE_READY) {
+    /* First call: initialize active modules */
+    transdecode_master_selection(cinfo);
+    cinfo->global_state = DSTATE_RDCOEFS;
+  }
+  if (cinfo->global_state == DSTATE_RDCOEFS) {
+    /* Absorb whole file into the coef buffer */
+    for (;;) {
+      int retcode;
+      /* Call progress monitor hook if present */
+      if (cinfo->progress != NULL)
+	(*cinfo->progress->progress_monitor) ((j_common_ptr) cinfo);
+      /* Absorb some more input */
+      retcode = (*cinfo->inputctl->consume_input) (cinfo);
+      if (retcode == JPEG_SUSPENDED)
+	return NULL;
+      if (retcode == JPEG_REACHED_EOI)
+	break;
+      /* Advance progress counter if appropriate */
+      if (cinfo->progress != NULL &&
+	  (retcode == JPEG_ROW_COMPLETED || retcode == JPEG_REACHED_SOS)) {
+	if (++cinfo->progress->pass_counter >= cinfo->progress->pass_limit) {
+	  /* startup underestimated number of scans; ratchet up one scan */
+	  cinfo->progress->pass_limit += (long) cinfo->total_iMCU_rows;
+	}
+      }
+    }
+    /* Set state so that jpeg_finish_decompress does the right thing */
+    cinfo->global_state = DSTATE_STOPPING;
+  }
+  /* At this point we should be in state DSTATE_STOPPING if being used
+   * standalone, or in state DSTATE_BUFIMAGE if being invoked to get access
+   * to the coefficients during a full buffered-image-mode decompression.
+   */
+  if ((cinfo->global_state == DSTATE_STOPPING ||
+       cinfo->global_state == DSTATE_BUFIMAGE) && cinfo->buffered_image) {
+    return cinfo->coef->coef_arrays;
+  }
+  /* Oops, improper usage */
+  ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+  return NULL;			/* keep compiler happy */
+}
+
+
+/*
+ * Master selection of decompression modules for transcoding.
+ * This substitutes for jdmaster.c's initialization of the full decompressor.
+ */
+
+LOCAL(void)
+transdecode_master_selection (j_decompress_ptr cinfo)
+{
+  /* This is effectively a buffered-image operation. */
+  cinfo->buffered_image = TRUE;
+
+  /* Compute output image dimensions and related values. */
+  jpeg_core_output_dimensions(cinfo);
+
+  /* Entropy decoding: either Huffman or arithmetic coding. */
+  if (cinfo->arith_code)
+    jinit_arith_decoder(cinfo);
+  else {
+    jinit_huff_decoder(cinfo);
+  }
+
+  /* Always get a full-image coefficient buffer. */
+  jinit_d_coef_controller(cinfo, TRUE);
+
+  /* We can now tell the memory manager to allocate virtual arrays. */
+  (*cinfo->mem->realize_virt_arrays) ((j_common_ptr) cinfo);
+
+  /* Initialize input side of decompressor to consume first scan. */
+  (*cinfo->inputctl->start_input_pass) (cinfo);
+
+  /* Initialize progress monitoring. */
+  if (cinfo->progress != NULL) {
+    int nscans;
+    /* Estimate number of scans to set pass_limit. */
+    if (cinfo->progressive_mode) {
+      /* Arbitrarily estimate 2 interleaved DC scans + 3 AC scans/component. */
+      nscans = 2 + 3 * cinfo->num_components;
+    } else if (cinfo->inputctl->has_multiple_scans) {
+      /* For a nonprogressive multiscan file, estimate 1 scan per component. */
+      nscans = cinfo->num_components;
+    } else {
+      nscans = 1;
+    }
+    cinfo->progress->pass_counter = 0L;
+    cinfo->progress->pass_limit = (long) cinfo->total_iMCU_rows * nscans;
+    cinfo->progress->completed_passes = 0;
+    cinfo->progress->total_passes = 1;
+  }
+}
@@ -0,0 +1,253 @@
+/*
+ * jerror.c
+ *
+ * Copyright (C) 1991-1998, Thomas G. Lane.
+ * Modified 2012-2015 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains simple error-reporting and trace-message routines.
+ * These are suitable for Unix-like systems and others where writing to
+ * stderr is the right thing to do.  Many applications will want to replace
+ * some or all of these routines.
+ *
+ * If you define USE_WINDOWS_MESSAGEBOX in jconfig.h or in the makefile,
+ * you get a Windows-specific hack to display error messages in a dialog box.
+ * It ain't much, but it beats dropping error messages into the bit bucket,
+ * which is what happens to output to stderr under most Windows C compilers.
+ *
+ * These routines are used by both the compression and decompression code.
+ */
+
+#ifdef USE_WINDOWS_MESSAGEBOX
+#include <windows.h>
+#endif
+
+/* this is not a core library module, so it doesn't define JPEG_INTERNALS */
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jversion.h"
+#include "jerror.h"
+
+#ifndef EXIT_FAILURE		/* define exit() codes if not provided */
+#define EXIT_FAILURE  1
+#endif
+
+
+/*
+ * Create the message string table.
+ * We do this from the master message list in jerror.h by re-reading
+ * jerror.h with a suitable definition for macro JMESSAGE.
+ * The message table is made an external symbol just in case any applications
+ * want to refer to it directly.
+ */
+
+#ifdef NEED_SHORT_EXTERNAL_NAMES
+#define jpeg_std_message_table	jMsgTable
+#endif
+
+#define JMESSAGE(code,string)	string ,
+
+const char * const jpeg_std_message_table[] = {
+#include "jerror.h"
+  NULL
+};
+
+
+/*
+ * Error exit handler: must not return to caller.
+ *
+ * Applications may override this if they want to get control back after
+ * an error.  Typically one would longjmp somewhere instead of exiting.
+ * The setjmp buffer can be made a private field within an expanded error
+ * handler object.  Note that the info needed to generate an error message
+ * is stored in the error object, so you can generate the message now or
+ * later, at your convenience.
+ * You should make sure that the JPEG object is cleaned up (with jpeg_abort
+ * or jpeg_destroy) at some point.
+ */
+
+METHODDEF(noreturn_t)
+error_exit (j_common_ptr cinfo)
+{
+  /* Always display the message */
+  (*cinfo->err->output_message) (cinfo);
+
+  /* Let the memory manager delete any temp files before we die */
+  jpeg_destroy(cinfo);
+
+  exit(EXIT_FAILURE);
+}
+
+
+/*
+ * Actual output of an error or trace message.
+ * Applications may override this method to send JPEG messages somewhere
+ * other than stderr.
+ *
+ * On Windows, printing to stderr is generally completely useless,
+ * so we provide optional code to produce an error-dialog popup.
+ * Most Windows applications will still prefer to override this routine,
+ * but if they don't, it'll do something at least marginally useful.
+ *
+ * NOTE: to use the library in an environment that doesn't support the
+ * C stdio library, you may have to delete the call to fprintf() entirely,
+ * not just not use this routine.
+ */
+
+METHODDEF(void)
+output_message (j_common_ptr cinfo)
+{
+  char buffer[JMSG_LENGTH_MAX];
+
+  /* Create the message */
+  (*cinfo->err->format_message) (cinfo, buffer);
+
+#ifdef USE_WINDOWS_MESSAGEBOX
+  /* Display it in a message dialog box */
+  MessageBox(GetActiveWindow(), buffer, "JPEG Library Error",
+	     MB_OK | MB_ICONERROR);
+#else
+  /* Send it to stderr, adding a newline */
+  fprintf(stderr, "%s\n", buffer);
+#endif
+}
+
+
+/*
+ * Decide whether to emit a trace or warning message.
+ * msg_level is one of:
+ *   -1: recoverable corrupt-data warning, may want to abort.
+ *    0: important advisory messages (always display to user).
+ *    1: first level of tracing detail.
+ *    2,3,...: successively more detailed tracing messages.
+ * An application might override this method if it wanted to abort on warnings
+ * or change the policy about which messages to display.
+ */
+
+METHODDEF(void)
+emit_message (j_common_ptr cinfo, int msg_level)
+{
+  struct jpeg_error_mgr * err = cinfo->err;
+
+  if (msg_level < 0) {
+    /* It's a warning message.  Since corrupt files may generate many warnings,
+     * the policy implemented here is to show only the first warning,
+     * unless trace_level >= 3.
+     */
+    if (err->num_warnings == 0 || err->trace_level >= 3)
+      (*err->output_message) (cinfo);
+    /* Always count warnings in num_warnings. */
+    err->num_warnings++;
+  } else {
+    /* It's a trace message.  Show it if trace_level >= msg_level. */
+    if (err->trace_level >= msg_level)
+      (*err->output_message) (cinfo);
+  }
+}
+
+
+/*
+ * Format a message string for the most recent JPEG error or message.
+ * The message is stored into buffer, which should be at least JMSG_LENGTH_MAX
+ * characters.  Note that no '\n' character is added to the string.
+ * Few applications should need to override this method.
+ */
+
+METHODDEF(void)
+format_message (j_common_ptr cinfo, char * buffer)
+{
+  struct jpeg_error_mgr * err = cinfo->err;
+  int msg_code = err->msg_code;
+  const char * msgtext = NULL;
+  const char * msgptr;
+  char ch;
+  boolean isstring;
+
+  /* Look up message string in proper table */
+  if (msg_code > 0 && msg_code <= err->last_jpeg_message) {
+    msgtext = err->jpeg_message_table[msg_code];
+  } else if (err->addon_message_table != NULL &&
+	     msg_code >= err->first_addon_message &&
+	     msg_code <= err->last_addon_message) {
+    msgtext = err->addon_message_table[msg_code - err->first_addon_message];
+  }
+
+  /* Defend against bogus message number */
+  if (msgtext == NULL) {
+    err->msg_parm.i[0] = msg_code;
+    msgtext = err->jpeg_message_table[0];
+  }
+
+  /* Check for string parameter, as indicated by %s in the message text */
+  isstring = FALSE;
+  msgptr = msgtext;
+  while ((ch = *msgptr++) != '\0') {
+    if (ch == '%') {
+      if (*msgptr == 's') isstring = TRUE;
+      break;
+    }
+  }
+
+  /* Format the message into the passed buffer */
+  if (isstring)
+    sprintf(buffer, msgtext, err->msg_parm.s);
+  else
+    sprintf(buffer, msgtext,
+	    err->msg_parm.i[0], err->msg_parm.i[1],
+	    err->msg_parm.i[2], err->msg_parm.i[3],
+	    err->msg_parm.i[4], err->msg_parm.i[5],
+	    err->msg_parm.i[6], err->msg_parm.i[7]);
+}
+
+
+/*
+ * Reset error state variables at start of a new image.
+ * This is called during compression startup to reset trace/error
+ * processing to default state, without losing any application-specific
+ * method pointers.  An application might possibly want to override
+ * this method if it has additional error processing state.
+ */
+
+METHODDEF(void)
+reset_error_mgr (j_common_ptr cinfo)
+{
+  cinfo->err->num_warnings = 0;
+  /* trace_level is not reset since it is an application-supplied parameter */
+  cinfo->err->msg_code = 0;	/* may be useful as a flag for "no error" */
+}
+
+
+/*
+ * Fill in the standard error-handling methods in a jpeg_error_mgr object.
+ * Typical call is:
+ *	struct jpeg_compress_struct cinfo;
+ *	struct jpeg_error_mgr err;
+ *
+ *	cinfo.err = jpeg_std_error(&err);
+ * after which the application may override some of the methods.
+ */
+
+GLOBAL(struct jpeg_error_mgr *)
+jpeg_std_error (struct jpeg_error_mgr * err)
+{
+  err->error_exit = error_exit;
+  err->emit_message = emit_message;
+  err->output_message = output_message;
+  err->format_message = format_message;
+  err->reset_error_mgr = reset_error_mgr;
+
+  err->trace_level = 0;		/* default = no tracing */
+  err->num_warnings = 0;	/* no warnings emitted yet */
+  err->msg_code = 0;		/* may be useful as a flag for "no error" */
+
+  /* Initialize message table pointers */
+  err->jpeg_message_table = jpeg_std_message_table;
+  err->last_jpeg_message = (int) JMSG_LASTMSGCODE - 1;
+
+  err->addon_message_table = NULL;
+  err->first_addon_message = 0;	/* for safety */
+  err->last_addon_message = 0;
+
+  return err;
+}
@@ -0,0 +1,304 @@
+/*
+ * jerror.h
+ *
+ * Copyright (C) 1994-1997, Thomas G. Lane.
+ * Modified 1997-2018 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file defines the error and message codes for the JPEG library.
+ * Edit this file to add new codes, or to translate the message strings to
+ * some other language.
+ * A set of error-reporting macros are defined too.  Some applications using
+ * the JPEG library may wish to include this file to get the error codes
+ * and/or the macros.
+ */
+
+/*
+ * To define the enum list of message codes, include this file without
+ * defining macro JMESSAGE.  To create a message string table, include it
+ * again with a suitable JMESSAGE definition (see jerror.c for an example).
+ */
+#ifndef JMESSAGE
+#ifndef JERROR_H
+/* First time through, define the enum list */
+#define JMAKE_ENUM_LIST
+#else
+/* Repeated inclusions of this file are no-ops unless JMESSAGE is defined */
+#define JMESSAGE(code,string)
+#endif /* JERROR_H */
+#endif /* JMESSAGE */
+
+#ifdef JMAKE_ENUM_LIST
+
+typedef enum {
+
+#define JMESSAGE(code,string)	code ,
+
+#endif /* JMAKE_ENUM_LIST */
+
+JMESSAGE(JMSG_NOMESSAGE, "Bogus message code %d") /* Must be first entry! */
+
+/* For maintenance convenience, list is alphabetical by message code name */
+JMESSAGE(JERR_BAD_ALIGN_TYPE, "ALIGN_TYPE is wrong, please fix")
+JMESSAGE(JERR_BAD_ALLOC_CHUNK, "MAX_ALLOC_CHUNK is wrong, please fix")
+JMESSAGE(JERR_BAD_BUFFER_MODE, "Bogus buffer control mode")
+JMESSAGE(JERR_BAD_COMPONENT_ID, "Invalid component ID %d in SOS")
+JMESSAGE(JERR_BAD_CROP_SPEC, "Invalid crop request")
+JMESSAGE(JERR_BAD_DCT_COEF, "DCT coefficient out of range")
+JMESSAGE(JERR_BAD_DCTSIZE, "DCT scaled block size %dx%d not supported")
+JMESSAGE(JERR_BAD_DROP_SAMPLING,
+	 "Component index %d: mismatching sampling ratio %d:%d, %d:%d, %c")
+JMESSAGE(JERR_BAD_HUFF_TABLE, "Bogus Huffman table definition")
+JMESSAGE(JERR_BAD_IN_COLORSPACE, "Bogus input colorspace")
+JMESSAGE(JERR_BAD_J_COLORSPACE, "Bogus JPEG colorspace")
+JMESSAGE(JERR_BAD_LENGTH, "Bogus marker length")
+JMESSAGE(JERR_BAD_LIB_VERSION,
+	 "Wrong JPEG library version: library is %d, caller expects %d")
+JMESSAGE(JERR_BAD_MCU_SIZE, "Sampling factors too large for interleaved scan")
+JMESSAGE(JERR_BAD_POOL_ID, "Invalid memory pool code %d")
+JMESSAGE(JERR_BAD_PRECISION, "Unsupported JPEG data precision %d")
+JMESSAGE(JERR_BAD_PROGRESSION,
+	 "Invalid progressive parameters Ss=%d Se=%d Ah=%d Al=%d")
+JMESSAGE(JERR_BAD_PROG_SCRIPT,
+	 "Invalid progressive parameters at scan script entry %d")
+JMESSAGE(JERR_BAD_SAMPLING, "Bogus sampling factors")
+JMESSAGE(JERR_BAD_SCAN_SCRIPT, "Invalid scan script at entry %d")
+JMESSAGE(JERR_BAD_STATE, "Improper call to JPEG library in state %d")
+JMESSAGE(JERR_BAD_STRUCT_SIZE,
+	 "JPEG parameter struct mismatch: library thinks size is %u, caller expects %u")
+JMESSAGE(JERR_BAD_VIRTUAL_ACCESS, "Bogus virtual array access")
+JMESSAGE(JERR_BUFFER_SIZE, "Buffer passed to JPEG library is too small")
+JMESSAGE(JERR_CANT_SUSPEND, "Suspension not allowed here")
+JMESSAGE(JERR_CCIR601_NOTIMPL, "CCIR601 sampling not implemented yet")
+JMESSAGE(JERR_COMPONENT_COUNT, "Too many color components: %d, max %d")
+JMESSAGE(JERR_CONVERSION_NOTIMPL, "Unsupported color conversion request")
+JMESSAGE(JERR_DAC_INDEX, "Bogus DAC index %d")
+JMESSAGE(JERR_DAC_VALUE, "Bogus DAC value 0x%x")
+JMESSAGE(JERR_DHT_INDEX, "Bogus DHT index %d")
+JMESSAGE(JERR_DQT_INDEX, "Bogus DQT index %d")
+JMESSAGE(JERR_EMPTY_IMAGE, "Empty JPEG image (DNL not supported)")
+JMESSAGE(JERR_EMS_READ, "Read from EMS failed")
+JMESSAGE(JERR_EMS_WRITE, "Write to EMS failed")
+JMESSAGE(JERR_EOI_EXPECTED, "Didn't expect more than one scan")
+JMESSAGE(JERR_FILE_READ, "Input file read error")
+JMESSAGE(JERR_FILE_WRITE, "Output file write error --- out of disk space?")
+JMESSAGE(JERR_FRACT_SAMPLE_NOTIMPL, "Fractional sampling not implemented yet")
+JMESSAGE(JERR_HUFF_CLEN_OUTOFBOUNDS, "Huffman code size table out of bounds")
+JMESSAGE(JERR_HUFF_MISSING_CODE, "Missing Huffman code table entry")
+JMESSAGE(JERR_IMAGE_TOO_BIG, "Maximum supported image dimension is %u pixels")
+JMESSAGE(JERR_INPUT_EMPTY, "Empty input file")
+JMESSAGE(JERR_INPUT_EOF, "Premature end of input file")
+JMESSAGE(JERR_MISMATCHED_QUANT_TABLE,
+	 "Cannot transcode due to multiple use of quantization table %d")
+JMESSAGE(JERR_MISSING_DATA, "Scan script does not transmit all data")
+JMESSAGE(JERR_MODE_CHANGE, "Invalid color quantization mode change")
+JMESSAGE(JERR_NOTIMPL, "Not implemented yet")
+JMESSAGE(JERR_NOT_COMPILED, "Requested feature was omitted at compile time")
+JMESSAGE(JERR_NO_ARITH_TABLE, "Arithmetic table 0x%02x was not defined")
+JMESSAGE(JERR_NO_BACKING_STORE, "Backing store not supported")
+JMESSAGE(JERR_NO_HUFF_TABLE, "Huffman table 0x%02x was not defined")
+JMESSAGE(JERR_NO_IMAGE, "JPEG datastream contains no image")
+JMESSAGE(JERR_NO_QUANT_TABLE, "Quantization table 0x%02x was not defined")
+JMESSAGE(JERR_NO_SOI, "Not a JPEG file: starts with 0x%02x 0x%02x")
+JMESSAGE(JERR_OUT_OF_MEMORY, "Insufficient memory (case %d)")
+JMESSAGE(JERR_QUANT_COMPONENTS,
+	 "Cannot quantize more than %d color components")
+JMESSAGE(JERR_QUANT_FEW_COLORS, "Cannot quantize to fewer than %d colors")
+JMESSAGE(JERR_QUANT_MANY_COLORS, "Cannot quantize to more than %d colors")
+JMESSAGE(JERR_SOF_BEFORE, "Invalid JPEG file structure: %s before SOF")
+JMESSAGE(JERR_SOF_DUPLICATE, "Invalid JPEG file structure: two SOF markers")
+JMESSAGE(JERR_SOF_NO_SOS, "Invalid JPEG file structure: missing SOS marker")
+JMESSAGE(JERR_SOF_UNSUPPORTED, "Unsupported JPEG process: SOF type 0x%02x")
+JMESSAGE(JERR_SOI_DUPLICATE, "Invalid JPEG file structure: two SOI markers")
+JMESSAGE(JERR_TFILE_CREATE, "Failed to create temporary file %s")
+JMESSAGE(JERR_TFILE_READ, "Read failed on temporary file")
+JMESSAGE(JERR_TFILE_SEEK, "Seek failed on temporary file")
+JMESSAGE(JERR_TFILE_WRITE,
+	 "Write failed on temporary file --- out of disk space?")
+JMESSAGE(JERR_TOO_LITTLE_DATA, "Application transferred too few scanlines")
+JMESSAGE(JERR_UNKNOWN_MARKER, "Unsupported marker type 0x%02x")
+JMESSAGE(JERR_VIRTUAL_BUG, "Virtual array controller messed up")
+JMESSAGE(JERR_WIDTH_OVERFLOW, "Image too wide for this implementation")
+JMESSAGE(JERR_XMS_READ, "Read from XMS failed")
+JMESSAGE(JERR_XMS_WRITE, "Write to XMS failed")
+JMESSAGE(JMSG_COPYRIGHT, JCOPYRIGHT)
+JMESSAGE(JMSG_VERSION, JVERSION)
+JMESSAGE(JTRC_16BIT_TABLES,
+	 "Caution: quantization tables are too coarse for baseline JPEG")
+JMESSAGE(JTRC_ADOBE,
+	 "Adobe APP14 marker: version %d, flags 0x%04x 0x%04x, transform %d")
+JMESSAGE(JTRC_APP0, "Unknown APP0 marker (not JFIF), length %u")
+JMESSAGE(JTRC_APP14, "Unknown APP14 marker (not Adobe), length %u")
+JMESSAGE(JTRC_DAC, "Define Arithmetic Table 0x%02x: 0x%02x")
+JMESSAGE(JTRC_DHT, "Define Huffman Table 0x%02x")
+JMESSAGE(JTRC_DQT, "Define Quantization Table %d  precision %d")
+JMESSAGE(JTRC_DRI, "Define Restart Interval %u")
+JMESSAGE(JTRC_EMS_CLOSE, "Freed EMS handle %u")
+JMESSAGE(JTRC_EMS_OPEN, "Obtained EMS handle %u")
+JMESSAGE(JTRC_EOI, "End Of Image")
+JMESSAGE(JTRC_HUFFBITS, "        %3d %3d %3d %3d %3d %3d %3d %3d")
+JMESSAGE(JTRC_JFIF, "JFIF APP0 marker: version %d.%02d, density %dx%d  %d")
+JMESSAGE(JTRC_JFIF_BADTHUMBNAILSIZE,
+	 "Warning: thumbnail image size does not match data length %u")
+JMESSAGE(JTRC_JFIF_EXTENSION,
+	 "JFIF extension marker: type 0x%02x, length %u")
+JMESSAGE(JTRC_JFIF_THUMBNAIL, "    with %d x %d thumbnail image")
+JMESSAGE(JTRC_MISC_MARKER, "Miscellaneous marker 0x%02x, length %u")
+JMESSAGE(JTRC_PARMLESS_MARKER, "Unexpected marker 0x%02x")
+JMESSAGE(JTRC_QUANTVALS, "        %4u %4u %4u %4u %4u %4u %4u %4u")
+JMESSAGE(JTRC_QUANT_3_NCOLORS, "Quantizing to %d = %d*%d*%d colors")
+JMESSAGE(JTRC_QUANT_NCOLORS, "Quantizing to %d colors")
+JMESSAGE(JTRC_QUANT_SELECTED, "Selected %d colors for quantization")
+JMESSAGE(JTRC_RECOVERY_ACTION, "At marker 0x%02x, recovery action %d")
+JMESSAGE(JTRC_RST, "RST%d")
+JMESSAGE(JTRC_SMOOTH_NOTIMPL,
+	 "Smoothing not supported with nonstandard sampling ratios")
+JMESSAGE(JTRC_SOF, "Start Of Frame 0x%02x: width=%u, height=%u, components=%d")
+JMESSAGE(JTRC_SOF_COMPONENT, "    Component %d: %dhx%dv q=%d")
+JMESSAGE(JTRC_SOI, "Start of Image")
+JMESSAGE(JTRC_SOS, "Start Of Scan: %d components")
+JMESSAGE(JTRC_SOS_COMPONENT, "    Component %d: dc=%d ac=%d")
+JMESSAGE(JTRC_SOS_PARAMS, "  Ss=%d, Se=%d, Ah=%d, Al=%d")
+JMESSAGE(JTRC_TFILE_CLOSE, "Closed temporary file %s")
+JMESSAGE(JTRC_TFILE_OPEN, "Opened temporary file %s")
+JMESSAGE(JTRC_THUMB_JPEG,
+	 "JFIF extension marker: JPEG-compressed thumbnail image, length %u")
+JMESSAGE(JTRC_THUMB_PALETTE,
+	 "JFIF extension marker: palette thumbnail image, length %u")
+JMESSAGE(JTRC_THUMB_RGB,
+	 "JFIF extension marker: RGB thumbnail image, length %u")
+JMESSAGE(JTRC_UNKNOWN_IDS,
+	 "Unrecognized component IDs %d %d %d, assuming YCbCr")
+JMESSAGE(JTRC_XMS_CLOSE, "Freed XMS handle %u")
+JMESSAGE(JTRC_XMS_OPEN, "Obtained XMS handle %u")
+JMESSAGE(JWRN_ADOBE_XFORM, "Unknown Adobe color transform code %d")
+JMESSAGE(JWRN_ARITH_BAD_CODE, "Corrupt JPEG data: bad arithmetic code")
+JMESSAGE(JWRN_BOGUS_PROGRESSION,
+	 "Inconsistent progression sequence for component %d coefficient %d")
+JMESSAGE(JWRN_EXTRANEOUS_DATA,
+	 "Corrupt JPEG data: %u extraneous bytes before marker 0x%02x")
+JMESSAGE(JWRN_HIT_MARKER, "Corrupt JPEG data: premature end of data segment")
+JMESSAGE(JWRN_HUFF_BAD_CODE, "Corrupt JPEG data: bad Huffman code")
+JMESSAGE(JWRN_JFIF_MAJOR, "Warning: unknown JFIF revision number %d.%02d")
+JMESSAGE(JWRN_JPEG_EOF, "Premature end of JPEG file")
+JMESSAGE(JWRN_MUST_RESYNC,
+	 "Corrupt JPEG data: found marker 0x%02x instead of RST%d")
+JMESSAGE(JWRN_NOT_SEQUENTIAL, "Invalid SOS parameters for sequential JPEG")
+JMESSAGE(JWRN_TOO_MUCH_DATA, "Application transferred too many scanlines")
+
+#ifdef JMAKE_ENUM_LIST
+
+  JMSG_LASTMSGCODE
+} J_MESSAGE_CODE;
+
+#undef JMAKE_ENUM_LIST
+#endif /* JMAKE_ENUM_LIST */
+
+/* Zap JMESSAGE macro so that future re-inclusions do nothing by default */
+#undef JMESSAGE
+
+
+#ifndef JERROR_H
+#define JERROR_H
+
+/* Macros to simplify using the error and trace message stuff */
+/* The first parameter is either type of cinfo pointer */
+
+/* Fatal errors (print message and exit) */
+#define ERREXIT(cinfo,code)  \
+  ((cinfo)->err->msg_code = (code), \
+   (*(cinfo)->err->error_exit) ((j_common_ptr) (cinfo)))
+#define ERREXIT1(cinfo,code,p1)  \
+  ((cinfo)->err->msg_code = (code), \
+   (cinfo)->err->msg_parm.i[0] = (p1), \
+   (*(cinfo)->err->error_exit) ((j_common_ptr) (cinfo)))
+#define ERREXIT2(cinfo,code,p1,p2)  \
+  ((cinfo)->err->msg_code = (code), \
+   (cinfo)->err->msg_parm.i[0] = (p1), \
+   (cinfo)->err->msg_parm.i[1] = (p2), \
+   (*(cinfo)->err->error_exit) ((j_common_ptr) (cinfo)))
+#define ERREXIT3(cinfo,code,p1,p2,p3)  \
+  ((cinfo)->err->msg_code = (code), \
+   (cinfo)->err->msg_parm.i[0] = (p1), \
+   (cinfo)->err->msg_parm.i[1] = (p2), \
+   (cinfo)->err->msg_parm.i[2] = (p3), \
+   (*(cinfo)->err->error_exit) ((j_common_ptr) (cinfo)))
+#define ERREXIT4(cinfo,code,p1,p2,p3,p4)  \
+  ((cinfo)->err->msg_code = (code), \
+   (cinfo)->err->msg_parm.i[0] = (p1), \
+   (cinfo)->err->msg_parm.i[1] = (p2), \
+   (cinfo)->err->msg_parm.i[2] = (p3), \
+   (cinfo)->err->msg_parm.i[3] = (p4), \
+   (*(cinfo)->err->error_exit) ((j_common_ptr) (cinfo)))
+#define ERREXIT6(cinfo,code,p1,p2,p3,p4,p5,p6)  \
+  ((cinfo)->err->msg_code = (code), \
+   (cinfo)->err->msg_parm.i[0] = (p1), \
+   (cinfo)->err->msg_parm.i[1] = (p2), \
+   (cinfo)->err->msg_parm.i[2] = (p3), \
+   (cinfo)->err->msg_parm.i[3] = (p4), \
+   (cinfo)->err->msg_parm.i[4] = (p5), \
+   (cinfo)->err->msg_parm.i[5] = (p6), \
+   (*(cinfo)->err->error_exit) ((j_common_ptr) (cinfo)))
+#define ERREXITS(cinfo,code,str)  \
+  ((cinfo)->err->msg_code = (code), \
+   strncpy((cinfo)->err->msg_parm.s, (str), JMSG_STR_PARM_MAX), \
+   (*(cinfo)->err->error_exit) ((j_common_ptr) (cinfo)))
+
+#define MAKESTMT(stuff)		do { stuff } while (0)
+
+/* Nonfatal errors (we can keep going, but the data is probably corrupt) */
+#define WARNMS(cinfo,code)  \
+  ((cinfo)->err->msg_code = (code), \
+   (*(cinfo)->err->emit_message) ((j_common_ptr) (cinfo), -1))
+#define WARNMS1(cinfo,code,p1)  \
+  ((cinfo)->err->msg_code = (code), \
+   (cinfo)->err->msg_parm.i[0] = (p1), \
+   (*(cinfo)->err->emit_message) ((j_common_ptr) (cinfo), -1))
+#define WARNMS2(cinfo,code,p1,p2)  \
+  ((cinfo)->err->msg_code = (code), \
+   (cinfo)->err->msg_parm.i[0] = (p1), \
+   (cinfo)->err->msg_parm.i[1] = (p2), \
+   (*(cinfo)->err->emit_message) ((j_common_ptr) (cinfo), -1))
+
+/* Informational/debugging messages */
+#define TRACEMS(cinfo,lvl,code)  \
+  ((cinfo)->err->msg_code = (code), \
+   (*(cinfo)->err->emit_message) ((j_common_ptr) (cinfo), (lvl)))
+#define TRACEMS1(cinfo,lvl,code,p1)  \
+  ((cinfo)->err->msg_code = (code), \
+   (cinfo)->err->msg_parm.i[0] = (p1), \
+   (*(cinfo)->err->emit_message) ((j_common_ptr) (cinfo), (lvl)))
+#define TRACEMS2(cinfo,lvl,code,p1,p2)  \
+  ((cinfo)->err->msg_code = (code), \
+   (cinfo)->err->msg_parm.i[0] = (p1), \
+   (cinfo)->err->msg_parm.i[1] = (p2), \
+   (*(cinfo)->err->emit_message) ((j_common_ptr) (cinfo), (lvl)))
+#define TRACEMS3(cinfo,lvl,code,p1,p2,p3)  \
+  MAKESTMT(int * _mp = (cinfo)->err->msg_parm.i; \
+	   _mp[0] = (p1); _mp[1] = (p2); _mp[2] = (p3); \
+	   (cinfo)->err->msg_code = (code); \
+	   (*(cinfo)->err->emit_message) ((j_common_ptr) (cinfo), (lvl)); )
+#define TRACEMS4(cinfo,lvl,code,p1,p2,p3,p4)  \
+  MAKESTMT(int * _mp = (cinfo)->err->msg_parm.i; \
+	   _mp[0] = (p1); _mp[1] = (p2); _mp[2] = (p3); _mp[3] = (p4); \
+	   (cinfo)->err->msg_code = (code); \
+	   (*(cinfo)->err->emit_message) ((j_common_ptr) (cinfo), (lvl)); )
+#define TRACEMS5(cinfo,lvl,code,p1,p2,p3,p4,p5)  \
+  MAKESTMT(int * _mp = (cinfo)->err->msg_parm.i; \
+	   _mp[0] = (p1); _mp[1] = (p2); _mp[2] = (p3); _mp[3] = (p4); \
+	   _mp[4] = (p5); \
+	   (cinfo)->err->msg_code = (code); \
+	   (*(cinfo)->err->emit_message) ((j_common_ptr) (cinfo), (lvl)); )
+#define TRACEMS8(cinfo,lvl,code,p1,p2,p3,p4,p5,p6,p7,p8)  \
+  MAKESTMT(int * _mp = (cinfo)->err->msg_parm.i; \
+	   _mp[0] = (p1); _mp[1] = (p2); _mp[2] = (p3); _mp[3] = (p4); \
+	   _mp[4] = (p5); _mp[5] = (p6); _mp[6] = (p7); _mp[7] = (p8); \
+	   (cinfo)->err->msg_code = (code); \
+	   (*(cinfo)->err->emit_message) ((j_common_ptr) (cinfo), (lvl)); )
+#define TRACEMSS(cinfo,lvl,code,str)  \
+  ((cinfo)->err->msg_code = (code), \
+   strncpy((cinfo)->err->msg_parm.s, (str), JMSG_STR_PARM_MAX), \
+   (*(cinfo)->err->emit_message) ((j_common_ptr) (cinfo), (lvl)))
+
+#endif /* JERROR_H */
@@ -0,0 +1,176 @@
+/*
+ * jfdctflt.c
+ *
+ * Copyright (C) 1994-1996, Thomas G. Lane.
+ * Modified 2003-2017 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains a floating-point implementation of the
+ * forward DCT (Discrete Cosine Transform).
+ *
+ * This implementation should be more accurate than either of the integer
+ * DCT implementations.  However, it may not give the same results on all
+ * machines because of differences in roundoff behavior.  Speed will depend
+ * on the hardware's floating point capacity.
+ *
+ * A 2-D DCT can be done by 1-D DCT on each row followed by 1-D DCT
+ * on each column.  Direct algorithms are also available, but they are
+ * much more complex and seem not to be any faster when reduced to code.
+ *
+ * This implementation is based on Arai, Agui, and Nakajima's algorithm for
+ * scaled DCT.  Their original paper (Trans. IEICE E-71(11):1095) is in
+ * Japanese, but the algorithm is described in the Pennebaker & Mitchell
+ * JPEG textbook (see REFERENCES section in file README).  The following code
+ * is based directly on figure 4-8 in P&M.
+ * While an 8-point DCT cannot be done in less than 11 multiplies, it is
+ * possible to arrange the computation so that many of the multiplies are
+ * simple scalings of the final outputs.  These multiplies can then be
+ * folded into the multiplications or divisions by the JPEG quantization
+ * table entries.  The AA&N method leaves only 5 multiplies and 29 adds
+ * to be done in the DCT itself.
+ * The primary disadvantage of this method is that with a fixed-point
+ * implementation, accuracy is lost due to imprecise representation of the
+ * scaled quantization values.  However, that problem does not arise if
+ * we use floating point arithmetic.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jdct.h"		/* Private declarations for DCT subsystem */
+
+#ifdef DCT_FLOAT_SUPPORTED
+
+
+/*
+ * This module is specialized to the case DCTSIZE = 8.
+ */
+
+#if DCTSIZE != 8
+  Sorry, this code only copes with 8x8 DCT blocks. /* deliberate syntax err */
+#endif
+
+
+/*
+ * Perform the forward DCT on one block of samples.
+ *
+ * cK represents cos(K*pi/16).
+ */
+
+GLOBAL(void)
+jpeg_fdct_float (FAST_FLOAT * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  FAST_FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
+  FAST_FLOAT tmp10, tmp11, tmp12, tmp13;
+  FAST_FLOAT z1, z2, z3, z4, z5, z11, z13;
+  FAST_FLOAT *dataptr;
+  JSAMPROW elemptr;
+  int ctr;
+
+  /* Pass 1: process rows. */
+
+  dataptr = data;
+  for (ctr = 0; ctr < DCTSIZE; ctr++) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Load data into workspace */
+    tmp0 = (FAST_FLOAT) (GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[7]));
+    tmp7 = (FAST_FLOAT) (GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[7]));
+    tmp1 = (FAST_FLOAT) (GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[6]));
+    tmp6 = (FAST_FLOAT) (GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[6]));
+    tmp2 = (FAST_FLOAT) (GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[5]));
+    tmp5 = (FAST_FLOAT) (GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[5]));
+    tmp3 = (FAST_FLOAT) (GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[4]));
+    tmp4 = (FAST_FLOAT) (GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[4]));
+
+    /* Even part */
+
+    tmp10 = tmp0 + tmp3;	/* phase 2 */
+    tmp13 = tmp0 - tmp3;
+    tmp11 = tmp1 + tmp2;
+    tmp12 = tmp1 - tmp2;
+
+    /* Apply unsigned->signed conversion. */
+    dataptr[0] = tmp10 + tmp11 - 8 * CENTERJSAMPLE; /* phase 3 */
+    dataptr[4] = tmp10 - tmp11;
+
+    z1 = (tmp12 + tmp13) * ((FAST_FLOAT) 0.707106781); /* c4 */
+    dataptr[2] = tmp13 + z1;	/* phase 5 */
+    dataptr[6] = tmp13 - z1;
+
+    /* Odd part */
+
+    tmp10 = tmp4 + tmp5;	/* phase 2 */
+    tmp11 = tmp5 + tmp6;
+    tmp12 = tmp6 + tmp7;
+
+    /* The rotator is modified from fig 4-8 to avoid extra negations. */
+    z5 = (tmp10 - tmp12) * ((FAST_FLOAT) 0.382683433); /* c6 */
+    z2 = ((FAST_FLOAT) 0.541196100) * tmp10 + z5; /* c2-c6 */
+    z4 = ((FAST_FLOAT) 1.306562965) * tmp12 + z5; /* c2+c6 */
+    z3 = tmp11 * ((FAST_FLOAT) 0.707106781); /* c4 */
+
+    z11 = tmp7 + z3;		/* phase 5 */
+    z13 = tmp7 - z3;
+
+    dataptr[5] = z13 + z2;	/* phase 6 */
+    dataptr[3] = z13 - z2;
+    dataptr[1] = z11 + z4;
+    dataptr[7] = z11 - z4;
+
+    dataptr += DCTSIZE;		/* advance pointer to next row */
+  }
+
+  /* Pass 2: process columns. */
+
+  dataptr = data;
+  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
+    tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*7];
+    tmp7 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*7];
+    tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*6];
+    tmp6 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*6];
+    tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*5];
+    tmp5 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*5];
+    tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*4];
+    tmp4 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*4];
+
+    /* Even part */
+
+    tmp10 = tmp0 + tmp3;	/* phase 2 */
+    tmp13 = tmp0 - tmp3;
+    tmp11 = tmp1 + tmp2;
+    tmp12 = tmp1 - tmp2;
+
+    dataptr[DCTSIZE*0] = tmp10 + tmp11; /* phase 3 */
+    dataptr[DCTSIZE*4] = tmp10 - tmp11;
+
+    z1 = (tmp12 + tmp13) * ((FAST_FLOAT) 0.707106781); /* c4 */
+    dataptr[DCTSIZE*2] = tmp13 + z1; /* phase 5 */
+    dataptr[DCTSIZE*6] = tmp13 - z1;
+
+    /* Odd part */
+
+    tmp10 = tmp4 + tmp5;	/* phase 2 */
+    tmp11 = tmp5 + tmp6;
+    tmp12 = tmp6 + tmp7;
+
+    /* The rotator is modified from fig 4-8 to avoid extra negations. */
+    z5 = (tmp10 - tmp12) * ((FAST_FLOAT) 0.382683433); /* c6 */
+    z2 = ((FAST_FLOAT) 0.541196100) * tmp10 + z5; /* c2-c6 */
+    z4 = ((FAST_FLOAT) 1.306562965) * tmp12 + z5; /* c2+c6 */
+    z3 = tmp11 * ((FAST_FLOAT) 0.707106781); /* c4 */
+
+    z11 = tmp7 + z3;		/* phase 5 */
+    z13 = tmp7 - z3;
+
+    dataptr[DCTSIZE*5] = z13 + z2; /* phase 6 */
+    dataptr[DCTSIZE*3] = z13 - z2;
+    dataptr[DCTSIZE*1] = z11 + z4;
+    dataptr[DCTSIZE*7] = z11 - z4;
+
+    dataptr++;			/* advance pointer to next column */
+  }
+}
+
+#endif /* DCT_FLOAT_SUPPORTED */
@@ -0,0 +1,232 @@
+/*
+ * jfdctfst.c
+ *
+ * Copyright (C) 1994-1996, Thomas G. Lane.
+ * Modified 2003-2017 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains a fast, not so accurate integer implementation of the
+ * forward DCT (Discrete Cosine Transform).
+ *
+ * A 2-D DCT can be done by 1-D DCT on each row followed by 1-D DCT
+ * on each column.  Direct algorithms are also available, but they are
+ * much more complex and seem not to be any faster when reduced to code.
+ *
+ * This implementation is based on Arai, Agui, and Nakajima's algorithm for
+ * scaled DCT.  Their original paper (Trans. IEICE E-71(11):1095) is in
+ * Japanese, but the algorithm is described in the Pennebaker & Mitchell
+ * JPEG textbook (see REFERENCES section in file README).  The following code
+ * is based directly on figure 4-8 in P&M.
+ * While an 8-point DCT cannot be done in less than 11 multiplies, it is
+ * possible to arrange the computation so that many of the multiplies are
+ * simple scalings of the final outputs.  These multiplies can then be
+ * folded into the multiplications or divisions by the JPEG quantization
+ * table entries.  The AA&N method leaves only 5 multiplies and 29 adds
+ * to be done in the DCT itself.
+ * The primary disadvantage of this method is that with fixed-point math,
+ * accuracy is lost due to imprecise representation of the scaled
+ * quantization values.  The smaller the quantization table entry, the less
+ * precise the scaled value, so this implementation does worse with high-
+ * quality-setting files than with low-quality ones.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jdct.h"		/* Private declarations for DCT subsystem */
+
+#ifdef DCT_IFAST_SUPPORTED
+
+
+/*
+ * This module is specialized to the case DCTSIZE = 8.
+ */
+
+#if DCTSIZE != 8
+  Sorry, this code only copes with 8x8 DCT blocks. /* deliberate syntax err */
+#endif
+
+
+/* Scaling decisions are generally the same as in the LL&M algorithm;
+ * see jfdctint.c for more details.  However, we choose to descale
+ * (right shift) multiplication products as soon as they are formed,
+ * rather than carrying additional fractional bits into subsequent additions.
+ * This compromises accuracy slightly, but it lets us save a few shifts.
+ * More importantly, 16-bit arithmetic is then adequate (for 8-bit samples)
+ * everywhere except in the multiplications proper; this saves a good deal
+ * of work on 16-bit-int machines.
+ *
+ * Again to save a few shifts, the intermediate results between pass 1 and
+ * pass 2 are not upscaled, but are represented only to integral precision.
+ *
+ * A final compromise is to represent the multiplicative constants to only
+ * 8 fractional bits, rather than 13.  This saves some shifting work on some
+ * machines, and may also reduce the cost of multiplication (since there
+ * are fewer one-bits in the constants).
+ */
+
+#define CONST_BITS  8
+
+
+/* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
+ * causing a lot of useless floating-point operations at run time.
+ * To get around this we use the following pre-calculated constants.
+ * If you change CONST_BITS you may want to add appropriate values.
+ * (With a reasonable C compiler, you can just rely on the FIX() macro...)
+ */
+
+#if CONST_BITS == 8
+#define FIX_0_382683433  ((INT32)   98)		/* FIX(0.382683433) */
+#define FIX_0_541196100  ((INT32)  139)		/* FIX(0.541196100) */
+#define FIX_0_707106781  ((INT32)  181)		/* FIX(0.707106781) */
+#define FIX_1_306562965  ((INT32)  334)		/* FIX(1.306562965) */
+#else
+#define FIX_0_382683433  FIX(0.382683433)
+#define FIX_0_541196100  FIX(0.541196100)
+#define FIX_0_707106781  FIX(0.707106781)
+#define FIX_1_306562965  FIX(1.306562965)
+#endif
+
+
+/* We can gain a little more speed, with a further compromise in accuracy,
+ * by omitting the addition in a descaling shift.  This yields an incorrectly
+ * rounded result half the time...
+ */
+
+#ifndef USE_ACCURATE_ROUNDING
+#undef DESCALE
+#define DESCALE(x,n)  RIGHT_SHIFT(x, n)
+#endif
+
+
+/* Multiply a DCTELEM variable by an INT32 constant, and immediately
+ * descale to yield a DCTELEM result.
+ */
+
+#define MULTIPLY(var,const)  ((DCTELEM) DESCALE((var) * (const), CONST_BITS))
+
+
+/*
+ * Perform the forward DCT on one block of samples.
+ *
+ * cK represents cos(K*pi/16).
+ */
+
+GLOBAL(void)
+jpeg_fdct_ifast (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  DCTELEM tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
+  DCTELEM tmp10, tmp11, tmp12, tmp13;
+  DCTELEM z1, z2, z3, z4, z5, z11, z13;
+  DCTELEM *dataptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pass 1: process rows. */
+
+  dataptr = data;
+  for (ctr = 0; ctr < DCTSIZE; ctr++) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Load data into workspace */
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[7]);
+    tmp7 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[7]);
+    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[6]);
+    tmp6 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[6]);
+    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[5]);
+    tmp5 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[5]);
+    tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[4]);
+    tmp4 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[4]);
+
+    /* Even part */
+
+    tmp10 = tmp0 + tmp3;	/* phase 2 */
+    tmp13 = tmp0 - tmp3;
+    tmp11 = tmp1 + tmp2;
+    tmp12 = tmp1 - tmp2;
+
+    /* Apply unsigned->signed conversion. */
+    dataptr[0] = tmp10 + tmp11 - 8 * CENTERJSAMPLE; /* phase 3 */
+    dataptr[4] = tmp10 - tmp11;
+
+    z1 = MULTIPLY(tmp12 + tmp13, FIX_0_707106781); /* c4 */
+    dataptr[2] = tmp13 + z1;	/* phase 5 */
+    dataptr[6] = tmp13 - z1;
+
+    /* Odd part */
+
+    tmp10 = tmp4 + tmp5;	/* phase 2 */
+    tmp11 = tmp5 + tmp6;
+    tmp12 = tmp6 + tmp7;
+
+    /* The rotator is modified from fig 4-8 to avoid extra negations. */
+    z5 = MULTIPLY(tmp10 - tmp12, FIX_0_382683433); /* c6 */
+    z2 = MULTIPLY(tmp10, FIX_0_541196100) + z5; /* c2-c6 */
+    z4 = MULTIPLY(tmp12, FIX_1_306562965) + z5; /* c2+c6 */
+    z3 = MULTIPLY(tmp11, FIX_0_707106781); /* c4 */
+
+    z11 = tmp7 + z3;		/* phase 5 */
+    z13 = tmp7 - z3;
+
+    dataptr[5] = z13 + z2;	/* phase 6 */
+    dataptr[3] = z13 - z2;
+    dataptr[1] = z11 + z4;
+    dataptr[7] = z11 - z4;
+
+    dataptr += DCTSIZE;		/* advance pointer to next row */
+  }
+
+  /* Pass 2: process columns. */
+
+  dataptr = data;
+  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
+    tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*7];
+    tmp7 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*7];
+    tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*6];
+    tmp6 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*6];
+    tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*5];
+    tmp5 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*5];
+    tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*4];
+    tmp4 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*4];
+
+    /* Even part */
+
+    tmp10 = tmp0 + tmp3;	/* phase 2 */
+    tmp13 = tmp0 - tmp3;
+    tmp11 = tmp1 + tmp2;
+    tmp12 = tmp1 - tmp2;
+
+    dataptr[DCTSIZE*0] = tmp10 + tmp11; /* phase 3 */
+    dataptr[DCTSIZE*4] = tmp10 - tmp11;
+
+    z1 = MULTIPLY(tmp12 + tmp13, FIX_0_707106781); /* c4 */
+    dataptr[DCTSIZE*2] = tmp13 + z1; /* phase 5 */
+    dataptr[DCTSIZE*6] = tmp13 - z1;
+
+    /* Odd part */
+
+    tmp10 = tmp4 + tmp5;	/* phase 2 */
+    tmp11 = tmp5 + tmp6;
+    tmp12 = tmp6 + tmp7;
+
+    /* The rotator is modified from fig 4-8 to avoid extra negations. */
+    z5 = MULTIPLY(tmp10 - tmp12, FIX_0_382683433); /* c6 */
+    z2 = MULTIPLY(tmp10, FIX_0_541196100) + z5; /* c2-c6 */
+    z4 = MULTIPLY(tmp12, FIX_1_306562965) + z5; /* c2+c6 */
+    z3 = MULTIPLY(tmp11, FIX_0_707106781); /* c4 */
+
+    z11 = tmp7 + z3;		/* phase 5 */
+    z13 = tmp7 - z3;
+
+    dataptr[DCTSIZE*5] = z13 + z2; /* phase 6 */
+    dataptr[DCTSIZE*3] = z13 - z2;
+    dataptr[DCTSIZE*1] = z11 + z4;
+    dataptr[DCTSIZE*7] = z11 - z4;
+
+    dataptr++;			/* advance pointer to next column */
+  }
+}
+
+#endif /* DCT_IFAST_SUPPORTED */
@@ -0,0 +1,238 @@
+/*
+ * jidctflt.c
+ *
+ * Copyright (C) 1994-1998, Thomas G. Lane.
+ * Modified 2010-2017 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains a floating-point implementation of the
+ * inverse DCT (Discrete Cosine Transform).  In the IJG code, this routine
+ * must also perform dequantization of the input coefficients.
+ *
+ * This implementation should be more accurate than either of the integer
+ * IDCT implementations.  However, it may not give the same results on all
+ * machines because of differences in roundoff behavior.  Speed will depend
+ * on the hardware's floating point capacity.
+ *
+ * A 2-D IDCT can be done by 1-D IDCT on each column followed by 1-D IDCT
+ * on each row (or vice versa, but it's more convenient to emit a row at
+ * a time).  Direct algorithms are also available, but they are much more
+ * complex and seem not to be any faster when reduced to code.
+ *
+ * This implementation is based on Arai, Agui, and Nakajima's algorithm for
+ * scaled DCT.  Their original paper (Trans. IEICE E-71(11):1095) is in
+ * Japanese, but the algorithm is described in the Pennebaker & Mitchell
+ * JPEG textbook (see REFERENCES section in file README).  The following code
+ * is based directly on figure 4-8 in P&M.
+ * While an 8-point DCT cannot be done in less than 11 multiplies, it is
+ * possible to arrange the computation so that many of the multiplies are
+ * simple scalings of the final outputs.  These multiplies can then be
+ * folded into the multiplications or divisions by the JPEG quantization
+ * table entries.  The AA&N method leaves only 5 multiplies and 29 adds
+ * to be done in the DCT itself.
+ * The primary disadvantage of this method is that with a fixed-point
+ * implementation, accuracy is lost due to imprecise representation of the
+ * scaled quantization values.  However, that problem does not arise if
+ * we use floating point arithmetic.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jdct.h"		/* Private declarations for DCT subsystem */
+
+#ifdef DCT_FLOAT_SUPPORTED
+
+
+/*
+ * This module is specialized to the case DCTSIZE = 8.
+ */
+
+#if DCTSIZE != 8
+  Sorry, this code only copes with 8x8 DCT blocks. /* deliberate syntax err */
+#endif
+
+
+/* Dequantize a coefficient by multiplying it by the multiplier-table
+ * entry; produce a float result.
+ */
+
+#define DEQUANTIZE(coef,quantval)  (((FAST_FLOAT) (coef)) * (quantval))
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients.
+ *
+ * cK represents cos(K*pi/16).
+ */
+
+GLOBAL(void)
+jpeg_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+		 JCOEFPTR coef_block,
+		 JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  FAST_FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
+  FAST_FLOAT tmp10, tmp11, tmp12, tmp13;
+  FAST_FLOAT z5, z10, z11, z12, z13;
+  JCOEFPTR inptr;
+  FLOAT_MULT_TYPE * quantptr;
+  FAST_FLOAT * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  FAST_FLOAT workspace[DCTSIZE2]; /* buffers data between passes */
+
+  /* Pass 1: process columns from input, store into work array. */
+
+  inptr = coef_block;
+  quantptr = (FLOAT_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = DCTSIZE; ctr > 0; ctr--) {
+    /* Due to quantization, we will usually find that many of the input
+     * coefficients are zero, especially the AC terms.  We can exploit this
+     * by short-circuiting the IDCT calculation for any column in which all
+     * the AC terms are zero.  In that case each output is equal to the
+     * DC coefficient (with scale factor as needed).
+     * With typical images and quantization tables, half or more of the
+     * column DCT calculations can be simplified this way.
+     */
+
+    if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
+	inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*4] == 0 &&
+	inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*6] == 0 &&
+	inptr[DCTSIZE*7] == 0) {
+      /* AC terms all zero */
+      FAST_FLOAT dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+
+      wsptr[DCTSIZE*0] = dcval;
+      wsptr[DCTSIZE*1] = dcval;
+      wsptr[DCTSIZE*2] = dcval;
+      wsptr[DCTSIZE*3] = dcval;
+      wsptr[DCTSIZE*4] = dcval;
+      wsptr[DCTSIZE*5] = dcval;
+      wsptr[DCTSIZE*6] = dcval;
+      wsptr[DCTSIZE*7] = dcval;
+
+      inptr++;			/* advance pointers to next column */
+      quantptr++;
+      wsptr++;
+      continue;
+    }
+
+    /* Even part */
+
+    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    tmp1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    tmp2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
+    tmp3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
+
+    tmp10 = tmp0 + tmp2;	/* phase 3 */
+    tmp11 = tmp0 - tmp2;
+
+    tmp13 = tmp1 + tmp3;	/* phases 5-3 */
+    tmp12 = (tmp1 - tmp3) * ((FAST_FLOAT) 1.414213562) - tmp13; /* 2*c4 */
+
+    tmp0 = tmp10 + tmp13;	/* phase 2 */
+    tmp3 = tmp10 - tmp13;
+    tmp1 = tmp11 + tmp12;
+    tmp2 = tmp11 - tmp12;
+
+    /* Odd part */
+
+    tmp4 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    tmp5 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+    tmp6 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
+    tmp7 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
+
+    z13 = tmp6 + tmp5;		/* phase 6 */
+    z10 = tmp6 - tmp5;
+    z11 = tmp4 + tmp7;
+    z12 = tmp4 - tmp7;
+
+    tmp7 = z11 + z13;		/* phase 5 */
+    tmp11 = (z11 - z13) * ((FAST_FLOAT) 1.414213562); /* 2*c4 */
+
+    z5 = (z10 + z12) * ((FAST_FLOAT) 1.847759065); /* 2*c2 */
+    tmp10 = z5 - z12 * ((FAST_FLOAT) 1.082392200); /* 2*(c2-c6) */
+    tmp12 = z5 - z10 * ((FAST_FLOAT) 2.613125930); /* 2*(c2+c6) */
+
+    tmp6 = tmp12 - tmp7;	/* phase 2 */
+    tmp5 = tmp11 - tmp6;
+    tmp4 = tmp10 - tmp5;
+
+    wsptr[DCTSIZE*0] = tmp0 + tmp7;
+    wsptr[DCTSIZE*7] = tmp0 - tmp7;
+    wsptr[DCTSIZE*1] = tmp1 + tmp6;
+    wsptr[DCTSIZE*6] = tmp1 - tmp6;
+    wsptr[DCTSIZE*2] = tmp2 + tmp5;
+    wsptr[DCTSIZE*5] = tmp2 - tmp5;
+    wsptr[DCTSIZE*3] = tmp3 + tmp4;
+    wsptr[DCTSIZE*4] = tmp3 - tmp4;
+
+    inptr++;			/* advance pointers to next column */
+    quantptr++;
+    wsptr++;
+  }
+
+  /* Pass 2: process rows from work array, store into output array. */
+
+  wsptr = workspace;
+  for (ctr = 0; ctr < DCTSIZE; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+    /* Rows of zeroes can be exploited in the same way as we did with columns.
+     * However, the column calculation has created many nonzero AC terms, so
+     * the simplification applies less often (typically 5% to 10% of the time).
+     * And testing floats for zero is relatively expensive, so we don't bother.
+     */
+
+    /* Even part */
+
+    /* Prepare range-limit and float->int conversion */
+    z5 = wsptr[0] + (((FAST_FLOAT) RANGE_CENTER) + ((FAST_FLOAT) 0.5));
+    tmp10 = z5 + wsptr[4];
+    tmp11 = z5 - wsptr[4];
+
+    tmp13 = wsptr[2] + wsptr[6];
+    tmp12 = (wsptr[2] - wsptr[6]) *
+	      ((FAST_FLOAT) 1.414213562) - tmp13; /* 2*c4 */
+
+    tmp0 = tmp10 + tmp13;
+    tmp3 = tmp10 - tmp13;
+    tmp1 = tmp11 + tmp12;
+    tmp2 = tmp11 - tmp12;
+
+    /* Odd part */
+
+    z13 = wsptr[5] + wsptr[3];
+    z10 = wsptr[5] - wsptr[3];
+    z11 = wsptr[1] + wsptr[7];
+    z12 = wsptr[1] - wsptr[7];
+
+    tmp7 = z11 + z13;		/* phase 5 */
+    tmp11 = (z11 - z13) * ((FAST_FLOAT) 1.414213562); /* 2*c4 */
+
+    z5 = (z10 + z12) * ((FAST_FLOAT) 1.847759065); /* 2*c2 */
+    tmp10 = z5 - z12 * ((FAST_FLOAT) 1.082392200); /* 2*(c2-c6) */
+    tmp12 = z5 - z10 * ((FAST_FLOAT) 2.613125930); /* 2*(c2+c6) */
+
+    tmp6 = tmp12 - tmp7;	/* phase 2 */
+    tmp5 = tmp11 - tmp6;
+    tmp4 = tmp10 - tmp5;
+
+    /* Final output stage: float->int conversion and range-limit */
+
+    outptr[0] = range_limit[(int) (tmp0 + tmp7) & RANGE_MASK];
+    outptr[7] = range_limit[(int) (tmp0 - tmp7) & RANGE_MASK];
+    outptr[1] = range_limit[(int) (tmp1 + tmp6) & RANGE_MASK];
+    outptr[6] = range_limit[(int) (tmp1 - tmp6) & RANGE_MASK];
+    outptr[2] = range_limit[(int) (tmp2 + tmp5) & RANGE_MASK];
+    outptr[5] = range_limit[(int) (tmp2 - tmp5) & RANGE_MASK];
+    outptr[3] = range_limit[(int) (tmp3 + tmp4) & RANGE_MASK];
+    outptr[4] = range_limit[(int) (tmp3 - tmp4) & RANGE_MASK];
+
+    wsptr += DCTSIZE;		/* advance pointer to next row */
+  }
+}
+
+#endif /* DCT_FLOAT_SUPPORTED */
@@ -0,0 +1,351 @@
+/*
+ * jidctfst.c
+ *
+ * Copyright (C) 1994-1998, Thomas G. Lane.
+ * Modified 2015-2017 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains a fast, not so accurate integer implementation of the
+ * inverse DCT (Discrete Cosine Transform).  In the IJG code, this routine
+ * must also perform dequantization of the input coefficients.
+ *
+ * A 2-D IDCT can be done by 1-D IDCT on each column followed by 1-D IDCT
+ * on each row (or vice versa, but it's more convenient to emit a row at
+ * a time).  Direct algorithms are also available, but they are much more
+ * complex and seem not to be any faster when reduced to code.
+ *
+ * This implementation is based on Arai, Agui, and Nakajima's algorithm for
+ * scaled DCT.  Their original paper (Trans. IEICE E-71(11):1095) is in
+ * Japanese, but the algorithm is described in the Pennebaker & Mitchell
+ * JPEG textbook (see REFERENCES section in file README).  The following code
+ * is based directly on figure 4-8 in P&M.
+ * While an 8-point DCT cannot be done in less than 11 multiplies, it is
+ * possible to arrange the computation so that many of the multiplies are
+ * simple scalings of the final outputs.  These multiplies can then be
+ * folded into the multiplications or divisions by the JPEG quantization
+ * table entries.  The AA&N method leaves only 5 multiplies and 29 adds
+ * to be done in the DCT itself.
+ * The primary disadvantage of this method is that with fixed-point math,
+ * accuracy is lost due to imprecise representation of the scaled
+ * quantization values.  The smaller the quantization table entry, the less
+ * precise the scaled value, so this implementation does worse with high-
+ * quality-setting files than with low-quality ones.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jdct.h"		/* Private declarations for DCT subsystem */
+
+#ifdef DCT_IFAST_SUPPORTED
+
+
+/*
+ * This module is specialized to the case DCTSIZE = 8.
+ */
+
+#if DCTSIZE != 8
+  Sorry, this code only copes with 8x8 DCT blocks. /* deliberate syntax err */
+#endif
+
+
+/* Scaling decisions are generally the same as in the LL&M algorithm;
+ * see jidctint.c for more details.  However, we choose to descale
+ * (right shift) multiplication products as soon as they are formed,
+ * rather than carrying additional fractional bits into subsequent additions.
+ * This compromises accuracy slightly, but it lets us save a few shifts.
+ * More importantly, 16-bit arithmetic is then adequate (for 8-bit samples)
+ * everywhere except in the multiplications proper; this saves a good deal
+ * of work on 16-bit-int machines.
+ *
+ * The dequantized coefficients are not integers because the AA&N scaling
+ * factors have been incorporated.  We represent them scaled up by PASS1_BITS,
+ * so that the first and second IDCT rounds have the same input scaling.
+ * For 8-bit JSAMPLEs, we choose IFAST_SCALE_BITS = PASS1_BITS so as to
+ * avoid a descaling shift; this compromises accuracy rather drastically
+ * for small quantization table entries, but it saves a lot of shifts.
+ * For 12-bit JSAMPLEs, there's no hope of using 16x16 multiplies anyway,
+ * so we use a much larger scaling factor to preserve accuracy.
+ *
+ * A final compromise is to represent the multiplicative constants to only
+ * 8 fractional bits, rather than 13.  This saves some shifting work on some
+ * machines, and may also reduce the cost of multiplication (since there
+ * are fewer one-bits in the constants).
+ */
+
+#if BITS_IN_JSAMPLE == 8
+#define CONST_BITS  8
+#define PASS1_BITS  2
+#else
+#define CONST_BITS  8
+#define PASS1_BITS  1		/* lose a little precision to avoid overflow */
+#endif
+
+/* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
+ * causing a lot of useless floating-point operations at run time.
+ * To get around this we use the following pre-calculated constants.
+ * If you change CONST_BITS you may want to add appropriate values.
+ * (With a reasonable C compiler, you can just rely on the FIX() macro...)
+ */
+
+#if CONST_BITS == 8
+#define FIX_1_082392200  ((INT32)  277)		/* FIX(1.082392200) */
+#define FIX_1_414213562  ((INT32)  362)		/* FIX(1.414213562) */
+#define FIX_1_847759065  ((INT32)  473)		/* FIX(1.847759065) */
+#define FIX_2_613125930  ((INT32)  669)		/* FIX(2.613125930) */
+#else
+#define FIX_1_082392200  FIX(1.082392200)
+#define FIX_1_414213562  FIX(1.414213562)
+#define FIX_1_847759065  FIX(1.847759065)
+#define FIX_2_613125930  FIX(2.613125930)
+#endif
+
+
+/* We can gain a little more speed, with a further compromise in accuracy,
+ * by omitting the addition in a descaling shift.  This yields an incorrectly
+ * rounded result half the time...
+ */
+
+#ifndef USE_ACCURATE_ROUNDING
+#undef DESCALE
+#define DESCALE(x,n)  RIGHT_SHIFT(x, n)
+#endif
+
+
+/* Multiply a DCTELEM variable by an INT32 constant, and immediately
+ * descale to yield a DCTELEM result.
+ */
+
+#define MULTIPLY(var,const)  ((DCTELEM) DESCALE((var) * (const), CONST_BITS))
+
+
+/* Dequantize a coefficient by multiplying it by the multiplier-table
+ * entry; produce a DCTELEM result.  For 8-bit data a 16x16->16
+ * multiplication will do.  For 12-bit data, the multiplier table is
+ * declared INT32, so a 32-bit multiply will be used.
+ */
+
+#if BITS_IN_JSAMPLE == 8
+#define DEQUANTIZE(coef,quantval)  (((IFAST_MULT_TYPE) (coef)) * (quantval))
+#else
+#define DEQUANTIZE(coef,quantval)  \
+	DESCALE((coef)*(quantval), IFAST_SCALE_BITS-PASS1_BITS)
+#endif
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients.
+ *
+ * cK represents cos(K*pi/16).
+ */
+
+GLOBAL(void)
+jpeg_idct_ifast (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+		 JCOEFPTR coef_block,
+		 JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  DCTELEM tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
+  DCTELEM tmp10, tmp11, tmp12, tmp13;
+  DCTELEM z5, z10, z11, z12, z13;
+  JCOEFPTR inptr;
+  IFAST_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[DCTSIZE2];	/* buffers data between passes */
+  SHIFT_TEMPS			/* for DESCALE */
+  ISHIFT_TEMPS			/* for IRIGHT_SHIFT */
+
+  /* Pass 1: process columns from input, store into work array. */
+
+  inptr = coef_block;
+  quantptr = (IFAST_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = DCTSIZE; ctr > 0; ctr--) {
+    /* Due to quantization, we will usually find that many of the input
+     * coefficients are zero, especially the AC terms.  We can exploit this
+     * by short-circuiting the IDCT calculation for any column in which all
+     * the AC terms are zero.  In that case each output is equal to the
+     * DC coefficient (with scale factor as needed).
+     * With typical images and quantization tables, half or more of the
+     * column DCT calculations can be simplified this way.
+     */
+    
+    if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
+	inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*4] == 0 &&
+	inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*6] == 0 &&
+	inptr[DCTSIZE*7] == 0) {
+      /* AC terms all zero */
+      int dcval = (int) DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+
+      wsptr[DCTSIZE*0] = dcval;
+      wsptr[DCTSIZE*1] = dcval;
+      wsptr[DCTSIZE*2] = dcval;
+      wsptr[DCTSIZE*3] = dcval;
+      wsptr[DCTSIZE*4] = dcval;
+      wsptr[DCTSIZE*5] = dcval;
+      wsptr[DCTSIZE*6] = dcval;
+      wsptr[DCTSIZE*7] = dcval;
+      
+      inptr++;			/* advance pointers to next column */
+      quantptr++;
+      wsptr++;
+      continue;
+    }
+    
+    /* Even part */
+
+    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    tmp1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    tmp2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
+    tmp3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
+
+    tmp10 = tmp0 + tmp2;	/* phase 3 */
+    tmp11 = tmp0 - tmp2;
+
+    tmp13 = tmp1 + tmp3;	/* phases 5-3 */
+    tmp12 = MULTIPLY(tmp1 - tmp3, FIX_1_414213562) - tmp13; /* 2*c4 */
+
+    tmp0 = tmp10 + tmp13;	/* phase 2 */
+    tmp3 = tmp10 - tmp13;
+    tmp1 = tmp11 + tmp12;
+    tmp2 = tmp11 - tmp12;
+    
+    /* Odd part */
+
+    tmp4 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    tmp5 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+    tmp6 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
+    tmp7 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
+
+    z13 = tmp6 + tmp5;		/* phase 6 */
+    z10 = tmp6 - tmp5;
+    z11 = tmp4 + tmp7;
+    z12 = tmp4 - tmp7;
+
+    tmp7 = z11 + z13;		/* phase 5 */
+    tmp11 = MULTIPLY(z11 - z13, FIX_1_414213562); /* 2*c4 */
+
+    z5 = MULTIPLY(z10 + z12, FIX_1_847759065); /* 2*c2 */
+    tmp10 = z5 - MULTIPLY(z12, FIX_1_082392200); /* 2*(c2-c6) */
+    tmp12 = z5 - MULTIPLY(z10, FIX_2_613125930); /* 2*(c2+c6) */
+
+    tmp6 = tmp12 - tmp7;	/* phase 2 */
+    tmp5 = tmp11 - tmp6;
+    tmp4 = tmp10 - tmp5;
+
+    wsptr[DCTSIZE*0] = (int) (tmp0 + tmp7);
+    wsptr[DCTSIZE*7] = (int) (tmp0 - tmp7);
+    wsptr[DCTSIZE*1] = (int) (tmp1 + tmp6);
+    wsptr[DCTSIZE*6] = (int) (tmp1 - tmp6);
+    wsptr[DCTSIZE*2] = (int) (tmp2 + tmp5);
+    wsptr[DCTSIZE*5] = (int) (tmp2 - tmp5);
+    wsptr[DCTSIZE*3] = (int) (tmp3 + tmp4);
+    wsptr[DCTSIZE*4] = (int) (tmp3 - tmp4);
+
+    inptr++;			/* advance pointers to next column */
+    quantptr++;
+    wsptr++;
+  }
+  
+  /* Pass 2: process rows from work array, store into output array.
+   * Note that we must descale the results by a factor of 8 == 2**3,
+   * and also undo the PASS1_BITS scaling.
+   */
+
+  wsptr = workspace;
+  for (ctr = 0; ctr < DCTSIZE; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Add range center and fudge factor for final descale and range-limit. */
+    z5 = (DCTELEM) wsptr[0] +
+	   ((((DCTELEM) RANGE_CENTER) << (PASS1_BITS+3)) +
+	    (1 << (PASS1_BITS+2)));
+
+    /* Rows of zeroes can be exploited in the same way as we did with columns.
+     * However, the column calculation has created many nonzero AC terms, so
+     * the simplification applies less often (typically 5% to 10% of the time).
+     * On machines with very fast multiplication, it's possible that the
+     * test takes more time than it's worth.  In that case this section
+     * may be commented out.
+     */
+    
+#ifndef NO_ZERO_ROW_TEST
+    if (wsptr[1] == 0 && wsptr[2] == 0 && wsptr[3] == 0 && wsptr[4] == 0 &&
+	wsptr[5] == 0 && wsptr[6] == 0 && wsptr[7] == 0) {
+      /* AC terms all zero */
+      JSAMPLE dcval = range_limit[(int) IRIGHT_SHIFT(z5, PASS1_BITS+3)
+				  & RANGE_MASK];
+      
+      outptr[0] = dcval;
+      outptr[1] = dcval;
+      outptr[2] = dcval;
+      outptr[3] = dcval;
+      outptr[4] = dcval;
+      outptr[5] = dcval;
+      outptr[6] = dcval;
+      outptr[7] = dcval;
+
+      wsptr += DCTSIZE;		/* advance pointer to next row */
+      continue;
+    }
+#endif
+    
+    /* Even part */
+
+    tmp10 = z5 + (DCTELEM) wsptr[4];
+    tmp11 = z5 - (DCTELEM) wsptr[4];
+
+    tmp13 = (DCTELEM) wsptr[2] + (DCTELEM) wsptr[6];
+    tmp12 = MULTIPLY((DCTELEM) wsptr[2] - (DCTELEM) wsptr[6],
+		     FIX_1_414213562) - tmp13; /* 2*c4 */
+
+    tmp0 = tmp10 + tmp13;
+    tmp3 = tmp10 - tmp13;
+    tmp1 = tmp11 + tmp12;
+    tmp2 = tmp11 - tmp12;
+
+    /* Odd part */
+
+    z13 = (DCTELEM) wsptr[5] + (DCTELEM) wsptr[3];
+    z10 = (DCTELEM) wsptr[5] - (DCTELEM) wsptr[3];
+    z11 = (DCTELEM) wsptr[1] + (DCTELEM) wsptr[7];
+    z12 = (DCTELEM) wsptr[1] - (DCTELEM) wsptr[7];
+
+    tmp7 = z11 + z13;		/* phase 5 */
+    tmp11 = MULTIPLY(z11 - z13, FIX_1_414213562); /* 2*c4 */
+
+    z5 = MULTIPLY(z10 + z12, FIX_1_847759065); /* 2*c2 */
+    tmp10 = z5 - MULTIPLY(z12, FIX_1_082392200); /* 2*(c2-c6) */
+    tmp12 = z5 - MULTIPLY(z10, FIX_2_613125930); /* 2*(c2+c6) */
+
+    tmp6 = tmp12 - tmp7;	/* phase 2 */
+    tmp5 = tmp11 - tmp6;
+    tmp4 = tmp10 - tmp5;
+
+    /* Final output stage: scale down by a factor of 8 and range-limit */
+
+    outptr[0] = range_limit[(int) IRIGHT_SHIFT(tmp0 + tmp7, PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[7] = range_limit[(int) IRIGHT_SHIFT(tmp0 - tmp7, PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[1] = range_limit[(int) IRIGHT_SHIFT(tmp1 + tmp6, PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[6] = range_limit[(int) IRIGHT_SHIFT(tmp1 - tmp6, PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[2] = range_limit[(int) IRIGHT_SHIFT(tmp2 + tmp5, PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[5] = range_limit[(int) IRIGHT_SHIFT(tmp2 - tmp5, PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[3] = range_limit[(int) IRIGHT_SHIFT(tmp3 + tmp4, PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[4] = range_limit[(int) IRIGHT_SHIFT(tmp3 - tmp4, PASS1_BITS+3)
+			    & RANGE_MASK];
+
+    wsptr += DCTSIZE;		/* advance pointer to next row */
+  }
+}
+
+#endif /* DCT_IFAST_SUPPORTED */
@@ -0,0 +1,157 @@
+/*
+ * jinclude.h
+ *
+ * Copyright (C) 1991-1994, Thomas G. Lane.
+ * Modified 2017-2022 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file exists to provide a single place to fix any problems with
+ * including the wrong system include files.  (Common problems are taken
+ * care of by the standard jconfig symbols, but on really weird systems
+ * you may have to edit this file.)
+ *
+ * NOTE: this file is NOT intended to be included by applications using
+ * the JPEG library.  Most applications need only include jpeglib.h.
+ */
+
+
+/* Include auto-config file to find out which system include files we need. */
+
+#include "jconfig.h"		/* auto configuration options */
+#define JCONFIG_INCLUDED	/* so that jpeglib.h doesn't do it again */
+
+/*
+ * We need the NULL macro and size_t typedef.
+ * On an ANSI-conforming system it is sufficient to include <stddef.h>.
+ * Otherwise, we get them from <stdlib.h> or <stdio.h>; we may have to
+ * pull in <sys/types.h> as well.
+ * Note that the core JPEG library does not require <stdio.h>;
+ * only the default error handler and data source/destination modules do.
+ * But we must pull it in because of the references to FILE in jpeglib.h.
+ * You can remove those references if you want to compile without <stdio.h>.
+ */
+
+#ifdef HAVE_STDDEF_H
+#include <stddef.h>
+#endif
+
+#ifdef HAVE_STDLIB_H
+#include <stdlib.h>
+#endif
+
+#ifdef NEED_SYS_TYPES_H
+#include <sys/types.h>
+#endif
+
+#include <stdio.h>
+
+/*
+ * We need memory copying and zeroing functions, plus strncpy().
+ * ANSI and System V implementations declare these in <string.h>.
+ * BSD doesn't have the mem() functions, but it does have bcopy()/bzero().
+ * Some systems may declare memset and memcpy in <memory.h>.
+ *
+ * NOTE: we assume the size parameters to these functions are of type size_t.
+ * Change the casts in these macros if not!
+ */
+
+#ifdef NEED_BSD_STRINGS
+
+#include <strings.h>
+#define MEMZERO(target,size)	bzero((void *)(target), (size_t)(size))
+#define MEMCOPY(dest,src,size)	bcopy((const void *)(src), (void *)(dest), (size_t)(size))
+
+#else /* not BSD, assume ANSI/SysV string lib */
+
+#include <string.h>
+#define MEMZERO(target,size)	memset((void *)(target), 0, (size_t)(size))
+#define MEMCOPY(dest,src,size)	memcpy((void *)(dest), (const void *)(src), (size_t)(size))
+
+#endif
+
+/*
+ * In ANSI C, and indeed any rational implementation, size_t is also the
+ * type returned by sizeof().  However, it seems there are some irrational
+ * implementations out there, in which sizeof() returns an int even though
+ * size_t is defined as long or unsigned long.  To ensure consistent results
+ * we always use this SIZEOF() macro in place of using sizeof() directly.
+ */
+
+#define SIZEOF(object)	((size_t) sizeof(object))
+
+/*
+ * The modules that use fread() and fwrite() always invoke them through
+ * these macros.  On some systems you may need to twiddle the argument casts.
+ * CAUTION: argument order is different from underlying functions!
+ *
+ * Furthermore, macros are provided for fflush() and ferror() in order
+ * to facilitate adaption by applications using an own FILE class.
+ *
+ * You can define your own custom file I/O functions in jconfig.h and
+ * #define JPEG_HAVE_FILE_IO_CUSTOM there to prevent redefinition here.
+ *
+ * You can #define JPEG_USE_FILE_IO_CUSTOM in jconfig.h to use custom file
+ * I/O functions implemented in Delphi VCL (Visual Component Library)
+ * in Vcl.Imaging.jpeg.pas for the TJPEGImage component utilizing
+ * the Delphi RTL (Run-Time Library) TMemoryStream component:
+ *
+ *   procedure jpeg_stdio_src(var cinfo: jpeg_decompress_struct;
+ *     input_file: TStream); external;
+ *
+ *   procedure jpeg_stdio_dest(var cinfo: jpeg_compress_struct;
+ *     output_file: TStream); external;
+ *
+ *   function jfread(var buf; recsize, reccount: Integer; S: TStream): Integer;
+ *   begin
+ *     Result := S.Read(buf, recsize * reccount);
+ *   end;
+ *
+ *   function jfwrite(const buf; recsize, reccount: Integer; S: TStream): Integer;
+ *   begin
+ *     Result := S.Write(buf, recsize * reccount);
+ *   end;
+ *
+ *   function jfflush(S: TStream): Integer;
+ *   begin
+ *     Result := 0;
+ *   end;
+ *
+ *   function jferror(S: TStream): Integer;
+ *   begin
+ *     Result := 0;
+ *   end;
+ *
+ * TMemoryStream of Delphi RTL has the distinctive feature to provide dynamic
+ * memory buffer management with a file/stream-based interface, particularly for
+ * the write (output) operation, which is easier to apply compared with direct
+ * implementations as given in jdatadst.c for memory destination.  Those direct
+ * implementations of dynamic memory write tend to be more difficult to use,
+ * so providing an option like TMemoryStream may be a useful alternative.
+ *
+ * The CFile/CMemFile classes of the Microsoft Foundation Class (MFC) Library
+ * may be used in a similar fashion.
+ */
+
+#ifndef JPEG_HAVE_FILE_IO_CUSTOM
+#ifdef JPEG_USE_FILE_IO_CUSTOM
+extern size_t jfread(void * __ptr, size_t __size, size_t __n, FILE * __stream);
+extern size_t jfwrite(const void * __ptr, size_t __size, size_t __n, FILE * __stream);
+extern int    jfflush(FILE * __stream);
+extern int    jferror(FILE * __fp);
+
+#define JFREAD(file,buf,sizeofbuf)  \
+  ((size_t) jfread((void *) (buf), (size_t) 1, (size_t) (sizeofbuf), (file)))
+#define JFWRITE(file,buf,sizeofbuf)  \
+  ((size_t) jfwrite((const void *) (buf), (size_t) 1, (size_t) (sizeofbuf), (file)))
+#define JFFLUSH(file)	jfflush(file)
+#define JFERROR(file)	jferror(file)
+#else
+#define JFREAD(file,buf,sizeofbuf)  \
+  ((size_t) fread((void *) (buf), (size_t) 1, (size_t) (sizeofbuf), (file)))
+#define JFWRITE(file,buf,sizeofbuf)  \
+  ((size_t) fwrite((const void *) (buf), (size_t) 1, (size_t) (sizeofbuf), (file)))
+#define JFFLUSH(file)	fflush(file)
+#define JFERROR(file)	ferror(file)
+#endif
+#endif
--- a/Show More
+++ b/Show More