release: OpenCV 3.4.17

Merge pull request #21341 from alalek:issue_20930_3.4
python: use '((x,y), (w,h), angle)' in std::vector<RotatedRect>
2026-07-21 19:33:03 +04:00 · 2021-12-24 16:39:15 +00:00 · 2021-12-24 16:33:19 +00:00 · 2021-12-24 15:01:45 +00:00 · 2021-12-24 11:56:12 +00:00 · 2021-12-24 11:55:55 +00:00
1780 changed files with 121153 additions and 32165 deletions
@@ -1,5 +1,5 @@
 <!--
-If you have a question rather than reporting a bug please go to http://answers.opencv.org where you get much faster responses.
+If you have a question rather than reporting a bug please go to https://forum.opencv.org where you get much faster responses.
 If you need further assistance please read [How To Contribute](https://github.com/opencv/opencv/wiki/How_to_contribute).

 This is a template helping you to create an issue which can be processed as quickly as possible. This is the bug reporting section for the OpenCV library.
@@ -27,4 +27,4 @@ This is a template helping you to create an issue which can be processed as quic
    // C++ code example
    ```
 or attach as .txt or .zip file
-->
+-->
@@ -22,3 +22,5 @@ bin/
 *.log
 *.tlog
 build
+node_modules
+CMakeSettings.json
@@ -20,11 +20,14 @@ if(CMAKE_COMPILER_IS_GNUCC)
    # - matchTemplate about 5-10%
    # - goodFeaturesToTrack 10-20%
    # - cornerHarris 30% for some cases
-
-    set_source_files_properties(${carotene_sources} COMPILE_FLAGS "--param ipcp-unit-growth=100000 --param inline-unit-growth=100000 --param large-stack-frame-growth=5000")
+    if(CMAKE_CXX_COMPILER_VERSION VERSION_LESS "10.0.0")
+        set_source_files_properties(${carotene_sources} COMPILE_FLAGS "--param ipcp-unit-growth=100000 --param inline-unit-growth=100000 --param large-stack-frame-growth=5000")
+    else()
+        set_source_files_properties(${carotene_sources} COMPILE_FLAGS "--param ipa-cp-unit-growth=100000 --param inline-unit-growth=100000 --param large-stack-frame-growth=5000")
+    endif()
 endif()

-add_library(carotene_objs OBJECT
+add_library(carotene_objs OBJECT EXCLUDE_FROM_ALL
  ${carotene_headers}
  ${carotene_sources}
 )
@@ -37,4 +40,5 @@ if(WITH_NEON)
    target_compile_definitions(carotene_objs PRIVATE "-DWITH_NEON")
 endif()

-add_library(carotene STATIC EXCLUDE_FROM_ALL "$<TARGET_OBJECTS:carotene_objs>")
+# we add dummy file to fix XCode build
+add_library(carotene STATIC ${OPENCV_3RDPARTY_EXCLUDE_FROM_ALL} "$<TARGET_OBJECTS:carotene_objs>" "${CAROTENE_SOURCE_DIR}/dummy.cpp")
@@ -58,22 +58,6 @@ function(compile_carotene)
  endif()

  add_subdirectory("${CAROTENE_DIR}" "${CMAKE_CURRENT_BINARY_DIR}/carotene")
-
-  if(ARM OR AARCH64)
-    if(CMAKE_BUILD_TYPE)
-      set(CMAKE_TRY_COMPILE_CONFIGURATION ${CMAKE_BUILD_TYPE})
-    endif()
-    check_cxx_compiler_flag("-mfpu=neon" CXX_HAS_MFPU_NEON)
-    check_c_compiler_flag("-mfpu=neon" C_HAS_MFPU_NEON)
-    if(${CXX_HAS_MFPU_NEON} AND ${C_HAS_MFPU_NEON} AND NOT "${CMAKE_CXX_FLAGS} " MATCHES "-mfpu=neon[^ ]*")
-      get_target_property(old_flags "carotene_objs" COMPILE_FLAGS)
-      if(old_flags)
-        set_target_properties("carotene_objs" PROPERTIES COMPILE_FLAGS "${old_flags} -mfpu=neon")
-      else()
-        set_target_properties("carotene_objs" PROPERTIES COMPILE_FLAGS "-mfpu=neon")
-      endif()
-    endif()
-  endif()
 endfunction()

 compile_carotene()
@@ -88,11 +72,16 @@ set_property(DIRECTORY APPEND PROPERTY COMPILE_DEFINITIONS ${carotene_defs})
    #   matchTemplate about 5-10%
    #   goodFeaturesToTrack 10-20%
    #   cornerHarris 30% for some cases
-    set_source_files_properties(impl.cpp $<TARGET_OBJECTS:carotene_objs> COMPILE_FLAGS "--param ipcp-unit-growth=100000 --param inline-unit-growth=100000 --param large-stack-frame-growth=5000")
+    if(CMAKE_CXX_COMPILER_VERSION VERSION_LESS "10.0.0")
+      set_source_files_properties(impl.cpp $<TARGET_OBJECTS:carotene_objs> COMPILE_FLAGS "--param ipcp-unit-growth=100000 --param inline-unit-growth=100000 --param large-stack-frame-growth=5000")
+    else()
+      set_source_files_properties(impl.cpp $<TARGET_OBJECTS:carotene_objs> COMPILE_FLAGS "--param ipa-cp-unit-growth=100000 --param inline-unit-growth=100000 --param large-stack-frame-growth=5000")
+    endif()
 #    set_source_files_properties(impl.cpp $<TARGET_OBJECTS:carotene_objs> COMPILE_FLAGS "--param ipcp-unit-growth=100000 --param inline-unit-growth=100000 --param large-stack-frame-growth=5000")
  endif()

-add_library(tegra_hal STATIC $<TARGET_OBJECTS:carotene_objs>)
+# we add dummy file to fix XCode build
+add_library(tegra_hal STATIC $<TARGET_OBJECTS:carotene_objs> "dummy.cpp")
 set_target_properties(tegra_hal PROPERTIES ARCHIVE_OUTPUT_DIRECTORY ${3P_LIBRARY_OUTPUT_PATH})
 set(OPENCV_SRC_DIR "${CMAKE_SOURCE_DIR}")
 if(NOT BUILD_SHARED_LIBS)
@@ -0,0 +1,2 @@
+// This file is needed for compilation on some platforms e.g. with XCode generator
+// Related issue: https://gitlab.kitware.com/cmake/cmake/-/issues/17457
@@ -1127,6 +1127,7 @@ inline int TEGRA_FILTERINIT(cvhalFilter2D **context, uchar *kernel_data, size_t
        {
            std::memcpy(ctx->kernel_data + kernel_width * j, kernel_data + kernel_step * j, kernel_width * sizeof(int16_t));
        }
+        break;
    default:
        delete[] ctx->kernel_data;
        delete ctx;
@@ -1243,6 +1244,7 @@ inline int TEGRA_SEPFILTERINIT(cvhalFilter2D **context, int src_type, int dst_ty
        ctx->kernely_data[0]=((int16_t*)kernely_data)[0];
        ctx->kernely_data[1]=((int16_t*)kernely_data)[1];
        ctx->kernely_data[2]=((int16_t*)kernely_data)[2];
+        break;
    default:
        delete ctx;
        return CV_HAL_ERROR_NOT_IMPLEMENTED;
@@ -1776,30 +1778,30 @@ TegraCvtColor_Invoker(bgrx2hsvf, bgrx2hsv, src_data + static_cast<size_t>(range.
    : CV_HAL_ERROR_NOT_IMPLEMENTED \
 )

-#define TEGRA_CVT2PYUVTOBGR(src_data, src_step, dst_data, dst_step, dst_width, dst_height, dcn, swapBlue, uIdx) \
+#define TEGRA_CVT2PYUVTOBGR_EX(y_data, y_step, uv_data, uv_step, dst_data, dst_step, dst_width, dst_height, dcn, swapBlue, uIdx) \
 ( \
    CAROTENE_NS::isSupportedConfiguration() ? \
        dcn == 3 ? \
            uIdx == 0 ? \
                (swapBlue ? \
                    CAROTENE_NS::yuv420i2rgb(CAROTENE_NS::Size2D(dst_width, dst_height), \
-                                             src_data, src_step, \
-                                             src_data + src_step * dst_height, src_step, \
+                                             y_data, y_step, \
+                                             uv_data, uv_step, \
                                             dst_data, dst_step) : \
                    CAROTENE_NS::yuv420i2bgr(CAROTENE_NS::Size2D(dst_width, dst_height), \
-                                             src_data, src_step, \
-                                             src_data + src_step * dst_height, src_step, \
+                                             y_data, y_step, \
+                                             uv_data, uv_step, \
                                             dst_data, dst_step)), \
                CV_HAL_ERROR_OK : \
            uIdx == 1 ? \
                (swapBlue ? \
                    CAROTENE_NS::yuv420sp2rgb(CAROTENE_NS::Size2D(dst_width, dst_height), \
-                                              src_data, src_step, \
-                                              src_data + src_step * dst_height, src_step, \
+                                              y_data, y_step, \
+                                              uv_data, uv_step, \
                                              dst_data, dst_step) : \
                    CAROTENE_NS::yuv420sp2bgr(CAROTENE_NS::Size2D(dst_width, dst_height), \
-                                              src_data, src_step, \
-                                              src_data + src_step * dst_height, src_step, \
+                                              y_data, y_step, \
+                                              uv_data, uv_step, \
                                              dst_data, dst_step)), \
                CV_HAL_ERROR_OK : \
            CV_HAL_ERROR_NOT_IMPLEMENTED : \
@@ -1807,29 +1809,32 @@ TegraCvtColor_Invoker(bgrx2hsvf, bgrx2hsv, src_data + static_cast<size_t>(range.
            uIdx == 0 ? \
                (swapBlue ? \
                    CAROTENE_NS::yuv420i2rgbx(CAROTENE_NS::Size2D(dst_width, dst_height), \
-                                              src_data, src_step, \
-                                              src_data + src_step * dst_height, src_step, \
+                                              y_data, y_step, \
+                                              uv_data, uv_step, \
                                              dst_data, dst_step) : \
                    CAROTENE_NS::yuv420i2bgrx(CAROTENE_NS::Size2D(dst_width, dst_height), \
-                                              src_data, src_step, \
-                                              src_data + src_step * dst_height, src_step, \
+                                              y_data, y_step, \
+                                              uv_data, uv_step, \
                                              dst_data, dst_step)), \
                CV_HAL_ERROR_OK : \
            uIdx == 1 ? \
                (swapBlue ? \
                    CAROTENE_NS::yuv420sp2rgbx(CAROTENE_NS::Size2D(dst_width, dst_height), \
-                                               src_data, src_step, \
-                                               src_data + src_step * dst_height, src_step, \
+                                               y_data, y_step, \
+                                               uv_data, uv_step, \
                                               dst_data, dst_step) : \
                    CAROTENE_NS::yuv420sp2bgrx(CAROTENE_NS::Size2D(dst_width, dst_height), \
-                                               src_data, src_step, \
-                                               src_data + src_step * dst_height, src_step, \
+                                               y_data, y_step, \
+                                               uv_data, uv_step, \
                                               dst_data, dst_step)), \
                CV_HAL_ERROR_OK : \
            CV_HAL_ERROR_NOT_IMPLEMENTED : \
        CV_HAL_ERROR_NOT_IMPLEMENTED \
    : CV_HAL_ERROR_NOT_IMPLEMENTED \
 )
+#define TEGRA_CVT2PYUVTOBGR(src_data, src_step, dst_data, dst_step, dst_width, dst_height, dcn, swapBlue, uIdx) \
+    TEGRA_CVT2PYUVTOBGR_EX(src_data, src_step, src_data + src_step * dst_height, src_step, dst_data, dst_step, \
+            dst_width, dst_height, dcn, swapBlue, uIdx);

 #undef cv_hal_cvtBGRtoBGR
 #define cv_hal_cvtBGRtoBGR TEGRA_CVTBGRTOBGR
@@ -1839,12 +1844,18 @@ TegraCvtColor_Invoker(bgrx2hsvf, bgrx2hsv, src_data + static_cast<size_t>(range.
 #define cv_hal_cvtBGRtoGray TEGRA_CVTBGRTOGRAY
 #undef cv_hal_cvtGraytoBGR
 #define cv_hal_cvtGraytoBGR TEGRA_CVTGRAYTOBGR
+#if 0  // bit-exact tests are failed
 #undef cv_hal_cvtBGRtoYUV
 #define cv_hal_cvtBGRtoYUV TEGRA_CVTBGRTOYUV
+#endif
 #undef cv_hal_cvtBGRtoHSV
 #define cv_hal_cvtBGRtoHSV TEGRA_CVTBGRTOHSV
+#if 0  // bit-exact tests are failed
 #undef cv_hal_cvtTwoPlaneYUVtoBGR
 #define cv_hal_cvtTwoPlaneYUVtoBGR TEGRA_CVT2PYUVTOBGR
+#undef cv_hal_cvtTwoPlaneYUVtoBGREx
+#define cv_hal_cvtTwoPlaneYUVtoBGREx TEGRA_CVT2PYUVTOBGR_EX
+#endif

 #endif // OPENCV_IMGPROC_HAL_INTERFACE_H

@@ -0,0 +1,2 @@
+// This file is needed for compilation on some platforms e.g. with XCode generator
+// Related issue: https://gitlab.kitware.com/cmake/cmake/-/issues/17457
@@ -758,7 +758,7 @@ inline void resizeAreaRounding(const Size2D &ssize, const Size2D &dsize,
    }
    else if (channels == 3)
    {
-        if ((wr == 2.0f) && (wr == 2.0f))
+        if ((wr == 2.0f) && (hr == 2.0f))
        {
 #ifndef __ANDROID__
            size_t roiw16 = dsize.width >= 15 ? (dsize.width - 15) * 3 : 0;
@@ -14,7 +14,7 @@ if(NOT DEFINED CPUFEATURES_SOURCES)
 endif()

 include_directories(${CPUFEATURES_INCLUDE_DIRS})
-add_library(${OPENCV_CPUFEATURES_TARGET_NAME} STATIC ${CPUFEATURES_SOURCES})
+add_library(${OPENCV_CPUFEATURES_TARGET_NAME} STATIC ${OPENCV_3RDPARTY_EXCLUDE_FROM_ALL} ${CPUFEATURES_SOURCES})

 set_target_properties(${OPENCV_CPUFEATURES_TARGET_NAME}
  PROPERTIES OUTPUT_NAME cpufeatures
@@ -29,7 +29,7 @@ if(ENABLE_SOLUTION_FOLDERS)
 endif()

 if(NOT BUILD_SHARED_LIBS)
-  ocv_install_target(${OPENCV_CPUFEATURES_TARGET_NAME} EXPORT OpenCVModules ARCHIVE DESTINATION ${OPENCV_3P_LIB_INSTALL_PATH} COMPONENT dev)
+  ocv_install_target(${OPENCV_CPUFEATURES_TARGET_NAME} EXPORT OpenCVModules ARCHIVE DESTINATION ${OPENCV_3P_LIB_INSTALL_PATH} COMPONENT dev OPTIONAL)
 endif()

 ocv_install_3rdparty_licenses(cpufeatures LICENSE README.md)
@@ -1,8 +1,8 @@
-# Binaries branch name: ffmpeg/3.4_20190612
-# Binaries were created for OpenCV: 1c661e754812d3423e8c15ccb9cdab57fc122c44
-ocv_update(FFMPEG_BINARIES_COMMIT "d284fd09e76966e95f08e1ac3a1a28b227447501")
-ocv_update(FFMPEG_FILE_HASH_BIN32 "a7032bf6cc1cb010dc055f352277792c")
-ocv_update(FFMPEG_FILE_HASH_BIN64 "40d1a38b83a0413a3fcf9df1044fff2a")
+# Binaries branch name: ffmpeg/3.4_20211220
+# Binaries were created for OpenCV: a22dd28e0272ec0f1cfee8811d3f5f0392827c65
+ocv_update(FFMPEG_BINARIES_COMMIT "5a7644ec3940c6eed41c6ebb5a0602a5615fdb3f")
+ocv_update(FFMPEG_FILE_HASH_BIN32 "8ad9de6f1f2ca77786748d1f3a4e83ea")
+ocv_update(FFMPEG_FILE_HASH_BIN64 "2c670f068252e7cd28d3883993dc1d6e")
 ocv_update(FFMPEG_FILE_HASH_CMAKE "3b90f67f4b429e77d3da36698cef700c")

 function(download_win_ffmpeg script_var)
@@ -17,7 +17,7 @@ file(GLOB lib_hdrs ${IPP_IW_PATH}/include/*.h ${IPP_IW_PATH}/include/iw/*.h ${IP
 #         Define the library target:
 # ----------------------------------------------------------------------------------

-add_library(${IPP_IW_LIBRARY} STATIC ${lib_srcs} ${lib_hdrs})
+add_library(${IPP_IW_LIBRARY} STATIC ${OPENCV_3RDPARTY_EXCLUDE_FROM_ALL} ${lib_srcs} ${lib_hdrs})

 if(UNIX)
  if(CV_GCC OR CV_CLANG OR CV_ICC)
@@ -41,5 +41,5 @@ if(ENABLE_SOLUTION_FOLDERS)
 endif()

 if(NOT BUILD_SHARED_LIBS)
-  ocv_install_target(${IPP_IW_LIBRARY} EXPORT OpenCVModules ARCHIVE DESTINATION ${OPENCV_3P_LIB_INSTALL_PATH} COMPONENT dev)
+  ocv_install_target(${IPP_IW_LIBRARY} EXPORT OpenCVModules ARCHIVE DESTINATION ${OPENCV_3P_LIB_INSTALL_PATH} COMPONENT dev OPTIONAL)
 endif()
@@ -2,37 +2,32 @@ function(download_ippicv root_var)
  set(${root_var} "" PARENT_SCOPE)

  # Commit SHA in the opencv_3rdparty repo
-  set(IPPICV_COMMIT "32e315a5b106a7b89dbed51c28f8120a48b368b4")
+  set(IPPICV_COMMIT "a56b6ac6f030c312b2dce17430eef13aed9af274")
  # Define actual ICV versions
  if(APPLE)
    set(OPENCV_ICV_PLATFORM "macosx")
    set(OPENCV_ICV_PACKAGE_SUBDIR "ippicv_mac")
-    if(X86_64)
-      set(OPENCV_ICV_NAME "ippicv_2019_mac_intel64_general_20180723.tgz")
-      set(OPENCV_ICV_HASH "fe6b2bb75ae0e3f19ad3ae1a31dfa4a2")
-    else()
-      set(OPENCV_ICV_NAME "ippicv_2019_mac_ia32_general_20180723.tgz")
-      set(OPENCV_ICV_HASH "b5dfa78c87eb75c64470cbe5ec876f4f")
-    endif()
+    set(OPENCV_ICV_NAME "ippicv_2020_mac_intel64_20191018_general.tgz")
+    set(OPENCV_ICV_HASH "1c3d675c2a2395d094d523024896e01b")
  elseif((UNIX AND NOT ANDROID) OR (UNIX AND ANDROID_ABI MATCHES "x86"))
    set(OPENCV_ICV_PLATFORM "linux")
    set(OPENCV_ICV_PACKAGE_SUBDIR "ippicv_lnx")
    if(X86_64)
-      set(OPENCV_ICV_NAME "ippicv_2019_lnx_intel64_general_20180723.tgz")
-      set(OPENCV_ICV_HASH "c0bd78adb4156bbf552c1dfe90599607")
+      set(OPENCV_ICV_NAME "ippicv_2020_lnx_intel64_20191018_general.tgz")
+      set(OPENCV_ICV_HASH "7421de0095c7a39162ae13a6098782f9")
    else()
-      set(OPENCV_ICV_NAME "ippicv_2019_lnx_ia32_general_20180723.tgz")
-      set(OPENCV_ICV_HASH "4f38432c30bfd6423164b7a24bbc98a0")
+      set(OPENCV_ICV_NAME "ippicv_2020_lnx_ia32_20191018_general.tgz")
+      set(OPENCV_ICV_HASH "ad189a940fb60eb71f291321322fe3e8")
    endif()
  elseif(WIN32 AND NOT ARM)
    set(OPENCV_ICV_PLATFORM "windows")
    set(OPENCV_ICV_PACKAGE_SUBDIR "ippicv_win")
    if(X86_64)
-      set(OPENCV_ICV_NAME "ippicv_2019_win_intel64_20180723_general.zip")
-      set(OPENCV_ICV_HASH "1d222685246896fe089f88b8858e4b2f")
+      set(OPENCV_ICV_NAME "ippicv_2020_win_intel64_20191018_general.zip")
+      set(OPENCV_ICV_HASH "879741a7946b814455eee6c6ffde2984")
    else()
-      set(OPENCV_ICV_NAME "ippicv_2019_win_ia32_20180723_general.zip")
-      set(OPENCV_ICV_HASH "0157251a2eb9cd63a3ebc7eed0f3e59e")
+      set(OPENCV_ICV_NAME "ippicv_2020_win_ia32_20191018_general.zip")
+      set(OPENCV_ICV_HASH "cd39bdf0c2e1cac9a61101dad7a2413e")
    endif()
  else()
    return()
@@ -37,7 +37,7 @@ set(ITT_SRCS
    src/ittnotify/jitprofiling.c
 )

-add_library(${ITT_LIBRARY} STATIC ${ITT_SRCS} ${ITT_PUBLIC_HDRS} ${ITT_PRIVATE_HDRS})
+add_library(${ITT_LIBRARY} STATIC ${OPENCV_3RDPARTY_EXCLUDE_FROM_ALL} ${ITT_SRCS} ${ITT_PUBLIC_HDRS} ${ITT_PRIVATE_HDRS})

 if(NOT WIN32)
  if(HAVE_DL_LIBRARY)
@@ -60,7 +60,7 @@ if(ENABLE_SOLUTION_FOLDERS)
 endif()

 if(NOT BUILD_SHARED_LIBS)
-  ocv_install_target(${ITT_LIBRARY} EXPORT OpenCVModules ARCHIVE DESTINATION ${OPENCV_3P_LIB_INSTALL_PATH} COMPONENT dev)
+  ocv_install_target(${ITT_LIBRARY} EXPORT OpenCVModules ARCHIVE DESTINATION ${OPENCV_3P_LIB_INSTALL_PATH} COMPONENT dev OPTIONAL)
 endif()

 ocv_install_3rdparty_licenses(ittnotify src/ittnotify/LICENSE.BSD src/ittnotify/LICENSE.GPL)
@@ -17,7 +17,7 @@ file(GLOB lib_ext_hdrs jasper/*.h)
 #         Define the library target:
 # ----------------------------------------------------------------------------------

-add_library(${JASPER_LIBRARY} STATIC ${lib_srcs} ${lib_hdrs} ${lib_ext_hdrs})
+add_library(${JASPER_LIBRARY} STATIC ${OPENCV_3RDPARTY_EXCLUDE_FROM_ALL} ${lib_srcs} ${lib_hdrs} ${lib_ext_hdrs})

 if(WIN32 AND NOT MINGW)
  add_definitions(-DJAS_WIN_MSVC_BUILD)
@@ -46,7 +46,7 @@ if(ENABLE_SOLUTION_FOLDERS)
 endif()

 if(NOT BUILD_SHARED_LIBS)
-  ocv_install_target(${JASPER_LIBRARY} EXPORT OpenCVModules ARCHIVE DESTINATION ${OPENCV_3P_LIB_INSTALL_PATH} COMPONENT dev)
+  ocv_install_target(${JASPER_LIBRARY} EXPORT OpenCVModules ARCHIVE DESTINATION ${OPENCV_3P_LIB_INSTALL_PATH} COMPONENT dev OPTIONAL)
 endif()

 ocv_install_3rdparty_licenses(jasper LICENSE README copyright)
@@ -17,6 +17,12 @@
 #if !defined(JAS_WIN_MSVC_BUILD)
 /* A configure-based build is being used. */

+#include <stdio.h>
+
+// uClibc-ng workaround: https://github.com/opencv/opencv/pull/15279
+#ifndef L_tmpnam
+#define L_tmpnam 20
+#endif


 /* Extra debugging support */
@@ -3,10 +3,10 @@ project(${JPEG_LIBRARY} C)
 ocv_warnings_disable(CMAKE_C_FLAGS -Wunused-parameter -Wsign-compare -Wshorten-64-to-32 -Wimplicit-fallthrough)

 set(VERSION_MAJOR 2)
-set(VERSION_MINOR 0)
+set(VERSION_MINOR 1)
 set(VERSION_REVISION 2)
 set(VERSION ${VERSION_MAJOR}.${VERSION_MINOR}.${VERSION_REVISION})
-set(LIBJPEG_TURBO_VERSION_NUMBER 2000002)
+set(LIBJPEG_TURBO_VERSION_NUMBER 2001002)

 string(TIMESTAMP BUILD "opencv-${OPENCV_VERSION}-libjpeg-turbo")
 if(CMAKE_BUILD_TYPE STREQUAL "Debug")
@@ -46,7 +46,6 @@ if(UNIX)
  ocv_update(HAVE_UNSIGNED_SHORT 1)
  # undef INCOMPLETE_TYPES_BROKEN
  ocv_update(RIGHT_SHIFT_IS_UNSIGNED 0)
-  ocv_update(__CHAR_UNSIGNED__ 0)
 endif()


@@ -65,6 +64,8 @@ set(JPEG_LIB_VERSION 62)
 # OpenCV
 set(JPEG_LIB_VERSION "${VERSION}-${JPEG_LIB_VERSION}" PARENT_SCOPE)

+set(THREAD_LOCAL "")  # WITH_TURBOJPEG is not used
+
 if(MSVC)
  add_definitions(-W3 -wd4996 -wd4018)
 endif()
@@ -104,7 +105,7 @@ set(JPEG_SOURCES ${JPEG_SOURCES} jsimd_none.c)

 ocv_list_add_prefix(JPEG_SOURCES src/)

-add_library(${JPEG_LIBRARY} STATIC ${JPEG_SOURCES} ${SIMD_OBJS})
+add_library(${JPEG_LIBRARY} STATIC ${OPENCV_3RDPARTY_EXCLUDE_FROM_ALL} ${JPEG_SOURCES} ${SIMD_OBJS})

 set_target_properties(${JPEG_LIBRARY}
  PROPERTIES OUTPUT_NAME ${JPEG_LIBRARY}
@@ -119,7 +120,7 @@ if(ENABLE_SOLUTION_FOLDERS)
 endif()

 if(NOT BUILD_SHARED_LIBS)
-  ocv_install_target(${JPEG_LIBRARY} EXPORT OpenCVModules ARCHIVE DESTINATION ${OPENCV_3P_LIB_INSTALL_PATH} COMPONENT dev)
+  ocv_install_target(${JPEG_LIBRARY} EXPORT OpenCVModules ARCHIVE DESTINATION ${OPENCV_3P_LIB_INSTALL_PATH} COMPONENT dev OPTIONAL)
 endif()

 ocv_install_3rdparty_licenses(libjpeg-turbo README.md LICENSE.md README.ijg)
@@ -91,7 +91,7 @@ best of our understanding.
 The Modified (3-clause) BSD License
 ===================================

-Copyright (C)2009-2019 D. R. Commander.  All Rights Reserved.
+Copyright (C)2009-2021 D. R. Commander.  All Rights Reserved.<br>
 Copyright (C)2015 Viktor Szathmáry.  All Rights Reserved.

 Redistribution and use in source and binary forms, with or without
@@ -128,7 +128,7 @@ 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) 1991-2016, Thomas G. Lane, Guido Vollbeding.
+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
@@ -159,19 +159,6 @@ commercial products, provided that all warranty or liability claims are
 assumed by the product vendor.


-The IJG distribution formerly included code to read and write GIF files.
-To avoid entanglement with the Unisys LZW patent (now expired), GIF reading
-support has been removed altogether, and the GIF writer has been simplified
-to produce "uncompressed GIFs".  This technique does not use the LZW
-algorithm; the resulting GIF files are larger than usual, but are readable
-by all standard GIF decoders.
-
-We are required to state that
-    "The Graphics Interchange Format(c) is the Copyright property of
-    CompuServe Incorporated.  GIF(sm) is a Service Mark property of
-    CompuServe Incorporated."
-
-
 REFERENCES
 ==========

@@ -223,12 +210,12 @@ https://www.iso.org/standard/54989.html and http://www.itu.int/rec/T-REC-T.871.
 A PDF file of the older JFIF 1.02 specification 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
+The TIFF 6.0 file format specification can be obtained from
+http://mirrors.ctan.org/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
@@ -243,14 +230,8 @@ The most recent released version can always be found there in
 directory "files".

 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
+general information about JPEG.  It is available at
+http://www.faqs.org/faqs/jpeg-faq.


 FILE FORMAT COMPATIBILITY
@@ -1,14 +1,14 @@
 Background
 ==========

-libjpeg-turbo is a JPEG image codec that uses SIMD instructions (MMX, SSE2,
-AVX2, NEON, AltiVec) to accelerate baseline JPEG compression and decompression
-on x86, x86-64, ARM, and PowerPC systems, as well as progressive JPEG
-compression on x86 and x86-64 systems.  On such systems, libjpeg-turbo is
-generally 2-6x as fast as libjpeg, all else being equal.  On other types of
-systems, libjpeg-turbo can still outperform libjpeg by a significant amount, by
-virtue of its highly-optimized Huffman coding routines.  In many cases, the
-performance of libjpeg-turbo rivals that of proprietary high-speed JPEG codecs.
+libjpeg-turbo is a JPEG image codec that uses SIMD instructions to accelerate
+baseline JPEG compression and decompression on x86, x86-64, Arm, PowerPC, and
+MIPS systems, as well as progressive JPEG compression on x86, x86-64, and Arm
+systems.  On such systems, libjpeg-turbo is generally 2-6x as fast as libjpeg,
+all else being equal.  On other types of systems, libjpeg-turbo can still
+outperform libjpeg by a significant amount, by virtue of its highly-optimized
+Huffman coding routines.  In many cases, the performance of libjpeg-turbo
+rivals that of proprietary high-speed JPEG codecs.

 libjpeg-turbo implements both the traditional libjpeg API as well as the less
 powerful but more straightforward TurboJPEG API.  libjpeg-turbo also features
@@ -135,25 +135,24 @@ without recompiling.  libjpeg-turbo does not claim to support all of the
 libjpeg v7+ features, nor to produce identical output to libjpeg v7+ in all
 cases (see below.)

-By passing an argument of `--with-jpeg7` or `--with-jpeg8` to `configure`, or
-an argument of `-DWITH_JPEG7=1` or `-DWITH_JPEG8=1` to `cmake`, you can build a
-version of libjpeg-turbo that emulates the libjpeg v7 or v8 ABI, so that
-programs that are built against libjpeg v7 or v8 can be run with libjpeg-turbo.
-The following section describes which libjpeg v7+ features are supported and
-which aren't.
+By passing an argument of `-DWITH_JPEG7=1` or `-DWITH_JPEG8=1` to `cmake`, you
+can build a version of libjpeg-turbo that emulates the libjpeg v7 or v8 ABI, so
+that programs that are built against libjpeg v7 or v8 can be run with
+libjpeg-turbo.  The following section describes which libjpeg v7+ features are
+supported and which aren't.

 ### Support for libjpeg v7 and v8 Features

 #### Fully supported

- **libjpeg: IDCT scaling extensions in decompressor**<br>
+- **libjpeg API: IDCT scaling extensions in decompressor**<br>
  libjpeg-turbo supports IDCT scaling with scaling factors of 1/8, 1/4, 3/8,
  1/2, 5/8, 3/4, 7/8, 9/8, 5/4, 11/8, 3/2, 13/8, 7/4, 15/8, and 2/1 (only 1/4
  and 1/2 are SIMD-accelerated.)

- **libjpeg: Arithmetic coding**
+- **libjpeg API: Arithmetic coding**

- **libjpeg: In-memory source and destination managers**<br>
+- **libjpeg API: In-memory source and destination managers**<br>
  See notes below.

 - **cjpeg: Separate quality settings for luminance and chrominance**<br>
@@ -180,19 +179,19 @@ which aren't.

 NOTE:  As of this writing, extensive research has been conducted into the
 usefulness of DCT scaling as a means of data reduction and SmartScale as a
-means of quality improvement.  The reader is invited to peruse the research at
-<http://www.libjpeg-turbo.org/About/SmartScale> and draw his/her own conclusions,
+means of quality improvement.  Readers are invited to peruse the research at
+<http://www.libjpeg-turbo.org/About/SmartScale> and draw their own conclusions,
 but it is the general belief of our project that these features have not
 demonstrated sufficient usefulness to justify inclusion in libjpeg-turbo.

- **libjpeg: DCT scaling in compressor**<br>
+- **libjpeg API: DCT scaling in compressor**<br>
  `cinfo.scale_num` and `cinfo.scale_denom` are silently ignored.
  There is no technical reason why DCT scaling could not be supported when
  emulating the libjpeg v7+ API/ABI, but without the SmartScale extension (see
  below), only scaling factors of 1/2, 8/15, 4/7, 8/13, 2/3, 8/11, 4/5, and
  8/9 would be available, which is of limited usefulness.

- **libjpeg: SmartScale**<br>
+- **libjpeg API: SmartScale**<br>
  `cinfo.block_size` is silently ignored.
  SmartScale is an extension to the JPEG format that allows for DCT block
  sizes other than 8x8.  Providing support for this new format would be
@@ -205,7 +204,7 @@ demonstrated sufficient usefulness to justify inclusion in libjpeg-turbo.
  interest in providing this feature would be as a means of supporting
  additional DCT scaling factors.

- **libjpeg: Fancy downsampling in compressor**<br>
+- **libjpeg API: Fancy downsampling in compressor**<br>
  `cinfo.do_fancy_downsampling` is silently ignored.
  This requires the DCT scaling feature, which is not supported.

@@ -247,15 +246,14 @@ don't, and it allows those functions to be provided in the "official"
 libjpeg-turbo binaries.

 Those who are concerned about maintaining strict conformance with the libjpeg
-v6b or v7 API can pass an argument of `--without-mem-srcdst` to `configure` or
-an argument of `-DWITH_MEM_SRCDST=0` to `cmake` prior to building
-libjpeg-turbo.  This will restore the pre-1.3 behavior, in which
+v6b or v7 API can pass an argument of `-DWITH_MEM_SRCDST=0` to `cmake` prior to
+building libjpeg-turbo.  This will restore the pre-1.3 behavior, in which
 `jpeg_mem_src()` and `jpeg_mem_dest()` are only included when emulating the
 libjpeg v8 API/ABI.

 On Un*x systems, including the in-memory source/destination managers changes
-the dynamic library version from 62.1.0 to 62.2.0 if using libjpeg v6b API/ABI
-emulation and from 7.1.0 to 7.2.0 if using libjpeg v7 API/ABI emulation.
+the dynamic library version from 62.2.0 to 62.3.0 if using libjpeg v6b API/ABI
+emulation and from 7.2.0 to 7.3.0 if using libjpeg v7 API/ABI emulation.

 Note that, on most Un*x systems, the dynamic linker will not look for a
 function in a library until that function is actually used.  Thus, if a program
@@ -289,12 +287,13 @@ following reasons:
  (and slightly faster) floating point IDCT algorithm introduced in libjpeg
  v8a as opposed to the algorithm used in libjpeg v6b.  It should be noted,
  however, that this algorithm basically brings the accuracy of the floating
-  point IDCT in line with the accuracy of the slow integer IDCT.  The floating
-  point DCT/IDCT algorithms are mainly a legacy feature, and they do not
-  produce significantly more accuracy than the slow integer algorithms (to put
-  numbers on this, the typical difference in PNSR between the two algorithms
-  is less than 0.10 dB, whereas changing the quality level by 1 in the upper
-  range of the quality scale is typically more like a 1.0 dB difference.)
+  point IDCT in line with the accuracy of the accurate integer IDCT.  The
+  floating point DCT/IDCT algorithms are mainly a legacy feature, and they do
+  not produce significantly more accuracy than the accurate integer algorithms
+  (to put numbers on this, the typical difference in PNSR between the two
+  algorithms is less than 0.10 dB, whereas changing the quality level by 1 in
+  the upper range of the quality scale is typically more like a 1.0 dB
+  difference.)

 - If the floating point algorithms in libjpeg-turbo are not implemented using
  SIMD instructions on a particular platform, then the accuracy of the
@@ -331,7 +330,7 @@ in a way that makes the rest of the libjpeg infrastructure happy, so it is
 necessary to use the slow Huffman decoder when decompressing a JPEG image that
 has restart markers.  This can cause the decompression performance to drop by
 as much as 20%, but the performance will still be much greater than that of
-libjpeg.  Many consumer packages, such as PhotoShop, use restart markers when
+libjpeg.  Many consumer packages, such as Photoshop, use restart markers when
 generating JPEG images, so images generated by those programs will experience
 this issue.

@@ -342,5 +341,17 @@ The algorithm used by the SIMD-accelerated quantization function cannot produce
 correct results whenever the fast integer forward DCT is used along with a JPEG
 quality of 98-100.  Thus, libjpeg-turbo must use the non-SIMD quantization
 function in those cases.  This causes performance to drop by as much as 40%.
-It is therefore strongly advised that you use the slow integer forward DCT
+It is therefore strongly advised that you use the accurate integer forward DCT
 whenever encoding images with a JPEG quality of 98 or higher.
+
+
+Memory Debugger Pitfalls
+========================
+
+Valgrind and Memory Sanitizer (MSan) can generate false positives
+(specifically, incorrect reports of uninitialized memory accesses) when used
+with libjpeg-turbo's SIMD extensions.  It is generally recommended that the
+SIMD extensions be disabled, either by passing an argument of `-DWITH_SIMD=0`
+to `cmake` when configuring the build or by setting the environment variable
+`JSIMD_FORCENONE` to `1` at run time, when testing libjpeg-turbo with Valgrind,
+MSan, or other memory debuggers.
@@ -61,11 +61,6 @@
   unsigned. */
 #cmakedefine RIGHT_SHIFT_IS_UNSIGNED 1

-/* Define to 1 if type `char' is unsigned and you are not using gcc.  */
-#ifndef __CHAR_UNSIGNED__
-  #cmakedefine __CHAR_UNSIGNED__ 1
-#endif
-
 /* Define to empty if `const' does not conform to ANSI C. */
 /* #undef const */

@@ -18,7 +18,6 @@
 #define HAVE_UNSIGNED_SHORT
 #undef INCOMPLETE_TYPES_BROKEN
 #undef RIGHT_SHIFT_IS_UNSIGNED
-#undef __CHAR_UNSIGNED__

 /* Define "boolean" as unsigned char, not int, per Windows custom */
 #ifndef __RPCNDR_H__            /* don't conflict if rpcndr.h already read */
@@ -15,6 +15,9 @@
 #endif
 #endif

+/* How to obtain thread-local storage */
+#define THREAD_LOCAL  @THREAD_LOCAL@
+
 /* Define to the full name of this package. */
 #define PACKAGE_NAME  "@CMAKE_PROJECT_NAME@"

@@ -37,3 +40,13 @@
 #define HAVE_BITSCANFORWARD
 #endif
 #endif
+
+#if defined(__has_attribute)
+#if __has_attribute(fallthrough)
+#define FALLTHROUGH  __attribute__((fallthrough));
+#else
+#define FALLTHROUGH
+#endif
+#else
+#define FALLTHROUGH
+#endif
@@ -48,9 +48,9 @@ rgb_ycc_convert_internal(j_compress_ptr cinfo, JSAMPARRAY input_buf,
    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]);
+      r = inptr[RGB_RED];
+      g = inptr[RGB_GREEN];
+      b = 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.
@@ -100,9 +100,9 @@ rgb_gray_convert_internal(j_compress_ptr cinfo, JSAMPARRAY input_buf,
    outptr = output_buf[0][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]);
+      r = inptr[RGB_RED];
+      g = inptr[RGB_GREEN];
+      b = inptr[RGB_BLUE];
      inptr += RGB_PIXELSIZE;
      /* Y */
      outptr[col] = (JSAMPLE)((ctab[r + R_Y_OFF] + ctab[g + G_Y_OFF] +
@@ -135,9 +135,9 @@ rgb_rgb_convert_internal(j_compress_ptr cinfo, JSAMPARRAY input_buf,
    outptr2 = output_buf[2][output_row];
    output_row++;
    for (col = 0; col < num_cols; col++) {
-      outptr0[col] = GETJSAMPLE(inptr[RGB_RED]);
-      outptr1[col] = GETJSAMPLE(inptr[RGB_GREEN]);
-      outptr2[col] = GETJSAMPLE(inptr[RGB_BLUE]);
+      outptr0[col] = inptr[RGB_RED];
+      outptr1[col] = inptr[RGB_GREEN];
+      outptr2[col] = inptr[RGB_BLUE];
      inptr += RGB_PIXELSIZE;
    }
  }
@@ -392,11 +392,11 @@ cmyk_ycck_convert(j_compress_ptr cinfo, JSAMPARRAY input_buf,
    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]);
+      r = MAXJSAMPLE - inptr[0];
+      g = MAXJSAMPLE - inptr[1];
+      b = MAXJSAMPLE - inptr[2];
      /* K passes through as-is */
-      outptr3[col] = inptr[3];  /* don't need GETJSAMPLE here */
+      outptr3[col] = inptr[3];
      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.
@@ -438,7 +438,7 @@ grayscale_convert(j_compress_ptr cinfo, JSAMPARRAY input_buf,
    outptr = output_buf[0][output_row];
    output_row++;
    for (col = 0; col < num_cols; col++) {
-      outptr[col] = inptr[0];   /* don't need GETJSAMPLE() here */
+      outptr[col] = inptr[0];
      inptr += instride;
    }
  }
@@ -497,7 +497,7 @@ null_convert(j_compress_ptr cinfo, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
        inptr = *input_buf;
        outptr = output_buf[ci][output_row];
        for (col = 0; col < num_cols; col++) {
-          outptr[col] = inptr[ci]; /* don't need GETJSAMPLE() here */
+          outptr[col] = inptr[ci];
          inptr += nc;
        }
      }
@@ -381,19 +381,19 @@ convsamp(JSAMPARRAY sample_data, JDIMENSION start_col, DCTELEM *workspace)
    elemptr = sample_data[elemr] + start_col;

 #if DCTSIZE == 8                /* unroll the inner loop */
-    *workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
-    *workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
-    *workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
-    *workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
-    *workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
-    *workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
-    *workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
-    *workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
+    *workspaceptr++ = (*elemptr++) - CENTERJSAMPLE;
+    *workspaceptr++ = (*elemptr++) - CENTERJSAMPLE;
+    *workspaceptr++ = (*elemptr++) - CENTERJSAMPLE;
+    *workspaceptr++ = (*elemptr++) - CENTERJSAMPLE;
+    *workspaceptr++ = (*elemptr++) - CENTERJSAMPLE;
+    *workspaceptr++ = (*elemptr++) - CENTERJSAMPLE;
+    *workspaceptr++ = (*elemptr++) - CENTERJSAMPLE;
+    *workspaceptr++ = (*elemptr++) - CENTERJSAMPLE;
 #else
    {
      register int elemc;
      for (elemc = DCTSIZE; elemc > 0; elemc--)
-        *workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
+        *workspaceptr++ = (*elemptr++) - CENTERJSAMPLE;
    }
 #endif
  }
@@ -533,20 +533,19 @@ convsamp_float(JSAMPARRAY sample_data, JDIMENSION start_col,
  for (elemr = 0; elemr < DCTSIZE; elemr++) {
    elemptr = sample_data[elemr] + start_col;
 #if DCTSIZE == 8                /* unroll the inner loop */
-    *workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
-    *workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
-    *workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
-    *workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
-    *workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
-    *workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
-    *workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
-    *workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
+    *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - CENTERJSAMPLE);
+    *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - CENTERJSAMPLE);
+    *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - CENTERJSAMPLE);
+    *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - CENTERJSAMPLE);
+    *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - CENTERJSAMPLE);
+    *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - CENTERJSAMPLE);
+    *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - CENTERJSAMPLE);
+    *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - CENTERJSAMPLE);
 #else
    {
      register int elemc;
      for (elemc = DCTSIZE; elemc > 0; elemc--)
-        *workspaceptr++ = (FAST_FLOAT)
-                          (GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
+        *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - CENTERJSAMPLE);
    }
 #endif
  }
@@ -4,8 +4,10 @@
 * This file was part of the Independent JPEG Group's software:
 * Copyright (C) 1991-1997, Thomas G. Lane.
 * libjpeg-turbo Modifications:
- * Copyright (C) 2009-2011, 2014-2016, 2018, D. R. Commander.
+ * Copyright (C) 2009-2011, 2014-2016, 2018-2021, D. R. Commander.
 * Copyright (C) 2015, Matthieu Darbois.
+ * Copyright (C) 2018, Matthias Räncker.
+ * Copyright (C) 2020, Arm Limited.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -34,23 +36,28 @@
 * memory footprint by 64k, which is important for some mobile applications
 * that create many isolated instances of libjpeg-turbo (web browsers, for
 * instance.)  This may improve performance on some mobile platforms as well.
- * This feature is enabled by default only on ARM processors, because some x86
+ * This feature is enabled by default only on Arm processors, because some x86
 * chips have a slow implementation of bsr, and the use of clz/bsr cannot be
 * shown to have a significant performance impact even on the x86 chips that
- * have a fast implementation of it.  When building for ARMv6, you can
+ * have a fast implementation of it.  When building for Armv6, you can
 * explicitly disable the use of clz/bsr by adding -mthumb to the compiler
 * flags (this defines __thumb__).
 */

 /* NOTE: Both GCC and Clang define __GNUC__ */
-#if defined __GNUC__ && (defined __arm__ || defined __aarch64__)
-#if !defined __thumb__ || defined __thumb2__
+#if (defined(__GNUC__) && (defined(__arm__) || defined(__aarch64__))) || \
+    defined(_M_ARM) || defined(_M_ARM64)
+#if !defined(__thumb__) || defined(__thumb2__)
 #define USE_CLZ_INTRINSIC
 #endif
 #endif

 #ifdef USE_CLZ_INTRINSIC
+#if defined(_MSC_VER) && !defined(__clang__)
+#define JPEG_NBITS_NONZERO(x)  (32 - _CountLeadingZeros(x))
+#else
 #define JPEG_NBITS_NONZERO(x)  (32 - __builtin_clz(x))
+#endif
 #define JPEG_NBITS(x)          (x ? JPEG_NBITS_NONZERO(x) : 0)
 #else
 #include "jpeg_nbits_table.h"
@@ -65,32 +72,43 @@
 * but must not be updated permanently until we complete the MCU.
 */

+#if defined(__x86_64__) && defined(__ILP32__)
+typedef unsigned long long bit_buf_type;
+#else
+typedef size_t bit_buf_type;
+#endif
+
+/* NOTE: The more optimal Huffman encoding algorithm is only used by the
+ * intrinsics implementation of the Arm Neon SIMD extensions, which is why we
+ * retain the old Huffman encoder behavior when using the GAS implementation.
+ */
+#if defined(WITH_SIMD) && !(defined(__arm__) || defined(__aarch64__) || \
+                            defined(_M_ARM) || defined(_M_ARM64))
+typedef unsigned long long simd_bit_buf_type;
+#else
+typedef bit_buf_type simd_bit_buf_type;
+#endif
+
+#if (defined(SIZEOF_SIZE_T) && SIZEOF_SIZE_T == 8) || defined(_WIN64) || \
+    (defined(__x86_64__) && defined(__ILP32__))
+#define BIT_BUF_SIZE  64
+#elif (defined(SIZEOF_SIZE_T) && SIZEOF_SIZE_T == 4) || defined(_WIN32)
+#define BIT_BUF_SIZE  32
+#else
+#error Cannot determine word size
+#endif
+#define SIMD_BIT_BUF_SIZE  (sizeof(simd_bit_buf_type) * 8)
+
 typedef struct {
-  size_t put_buffer;                    /* current bit-accumulation buffer */
-  int put_bits;                         /* # of bits now in it */
+  union {
+    bit_buf_type c;
+    simd_bit_buf_type simd;
+  } put_buffer;                         /* current bit accumulation buffer */
+  int free_bits;                        /* # of bits available in it */
+                                        /* (Neon GAS: # of bits now in it) */
  int last_dc_val[MAX_COMPS_IN_SCAN];   /* last DC coef for each component */
 } savable_state;

-/* This macro is to work around compilers with missing or broken
- * structure assignment.  You'll need to fix this code if you have
- * such a compiler and you change MAX_COMPS_IN_SCAN.
- */
-
-#ifndef NO_STRUCT_ASSIGN
-#define ASSIGN_STATE(dest, src)  ((dest) = (src))
-#else
-#if MAX_COMPS_IN_SCAN == 4
-#define ASSIGN_STATE(dest, src) \
-  ((dest).put_buffer = (src).put_buffer, \
-   (dest).put_bits = (src).put_bits, \
-   (dest).last_dc_val[0] = (src).last_dc_val[0], \
-   (dest).last_dc_val[1] = (src).last_dc_val[1], \
-   (dest).last_dc_val[2] = (src).last_dc_val[2], \
-   (dest).last_dc_val[3] = (src).last_dc_val[3])
-#endif
-#endif
-
-
 typedef struct {
  struct jpeg_entropy_encoder pub; /* public fields */

@@ -123,6 +141,7 @@ typedef struct {
  size_t free_in_buffer;        /* # of byte spaces remaining in buffer */
  savable_state cur;            /* Current bit buffer & DC state */
  j_compress_ptr cinfo;         /* dump_buffer needs access to this */
+  int simd;
 } working_state;


@@ -201,8 +220,17 @@ start_pass_huff(j_compress_ptr cinfo, boolean gather_statistics)
  }

  /* Initialize bit buffer to empty */
-  entropy->saved.put_buffer = 0;
-  entropy->saved.put_bits = 0;
+  if (entropy->simd) {
+    entropy->saved.put_buffer.simd = 0;
+#if defined(__aarch64__) && !defined(NEON_INTRINSICS)
+    entropy->saved.free_bits = 0;
+#else
+    entropy->saved.free_bits = SIMD_BIT_BUF_SIZE;
+#endif
+  } else {
+    entropy->saved.put_buffer.c = 0;
+    entropy->saved.free_bits = BIT_BUF_SIZE;
+  }

  /* Initialize restart stuff */
  entropy->restarts_to_go = cinfo->restart_interval;
@@ -287,6 +315,7 @@ jpeg_make_c_derived_tbl(j_compress_ptr cinfo, boolean isDC, int tblno,
   * this lets us detect duplicate VAL entries here, and later
   * allows emit_bits to detect any attempt to emit such symbols.
   */
+  MEMZERO(dtbl->ehufco, sizeof(dtbl->ehufco));
  MEMZERO(dtbl->ehufsi, sizeof(dtbl->ehufsi));

  /* This is also a convenient place to check for out-of-range
@@ -334,90 +363,94 @@ dump_buffer(working_state *state)

 /* Outputting bits to the file */

-/* These macros perform the same task as the emit_bits() function in the
- * original libjpeg code.  In addition to reducing overhead by explicitly
- * inlining the code, additional performance is achieved by taking into
- * account the size of the bit buffer and waiting until it is almost full
- * before emptying it.  This mostly benefits 64-bit platforms, since 6
- * bytes can be stored in a 64-bit bit buffer before it has to be emptied.
+/* Output byte b and, speculatively, an additional 0 byte.  0xFF must be
+ * encoded as 0xFF 0x00, so the output buffer pointer is advanced by 2 if the
+ * byte is 0xFF.  Otherwise, the output buffer pointer is advanced by 1, and
+ * the speculative 0 byte will be overwritten by the next byte.
 */
-
-#define EMIT_BYTE() { \
-  JOCTET c; \
-  put_bits -= 8; \
-  c = (JOCTET)GETJOCTET(put_buffer >> put_bits); \
-  *buffer++ = c; \
-  if (c == 0xFF)  /* need to stuff a zero byte? */ \
-    *buffer++ = 0; \
+#define EMIT_BYTE(b) { \
+  buffer[0] = (JOCTET)(b); \
+  buffer[1] = 0; \
+  buffer -= -2 + ((JOCTET)(b) < 0xFF); \
 }

-#define PUT_BITS(code, size) { \
-  put_bits += size; \
-  put_buffer = (put_buffer << size) | code; \
-}
+/* Output the entire bit buffer.  If there are no 0xFF bytes in it, then write
+ * directly to the output buffer.  Otherwise, use the EMIT_BYTE() macro to
+ * encode 0xFF as 0xFF 0x00.
+ */
+#if BIT_BUF_SIZE == 64

-#define CHECKBUF15() { \
-  if (put_bits > 15) { \
-    EMIT_BYTE() \
-    EMIT_BYTE() \
+#define FLUSH() { \
+  if (put_buffer & 0x8080808080808080 & ~(put_buffer + 0x0101010101010101)) { \
+    EMIT_BYTE(put_buffer >> 56) \
+    EMIT_BYTE(put_buffer >> 48) \
+    EMIT_BYTE(put_buffer >> 40) \
+    EMIT_BYTE(put_buffer >> 32) \
+    EMIT_BYTE(put_buffer >> 24) \
+    EMIT_BYTE(put_buffer >> 16) \
+    EMIT_BYTE(put_buffer >>  8) \
+    EMIT_BYTE(put_buffer      ) \
+  } else { \
+    buffer[0] = (JOCTET)(put_buffer >> 56); \
+    buffer[1] = (JOCTET)(put_buffer >> 48); \
+    buffer[2] = (JOCTET)(put_buffer >> 40); \
+    buffer[3] = (JOCTET)(put_buffer >> 32); \
+    buffer[4] = (JOCTET)(put_buffer >> 24); \
+    buffer[5] = (JOCTET)(put_buffer >> 16); \
+    buffer[6] = (JOCTET)(put_buffer >> 8); \
+    buffer[7] = (JOCTET)(put_buffer); \
+    buffer += 8; \
  } \
 }

-#define CHECKBUF31() { \
-  if (put_bits > 31) { \
-    EMIT_BYTE() \
-    EMIT_BYTE() \
-    EMIT_BYTE() \
-    EMIT_BYTE() \
-  } \
-}
-
-#define CHECKBUF47() { \
-  if (put_bits > 47) { \
-    EMIT_BYTE() \
-    EMIT_BYTE() \
-    EMIT_BYTE() \
-    EMIT_BYTE() \
-    EMIT_BYTE() \
-    EMIT_BYTE() \
-  } \
-}
-
-#if !defined(_WIN32) && !defined(SIZEOF_SIZE_T)
-#error Cannot determine word size
-#endif
-
-#if SIZEOF_SIZE_T == 8 || defined(_WIN64)
-
-#define EMIT_BITS(code, size) { \
-  CHECKBUF47() \
-  PUT_BITS(code, size) \
-}
-
-#define EMIT_CODE(code, size) { \
-  temp2 &= (((JLONG)1) << nbits) - 1; \
-  CHECKBUF31() \
-  PUT_BITS(code, size) \
-  PUT_BITS(temp2, nbits) \
-}
-
 #else

-#define EMIT_BITS(code, size) { \
-  PUT_BITS(code, size) \
-  CHECKBUF15() \
-}
-
-#define EMIT_CODE(code, size) { \
-  temp2 &= (((JLONG)1) << nbits) - 1; \
-  PUT_BITS(code, size) \
-  CHECKBUF15() \
-  PUT_BITS(temp2, nbits) \
-  CHECKBUF15() \
+#define FLUSH() { \
+  if (put_buffer & 0x80808080 & ~(put_buffer + 0x01010101)) { \
+    EMIT_BYTE(put_buffer >> 24) \
+    EMIT_BYTE(put_buffer >> 16) \
+    EMIT_BYTE(put_buffer >>  8) \
+    EMIT_BYTE(put_buffer      ) \
+  } else { \
+    buffer[0] = (JOCTET)(put_buffer >> 24); \
+    buffer[1] = (JOCTET)(put_buffer >> 16); \
+    buffer[2] = (JOCTET)(put_buffer >> 8); \
+    buffer[3] = (JOCTET)(put_buffer); \
+    buffer += 4; \
+  } \
 }

 #endif

+/* Fill the bit buffer to capacity with the leading bits from code, then output
+ * the bit buffer and put the remaining bits from code into the bit buffer.
+ */
+#define PUT_AND_FLUSH(code, size) { \
+  put_buffer = (put_buffer << (size + free_bits)) | (code >> -free_bits); \
+  FLUSH() \
+  free_bits += BIT_BUF_SIZE; \
+  put_buffer = code; \
+}
+
+/* Insert code into the bit buffer and output the bit buffer if needed.
+ * NOTE: We can't flush with free_bits == 0, since the left shift in
+ * PUT_AND_FLUSH() would have undefined behavior.
+ */
+#define PUT_BITS(code, size) { \
+  free_bits -= size; \
+  if (free_bits < 0) \
+    PUT_AND_FLUSH(code, size) \
+  else \
+    put_buffer = (put_buffer << size) | code; \
+}
+
+#define PUT_CODE(code, size) { \
+  temp &= (((JLONG)1) << nbits) - 1; \
+  temp |= code << nbits; \
+  nbits += size; \
+  PUT_BITS(temp, nbits) \
+}
+

 /* Although it is exceedingly rare, it is possible for a Huffman-encoded
 * coefficient block to be larger than the 128-byte unencoded block.  For each
@@ -428,7 +461,7 @@ dump_buffer(working_state *state)
 * scanning order-- 1, 8, 16, etc.), then this will produce an encoded block
 * larger than 200 bytes.
 */
-#define BUFSIZE  (DCTSIZE2 * 4)
+#define BUFSIZE  (DCTSIZE2 * 8)

 #define LOAD_BUFFER() { \
  if (state->free_in_buffer < BUFSIZE) { \
@@ -440,6 +473,7 @@ dump_buffer(working_state *state)

 #define STORE_BUFFER() { \
  if (localbuf) { \
+    size_t bytes, bytestocopy; \
    bytes = buffer - _buffer; \
    buffer = _buffer; \
    while (bytes > 0) { \
@@ -462,20 +496,46 @@ dump_buffer(working_state *state)
 LOCAL(boolean)
 flush_bits(working_state *state)
 {
-  JOCTET _buffer[BUFSIZE], *buffer;
-  size_t put_buffer;  int put_bits;
-  size_t bytes, bytestocopy;  int localbuf = 0;
+  JOCTET _buffer[BUFSIZE], *buffer, temp;
+  simd_bit_buf_type put_buffer;  int put_bits;
+  int localbuf = 0;
+
+  if (state->simd) {
+#if defined(__aarch64__) && !defined(NEON_INTRINSICS)
+    put_bits = state->cur.free_bits;
+#else
+    put_bits = SIMD_BIT_BUF_SIZE - state->cur.free_bits;
+#endif
+    put_buffer = state->cur.put_buffer.simd;
+  } else {
+    put_bits = BIT_BUF_SIZE - state->cur.free_bits;
+    put_buffer = state->cur.put_buffer.c;
+  }

-  put_buffer = state->cur.put_buffer;
-  put_bits = state->cur.put_bits;
  LOAD_BUFFER()

-  /* fill any partial byte with ones */
-  PUT_BITS(0x7F, 7)
-  while (put_bits >= 8) EMIT_BYTE()
+  while (put_bits >= 8) {
+    put_bits -= 8;
+    temp = (JOCTET)(put_buffer >> put_bits);
+    EMIT_BYTE(temp)
+  }
+  if (put_bits) {
+    /* fill partial byte with ones */
+    temp = (JOCTET)((put_buffer << (8 - put_bits)) | (0xFF >> put_bits));
+    EMIT_BYTE(temp)
+  }

-  state->cur.put_buffer = 0;    /* and reset bit-buffer to empty */
-  state->cur.put_bits = 0;
+  if (state->simd) {                    /* and reset bit buffer to empty */
+    state->cur.put_buffer.simd = 0;
+#if defined(__aarch64__) && !defined(NEON_INTRINSICS)
+    state->cur.free_bits = 0;
+#else
+    state->cur.free_bits = SIMD_BIT_BUF_SIZE;
+#endif
+  } else {
+    state->cur.put_buffer.c = 0;
+    state->cur.free_bits = BIT_BUF_SIZE;
+  }
  STORE_BUFFER()

  return TRUE;
@@ -489,7 +549,7 @@ encode_one_block_simd(working_state *state, JCOEFPTR block, int last_dc_val,
                      c_derived_tbl *dctbl, c_derived_tbl *actbl)
 {
  JOCTET _buffer[BUFSIZE], *buffer;
-  size_t bytes, bytestocopy;  int localbuf = 0;
+  int localbuf = 0;

  LOAD_BUFFER()

@@ -505,53 +565,41 @@ LOCAL(boolean)
 encode_one_block(working_state *state, JCOEFPTR block, int last_dc_val,
                 c_derived_tbl *dctbl, c_derived_tbl *actbl)
 {
-  int temp, temp2, temp3;
-  int nbits;
-  int r, code, size;
+  int temp, nbits, free_bits;
+  bit_buf_type put_buffer;
  JOCTET _buffer[BUFSIZE], *buffer;
-  size_t put_buffer;  int put_bits;
-  int code_0xf0 = actbl->ehufco[0xf0], size_0xf0 = actbl->ehufsi[0xf0];
-  size_t bytes, bytestocopy;  int localbuf = 0;
+  int localbuf = 0;

-  put_buffer = state->cur.put_buffer;
-  put_bits = state->cur.put_bits;
+  free_bits = state->cur.free_bits;
+  put_buffer = state->cur.put_buffer.c;
  LOAD_BUFFER()

  /* Encode the DC coefficient difference per section F.1.2.1 */

-  temp = temp2 = block[0] - last_dc_val;
+  temp = block[0] - last_dc_val;

  /* This is a well-known technique for obtaining the absolute value without a
   * branch.  It is derived from an assembly language technique presented in
   * "How to Optimize for the Pentium Processors", Copyright (c) 1996, 1997 by
-   * Agner Fog.
+   * Agner Fog.  This code assumes we are on a two's complement machine.
   */
-  temp3 = temp >> (CHAR_BIT * sizeof(int) - 1);
-  temp ^= temp3;
-  temp -= temp3;
-
-  /* For a negative input, want temp2 = bitwise complement of abs(input) */
-  /* This code assumes we are on a two's complement machine */
-  temp2 += temp3;
+  nbits = temp >> (CHAR_BIT * sizeof(int) - 1);
+  temp += nbits;
+  nbits ^= temp;

  /* Find the number of bits needed for the magnitude of the coefficient */
-  nbits = JPEG_NBITS(temp);
+  nbits = JPEG_NBITS(nbits);

-  /* Emit the Huffman-coded symbol for the number of bits */
-  code = dctbl->ehufco[nbits];
-  size = dctbl->ehufsi[nbits];
-  EMIT_BITS(code, size)
-
-  /* Mask off any extra bits in code */
-  temp2 &= (((JLONG)1) << nbits) - 1;
-
-  /* Emit that number of bits of the value, if positive, */
-  /* or the complement of its magnitude, if negative. */
-  EMIT_BITS(temp2, nbits)
+  /* Emit the Huffman-coded symbol for the number of bits.
+   * Emit that number of bits of the value, if positive,
+   * or the complement of its magnitude, if negative.
+   */
+  PUT_CODE(dctbl->ehufco[nbits], dctbl->ehufsi[nbits])

  /* Encode the AC coefficients per section F.1.2.2 */

-  r = 0;                        /* r = run length of zeros */
+  {
+    int r = 0;                  /* r = run length of zeros */

 /* Manually unroll the k loop to eliminate the counter variable.  This
 * improves performance greatly on systems with a limited number of
@@ -559,51 +607,46 @@ encode_one_block(working_state *state, JCOEFPTR block, int last_dc_val,
 */
 #define kloop(jpeg_natural_order_of_k) { \
  if ((temp = block[jpeg_natural_order_of_k]) == 0) { \
-    r++; \
+    r += 16; \
  } else { \
-    temp2 = temp; \
    /* Branch-less absolute value, bitwise complement, etc., same as above */ \
-    temp3 = temp >> (CHAR_BIT * sizeof(int) - 1); \
-    temp ^= temp3; \
-    temp -= temp3; \
-    temp2 += temp3; \
-    nbits = JPEG_NBITS_NONZERO(temp); \
+    nbits = temp >> (CHAR_BIT * sizeof(int) - 1); \
+    temp += nbits; \
+    nbits ^= temp; \
+    nbits = JPEG_NBITS_NONZERO(nbits); \
    /* if run length > 15, must emit special run-length-16 codes (0xF0) */ \
-    while (r > 15) { \
-      EMIT_BITS(code_0xf0, size_0xf0) \
-      r -= 16; \
+    while (r >= 16 * 16) { \
+      r -= 16 * 16; \
+      PUT_BITS(actbl->ehufco[0xf0], actbl->ehufsi[0xf0]) \
    } \
    /* Emit Huffman symbol for run length / number of bits */ \
-    temp3 = (r << 4) + nbits; \
-    code = actbl->ehufco[temp3]; \
-    size = actbl->ehufsi[temp3]; \
-    EMIT_CODE(code, size) \
+    r += nbits; \
+    PUT_CODE(actbl->ehufco[r], actbl->ehufsi[r]) \
    r = 0; \
  } \
 }

-  /* One iteration for each value in jpeg_natural_order[] */
-  kloop(1);   kloop(8);   kloop(16);  kloop(9);   kloop(2);   kloop(3);
-  kloop(10);  kloop(17);  kloop(24);  kloop(32);  kloop(25);  kloop(18);
-  kloop(11);  kloop(4);   kloop(5);   kloop(12);  kloop(19);  kloop(26);
-  kloop(33);  kloop(40);  kloop(48);  kloop(41);  kloop(34);  kloop(27);
-  kloop(20);  kloop(13);  kloop(6);   kloop(7);   kloop(14);  kloop(21);
-  kloop(28);  kloop(35);  kloop(42);  kloop(49);  kloop(56);  kloop(57);
-  kloop(50);  kloop(43);  kloop(36);  kloop(29);  kloop(22);  kloop(15);
-  kloop(23);  kloop(30);  kloop(37);  kloop(44);  kloop(51);  kloop(58);
-  kloop(59);  kloop(52);  kloop(45);  kloop(38);  kloop(31);  kloop(39);
-  kloop(46);  kloop(53);  kloop(60);  kloop(61);  kloop(54);  kloop(47);
-  kloop(55);  kloop(62);  kloop(63);
+    /* One iteration for each value in jpeg_natural_order[] */
+    kloop(1);   kloop(8);   kloop(16);  kloop(9);   kloop(2);   kloop(3);
+    kloop(10);  kloop(17);  kloop(24);  kloop(32);  kloop(25);  kloop(18);
+    kloop(11);  kloop(4);   kloop(5);   kloop(12);  kloop(19);  kloop(26);
+    kloop(33);  kloop(40);  kloop(48);  kloop(41);  kloop(34);  kloop(27);
+    kloop(20);  kloop(13);  kloop(6);   kloop(7);   kloop(14);  kloop(21);
+    kloop(28);  kloop(35);  kloop(42);  kloop(49);  kloop(56);  kloop(57);
+    kloop(50);  kloop(43);  kloop(36);  kloop(29);  kloop(22);  kloop(15);
+    kloop(23);  kloop(30);  kloop(37);  kloop(44);  kloop(51);  kloop(58);
+    kloop(59);  kloop(52);  kloop(45);  kloop(38);  kloop(31);  kloop(39);
+    kloop(46);  kloop(53);  kloop(60);  kloop(61);  kloop(54);  kloop(47);
+    kloop(55);  kloop(62);  kloop(63);

-  /* If the last coef(s) were zero, emit an end-of-block code */
-  if (r > 0) {
-    code = actbl->ehufco[0];
-    size = actbl->ehufsi[0];
-    EMIT_BITS(code, size)
+    /* If the last coef(s) were zero, emit an end-of-block code */
+    if (r > 0) {
+      PUT_BITS(actbl->ehufco[0], actbl->ehufsi[0])
+    }
  }

-  state->cur.put_buffer = put_buffer;
-  state->cur.put_bits = put_bits;
+  state->cur.put_buffer.c = put_buffer;
+  state->cur.free_bits = free_bits;
  STORE_BUFFER()

  return TRUE;
@@ -650,8 +693,9 @@ encode_mcu_huff(j_compress_ptr cinfo, JBLOCKROW *MCU_data)
  /* Load up working state */
  state.next_output_byte = cinfo->dest->next_output_byte;
  state.free_in_buffer = cinfo->dest->free_in_buffer;
-  ASSIGN_STATE(state.cur, entropy->saved);
+  state.cur = entropy->saved;
  state.cinfo = cinfo;
+  state.simd = entropy->simd;

  /* Emit restart marker if needed */
  if (cinfo->restart_interval) {
@@ -690,7 +734,7 @@ encode_mcu_huff(j_compress_ptr cinfo, JBLOCKROW *MCU_data)
  /* Completed MCU, so update state */
  cinfo->dest->next_output_byte = state.next_output_byte;
  cinfo->dest->free_in_buffer = state.free_in_buffer;
-  ASSIGN_STATE(entropy->saved, state.cur);
+  entropy->saved = state.cur;

  /* Update restart-interval state too */
  if (cinfo->restart_interval) {
@@ -719,8 +763,9 @@ finish_pass_huff(j_compress_ptr cinfo)
  /* Load up working state ... flush_bits needs it */
  state.next_output_byte = cinfo->dest->next_output_byte;
  state.free_in_buffer = cinfo->dest->free_in_buffer;
-  ASSIGN_STATE(state.cur, entropy->saved);
+  state.cur = entropy->saved;
  state.cinfo = cinfo;
+  state.simd = entropy->simd;

  /* Flush out the last data */
  if (!flush_bits(&state))
@@ -729,7 +774,7 @@ finish_pass_huff(j_compress_ptr cinfo)
  /* Update state */
  cinfo->dest->next_output_byte = state.next_output_byte;
  cinfo->dest->free_in_buffer = state.free_in_buffer;
-  ASSIGN_STATE(entropy->saved, state.cur);
+  entropy->saved = state.cur;
 }


@@ -1,8 +1,10 @@
 /*
 * jcinit.c
 *
+ * This file was part of the Independent JPEG Group's software:
 * Copyright (C) 1991-1997, Thomas G. Lane.
- * This file is part of the Independent JPEG Group's software.
+ * libjpeg-turbo Modifications:
+ * Copyright (C) 2020, D. R. Commander.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -19,6 +21,7 @@
 #define JPEG_INTERNALS
 #include "jinclude.h"
 #include "jpeglib.h"
+#include "jpegcomp.h"


 /*
@@ -492,8 +492,8 @@ prepare_for_pass(j_compress_ptr cinfo)
     */
    master->pass_type = output_pass;
    master->pass_number++;
-    /*FALLTHROUGH*/
 #endif
+    FALLTHROUGH                 /*FALLTHROUGH*/
  case output_pass:
    /* Do a data-output pass. */
    /* We need not repeat per-scan setup if prior optimization pass did it. */
@@ -4,8 +4,10 @@
 * This file was part of the Independent JPEG Group's software:
 * Copyright (C) 1995-1997, Thomas G. Lane.
 * libjpeg-turbo Modifications:
- * Copyright (C) 2011, 2015, 2018, D. R. Commander.
+ * Copyright (C) 2011, 2015, 2018, 2021, D. R. Commander.
 * Copyright (C) 2016, 2018, Matthieu Darbois.
+ * Copyright (C) 2020, Arm Limited.
+ * Copyright (C) 2021, Alex Richardson.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -43,23 +45,28 @@
 * memory footprint by 64k, which is important for some mobile applications
 * that create many isolated instances of libjpeg-turbo (web browsers, for
 * instance.)  This may improve performance on some mobile platforms as well.
- * This feature is enabled by default only on ARM processors, because some x86
+ * This feature is enabled by default only on Arm processors, because some x86
 * chips have a slow implementation of bsr, and the use of clz/bsr cannot be
 * shown to have a significant performance impact even on the x86 chips that
- * have a fast implementation of it.  When building for ARMv6, you can
+ * have a fast implementation of it.  When building for Armv6, you can
 * explicitly disable the use of clz/bsr by adding -mthumb to the compiler
 * flags (this defines __thumb__).
 */

 /* NOTE: Both GCC and Clang define __GNUC__ */
-#if defined __GNUC__ && (defined __arm__ || defined __aarch64__)
-#if !defined __thumb__ || defined __thumb2__
+#if (defined(__GNUC__) && (defined(__arm__) || defined(__aarch64__))) || \
+    defined(_M_ARM) || defined(_M_ARM64)
+#if !defined(__thumb__) || defined(__thumb2__)
 #define USE_CLZ_INTRINSIC
 #endif
 #endif

 #ifdef USE_CLZ_INTRINSIC
+#if defined(_MSC_VER) && !defined(__clang__)
+#define JPEG_NBITS_NONZERO(x)  (32 - _CountLeadingZeros(x))
+#else
 #define JPEG_NBITS_NONZERO(x)  (32 - __builtin_clz(x))
+#endif
 #define JPEG_NBITS(x)          (x ? JPEG_NBITS_NONZERO(x) : 0)
 #else
 #include "jpeg_nbits_table.h"
@@ -169,24 +176,26 @@ INLINE
 METHODDEF(int)
 count_zeroes(size_t *x)
 {
-  int result;
 #if defined(HAVE_BUILTIN_CTZL)
+  int result;
  result = __builtin_ctzl(*x);
  *x >>= result;
 #elif defined(HAVE_BITSCANFORWARD64)
+  unsigned long result;
  _BitScanForward64(&result, *x);
  *x >>= result;
 #elif defined(HAVE_BITSCANFORWARD)
+  unsigned long result;
  _BitScanForward(&result, *x);
  *x >>= result;
 #else
-  result = 0;
+  int result = 0;
  while ((*x & 1) == 0) {
    ++result;
    *x >>= 1;
  }
 #endif
-  return result;
+  return (int)result;
 }


@@ -672,7 +681,7 @@ encode_mcu_AC_first(j_compress_ptr cinfo, JBLOCKROW *MCU_data)
      emit_restart(entropy, entropy->next_restart_num);

 #ifdef WITH_SIMD
-  cvalue = values = (JCOEF *)PAD((size_t)values_unaligned, 16);
+  cvalue = values = (JCOEF *)PAD((JUINTPTR)values_unaligned, 16);
 #else
  /* Not using SIMD, so alignment is not needed */
  cvalue = values = values_unaligned;
@@ -860,7 +869,7 @@ encode_mcu_AC_refine_prepare(const JCOEF *block,

 #define ENCODE_COEFS_AC_REFINE(label) { \
  while (zerobits) { \
-    int idx = count_zeroes(&zerobits); \
+    idx = count_zeroes(&zerobits); \
    r += idx; \
    cabsvalue += idx; \
    signbits >>= idx; \
@@ -917,7 +926,7 @@ METHODDEF(boolean)
 encode_mcu_AC_refine(j_compress_ptr cinfo, JBLOCKROW *MCU_data)
 {
  phuff_entropy_ptr entropy = (phuff_entropy_ptr)cinfo->entropy;
-  register int temp, r;
+  register int temp, r, idx;
  char *BR_buffer;
  unsigned int BR;
  int Sl = cinfo->Se - cinfo->Ss + 1;
@@ -937,7 +946,7 @@ encode_mcu_AC_refine(j_compress_ptr cinfo, JBLOCKROW *MCU_data)
      emit_restart(entropy, entropy->next_restart_num);

 #ifdef WITH_SIMD
-  cabsvalue = absvalues = (JCOEF *)PAD((size_t)absvalues_unaligned, 16);
+  cabsvalue = absvalues = (JCOEF *)PAD((JUINTPTR)absvalues_unaligned, 16);
 #else
  /* Not using SIMD, so alignment is not needed */
  cabsvalue = absvalues = absvalues_unaligned;
@@ -968,7 +977,7 @@ encode_mcu_AC_refine(j_compress_ptr cinfo, JBLOCKROW *MCU_data)

  if (zerobits) {
    int diff = ((absvalues + DCTSIZE2 / 2) - cabsvalue);
-    int idx = count_zeroes(&zerobits);
+    idx = count_zeroes(&zerobits);
    signbits >>= idx;
    idx += diff;
    r += idx;
@@ -6,7 +6,7 @@
 * libjpeg-turbo Modifications:
 * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
 * Copyright (C) 2014, MIPS Technologies, Inc., California.
- * Copyright (C) 2015, D. R. Commander.
+ * Copyright (C) 2015, 2019, D. R. Commander.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -103,7 +103,7 @@ expand_right_edge(JSAMPARRAY image_data, int num_rows, JDIMENSION 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 */
+      pixval = ptr[-1];
      for (count = numcols; count > 0; count--)
        *ptr++ = pixval;
    }
@@ -174,7 +174,7 @@ int_downsample(j_compress_ptr cinfo, jpeg_component_info *compptr,
      for (v = 0; v < v_expand; v++) {
        inptr = input_data[inrow + v] + outcol_h;
        for (h = 0; h < h_expand; h++) {
-          outvalue += (JLONG)GETJSAMPLE(*inptr++);
+          outvalue += (JLONG)(*inptr++);
        }
      }
      *outptr++ = (JSAMPLE)((outvalue + numpix2) / numpix);
@@ -237,8 +237,7 @@ h2v1_downsample(j_compress_ptr cinfo, jpeg_component_info *compptr,
    inptr = input_data[outrow];
    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);
+      *outptr++ = (JSAMPLE)((inptr[0] + inptr[1] + bias) >> 1);
      bias ^= 1;                /* 0=>1, 1=>0 */
      inptr += 2;
    }
@@ -277,8 +276,7 @@ h2v2_downsample(j_compress_ptr cinfo, jpeg_component_info *compptr,
    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);
+        (JSAMPLE)((inptr0[0] + inptr0[1] + inptr1[0] + inptr1[1] + bias) >> 2);
      bias ^= 3;                /* 1=>2, 2=>1 */
      inptr0 += 2;  inptr1 += 2;
    }
@@ -337,33 +335,25 @@ h2v2_smooth_downsample(j_compress_ptr cinfo, jpeg_component_info *compptr,
    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]);
+    membersum = inptr0[0] + inptr0[1] + inptr1[0] + inptr1[1];
+    neighsum = above_ptr[0] + above_ptr[1] + below_ptr[0] + below_ptr[1] +
+               inptr0[0] + inptr0[2] + inptr1[0] + inptr1[2];
    neighsum += neighsum;
-    neighsum += GETJSAMPLE(*above_ptr) + GETJSAMPLE(above_ptr[2]) +
-                GETJSAMPLE(*below_ptr) + GETJSAMPLE(below_ptr[2]);
+    neighsum += above_ptr[0] + above_ptr[2] + below_ptr[0] + 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]);
+      membersum = inptr0[0] + inptr0[1] + inptr1[0] + 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]);
+      neighsum = above_ptr[0] + above_ptr[1] + below_ptr[0] + below_ptr[1] +
+                 inptr0[-1] + inptr0[2] + inptr1[-1] + 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]);
+      neighsum += above_ptr[-1] + above_ptr[2] + below_ptr[-1] + below_ptr[2];
      /* form final output scaled up by 2^16 */
      membersum = membersum * memberscale + neighsum * neighscale;
      /* round, descale and output it */
@@ -372,15 +362,11 @@ h2v2_smooth_downsample(j_compress_ptr cinfo, jpeg_component_info *compptr,
    }

    /* 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]);
+    membersum = inptr0[0] + inptr0[1] + inptr1[0] + inptr1[1];
+    neighsum = above_ptr[0] + above_ptr[1] + below_ptr[0] + below_ptr[1] +
+               inptr0[-1] + inptr0[1] + inptr1[-1] + inptr1[1];
    neighsum += neighsum;
-    neighsum += GETJSAMPLE(above_ptr[-1]) + GETJSAMPLE(above_ptr[1]) +
-                GETJSAMPLE(below_ptr[-1]) + GETJSAMPLE(below_ptr[1]);
+    neighsum += above_ptr[-1] + above_ptr[1] + below_ptr[-1] + below_ptr[1];
    membersum = membersum * memberscale + neighsum * neighscale;
    *outptr = (JSAMPLE)((membersum + 32768) >> 16);

@@ -429,21 +415,18 @@ fullsize_smooth_downsample(j_compress_ptr cinfo, jpeg_component_info *compptr,
    below_ptr = input_data[outrow + 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);
+    colsum = (*above_ptr++) + (*below_ptr++) + inptr[0];
+    membersum = *inptr++;
+    nextcolsum = above_ptr[0] + below_ptr[0] + inptr[0];
    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++);
+      membersum = *inptr++;
      above_ptr++;  below_ptr++;
-      nextcolsum = GETJSAMPLE(*above_ptr) + GETJSAMPLE(*below_ptr) +
-                   GETJSAMPLE(*inptr);
+      nextcolsum = above_ptr[0] + below_ptr[0] + inptr[0];
      neighsum = lastcolsum + (colsum - membersum) + nextcolsum;
      membersum = membersum * memberscale + neighsum * neighscale;
      *outptr++ = (JSAMPLE)((membersum + 32768) >> 16);
@@ -451,7 +434,7 @@ fullsize_smooth_downsample(j_compress_ptr cinfo, jpeg_component_info *compptr,
    }

    /* Special case for last column */
-    membersum = GETJSAMPLE(*inptr);
+    membersum = *inptr;
    neighsum = lastcolsum + (colsum - membersum) + colsum;
    membersum = membersum * memberscale + neighsum * neighscale;
    *outptr = (JSAMPLE)((membersum + 32768) >> 16);
@@ -4,8 +4,8 @@
 * This file was part of the Independent JPEG Group's software:
 * Copyright (C) 1995-1998, Thomas G. Lane.
 * Modified 2000-2009 by Guido Vollbeding.
- * It was modified by The libjpeg-turbo Project to include only code relevant
- * to libjpeg-turbo.
+ * libjpeg-turbo Modifications:
+ * Copyright (C) 2020, D. R. Commander.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -17,6 +17,7 @@
 #define JPEG_INTERNALS
 #include "jinclude.h"
 #include "jpeglib.h"
+#include "jpegcomp.h"


 /* Forward declarations */
@@ -23,6 +23,7 @@
 #include "jinclude.h"
 #include "jpeglib.h"
 #include "jdmaster.h"
+#include "jconfigint.h"


 /*
@@ -308,7 +309,7 @@ jpeg_consume_input(j_decompress_ptr cinfo)
    /* Initialize application's data source module */
    (*cinfo->src->init_source) (cinfo);
    cinfo->global_state = DSTATE_INHEADER;
-    /*FALLTHROUGH*/
+    FALLTHROUGH                 /*FALLTHROUGH*/
  case DSTATE_INHEADER:
    retcode = (*cinfo->inputctl->consume_input) (cinfo);
    if (retcode == JPEG_REACHED_SOS) { /* Found SOS, prepare to decompress */
@@ -4,7 +4,7 @@
 * This file was part of the Independent JPEG Group's software:
 * Copyright (C) 1994-1996, Thomas G. Lane.
 * libjpeg-turbo Modifications:
- * Copyright (C) 2010, 2015-2018, D. R. Commander.
+ * Copyright (C) 2010, 2015-2020, D. R. Commander.
 * Copyright (C) 2015, Google, Inc.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
@@ -21,6 +21,8 @@
 #include "jinclude.h"
 #include "jdmainct.h"
 #include "jdcoefct.h"
+#include "jdmaster.h"
+#include "jdmerge.h"
 #include "jdsample.h"
 #include "jmemsys.h"

@@ -316,6 +318,10 @@ LOCAL(void)
 read_and_discard_scanlines(j_decompress_ptr cinfo, JDIMENSION num_lines)
 {
  JDIMENSION n;
+  my_master_ptr master = (my_master_ptr)cinfo->master;
+  JSAMPLE dummy_sample[1] = { 0 };
+  JSAMPROW dummy_row = dummy_sample;
+  JSAMPARRAY scanlines = NULL;
  void (*color_convert) (j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
                         JDIMENSION input_row, JSAMPARRAY output_buf,
                         int num_rows) = NULL;
@@ -325,6 +331,10 @@ read_and_discard_scanlines(j_decompress_ptr cinfo, JDIMENSION num_lines)
  if (cinfo->cconvert && cinfo->cconvert->color_convert) {
    color_convert = cinfo->cconvert->color_convert;
    cinfo->cconvert->color_convert = noop_convert;
+    /* This just prevents UBSan from complaining about adding 0 to a NULL
+     * pointer.  The pointer isn't actually used.
+     */
+    scanlines = &dummy_row;
  }

  if (cinfo->cquantize && cinfo->cquantize->color_quantize) {
@@ -332,8 +342,13 @@ read_and_discard_scanlines(j_decompress_ptr cinfo, JDIMENSION num_lines)
    cinfo->cquantize->color_quantize = noop_quantize;
  }

+  if (master->using_merged_upsample && cinfo->max_v_samp_factor == 2) {
+    my_merged_upsample_ptr upsample = (my_merged_upsample_ptr)cinfo->upsample;
+    scanlines = &upsample->spare_row;
+  }
+
  for (n = 0; n < num_lines; n++)
-    jpeg_read_scanlines(cinfo, NULL, 1);
+    jpeg_read_scanlines(cinfo, scanlines, 1);

  if (color_convert)
    cinfo->cconvert->color_convert = color_convert;
@@ -353,6 +368,12 @@ increment_simple_rowgroup_ctr(j_decompress_ptr cinfo, JDIMENSION rows)
 {
  JDIMENSION rows_left;
  my_main_ptr main_ptr = (my_main_ptr)cinfo->main;
+  my_master_ptr master = (my_master_ptr)cinfo->master;
+
+  if (master->using_merged_upsample && cinfo->max_v_samp_factor == 2) {
+    read_and_discard_scanlines(cinfo, rows);
+    return;
+  }

  /* Increment the counter to the next row group after the skipped rows. */
  main_ptr->rowgroup_ctr += rows / cinfo->max_v_samp_factor;
@@ -382,21 +403,27 @@ jpeg_skip_scanlines(j_decompress_ptr cinfo, JDIMENSION num_lines)
 {
  my_main_ptr main_ptr = (my_main_ptr)cinfo->main;
  my_coef_ptr coef = (my_coef_ptr)cinfo->coef;
+  my_master_ptr master = (my_master_ptr)cinfo->master;
  my_upsample_ptr upsample = (my_upsample_ptr)cinfo->upsample;
  JDIMENSION i, x;
  int y;
  JDIMENSION lines_per_iMCU_row, lines_left_in_iMCU_row, lines_after_iMCU_row;
  JDIMENSION lines_to_skip, lines_to_read;

+  /* Two-pass color quantization is not supported. */
+  if (cinfo->quantize_colors && cinfo->two_pass_quantize)
+    ERREXIT(cinfo, JERR_NOTIMPL);
+
  if (cinfo->global_state != DSTATE_SCANNING)
    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);

  /* Do not skip past the bottom of the image. */
  if (cinfo->output_scanline + num_lines >= cinfo->output_height) {
+    num_lines = cinfo->output_height - cinfo->output_scanline;
    cinfo->output_scanline = cinfo->output_height;
    (*cinfo->inputctl->finish_input_pass) (cinfo);
    cinfo->inputctl->eoi_reached = TRUE;
-    return cinfo->output_height - cinfo->output_scanline;
+    return num_lines;
  }

  if (num_lines == 0)
@@ -445,8 +472,10 @@ jpeg_skip_scanlines(j_decompress_ptr cinfo, JDIMENSION num_lines)
    main_ptr->buffer_full = FALSE;
    main_ptr->rowgroup_ctr = 0;
    main_ptr->context_state = CTX_PREPARE_FOR_IMCU;
-    upsample->next_row_out = cinfo->max_v_samp_factor;
-    upsample->rows_to_go = cinfo->output_height - cinfo->output_scanline;
+    if (!master->using_merged_upsample) {
+      upsample->next_row_out = cinfo->max_v_samp_factor;
+      upsample->rows_to_go = cinfo->output_height - cinfo->output_scanline;
+    }
  }

  /* Skipping is much simpler when context rows are not required. */
@@ -458,8 +487,10 @@ jpeg_skip_scanlines(j_decompress_ptr cinfo, JDIMENSION num_lines)
      cinfo->output_scanline += lines_left_in_iMCU_row;
      main_ptr->buffer_full = FALSE;
      main_ptr->rowgroup_ctr = 0;
-      upsample->next_row_out = cinfo->max_v_samp_factor;
-      upsample->rows_to_go = cinfo->output_height - cinfo->output_scanline;
+      if (!master->using_merged_upsample) {
+        upsample->next_row_out = cinfo->max_v_samp_factor;
+        upsample->rows_to_go = cinfo->output_height - cinfo->output_scanline;
+      }
    }
  }

@@ -494,7 +525,8 @@ jpeg_skip_scanlines(j_decompress_ptr cinfo, JDIMENSION num_lines)
      cinfo->output_iMCU_row += lines_to_skip / lines_per_iMCU_row;
      increment_simple_rowgroup_ctr(cinfo, lines_to_read);
    }
-    upsample->rows_to_go = cinfo->output_height - cinfo->output_scanline;
+    if (!master->using_merged_upsample)
+      upsample->rows_to_go = cinfo->output_height - cinfo->output_scanline;
    return num_lines;
  }

@@ -506,6 +538,8 @@ jpeg_skip_scanlines(j_decompress_ptr cinfo, JDIMENSION num_lines)
         * decoded coefficients.  This is ~5% faster for large subsets, but
         * it's tough to tell a difference for smaller images.
         */
+        if (!cinfo->entropy->insufficient_data)
+          cinfo->master->last_good_iMCU_row = cinfo->input_iMCU_row;
        (*cinfo->entropy->decode_mcu) (cinfo, NULL);
      }
    }
@@ -535,7 +569,8 @@ jpeg_skip_scanlines(j_decompress_ptr cinfo, JDIMENSION num_lines)
   * bit odd, since "rows_to_go" seems to be redundantly keeping track of
   * output_scanline.
   */
-  upsample->rows_to_go = cinfo->output_height - cinfo->output_scanline;
+  if (!master->using_merged_upsample)
+    upsample->rows_to_go = cinfo->output_height - cinfo->output_scanline;

  /* Always skip the requested number of lines. */
  return num_lines;
@@ -4,7 +4,7 @@
 * This file was part of the Independent JPEG Group's software:
 * Developed 1997-2015 by Guido Vollbeding.
 * libjpeg-turbo Modifications:
- * Copyright (C) 2015-2018, D. R. Commander.
+ * Copyright (C) 2015-2020, D. R. Commander.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -80,7 +80,7 @@ get_byte(j_decompress_ptr cinfo)
    if (!(*src->fill_input_buffer) (cinfo))
      ERREXIT(cinfo, JERR_CANT_SUSPEND);
  src->bytes_in_buffer--;
-  return GETJOCTET(*src->next_input_byte++);
+  return *src->next_input_byte++;
 }


@@ -665,8 +665,16 @@ bad:
    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];
+      int *prev_coef_bit_ptr =
+        &cinfo->coef_bits[cindex + cinfo->num_components][0];
      if (cinfo->Ss && coef_bit_ptr[0] < 0) /* AC without prior DC scan */
        WARNMS2(cinfo, JWRN_BOGUS_PROGRESSION, cindex, 0);
+      for (coefi = MIN(cinfo->Ss, 1); coefi <= MAX(cinfo->Se, 9); coefi++) {
+        if (cinfo->input_scan_number > 1)
+          prev_coef_bit_ptr[coefi] = coef_bit_ptr[coefi];
+        else
+          prev_coef_bit_ptr[coefi] = 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)
@@ -727,6 +735,7 @@ bad:
  entropy->c = 0;
  entropy->a = 0;
  entropy->ct = -16;    /* force reading 2 initial bytes to fill C */
+  entropy->pub.insufficient_data = FALSE;

  /* Initialize restart counter */
  entropy->restarts_to_go = cinfo->restart_interval;
@@ -763,7 +772,7 @@ jinit_arith_decoder(j_decompress_ptr cinfo)
    int *coef_bit_ptr, ci;
    cinfo->coef_bits = (int (*)[DCTSIZE2])
      (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
-                                  cinfo->num_components * DCTSIZE2 *
+                                  cinfo->num_components * 2 * DCTSIZE2 *
                                  sizeof(int));
    coef_bit_ptr = &cinfo->coef_bits[0][0];
    for (ci = 0; ci < cinfo->num_components; ci++)
@@ -143,8 +143,7 @@ empty_mem_output_buffer(j_compress_ptr cinfo)

  MEMCOPY(nextbuffer, dest->buffer, dest->bufsize);

-  if (dest->newbuffer != NULL)
-    free(dest->newbuffer);
+  free(dest->newbuffer);

  dest->newbuffer = nextbuffer;

@@ -5,8 +5,8 @@
 * Copyright (C) 1994-1997, Thomas G. Lane.
 * libjpeg-turbo Modifications:
 * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
- * Copyright (C) 2010, 2015-2016, D. R. Commander.
- * Copyright (C) 2015, Google, Inc.
+ * Copyright (C) 2010, 2015-2016, 2019-2020, D. R. Commander.
+ * Copyright (C) 2015, 2020, Google, Inc.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -102,6 +102,8 @@ decompress_onepass(j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
      /* Try to fetch an MCU.  Entropy decoder expects buffer to be zeroed. */
      jzero_far((void *)coef->MCU_buffer[0],
                (size_t)(cinfo->blocks_in_MCU * sizeof(JBLOCK)));
+      if (!cinfo->entropy->insufficient_data)
+        cinfo->master->last_good_iMCU_row = cinfo->input_iMCU_row;
      if (!(*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) {
        /* Suspension forced; update state counters and exit */
        coef->MCU_vert_offset = yoffset;
@@ -227,6 +229,8 @@ consume_data(j_decompress_ptr cinfo)
          }
        }
      }
+      if (!cinfo->entropy->insufficient_data)
+        cinfo->master->last_good_iMCU_row = cinfo->input_iMCU_row;
      /* Try to fetch the MCU. */
      if (!(*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) {
        /* Suspension forced; update state counters and exit */
@@ -326,19 +330,22 @@ decompress_data(j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
 #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.
+ * This code applies interblock smoothing; the first 9 AC coefficients are
+ * estimated from the DC values of a DCT block and its 24 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 */
+/* Natural-order array positions of the first 9 zigzag-order coefficients */
 #define Q01_POS  1
 #define Q10_POS  8
 #define Q20_POS  16
 #define Q11_POS  9
 #define Q02_POS  2
+#define Q03_POS  3
+#define Q12_POS  10
+#define Q21_POS  17
+#define Q30_POS  24

 /*
 * Determine whether block smoothing is applicable and safe.
@@ -356,8 +363,8 @@ smoothing_ok(j_decompress_ptr cinfo)
  int ci, coefi;
  jpeg_component_info *compptr;
  JQUANT_TBL *qtable;
-  int *coef_bits;
-  int *coef_bits_latch;
+  int *coef_bits, *prev_coef_bits;
+  int *coef_bits_latch, *prev_coef_bits_latch;

  if (!cinfo->progressive_mode || cinfo->coef_bits == NULL)
    return FALSE;
@@ -366,34 +373,47 @@ smoothing_ok(j_decompress_ptr cinfo)
  if (coef->coef_bits_latch == NULL)
    coef->coef_bits_latch = (int *)
      (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
-                                  cinfo->num_components *
+                                  cinfo->num_components * 2 *
                                  (SAVED_COEFS * sizeof(int)));
  coef_bits_latch = coef->coef_bits_latch;
+  prev_coef_bits_latch =
+    &coef->coef_bits_latch[cinfo->num_components * SAVED_COEFS];

  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. */
+    /* Verify DC & first 9 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)
+        qtable->quantval[Q02_POS] == 0 ||
+        qtable->quantval[Q03_POS] == 0 ||
+        qtable->quantval[Q12_POS] == 0 ||
+        qtable->quantval[Q21_POS] == 0 ||
+        qtable->quantval[Q30_POS] == 0)
      return FALSE;
    /* DC values must be at least partly known for all components. */
    coef_bits = cinfo->coef_bits[ci];
+    prev_coef_bits = cinfo->coef_bits[ci + cinfo->num_components];
    if (coef_bits[0] < 0)
      return FALSE;
+    coef_bits_latch[0] = coef_bits[0];
    /* Block smoothing is helpful if some AC coefficients remain inaccurate. */
-    for (coefi = 1; coefi <= 5; coefi++) {
+    for (coefi = 1; coefi < SAVED_COEFS; coefi++) {
+      if (cinfo->input_scan_number > 1)
+        prev_coef_bits_latch[coefi] = prev_coef_bits[coefi];
+      else
+        prev_coef_bits_latch[coefi] = -1;
      coef_bits_latch[coefi] = coef_bits[coefi];
      if (coef_bits[coefi] != 0)
        smoothing_useful = TRUE;
    }
    coef_bits_latch += SAVED_COEFS;
+    prev_coef_bits_latch += SAVED_COEFS;
  }

  return smoothing_useful;
@@ -412,17 +432,20 @@ decompress_smooth_data(j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
  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;
+  JBLOCKROW buffer_ptr, prev_prev_block_row, prev_block_row;
+  JBLOCKROW next_block_row, next_next_block_row;
  JSAMPARRAY output_ptr;
  JDIMENSION output_col;
  jpeg_component_info *compptr;
  inverse_DCT_method_ptr inverse_DCT;
-  boolean first_row, last_row;
+  boolean change_dc;
  JCOEF *workspace;
  int *coef_bits;
  JQUANT_TBL *quanttbl;
-  JLONG Q00, Q01, Q02, Q10, Q11, Q20, num;
-  int DC1, DC2, DC3, DC4, DC5, DC6, DC7, DC8, DC9;
+  JLONG Q00, Q01, Q02, Q03 = 0, Q10, Q11, Q12 = 0, Q20, Q21 = 0, Q30 = 0, num;
+  int DC01, DC02, DC03, DC04, DC05, DC06, DC07, DC08, DC09, DC10, DC11, DC12,
+      DC13, DC14, DC15, DC16, DC17, DC18, DC19, DC20, DC21, DC22, DC23, DC24,
+      DC25;
  int Al, pred;

  /* Keep a local variable to avoid looking it up more than once */
@@ -434,10 +457,10 @@ decompress_smooth_data(j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
    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
+       * we want it to keep two rows ahead so that next two block rows' DC
       * values are up to date.
       */
-      JDIMENSION delta = (cinfo->Ss == 0) ? 1 : 0;
+      JDIMENSION delta = (cinfo->Ss == 0) ? 2 : 0;
      if (cinfo->input_iMCU_row > cinfo->output_iMCU_row + delta)
        break;
    }
@@ -452,34 +475,53 @@ decompress_smooth_data(j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
    if (!compptr->component_needed)
      continue;
    /* Count non-dummy DCT block rows in this iMCU row. */
-    if (cinfo->output_iMCU_row < last_iMCU_row) {
+    if (cinfo->output_iMCU_row < last_iMCU_row - 1) {
+      block_rows = compptr->v_samp_factor;
+      access_rows = block_rows * 3; /* this and next two iMCU rows */
+    } else 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 */
+    if (cinfo->output_iMCU_row > 1) {
+      access_rows += 2 * compptr->v_samp_factor; /* prior two iMCU rows too */
+      buffer = (*cinfo->mem->access_virt_barray)
+        ((j_common_ptr)cinfo, coef->whole_image[ci],
+         (cinfo->output_iMCU_row - 2) * compptr->v_samp_factor,
+         (JDIMENSION)access_rows, FALSE);
+      buffer += 2 * compptr->v_samp_factor; /* point to current iMCU row */
+    } else if (cinfo->output_iMCU_row > 0) {
      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);
+    /* Fetch component-dependent info.
+     * If the current scan is incomplete, then we use the component-dependent
+     * info from the previous scan.
+     */
+    if (cinfo->output_iMCU_row > cinfo->master->last_good_iMCU_row)
+      coef_bits =
+        coef->coef_bits_latch + ((ci + cinfo->num_components) * SAVED_COEFS);
+    else
+      coef_bits = coef->coef_bits_latch + (ci * SAVED_COEFS);
+
+    /* We only do DC interpolation if no AC coefficient data is available. */
+    change_dc =
+      coef_bits[1] == -1 && coef_bits[2] == -1 && coef_bits[3] == -1 &&
+      coef_bits[4] == -1 && coef_bits[5] == -1 && coef_bits[6] == -1 &&
+      coef_bits[7] == -1 && coef_bits[8] == -1 && coef_bits[9] == -1;
+
    quanttbl = compptr->quant_table;
    Q00 = quanttbl->quantval[0];
    Q01 = quanttbl->quantval[Q01_POS];
@@ -487,25 +529,51 @@ decompress_smooth_data(j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
    Q20 = quanttbl->quantval[Q20_POS];
    Q11 = quanttbl->quantval[Q11_POS];
    Q02 = quanttbl->quantval[Q02_POS];
+    if (change_dc) {
+      Q03 = quanttbl->quantval[Q03_POS];
+      Q12 = quanttbl->quantval[Q12_POS];
+      Q21 = quanttbl->quantval[Q21_POS];
+      Q30 = quanttbl->quantval[Q30_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] + cinfo->master->first_MCU_col[ci];
-      if (first_row && block_row == 0)
+
+      if (block_row > 0 || cinfo->output_iMCU_row > 0)
+        prev_block_row =
+          buffer[block_row - 1] + cinfo->master->first_MCU_col[ci];
+      else
        prev_block_row = buffer_ptr;
+
+      if (block_row > 1 || cinfo->output_iMCU_row > 1)
+        prev_prev_block_row =
+          buffer[block_row - 2] + cinfo->master->first_MCU_col[ci];
+      else
+        prev_prev_block_row = prev_block_row;
+
+      if (block_row < block_rows - 1 || cinfo->output_iMCU_row < last_iMCU_row)
+        next_block_row =
+          buffer[block_row + 1] + cinfo->master->first_MCU_col[ci];
      else
-        prev_block_row = buffer[block_row - 1];
-      if (last_row && block_row == block_rows - 1)
        next_block_row = buffer_ptr;
+
+      if (block_row < block_rows - 2 ||
+          cinfo->output_iMCU_row < last_iMCU_row - 1)
+        next_next_block_row =
+          buffer[block_row + 2] + cinfo->master->first_MCU_col[ci];
      else
-        next_block_row = buffer[block_row + 1];
+        next_next_block_row = next_block_row;
+
      /* 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.
+       * Initialize all 25 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];
+      DC01 = DC02 = DC03 = DC04 = DC05 = (int)prev_prev_block_row[0][0];
+      DC06 = DC07 = DC08 = DC09 = DC10 = (int)prev_block_row[0][0];
+      DC11 = DC12 = DC13 = DC14 = DC15 = (int)buffer_ptr[0][0];
+      DC16 = DC17 = DC18 = DC19 = DC20 = (int)next_block_row[0][0];
+      DC21 = DC22 = DC23 = DC24 = DC25 = (int)next_next_block_row[0][0];
      output_col = 0;
      last_block_column = compptr->width_in_blocks - 1;
      for (block_num = cinfo->master->first_MCU_col[ci];
@@ -513,18 +581,39 @@ decompress_smooth_data(j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
        /* 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];
+        if (block_num == cinfo->master->first_MCU_col[ci] &&
+            block_num < last_block_column) {
+          DC04 = (int)prev_prev_block_row[1][0];
+          DC09 = (int)prev_block_row[1][0];
+          DC14 = (int)buffer_ptr[1][0];
+          DC19 = (int)next_block_row[1][0];
+          DC24 = (int)next_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.
+        if (block_num + 1 < last_block_column) {
+          DC05 = (int)prev_prev_block_row[2][0];
+          DC10 = (int)prev_block_row[2][0];
+          DC15 = (int)buffer_ptr[2][0];
+          DC20 = (int)next_block_row[2][0];
+          DC25 = (int)next_next_block_row[2][0];
+        }
+        /* If DC interpolation is enabled, compute coefficient estimates using
+         * a Gaussian-like kernel, keeping the averages of the DC values.
+         *
+         * If DC interpolation is disabled, compute coefficient estimates using
+         * an algorithm similar to the one described in Section K.8 of the JPEG
+         * standard, except applied to a 5x5 window rather than a 3x3 window.
+         *
+         * An estimate is applied only if the 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);
+          num = Q00 * (change_dc ?
+                (-DC01 - DC02 + DC04 + DC05 - 3 * DC06 + 13 * DC07 -
+                 13 * DC09 + 3 * DC10 - 3 * DC11 + 38 * DC12 - 38 * DC14 +
+                 3 * DC15 - 3 * DC16 + 13 * DC17 - 13 * DC19 + 3 * DC20 -
+                 DC21 - DC22 + DC24 + DC25) :
+                (-7 * DC11 + 50 * DC12 - 50 * DC14 + 7 * DC15));
          if (num >= 0) {
            pred = (int)(((Q01 << 7) + num) / (Q01 << 8));
            if (Al > 0 && pred >= (1 << Al))
@@ -539,7 +628,12 @@ decompress_smooth_data(j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
        }
        /* AC10 */
        if ((Al = coef_bits[2]) != 0 && workspace[8] == 0) {
-          num = 36 * Q00 * (DC2 - DC8);
+          num = Q00 * (change_dc ?
+                (-DC01 - 3 * DC02 - 3 * DC03 - 3 * DC04 - DC05 - DC06 +
+                 13 * DC07 + 38 * DC08 + 13 * DC09 - DC10 + DC16 -
+                 13 * DC17 - 38 * DC18 - 13 * DC19 + DC20 + DC21 +
+                 3 * DC22 + 3 * DC23 + 3 * DC24 + DC25) :
+                (-7 * DC03 + 50 * DC08 - 50 * DC18 + 7 * DC23));
          if (num >= 0) {
            pred = (int)(((Q10 << 7) + num) / (Q10 << 8));
            if (Al > 0 && pred >= (1 << Al))
@@ -554,7 +648,10 @@ decompress_smooth_data(j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
        }
        /* AC20 */
        if ((Al = coef_bits[3]) != 0 && workspace[16] == 0) {
-          num = 9 * Q00 * (DC2 + DC8 - 2 * DC5);
+          num = Q00 * (change_dc ?
+                (DC03 + 2 * DC07 + 7 * DC08 + 2 * DC09 - 5 * DC12 - 14 * DC13 -
+                 5 * DC14 + 2 * DC17 + 7 * DC18 + 2 * DC19 + DC23) :
+                (-DC03 + 13 * DC08 - 24 * DC13 + 13 * DC18 - DC23));
          if (num >= 0) {
            pred = (int)(((Q20 << 7) + num) / (Q20 << 8));
            if (Al > 0 && pred >= (1 << Al))
@@ -569,7 +666,11 @@ decompress_smooth_data(j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
        }
        /* AC11 */
        if ((Al = coef_bits[4]) != 0 && workspace[9] == 0) {
-          num = 5 * Q00 * (DC1 - DC3 - DC7 + DC9);
+          num = Q00 * (change_dc ?
+                (-DC01 + DC05 + 9 * DC07 - 9 * DC09 - 9 * DC17 +
+                 9 * DC19 + DC21 - DC25) :
+                (DC10 + DC16 - 10 * DC17 + 10 * DC19 - DC02 - DC20 + DC22 -
+                 DC24 + DC04 - DC06 + 10 * DC07 - 10 * DC09));
          if (num >= 0) {
            pred = (int)(((Q11 << 7) + num) / (Q11 << 8));
            if (Al > 0 && pred >= (1 << Al))
@@ -584,7 +685,10 @@ decompress_smooth_data(j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
        }
        /* AC02 */
        if ((Al = coef_bits[5]) != 0 && workspace[2] == 0) {
-          num = 9 * Q00 * (DC4 + DC6 - 2 * DC5);
+          num = Q00 * (change_dc ?
+                (2 * DC07 - 5 * DC08 + 2 * DC09 + DC11 + 7 * DC12 - 14 * DC13 +
+                 7 * DC14 + DC15 + 2 * DC17 - 5 * DC18 + 2 * DC19) :
+                (-DC11 + 13 * DC12 - 24 * DC13 + 13 * DC14 - DC15));
          if (num >= 0) {
            pred = (int)(((Q02 << 7) + num) / (Q02 << 8));
            if (Al > 0 && pred >= (1 << Al))
@@ -597,14 +701,96 @@ decompress_smooth_data(j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
          }
          workspace[2] = (JCOEF)pred;
        }
+        if (change_dc) {
+          /* AC03 */
+          if ((Al = coef_bits[6]) != 0 && workspace[3] == 0) {
+            num = Q00 * (DC07 - DC09 + 2 * DC12 - 2 * DC14 + DC17 - DC19);
+            if (num >= 0) {
+              pred = (int)(((Q03 << 7) + num) / (Q03 << 8));
+              if (Al > 0 && pred >= (1 << Al))
+                pred = (1 << Al) - 1;
+            } else {
+              pred = (int)(((Q03 << 7) - num) / (Q03 << 8));
+              if (Al > 0 && pred >= (1 << Al))
+                pred = (1 << Al) - 1;
+              pred = -pred;
+            }
+            workspace[3] = (JCOEF)pred;
+          }
+          /* AC12 */
+          if ((Al = coef_bits[7]) != 0 && workspace[10] == 0) {
+            num = Q00 * (DC07 - 3 * DC08 + DC09 - DC17 + 3 * DC18 - DC19);
+            if (num >= 0) {
+              pred = (int)(((Q12 << 7) + num) / (Q12 << 8));
+              if (Al > 0 && pred >= (1 << Al))
+                pred = (1 << Al) - 1;
+            } else {
+              pred = (int)(((Q12 << 7) - num) / (Q12 << 8));
+              if (Al > 0 && pred >= (1 << Al))
+                pred = (1 << Al) - 1;
+              pred = -pred;
+            }
+            workspace[10] = (JCOEF)pred;
+          }
+          /* AC21 */
+          if ((Al = coef_bits[8]) != 0 && workspace[17] == 0) {
+            num = Q00 * (DC07 - DC09 - 3 * DC12 + 3 * DC14 + DC17 - DC19);
+            if (num >= 0) {
+              pred = (int)(((Q21 << 7) + num) / (Q21 << 8));
+              if (Al > 0 && pred >= (1 << Al))
+                pred = (1 << Al) - 1;
+            } else {
+              pred = (int)(((Q21 << 7) - num) / (Q21 << 8));
+              if (Al > 0 && pred >= (1 << Al))
+                pred = (1 << Al) - 1;
+              pred = -pred;
+            }
+            workspace[17] = (JCOEF)pred;
+          }
+          /* AC30 */
+          if ((Al = coef_bits[9]) != 0 && workspace[24] == 0) {
+            num = Q00 * (DC07 + 2 * DC08 + DC09 - DC17 - 2 * DC18 - DC19);
+            if (num >= 0) {
+              pred = (int)(((Q30 << 7) + num) / (Q30 << 8));
+              if (Al > 0 && pred >= (1 << Al))
+                pred = (1 << Al) - 1;
+            } else {
+              pred = (int)(((Q30 << 7) - num) / (Q30 << 8));
+              if (Al > 0 && pred >= (1 << Al))
+                pred = (1 << Al) - 1;
+              pred = -pred;
+            }
+            workspace[24] = (JCOEF)pred;
+          }
+          /* coef_bits[0] is non-negative.  Otherwise this function would not
+           * be called.
+           */
+          num = Q00 *
+                (-2 * DC01 - 6 * DC02 - 8 * DC03 - 6 * DC04 - 2 * DC05 -
+                 6 * DC06 + 6 * DC07 + 42 * DC08 + 6 * DC09 - 6 * DC10 -
+                 8 * DC11 + 42 * DC12 + 152 * DC13 + 42 * DC14 - 8 * DC15 -
+                 6 * DC16 + 6 * DC17 + 42 * DC18 + 6 * DC19 - 6 * DC20 -
+                 2 * DC21 - 6 * DC22 - 8 * DC23 - 6 * DC24 - 2 * DC25);
+          if (num >= 0) {
+            pred = (int)(((Q00 << 7) + num) / (Q00 << 8));
+          } else {
+            pred = (int)(((Q00 << 7) - num) / (Q00 << 8));
+            pred = -pred;
+          }
+          workspace[0] = (JCOEF)pred;
+        }  /* change_dc */
+
        /* 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++;
+        DC01 = DC02;  DC02 = DC03;  DC03 = DC04;  DC04 = DC05;
+        DC06 = DC07;  DC07 = DC08;  DC08 = DC09;  DC09 = DC10;
+        DC11 = DC12;  DC12 = DC13;  DC13 = DC14;  DC14 = DC15;
+        DC16 = DC17;  DC17 = DC18;  DC18 = DC19;  DC19 = DC20;
+        DC21 = DC22;  DC22 = DC23;  DC23 = DC24;  DC24 = DC25;
+        buffer_ptr++, prev_block_row++, next_block_row++,
+          prev_prev_block_row++, next_next_block_row++;
        output_col += compptr->_DCT_scaled_size;
      }
      output_ptr += compptr->_DCT_scaled_size;
@@ -653,7 +839,7 @@ jinit_d_coef_controller(j_decompress_ptr cinfo, boolean need_full_buffer)
 #ifdef BLOCK_SMOOTHING_SUPPORTED
      /* If block smoothing could be used, need a bigger window */
      if (cinfo->progressive_mode)
-        access_rows *= 3;
+        access_rows *= 5;
 #endif
      coef->whole_image[ci] = (*cinfo->mem->request_virt_barray)
        ((j_common_ptr)cinfo, JPOOL_IMAGE, TRUE,
@@ -5,6 +5,7 @@
 * Copyright (C) 1994-1997, Thomas G. Lane.
 * libjpeg-turbo Modifications:
 * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+ * Copyright (C) 2020, Google, Inc.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 */
@@ -51,7 +52,7 @@ typedef struct {
 #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] */
+#define SAVED_COEFS  10         /* we save coef_bits[0..9] */
 #endif
 } my_coef_controller;

@@ -45,9 +45,9 @@ ycc_rgb565_convert_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
    outptr = *output_buf++;

    if (PACK_NEED_ALIGNMENT(outptr)) {
-      y  = GETJSAMPLE(*inptr0++);
-      cb = GETJSAMPLE(*inptr1++);
-      cr = GETJSAMPLE(*inptr2++);
+      y  = *inptr0++;
+      cb = *inptr1++;
+      cr = *inptr2++;
      r = range_limit[y + Crrtab[cr]];
      g = range_limit[y + ((int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr],
                                            SCALEBITS))];
@@ -58,18 +58,18 @@ ycc_rgb565_convert_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
      num_cols--;
    }
    for (col = 0; col < (num_cols >> 1); col++) {
-      y  = GETJSAMPLE(*inptr0++);
-      cb = GETJSAMPLE(*inptr1++);
-      cr = GETJSAMPLE(*inptr2++);
+      y  = *inptr0++;
+      cb = *inptr1++;
+      cr = *inptr2++;
      r = range_limit[y + Crrtab[cr]];
      g = range_limit[y + ((int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr],
                                            SCALEBITS))];
      b = range_limit[y + Cbbtab[cb]];
      rgb = PACK_SHORT_565(r, g, b);

-      y  = GETJSAMPLE(*inptr0++);
-      cb = GETJSAMPLE(*inptr1++);
-      cr = GETJSAMPLE(*inptr2++);
+      y  = *inptr0++;
+      cb = *inptr1++;
+      cr = *inptr2++;
      r = range_limit[y + Crrtab[cr]];
      g = range_limit[y + ((int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr],
                                            SCALEBITS))];
@@ -80,9 +80,9 @@ ycc_rgb565_convert_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
      outptr += 4;
    }
    if (num_cols & 1) {
-      y  = GETJSAMPLE(*inptr0);
-      cb = GETJSAMPLE(*inptr1);
-      cr = GETJSAMPLE(*inptr2);
+      y  = *inptr0;
+      cb = *inptr1;
+      cr = *inptr2;
      r = range_limit[y + Crrtab[cr]];
      g = range_limit[y + ((int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr],
                                            SCALEBITS))];
@@ -125,9 +125,9 @@ ycc_rgb565D_convert_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
    input_row++;
    outptr = *output_buf++;
    if (PACK_NEED_ALIGNMENT(outptr)) {
-      y  = GETJSAMPLE(*inptr0++);
-      cb = GETJSAMPLE(*inptr1++);
-      cr = GETJSAMPLE(*inptr2++);
+      y  = *inptr0++;
+      cb = *inptr1++;
+      cr = *inptr2++;
      r = range_limit[DITHER_565_R(y + Crrtab[cr], d0)];
      g = range_limit[DITHER_565_G(y +
                                   ((int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr],
@@ -139,9 +139,9 @@ ycc_rgb565D_convert_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
      num_cols--;
    }
    for (col = 0; col < (num_cols >> 1); col++) {
-      y  = GETJSAMPLE(*inptr0++);
-      cb = GETJSAMPLE(*inptr1++);
-      cr = GETJSAMPLE(*inptr2++);
+      y  = *inptr0++;
+      cb = *inptr1++;
+      cr = *inptr2++;
      r = range_limit[DITHER_565_R(y + Crrtab[cr], d0)];
      g = range_limit[DITHER_565_G(y +
                                   ((int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr],
@@ -150,9 +150,9 @@ ycc_rgb565D_convert_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
      d0 = DITHER_ROTATE(d0);
      rgb = PACK_SHORT_565(r, g, b);

-      y  = GETJSAMPLE(*inptr0++);
-      cb = GETJSAMPLE(*inptr1++);
-      cr = GETJSAMPLE(*inptr2++);
+      y  = *inptr0++;
+      cb = *inptr1++;
+      cr = *inptr2++;
      r = range_limit[DITHER_565_R(y + Crrtab[cr], d0)];
      g = range_limit[DITHER_565_G(y +
                                   ((int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr],
@@ -165,9 +165,9 @@ ycc_rgb565D_convert_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
      outptr += 4;
    }
    if (num_cols & 1) {
-      y  = GETJSAMPLE(*inptr0);
-      cb = GETJSAMPLE(*inptr1);
-      cr = GETJSAMPLE(*inptr2);
+      y  = *inptr0;
+      cb = *inptr1;
+      cr = *inptr2;
      r = range_limit[DITHER_565_R(y + Crrtab[cr], d0)];
      g = range_limit[DITHER_565_G(y +
                                   ((int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr],
@@ -202,32 +202,32 @@ rgb_rgb565_convert_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
    input_row++;
    outptr = *output_buf++;
    if (PACK_NEED_ALIGNMENT(outptr)) {
-      r = GETJSAMPLE(*inptr0++);
-      g = GETJSAMPLE(*inptr1++);
-      b = GETJSAMPLE(*inptr2++);
+      r = *inptr0++;
+      g = *inptr1++;
+      b = *inptr2++;
      rgb = PACK_SHORT_565(r, g, b);
      *(INT16 *)outptr = (INT16)rgb;
      outptr += 2;
      num_cols--;
    }
    for (col = 0; col < (num_cols >> 1); col++) {
-      r = GETJSAMPLE(*inptr0++);
-      g = GETJSAMPLE(*inptr1++);
-      b = GETJSAMPLE(*inptr2++);
+      r = *inptr0++;
+      g = *inptr1++;
+      b = *inptr2++;
      rgb = PACK_SHORT_565(r, g, b);

-      r = GETJSAMPLE(*inptr0++);
-      g = GETJSAMPLE(*inptr1++);
-      b = GETJSAMPLE(*inptr2++);
+      r = *inptr0++;
+      g = *inptr1++;
+      b = *inptr2++;
      rgb = PACK_TWO_PIXELS(rgb, PACK_SHORT_565(r, g, b));

      WRITE_TWO_ALIGNED_PIXELS(outptr, rgb);
      outptr += 4;
    }
    if (num_cols & 1) {
-      r = GETJSAMPLE(*inptr0);
-      g = GETJSAMPLE(*inptr1);
-      b = GETJSAMPLE(*inptr2);
+      r = *inptr0;
+      g = *inptr1;
+      b = *inptr2;
      rgb = PACK_SHORT_565(r, g, b);
      *(INT16 *)outptr = (INT16)rgb;
    }
@@ -259,24 +259,24 @@ rgb_rgb565D_convert_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
    input_row++;
    outptr = *output_buf++;
    if (PACK_NEED_ALIGNMENT(outptr)) {
-      r = range_limit[DITHER_565_R(GETJSAMPLE(*inptr0++), d0)];
-      g = range_limit[DITHER_565_G(GETJSAMPLE(*inptr1++), d0)];
-      b = range_limit[DITHER_565_B(GETJSAMPLE(*inptr2++), d0)];
+      r = range_limit[DITHER_565_R(*inptr0++, d0)];
+      g = range_limit[DITHER_565_G(*inptr1++, d0)];
+      b = range_limit[DITHER_565_B(*inptr2++, d0)];
      rgb = PACK_SHORT_565(r, g, b);
      *(INT16 *)outptr = (INT16)rgb;
      outptr += 2;
      num_cols--;
    }
    for (col = 0; col < (num_cols >> 1); col++) {
-      r = range_limit[DITHER_565_R(GETJSAMPLE(*inptr0++), d0)];
-      g = range_limit[DITHER_565_G(GETJSAMPLE(*inptr1++), d0)];
-      b = range_limit[DITHER_565_B(GETJSAMPLE(*inptr2++), d0)];
+      r = range_limit[DITHER_565_R(*inptr0++, d0)];
+      g = range_limit[DITHER_565_G(*inptr1++, d0)];
+      b = range_limit[DITHER_565_B(*inptr2++, d0)];
      d0 = DITHER_ROTATE(d0);
      rgb = PACK_SHORT_565(r, g, b);

-      r = range_limit[DITHER_565_R(GETJSAMPLE(*inptr0++), d0)];
-      g = range_limit[DITHER_565_G(GETJSAMPLE(*inptr1++), d0)];
-      b = range_limit[DITHER_565_B(GETJSAMPLE(*inptr2++), d0)];
+      r = range_limit[DITHER_565_R(*inptr0++, d0)];
+      g = range_limit[DITHER_565_G(*inptr1++, d0)];
+      b = range_limit[DITHER_565_B(*inptr2++, d0)];
      d0 = DITHER_ROTATE(d0);
      rgb = PACK_TWO_PIXELS(rgb, PACK_SHORT_565(r, g, b));

@@ -284,9 +284,9 @@ rgb_rgb565D_convert_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
      outptr += 4;
    }
    if (num_cols & 1) {
-      r = range_limit[DITHER_565_R(GETJSAMPLE(*inptr0), d0)];
-      g = range_limit[DITHER_565_G(GETJSAMPLE(*inptr1), d0)];
-      b = range_limit[DITHER_565_B(GETJSAMPLE(*inptr2), d0)];
+      r = range_limit[DITHER_565_R(*inptr0, d0)];
+      g = range_limit[DITHER_565_G(*inptr1, d0)];
+      b = range_limit[DITHER_565_B(*inptr2, d0)];
      rgb = PACK_SHORT_565(r, g, b);
      *(INT16 *)outptr = (INT16)rgb;
    }
@@ -53,9 +53,9 @@ ycc_rgb_convert_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
    input_row++;
    outptr = *output_buf++;
    for (col = 0; col < num_cols; col++) {
-      y  = GETJSAMPLE(inptr0[col]);
-      cb = GETJSAMPLE(inptr1[col]);
-      cr = GETJSAMPLE(inptr2[col]);
+      y  = inptr0[col];
+      cb = inptr1[col];
+      cr = inptr2[col];
      /* Range-limiting is essential due to noise introduced by DCT losses. */
      outptr[RGB_RED] =   range_limit[y + Crrtab[cr]];
      outptr[RGB_GREEN] = range_limit[y +
@@ -93,7 +93,6 @@ gray_rgb_convert_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
    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];
      /* Set unused byte to 0xFF so it can be interpreted as an opaque */
      /* alpha channel value */
@@ -128,7 +127,6 @@ rgb_rgb_convert_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
    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];
@@ -341,9 +341,9 @@ rgb_gray_convert(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
    input_row++;
    outptr = *output_buf++;
    for (col = 0; col < num_cols; col++) {
-      r = GETJSAMPLE(inptr0[col]);
-      g = GETJSAMPLE(inptr1[col]);
-      b = GETJSAMPLE(inptr2[col]);
+      r = inptr0[col];
+      g = inptr1[col];
+      b = inptr2[col];
      /* Y */
      outptr[col] = (JSAMPLE)((ctab[r + R_Y_OFF] + ctab[g + G_Y_OFF] +
                               ctab[b + B_Y_OFF]) >> SCALEBITS);
@@ -550,9 +550,9 @@ ycck_cmyk_convert(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
    input_row++;
    outptr = *output_buf++;
    for (col = 0; col < num_cols; col++) {
-      y  = GETJSAMPLE(inptr0[col]);
-      cb = GETJSAMPLE(inptr1[col]);
-      cr = GETJSAMPLE(inptr2[col]);
+      y  = inptr0[col];
+      cb = inptr1[col];
+      cr = inptr2[col];
      /* Range-limiting is essential due to noise introduced by DCT losses. */
      outptr[0] = range_limit[MAXJSAMPLE - (y + Crrtab[cr])];   /* red */
      outptr[1] = range_limit[MAXJSAMPLE - (y +                 /* green */
@@ -560,7 +560,7 @@ ycck_cmyk_convert(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
                                                 SCALEBITS)))];
      outptr[2] = range_limit[MAXJSAMPLE - (y + Cbbtab[cb])];   /* blue */
      /* K passes through unchanged */
-      outptr[3] = inptr3[col];  /* don't need GETJSAMPLE here */
+      outptr[3] = inptr3[col];
      outptr += 4;
    }
  }
@@ -571,11 +571,10 @@ ycck_cmyk_convert(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
 * RGB565 conversion
 */

-#define PACK_SHORT_565_LE(r, g, b)  ((((r) << 8) & 0xF800) | \
-                                     (((g) << 3) & 0x7E0) | ((b) >> 3))
-#define PACK_SHORT_565_BE(r, g, b)  (((r) & 0xF8) | ((g) >> 5) | \
-                                     (((g) << 11) & 0xE000) | \
-                                     (((b) << 5) & 0x1F00))
+#define PACK_SHORT_565_LE(r, g, b) \
+  ((((r) << 8) & 0xF800) | (((g) << 3) & 0x7E0) | ((b) >> 3))
+#define PACK_SHORT_565_BE(r, g, b) \
+  (((r) & 0xF8) | ((g) >> 5) | (((g) << 11) & 0xE000) | (((b) << 5) & 0x1F00))

 #define PACK_TWO_PIXELS_LE(l, r)    ((r << 16) | l)
 #define PACK_TWO_PIXELS_BE(l, r)    ((l << 16) | r)
@@ -4,7 +4,8 @@
 * This file was part of the Independent JPEG Group's software:
 * Copyright (C) 1991-1997, Thomas G. Lane.
 * libjpeg-turbo Modifications:
- * Copyright (C) 2009-2011, 2016, 2018, D. R. Commander.
+ * Copyright (C) 2009-2011, 2016, 2018-2019, D. R. Commander.
+ * Copyright (C) 2018, Matthias Räncker.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -39,24 +40,6 @@ typedef struct {
  int last_dc_val[MAX_COMPS_IN_SCAN]; /* last DC coef for each component */
 } savable_state;

-/* This macro is to work around compilers with missing or broken
- * structure assignment.  You'll need to fix this code if you have
- * such a compiler and you change MAX_COMPS_IN_SCAN.
- */
-
-#ifndef NO_STRUCT_ASSIGN
-#define ASSIGN_STATE(dest, src)  ((dest) = (src))
-#else
-#if MAX_COMPS_IN_SCAN == 4
-#define ASSIGN_STATE(dest, src) \
-  ((dest).last_dc_val[0] = (src).last_dc_val[0], \
-   (dest).last_dc_val[1] = (src).last_dc_val[1], \
-   (dest).last_dc_val[2] = (src).last_dc_val[2], \
-   (dest).last_dc_val[3] = (src).last_dc_val[3])
-#endif
-#endif
-
-
 typedef struct {
  struct jpeg_entropy_decoder pub; /* public fields */

@@ -325,7 +308,7 @@ jpeg_fill_bit_buffer(bitread_working_state *state,
        bytes_in_buffer = cinfo->src->bytes_in_buffer;
      }
      bytes_in_buffer--;
-      c = GETJOCTET(*next_input_byte++);
+      c = *next_input_byte++;

      /* If it's 0xFF, check and discard stuffed zero byte */
      if (c == 0xFF) {
@@ -342,7 +325,7 @@ jpeg_fill_bit_buffer(bitread_working_state *state,
            bytes_in_buffer = cinfo->src->bytes_in_buffer;
          }
          bytes_in_buffer--;
-          c = GETJOCTET(*next_input_byte++);
+          c = *next_input_byte++;
        } while (c == 0xFF);

        if (c == 0) {
@@ -405,8 +388,8 @@ no_more_bytes:

 #define GET_BYTE { \
  register int c0, c1; \
-  c0 = GETJOCTET(*buffer++); \
-  c1 = GETJOCTET(*buffer); \
+  c0 = *buffer++; \
+  c1 = *buffer; \
  /* Pre-execute most common case */ \
  get_buffer = (get_buffer << 8) | c0; \
  bits_left += 8; \
@@ -423,7 +406,7 @@ no_more_bytes:
  } \
 }

-#if SIZEOF_SIZE_T == 8 || defined(_WIN64)
+#if SIZEOF_SIZE_T == 8 || defined(_WIN64) || (defined(__x86_64__) && defined(__ILP32__))

 /* Pre-fetch 48 bytes, because the holding register is 64-bit */
 #define FILL_BIT_BUFFER_FAST \
@@ -557,6 +540,12 @@ process_restart(j_decompress_ptr cinfo)
 }


+#if defined(__has_feature)
+#if __has_feature(undefined_behavior_sanitizer)
+__attribute__((no_sanitize("signed-integer-overflow"),
+               no_sanitize("unsigned-integer-overflow")))
+#endif
+#endif
 LOCAL(boolean)
 decode_mcu_slow(j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
 {
@@ -568,7 +557,7 @@ decode_mcu_slow(j_decompress_ptr cinfo, JBLOCKROW *MCU_data)

  /* Load up working state */
  BITREAD_LOAD_STATE(cinfo, entropy->bitstate);
-  ASSIGN_STATE(state, entropy->saved);
+  state = entropy->saved;

  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
    JBLOCKROW block = MCU_data ? MCU_data[blkn] : NULL;
@@ -589,6 +578,14 @@ decode_mcu_slow(j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
    if (entropy->dc_needed[blkn]) {
      /* Convert DC difference to actual value, update last_dc_val */
      int ci = cinfo->MCU_membership[blkn];
+      /* Certain malformed JPEG images produce repeated DC coefficient
+       * differences of 2047 or -2047, which causes state.last_dc_val[ci] to
+       * grow until it overflows or underflows a 32-bit signed integer.  This
+       * behavior is, to the best of our understanding, innocuous, and it is
+       * unclear how to work around it without potentially affecting
+       * performance.  Thus, we (hopefully temporarily) suppress UBSan integer
+       * overflow errors for this function and decode_mcu_fast().
+       */
      s += state.last_dc_val[ci];
      state.last_dc_val[ci] = s;
      if (block) {
@@ -649,11 +646,17 @@ decode_mcu_slow(j_decompress_ptr cinfo, JBLOCKROW *MCU_data)

  /* Completed MCU, so update state */
  BITREAD_SAVE_STATE(cinfo, entropy->bitstate);
-  ASSIGN_STATE(entropy->saved, state);
+  entropy->saved = state;
  return TRUE;
 }


+#if defined(__has_feature)
+#if __has_feature(undefined_behavior_sanitizer)
+__attribute__((no_sanitize("signed-integer-overflow"),
+               no_sanitize("unsigned-integer-overflow")))
+#endif
+#endif
 LOCAL(boolean)
 decode_mcu_fast(j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
 {
@@ -667,7 +670,7 @@ decode_mcu_fast(j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
  /* Load up working state */
  BITREAD_LOAD_STATE(cinfo, entropy->bitstate);
  buffer = (JOCTET *)br_state.next_input_byte;
-  ASSIGN_STATE(state, entropy->saved);
+  state = entropy->saved;

  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
    JBLOCKROW block = MCU_data ? MCU_data[blkn] : NULL;
@@ -684,6 +687,9 @@ decode_mcu_fast(j_decompress_ptr cinfo, JBLOCKROW *MCU_data)

    if (entropy->dc_needed[blkn]) {
      int ci = cinfo->MCU_membership[blkn];
+      /* Refer to the comment in decode_mcu_slow() regarding the supression of
+       * a UBSan integer overflow error in this line of code.
+       */
      s += state.last_dc_val[ci];
      state.last_dc_val[ci] = s;
      if (block)
@@ -736,7 +742,7 @@ decode_mcu_fast(j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
  br_state.bytes_in_buffer -= (buffer - br_state.next_input_byte);
  br_state.next_input_byte = buffer;
  BITREAD_SAVE_STATE(cinfo, entropy->bitstate);
-  ASSIGN_STATE(entropy->saved, state);
+  entropy->saved = state;
  return TRUE;
 }

@@ -791,7 +797,8 @@ use_slow:
  }

  /* Account for restart interval (no-op if not using restarts) */
-  entropy->restarts_to_go--;
+  if (cinfo->restart_interval)
+    entropy->restarts_to_go--;

  return TRUE;
 }
@@ -4,7 +4,8 @@
 * This file was part of the Independent JPEG Group's software:
 * Copyright (C) 1991-1997, Thomas G. Lane.
 * libjpeg-turbo Modifications:
- * Copyright (C) 2010-2011, 2015-2016, D. R. Commander.
+ * Copyright (C) 2010-2011, 2015-2016, 2021, D. R. Commander.
+ * Copyright (C) 2018, Matthias Räncker.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -78,6 +79,11 @@ EXTERN(void) jpeg_make_d_derived_tbl(j_decompress_ptr cinfo, boolean isDC,
 typedef size_t bit_buf_type;            /* type of bit-extraction buffer */
 #define BIT_BUF_SIZE  64                /* size of buffer in bits */

+#elif defined(__x86_64__) && defined(__ILP32__)
+
+typedef unsigned long long bit_buf_type; /* type of bit-extraction buffer */
+#define BIT_BUF_SIZE  64                 /* size of buffer in bits */
+
 #else

 typedef unsigned long bit_buf_type;     /* type of bit-extraction buffer */
@@ -228,7 +234,10 @@ slowlabel: \
      s |= GET_BITS(1); \
      nb++; \
    } \
-    s = htbl->pub->huffval[(int)(s + htbl->valoffset[nb]) & 0xFF]; \
+    if (nb > 16) \
+      s = 0; \
+    else \
+      s = htbl->pub->huffval[(int)(s + htbl->valoffset[nb]) & 0xFF]; \
  }

 /* Out-of-line case for Huffman code fetching */
@@ -38,18 +38,18 @@ marker_is_icc(jpeg_saved_marker_ptr marker)
    marker->marker == ICC_MARKER &&
    marker->data_length >= ICC_OVERHEAD_LEN &&
    /* verify the identifying string */
-    GETJOCTET(marker->data[0]) == 0x49 &&
-    GETJOCTET(marker->data[1]) == 0x43 &&
-    GETJOCTET(marker->data[2]) == 0x43 &&
-    GETJOCTET(marker->data[3]) == 0x5F &&
-    GETJOCTET(marker->data[4]) == 0x50 &&
-    GETJOCTET(marker->data[5]) == 0x52 &&
-    GETJOCTET(marker->data[6]) == 0x4F &&
-    GETJOCTET(marker->data[7]) == 0x46 &&
-    GETJOCTET(marker->data[8]) == 0x49 &&
-    GETJOCTET(marker->data[9]) == 0x4C &&
-    GETJOCTET(marker->data[10]) == 0x45 &&
-    GETJOCTET(marker->data[11]) == 0x0;
+    marker->data[0] == 0x49 &&
+    marker->data[1] == 0x43 &&
+    marker->data[2] == 0x43 &&
+    marker->data[3] == 0x5F &&
+    marker->data[4] == 0x50 &&
+    marker->data[5] == 0x52 &&
+    marker->data[6] == 0x4F &&
+    marker->data[7] == 0x46 &&
+    marker->data[8] == 0x49 &&
+    marker->data[9] == 0x4C &&
+    marker->data[10] == 0x45 &&
+    marker->data[11] == 0x0;
 }


@@ -102,12 +102,12 @@ jpeg_read_icc_profile(j_decompress_ptr cinfo, JOCTET **icc_data_ptr,
  for (marker = cinfo->marker_list; marker != NULL; marker = marker->next) {
    if (marker_is_icc(marker)) {
      if (num_markers == 0)
-        num_markers = GETJOCTET(marker->data[13]);
-      else if (num_markers != GETJOCTET(marker->data[13])) {
+        num_markers = marker->data[13];
+      else if (num_markers != marker->data[13]) {
        WARNMS(cinfo, JWRN_BOGUS_ICC);  /* inconsistent num_markers fields */
        return FALSE;
      }
-      seq_no = GETJOCTET(marker->data[12]);
+      seq_no = marker->data[12];
      if (seq_no <= 0 || seq_no > num_markers) {
        WARNMS(cinfo, JWRN_BOGUS_ICC);  /* bogus sequence number */
        return FALSE;
@@ -154,7 +154,7 @@ jpeg_read_icc_profile(j_decompress_ptr cinfo, JOCTET **icc_data_ptr,
      JOCTET FAR *src_ptr;
      JOCTET *dst_ptr;
      unsigned int length;
-      seq_no = GETJOCTET(marker->data[12]);
+      seq_no = marker->data[12];
      dst_ptr = icc_data + data_offset[seq_no];
      src_ptr = marker->data + ICC_OVERHEAD_LEN;
      length = data_length[seq_no];
@@ -18,6 +18,7 @@

 #include "jinclude.h"
 #include "jdmainct.h"
+#include "jconfigint.h"


 /*
@@ -360,7 +361,7 @@ process_data_context_main(j_decompress_ptr cinfo, JSAMPARRAY output_buf,
    main_ptr->context_state = CTX_PREPARE_FOR_IMCU;
    if (*out_row_ctr >= out_rows_avail)
      return;                   /* Postprocessor exactly filled output buf */
-    /*FALLTHROUGH*/
+    FALLTHROUGH                 /*FALLTHROUGH*/
  case CTX_PREPARE_FOR_IMCU:
    /* Prepare to process first M-1 row groups of this iMCU row */
    main_ptr->rowgroup_ctr = 0;
@@ -371,7 +372,7 @@ process_data_context_main(j_decompress_ptr cinfo, JSAMPARRAY output_buf,
    if (main_ptr->iMCU_row_ctr == cinfo->total_iMCU_rows)
      set_bottom_pointers(cinfo);
    main_ptr->context_state = CTX_PROCESS_IMCU;
-    /*FALLTHROUGH*/
+    FALLTHROUGH                 /*FALLTHROUGH*/
  case CTX_PROCESS_IMCU:
    /* Call postprocessor using previously set pointers */
    (*cinfo->post->post_process_data) (cinfo,
@@ -151,7 +151,7 @@ typedef my_marker_reader *my_marker_ptr;
 #define INPUT_BYTE(cinfo, V, action) \
  MAKESTMT( MAKE_BYTE_AVAIL(cinfo, action); \
            bytes_in_buffer--; \
-            V = GETJOCTET(*next_input_byte++); )
+            V = *next_input_byte++; )

 /* As above, but read two bytes interpreted as an unsigned 16-bit integer.
 * V should be declared unsigned int or perhaps JLONG.
@@ -159,10 +159,10 @@ typedef my_marker_reader *my_marker_ptr;
 #define INPUT_2BYTES(cinfo, V, action) \
  MAKESTMT( MAKE_BYTE_AVAIL(cinfo, action); \
            bytes_in_buffer--; \
-            V = ((unsigned int)GETJOCTET(*next_input_byte++)) << 8; \
+            V = ((unsigned int)(*next_input_byte++)) << 8; \
            MAKE_BYTE_AVAIL(cinfo, action); \
            bytes_in_buffer--; \
-            V += GETJOCTET(*next_input_byte++); )
+            V += *next_input_byte++; )


 /*
@@ -608,18 +608,18 @@ examine_app0(j_decompress_ptr cinfo, JOCTET *data, unsigned int datalen,
  JLONG totallen = (JLONG)datalen + remaining;

  if (datalen >= APP0_DATA_LEN &&
-      GETJOCTET(data[0]) == 0x4A &&
-      GETJOCTET(data[1]) == 0x46 &&
-      GETJOCTET(data[2]) == 0x49 &&
-      GETJOCTET(data[3]) == 0x46 &&
-      GETJOCTET(data[4]) == 0) {
+      data[0] == 0x4A &&
+      data[1] == 0x46 &&
+      data[2] == 0x49 &&
+      data[3] == 0x46 &&
+      data[4] == 0) {
    /* Found JFIF APP0 marker: save info */
    cinfo->saw_JFIF_marker = TRUE;
-    cinfo->JFIF_major_version = GETJOCTET(data[5]);
-    cinfo->JFIF_minor_version = GETJOCTET(data[6]);
-    cinfo->density_unit = GETJOCTET(data[7]);
-    cinfo->X_density = (GETJOCTET(data[8]) << 8) + GETJOCTET(data[9]);
-    cinfo->Y_density = (GETJOCTET(data[10]) << 8) + GETJOCTET(data[11]);
+    cinfo->JFIF_major_version = data[5];
+    cinfo->JFIF_minor_version = data[6];
+    cinfo->density_unit = data[7];
+    cinfo->X_density = (data[8] << 8) + data[9];
+    cinfo->Y_density = (data[10] << 8) + data[11];
    /* Check version.
     * Major version must be 1, anything else signals an incompatible change.
     * (We used to treat this as an error, but now it's a nonfatal warning,
@@ -634,24 +634,22 @@ examine_app0(j_decompress_ptr cinfo, JOCTET *data, unsigned int datalen,
             cinfo->JFIF_major_version, cinfo->JFIF_minor_version,
             cinfo->X_density, cinfo->Y_density, cinfo->density_unit);
    /* Validate thumbnail dimensions and issue appropriate messages */
-    if (GETJOCTET(data[12]) | GETJOCTET(data[13]))
-      TRACEMS2(cinfo, 1, JTRC_JFIF_THUMBNAIL,
-               GETJOCTET(data[12]), GETJOCTET(data[13]));
+    if (data[12] | data[13])
+      TRACEMS2(cinfo, 1, JTRC_JFIF_THUMBNAIL, data[12], data[13]);
    totallen -= APP0_DATA_LEN;
-    if (totallen !=
-        ((JLONG)GETJOCTET(data[12]) * (JLONG)GETJOCTET(data[13]) * (JLONG)3))
+    if (totallen != ((JLONG)data[12] * (JLONG)data[13] * (JLONG)3))
      TRACEMS1(cinfo, 1, JTRC_JFIF_BADTHUMBNAILSIZE, (int)totallen);
  } else if (datalen >= 6 &&
-             GETJOCTET(data[0]) == 0x4A &&
-             GETJOCTET(data[1]) == 0x46 &&
-             GETJOCTET(data[2]) == 0x58 &&
-             GETJOCTET(data[3]) == 0x58 &&
-             GETJOCTET(data[4]) == 0) {
+             data[0] == 0x4A &&
+             data[1] == 0x46 &&
+             data[2] == 0x58 &&
+             data[3] == 0x58 &&
+             data[4] == 0) {
    /* Found JFIF "JFXX" extension APP0 marker */
    /* The library doesn't actually do anything with these,
     * but we try to produce a helpful trace message.
     */
-    switch (GETJOCTET(data[5])) {
+    switch (data[5]) {
    case 0x10:
      TRACEMS1(cinfo, 1, JTRC_THUMB_JPEG, (int)totallen);
      break;
@@ -662,8 +660,7 @@ examine_app0(j_decompress_ptr cinfo, JOCTET *data, unsigned int datalen,
      TRACEMS1(cinfo, 1, JTRC_THUMB_RGB, (int)totallen);
      break;
    default:
-      TRACEMS2(cinfo, 1, JTRC_JFIF_EXTENSION,
-               GETJOCTET(data[5]), (int)totallen);
+      TRACEMS2(cinfo, 1, JTRC_JFIF_EXTENSION, data[5], (int)totallen);
      break;
    }
  } else {
@@ -684,16 +681,16 @@ examine_app14(j_decompress_ptr cinfo, JOCTET *data, unsigned int datalen,
  unsigned int version, flags0, flags1, transform;

  if (datalen >= APP14_DATA_LEN &&
-      GETJOCTET(data[0]) == 0x41 &&
-      GETJOCTET(data[1]) == 0x64 &&
-      GETJOCTET(data[2]) == 0x6F &&
-      GETJOCTET(data[3]) == 0x62 &&
-      GETJOCTET(data[4]) == 0x65) {
+      data[0] == 0x41 &&
+      data[1] == 0x64 &&
+      data[2] == 0x6F &&
+      data[3] == 0x62 &&
+      data[4] == 0x65) {
    /* Found Adobe APP14 marker */
-    version = (GETJOCTET(data[5]) << 8) + GETJOCTET(data[6]);
-    flags0 = (GETJOCTET(data[7]) << 8) + GETJOCTET(data[8]);
-    flags1 = (GETJOCTET(data[9]) << 8) + GETJOCTET(data[10]);
-    transform = GETJOCTET(data[11]);
+    version = (data[5] << 8) + data[6];
+    flags0 = (data[7] << 8) + data[8];
+    flags1 = (data[9] << 8) + data[10];
+    transform = data[11];
    TRACEMS4(cinfo, 1, JTRC_ADOBE, version, flags0, flags1, transform);
    cinfo->saw_Adobe_marker = TRUE;
    cinfo->Adobe_transform = (UINT8)transform;
@@ -5,7 +5,7 @@
 * Copyright (C) 1991-1997, Thomas G. Lane.
 * Modified 2002-2009 by Guido Vollbeding.
 * libjpeg-turbo Modifications:
- * Copyright (C) 2009-2011, 2016, D. R. Commander.
+ * Copyright (C) 2009-2011, 2016, 2019, D. R. Commander.
 * Copyright (C) 2013, Linaro Limited.
 * Copyright (C) 2015, Google, Inc.
 * For conditions of distribution and use, see the accompanying README.ijg
@@ -22,7 +22,6 @@
 #include "jpeglib.h"
 #include "jpegcomp.h"
 #include "jdmaster.h"
-#include "jsimd.h"


 /*
@@ -70,17 +69,6 @@ use_merged_upsample(j_decompress_ptr cinfo)
      cinfo->comp_info[1]._DCT_scaled_size != cinfo->_min_DCT_scaled_size ||
      cinfo->comp_info[2]._DCT_scaled_size != cinfo->_min_DCT_scaled_size)
    return FALSE;
-#ifdef WITH_SIMD
-  /* If YCbCr-to-RGB color conversion is SIMD-accelerated but merged upsampling
-     isn't, then disabling merged upsampling is likely to be faster when
-     decompressing YCbCr JPEG images. */
-  if (!jsimd_can_h2v2_merged_upsample() && !jsimd_can_h2v1_merged_upsample() &&
-      jsimd_can_ycc_rgb() && cinfo->jpeg_color_space == JCS_YCbCr &&
-      (cinfo->out_color_space == JCS_RGB ||
-       (cinfo->out_color_space >= JCS_EXT_RGB &&
-        cinfo->out_color_space <= JCS_EXT_ARGB)))
-    return FALSE;
-#endif
  /* ??? also need to test for upsample-time rescaling, when & if supported */
  return TRUE;                  /* by golly, it'll work... */
 #else
@@ -580,6 +568,7 @@ master_selection(j_decompress_ptr cinfo)
   */
  cinfo->master->first_iMCU_col = 0;
  cinfo->master->last_iMCU_col = cinfo->MCUs_per_row - 1;
+  cinfo->master->last_good_iMCU_row = 0;

 #ifdef D_MULTISCAN_FILES_SUPPORTED
  /* If jpeg_start_decompress will read the whole file, initialize
@@ -5,7 +5,7 @@
 * Copyright (C) 1994-1996, Thomas G. Lane.
 * libjpeg-turbo Modifications:
 * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
- * Copyright (C) 2009, 2011, 2014-2015, D. R. Commander.
+ * Copyright (C) 2009, 2011, 2014-2015, 2020, D. R. Commander.
 * Copyright (C) 2013, Linaro Limited.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
@@ -40,41 +40,13 @@
 #define JPEG_INTERNALS
 #include "jinclude.h"
 #include "jpeglib.h"
+#include "jdmerge.h"
 #include "jsimd.h"
 #include "jconfigint.h"

 #ifdef UPSAMPLE_MERGING_SUPPORTED


-/* Private subobject */
-
-typedef struct {
-  struct jpeg_upsampler pub;    /* public fields */
-
-  /* Pointer to routine to do actual upsampling/conversion of one row group */
-  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 */
-  JLONG *Cr_g_tab;              /* => table for Cr to G conversion */
-  JLONG *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        ((JLONG)1 << (SCALEBITS - 1))
 #define FIX(x)          ((JLONG)((x) * (1L << SCALEBITS) + 0.5))
@@ -189,7 +161,7 @@ typedef my_upsampler *my_upsample_ptr;
 LOCAL(void)
 build_ycc_rgb_table(j_decompress_ptr cinfo)
 {
-  my_upsample_ptr upsample = (my_upsample_ptr)cinfo->upsample;
+  my_merged_upsample_ptr upsample = (my_merged_upsample_ptr)cinfo->upsample;
  int i;
  JLONG x;
  SHIFT_TEMPS
@@ -232,7 +204,7 @@ build_ycc_rgb_table(j_decompress_ptr cinfo)
 METHODDEF(void)
 start_pass_merged_upsample(j_decompress_ptr cinfo)
 {
-  my_upsample_ptr upsample = (my_upsample_ptr)cinfo->upsample;
+  my_merged_upsample_ptr upsample = (my_merged_upsample_ptr)cinfo->upsample;

  /* Mark the spare buffer empty */
  upsample->spare_full = FALSE;
@@ -254,7 +226,7 @@ merged_2v_upsample(j_decompress_ptr cinfo, JSAMPIMAGE input_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;
+  my_merged_upsample_ptr upsample = (my_merged_upsample_ptr)cinfo->upsample;
  JSAMPROW work_ptrs[2];
  JDIMENSION num_rows;          /* number of rows returned to caller */

@@ -305,7 +277,7 @@ merged_1v_upsample(j_decompress_ptr cinfo, JSAMPIMAGE input_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;
+  my_merged_upsample_ptr upsample = (my_merged_upsample_ptr)cinfo->upsample;

  /* Just do the upsampling. */
  (*upsample->upmethod) (cinfo, input_buf, *in_row_group_ctr,
@@ -420,17 +392,14 @@ h2v2_merged_upsample(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
 * RGB565 conversion
 */

-#define PACK_SHORT_565_LE(r, g, b)  ((((r) << 8) & 0xF800) | \
-                                     (((g) << 3) & 0x7E0) | ((b) >> 3))
-#define PACK_SHORT_565_BE(r, g, b)  (((r) & 0xF8) | ((g) >> 5) | \
-                                     (((g) << 11) & 0xE000) | \
-                                     (((b) << 5) & 0x1F00))
+#define PACK_SHORT_565_LE(r, g, b) \
+  ((((r) << 8) & 0xF800) | (((g) << 3) & 0x7E0) | ((b) >> 3))
+#define PACK_SHORT_565_BE(r, g, b) \
+  (((r) & 0xF8) | ((g) >> 5) | (((g) << 11) & 0xE000) | (((b) << 5) & 0x1F00))

 #define PACK_TWO_PIXELS_LE(l, r)    ((r << 16) | l)
 #define PACK_TWO_PIXELS_BE(l, r)    ((l << 16) | r)

-#define PACK_NEED_ALIGNMENT(ptr)    (((size_t)(ptr)) & 3)
-
 #define WRITE_TWO_PIXELS_LE(addr, pixels) { \
  ((INT16 *)(addr))[0] = (INT16)(pixels); \
  ((INT16 *)(addr))[1] = (INT16)((pixels) >> 16); \
@@ -568,11 +537,11 @@ h2v2_merged_upsample_565D(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
 GLOBAL(void)
 jinit_merged_upsampler(j_decompress_ptr cinfo)
 {
-  my_upsample_ptr upsample;
+  my_merged_upsample_ptr upsample;

-  upsample = (my_upsample_ptr)
+  upsample = (my_merged_upsample_ptr)
    (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
-                                sizeof(my_upsampler));
+                                sizeof(my_merged_upsampler));
  cinfo->upsample = (struct jpeg_upsampler *)upsample;
  upsample->pub.start_pass = start_pass_merged_upsample;
  upsample->pub.need_context_rows = FALSE;
@@ -0,0 +1,47 @@
+/*
+ * jdmerge.h
+ *
+ * This file was part of the Independent JPEG Group's software:
+ * Copyright (C) 1994-1996, Thomas G. Lane.
+ * libjpeg-turbo Modifications:
+ * Copyright (C) 2020, D. R. Commander.
+ * For conditions of distribution and use, see the accompanying README.ijg
+ * file.
+ */
+
+#define JPEG_INTERNALS
+#include "jpeglib.h"
+
+#ifdef UPSAMPLE_MERGING_SUPPORTED
+
+
+/* Private subobject */
+
+typedef struct {
+  struct jpeg_upsampler pub;    /* public fields */
+
+  /* Pointer to routine to do actual upsampling/conversion of one row group */
+  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 */
+  JLONG *Cr_g_tab;              /* => table for Cr to G conversion */
+  JLONG *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_merged_upsampler;
+
+typedef my_merged_upsampler *my_merged_upsample_ptr;
+
+#endif /* UPSAMPLE_MERGING_SUPPORTED */
@@ -5,7 +5,7 @@
 * Copyright (C) 1994-1996, Thomas G. Lane.
 * libjpeg-turbo Modifications:
 * Copyright (C) 2013, Linaro Limited.
- * Copyright (C) 2014-2015, 2018, D. R. Commander.
+ * Copyright (C) 2014-2015, 2018, 2020, D. R. Commander.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -19,7 +19,7 @@ h2v1_merged_upsample_565_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
                                  JDIMENSION in_row_group_ctr,
                                  JSAMPARRAY output_buf)
 {
-  my_upsample_ptr upsample = (my_upsample_ptr)cinfo->upsample;
+  my_merged_upsample_ptr upsample = (my_merged_upsample_ptr)cinfo->upsample;
  register int y, cred, cgreen, cblue;
  int cb, cr;
  register JSAMPROW outptr;
@@ -43,20 +43,20 @@ h2v1_merged_upsample_565_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
  /* 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++);
+    cb = *inptr1++;
+    cr = *inptr2++;
    cred = Crrtab[cr];
    cgreen = (int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS);
    cblue = Cbbtab[cb];

    /* Fetch 2 Y values and emit 2 pixels */
-    y  = GETJSAMPLE(*inptr0++);
+    y  = *inptr0++;
    r = range_limit[y + cred];
    g = range_limit[y + cgreen];
    b = range_limit[y + cblue];
    rgb = PACK_SHORT_565(r, g, b);

-    y  = GETJSAMPLE(*inptr0++);
+    y  = *inptr0++;
    r = range_limit[y + cred];
    g = range_limit[y + cgreen];
    b = range_limit[y + cblue];
@@ -68,12 +68,12 @@ h2v1_merged_upsample_565_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,

  /* If image width is odd, do the last output column separately */
  if (cinfo->output_width & 1) {
-    cb = GETJSAMPLE(*inptr1);
-    cr = GETJSAMPLE(*inptr2);
+    cb = *inptr1;
+    cr = *inptr2;
    cred = Crrtab[cr];
    cgreen = (int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS);
    cblue = Cbbtab[cb];
-    y  = GETJSAMPLE(*inptr0);
+    y  = *inptr0;
    r = range_limit[y + cred];
    g = range_limit[y + cgreen];
    b = range_limit[y + cblue];
@@ -90,7 +90,7 @@ h2v1_merged_upsample_565D_internal(j_decompress_ptr cinfo,
                                   JDIMENSION in_row_group_ctr,
                                   JSAMPARRAY output_buf)
 {
-  my_upsample_ptr upsample = (my_upsample_ptr)cinfo->upsample;
+  my_merged_upsample_ptr upsample = (my_merged_upsample_ptr)cinfo->upsample;
  register int y, cred, cgreen, cblue;
  int cb, cr;
  register JSAMPROW outptr;
@@ -115,21 +115,21 @@ h2v1_merged_upsample_565D_internal(j_decompress_ptr cinfo,
  /* 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++);
+    cb = *inptr1++;
+    cr = *inptr2++;
    cred = Crrtab[cr];
    cgreen = (int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS);
    cblue = Cbbtab[cb];

    /* Fetch 2 Y values and emit 2 pixels */
-    y  = GETJSAMPLE(*inptr0++);
+    y  = *inptr0++;
    r = range_limit[DITHER_565_R(y + cred, d0)];
    g = range_limit[DITHER_565_G(y + cgreen, d0)];
    b = range_limit[DITHER_565_B(y + cblue, d0)];
    d0 = DITHER_ROTATE(d0);
    rgb = PACK_SHORT_565(r, g, b);

-    y  = GETJSAMPLE(*inptr0++);
+    y  = *inptr0++;
    r = range_limit[DITHER_565_R(y + cred, d0)];
    g = range_limit[DITHER_565_G(y + cgreen, d0)];
    b = range_limit[DITHER_565_B(y + cblue, d0)];
@@ -142,12 +142,12 @@ h2v1_merged_upsample_565D_internal(j_decompress_ptr cinfo,

  /* If image width is odd, do the last output column separately */
  if (cinfo->output_width & 1) {
-    cb = GETJSAMPLE(*inptr1);
-    cr = GETJSAMPLE(*inptr2);
+    cb = *inptr1;
+    cr = *inptr2;
    cred = Crrtab[cr];
    cgreen = (int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS);
    cblue = Cbbtab[cb];
-    y  = GETJSAMPLE(*inptr0);
+    y  = *inptr0;
    r = range_limit[DITHER_565_R(y + cred, d0)];
    g = range_limit[DITHER_565_G(y + cgreen, d0)];
    b = range_limit[DITHER_565_B(y + cblue, d0)];
@@ -163,7 +163,7 @@ h2v2_merged_upsample_565_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
                                  JDIMENSION in_row_group_ctr,
                                  JSAMPARRAY output_buf)
 {
-  my_upsample_ptr upsample = (my_upsample_ptr)cinfo->upsample;
+  my_merged_upsample_ptr upsample = (my_merged_upsample_ptr)cinfo->upsample;
  register int y, cred, cgreen, cblue;
  int cb, cr;
  register JSAMPROW outptr0, outptr1;
@@ -189,20 +189,20 @@ h2v2_merged_upsample_565_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
  /* 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++);
+    cb = *inptr1++;
+    cr = *inptr2++;
    cred = Crrtab[cr];
    cgreen = (int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS);
    cblue = Cbbtab[cb];

    /* Fetch 4 Y values and emit 4 pixels */
-    y  = GETJSAMPLE(*inptr00++);
+    y  = *inptr00++;
    r = range_limit[y + cred];
    g = range_limit[y + cgreen];
    b = range_limit[y + cblue];
    rgb = PACK_SHORT_565(r, g, b);

-    y  = GETJSAMPLE(*inptr00++);
+    y  = *inptr00++;
    r = range_limit[y + cred];
    g = range_limit[y + cgreen];
    b = range_limit[y + cblue];
@@ -211,13 +211,13 @@ h2v2_merged_upsample_565_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
    WRITE_TWO_PIXELS(outptr0, rgb);
    outptr0 += 4;

-    y  = GETJSAMPLE(*inptr01++);
+    y  = *inptr01++;
    r = range_limit[y + cred];
    g = range_limit[y + cgreen];
    b = range_limit[y + cblue];
    rgb = PACK_SHORT_565(r, g, b);

-    y  = GETJSAMPLE(*inptr01++);
+    y  = *inptr01++;
    r = range_limit[y + cred];
    g = range_limit[y + cgreen];
    b = range_limit[y + cblue];
@@ -229,20 +229,20 @@ h2v2_merged_upsample_565_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,

  /* If image width is odd, do the last output column separately */
  if (cinfo->output_width & 1) {
-    cb = GETJSAMPLE(*inptr1);
-    cr = GETJSAMPLE(*inptr2);
+    cb = *inptr1;
+    cr = *inptr2;
    cred = Crrtab[cr];
    cgreen = (int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS);
    cblue = Cbbtab[cb];

-    y  = GETJSAMPLE(*inptr00);
+    y  = *inptr00;
    r = range_limit[y + cred];
    g = range_limit[y + cgreen];
    b = range_limit[y + cblue];
    rgb = PACK_SHORT_565(r, g, b);
    *(INT16 *)outptr0 = (INT16)rgb;

-    y  = GETJSAMPLE(*inptr01);
+    y  = *inptr01;
    r = range_limit[y + cred];
    g = range_limit[y + cgreen];
    b = range_limit[y + cblue];
@@ -259,7 +259,7 @@ h2v2_merged_upsample_565D_internal(j_decompress_ptr cinfo,
                                   JDIMENSION in_row_group_ctr,
                                   JSAMPARRAY output_buf)
 {
-  my_upsample_ptr upsample = (my_upsample_ptr)cinfo->upsample;
+  my_merged_upsample_ptr upsample = (my_merged_upsample_ptr)cinfo->upsample;
  register int y, cred, cgreen, cblue;
  int cb, cr;
  register JSAMPROW outptr0, outptr1;
@@ -287,21 +287,21 @@ h2v2_merged_upsample_565D_internal(j_decompress_ptr cinfo,
  /* 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++);
+    cb = *inptr1++;
+    cr = *inptr2++;
    cred = Crrtab[cr];
    cgreen = (int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS);
    cblue = Cbbtab[cb];

    /* Fetch 4 Y values and emit 4 pixels */
-    y  = GETJSAMPLE(*inptr00++);
+    y  = *inptr00++;
    r = range_limit[DITHER_565_R(y + cred, d0)];
    g = range_limit[DITHER_565_G(y + cgreen, d0)];
    b = range_limit[DITHER_565_B(y + cblue, d0)];
    d0 = DITHER_ROTATE(d0);
    rgb = PACK_SHORT_565(r, g, b);

-    y  = GETJSAMPLE(*inptr00++);
+    y  = *inptr00++;
    r = range_limit[DITHER_565_R(y + cred, d0)];
    g = range_limit[DITHER_565_G(y + cgreen, d0)];
    b = range_limit[DITHER_565_B(y + cblue, d0)];
@@ -311,14 +311,14 @@ h2v2_merged_upsample_565D_internal(j_decompress_ptr cinfo,
    WRITE_TWO_PIXELS(outptr0, rgb);
    outptr0 += 4;

-    y  = GETJSAMPLE(*inptr01++);
+    y  = *inptr01++;
    r = range_limit[DITHER_565_R(y + cred, d1)];
    g = range_limit[DITHER_565_G(y + cgreen, d1)];
    b = range_limit[DITHER_565_B(y + cblue, d1)];
    d1 = DITHER_ROTATE(d1);
    rgb = PACK_SHORT_565(r, g, b);

-    y  = GETJSAMPLE(*inptr01++);
+    y  = *inptr01++;
    r = range_limit[DITHER_565_R(y + cred, d1)];
    g = range_limit[DITHER_565_G(y + cgreen, d1)];
    b = range_limit[DITHER_565_B(y + cblue, d1)];
@@ -331,20 +331,20 @@ h2v2_merged_upsample_565D_internal(j_decompress_ptr cinfo,

  /* If image width is odd, do the last output column separately */
  if (cinfo->output_width & 1) {
-    cb = GETJSAMPLE(*inptr1);
-    cr = GETJSAMPLE(*inptr2);
+    cb = *inptr1;
+    cr = *inptr2;
    cred = Crrtab[cr];
    cgreen = (int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS);
    cblue = Cbbtab[cb];

-    y  = GETJSAMPLE(*inptr00);
+    y  = *inptr00;
    r = range_limit[DITHER_565_R(y + cred, d0)];
    g = range_limit[DITHER_565_G(y + cgreen, d0)];
    b = range_limit[DITHER_565_B(y + cblue, d0)];
    rgb = PACK_SHORT_565(r, g, b);
    *(INT16 *)outptr0 = (INT16)rgb;

-    y  = GETJSAMPLE(*inptr01);
+    y  = *inptr01;
    r = range_limit[DITHER_565_R(y + cred, d1)];
    g = range_limit[DITHER_565_G(y + cgreen, d1)];
    b = range_limit[DITHER_565_B(y + cblue, d1)];
@@ -4,7 +4,7 @@
 * This file was part of the Independent JPEG Group's software:
 * Copyright (C) 1994-1996, Thomas G. Lane.
 * libjpeg-turbo Modifications:
- * Copyright (C) 2011, 2015, D. R. Commander.
+ * Copyright (C) 2011, 2015, 2020, D. R. Commander.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -25,7 +25,7 @@ h2v1_merged_upsample_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
                              JDIMENSION in_row_group_ctr,
                              JSAMPARRAY output_buf)
 {
-  my_upsample_ptr upsample = (my_upsample_ptr)cinfo->upsample;
+  my_merged_upsample_ptr upsample = (my_merged_upsample_ptr)cinfo->upsample;
  register int y, cred, cgreen, cblue;
  int cb, cr;
  register JSAMPROW outptr;
@@ -46,13 +46,13 @@ h2v1_merged_upsample_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
  /* 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++);
+    cb = *inptr1++;
+    cr = *inptr2++;
    cred = Crrtab[cr];
    cgreen = (int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS);
    cblue = Cbbtab[cb];
    /* Fetch 2 Y values and emit 2 pixels */
-    y  = GETJSAMPLE(*inptr0++);
+    y  = *inptr0++;
    outptr[RGB_RED] =   range_limit[y + cred];
    outptr[RGB_GREEN] = range_limit[y + cgreen];
    outptr[RGB_BLUE] =  range_limit[y + cblue];
@@ -60,7 +60,7 @@ h2v1_merged_upsample_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
    outptr[RGB_ALPHA] = 0xFF;
 #endif
    outptr += RGB_PIXELSIZE;
-    y  = GETJSAMPLE(*inptr0++);
+    y  = *inptr0++;
    outptr[RGB_RED] =   range_limit[y + cred];
    outptr[RGB_GREEN] = range_limit[y + cgreen];
    outptr[RGB_BLUE] =  range_limit[y + cblue];
@@ -71,12 +71,12 @@ h2v1_merged_upsample_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
  }
  /* If image width is odd, do the last output column separately */
  if (cinfo->output_width & 1) {
-    cb = GETJSAMPLE(*inptr1);
-    cr = GETJSAMPLE(*inptr2);
+    cb = *inptr1;
+    cr = *inptr2;
    cred = Crrtab[cr];
    cgreen = (int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS);
    cblue = Cbbtab[cb];
-    y  = GETJSAMPLE(*inptr0);
+    y  = *inptr0;
    outptr[RGB_RED] =   range_limit[y + cred];
    outptr[RGB_GREEN] = range_limit[y + cgreen];
    outptr[RGB_BLUE] =  range_limit[y + cblue];
@@ -97,7 +97,7 @@ h2v2_merged_upsample_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
                              JDIMENSION in_row_group_ctr,
                              JSAMPARRAY output_buf)
 {
-  my_upsample_ptr upsample = (my_upsample_ptr)cinfo->upsample;
+  my_merged_upsample_ptr upsample = (my_merged_upsample_ptr)cinfo->upsample;
  register int y, cred, cgreen, cblue;
  int cb, cr;
  register JSAMPROW outptr0, outptr1;
@@ -120,13 +120,13 @@ h2v2_merged_upsample_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
  /* 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++);
+    cb = *inptr1++;
+    cr = *inptr2++;
    cred = Crrtab[cr];
    cgreen = (int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS);
    cblue = Cbbtab[cb];
    /* Fetch 4 Y values and emit 4 pixels */
-    y  = GETJSAMPLE(*inptr00++);
+    y  = *inptr00++;
    outptr0[RGB_RED] =   range_limit[y + cred];
    outptr0[RGB_GREEN] = range_limit[y + cgreen];
    outptr0[RGB_BLUE] =  range_limit[y + cblue];
@@ -134,7 +134,7 @@ h2v2_merged_upsample_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
    outptr0[RGB_ALPHA] = 0xFF;
 #endif
    outptr0 += RGB_PIXELSIZE;
-    y  = GETJSAMPLE(*inptr00++);
+    y  = *inptr00++;
    outptr0[RGB_RED] =   range_limit[y + cred];
    outptr0[RGB_GREEN] = range_limit[y + cgreen];
    outptr0[RGB_BLUE] =  range_limit[y + cblue];
@@ -142,7 +142,7 @@ h2v2_merged_upsample_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
    outptr0[RGB_ALPHA] = 0xFF;
 #endif
    outptr0 += RGB_PIXELSIZE;
-    y  = GETJSAMPLE(*inptr01++);
+    y  = *inptr01++;
    outptr1[RGB_RED] =   range_limit[y + cred];
    outptr1[RGB_GREEN] = range_limit[y + cgreen];
    outptr1[RGB_BLUE] =  range_limit[y + cblue];
@@ -150,7 +150,7 @@ h2v2_merged_upsample_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
    outptr1[RGB_ALPHA] = 0xFF;
 #endif
    outptr1 += RGB_PIXELSIZE;
-    y  = GETJSAMPLE(*inptr01++);
+    y  = *inptr01++;
    outptr1[RGB_RED] =   range_limit[y + cred];
    outptr1[RGB_GREEN] = range_limit[y + cgreen];
    outptr1[RGB_BLUE] =  range_limit[y + cblue];
@@ -161,19 +161,19 @@ h2v2_merged_upsample_internal(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
  }
  /* If image width is odd, do the last output column separately */
  if (cinfo->output_width & 1) {
-    cb = GETJSAMPLE(*inptr1);
-    cr = GETJSAMPLE(*inptr2);
+    cb = *inptr1;
+    cr = *inptr2;
    cred = Crrtab[cr];
    cgreen = (int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS);
    cblue = Cbbtab[cb];
-    y  = GETJSAMPLE(*inptr00);
+    y  = *inptr00;
    outptr0[RGB_RED] =   range_limit[y + cred];
    outptr0[RGB_GREEN] = range_limit[y + cgreen];
    outptr0[RGB_BLUE] =  range_limit[y + cblue];
 #ifdef RGB_ALPHA
    outptr0[RGB_ALPHA] = 0xFF;
 #endif
-    y  = GETJSAMPLE(*inptr01);
+    y  = *inptr01;
    outptr1[RGB_RED] =   range_limit[y + cred];
    outptr1[RGB_GREEN] = range_limit[y + cgreen];
    outptr1[RGB_BLUE] =  range_limit[y + cblue];
@@ -4,7 +4,7 @@
 * This file was part of the Independent JPEG Group's software:
 * Copyright (C) 1995-1997, Thomas G. Lane.
 * libjpeg-turbo Modifications:
- * Copyright (C) 2015-2016, 2018, D. R. Commander.
+ * Copyright (C) 2015-2016, 2018-2021, D. R. Commander.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -41,25 +41,6 @@ typedef struct {
  int last_dc_val[MAX_COMPS_IN_SCAN];   /* last DC coef for each component */
 } savable_state;

-/* This macro is to work around compilers with missing or broken
- * structure assignment.  You'll need to fix this code if you have
- * such a compiler and you change MAX_COMPS_IN_SCAN.
- */
-
-#ifndef NO_STRUCT_ASSIGN
-#define ASSIGN_STATE(dest, src)  ((dest) = (src))
-#else
-#if MAX_COMPS_IN_SCAN == 4
-#define ASSIGN_STATE(dest, src) \
-  ((dest).EOBRUN = (src).EOBRUN, \
-   (dest).last_dc_val[0] = (src).last_dc_val[0], \
-   (dest).last_dc_val[1] = (src).last_dc_val[1], \
-   (dest).last_dc_val[2] = (src).last_dc_val[2], \
-   (dest).last_dc_val[3] = (src).last_dc_val[3])
-#endif
-#endif
-
-
 typedef struct {
  struct jpeg_entropy_decoder pub; /* public fields */

@@ -102,7 +83,7 @@ start_pass_phuff_decoder(j_decompress_ptr cinfo)
  boolean is_DC_band, bad;
  int ci, coefi, tbl;
  d_derived_tbl **pdtbl;
-  int *coef_bit_ptr;
+  int *coef_bit_ptr, *prev_coef_bit_ptr;
  jpeg_component_info *compptr;

  is_DC_band = (cinfo->Ss == 0);
@@ -143,8 +124,15 @@ start_pass_phuff_decoder(j_decompress_ptr cinfo)
  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
    int cindex = cinfo->cur_comp_info[ci]->component_index;
    coef_bit_ptr = &cinfo->coef_bits[cindex][0];
+    prev_coef_bit_ptr = &cinfo->coef_bits[cindex + cinfo->num_components][0];
    if (!is_DC_band && coef_bit_ptr[0] < 0) /* AC without prior DC scan */
      WARNMS2(cinfo, JWRN_BOGUS_PROGRESSION, cindex, 0);
+    for (coefi = MIN(cinfo->Ss, 1); coefi <= MAX(cinfo->Se, 9); coefi++) {
+      if (cinfo->input_scan_number > 1)
+        prev_coef_bit_ptr[coefi] = coef_bit_ptr[coefi];
+      else
+        prev_coef_bit_ptr[coefi] = 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)
@@ -323,7 +311,7 @@ decode_mcu_DC_first(j_decompress_ptr cinfo, JBLOCKROW *MCU_data)

    /* Load up working state */
    BITREAD_LOAD_STATE(cinfo, entropy->bitstate);
-    ASSIGN_STATE(state, entropy->saved);
+    state = entropy->saved;

    /* Outer loop handles each block in the MCU */

@@ -356,11 +344,12 @@ decode_mcu_DC_first(j_decompress_ptr cinfo, JBLOCKROW *MCU_data)

    /* Completed MCU, so update state */
    BITREAD_SAVE_STATE(cinfo, entropy->bitstate);
-    ASSIGN_STATE(entropy->saved, state);
+    entropy->saved = state;
  }

  /* Account for restart interval (no-op if not using restarts) */
-  entropy->restarts_to_go--;
+  if (cinfo->restart_interval)
+    entropy->restarts_to_go--;

  return TRUE;
 }
@@ -444,7 +433,8 @@ decode_mcu_AC_first(j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
  }

  /* Account for restart interval (no-op if not using restarts) */
-  entropy->restarts_to_go--;
+  if (cinfo->restart_interval)
+    entropy->restarts_to_go--;

  return TRUE;
 }
@@ -495,7 +485,8 @@ decode_mcu_DC_refine(j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
  BITREAD_SAVE_STATE(cinfo, entropy->bitstate);

  /* Account for restart interval (no-op if not using restarts) */
-  entropy->restarts_to_go--;
+  if (cinfo->restart_interval)
+    entropy->restarts_to_go--;

  return TRUE;
 }
@@ -638,7 +629,8 @@ decode_mcu_AC_refine(j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
  }

  /* Account for restart interval (no-op if not using restarts) */
-  entropy->restarts_to_go--;
+  if (cinfo->restart_interval)
+    entropy->restarts_to_go--;

  return TRUE;

@@ -676,7 +668,7 @@ jinit_phuff_decoder(j_decompress_ptr cinfo)
  /* Create progression status table */
  cinfo->coef_bits = (int (*)[DCTSIZE2])
    (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
-                                cinfo->num_components * DCTSIZE2 *
+                                cinfo->num_components * 2 * DCTSIZE2 *
                                sizeof(int));
  coef_bit_ptr = &cinfo->coef_bits[0][0];
  for (ci = 0; ci < cinfo->num_components; ci++)
@@ -8,6 +8,7 @@
 * Copyright (C) 2010, 2015-2016, D. R. Commander.
 * Copyright (C) 2014, MIPS Technologies, Inc., California.
 * Copyright (C) 2015, Google, Inc.
+ * Copyright (C) 2019-2020, Arm Limited.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -176,7 +177,7 @@ int_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr,
    outptr = output_data[outrow];
    outend = outptr + cinfo->output_width;
    while (outptr < outend) {
-      invalue = *inptr++;       /* don't need GETJSAMPLE() here */
+      invalue = *inptr++;
      for (h = h_expand; h > 0; h--) {
        *outptr++ = invalue;
      }
@@ -212,7 +213,7 @@ h2v1_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr,
    outptr = output_data[inrow];
    outend = outptr + cinfo->output_width;
    while (outptr < outend) {
-      invalue = *inptr++;       /* don't need GETJSAMPLE() here */
+      invalue = *inptr++;
      *outptr++ = invalue;
      *outptr++ = invalue;
    }
@@ -241,7 +242,7 @@ h2v2_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr,
    outptr = output_data[outrow];
    outend = outptr + cinfo->output_width;
    while (outptr < outend) {
-      invalue = *inptr++;       /* don't need GETJSAMPLE() here */
+      invalue = *inptr++;
      *outptr++ = invalue;
      *outptr++ = invalue;
    }
@@ -282,20 +283,20 @@ h2v1_fancy_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr,
    inptr = input_data[inrow];
    outptr = output_data[inrow];
    /* Special case for first column */
-    invalue = GETJSAMPLE(*inptr++);
+    invalue = *inptr++;
    *outptr++ = (JSAMPLE)invalue;
-    *outptr++ = (JSAMPLE)((invalue * 3 + GETJSAMPLE(*inptr) + 2) >> 2);
+    *outptr++ = (JSAMPLE)((invalue * 3 + inptr[0] + 2) >> 2);

    for (colctr = compptr->downsampled_width - 2; colctr > 0; colctr--) {
      /* General case: 3/4 * nearer pixel + 1/4 * further pixel */
-      invalue = GETJSAMPLE(*inptr++) * 3;
-      *outptr++ = (JSAMPLE)((invalue + GETJSAMPLE(inptr[-2]) + 1) >> 2);
-      *outptr++ = (JSAMPLE)((invalue + GETJSAMPLE(*inptr) + 2) >> 2);
+      invalue = (*inptr++) * 3;
+      *outptr++ = (JSAMPLE)((invalue + inptr[-2] + 1) >> 2);
+      *outptr++ = (JSAMPLE)((invalue + inptr[0] + 2) >> 2);
    }

    /* Special case for last column */
-    invalue = GETJSAMPLE(*inptr);
-    *outptr++ = (JSAMPLE)((invalue * 3 + GETJSAMPLE(inptr[-1]) + 1) >> 2);
+    invalue = *inptr;
+    *outptr++ = (JSAMPLE)((invalue * 3 + inptr[-1] + 1) >> 2);
    *outptr++ = (JSAMPLE)invalue;
  }
 }
@@ -315,9 +316,9 @@ h1v2_fancy_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr,
  JSAMPARRAY output_data = *output_data_ptr;
  JSAMPROW inptr0, inptr1, outptr;
 #if BITS_IN_JSAMPLE == 8
-  int thiscolsum;
+  int thiscolsum, bias;
 #else
-  JLONG thiscolsum;
+  JLONG thiscolsum, bias;
 #endif
  JDIMENSION colctr;
  int inrow, outrow, v;
@@ -327,15 +328,18 @@ h1v2_fancy_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr,
    for (v = 0; v < 2; v++) {
      /* inptr0 points to nearest input row, inptr1 points to next nearest */
      inptr0 = input_data[inrow];
-      if (v == 0)               /* next nearest is row above */
+      if (v == 0) {             /* next nearest is row above */
        inptr1 = input_data[inrow - 1];
-      else                      /* next nearest is row below */
+        bias = 1;
+      } else {                  /* next nearest is row below */
        inptr1 = input_data[inrow + 1];
+        bias = 2;
+      }
      outptr = output_data[outrow++];

      for (colctr = 0; colctr < compptr->downsampled_width; colctr++) {
-        thiscolsum = GETJSAMPLE(*inptr0++) * 3 + GETJSAMPLE(*inptr1++);
-        *outptr++ = (JSAMPLE)((thiscolsum + 1) >> 2);
+        thiscolsum = (*inptr0++) * 3 + (*inptr1++);
+        *outptr++ = (JSAMPLE)((thiscolsum + bias) >> 2);
      }
    }
    inrow++;
@@ -377,8 +381,8 @@ h2v2_fancy_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr,
      outptr = output_data[outrow++];

      /* Special case for first column */
-      thiscolsum = GETJSAMPLE(*inptr0++) * 3 + GETJSAMPLE(*inptr1++);
-      nextcolsum = GETJSAMPLE(*inptr0++) * 3 + GETJSAMPLE(*inptr1++);
+      thiscolsum = (*inptr0++) * 3 + (*inptr1++);
+      nextcolsum = (*inptr0++) * 3 + (*inptr1++);
      *outptr++ = (JSAMPLE)((thiscolsum * 4 + 8) >> 4);
      *outptr++ = (JSAMPLE)((thiscolsum * 3 + nextcolsum + 7) >> 4);
      lastcolsum = thiscolsum;  thiscolsum = nextcolsum;
@@ -386,7 +390,7 @@ h2v2_fancy_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr,
      for (colctr = compptr->downsampled_width - 2; colctr > 0; colctr--) {
        /* General case: 3/4 * nearer pixel + 1/4 * further pixel in each */
        /* dimension, thus 9/16, 3/16, 3/16, 1/16 overall */
-        nextcolsum = GETJSAMPLE(*inptr0++) * 3 + GETJSAMPLE(*inptr1++);
+        nextcolsum = (*inptr0++) * 3 + (*inptr1++);
        *outptr++ = (JSAMPLE)((thiscolsum * 3 + lastcolsum + 8) >> 4);
        *outptr++ = (JSAMPLE)((thiscolsum * 3 + nextcolsum + 7) >> 4);
        lastcolsum = thiscolsum;  thiscolsum = nextcolsum;
@@ -473,7 +477,13 @@ jinit_upsampler(j_decompress_ptr cinfo)
    } else if (h_in_group == h_out_group &&
               v_in_group * 2 == v_out_group && do_fancy) {
      /* Non-fancy upsampling is handled by the generic method */
-      upsample->methods[ci] = h1v2_fancy_upsample;
+#if defined(__arm__) || defined(__aarch64__) || \
+    defined(_M_ARM) || defined(_M_ARM64)
+      if (jsimd_can_h1v2_fancy_upsample())
+        upsample->methods[ci] = jsimd_h1v2_fancy_upsample;
+      else
+#endif
+        upsample->methods[ci] = h1v2_fancy_upsample;
      upsample->pub.need_context_rows = TRUE;
    } else if (h_in_group * 2 == h_out_group &&
               v_in_group * 2 == v_out_group) {
@@ -3,8 +3,8 @@
 *
 * This file was part of the Independent JPEG Group's software:
 * Copyright (C) 1995-1997, Thomas G. Lane.
- * It was modified by The libjpeg-turbo Project to include only code relevant
- * to libjpeg-turbo.
+ * libjpeg-turbo Modifications:
+ * Copyright (C) 2020, D. R. Commander.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -16,6 +16,7 @@
 #define JPEG_INTERNALS
 #include "jinclude.h"
 #include "jpeglib.h"
+#include "jpegcomp.h"


 /* Forward declarations */
@@ -207,6 +207,10 @@ JMESSAGE(JWRN_ARITH_BAD_CODE, "Corrupt JPEG data: bad arithmetic code")
 #endif
 #endif
 JMESSAGE(JWRN_BOGUS_ICC, "Corrupt JPEG data: bad ICC marker")
+#if JPEG_LIB_VERSION < 70
+JMESSAGE(JERR_BAD_DROP_SAMPLING,
+         "Component index %d: mismatching sampling ratio %d:%d, %d:%d, %c")
+#endif

 #ifdef JMAKE_ENUM_LIST

@@ -252,6 +256,15 @@ JMESSAGE(JWRN_BOGUS_ICC, "Corrupt JPEG data: bad ICC marker")
   (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), \
@@ -1,14 +1,14 @@
 /*
 * jfdctint.c
 *
- * This file was part of the Independent JPEG Group's software.
+ * This file was part of the Independent JPEG Group's software:
 * Copyright (C) 1991-1996, Thomas G. Lane.
 * libjpeg-turbo Modifications:
- * Copyright (C) 2015, D. R. Commander.
+ * Copyright (C) 2015, 2020, D. R. Commander.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
- * This file contains a slow-but-accurate integer implementation of the
+ * This file contains a slower but more 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
@@ -1,15 +1,15 @@
 /*
 * jidctint.c
 *
- * This file was part of the Independent JPEG Group's software.
+ * This file was part of the Independent JPEG Group's software:
 * Copyright (C) 1991-1998, Thomas G. Lane.
- * Modification developed 2002-2009 by Guido Vollbeding.
+ * Modification developed 2002-2018 by Guido Vollbeding.
 * libjpeg-turbo Modifications:
- * Copyright (C) 2015, D. R. Commander.
+ * Copyright (C) 2015, 2020, D. R. Commander.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
- * This file contains a slow-but-accurate integer implementation of the
+ * This file contains a slower but more accurate integer implementation of the
 * inverse DCT (Discrete Cosine Transform).  In the IJG code, this routine
 * must also perform dequantization of the input coefficients.
 *
@@ -417,7 +417,7 @@ jpeg_idct_islow(j_decompress_ptr cinfo, jpeg_component_info *compptr,

 /*
 * Perform dequantization and inverse DCT on one block of coefficients,
- * producing a 7x7 output block.
+ * producing a reduced-size 7x7 output block.
 *
 * Optimized algorithm with 12 multiplications in the 1-D kernel.
 * cK represents sqrt(2) * cos(K*pi/14).
@@ -1258,7 +1258,7 @@ jpeg_idct_10x10(j_decompress_ptr cinfo, jpeg_component_info *compptr,

 /*
 * Perform dequantization and inverse DCT on one block of coefficients,
- * producing a 11x11 output block.
+ * producing an 11x11 output block.
 *
 * Optimized algorithm with 24 multiplications in the 1-D kernel.
 * cK represents sqrt(2) * cos(K*pi/22).
@@ -2398,7 +2398,7 @@ jpeg_idct_16x16(j_decompress_ptr cinfo, jpeg_component_info *compptr,
    tmp0 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]);
    tmp0 = LEFT_SHIFT(tmp0, CONST_BITS);
    /* Add fudge factor here for final descale. */
-    tmp0 += 1 << (CONST_BITS - PASS1_BITS - 1);
+    tmp0 += ONE << (CONST_BITS - PASS1_BITS - 1);

    z1 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]);
    tmp1 = MULTIPLY(z1, FIX(1.306562965));      /* c4[16] = c2[8] */
@@ -1,7 +1,7 @@
 /*
 * jidctred.c
 *
- * This file was part of the Independent JPEG Group's software.
+ * This file was part of the Independent JPEG Group's software:
 * Copyright (C) 1994-1998, Thomas G. Lane.
 * libjpeg-turbo Modifications:
 * Copyright (C) 2015, D. R. Commander.
@@ -4,7 +4,7 @@
 * This file was part of the Independent JPEG Group's software:
 * Copyright (C) 1991-1997, Thomas G. Lane.
 * libjpeg-turbo Modifications:
- * Copyright (C) 2016, D. R. Commander.
+ * Copyright (C) 2016, 2021, D. R. Commander.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -1032,7 +1032,7 @@ free_pool(j_common_ptr cinfo, int pool_id)
    large_pool_ptr next_lhdr_ptr = lhdr_ptr->next;
    space_freed = lhdr_ptr->bytes_used +
                  lhdr_ptr->bytes_left +
-                  sizeof(large_pool_hdr);
+                  sizeof(large_pool_hdr) + ALIGN_SIZE - 1;
    jpeg_free_large(cinfo, (void *)lhdr_ptr, space_freed);
    mem->total_space_allocated -= space_freed;
    lhdr_ptr = next_lhdr_ptr;
@@ -1045,7 +1045,7 @@ free_pool(j_common_ptr cinfo, int pool_id)
  while (shdr_ptr != NULL) {
    small_pool_ptr next_shdr_ptr = shdr_ptr->next;
    space_freed = shdr_ptr->bytes_used + shdr_ptr->bytes_left +
-                  sizeof(small_pool_hdr);
+                  sizeof(small_pool_hdr) + ALIGN_SIZE - 1;
    jpeg_free_small(cinfo, (void *)shdr_ptr, space_freed);
    mem->total_space_allocated -= space_freed;
    shdr_ptr = next_shdr_ptr;
@@ -5,7 +5,7 @@
 * Copyright (C) 1991-1997, Thomas G. Lane.
 * Modified 1997-2009 by Guido Vollbeding.
 * libjpeg-turbo Modifications:
- * Copyright (C) 2009, 2011, 2014-2015, 2018, D. R. Commander.
+ * Copyright (C) 2009, 2011, 2014-2015, 2018, 2020, D. R. Commander.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -43,25 +43,11 @@

 #if BITS_IN_JSAMPLE == 8
 /* JSAMPLE should be the smallest type that will hold the values 0..255.
- * You can use a signed char by having GETJSAMPLE mask it with 0xFF.
 */

-#ifdef HAVE_UNSIGNED_CHAR
-
 typedef unsigned char JSAMPLE;
 #define GETJSAMPLE(value)  ((int)(value))

-#else /* not HAVE_UNSIGNED_CHAR */
-
-typedef char JSAMPLE;
-#ifdef __CHAR_UNSIGNED__
-#define GETJSAMPLE(value)  ((int)(value))
-#else
-#define GETJSAMPLE(value)  ((int)(value) & 0xFF)
-#endif /* __CHAR_UNSIGNED__ */
-
-#endif /* HAVE_UNSIGNED_CHAR */
-
 #define MAXJSAMPLE      255
 #define CENTERJSAMPLE   128

@@ -97,22 +83,9 @@ typedef short JCOEF;
 * managers, this is also the data type passed to fread/fwrite.
 */

-#ifdef HAVE_UNSIGNED_CHAR
-
 typedef unsigned char JOCTET;
 #define GETJOCTET(value)  (value)

-#else /* not HAVE_UNSIGNED_CHAR */
-
-typedef char JOCTET;
-#ifdef __CHAR_UNSIGNED__
-#define GETJOCTET(value)  (value)
-#else
-#define GETJOCTET(value)  ((value) & 0xFF)
-#endif /* __CHAR_UNSIGNED__ */
-
-#endif /* HAVE_UNSIGNED_CHAR */
-

 /* These typedefs are used for various table entries and so forth.
 * They must be at least as wide as specified; but making them too big
@@ -123,15 +96,7 @@ typedef char JOCTET;

 /* UINT8 must hold at least the values 0..255. */

-#ifdef HAVE_UNSIGNED_CHAR
 typedef unsigned char UINT8;
-#else /* not HAVE_UNSIGNED_CHAR */
-#ifdef __CHAR_UNSIGNED__
-typedef char UINT8;
-#else /* not __CHAR_UNSIGNED__ */
-typedef short UINT8;
-#endif /* __CHAR_UNSIGNED__ */
-#endif /* HAVE_UNSIGNED_CHAR */

 /* UINT16 must hold at least the values 0..65535. */

@@ -273,9 +238,9 @@ typedef int boolean;

 /* Capability options common to encoder and decoder: */

-#define DCT_ISLOW_SUPPORTED     /* slow but accurate integer algorithm */
-#define DCT_IFAST_SUPPORTED     /* faster, less accurate integer method */
-#define DCT_FLOAT_SUPPORTED     /* floating-point: accurate, fast on fast HW */
+#define DCT_ISLOW_SUPPORTED     /* accurate integer method */
+#define DCT_IFAST_SUPPORTED     /* less accurate int method [legacy feature] */
+#define DCT_FLOAT_SUPPORTED     /* floating-point method [legacy feature] */

 /* Encoder capability options: */

@@ -1,7 +1,7 @@
 /*
 * jpegcomp.h
 *
- * Copyright (C) 2010, D. R. Commander.
+ * Copyright (C) 2010, 2020, D. R. Commander.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -19,6 +19,7 @@
 #define _min_DCT_v_scaled_size  min_DCT_v_scaled_size
 #define _jpeg_width  jpeg_width
 #define _jpeg_height  jpeg_height
+#define JERR_ARITH_NOTIMPL  JERR_NOT_COMPILED
 #else
 #define _DCT_scaled_size  DCT_scaled_size
 #define _DCT_h_scaled_size  DCT_scaled_size
@@ -5,8 +5,9 @@
 * Copyright (C) 1991-1997, Thomas G. Lane.
 * Modified 1997-2009 by Guido Vollbeding.
 * libjpeg-turbo Modifications:
- * Copyright (C) 2015-2016, D. R. Commander.
+ * Copyright (C) 2015-2016, 2019, 2021, D. R. Commander.
 * Copyright (C) 2015, Google, Inc.
+ * Copyright (C) 2021, Alex Richardson.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -47,6 +48,18 @@ typedef enum {            /* Operating modes for buffer controllers */
 /* JLONG must hold at least signed 32-bit values. */
 typedef long JLONG;

+/* JUINTPTR must hold pointer values. */
+#ifdef __UINTPTR_TYPE__
+/*
+ * __UINTPTR_TYPE__ is GNU-specific and available in GCC 4.6+ and Clang 3.0+.
+ * Fortunately, that is sufficient to support the few architectures for which
+ * sizeof(void *) != sizeof(size_t).  The only other options would require C99
+ * or Clang-specific builtins.
+ */
+typedef __UINTPTR_TYPE__ JUINTPTR;
+#else
+typedef size_t JUINTPTR;
+#endif

 /*
 * Left shift macro that handles a negative operand without causing any
@@ -158,6 +171,9 @@ struct jpeg_decomp_master {
  JDIMENSION first_MCU_col[MAX_COMPONENTS];
  JDIMENSION last_MCU_col[MAX_COMPONENTS];
  boolean jinit_upsampler_no_alloc;
+
+  /* Last iMCU row that was successfully decoded */
+  JDIMENSION last_good_iMCU_row;
 };

 /* Input control module */
@@ -5,7 +5,7 @@
 * Copyright (C) 1991-1998, Thomas G. Lane.
 * Modified 2002-2009 by Guido Vollbeding.
 * libjpeg-turbo Modifications:
- * Copyright (C) 2009-2011, 2013-2014, 2016-2017, D. R. Commander.
+ * Copyright (C) 2009-2011, 2013-2014, 2016-2017, 2020, D. R. Commander.
 * Copyright (C) 2015, Google, Inc.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
@@ -244,9 +244,9 @@ typedef enum {
 /* DCT/IDCT algorithm options. */

 typedef enum {
-  JDCT_ISLOW,             /* slow but accurate integer algorithm */
-  JDCT_IFAST,             /* faster, less accurate integer method */
-  JDCT_FLOAT              /* floating-point: accurate, fast on fast HW */
+  JDCT_ISLOW,             /* accurate integer method */
+  JDCT_IFAST,             /* less accurate integer method [legacy feature] */
+  JDCT_FLOAT              /* floating-point method [legacy feature] */
 } J_DCT_METHOD;

 #ifndef JDCT_DEFAULT            /* may be overridden in jconfig.h */
@@ -479,7 +479,7 @@ color_quantize(j_decompress_ptr cinfo, JSAMPARRAY input_buf,
    for (col = width; col > 0; col--) {
      pixcode = 0;
      for (ci = 0; ci < nc; ci++) {
-        pixcode += GETJSAMPLE(colorindex[ci][GETJSAMPLE(*ptrin++)]);
+        pixcode += colorindex[ci][*ptrin++];
      }
      *ptrout++ = (JSAMPLE)pixcode;
    }
@@ -506,9 +506,9 @@ color_quantize3(j_decompress_ptr cinfo, JSAMPARRAY input_buf,
    ptrin = input_buf[row];
    ptrout = output_buf[row];
    for (col = width; col > 0; col--) {
-      pixcode  = GETJSAMPLE(colorindex0[GETJSAMPLE(*ptrin++)]);
-      pixcode += GETJSAMPLE(colorindex1[GETJSAMPLE(*ptrin++)]);
-      pixcode += GETJSAMPLE(colorindex2[GETJSAMPLE(*ptrin++)]);
+      pixcode  = colorindex0[*ptrin++];
+      pixcode += colorindex1[*ptrin++];
+      pixcode += colorindex2[*ptrin++];
      *ptrout++ = (JSAMPLE)pixcode;
    }
  }
@@ -552,7 +552,7 @@ quantize_ord_dither(j_decompress_ptr cinfo, JSAMPARRAY input_buf,
         * required amount of padding.
         */
        *output_ptr +=
-          colorindex_ci[GETJSAMPLE(*input_ptr) + dither[col_index]];
+          colorindex_ci[*input_ptr + dither[col_index]];
        input_ptr += nc;
        output_ptr++;
        col_index = (col_index + 1) & ODITHER_MASK;
@@ -595,12 +595,9 @@ quantize3_ord_dither(j_decompress_ptr cinfo, JSAMPARRAY input_buf,
    col_index = 0;

    for (col = width; col > 0; col--) {
-      pixcode  =
-        GETJSAMPLE(colorindex0[GETJSAMPLE(*input_ptr++) + dither0[col_index]]);
-      pixcode +=
-        GETJSAMPLE(colorindex1[GETJSAMPLE(*input_ptr++) + dither1[col_index]]);
-      pixcode +=
-        GETJSAMPLE(colorindex2[GETJSAMPLE(*input_ptr++) + dither2[col_index]]);
+      pixcode  = colorindex0[(*input_ptr++) + dither0[col_index]];
+      pixcode += colorindex1[(*input_ptr++) + dither1[col_index]];
+      pixcode += colorindex2[(*input_ptr++) + dither2[col_index]];
      *output_ptr++ = (JSAMPLE)pixcode;
      col_index = (col_index + 1) & ODITHER_MASK;
    }
@@ -677,15 +674,15 @@ quantize_fs_dither(j_decompress_ptr cinfo, JSAMPARRAY input_buf,
         * The maximum error is +- MAXJSAMPLE; this sets the required size
         * of the range_limit array.
         */
-        cur += GETJSAMPLE(*input_ptr);
-        cur = GETJSAMPLE(range_limit[cur]);
+        cur += *input_ptr;
+        cur = range_limit[cur];
        /* Select output value, accumulate into output code for this pixel */
-        pixcode = GETJSAMPLE(colorindex_ci[cur]);
+        pixcode = colorindex_ci[cur];
        *output_ptr += (JSAMPLE)pixcode;
        /* Compute actual representation error at this pixel */
        /* Note: we can do this even though we don't have the final */
        /* pixel code, because the colormap is orthogonal. */
-        cur -= GETJSAMPLE(colormap_ci[pixcode]);
+        cur -= colormap_ci[pixcode];
        /* Compute error fractions to be propagated to adjacent pixels.
         * Add these into the running sums, and simultaneously shift the
         * next-line error sums left by 1 column.
@@ -4,7 +4,7 @@
 * This file was part of the Independent JPEG Group's software:
 * Copyright (C) 1991-1996, Thomas G. Lane.
 * libjpeg-turbo Modifications:
- * Copyright (C) 2009, 2014-2015, D. R. Commander.
+ * Copyright (C) 2009, 2014-2015, 2020, D. R. Commander.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -215,9 +215,9 @@ prescan_quantize(j_decompress_ptr cinfo, JSAMPARRAY input_buf,
    ptr = input_buf[row];
    for (col = width; col > 0; col--) {
      /* get pixel value and index into the histogram */
-      histp = &histogram[GETJSAMPLE(ptr[0]) >> C0_SHIFT]
-                        [GETJSAMPLE(ptr[1]) >> C1_SHIFT]
-                        [GETJSAMPLE(ptr[2]) >> C2_SHIFT];
+      histp = &histogram[ptr[0] >> C0_SHIFT]
+                        [ptr[1] >> C1_SHIFT]
+                        [ptr[2] >> C2_SHIFT];
      /* increment, check for overflow and undo increment if so. */
      if (++(*histp) <= 0)
        (*histp)--;
@@ -665,7 +665,7 @@ find_nearby_colors(j_decompress_ptr cinfo, int minc0, int minc1, int minc2,

  for (i = 0; i < numcolors; i++) {
    /* We compute the squared-c0-distance term, then add in the other two. */
-    x = GETJSAMPLE(cinfo->colormap[0][i]);
+    x = cinfo->colormap[0][i];
    if (x < minc0) {
      tdist = (x - minc0) * C0_SCALE;
      min_dist = tdist * tdist;
@@ -688,7 +688,7 @@ find_nearby_colors(j_decompress_ptr cinfo, int minc0, int minc1, int minc2,
      }
    }

-    x = GETJSAMPLE(cinfo->colormap[1][i]);
+    x = cinfo->colormap[1][i];
    if (x < minc1) {
      tdist = (x - minc1) * C1_SCALE;
      min_dist += tdist * tdist;
@@ -710,7 +710,7 @@ find_nearby_colors(j_decompress_ptr cinfo, int minc0, int minc1, int minc2,
      }
    }

-    x = GETJSAMPLE(cinfo->colormap[2][i]);
+    x = cinfo->colormap[2][i];
    if (x < minc2) {
      tdist = (x - minc2) * C2_SCALE;
      min_dist += tdist * tdist;
@@ -788,13 +788,13 @@ find_best_colors(j_decompress_ptr cinfo, int minc0, int minc1, int minc2,
 #define STEP_C2  ((1 << C2_SHIFT) * C2_SCALE)

  for (i = 0; i < numcolors; i++) {
-    icolor = GETJSAMPLE(colorlist[i]);
+    icolor = colorlist[i];
    /* Compute (square of) distance from minc0/c1/c2 to this color */
-    inc0 = (minc0 - GETJSAMPLE(cinfo->colormap[0][icolor])) * C0_SCALE;
+    inc0 = (minc0 - cinfo->colormap[0][icolor]) * C0_SCALE;
    dist0 = inc0 * inc0;
-    inc1 = (minc1 - GETJSAMPLE(cinfo->colormap[1][icolor])) * C1_SCALE;
+    inc1 = (minc1 - cinfo->colormap[1][icolor]) * C1_SCALE;
    dist0 += inc1 * inc1;
-    inc2 = (minc2 - GETJSAMPLE(cinfo->colormap[2][icolor])) * C2_SCALE;
+    inc2 = (minc2 - cinfo->colormap[2][icolor]) * C2_SCALE;
    dist0 += inc2 * inc2;
    /* Form the initial difference increments */
    inc0 = inc0 * (2 * STEP_C0) + STEP_C0 * STEP_C0;
@@ -879,7 +879,7 @@ fill_inverse_cmap(j_decompress_ptr cinfo, int c0, int c1, int c2)
    for (ic1 = 0; ic1 < BOX_C1_ELEMS; ic1++) {
      cachep = &histogram[c0 + ic0][c1 + ic1][c2];
      for (ic2 = 0; ic2 < BOX_C2_ELEMS; ic2++) {
-        *cachep++ = (histcell)(GETJSAMPLE(*cptr++) + 1);
+        *cachep++ = (histcell)((*cptr++) + 1);
      }
    }
  }
@@ -909,9 +909,9 @@ pass2_no_dither(j_decompress_ptr cinfo, JSAMPARRAY input_buf,
    outptr = output_buf[row];
    for (col = width; col > 0; col--) {
      /* get pixel value and index into the cache */
-      c0 = GETJSAMPLE(*inptr++) >> C0_SHIFT;
-      c1 = GETJSAMPLE(*inptr++) >> C1_SHIFT;
-      c2 = GETJSAMPLE(*inptr++) >> C2_SHIFT;
+      c0 = (*inptr++) >> C0_SHIFT;
+      c1 = (*inptr++) >> C1_SHIFT;
+      c2 = (*inptr++) >> C2_SHIFT;
      cachep = &histogram[c0][c1][c2];
      /* If we have not seen this color before, find nearest colormap entry */
      /* and update the cache */
@@ -996,12 +996,12 @@ pass2_fs_dither(j_decompress_ptr cinfo, JSAMPARRAY input_buf,
       * The maximum error is +- MAXJSAMPLE (or less with error limiting);
       * this sets the required size of the range_limit array.
       */
-      cur0 += GETJSAMPLE(inptr[0]);
-      cur1 += GETJSAMPLE(inptr[1]);
-      cur2 += GETJSAMPLE(inptr[2]);
-      cur0 = GETJSAMPLE(range_limit[cur0]);
-      cur1 = GETJSAMPLE(range_limit[cur1]);
-      cur2 = GETJSAMPLE(range_limit[cur2]);
+      cur0 += inptr[0];
+      cur1 += inptr[1];
+      cur2 += inptr[2];
+      cur0 = range_limit[cur0];
+      cur1 = range_limit[cur1];
+      cur2 = range_limit[cur2];
      /* Index into the cache with adjusted pixel value */
      cachep =
        &histogram[cur0 >> C0_SHIFT][cur1 >> C1_SHIFT][cur2 >> C2_SHIFT];
@@ -1015,9 +1015,9 @@ pass2_fs_dither(j_decompress_ptr cinfo, JSAMPARRAY input_buf,
        register int pixcode = *cachep - 1;
        *outptr = (JSAMPLE)pixcode;
        /* Compute representation error for this pixel */
-        cur0 -= GETJSAMPLE(colormap0[pixcode]);
-        cur1 -= GETJSAMPLE(colormap1[pixcode]);
-        cur2 -= GETJSAMPLE(colormap2[pixcode]);
+        cur0 -= colormap0[pixcode];
+        cur1 -= colormap1[pixcode];
+        cur2 -= colormap2[pixcode];
      }
      /* Compute error fractions to be propagated to adjacent pixels.
       * Add these into the running sums, and simultaneously shift the
@@ -1145,7 +1145,7 @@ start_pass_2_quant(j_decompress_ptr cinfo, boolean is_pre_scan)
  int i;

  /* Only F-S dithering or no dithering is supported. */
-  /* If user asks for ordered dither, give him F-S. */
+  /* If user asks for ordered dither, give them F-S. */
  if (cinfo->dither_mode != JDITHER_NONE)
    cinfo->dither_mode = JDITHER_FS;

@@ -1263,7 +1263,7 @@ jinit_2pass_quantizer(j_decompress_ptr cinfo)
    cquantize->sv_colormap = NULL;

  /* Only F-S dithering or no dithering is supported. */
-  /* If user asks for ordered dither, give him F-S. */
+  /* If user asks for ordered dither, give them F-S. */
  if (cinfo->dither_mode != JDITHER_NONE)
    cinfo->dither_mode = JDITHER_FS;

@@ -4,6 +4,7 @@
 * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
 * Copyright (C) 2011, 2014, D. R. Commander.
 * Copyright (C) 2015-2016, 2018, Matthieu Darbois.
+ * Copyright (C) 2020, Arm Limited.
 *
 * Based on the x86 SIMD extension for IJG JPEG library,
 * Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -75,6 +76,7 @@ EXTERN(void) jsimd_int_upsample(j_decompress_ptr cinfo,

 EXTERN(int) jsimd_can_h2v2_fancy_upsample(void);
 EXTERN(int) jsimd_can_h2v1_fancy_upsample(void);
+EXTERN(int) jsimd_can_h1v2_fancy_upsample(void);

 EXTERN(void) jsimd_h2v2_fancy_upsample(j_decompress_ptr cinfo,
                                       jpeg_component_info *compptr,
@@ -84,6 +86,10 @@ EXTERN(void) jsimd_h2v1_fancy_upsample(j_decompress_ptr cinfo,
                                       jpeg_component_info *compptr,
                                       JSAMPARRAY input_data,
                                       JSAMPARRAY *output_data_ptr);
+EXTERN(void) jsimd_h1v2_fancy_upsample(j_decompress_ptr cinfo,
+                                       jpeg_component_info *compptr,
+                                       JSAMPARRAY input_data,
+                                       JSAMPARRAY *output_data_ptr);

 EXTERN(int) jsimd_can_h2v2_merged_upsample(void);
 EXTERN(int) jsimd_can_h2v1_merged_upsample(void);
@@ -4,6 +4,7 @@
 * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
 * Copyright (C) 2009-2011, 2014, D. R. Commander.
 * Copyright (C) 2015-2016, 2018, Matthieu Darbois.
+ * Copyright (C) 2020, Arm Limited.
 *
 * Based on the x86 SIMD extension for IJG JPEG library,
 * Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -169,6 +170,12 @@ jsimd_can_h2v1_fancy_upsample(void)
  return 0;
 }

+GLOBAL(int)
+jsimd_can_h1v2_fancy_upsample(void)
+{
+  return 0;
+}
+
 GLOBAL(void)
 jsimd_h2v2_fancy_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr,
                          JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr)
@@ -181,6 +188,12 @@ jsimd_h2v1_fancy_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr,
 {
 }

+GLOBAL(void)
+jsimd_h1v2_fancy_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr,
+                          JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr)
+{
+}
+
 GLOBAL(int)
 jsimd_can_h2v2_merged_upsample(void)
 {
@@ -2,9 +2,9 @@
 * jversion.h
 *
 * This file was part of the Independent JPEG Group's software:
- * Copyright (C) 1991-2012, Thomas G. Lane, Guido Vollbeding.
+ * Copyright (C) 1991-2020, Thomas G. Lane, Guido Vollbeding.
 * libjpeg-turbo Modifications:
- * Copyright (C) 2010, 2012-2019, D. R. Commander.
+ * Copyright (C) 2010, 2012-2021, D. R. Commander.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -30,23 +30,25 @@
 * NOTE: It is our convention to place the authors in the following order:
 * - libjpeg-turbo authors (2009-) in descending order of the date of their
 *   most recent contribution to the project, then in ascending order of the
- *   date of their first contribution to the project
+ *   date of their first contribution to the project, then in alphabetical
+ *   order
 * - Upstream authors in descending order of the date of the first inclusion of
 *   their code
 */

 #define JCOPYRIGHT \
-  "Copyright (C) 2009-2019 D. R. Commander\n" \
-  "Copyright (C) 2011-2016 Siarhei Siamashka\n" \
+  "Copyright (C) 2009-2021 D. R. Commander\n" \
+  "Copyright (C) 2015, 2020 Google, Inc.\n" \
+  "Copyright (C) 2019-2020 Arm Limited\n" \
  "Copyright (C) 2015-2016, 2018 Matthieu Darbois\n" \
+  "Copyright (C) 2011-2016 Siarhei Siamashka\n" \
  "Copyright (C) 2015 Intel Corporation\n" \
-  "Copyright (C) 2015 Google, Inc.\n" \
+  "Copyright (C) 2013-2014 Linaro Limited\n" \
  "Copyright (C) 2013-2014 MIPS Technologies, Inc.\n" \
-  "Copyright (C) 2013 Linaro Limited\n" \
+  "Copyright (C) 2009, 2012 Pierre Ossman for Cendio AB\n" \
  "Copyright (C) 2009-2011 Nokia Corporation and/or its subsidiary(-ies)\n" \
-  "Copyright (C) 2009 Pierre Ossman for Cendio AB\n" \
  "Copyright (C) 1999-2006 MIYASAKA Masaru\n" \
-  "Copyright (C) 1991-2016 Thomas G. Lane, Guido Vollbeding"
+  "Copyright (C) 1991-2020 Thomas G. Lane, Guido Vollbeding"

 #define JCOPYRIGHT_SHORT \
-  "Copyright (C) 1991-2019 The libjpeg-turbo Project and many others"
+  "Copyright (C) 1991-2021 The libjpeg-turbo Project and many others"
@@ -15,18 +15,11 @@ else()
  ocv_list_filterout(lib_srcs jmemnobs.c)
 endif()

-if(WINRT)
-    add_definitions(-DNO_GETENV)
-    get_directory_property( DirDefs COMPILE_DEFINITIONS )
-    message(STATUS "Adding NO_GETENV to compiler definitions for WINRT:")
-    message(STATUS "   COMPILE_DEFINITIONS = ${DirDefs}")
-endif()
-
 # ----------------------------------------------------------------------------------
 #         Define the library target:
 # ----------------------------------------------------------------------------------

-add_library(${JPEG_LIBRARY} STATIC ${lib_srcs} ${lib_hdrs})
+add_library(${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")
@@ -49,7 +42,7 @@ if(ENABLE_SOLUTION_FOLDERS)
 endif()

 if(NOT BUILD_SHARED_LIBS)
-  ocv_install_target(${JPEG_LIBRARY} EXPORT OpenCVModules ARCHIVE DESTINATION ${OPENCV_3P_LIB_INSTALL_PATH} COMPONENT dev)
+  ocv_install_target(${JPEG_LIBRARY} EXPORT OpenCVModules ARCHIVE DESTINATION ${OPENCV_3P_LIB_INSTALL_PATH} COMPONENT dev OPTIONAL)
 endif()

 ocv_install_3rdparty_licenses(libjpeg README)
@@ -1,7 +1,7 @@
 The Independent JPEG Group's JPEG software
 ==========================================

-README for release 9c of 14-Jan-2018
+README for release 9d of 12-Jan-2020
 ====================================

 This distribution contains the ninth public release of the Independent JPEG
@@ -10,8 +10,8 @@ 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,
-Julian Minguillon, Luis Ortiz, George Phillips, Davide Rossi, Ge' Weijers,
-and other members of the Independent JPEG Group.
+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).
@@ -115,7 +115,7 @@ 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) 1991-2018, Thomas G. Lane, Guido Vollbeding.
+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
@@ -152,13 +152,6 @@ 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.

-The IJG distribution formerly included code to read and write GIF files.
-To avoid entanglement with the Unisys LZW patent (now expired), GIF reading
-support has been removed altogether, and the GIF writer has been simplified
-to produce "uncompressed GIFs".  This technique does not use the LZW
-algorithm; the resulting GIF files are larger than usual, but are readable
-by all standard GIF decoders.
-

 REFERENCES
 ==========
@@ -246,8 +239,8 @@ 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.v9c.tar.gz, and in Windows-compatible
-"zip" archive format as http://www.ijg.org/files/jpegsr9c.zip.
+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.
@@ -1,6 +1,55 @@
 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
 -----------------------

@@ -1,7 +1,7 @@
 /*
 * jcarith.c
 *
- * Developed 1997-2013 by Guido Vollbeding.
+ * Developed 1997-2019 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.
 *
@@ -181,11 +181,11 @@ finish_pass (j_compress_ptr cinfo)
    if (e->zc)  /* output final pending zero bytes */
      do emit_byte(0x00, cinfo);
      while (--e->zc);
-    emit_byte((e->c >> 19) & 0xFF, cinfo);
+    emit_byte((int) ((e->c >> 19) & 0xFF), cinfo);
    if (((e->c >> 19) & 0xFF) == 0xFF)
      emit_byte(0x00, cinfo);
    if (e->c & 0x7F800L) {
-      emit_byte((e->c >> 11) & 0xFF, cinfo);
+      emit_byte((int) ((e->c >> 11) & 0xFF), cinfo);
      if (((e->c >> 11) & 0xFF) == 0xFF)
 	emit_byte(0x00, cinfo);
    }
@@ -280,7 +280,8 @@ arith_encode (j_compress_ptr cinfo, unsigned char *st, int val)
 	/* 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). */
-	e->buffer = temp & 0xFF;  /* new output byte, might overflow later */
+	/* 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 {
@@ -302,7 +303,8 @@ arith_encode (j_compress_ptr cinfo, unsigned char *st, int val)
 	    emit_byte(0x00, cinfo);
 	  } while (--e->sc);
 	}
-	e->buffer = temp & 0xFF;  /* new output byte (can still overflow) */
+	/* New output byte (can still overflow) */
+	e->buffer = (int) (temp & 0xFF);
      }
      e->c &= 0x7FFFFL;
      e->ct += 8;
@@ -926,9 +928,8 @@ 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));
+  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;
@@ -2,7 +2,7 @@
 * jccolor.c
 *
 * Copyright (C) 1991-1996, Thomas G. Lane.
- * Modified 2011-2013 by Guido Vollbeding.
+ * Modified 2011-2019 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.
 *
@@ -105,14 +105,14 @@ rgb_ycc_start (j_compress_ptr cinfo)
  /* 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)));
+				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;
+    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.
@@ -121,8 +121,8 @@ rgb_ycc_start (j_compress_ptr cinfo)
 /*  B=>Cb and R=>Cr tables are the same
    rgb_ycc_tab[i+R_CR_OFF] = FIX(0.5) * i    + 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;
+    rgb_ycc_tab[i+G_CR_OFF] = (- FIX(0.418687589)) * i;
+    rgb_ycc_tab[i+B_CR_OFF] = (- FIX(0.081312411)) * i;
  }
 }

@@ -131,12 +131,12 @@ rgb_ycc_start (j_compress_ptr cinfo)
 * 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.
+ * 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.
+ * 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)
@@ -145,8 +145,8 @@ rgb_ycc_convert (j_compress_ptr cinfo,
 		 JDIMENSION output_row, int num_rows)
 {
  my_cconvert_ptr cconvert = (my_cconvert_ptr) cinfo->cconvert;
-  register INT32 * ctab = cconvert->rgb_ycc_tab;
  register int r, g, b;
+  register INT32 * ctab = cconvert->rgb_ycc_tab;
  register JSAMPROW inptr;
  register JSAMPROW outptr0, outptr1, outptr2;
  register JDIMENSION col;
@@ -162,6 +162,7 @@ rgb_ycc_convert (j_compress_ptr cinfo,
      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
@@ -179,7 +180,6 @@ rgb_ycc_convert (j_compress_ptr cinfo,
      outptr2[col] = (JSAMPLE)
 		((ctab[r+R_CR_OFF] + ctab[g+G_CR_OFF] + ctab[b+B_CR_OFF])
 		 >> SCALEBITS);
-      inptr += RGB_PIXELSIZE;
    }
  }
 }
@@ -201,8 +201,8 @@ rgb_gray_convert (j_compress_ptr cinfo,
 		  JDIMENSION output_row, int num_rows)
 {
  my_cconvert_ptr cconvert = (my_cconvert_ptr) cinfo->cconvert;
-  register INT32 * ctab = cconvert->rgb_ycc_tab;
  register int r, g, b;
+  register INT32 * ctab = cconvert->rgb_ycc_tab;
  register JSAMPROW inptr;
  register JSAMPROW outptr;
  register JDIMENSION col;
@@ -215,11 +215,11 @@ rgb_gray_convert (j_compress_ptr cinfo,
      r = GETJSAMPLE(inptr[RGB_RED]);
      g = GETJSAMPLE(inptr[RGB_GREEN]);
      b = GETJSAMPLE(inptr[RGB_BLUE]);
+      inptr += RGB_PIXELSIZE;
      /* Y */
      outptr[col] = (JSAMPLE)
 		((ctab[r+R_Y_OFF] + ctab[g+G_Y_OFF] + ctab[b+B_Y_OFF])
 		 >> SCALEBITS);
-      inptr += RGB_PIXELSIZE;
    }
  }
 }
@@ -228,8 +228,8 @@ rgb_gray_convert (j_compress_ptr cinfo,
 /*
 * 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.
+ * 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.
 */

@@ -239,8 +239,8 @@ cmyk_ycck_convert (j_compress_ptr cinfo,
 		   JDIMENSION output_row, int num_rows)
 {
  my_cconvert_ptr cconvert = (my_cconvert_ptr) cinfo->cconvert;
-  register INT32 * ctab = cconvert->rgb_ycc_tab;
  register int r, g, b;
+  register INT32 * ctab = cconvert->rgb_ycc_tab;
  register JSAMPROW inptr;
  register JSAMPROW outptr0, outptr1, outptr2, outptr3;
  register JDIMENSION col;
@@ -259,6 +259,7 @@ cmyk_ycck_convert (j_compress_ptr cinfo,
      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
@@ -276,7 +277,6 @@ cmyk_ycck_convert (j_compress_ptr cinfo,
      outptr2[col] = (JSAMPLE)
 		((ctab[r+R_CR_OFF] + ctab[g+G_CR_OFF] + ctab[b+B_CR_OFF])
 		 >> SCALEBITS);
-      inptr += 4;
    }
  }
 }
@@ -312,13 +312,13 @@ rgb_rgb1_convert (j_compress_ptr cinfo,
      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);
-      inptr += RGB_PIXELSIZE;
    }
  }
 }
@@ -335,17 +335,17 @@ grayscale_convert (j_compress_ptr cinfo,
 		   JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
 		   JDIMENSION output_row, int num_rows)
 {
-  int instride = cinfo->input_components;
  register JSAMPROW inptr;
  register JSAMPROW outptr;
-  register JDIMENSION col;
+  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 (col = 0; col < num_cols; col++) {
-      outptr[col] = inptr[0];	/* don't need GETJSAMPLE() here */
+    for (count = num_cols; count > 0; count--) {
+      *outptr++ = *inptr;	/* don't need GETJSAMPLE() here */
      inptr += instride;
    }
  }
@@ -396,21 +396,21 @@ null_convert (j_compress_ptr cinfo,
 	      JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
 	      JDIMENSION output_row, int num_rows)
 {
-  int ci;
-  register int nc = cinfo->num_components;
  register JSAMPROW inptr;
  register JSAMPROW outptr;
-  register JDIMENSION col;
+  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 < nc; ci++) {
+    for (ci = 0; ci < num_comps; ci++) {
      inptr = input_buf[0] + ci;
      outptr = output_buf[ci][output_row];
-      for (col = 0; col < num_cols; col++) {
+      for (count = num_cols; count > 0; count--) {
 	*outptr++ = *inptr;	/* don't need GETJSAMPLE() here */
-	inptr += nc;
+	inptr += num_comps;
      }
    }
    input_buf++;
@@ -439,9 +439,8 @@ 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));
+  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;
@@ -455,9 +454,11 @@ jinit_color_converter (j_compress_ptr cinfo)

  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:
@@ -474,7 +475,6 @@ jinit_color_converter (j_compress_ptr cinfo)
  default:			/* JCS_UNKNOWN can be anything */
    if (cinfo->input_components < 1)
      ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE);
-    break;
  }

  /* Support color transform only for RGB colorspaces */
@@ -507,19 +507,18 @@ jinit_color_converter (j_compress_ptr cinfo)
  case JCS_BG_RGB:
    if (cinfo->num_components != 3)
      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
-    if (cinfo->in_color_space == cinfo->jpeg_color_space) {
-      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);
-      }
-    } else
+    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:
@@ -572,10 +571,9 @@ jinit_color_converter (j_compress_ptr cinfo)
  case JCS_CMYK:
    if (cinfo->num_components != 4)
      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
-    if (cinfo->in_color_space == JCS_CMYK)
-      cconvert->pub.color_convert = null_convert;
-    else
+    if (cinfo->in_color_space != JCS_CMYK)
      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    cconvert->pub.color_convert = null_convert;
    break;

  case JCS_YCCK:
@@ -599,6 +597,5 @@ jinit_color_converter (j_compress_ptr cinfo)
 	cinfo->num_components != cinfo->input_components)
      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
    cconvert->pub.color_convert = null_convert;
-    break;
  }
 }
@@ -2,7 +2,7 @@
 * jchuff.c
 *
 * Copyright (C) 1991-1997, Thomas G. Lane.
- * Modified 2006-2013 by Guido Vollbeding.
+ * Modified 2006-2019 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.
 *
@@ -178,13 +178,12 @@ jpeg_make_c_derived_tbl (j_compress_ptr cinfo, boolean isDC, int tblno,
  htbl =
    isDC ? cinfo->dc_huff_tbl_ptrs[tblno] : cinfo->ac_huff_tbl_ptrs[tblno];
  if (htbl == NULL)
-    ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tblno);
+    htbl = jpeg_std_huff_table((j_common_ptr) cinfo, isDC, tblno);

  /* Allocate a workspace if we haven't already done so. */
  if (*pdtbl == NULL)
-    *pdtbl = (c_derived_tbl *)
-      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-				  SIZEOF(c_derived_tbl));
+    *pdtbl = (c_derived_tbl *) (*cinfo->mem->alloc_small)
+      ((j_common_ptr) cinfo, JPOOL_IMAGE, SIZEOF(c_derived_tbl));
  dtbl = *pdtbl;
  
  /* Figure C.1: make table of Huffman code length for each symbol */
@@ -1256,22 +1255,88 @@ jpeg_gen_optimal_table (j_compress_ptr cinfo, JHUFF_TBL * htbl, long freq[])
  UINT8 bits[MAX_CLEN+1];	/* bits[k] = # of symbols with code length k */
  int codesize[257];		/* codesize[k] = code length of symbol k */
  int others[257];		/* next symbol in current branch of tree */
-  int c1, c2;
-  int p, i, j;
+  int c1, c2, i, j;
+  UINT8 *p;
  long v;

+  freq[256] = 1;		/* make sure 256 has a nonzero count */
+  /* Including the pseudo-symbol 256 in the Huffman procedure guarantees
+   * that no real symbol is given code-value of all ones, because 256
+   * will be placed last in the largest codeword category.
+   * In the symbol list build procedure this element serves as sentinel
+   * for the zero run loop.
+   */
+
+#ifndef DONT_USE_FANCY_HUFF_OPT
+
+  /* Build list of symbols sorted in order of descending frequency */
+  /* This approach has several benefits (thank to John Korejwa for the idea):
+   *     1.
+   * If a codelength category is split during the length limiting procedure
+   * below, the feature that more frequent symbols are assigned shorter
+   * codewords remains valid for the adjusted code.
+   *     2.
+   * To reduce consecutive ones in a Huffman data stream (thus reducing the
+   * number of stuff bytes in JPEG) it is preferable to follow 0 branches
+   * (and avoid 1 branches) as much as possible.  This is easily done by
+   * assigning symbols to leaves of the Huffman tree in order of decreasing
+   * frequency, with no secondary sort based on codelengths.
+   *     3.
+   * The symbol list can be built independently from the assignment of code
+   * lengths by the Huffman procedure below.
+   * Note: The symbol list build procedure must be performed first, because
+   * the Huffman procedure assigning the codelengths clobbers the frequency
+   * counts!
+   */
+
+  /* Here we use the others array as a linked list of nonzero frequencies
+   * to be sorted.  Already sorted elements are removed from the list.
+   */
+
+  /* Building list */
+
+  /* This item does not correspond to a valid symbol frequency and is used
+   * as starting index.
+   */
+  j = 256;
+
+  for (i = 0;; i++) {
+    if (freq[i] == 0)		/* skip zero frequencies */
+      continue;
+    if (i > 255)
+      break;
+    others[j] = i;		/* this symbol value */
+    j = i;			/* previous symbol value */
+  }
+  others[j] = -1;		/* mark end of list */
+
+  /* Sorting list */
+
+  p = htbl->huffval;
+  while ((c1 = others[256]) >= 0) {
+    v = freq[c1];
+    i = c1;			/* first symbol value */
+    j = 256;			/* pseudo symbol value for starting index */
+    while ((c2 = others[c1]) >= 0) {
+      if (freq[c2] > v) {
+	v = freq[c2];
+	i = c2;			/* this symbol value */
+	j = c1;			/* previous symbol value */
+      }
+      c1 = c2;
+    }
+    others[j] = others[i];	/* remove this symbol i from list */
+    *p++ = (UINT8) i;
+  }
+
+#endif /* DONT_USE_FANCY_HUFF_OPT */
+
  /* This algorithm is explained in section K.2 of the JPEG standard */

  MEMZERO(bits, SIZEOF(bits));
  MEMZERO(codesize, SIZEOF(codesize));
  for (i = 0; i < 257; i++)
    others[i] = -1;		/* init links to empty */
-  
-  freq[256] = 1;		/* make sure 256 has a nonzero count */
-  /* Including the pseudo-symbol 256 in the Huffman procedure guarantees
-   * that no real symbol is given code-value of all ones, because 256
-   * will be placed last in the largest codeword category.
-   */

  /* Huffman's basic algorithm to assign optimal code lengths to symbols */

@@ -1301,7 +1366,7 @@ jpeg_gen_optimal_table (j_compress_ptr cinfo, JHUFF_TBL * htbl, long freq[])
    /* Done if we've merged everything into one frequency */
    if (c2 < 0)
      break;
-    
+
    /* Else merge the two counts/trees */
    freq[c1] += freq[c2];
    freq[c2] = 0;
@@ -1312,9 +1377,9 @@ jpeg_gen_optimal_table (j_compress_ptr cinfo, JHUFF_TBL * htbl, long freq[])
      c1 = others[c1];
      codesize[c1]++;
    }
-    
+
    others[c1] = c2;		/* chain c2 onto c1's tree branch */
-    
+
    /* Increment the codesize of everything in c2's tree branch */
    codesize[c2]++;
    while (others[c2] >= 0) {
@@ -1329,7 +1394,7 @@ jpeg_gen_optimal_table (j_compress_ptr cinfo, JHUFF_TBL * htbl, long freq[])
      /* The JPEG standard seems to think that this can't happen, */
      /* but I'm paranoid... */
      if (codesize[i] > MAX_CLEN)
-	ERREXIT(cinfo, JERR_HUFF_CLEN_OVERFLOW);
+	ERREXIT(cinfo, JERR_HUFF_CLEN_OUTOFBOUNDS);

      bits[codesize[i]]++;
    }
@@ -1345,13 +1410,16 @@ jpeg_gen_optimal_table (j_compress_ptr cinfo, JHUFF_TBL * htbl, long freq[])
   * shortest nonzero BITS entry is converted into a prefix for two code words
   * one bit longer.
   */
-  
+
  for (i = MAX_CLEN; i > 16; i--) {
    while (bits[i] > 0) {
      j = i - 2;		/* find length of new prefix to be used */
-      while (bits[j] == 0)
+      while (bits[j] == 0) {
+	if (j == 0)
+	  ERREXIT(cinfo, JERR_HUFF_CLEN_OUTOFBOUNDS);
 	j--;
-      
+      }
+
      bits[i] -= 2;		/* remove two symbols */
      bits[i-1]++;		/* one goes in this length */
      bits[j+1] += 2;		/* two new symbols in this length */
@@ -1363,24 +1431,27 @@ jpeg_gen_optimal_table (j_compress_ptr cinfo, JHUFF_TBL * htbl, long freq[])
  while (bits[i] == 0)		/* find largest codelength still in use */
    i--;
  bits[i]--;
-  
+
  /* Return final symbol counts (only for lengths 0..16) */
  MEMCOPY(htbl->bits, bits, SIZEOF(htbl->bits));
-  
+
+#ifdef DONT_USE_FANCY_HUFF_OPT
+
  /* Return a list of the symbols sorted by code length */
-  /* It's not real clear to me why we don't need to consider the codelength
-   * changes made above, but the JPEG spec seems to think this works.
+  /* Note: Due to the codelength changes made above, it can happen
+   * that more frequent symbols are assigned longer codewords.
   */
-  p = 0;
+  p = htbl->huffval;
  for (i = 1; i <= MAX_CLEN; i++) {
    for (j = 0; j <= 255; j++) {
      if (codesize[j] == i) {
-	htbl->huffval[p] = (UINT8) j;
-	p++;
+	*p++ = (UINT8) j;
      }
    }
  }

+#endif /* DONT_USE_FANCY_HUFF_OPT */
+
  /* Set sent_table FALSE so updated table will be written to JPEG file. */
  htbl->sent_table = FALSE;
 }
@@ -1400,13 +1471,13 @@ finish_pass_gather (j_compress_ptr cinfo)
  boolean did_dc[NUM_HUFF_TBLS];
  boolean did_ac[NUM_HUFF_TBLS];

-  /* It's important not to apply jpeg_gen_optimal_table more than once
-   * per table, because it clobbers the input frequency counts!
-   */
  if (cinfo->progressive_mode)
    /* Flush out buffered data (all we care about is counting the EOB symbol) */
    emit_eobrun(entropy);

+  /* It's important not to apply jpeg_gen_optimal_table more than once
+   * per table, because it clobbers the input frequency counts!
+   */
  MEMZERO(did_dc, SIZEOF(did_dc));
  MEMZERO(did_ac, SIZEOF(did_ac));

@@ -1475,9 +1546,8 @@ start_pass_huff (j_compress_ptr cinfo, boolean gather_statistics)
 	entropy->pub.encode_mcu = encode_mcu_AC_refine;
 	/* AC refinement needs a correction bit buffer */
 	if (entropy->bit_buffer == NULL)
-	  entropy->bit_buffer = (char *)
-	    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-					MAX_CORR_BITS * SIZEOF(char));
+	  entropy->bit_buffer = (char *) (*cinfo->mem->alloc_small)
+	    ((j_common_ptr) cinfo, JPOOL_IMAGE, MAX_CORR_BITS * SIZEOF(char));
      }
    }

@@ -1505,9 +1575,8 @@ start_pass_huff (j_compress_ptr cinfo, boolean gather_statistics)
 	/* Allocate and zero the statistics tables */
 	/* Note that jpeg_gen_optimal_table expects 257 entries in each table! */
 	if (entropy->dc_count_ptrs[tbl] == NULL)
-	  entropy->dc_count_ptrs[tbl] = (long *)
-	    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-					257 * SIZEOF(long));
+	  entropy->dc_count_ptrs[tbl] = (long *) (*cinfo->mem->alloc_small)
+	    ((j_common_ptr) cinfo, JPOOL_IMAGE, 257 * SIZEOF(long));
 	MEMZERO(entropy->dc_count_ptrs[tbl], 257 * SIZEOF(long));
      } else {
 	/* Compute derived values for Huffman tables */
@@ -1525,9 +1594,8 @@ start_pass_huff (j_compress_ptr cinfo, boolean gather_statistics)
 	if (tbl < 0 || tbl >= NUM_HUFF_TBLS)
 	  ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tbl);
 	if (entropy->ac_count_ptrs[tbl] == NULL)
-	  entropy->ac_count_ptrs[tbl] = (long *)
-	    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-					257 * SIZEOF(long));
+	  entropy->ac_count_ptrs[tbl] = (long *) (*cinfo->mem->alloc_small)
+	    ((j_common_ptr) cinfo, JPOOL_IMAGE, 257 * SIZEOF(long));
 	MEMZERO(entropy->ac_count_ptrs[tbl], 257 * SIZEOF(long));
      } else {
 	jpeg_make_c_derived_tbl(cinfo, FALSE, tbl,
@@ -1556,9 +1624,8 @@ jinit_huff_encoder (j_compress_ptr cinfo)
  huff_entropy_ptr entropy;
  int i;

-  entropy = (huff_entropy_ptr)
-    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-				SIZEOF(huff_entropy_encoder));
+  entropy = (huff_entropy_ptr) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, SIZEOF(huff_entropy_encoder));
  cinfo->entropy = &entropy->pub;
  entropy->pub.start_pass = start_pass_huff;

@@ -2,7 +2,7 @@
 * jcmarker.c
 *
 * Copyright (C) 1991-1998, Thomas G. Lane.
- * Modified 2003-2013 by Guido Vollbeding.
+ * 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.
 *
@@ -471,7 +471,6 @@ emit_adobe_app14 (j_compress_ptr cinfo)
    break;
  default:
    emit_byte(cinfo, 0);	/* Color transform = 0 */
-    break;
  }
 }

@@ -702,9 +701,8 @@ 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));
+  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;
@@ -2,7 +2,7 @@
 * jcmaster.c
 *
 * Copyright (C) 1991-1997, Thomas G. Lane.
- * Modified 2003-2017 by Guido Vollbeding.
+ * 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.
 *
@@ -62,7 +62,7 @@ initial_setup (j_compress_ptr cinfo)
  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; break;
+  default: cinfo->natural_order = jpeg_natural_order;
  }

  /* Derive lim_Se from block_size */
@@ -114,20 +114,24 @@ initial_setup (j_compress_ptr cinfo)
     */
    ssize = 1;
 #ifdef DCT_SCALING_SUPPORTED
-    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;
-    }
+    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
-    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;
-    }
+    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;

@@ -620,9 +624,8 @@ 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));
+  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;
@@ -2,6 +2,7 @@
 * 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.
 *
@@ -104,3 +105,140 @@ jpeg_alloc_huff_table (j_common_ptr cinfo)
  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;
+}
@@ -2,7 +2,7 @@
 * jcparam.c
 *
 * Copyright (C) 1991-1998, Thomas G. Lane.
- * Modified 2003-2013 by Guido Vollbeding.
+ * 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.
 *
@@ -162,112 +162,23 @@ jpeg_set_quality (j_compress_ptr cinfo, int quality, boolean force_baseline)


 /*
- * Huffman table setup routines
+ * Reset standard Huffman tables
 */

-LOCAL(void)
-add_huff_table (j_compress_ptr cinfo,
-		JHUFF_TBL **htblptr, const UINT8 *bits, const UINT8 *val)
-/* Define a Huffman table */
-{
-  int nsymbols, len;
-
-  if (*htblptr == NULL)
-    *htblptr = jpeg_alloc_huff_table((j_common_ptr) cinfo);
-
-  /* Copy the number-of-symbols-of-each-code-length counts */
-  MEMCOPY((*htblptr)->bits, bits, SIZEOF((*htblptr)->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.  jchuff.c will do a more thorough test later.
-   */
-  nsymbols = 0;
-  for (len = 1; len <= 16; len++)
-    nsymbols += bits[len];
-  if (nsymbols < 1 || nsymbols > 256)
-    ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);
-
-  MEMCOPY((*htblptr)->huffval, val, nsymbols * SIZEOF(UINT8));
-
-  /* Initialize sent_table FALSE so table will be written to JPEG file. */
-  (*htblptr)->sent_table = FALSE;
-}
-
-
 LOCAL(void)
 std_huff_tables (j_compress_ptr cinfo)
-/* Set up the standard Huffman tables (cf. JPEG standard section K.3) */
-/* IMPORTANT: these are only valid for 8-bit data precision! */
 {
-  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 };
-  
-  add_huff_table(cinfo, &cinfo->dc_huff_tbl_ptrs[0],
-		 bits_dc_luminance, val_dc_luminance);
-  add_huff_table(cinfo, &cinfo->ac_huff_tbl_ptrs[0],
-		 bits_ac_luminance, val_ac_luminance);
-  add_huff_table(cinfo, &cinfo->dc_huff_tbl_ptrs[1],
-		 bits_dc_chrominance, val_dc_chrominance);
-  add_huff_table(cinfo, &cinfo->ac_huff_tbl_ptrs[1],
-		 bits_ac_chrominance, val_ac_chrominance);
+  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);
 }


@@ -306,7 +217,7 @@ jpeg_set_defaults (j_compress_ptr cinfo)
  cinfo->data_precision = BITS_IN_JSAMPLE;
  /* Set up two quantization tables using default quality of 75 */
  jpeg_set_quality(cinfo, 75, TRUE);
-  /* Set up two Huffman tables */
+  /* Reset standard Huffman tables */
  std_huff_tables(cinfo);

  /* Initialize default arithmetic coding conditioning */
@@ -1,7 +1,7 @@
 /*
 * jdarith.c
 *
- * Developed 1997-2015 by Guido Vollbeding.
+ * Developed 1997-2019 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.
 *
@@ -280,7 +280,7 @@ decode_mcu_DC_first (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
      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) == 0x8000) {
+	  if ((m <<= 1) == (int) 0x8000U) {
 	    WARNMS(cinfo, JWRN_ARITH_BAD_CODE);
 	    entropy->ct = -1;			/* magnitude overflow */
 	    return TRUE;
@@ -370,7 +370,7 @@ decode_mcu_AC_first (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
 	st = entropy->ac_stats[tbl] +
 	     (k <= cinfo->arith_ac_K[tbl] ? 189 : 217);
 	while (arith_decode(cinfo, st)) {
-	  if ((m <<= 1) == 0x8000) {
+	  if ((m <<= 1) == (int) 0x8000U) {
 	    WARNMS(cinfo, JWRN_ARITH_BAD_CODE);
 	    entropy->ct = -1;			/* magnitude overflow */
 	    return TRUE;
@@ -404,7 +404,8 @@ decode_mcu_DC_refine (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
 {
  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
  unsigned char *st;
-  int p1, blkn;
+  JCOEF p1;
+  int blkn;

  /* Process restart marker if needed */
  if (cinfo->restart_interval) {
@@ -440,7 +441,7 @@ decode_mcu_AC_refine (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
  JCOEFPTR thiscoef;
  unsigned char *st;
  int tbl, k, kex;
-  int p1, m1;
+  JCOEF p1, m1;
  const int * natural_order;

  /* Process restart marker if needed */
@@ -459,7 +460,7 @@ decode_mcu_AC_refine (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
  tbl = cinfo->cur_comp_info[0]->ac_tbl_no;

  p1 = 1 << cinfo->Al;		/* 1 in the bit position being coded */
-  m1 = (-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;
@@ -555,7 +556,7 @@ decode_mcu (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
      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) == 0x8000) {
+	  if ((m <<= 1) == (int) 0x8000U) {
 	    WARNMS(cinfo, JWRN_ARITH_BAD_CODE);
 	    entropy->ct = -1;			/* magnitude overflow */
 	    return TRUE;
@@ -612,7 +613,7 @@ decode_mcu (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
 	  st = entropy->ac_stats[tbl] +
 	       (k <= cinfo->arith_ac_K[tbl] ? 189 : 217);
 	  while (arith_decode(cinfo, st)) {
-	    if ((m <<= 1) == 0x8000) {
+	    if ((m <<= 1) == (int) 0x8000U) {
 	      WARNMS(cinfo, JWRN_ARITH_BAD_CODE);
 	      entropy->ct = -1;			/* magnitude overflow */
 	      return TRUE;
@@ -766,9 +767,8 @@ 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));
+  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;
@@ -785,9 +785,9 @@ jinit_arith_decoder (j_decompress_ptr cinfo)
  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));
+    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++)
@@ -2,7 +2,7 @@
 * jdatadst.c
 *
 * Copyright (C) 1994-1996, Thomas G. Lane.
- * Modified 2009-2017 by Guido Vollbeding.
+ * Modified 2009-2019 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.
 *
@@ -46,7 +46,7 @@ typedef struct {
  struct jpeg_destination_mgr pub; /* public fields */

  unsigned char ** outbuffer;	/* target buffer */
-  unsigned long * outsize;
+  size_t * outsize;
  unsigned char * newbuffer;	/* newly allocated buffer */
  JOCTET * buffer;		/* start of buffer */
  size_t bufsize;
@@ -66,9 +66,8 @@ init_destination (j_compress_ptr cinfo)
  my_dest_ptr dest = (my_dest_ptr) cinfo->dest;

  /* Allocate the output buffer --- it will be released when done with image */
-  dest->buffer = (JOCTET *)
-      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-				  OUTPUT_BUF_SIZE * SIZEOF(JOCTET));
+  dest->buffer = (JOCTET *) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, OUTPUT_BUF_SIZE * SIZEOF(JOCTET));

  dest->pub.next_output_byte = dest->buffer;
  dest->pub.free_in_buffer = OUTPUT_BUF_SIZE;
@@ -131,7 +130,7 @@ empty_mem_output_buffer (j_compress_ptr cinfo)
  nextbuffer = (JOCTET *) malloc(nextsize);

  if (nextbuffer == NULL)
-    ERREXIT1(cinfo, JERR_OUT_OF_MEMORY, 10);
+    ERREXIT1(cinfo, JERR_OUT_OF_MEMORY, 11);

  MEMCOPY(nextbuffer, dest->buffer, dest->bufsize);

@@ -204,9 +203,8 @@ jpeg_stdio_dest (j_compress_ptr cinfo, FILE * outfile)
   * 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));
+    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;
@@ -233,7 +231,7 @@ jpeg_stdio_dest (j_compress_ptr cinfo, FILE * outfile)

 GLOBAL(void)
 jpeg_mem_dest (j_compress_ptr cinfo,
-	       unsigned char ** outbuffer, unsigned long * outsize)
+	       unsigned char ** outbuffer, size_t * outsize)
 {
  my_mem_dest_ptr dest;

@@ -244,9 +242,8 @@ jpeg_mem_dest (j_compress_ptr cinfo,
   * 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));
+    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;
@@ -2,7 +2,7 @@
 * jdatasrc.c
 *
 * Copyright (C) 1994-1996, Thomas G. Lane.
- * Modified 2009-2015 by Guido Vollbeding.
+ * Modified 2009-2019 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.
 *
@@ -156,21 +156,23 @@ 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) {
-    while (num_bytes > (long) src->bytes_in_buffer) {
-      num_bytes -= (long) src->bytes_in_buffer;
+    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 += (size_t) num_bytes;
-    src->bytes_in_buffer -= (size_t) num_bytes;
+    src->next_input_byte += nbytes;
+    src->bytes_in_buffer -= nbytes;
  }
 }

@@ -219,13 +221,11 @@ jpeg_stdio_src (j_decompress_ptr cinfo, FILE * infile)
   * 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));
+    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->buffer = (JOCTET *)
-      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_PERMANENT,
-				  INPUT_BUF_SIZE * SIZEOF(JOCTET));
+    src->buffer = (JOCTET *) (*cinfo->mem->alloc_small)
+      ((j_common_ptr) cinfo, JPOOL_PERMANENT, INPUT_BUF_SIZE * SIZEOF(JOCTET));
  }

  src = (my_src_ptr) cinfo->src;
@@ -247,7 +247,7 @@ jpeg_stdio_src (j_decompress_ptr cinfo, FILE * infile)

 GLOBAL(void)
 jpeg_mem_src (j_decompress_ptr cinfo,
-	      const unsigned char * inbuffer, unsigned long insize)
+	      const unsigned char * inbuffer, size_t insize)
 {
  struct jpeg_source_mgr * src;

@@ -259,9 +259,8 @@ jpeg_mem_src (j_decompress_ptr cinfo,
   * 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));
+    cinfo->src = (struct jpeg_source_mgr *) (*cinfo->mem->alloc_small)
+      ((j_common_ptr) cinfo, JPOOL_PERMANENT, SIZEOF(struct jpeg_source_mgr));
  }

  src = cinfo->src;
@@ -270,6 +269,6 @@ jpeg_mem_src (j_decompress_ptr cinfo,
  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 = (size_t) insize;
+  src->bytes_in_buffer = insize;
  src->next_input_byte = (const JOCTET *) inbuffer;
 }
@@ -2,7 +2,7 @@
 * jdcolor.c
 *
 * Copyright (C) 1991-1997, Thomas G. Lane.
- * Modified 2011-2017 by Guido Vollbeding.
+ * Modified 2011-2019 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.
 *
@@ -124,28 +124,22 @@ build_ycc_rgb_table (j_decompress_ptr cinfo)
  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));
+  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)
-		    RIGHT_SHIFT(FIX(1.402) * x + ONE_HALF, SCALEBITS);
+    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)
-		    RIGHT_SHIFT(FIX(1.772) * x + ONE_HALF, SCALEBITS);
+    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 */
@@ -164,28 +158,22 @@ build_bg_ycc_rgb_table (j_decompress_ptr cinfo)
  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));
+  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)
-		    RIGHT_SHIFT(FIX(2.804) * x + ONE_HALF, SCALEBITS);
+    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)
-		    RIGHT_SHIFT(FIX(3.544) * x + ONE_HALF, SCALEBITS);
+    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 */
@@ -201,6 +189,7 @@ build_bg_ycc_rgb_table (j_decompress_ptr cinfo)
 * 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.
@@ -264,9 +253,8 @@ build_rgb_y_table (j_decompress_ptr cinfo)
  INT32 i;

  /* Allocate and fill in the conversion tables. */
-  cconvert->rgb_y_tab = rgb_y_tab = (INT32 *)
-    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-				(TABLE_SIZE * SIZEOF(INT32)));
+  cconvert->rgb_y_tab = rgb_y_tab = (INT32 *) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, TABLE_SIZE * SIZEOF(INT32));

  for (i = 0; i <= MAXJSAMPLE; i++) {
    rgb_y_tab[i+R_Y_OFF] = FIX(0.299) * i;
@@ -286,8 +274,8 @@ rgb_gray_convert (j_decompress_ptr cinfo,
 		  JSAMPARRAY output_buf, int num_rows)
 {
  my_cconvert_ptr cconvert = (my_cconvert_ptr) cinfo->cconvert;
-  register INT32 * ctab = cconvert->rgb_y_tab;
  register int r, g, b;
+  register INT32 * ctab = cconvert->rgb_y_tab;
  register JSAMPROW outptr;
  register JSAMPROW inptr0, inptr1, inptr2;
  register JDIMENSION col;
@@ -313,6 +301,7 @@ rgb_gray_convert (j_decompress_ptr cinfo,


 /*
+ * 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
@@ -364,8 +353,8 @@ rgb1_gray_convert (j_decompress_ptr cinfo,
 		   JSAMPARRAY output_buf, int num_rows)
 {
  my_cconvert_ptr cconvert = (my_cconvert_ptr) cinfo->cconvert;
-  register INT32 * ctab = cconvert->rgb_y_tab;
  register int r, g, b;
+  register INT32 * ctab = cconvert->rgb_y_tab;
  register JSAMPROW outptr;
  register JSAMPROW inptr0, inptr1, inptr2;
  register JDIMENSION col;
@@ -396,6 +385,7 @@ rgb1_gray_convert (j_decompress_ptr cinfo,


 /*
+ * Convert some rows of samples to the output colorspace.
 * No colorspace change, but conversion from separate-planes
 * to interleaved representation.
 */
@@ -430,6 +420,7 @@ rgb_convert (j_decompress_ptr cinfo,
 /*
 * Color conversion for no colorspace change: just copy the data,
 * converting from separate-planes to interleaved representation.
+ * We assume out_color_components == num_components.
 */

 METHODDEF(void)
@@ -437,20 +428,21 @@ null_convert (j_decompress_ptr cinfo,
 	      JSAMPIMAGE input_buf, JDIMENSION input_row,
 	      JSAMPARRAY output_buf, int num_rows)
 {
-  int ci;
-  register int nc = cinfo->num_components;
  register JSAMPROW outptr;
  register JSAMPROW inptr;
-  register JDIMENSION col;
+  register JDIMENSION count;
+  register int num_comps = cinfo->num_components;
  JDIMENSION num_cols = cinfo->output_width;
+  int ci;

  while (--num_rows >= 0) {
-    for (ci = 0; ci < nc; ci++) {
+    /* It seems fastest to make a separate pass for each component. */
+    for (ci = 0; ci < num_comps; ci++) {
      inptr = input_buf[ci][input_row];
      outptr = output_buf[0] + ci;
-      for (col = 0; col < num_cols; col++) {
-	*outptr = *inptr++;	/* needn't bother with GETJSAMPLE() here */
-	outptr += nc;
+      for (count = num_cols; count > 0; count--) {
+	*outptr = *inptr++;	/* don't need GETJSAMPLE() here */
+	outptr += num_comps;
      }
    }
    input_row++;
@@ -504,9 +496,10 @@ gray_rgb_convert (j_decompress_ptr cinfo,


 /*
- * Adobe-style YCCK->CMYK conversion.
- * 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.
+ * 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.
 */

@@ -577,9 +570,8 @@ jinit_color_deconverter (j_decompress_ptr cinfo)
  my_cconvert_ptr cconvert;
  int ci;

-  cconvert = (my_cconvert_ptr)
-    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-				SIZEOF(my_color_deconverter));
+  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;

@@ -607,7 +599,6 @@ jinit_color_deconverter (j_decompress_ptr cinfo)
  default:			/* JCS_UNKNOWN can be anything */
    if (cinfo->num_components < 1)
      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
-    break;
  }

  /* Support color transform only for RGB colorspaces */
@@ -684,19 +675,18 @@ jinit_color_deconverter (j_decompress_ptr cinfo)

  case JCS_BG_RGB:
    cinfo->out_color_components = RGB_PIXELSIZE;
-    if (cinfo->jpeg_color_space == JCS_BG_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);
-      }
-    } else
+    if (cinfo->jpeg_color_space != JCS_BG_RGB)
      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 = rgb1_rgb_convert;
+      break;
+    default:
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    }
    break;

  case JCS_CMYK:
@@ -714,14 +704,12 @@ jinit_color_deconverter (j_decompress_ptr cinfo)
    }
    break;

-  default:
-    /* Permit null conversion to same output space */
-    if (cinfo->out_color_space == cinfo->jpeg_color_space) {
-      cinfo->out_color_components = cinfo->num_components;
-      cconvert->pub.color_convert = null_convert;
-    } else			/* unsupported non-null conversion */
+  default:		/* 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);
-    break;
+    cinfo->out_color_components = cinfo->num_components;
+    cconvert->pub.color_convert = null_convert;
  }

  if (cinfo->quantize_colors)
@@ -2,7 +2,7 @@
 * jdct.h
 *
 * Copyright (C) 1994-1996, Thomas G. Lane.
- * Modified 2002-2017 by Guido Vollbeding.
+ * Modified 2002-2019 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.
 *
@@ -358,13 +358,6 @@ EXTERN(void) jpeg_idct_1x2

 #define FIX(x)	((INT32) ((x) * CONST_SCALE + 0.5))

-/* Descale and correctly round an INT32 value that's scaled by N bits.
- * We assume RIGHT_SHIFT rounds towards minus infinity, so adding
- * the fudge factor is correct for either sign of X.
- */
-
-#define DESCALE(x,n)  RIGHT_SHIFT((x) + (ONE << ((n)-1)), n)
-
 /* 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
@@ -2,7 +2,7 @@
 * jdhuff.c
 *
 * Copyright (C) 1991-1997, Thomas G. Lane.
- * Modified 2006-2016 by Guido Vollbeding.
+ * Modified 2006-2019 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.
 *
@@ -341,13 +341,12 @@ jpeg_make_d_derived_tbl (j_decompress_ptr cinfo, boolean isDC, int tblno,
  htbl =
    isDC ? cinfo->dc_huff_tbl_ptrs[tblno] : cinfo->ac_huff_tbl_ptrs[tblno];
  if (htbl == NULL)
-    ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tblno);
+    htbl = jpeg_std_huff_table((j_common_ptr) cinfo, isDC, tblno);

  /* Allocate a workspace if we haven't already done so. */
  if (*pdtbl == NULL)
-    *pdtbl = (d_derived_tbl *)
-      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-				  SIZEOF(d_derived_tbl));
+    *pdtbl = (d_derived_tbl *) (*cinfo->mem->alloc_small)
+      ((j_common_ptr) cinfo, JPOOL_IMAGE, SIZEOF(d_derived_tbl));
  dtbl = *pdtbl;
  dtbl->pub = htbl;		/* fill in back link */
  
@@ -706,7 +705,7 @@ process_restart (j_decompress_ptr cinfo)

 METHODDEF(boolean)
 decode_mcu_DC_first (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
-{   
+{
  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
  int Al = cinfo->Al;
  register int s, r;
@@ -730,7 +729,7 @@ decode_mcu_DC_first (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
  if (! entropy->insufficient_data) {

    /* Load up working state */
-    BITREAD_LOAD_STATE(cinfo,entropy->bitstate);
+    BITREAD_LOAD_STATE(cinfo, entropy->bitstate);
    ASSIGN_STATE(state, entropy->saved);

    /* Outer loop handles each block in the MCU */
@@ -759,12 +758,13 @@ decode_mcu_DC_first (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
    }

    /* Completed MCU, so update state */
-    BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
+    BITREAD_SAVE_STATE(cinfo, entropy->bitstate);
    ASSIGN_STATE(entropy->saved, state);
  }

-  /* Account for restart interval (no-op if not using restarts) */
-  entropy->restarts_to_go--;
+  /* Account for restart interval if using restarts */
+  if (cinfo->restart_interval)
+    entropy->restarts_to_go--;

  return TRUE;
 }
@@ -777,7 +777,7 @@ decode_mcu_DC_first (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)

 METHODDEF(boolean)
 decode_mcu_AC_first (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
-{   
+{
  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
  register int s, k, r;
  unsigned int EOBRUN;
@@ -809,7 +809,7 @@ decode_mcu_AC_first (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
    if (EOBRUN)			/* if it's a band of zeroes... */
      EOBRUN--;			/* ...process it now (we do nothing) */
    else {
-      BITREAD_LOAD_STATE(cinfo,entropy->bitstate);
+      BITREAD_LOAD_STATE(cinfo, entropy->bitstate);
      Se = cinfo->Se;
      Al = cinfo->Al;
      natural_order = cinfo->natural_order;
@@ -842,15 +842,16 @@ decode_mcu_AC_first (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
 	}
      }

-      BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
+      BITREAD_SAVE_STATE(cinfo, entropy->bitstate);
    }

    /* Completed MCU, so update state */
    entropy->saved.EOBRUN = EOBRUN;	/* only part of saved state we need */
  }

-  /* Account for restart interval (no-op if not using restarts) */
-  entropy->restarts_to_go--;
+  /* Account for restart interval if using restarts */
+  if (cinfo->restart_interval)
+    entropy->restarts_to_go--;

  return TRUE;
 }
@@ -864,9 +865,10 @@ decode_mcu_AC_first (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)

 METHODDEF(boolean)
 decode_mcu_DC_refine (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
-{   
+{
  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
-  int p1, blkn;
+  JCOEF p1;
+  int blkn;
  BITREAD_STATE_VARS;

  /* Process restart marker if needed; may have to suspend */
@@ -881,7 +883,7 @@ decode_mcu_DC_refine (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
   */

  /* Load up working state */
-  BITREAD_LOAD_STATE(cinfo,entropy->bitstate);
+  BITREAD_LOAD_STATE(cinfo, entropy->bitstate);

  p1 = 1 << cinfo->Al;		/* 1 in the bit position being coded */

@@ -896,10 +898,11 @@ decode_mcu_DC_refine (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
  }

  /* Completed MCU, so update state */
-  BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
+  BITREAD_SAVE_STATE(cinfo, entropy->bitstate);

-  /* Account for restart interval (no-op if not using restarts) */
-  entropy->restarts_to_go--;
+  /* Account for restart interval if using restarts */
+  if (cinfo->restart_interval)
+    entropy->restarts_to_go--;

  return TRUE;
 }
@@ -911,11 +914,12 @@ decode_mcu_DC_refine (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)

 METHODDEF(boolean)
 decode_mcu_AC_refine (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
-{   
+{
  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
  register int s, k, r;
  unsigned int EOBRUN;
-  int Se, p1, m1;
+  int Se;
+  JCOEF p1, m1;
  const int * natural_order;
  JBLOCKROW block;
  JCOEFPTR thiscoef;
@@ -937,11 +941,11 @@ decode_mcu_AC_refine (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)

    Se = cinfo->Se;
    p1 = 1 << cinfo->Al;	/* 1 in the bit position being coded */
-    m1 = (-1) << cinfo->Al;	/* -1 in the bit position being coded */
+    m1 = -p1;			/* -1 in the bit position being coded */
    natural_order = cinfo->natural_order;

    /* Load up working state */
-    BITREAD_LOAD_STATE(cinfo,entropy->bitstate);
+    BITREAD_LOAD_STATE(cinfo, entropy->bitstate);
    EOBRUN = entropy->saved.EOBRUN; /* only part of saved state we need */

    /* There is always only one block per MCU */
@@ -1043,12 +1047,13 @@ decode_mcu_AC_refine (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
    }

    /* Completed MCU, so update state */
-    BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
+    BITREAD_SAVE_STATE(cinfo, entropy->bitstate);
    entropy->saved.EOBRUN = EOBRUN; /* only part of saved state we need */
  }

-  /* Account for restart interval (no-op if not using restarts) */
-  entropy->restarts_to_go--;
+  /* Account for restart interval if using restarts */
+  if (cinfo->restart_interval)
+    entropy->restarts_to_go--;

  return TRUE;

@@ -1091,7 +1096,7 @@ decode_mcu_sub (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
    Se = cinfo->lim_Se;

    /* Load up working state */
-    BITREAD_LOAD_STATE(cinfo,entropy->bitstate);
+    BITREAD_LOAD_STATE(cinfo, entropy->bitstate);
    ASSIGN_STATE(state, entropy->saved);

    /* Outer loop handles each block in the MCU */
@@ -1178,12 +1183,13 @@ decode_mcu_sub (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
    }

    /* Completed MCU, so update state */
-    BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
+    BITREAD_SAVE_STATE(cinfo, entropy->bitstate);
    ASSIGN_STATE(entropy->saved, state);
  }

-  /* Account for restart interval (no-op if not using restarts) */
-  entropy->restarts_to_go--;
+  /* Account for restart interval if using restarts */
+  if (cinfo->restart_interval)
+    entropy->restarts_to_go--;

  return TRUE;
 }
@@ -1215,7 +1221,7 @@ decode_mcu (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
  if (! entropy->insufficient_data) {

    /* Load up working state */
-    BITREAD_LOAD_STATE(cinfo,entropy->bitstate);
+    BITREAD_LOAD_STATE(cinfo, entropy->bitstate);
    ASSIGN_STATE(state, entropy->saved);

    /* Outer loop handles each block in the MCU */
@@ -1302,12 +1308,13 @@ decode_mcu (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
    }

    /* Completed MCU, so update state */
-    BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
+    BITREAD_SAVE_STATE(cinfo, entropy->bitstate);
    ASSIGN_STATE(entropy->saved, state);
  }

-  /* Account for restart interval (no-op if not using restarts) */
-  entropy->restarts_to_go--;
+  /* Account for restart interval if using restarts */
+  if (cinfo->restart_interval)
+    entropy->restarts_to_go--;

  return TRUE;
 }
@@ -1343,11 +1350,11 @@ start_pass_huff_decoder (j_decompress_ptr cinfo)
 	goto bad;
    }
    if (cinfo->Al > 13) {	/* need not check for < 0 */
-      /* Arguably the maximum Al value should be less than 13 for 8-bit precision,
-       * but the spec doesn't say so, and we try to be liberal about what we
-       * accept.  Note: large Al values could result in out-of-range DC
-       * coefficients during early scans, leading to bizarre displays due to
-       * overflows in the IDCT math.  But we won't crash.
+      /* Arguably the maximum Al value should be less than 13 for 8-bit
+       * precision, but the spec doesn't say so, and we try to be liberal
+       * about what we accept.  Note: large Al values could result in
+       * out-of-range DC coefficients during early scans, leading to bizarre
+       * displays due to overflows in the IDCT math.  But we won't crash.
       */
      bad:
      ERREXIT4(cinfo, JERR_BAD_PROGRESSION,
@@ -1451,7 +1458,8 @@ start_pass_huff_decoder (j_decompress_ptr cinfo)
      compptr = cinfo->cur_comp_info[ci];
      /* Precalculate which table to use for each block */
      entropy->dc_cur_tbls[blkn] = entropy->dc_derived_tbls[compptr->dc_tbl_no];
-      entropy->ac_cur_tbls[blkn] = entropy->ac_derived_tbls[compptr->ac_tbl_no];
+      entropy->ac_cur_tbls[blkn] =	/* AC needs no table when not present */
+	cinfo->lim_Se ? entropy->ac_derived_tbls[compptr->ac_tbl_no] : NULL;
      /* Decide whether we really care about the coefficient values */
      if (compptr->component_needed) {
 	ci = compptr->DCT_v_scaled_size;
@@ -1494,7 +1502,6 @@ start_pass_huff_decoder (j_decompress_ptr cinfo)
 	  if (ci <= 0 || ci > 8) ci = 8;
 	  if (i <= 0 || i > 8) i = 8;
 	  entropy->coef_limit[blkn] = 1 + jpeg_zigzag_order[ci - 1][i - 1];
-	  break;
 	}
      } else {
 	entropy->coef_limit[blkn] = 0;
@@ -1522,9 +1529,8 @@ jinit_huff_decoder (j_decompress_ptr cinfo)
  huff_entropy_ptr entropy;
  int i;

-  entropy = (huff_entropy_ptr)
-    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-				SIZEOF(huff_entropy_decoder));
+  entropy = (huff_entropy_ptr) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_IMAGE, SIZEOF(huff_entropy_decoder));
  cinfo->entropy = &entropy->pub;
  entropy->pub.start_pass = start_pass_huff_decoder;
  entropy->pub.finish_pass = finish_pass_huff;
@@ -1532,9 +1538,9 @@ jinit_huff_decoder (j_decompress_ptr cinfo)
  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));
+    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++)
@@ -1545,7 +1551,7 @@ jinit_huff_decoder (j_decompress_ptr cinfo)
      entropy->derived_tbls[i] = NULL;
    }
  } else {
-    /* Mark tables unallocated */
+    /* Mark derived tables unallocated */
    for (i = 0; i < NUM_HUFF_TBLS; i++) {
      entropy->dc_derived_tbls[i] = entropy->ac_derived_tbls[i] = NULL;
    }
@@ -2,7 +2,7 @@
 * jdmarker.c
 *
 * Copyright (C) 1991-1998, Thomas G. Lane.
- * Modified 2009-2013 by Guido Vollbeding.
+ * Modified 2009-2019 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.
 *
@@ -496,8 +496,6 @@ get_dht (j_decompress_ptr cinfo)
    if (count > 256 || ((INT32) count) > length)
      ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);

-    MEMZERO(huffval, SIZEOF(huffval)); /* pre-zero array for later copy */
-
    for (i = 0; i < count; i++)
      INPUT_BYTE(cinfo, huffval[i], return FALSE);

@@ -517,7 +515,8 @@ get_dht (j_decompress_ptr cinfo)
      *htblptr = jpeg_alloc_huff_table((j_common_ptr) cinfo);
  
    MEMCOPY((*htblptr)->bits, bits, SIZEOF((*htblptr)->bits));
-    MEMCOPY((*htblptr)->huffval, huffval, SIZEOF((*htblptr)->huffval));
+    if (count > 0)
+      MEMCOPY((*htblptr)->huffval, huffval, count * SIZEOF(UINT8));
  }

  if (length != 0)
@@ -577,14 +576,14 @@ get_dqt (j_decompress_ptr cinfo)
 	count = DCTSIZE2;
    }

-    switch (count) {
+    switch ((int) count) {
    case (2*2): natural_order = jpeg_natural_order2; break;
    case (3*3): natural_order = jpeg_natural_order3; break;
    case (4*4): natural_order = jpeg_natural_order4; break;
    case (5*5): natural_order = jpeg_natural_order5; break;
    case (6*6): natural_order = jpeg_natural_order6; break;
    case (7*7): natural_order = jpeg_natural_order7; break;
-    default:    natural_order = jpeg_natural_order;  break;
+    default:    natural_order = jpeg_natural_order;
    }

    for (i = 0; i < count; i++) {
@@ -784,7 +783,6 @@ examine_app0 (j_decompress_ptr cinfo, JOCTET FAR * data,
    default:
      TRACEMS2(cinfo, 1, JTRC_JFIF_EXTENSION,
 	       GETJOCTET(data[5]), (int) totallen);
-      break;
    }
  } else {
    /* Start of APP0 does not match "JFIF" or "JFXX", or too short */
@@ -858,7 +856,6 @@ get_interesting_appn (j_decompress_ptr cinfo)
  default:
    /* can't get here unless jpeg_save_markers chooses wrong processor */
    ERREXIT1(cinfo, JERR_UNKNOWN_MARKER, cinfo->unread_marker);
-    break;
  }

  /* skip any remaining data -- could be lots */
@@ -964,7 +961,6 @@ save_marker (j_decompress_ptr cinfo)
  default:
    TRACEMS2(cinfo, 1, JTRC_MISC_MARKER, cinfo->unread_marker,
 	     (int) (data_length + length));
-    break;
  }

  /* skip any remaining data -- could be lots */
@@ -1240,7 +1236,6 @@ read_markers (j_decompress_ptr cinfo)
       * ought to change!
       */
      ERREXIT1(cinfo, JERR_UNKNOWN_MARKER, cinfo->unread_marker);
-      break;
    }
    /* Successfully processed marker, so reset state variable */
    cinfo->unread_marker = 0;
@@ -1416,9 +1411,8 @@ jinit_marker_reader (j_decompress_ptr cinfo)
  int i;

  /* Create subobject in permanent pool */
-  marker = (my_marker_ptr)
-    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_PERMANENT,
-				SIZEOF(my_marker_reader));
+  marker = (my_marker_ptr) (*cinfo->mem->alloc_small)
+    ((j_common_ptr) cinfo, JPOOL_PERMANENT, SIZEOF(my_marker_reader));
  cinfo->marker = &marker->pub;
  /* Initialize public method pointers */
  marker->pub.reset_marker_reader = reset_marker_reader;
@@ -2,7 +2,7 @@
 * jdmaster.c
 *
 * Copyright (C) 1991-1997, Thomas G. Lane.
- * Modified 2002-2017 by Guido Vollbeding.
+ * Modified 2002-2019 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.
 *
@@ -104,7 +104,7 @@ jpeg_calc_output_dimensions (j_decompress_ptr cinfo)
 */
 {
 #ifdef IDCT_SCALING_SUPPORTED
-  int ci;
+  int ci, ssize;
  jpeg_component_info *compptr;
 #endif

@@ -124,19 +124,23 @@ jpeg_calc_output_dimensions (j_decompress_ptr cinfo)
   */
  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
       ci++, compptr++) {
-    int ssize = 1;
-    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;
-    }
+    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;
-    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;
-    }
+    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. */
@@ -144,13 +148,10 @@ jpeg_calc_output_dimensions (j_decompress_ptr cinfo)
 	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.
-   */
-  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
-       ci++, compptr++) {
+    /* 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 *
@@ -172,8 +173,10 @@ jpeg_calc_output_dimensions (j_decompress_ptr cinfo)
    break;
  case JCS_RGB:
  case JCS_BG_RGB:
+#if RGB_PIXELSIZE != 3
    cinfo->out_color_components = RGB_PIXELSIZE;
    break;
+#endif /* else share code with YCbCr */
  case JCS_YCbCr:
  case JCS_BG_YCC:
    cinfo->out_color_components = 3;
@@ -184,7 +187,6 @@ jpeg_calc_output_dimensions (j_decompress_ptr cinfo)
    break;
  default:			/* else must be same colorspace as in file */
    cinfo->out_color_components = cinfo->num_components;
-    break;
  }
  cinfo->output_components = (cinfo->quantize_colors ? 1 :
 			      cinfo->out_color_components);
@@ -525,9 +527,8 @@ 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));
+  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;
@@ -2,7 +2,7 @@
 * jdmerge.c
 *
 * Copyright (C) 1994-1996, Thomas G. Lane.
- * Modified 2013-2017 by Guido Vollbeding.
+ * Modified 2013-2019 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.
 *
@@ -95,28 +95,22 @@ build_ycc_rgb_table (j_decompress_ptr cinfo)
  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));
+  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)
-		    RIGHT_SHIFT(FIX(1.402) * x + ONE_HALF, SCALEBITS);
+    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)
-		    RIGHT_SHIFT(FIX(1.772) * x + ONE_HALF, SCALEBITS);
+    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 */
@@ -135,28 +129,22 @@ build_bg_ycc_rgb_table (j_decompress_ptr cinfo)
  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));
+  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)
-		    RIGHT_SHIFT(FIX(2.804) * x + ONE_HALF, SCALEBITS);
+    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)
-		    RIGHT_SHIFT(FIX(3.544) * x + ONE_HALF, SCALEBITS);
+    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 */
@@ -419,9 +407,8 @@ 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));
+  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;
@@ -432,9 +419,9 @@ jinit_merged_upsampler (j_decompress_ptr cinfo)
    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)));
+    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;
@@ -2,7 +2,7 @@
 * jerror.h
 *
 * Copyright (C) 1994-1997, Thomas G. Lane.
- * Modified 1997-2012 by Guido Vollbeding.
+ * 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.
 *
@@ -84,7 +84,7 @@ 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_OVERFLOW, "Huffman code size table overflow")
+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")
@@ -2,7 +2,7 @@
 * jfdctint.c
 *
 * Copyright (C) 1991-1996, Thomas G. Lane.
- * Modification developed 2003-2015 by Guido Vollbeding.
+ * Modification developed 2003-2018 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.
 *
@@ -3261,78 +3261,84 @@ jpeg_fdct_6x3 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
 GLOBAL(void)
 jpeg_fdct_4x2 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
 {
-  INT32 tmp0, tmp1;
-  INT32 tmp10, tmp11;
-  DCTELEM *dataptr;
+  DCTELEM tmp0, tmp2, tmp10, tmp12, tmp4, tmp5;
+  INT32 tmp1, tmp3, tmp11, tmp13;
+  INT32 z1, z2, z3;
  JSAMPROW elemptr;
-  int ctr;
  SHIFT_TEMPS

  /* Pre-zero output coefficient block. */
  MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2);

  /* Pass 1: process rows.
-   * Note results are scaled up by sqrt(8) compared to a true DCT;
-   * furthermore, we scale the results by 2**PASS1_BITS.
-   * We must also scale the output by (8/4)*(8/2) = 2**3, which we add here.
+   * Note results are scaled up by sqrt(8) compared to a true DCT.
   * 4-point FDCT kernel,
   * cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point FDCT].
   */

-  dataptr = data;
-  for (ctr = 0; ctr < 2; ctr++) {
-    elemptr = sample_data[ctr] + start_col;
+  /* Row 0 */
+  elemptr = sample_data[0] + start_col;

-    /* Even part */
+  /* Even part */

-    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[3]);
-    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[2]);
+  tmp4 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[3]);
+  tmp5 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[2]);

-    tmp10 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[3]);
-    tmp11 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[2]);
+  tmp0 = tmp4 + tmp5;
+  tmp2 = tmp4 - tmp5;

-    /* Apply unsigned->signed conversion. */
-    dataptr[0] = (DCTELEM)
-      ((tmp0 + tmp1 - 4 * CENTERJSAMPLE) << (PASS1_BITS+3));
-    dataptr[2] = (DCTELEM) ((tmp0 - tmp1) << (PASS1_BITS+3));
+  /* Odd part */

-    /* Odd part */
+  z2 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[3]);
+  z3 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[2]);

-    tmp0 = MULTIPLY(tmp10 + tmp11, FIX_0_541196100);       /* c6 */
-    /* Add fudge factor here for final descale. */
-    tmp0 += ONE << (CONST_BITS-PASS1_BITS-4);
+  z1 = MULTIPLY(z2 + z3, FIX_0_541196100);    /* c6 */
+  /* Add fudge factor here for final descale. */
+  z1 += ONE << (CONST_BITS-3-1);
+  tmp1 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */
+  tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */

-    dataptr[1] = (DCTELEM)
-      RIGHT_SHIFT(tmp0 + MULTIPLY(tmp10, FIX_0_765366865), /* c2-c6 */
-		  CONST_BITS-PASS1_BITS-3);
-    dataptr[3] = (DCTELEM)
-      RIGHT_SHIFT(tmp0 - MULTIPLY(tmp11, FIX_1_847759065), /* c2+c6 */
-		  CONST_BITS-PASS1_BITS-3);
+  /* Row 1 */
+  elemptr = sample_data[1] + start_col;

-    dataptr += DCTSIZE;		/* advance pointer to next row */
-  }
+  /* Even part */
+
+  tmp4 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[3]);
+  tmp5 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[2]);
+
+  tmp10 = tmp4 + tmp5;
+  tmp12 = tmp4 - tmp5;
+
+  /* Odd part */
+
+  z2 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[3]);
+  z3 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[2]);
+
+  z1 = MULTIPLY(z2 + z3, FIX_0_541196100);    /* c6 */
+  tmp11 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */
+  tmp13 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */

  /* Pass 2: process columns.
-   * We remove the PASS1_BITS scaling, but leave the results scaled up
-   * by an overall factor of 8.
+   * We leave the results scaled up by an overall factor of 8.
+   * We must also scale the output by (8/4)*(8/2) = 2**3.
   */

-  dataptr = data;
-  for (ctr = 0; ctr < 4; ctr++) {
-    /* Even part */
+  /* Column 0 */
+  /* Apply unsigned->signed conversion. */
+  data[DCTSIZE*0] = (tmp0 + tmp10 - 8 * CENTERJSAMPLE) << 3;
+  data[DCTSIZE*1] = (tmp0 - tmp10) << 3;

-    /* Add fudge factor here for final descale. */
-    tmp0 = dataptr[DCTSIZE*0] + (ONE << (PASS1_BITS-1));
-    tmp1 = dataptr[DCTSIZE*1];
+  /* Column 1 */
+  data[DCTSIZE*0+1] = (DCTELEM) RIGHT_SHIFT(tmp1 + tmp11, CONST_BITS-3);
+  data[DCTSIZE*1+1] = (DCTELEM) RIGHT_SHIFT(tmp1 - tmp11, CONST_BITS-3);

-    dataptr[DCTSIZE*0] = (DCTELEM) RIGHT_SHIFT(tmp0 + tmp1, PASS1_BITS);
+  /* Column 2 */
+  data[DCTSIZE*0+2] = (tmp2 + tmp12) << 3;
+  data[DCTSIZE*1+2] = (tmp2 - tmp12) << 3;

-    /* Odd part */
-
-    dataptr[DCTSIZE*1] = (DCTELEM) RIGHT_SHIFT(tmp0 - tmp1, PASS1_BITS);
-
-    dataptr++;			/* advance pointer to next column */
-  }
+  /* Column 3 */
+  data[DCTSIZE*0+3] = (DCTELEM) RIGHT_SHIFT(tmp3 + tmp13, CONST_BITS-3);
+  data[DCTSIZE*1+3] = (DCTELEM) RIGHT_SHIFT(tmp3 - tmp13, CONST_BITS-3);
 }


@@ -4312,7 +4318,6 @@ jpeg_fdct_2x4 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)

  /* Pass 1: process rows.
   * Note results are scaled up by sqrt(8) compared to a true DCT.
-   * We must also scale the output by (8/2)*(8/4) = 2**3, which we add here.
   */

  dataptr = data;
@@ -4325,17 +4330,18 @@ jpeg_fdct_2x4 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
    tmp1 = GETJSAMPLE(elemptr[1]);

    /* Apply unsigned->signed conversion. */
-    dataptr[0] = (DCTELEM) ((tmp0 + tmp1 - 2 * CENTERJSAMPLE) << 3);
+    dataptr[0] = (DCTELEM) (tmp0 + tmp1 - 2 * CENTERJSAMPLE);

    /* Odd part */

-    dataptr[1] = (DCTELEM) ((tmp0 - tmp1) << 3);
+    dataptr[1] = (DCTELEM) (tmp0 - tmp1);

    dataptr += DCTSIZE;		/* advance pointer to next row */
  }

  /* Pass 2: process columns.
   * We leave the results scaled up by an overall factor of 8.
+   * We must also scale the output by (8/2)*(8/4) = 2**3.
   * 4-point FDCT kernel,
   * cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point FDCT].
   */
@@ -4350,21 +4356,21 @@ jpeg_fdct_2x4 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
    tmp10 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*3];
    tmp11 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*2];

-    dataptr[DCTSIZE*0] = (DCTELEM) (tmp0 + tmp1);
-    dataptr[DCTSIZE*2] = (DCTELEM) (tmp0 - tmp1);
+    dataptr[DCTSIZE*0] = (DCTELEM) ((tmp0 + tmp1) << 3);
+    dataptr[DCTSIZE*2] = (DCTELEM) ((tmp0 - tmp1) << 3);

    /* Odd part */

    tmp0 = MULTIPLY(tmp10 + tmp11, FIX_0_541196100);       /* c6 */
    /* Add fudge factor here for final descale. */
-    tmp0 += ONE << (CONST_BITS-1);
+    tmp0 += ONE << (CONST_BITS-3-1);

    dataptr[DCTSIZE*1] = (DCTELEM)
      RIGHT_SHIFT(tmp0 + MULTIPLY(tmp10, FIX_0_765366865), /* c2-c6 */
-		  CONST_BITS);
+		  CONST_BITS-3);
    dataptr[DCTSIZE*3] = (DCTELEM)
      RIGHT_SHIFT(tmp0 - MULTIPLY(tmp11, FIX_1_847759065), /* c2+c6 */
-		  CONST_BITS);
+		  CONST_BITS-3);

    dataptr++;			/* advance pointer to next column */
  }
@@ -2,7 +2,7 @@
 * jidctint.c
 *
 * Copyright (C) 1991-1998, Thomas G. Lane.
- * Modification developed 2002-2016 by Guido Vollbeding.
+ * Modification developed 2002-2018 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.
 *
@@ -1474,7 +1474,7 @@ jpeg_idct_10x10 (j_decompress_ptr cinfo, jpeg_component_info * compptr,

 /*
 * Perform dequantization and inverse DCT on one block of coefficients,
- * producing a 11x11 output block.
+ * producing an 11x11 output block.
 *
 * Optimized algorithm with 24 multiplications in the 1-D kernel.
 * cK represents sqrt(2) * cos(K*pi/22).
@@ -3675,7 +3675,7 @@ jpeg_idct_10x5 (j_decompress_ptr cinfo, jpeg_component_info * compptr,

 /*
 * Perform dequantization and inverse DCT on one block of coefficients,
- * producing a 8x4 output block.
+ * producing an 8x4 output block.
 *
 * 4-point IDCT in pass 1 (columns), 8-point in pass 2 (rows).
 */
@@ -3835,7 +3835,7 @@ jpeg_idct_8x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,

 /*
 * Perform dequantization and inverse DCT on one block of coefficients,
- * producing a reduced-size 6x3 output block.
+ * producing a 6x3 output block.
 *
 * 3-point IDCT in pass 1 (columns), 6-point in pass 2 (rows).
 */
@@ -4082,7 +4082,7 @@ jpeg_idct_2x1 (j_decompress_ptr cinfo, jpeg_component_info * compptr,

 /*
 * Perform dequantization and inverse DCT on one block of coefficients,
- * producing a 8x16 output block.
+ * producing an 8x16 output block.
 *
 * 16-point IDCT in pass 1 (columns), 8-point in pass 2 (rows).
 */
@@ -5004,7 +5004,7 @@ jpeg_idct_4x8 (j_decompress_ptr cinfo, jpeg_component_info * compptr,

 /*
 * Perform dequantization and inverse DCT on one block of coefficients,
- * producing a reduced-size 3x6 output block.
+ * producing a 3x6 output block.
 *
 * 6-point IDCT in pass 1 (columns), 3-point in pass 2 (rows).
 */
@@ -2,7 +2,7 @@
 * jmemmgr.c
 *
 * Copyright (C) 1991-1997, Thomas G. Lane.
- * Modified 2011-2012 by Guido Vollbeding.
+ * Modified 2011-2019 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.
 *
@@ -130,7 +130,7 @@ typedef struct {
  jvirt_barray_ptr virt_barray_list;

  /* This counts total space obtained from jpeg_get_small/large */
-  long total_space_allocated;
+  size_t total_space_allocated;

  /* alloc_sarray and alloc_barray set this value for use by virtual
   * array routines.
@@ -195,7 +195,7 @@ print_mem_stats (j_common_ptr cinfo, int pool_id)
   * This is helpful because message parm array can't handle longs.
   */
  fprintf(stderr, "Freeing pool %d, total space = %ld\n",
-	  pool_id, mem->total_space_allocated);
+	  pool_id, (long) mem->total_space_allocated);

  for (lhdr_ptr = mem->large_list[pool_id]; lhdr_ptr != NULL;
       lhdr_ptr = lhdr_ptr->hdr.next) {
@@ -260,11 +260,11 @@ alloc_small (j_common_ptr cinfo, int pool_id, size_t sizeofobject)
 {
  my_mem_ptr mem = (my_mem_ptr) cinfo->mem;
  small_pool_ptr hdr_ptr, prev_hdr_ptr;
-  char * data_ptr;
  size_t odd_bytes, min_request, slop;
+  char * data_ptr;

  /* Check for unsatisfiable request (do now to ensure no overflow below) */
-  if (sizeofobject > (size_t) (MAX_ALLOC_CHUNK-SIZEOF(small_pool_hdr)))
+  if (sizeofobject > (size_t) MAX_ALLOC_CHUNK - SIZEOF(small_pool_hdr))
    out_of_memory(cinfo, 1);	/* request exceeds malloc's ability */

  /* Round up the requested size to a multiple of SIZEOF(ALIGN_TYPE) */
@@ -293,8 +293,8 @@ alloc_small (j_common_ptr cinfo, int pool_id, size_t sizeofobject)
    else
      slop = extra_pool_slop[pool_id];
    /* Don't ask for more than MAX_ALLOC_CHUNK */
-    if (slop > (size_t) (MAX_ALLOC_CHUNK-min_request))
-      slop = (size_t) (MAX_ALLOC_CHUNK-min_request);
+    if (slop > (size_t) MAX_ALLOC_CHUNK - min_request)
+      slop = (size_t) MAX_ALLOC_CHUNK - min_request;
    /* Try to get space, if fail reduce slop and try again */
    for (;;) {
      hdr_ptr = (small_pool_ptr) jpeg_get_small(cinfo, min_request + slop);
@@ -348,7 +348,7 @@ alloc_large (j_common_ptr cinfo, int pool_id, size_t sizeofobject)
  size_t odd_bytes;

  /* Check for unsatisfiable request (do now to ensure no overflow below) */
-  if (sizeofobject > (size_t) (MAX_ALLOC_CHUNK-SIZEOF(large_pool_hdr)))
+  if (sizeofobject > (size_t) MAX_ALLOC_CHUNK - SIZEOF(large_pool_hdr))
    out_of_memory(cinfo, 3);	/* request exceeds malloc's ability */

  /* Round up the requested size to a multiple of SIZEOF(ALIGN_TYPE) */
@@ -404,7 +404,7 @@ alloc_sarray (j_common_ptr cinfo, int pool_id,
  long ltemp;

  /* Calculate max # of rows allowed in one allocation chunk */
-  ltemp = (MAX_ALLOC_CHUNK-SIZEOF(large_pool_hdr)) /
+  ltemp = (MAX_ALLOC_CHUNK - SIZEOF(large_pool_hdr)) /
 	  ((long) samplesperrow * SIZEOF(JSAMPLE));
  if (ltemp <= 0)
    ERREXIT(cinfo, JERR_WIDTH_OVERFLOW);
@@ -416,15 +416,14 @@ alloc_sarray (j_common_ptr cinfo, int pool_id,

  /* Get space for row pointers (small object) */
  result = (JSAMPARRAY) alloc_small(cinfo, pool_id,
-				    (size_t) (numrows * SIZEOF(JSAMPROW)));
+				    (size_t) numrows * SIZEOF(JSAMPROW));

  /* Get the rows themselves (large objects) */
  currow = 0;
  while (currow < numrows) {
    rowsperchunk = MIN(rowsperchunk, numrows - currow);
    workspace = (JSAMPROW) alloc_large(cinfo, pool_id,
-	(size_t) ((size_t) rowsperchunk * (size_t) samplesperrow
-		  * SIZEOF(JSAMPLE)));
+      (size_t) rowsperchunk * (size_t) samplesperrow * SIZEOF(JSAMPLE));
    for (i = rowsperchunk; i > 0; i--) {
      result[currow++] = workspace;
      workspace += samplesperrow;
@@ -452,7 +451,7 @@ alloc_barray (j_common_ptr cinfo, int pool_id,
  long ltemp;

  /* Calculate max # of rows allowed in one allocation chunk */
-  ltemp = (MAX_ALLOC_CHUNK-SIZEOF(large_pool_hdr)) /
+  ltemp = (MAX_ALLOC_CHUNK - SIZEOF(large_pool_hdr)) /
 	  ((long) blocksperrow * SIZEOF(JBLOCK));
  if (ltemp <= 0)
    ERREXIT(cinfo, JERR_WIDTH_OVERFLOW);
@@ -464,15 +463,14 @@ alloc_barray (j_common_ptr cinfo, int pool_id,

  /* Get space for row pointers (small object) */
  result = (JBLOCKARRAY) alloc_small(cinfo, pool_id,
-				     (size_t) (numrows * SIZEOF(JBLOCKROW)));
+				     (size_t) numrows * SIZEOF(JBLOCKROW));

  /* Get the rows themselves (large objects) */
  currow = 0;
  while (currow < numrows) {
    rowsperchunk = MIN(rowsperchunk, numrows - currow);
    workspace = (JBLOCKROW) alloc_large(cinfo, pool_id,
-	(size_t) ((size_t) rowsperchunk * (size_t) blocksperrow
-		  * SIZEOF(JBLOCK)));
+      (size_t) rowsperchunk * (size_t) blocksperrow * SIZEOF(JBLOCK));
    for (i = rowsperchunk; i > 0; i--) {
      result[currow++] = workspace;
      workspace += blocksperrow;
@@ -585,8 +583,8 @@ realize_virt_arrays (j_common_ptr cinfo)
 /* Allocate the in-memory buffers for any unrealized virtual arrays */
 {
  my_mem_ptr mem = (my_mem_ptr) cinfo->mem;
-  long space_per_minheight, maximum_space, avail_mem;
-  long minheights, max_minheights;
+  long bytesperrow, space_per_minheight, maximum_space;
+  long avail_mem, minheights, max_minheights;
  jvirt_sarray_ptr sptr;
  jvirt_barray_ptr bptr;

@@ -598,18 +596,16 @@ realize_virt_arrays (j_common_ptr cinfo)
  maximum_space = 0;
  for (sptr = mem->virt_sarray_list; sptr != NULL; sptr = sptr->next) {
    if (sptr->mem_buffer == NULL) { /* if not realized yet */
-      space_per_minheight += (long) sptr->maxaccess *
-			     (long) sptr->samplesperrow * SIZEOF(JSAMPLE);
-      maximum_space += (long) sptr->rows_in_array *
-		       (long) sptr->samplesperrow * SIZEOF(JSAMPLE);
+      bytesperrow = (long) sptr->samplesperrow * SIZEOF(JSAMPLE);
+      space_per_minheight += (long) sptr->maxaccess * bytesperrow;
+      maximum_space += (long) sptr->rows_in_array * bytesperrow;
    }
  }
  for (bptr = mem->virt_barray_list; bptr != NULL; bptr = bptr->next) {
    if (bptr->mem_buffer == NULL) { /* if not realized yet */
-      space_per_minheight += (long) bptr->maxaccess *
-			     (long) bptr->blocksperrow * SIZEOF(JBLOCK);
-      maximum_space += (long) bptr->rows_in_array *
-		       (long) bptr->blocksperrow * SIZEOF(JBLOCK);
+      bytesperrow = (long) bptr->blocksperrow * SIZEOF(JBLOCK);
+      space_per_minheight += (long) bptr->maxaccess * bytesperrow;
+      maximum_space += (long) bptr->rows_in_array * bytesperrow;
    }
  }

@@ -618,7 +614,7 @@ realize_virt_arrays (j_common_ptr cinfo)

  /* Determine amount of memory to actually use; this is system-dependent. */
  avail_mem = jpeg_mem_available(cinfo, space_per_minheight, maximum_space,
-				 mem->total_space_allocated);
+				 (long) mem->total_space_allocated);

  /* If the maximum space needed is available, make all the buffers full
   * height; otherwise parcel it out with the same number of minheights
@@ -694,7 +690,7 @@ do_sarray_io (j_common_ptr cinfo, jvirt_sarray_ptr ptr, boolean writing)
  long bytesperrow, file_offset, byte_count, rows, thisrow, i;

  bytesperrow = (long) ptr->samplesperrow * SIZEOF(JSAMPLE);
-  file_offset = ptr->cur_start_row * bytesperrow;
+  file_offset = (long) ptr->cur_start_row * bytesperrow;
  /* Loop to read or write each allocation chunk in mem_buffer */
  for (i = 0; i < (long) ptr->rows_in_mem; i += ptr->rowsperchunk) {
    /* One chunk, but check for short chunk at end of buffer */
@@ -727,7 +723,7 @@ do_barray_io (j_common_ptr cinfo, jvirt_barray_ptr ptr, boolean writing)
  long bytesperrow, file_offset, byte_count, rows, thisrow, i;

  bytesperrow = (long) ptr->blocksperrow * SIZEOF(JBLOCK);
-  file_offset = ptr->cur_start_row * bytesperrow;
+  file_offset = (long) ptr->cur_start_row * bytesperrow;
  /* Loop to read or write each allocation chunk in mem_buffer */
  for (i = 0; i < (long) ptr->rows_in_mem; i += ptr->rowsperchunk) {
    /* One chunk, but check for short chunk at end of buffer */
@@ -771,7 +767,7 @@ access_virt_sarray (j_common_ptr cinfo, jvirt_sarray_ptr ptr,

  /* Make the desired part of the virtual array accessible */
  if (start_row < ptr->cur_start_row ||
-      end_row > ptr->cur_start_row+ptr->rows_in_mem) {
+      end_row > ptr->cur_start_row + ptr->rows_in_mem) {
    if (! ptr->b_s_open)
      ERREXIT(cinfo, JERR_VIRTUAL_BUG);
    /* Flush old buffer contents if necessary */
@@ -856,7 +852,7 @@ access_virt_barray (j_common_ptr cinfo, jvirt_barray_ptr ptr,

  /* Make the desired part of the virtual array accessible */
  if (start_row < ptr->cur_start_row ||
-      end_row > ptr->cur_start_row+ptr->rows_in_mem) {
+      end_row > ptr->cur_start_row + ptr->rows_in_mem) {
    if (! ptr->b_s_open)
      ERREXIT(cinfo, JERR_VIRTUAL_BUG);
    /* Flush old buffer contents if necessary */
@@ -1093,7 +1089,7 @@ jinit_memory_mgr (j_common_ptr cinfo)
  mem->total_space_allocated = SIZEOF(my_memory_mgr);

  /* Declare ourselves open for business */
-  cinfo->mem = & mem->pub;
+  cinfo->mem = &mem->pub;

  /* Check for an environment variable JPEGMEM; if found, override the
   * default max_memory setting from jpeg_mem_init.  Note that the
@@ -2,6 +2,7 @@
 * jmemnobs.c
 *
 * Copyright (C) 1992-1996, 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.
 *
@@ -12,7 +13,7 @@
 * This is very portable in the sense that it'll compile on almost anything,
 * but you'd better have lots of main memory (or virtual memory) if you want
 * to process big images.
- * Note that the max_memory_to_use option is ignored by this implementation.
+ * Note that the max_memory_to_use option is respected by this implementation.
 */

 #define JPEG_INTERNALS
@@ -66,13 +67,16 @@ jpeg_free_large (j_common_ptr cinfo, void FAR * object, size_t sizeofobject)

 /*
 * This routine computes the total memory space available for allocation.
- * Here we always say, "we got all you want bud!"
 */

 GLOBAL(long)
 jpeg_mem_available (j_common_ptr cinfo, long min_bytes_needed,
 		    long max_bytes_needed, long already_allocated)
 {
+  if (cinfo->mem->max_memory_to_use)
+    return cinfo->mem->max_memory_to_use - already_allocated;
+
+  /* Here we say, "we got all you want bud!" */
  return max_bytes_needed;
 }

@@ -2,7 +2,7 @@
 * jpegint.h
 *
 * Copyright (C) 1991-1997, Thomas G. Lane.
- * Modified 1997-2017 by Guido Vollbeding.
+ * Modified 1997-2019 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.
 *
@@ -302,6 +302,13 @@ struct jpeg_color_quantizer {
 #define RIGHT_SHIFT(x,shft)	((x) >> (shft))
 #endif

+/* Descale and correctly round an INT32 value that's scaled by N bits.
+ * We assume RIGHT_SHIFT rounds towards minus infinity, so adding
+ * the fudge factor is correct for either sign of X.
+ */
+
+#define DESCALE(x,n)	RIGHT_SHIFT((x) + ((INT32) 1 << ((n)-1)), n)
+

 /* Short forms of external names for systems with brain-damaged linkers. */

@@ -2,7 +2,7 @@
 * jpeglib.h
 *
 * Copyright (C) 1991-1998, Thomas G. Lane.
- * Modified 2002-2017 by Guido Vollbeding.
+ * Modified 2002-2019 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.
 *
@@ -39,7 +39,7 @@ extern "C" {

 #define JPEG_LIB_VERSION        90	/* Compatibility version 9.0 */
 #define JPEG_LIB_VERSION_MAJOR  9
-#define JPEG_LIB_VERSION_MINOR  3
+#define JPEG_LIB_VERSION_MINOR  4


 /* Various constants determining the sizes of things.
@@ -909,6 +909,7 @@ typedef JMETHOD(boolean, jpeg_marker_parser_method, (j_decompress_ptr cinfo));
 #define jpeg_suppress_tables	jSuppressTables
 #define jpeg_alloc_quant_table	jAlcQTable
 #define jpeg_alloc_huff_table	jAlcHTable
+#define jpeg_std_huff_table	jStdHTable
 #define jpeg_start_compress	jStrtCompress
 #define jpeg_write_scanlines	jWrtScanlines
 #define jpeg_finish_compress	jFinCompress
@@ -977,10 +978,10 @@ EXTERN(void) jpeg_stdio_src JPP((j_decompress_ptr cinfo, FILE * infile));
 /* Data source and destination managers: memory buffers. */
 EXTERN(void) jpeg_mem_dest JPP((j_compress_ptr cinfo,
 			       unsigned char ** outbuffer,
-			       unsigned long * outsize));
+			       size_t * outsize));
 EXTERN(void) jpeg_mem_src JPP((j_decompress_ptr cinfo,
 			      const unsigned char * inbuffer,
-			      unsigned long insize));
+			      size_t insize));

 /* Default parameter setup for compression */
 EXTERN(void) jpeg_set_defaults JPP((j_compress_ptr cinfo));
@@ -1005,6 +1006,8 @@ EXTERN(void) jpeg_suppress_tables JPP((j_compress_ptr cinfo,
 				       boolean suppress));
 EXTERN(JQUANT_TBL *) jpeg_alloc_quant_table JPP((j_common_ptr cinfo));
 EXTERN(JHUFF_TBL *) jpeg_alloc_huff_table JPP((j_common_ptr cinfo));
+EXTERN(JHUFF_TBL *) jpeg_std_huff_table JPP((j_common_ptr cinfo,
+					     boolean isDC, int tblno));

 /* Main entry points for compression */
 EXTERN(void) jpeg_start_compress JPP((j_compress_ptr cinfo,
@@ -2,7 +2,7 @@
 * jutils.c
 *
 * Copyright (C) 1991-1996, Thomas G. Lane.
- * Modified 2009-2011 by Guido Vollbeding.
+ * Modified 2009-2019 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.
 *
@@ -185,7 +185,7 @@ jcopy_sample_rows (JSAMPARRAY input_array, int source_row,
 {
  register JSAMPROW inptr, outptr;
 #ifdef FMEMCOPY
-  register size_t count = (size_t) (num_cols * SIZEOF(JSAMPLE));
+  register size_t count = (size_t) num_cols * SIZEOF(JSAMPLE);
 #else
  register JDIMENSION count;
 #endif
@@ -213,7 +213,7 @@ jcopy_block_row (JBLOCKROW input_row, JBLOCKROW output_row,
 /* Copy a row of coefficient blocks from one place to another. */
 {
 #ifdef FMEMCOPY
-  FMEMCOPY(output_row, input_row, num_blocks * (DCTSIZE2 * SIZEOF(JCOEF)));
+  FMEMCOPY(output_row, input_row, (size_t) num_blocks * (DCTSIZE2 * SIZEOF(JCOEF)));
 #else
  register JCOEFPTR inptr, outptr;
  register long count;
@@ -1,7 +1,7 @@
 /*
 * jversion.h
 *
- * Copyright (C) 1991-2018, Thomas G. Lane, Guido Vollbeding.
+ * Copyright (C) 1991-2020, Thomas G. Lane, Guido Vollbeding.
 * This file is part of the Independent JPEG Group's software.
 * For conditions of distribution and use, see the accompanying README file.
 *
@@ -9,6 +9,6 @@
 */


-#define JVERSION	"9c  14-Jan-2018"
+#define JVERSION	"9d  12-Jan-2020"

-#define JCOPYRIGHT	"Copyright (C) 2018, Thomas G. Lane, Guido Vollbeding"
+#define JCOPYRIGHT	"Copyright (C) 2020, Thomas G. Lane, Guido Vollbeding"
@@ -46,8 +46,19 @@ if(";${CPU_BASELINE_FINAL};" MATCHES "SSE2"
  add_definitions(-DPNG_INTEL_SSE)
 endif()

+# set definitions and sources for MIPS
+if(";${CPU_BASELINE_FINAL};" MATCHES "MSA")
+    list(APPEND lib_srcs mips/mips_init.c mips/filter_msa_intrinsics.c)
+    add_definitions(-DPNG_MIPS_MSA_OPT=2)
+    ocv_warnings_disable(CMAKE_C_FLAGS -Wshadow)
+else()
+    add_definitions(-DPNG_MIPS_MSA_OPT=0)
+endif()
+
 if(PPC64LE OR PPC64)
-  if(ENABLE_VSX AND NOT PPC64)
+  # VSX3 features are backwards compatible
+  if(";${CPU_BASELINE_FINAL};" MATCHES "VSX.*"
+      AND NOT PPC64)
    list(APPEND lib_srcs powerpc/powerpc_init.c powerpc/filter_vsx_intrinsics.c)
    add_definitions(-DPNG_POWERPC_VSX_OPT=2)
  else()
@@ -63,7 +74,7 @@ if(MSVC)
  add_definitions(-D_CRT_SECURE_NO_DEPRECATE)
 endif(MSVC)

-add_library(${PNG_LIBRARY} STATIC ${lib_srcs} ${lib_hdrs})
+add_library(${PNG_LIBRARY} STATIC ${OPENCV_3RDPARTY_EXCLUDE_FROM_ALL} ${lib_srcs} ${lib_hdrs})
 target_link_libraries(${PNG_LIBRARY} ${ZLIB_LIBRARIES})

 ocv_warnings_disable(CMAKE_C_FLAGS -Wundef -Wcast-align -Wimplicit-fallthrough -Wunused-parameter -Wsign-compare)
@@ -81,7 +92,7 @@ if(ENABLE_SOLUTION_FOLDERS)
 endif()

 if(NOT BUILD_SHARED_LIBS)
-  ocv_install_target(${PNG_LIBRARY} EXPORT OpenCVModules ARCHIVE DESTINATION ${OPENCV_3P_LIB_INSTALL_PATH} COMPONENT dev)
+  ocv_install_target(${PNG_LIBRARY} EXPORT OpenCVModules ARCHIVE DESTINATION ${OPENCV_3P_LIB_INSTALL_PATH} COMPONENT dev OPTIONAL)
 endif()

 ocv_install_3rdparty_licenses(libpng LICENSE README)
@@ -0,0 +1,808 @@
+
+/* filter_msa_intrinsics.c - MSA optimised filter functions
+ *
+ * Copyright (c) 2018 Cosmin Truta
+ * Copyright (c) 2016 Glenn Randers-Pehrson
+ * Written by Mandar Sahastrabuddhe, August 2016.
+ *
+ * This code is released under the libpng license.
+ * For conditions of distribution and use, see the disclaimer
+ * and license in png.h
+ */
+
+#include <stdio.h>
+#include <stdint.h>
+#include "../pngpriv.h"
+
+#ifdef PNG_READ_SUPPORTED
+
+/* This code requires -mfpu=msa on the command line: */
+#if PNG_MIPS_MSA_IMPLEMENTATION == 1 /* intrinsics code from pngpriv.h */
+
+#include <msa.h>
+
+/* libpng row pointers are not necessarily aligned to any particular boundary,
+ * however this code will only work with appropriate alignment. mips/mips_init.c
+ * checks for this (and will not compile unless it is done). This code uses
+ * variants of png_aligncast to avoid compiler warnings.
+ */
+#define png_ptr(type,pointer) png_aligncast(type *,pointer)
+#define png_ptrc(type,pointer) png_aligncastconst(const type *,pointer)
+
+/* The following relies on a variable 'temp_pointer' being declared with type
+ * 'type'.  This is written this way just to hide the GCC strict aliasing
+ * warning; note that the code is safe because there never is an alias between
+ * the input and output pointers.
+ */
+#define png_ldr(type,pointer)\
+   (temp_pointer = png_ptr(type,pointer), *temp_pointer)
+
+#if PNG_MIPS_MSA_OPT > 0
+
+#ifdef CLANG_BUILD
+   #define MSA_SRLI_B(a, b)   __msa_srli_b((v16i8) a, b)
+
+   #define LW(psrc)                              \
+   ( {                                           \
+       uint8_t *psrc_lw_m = (uint8_t *) (psrc);  \
+       uint32_t val_m;                           \
+                                                 \
+       asm volatile (                            \
+           "lw  %[val_m],  %[psrc_lw_m]  \n\t"   \
+                                                 \
+           : [val_m] "=r" (val_m)                \
+           : [psrc_lw_m] "m" (*psrc_lw_m)        \
+       );                                        \
+                                                 \
+       val_m;                                    \
+   } )
+
+   #define SH(val, pdst)                         \
+   {                                             \
+       uint8_t *pdst_sh_m = (uint8_t *) (pdst);  \
+       uint16_t val_m = (val);                   \
+                                                 \
+       asm volatile (                            \
+           "sh  %[val_m],  %[pdst_sh_m]  \n\t"   \
+                                                 \
+           : [pdst_sh_m] "=m" (*pdst_sh_m)       \
+           : [val_m] "r" (val_m)                 \
+       );                                        \
+   }
+
+   #define SW(val, pdst)                         \
+   {                                             \
+       uint8_t *pdst_sw_m = (uint8_t *) (pdst);  \
+       uint32_t val_m = (val);                   \
+                                                 \
+       asm volatile (                            \
+           "sw  %[val_m],  %[pdst_sw_m]  \n\t"   \
+                                                 \
+           : [pdst_sw_m] "=m" (*pdst_sw_m)       \
+           : [val_m] "r" (val_m)                 \
+       );                                        \
+   }
+
+       #if (__mips == 64)
+        #define SD(val, pdst)                         \
+        {                                             \
+            uint8_t *pdst_sd_m = (uint8_t *) (pdst);  \
+            uint64_t val_m = (val);                   \
+                                                      \
+            asm volatile (                            \
+                "sd  %[val_m],  %[pdst_sd_m]  \n\t"   \
+                                                      \
+                : [pdst_sd_m] "=m" (*pdst_sd_m)       \
+                : [val_m] "r" (val_m)                 \
+            );                                        \
+        }
+    #else
+        #define SD(val, pdst)                                          \
+        {                                                              \
+            uint8_t *pdst_sd_m = (uint8_t *) (pdst);                   \
+            uint32_t val0_m, val1_m;                                   \
+                                                                       \
+            val0_m = (uint32_t) ((val) & 0x00000000FFFFFFFF);          \
+            val1_m = (uint32_t) (((val) >> 32) & 0x00000000FFFFFFFF);  \
+                                                                       \
+            SW(val0_m, pdst_sd_m);                                     \
+            SW(val1_m, pdst_sd_m + 4);                                 \
+        }
+    #endif
+#else
+   #define MSA_SRLI_B(a, b)   (a >> b)
+
+#if (__mips_isa_rev >= 6)
+   #define LW(psrc)                              \
+   ( {                                           \
+       uint8_t *psrc_lw_m = (uint8_t *) (psrc);  \
+       uint32_t val_m;                           \
+                                                 \
+       asm volatile (                            \
+           "lw  %[val_m],  %[psrc_lw_m]  \n\t"   \
+                                                 \
+           : [val_m] "=r" (val_m)                \
+           : [psrc_lw_m] "m" (*psrc_lw_m)        \
+       );                                        \
+                                                 \
+       val_m;                                    \
+   } )
+
+   #define SH(val, pdst)                         \
+   {                                             \
+       uint8_t *pdst_sh_m = (uint8_t *) (pdst);  \
+       uint16_t val_m = (val);                   \
+                                                 \
+       asm volatile (                            \
+           "sh  %[val_m],  %[pdst_sh_m]  \n\t"   \
+                                                 \
+           : [pdst_sh_m] "=m" (*pdst_sh_m)       \
+           : [val_m] "r" (val_m)                 \
+       );                                        \
+   }
+
+   #define SW(val, pdst)                         \
+   {                                             \
+       uint8_t *pdst_sw_m = (uint8_t *) (pdst);  \
+       uint32_t val_m = (val);                   \
+                                                 \
+       asm volatile (                            \
+           "sw  %[val_m],  %[pdst_sw_m]  \n\t"   \
+                                                 \
+           : [pdst_sw_m] "=m" (*pdst_sw_m)       \
+           : [val_m] "r" (val_m)                 \
+       );                                        \
+   }
+
+   #if (__mips == 64)
+        #define SD(val, pdst)                         \
+        {                                             \
+            uint8_t *pdst_sd_m = (uint8_t *) (pdst);  \
+            uint64_t val_m = (val);                   \
+                                                      \
+            asm volatile (                            \
+                "sd  %[val_m],  %[pdst_sd_m]  \n\t"   \
+                                                      \
+                : [pdst_sd_m] "=m" (*pdst_sd_m)       \
+                : [val_m] "r" (val_m)                 \
+            );                                        \
+        }
+    #else
+        #define SD(val, pdst)                                          \
+        {                                                              \
+            uint8_t *pdst_sd_m = (uint8_t *) (pdst);                   \
+            uint32_t val0_m, val1_m;                                   \
+                                                                       \
+            val0_m = (uint32_t) ((val) & 0x00000000FFFFFFFF);          \
+            val1_m = (uint32_t) (((val) >> 32) & 0x00000000FFFFFFFF);  \
+                                                                       \
+            SW(val0_m, pdst_sd_m);                                     \
+            SW(val1_m, pdst_sd_m + 4);                                 \
+        }
+    #endif
+#else  // !(__mips_isa_rev >= 6)
+   #define LW(psrc)                              \
+   ( {                                           \
+       uint8_t *psrc_lw_m = (uint8_t *) (psrc);  \
+       uint32_t val_m;                           \
+                                                 \
+       asm volatile (                            \
+           "ulw  %[val_m],  %[psrc_lw_m]  \n\t"  \
+                                                 \
+           : [val_m] "=r" (val_m)                \
+           : [psrc_lw_m] "m" (*psrc_lw_m)        \
+       );                                        \
+                                                 \
+       val_m;                                    \
+   } )
+
+   #define SH(val, pdst)                         \
+   {                                             \
+       uint8_t *pdst_sh_m = (uint8_t *) (pdst);  \
+       uint16_t val_m = (val);                   \
+                                                 \
+       asm volatile (                            \
+           "ush  %[val_m],  %[pdst_sh_m]  \n\t"  \
+                                                 \
+           : [pdst_sh_m] "=m" (*pdst_sh_m)       \
+           : [val_m] "r" (val_m)                 \
+       );                                        \
+   }
+
+   #define SW(val, pdst)                         \
+   {                                             \
+       uint8_t *pdst_sw_m = (uint8_t *) (pdst);  \
+       uint32_t val_m = (val);                   \
+                                                 \
+       asm volatile (                            \
+           "usw  %[val_m],  %[pdst_sw_m]  \n\t"  \
+                                                 \
+           : [pdst_sw_m] "=m" (*pdst_sw_m)       \
+           : [val_m] "r" (val_m)                 \
+       );                                        \
+   }
+
+   #define SD(val, pdst)                                          \
+    {                                                              \
+        uint8_t *pdst_sd_m = (uint8_t *) (pdst);                   \
+        uint32_t val0_m, val1_m;                                   \
+                                                                   \
+        val0_m = (uint32_t) ((val) & 0x00000000FFFFFFFF);          \
+        val1_m = (uint32_t) (((val) >> 32) & 0x00000000FFFFFFFF);  \
+                                                                   \
+        SW(val0_m, pdst_sd_m);                                     \
+        SW(val1_m, pdst_sd_m + 4);                                 \
+    }
+
+    #define SW_ZERO(pdst)                      \
+    {                                          \
+        uint8_t *pdst_m = (uint8_t *) (pdst);  \
+                                               \
+        asm volatile (                         \
+            "usw  $0,  %[pdst_m]  \n\t"        \
+                                               \
+            : [pdst_m] "=m" (*pdst_m)          \
+            :                                  \
+        );                                     \
+    }
+#endif  // (__mips_isa_rev >= 6)
+#endif
+
+#define LD_B(RTYPE, psrc) *((RTYPE *) (psrc))
+#define LD_UB(...) LD_B(v16u8, __VA_ARGS__)
+#define LD_B2(RTYPE, psrc, stride, out0, out1)  \
+{                                               \
+    out0 = LD_B(RTYPE, (psrc));                 \
+    out1 = LD_B(RTYPE, (psrc) + stride);        \
+}
+#define LD_UB2(...) LD_B2(v16u8, __VA_ARGS__)
+#define LD_B4(RTYPE, psrc, stride, out0, out1, out2, out3)   \
+{                                                            \
+    LD_B2(RTYPE, (psrc), stride, out0, out1);                \
+    LD_B2(RTYPE, (psrc) + 2 * stride , stride, out2, out3);  \
+}
+#define LD_UB4(...) LD_B4(v16u8, __VA_ARGS__)
+
+#define ST_B(RTYPE, in, pdst) *((RTYPE *) (pdst)) = (in)
+#define ST_UB(...) ST_B(v16u8, __VA_ARGS__)
+#define ST_B2(RTYPE, in0, in1, pdst, stride)  \
+{                                             \
+    ST_B(RTYPE, in0, (pdst));                 \
+    ST_B(RTYPE, in1, (pdst) + stride);        \
+}
+#define ST_UB2(...) ST_B2(v16u8, __VA_ARGS__)
+#define ST_B4(RTYPE, in0, in1, in2, in3, pdst, stride)    \
+{                                                         \
+    ST_B2(RTYPE, in0, in1, (pdst), stride);               \
+    ST_B2(RTYPE, in2, in3, (pdst) + 2 * stride, stride);  \
+}
+#define ST_UB4(...) ST_B4(v16u8, __VA_ARGS__)
+
+#define ADD2(in0, in1, in2, in3, out0, out1)  \
+{                                             \
+    out0 = in0 + in1;                         \
+    out1 = in2 + in3;                         \
+}
+#define ADD3(in0, in1, in2, in3, in4, in5,  \
+             out0, out1, out2)              \
+{                                           \
+    ADD2(in0, in1, in2, in3, out0, out1);   \
+    out2 = in4 + in5;                       \
+}
+#define ADD4(in0, in1, in2, in3, in4, in5, in6, in7,  \
+             out0, out1, out2, out3)                  \
+{                                                     \
+    ADD2(in0, in1, in2, in3, out0, out1);             \
+    ADD2(in4, in5, in6, in7, out2, out3);             \
+}
+
+#define ILVR_B2(RTYPE, in0, in1, in2, in3, out0, out1)      \
+{                                                           \
+    out0 = (RTYPE) __msa_ilvr_b((v16i8) in0, (v16i8) in1);  \
+    out1 = (RTYPE) __msa_ilvr_b((v16i8) in2, (v16i8) in3);  \
+}
+#define ILVR_B2_SH(...) ILVR_B2(v8i16, __VA_ARGS__)
+
+#define HSUB_UB2(RTYPE, in0, in1, out0, out1)                 \
+{                                                             \
+    out0 = (RTYPE) __msa_hsub_u_h((v16u8) in0, (v16u8) in0);  \
+    out1 = (RTYPE) __msa_hsub_u_h((v16u8) in1, (v16u8) in1);  \
+}
+#define HSUB_UB2_SH(...) HSUB_UB2(v8i16, __VA_ARGS__)
+
+#define SLDI_B2_0(RTYPE, in0, in1, out0, out1, slide_val)                 \
+{                                                                         \
+    v16i8 zero_m = { 0 };                                                 \
+    out0 = (RTYPE) __msa_sldi_b((v16i8) zero_m, (v16i8) in0, slide_val);  \
+    out1 = (RTYPE) __msa_sldi_b((v16i8) zero_m, (v16i8) in1, slide_val);  \
+}
+#define SLDI_B2_0_UB(...) SLDI_B2_0(v16u8, __VA_ARGS__)
+
+#define SLDI_B3_0(RTYPE, in0, in1, in2, out0, out1, out2,  slide_val)     \
+{                                                                         \
+    v16i8 zero_m = { 0 };                                                 \
+    SLDI_B2_0(RTYPE, in0, in1, out0, out1, slide_val);                    \
+    out2 = (RTYPE) __msa_sldi_b((v16i8) zero_m, (v16i8) in2, slide_val);  \
+}
+#define SLDI_B3_0_UB(...) SLDI_B3_0(v16u8, __VA_ARGS__)
+
+#define ILVEV_W2(RTYPE, in0, in1, in2, in3, out0, out1)      \
+{                                                            \
+    out0 = (RTYPE) __msa_ilvev_w((v4i32) in1, (v4i32) in0);  \
+    out1 = (RTYPE) __msa_ilvev_w((v4i32) in3, (v4i32) in2);  \
+}
+#define ILVEV_W2_UB(...) ILVEV_W2(v16u8, __VA_ARGS__)
+
+#define ADD_ABS_H3(RTYPE, in0, in1, in2, out0, out1, out2)  \
+{                                                           \
+    RTYPE zero = {0};                                       \
+                                                            \
+    out0 = __msa_add_a_h((v8i16) zero, in0);                \
+    out1 = __msa_add_a_h((v8i16) zero, in1);                \
+    out2 = __msa_add_a_h((v8i16) zero, in2);                \
+}
+#define ADD_ABS_H3_SH(...) ADD_ABS_H3(v8i16, __VA_ARGS__)
+
+#define VSHF_B2(RTYPE, in0, in1, in2, in3, mask0, mask1, out0, out1)       \
+{                                                                          \
+    out0 = (RTYPE) __msa_vshf_b((v16i8) mask0, (v16i8) in1, (v16i8) in0);  \
+    out1 = (RTYPE) __msa_vshf_b((v16i8) mask1, (v16i8) in3, (v16i8) in2);  \
+}
+#define VSHF_B2_UB(...) VSHF_B2(v16u8, __VA_ARGS__)
+
+#define CMP_AND_SELECT(inp0, inp1, inp2, inp3, inp4, inp5, out0)              \
+{                                                                             \
+   v8i16 _sel_h0, _sel_h1;                                                    \
+   v16u8 _sel_b0, _sel_b1;                                                    \
+   _sel_h0 = (v8i16) __msa_clt_u_h((v8u16) inp1, (v8u16) inp0);               \
+   _sel_b0 = (v16u8) __msa_pckev_b((v16i8) _sel_h0, (v16i8) _sel_h0);         \
+   inp0 = (v8i16) __msa_bmnz_v((v16u8) inp0, (v16u8) inp1, (v16u8) _sel_h0);  \
+   inp4 = (v16u8) __msa_bmnz_v(inp3, inp4, _sel_b0);                          \
+   _sel_h1 = (v8i16) __msa_clt_u_h((v8u16) inp2, (v8u16) inp0);               \
+   _sel_b1 = (v16u8) __msa_pckev_b((v16i8) _sel_h1, (v16i8) _sel_h1);         \
+   inp4 = (v16u8) __msa_bmnz_v(inp4, inp5, _sel_b1);                          \
+   out0 += inp4;                                                              \
+}
+
+void png_read_filter_row_up_msa(png_row_infop row_info, png_bytep row,
+                                png_const_bytep prev_row)
+{
+   size_t i, cnt, cnt16, cnt32;
+   size_t istop = row_info->rowbytes;
+   png_bytep rp = row;
+   png_const_bytep pp = prev_row;
+   v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
+
+   for (i = 0; i < (istop >> 6); i++)
+   {
+      LD_UB4(rp, 16, src0, src1, src2, src3);
+      LD_UB4(pp, 16, src4, src5, src6, src7);
+      pp += 64;
+
+	  ADD4(src0, src4, src1, src5, src2, src6, src3, src7,
+	       src0, src1, src2, src3);
+
+      ST_UB4(src0, src1, src2, src3, rp, 16);
+      rp += 64;
+   }
+
+   if (istop & 0x3F)
+   {
+      cnt32 = istop & 0x20;
+      cnt16 = istop & 0x10;
+      cnt = istop & 0xF;
+
+      if(cnt32)
+      {
+         if (cnt16 && cnt)
+         {
+            LD_UB4(rp, 16, src0, src1, src2, src3);
+            LD_UB4(pp, 16, src4, src5, src6, src7);
+
+            ADD4(src0, src4, src1, src5, src2, src6, src3, src7,
+	             src0, src1, src2, src3);
+
+            ST_UB4(src0, src1, src2, src3, rp, 16);
+            rp += 64;
+         }
+         else if (cnt16 || cnt)
+         {
+            LD_UB2(rp, 16, src0, src1);
+            LD_UB2(pp, 16, src4, src5);
+            pp += 32;
+            src2 = LD_UB(rp + 32);
+            src6 = LD_UB(pp);
+
+            ADD3(src0, src4, src1, src5, src2, src6, src0, src1, src2);
+
+            ST_UB2(src0, src1, rp, 16);
+            rp += 32;
+            ST_UB(src2, rp);
+            rp += 16;
+         }
+         else
+         {
+            LD_UB2(rp, 16, src0, src1);
+            LD_UB2(pp, 16, src4, src5);
+
+			ADD2(src0, src4, src1, src5, src0, src1);
+
+            ST_UB2(src0, src1, rp, 16);
+            rp += 32;
+         }
+      }
+      else if (cnt16 && cnt)
+      {
+         LD_UB2(rp, 16, src0, src1);
+         LD_UB2(pp, 16, src4, src5);
+
+         ADD2(src0, src4, src1, src5, src0, src1);
+
+         ST_UB2(src0, src1, rp, 16);
+         rp += 32;
+      }
+      else if (cnt16 || cnt)
+      {
+         src0 = LD_UB(rp);
+         src4 = LD_UB(pp);
+         pp += 16;
+
+         src0 += src4;
+
+         ST_UB(src0, rp);
+         rp += 16;
+      }
+   }
+}
+
+void png_read_filter_row_sub4_msa(png_row_infop row_info, png_bytep row,
+                                  png_const_bytep prev_row)
+{
+   size_t count;
+   size_t istop = row_info->rowbytes;
+   png_bytep src = row;
+   png_bytep nxt = row + 4;
+   int32_t inp0;
+   v16u8 src0, src1, src2, src3, src4;
+   v16u8 dst0, dst1;
+   v16u8 zero = { 0 };
+
+   istop -= 4;
+
+   inp0 = LW(src);
+   src += 4;
+   src0 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp0);
+
+   for (count = 0; count < istop; count += 16)
+   {
+      src1 = LD_UB(src);
+      src += 16;
+
+      src2 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) src1, 4);
+      src3 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) src1, 8);
+      src4 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) src1, 12);
+      src1 += src0;
+      src2 += src1;
+      src3 += src2;
+      src4 += src3;
+      src0 = src4;
+      ILVEV_W2_UB(src1, src2, src3, src4, dst0, dst1);
+      dst0 = (v16u8) __msa_pckev_d((v2i64) dst1, (v2i64) dst0);
+
+      ST_UB(dst0, nxt);
+      nxt += 16;
+   }
+}
+
+void png_read_filter_row_sub3_msa(png_row_infop row_info, png_bytep row,
+                                  png_const_bytep prev_row)
+{
+   size_t count;
+   size_t istop = row_info->rowbytes;
+   png_bytep src = row;
+   png_bytep nxt = row + 3;
+   int64_t out0;
+   int32_t inp0, out1;
+   v16u8 src0, src1, src2, src3, src4, dst0, dst1;
+   v16u8 zero = { 0 };
+   v16i8 mask0 = { 0, 1, 2, 16, 17, 18, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+   v16i8 mask1 = { 0, 1, 2, 3, 4, 5, 16, 17, 18, 19, 20, 21, 0, 0, 0, 0 };
+
+   istop -= 3;
+
+   inp0 = LW(src);
+   src += 3;
+   src0 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp0);
+
+   for (count = 0; count < istop; count += 12)
+   {
+      src1 = LD_UB(src);
+      src += 12;
+
+      src2 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) src1, 3);
+      src3 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) src1, 6);
+      src4 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) src1, 9);
+      src1 += src0;
+      src2 += src1;
+      src3 += src2;
+      src4 += src3;
+      src0 = src4;
+      VSHF_B2_UB(src1, src2, src3, src4, mask0, mask0, dst0, dst1);
+      dst0 = (v16u8) __msa_vshf_b(mask1, (v16i8) dst1, (v16i8) dst0);
+      out0 = __msa_copy_s_d((v2i64) dst0, 0);
+      out1 = __msa_copy_s_w((v4i32) dst0, 2);
+
+      SD(out0, nxt);
+      nxt += 8;
+      SW(out1, nxt);
+      nxt += 4;
+   }
+}
+
+void png_read_filter_row_avg4_msa(png_row_infop row_info, png_bytep row,
+                                  png_const_bytep prev_row)
+{
+   size_t i;
+   png_bytep src = row;
+   png_bytep nxt = row;
+   png_const_bytep pp = prev_row;
+   size_t istop = row_info->rowbytes - 4;
+   int32_t inp0, inp1, out0;
+   v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, dst0, dst1;
+   v16u8 zero = { 0 };
+
+   inp0 = LW(pp);
+   pp += 4;
+   inp1 = LW(src);
+   src += 4;
+   src0 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp0);
+   src1 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp1);
+   src0 = (v16u8) MSA_SRLI_B(src0, 1);
+   src1 += src0;
+   out0 = __msa_copy_s_w((v4i32) src1, 0);
+   SW(out0, nxt);
+   nxt += 4;
+
+   for (i = 0; i < istop; i += 16)
+   {
+      src2 = LD_UB(pp);
+      pp += 16;
+      src6 = LD_UB(src);
+      src += 16;
+
+      SLDI_B2_0_UB(src2, src6, src3, src7, 4);
+      SLDI_B2_0_UB(src2, src6, src4, src8, 8);
+      SLDI_B2_0_UB(src2, src6, src5, src9, 12);
+      src2 = __msa_ave_u_b(src2, src1);
+      src6 += src2;
+      src3 = __msa_ave_u_b(src3, src6);
+      src7 += src3;
+      src4 = __msa_ave_u_b(src4, src7);
+      src8 += src4;
+      src5 = __msa_ave_u_b(src5, src8);
+      src9 += src5;
+      src1 = src9;
+      ILVEV_W2_UB(src6, src7, src8, src9, dst0, dst1);
+      dst0 = (v16u8) __msa_pckev_d((v2i64) dst1, (v2i64) dst0);
+
+      ST_UB(dst0, nxt);
+      nxt += 16;
+   }
+}
+
+void png_read_filter_row_avg3_msa(png_row_infop row_info, png_bytep row,
+                                  png_const_bytep prev_row)
+{
+   size_t i;
+   png_bytep src = row;
+   png_bytep nxt = row;
+   png_const_bytep pp = prev_row;
+   size_t istop = row_info->rowbytes - 3;
+   int64_t out0;
+   int32_t inp0, inp1, out1;
+   int16_t out2;
+   v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, dst0, dst1;
+   v16u8 zero = { 0 };
+   v16i8 mask0 = { 0, 1, 2, 16, 17, 18, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+   v16i8 mask1 = { 0, 1, 2, 3, 4, 5, 16, 17, 18, 19, 20, 21, 0, 0, 0, 0 };
+
+   inp0 = LW(pp);
+   pp += 3;
+   inp1 = LW(src);
+   src += 3;
+   src0 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp0);
+   src1 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp1);
+   src0 = (v16u8) MSA_SRLI_B(src0, 1);
+   src1 += src0;
+   out2 = __msa_copy_s_h((v8i16) src1, 0);
+   SH(out2, nxt);
+   nxt += 2;
+   nxt[0] = src1[2];
+   nxt++;
+
+   for (i = 0; i < istop; i += 12)
+   {
+      src2 = LD_UB(pp);
+      pp += 12;
+      src6 = LD_UB(src);
+      src += 12;
+
+      SLDI_B2_0_UB(src2, src6, src3, src7, 3);
+      SLDI_B2_0_UB(src2, src6, src4, src8, 6);
+      SLDI_B2_0_UB(src2, src6, src5, src9, 9);
+      src2 = __msa_ave_u_b(src2, src1);
+      src6 += src2;
+      src3 = __msa_ave_u_b(src3, src6);
+      src7 += src3;
+      src4 = __msa_ave_u_b(src4, src7);
+      src8 += src4;
+      src5 = __msa_ave_u_b(src5, src8);
+      src9 += src5;
+      src1 = src9;
+      VSHF_B2_UB(src6, src7, src8, src9, mask0, mask0, dst0, dst1);
+      dst0 = (v16u8) __msa_vshf_b(mask1, (v16i8) dst1, (v16i8) dst0);
+      out0 = __msa_copy_s_d((v2i64) dst0, 0);
+      out1 = __msa_copy_s_w((v4i32) dst0, 2);
+
+      SD(out0, nxt);
+      nxt += 8;
+      SW(out1, nxt);
+      nxt += 4;
+   }
+}
+
+void png_read_filter_row_paeth4_msa(png_row_infop row_info,
+                                    png_bytep row,
+                                    png_const_bytep prev_row)
+{
+   int32_t count, rp_end;
+   png_bytep nxt;
+   png_const_bytep prev_nxt;
+   int32_t inp0, inp1, res0;
+   v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9;
+   v16u8 src10, src11, src12, src13, dst0, dst1;
+   v8i16 vec0, vec1, vec2;
+   v16u8 zero = { 0 };
+
+   nxt = row;
+   prev_nxt = prev_row;
+
+   inp0 = LW(nxt);
+   inp1 = LW(prev_nxt);
+   prev_nxt += 4;
+   src0 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp0);
+   src1 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp1);
+
+   src1 += src0;
+   res0 = __msa_copy_s_w((v4i32) src1, 0);
+
+   SW(res0, nxt);
+   nxt += 4;
+
+   /* Remainder */
+   rp_end = row_info->rowbytes - 4;
+
+   for (count = 0; count < rp_end; count += 16)
+   {
+      src2 = LD_UB(prev_nxt);
+      prev_nxt += 16;
+      src6 = LD_UB(prev_row);
+      prev_row += 16;
+      src10 = LD_UB(nxt);
+
+      SLDI_B3_0_UB(src2, src6, src10, src3, src7, src11, 4);
+      SLDI_B3_0_UB(src2, src6, src10, src4, src8, src12, 8);
+      SLDI_B3_0_UB(src2, src6, src10, src5, src9, src13, 12);
+      ILVR_B2_SH(src2, src6, src1, src6, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src1, src2, src6, src10);
+      ILVR_B2_SH(src3, src7, src10, src7, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src10, src3, src7, src11);
+      ILVR_B2_SH(src4, src8, src11, src8, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src11, src4, src8, src12);
+      ILVR_B2_SH(src5, src9, src12, src9, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src12, src5, src9, src13);
+      src1 = src13;
+      ILVEV_W2_UB(src10, src11, src12, src1, dst0, dst1);
+      dst0 = (v16u8) __msa_pckev_d((v2i64) dst1, (v2i64) dst0);
+
+      ST_UB(dst0, nxt);
+      nxt += 16;
+   }
+}
+
+void png_read_filter_row_paeth3_msa(png_row_infop row_info,
+                                    png_bytep row,
+                                    png_const_bytep prev_row)
+{
+   int32_t count, rp_end;
+   png_bytep nxt;
+   png_const_bytep prev_nxt;
+   int64_t out0;
+   int32_t inp0, inp1, out1;
+   int16_t out2;
+   v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, dst0, dst1;
+   v16u8 src10, src11, src12, src13;
+   v8i16 vec0, vec1, vec2;
+   v16u8 zero = { 0 };
+   v16i8 mask0 = { 0, 1, 2, 16, 17, 18, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+   v16i8 mask1 = { 0, 1, 2, 3, 4, 5, 16, 17, 18, 19, 20, 21, 0, 0, 0, 0 };
+
+   nxt = row;
+   prev_nxt = prev_row;
+
+   inp0 = LW(nxt);
+   inp1 = LW(prev_nxt);
+   prev_nxt += 3;
+   src0 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp0);
+   src1 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp1);
+
+   src1 += src0;
+   out2 = __msa_copy_s_h((v8i16) src1, 0);
+
+   SH(out2, nxt);
+   nxt += 2;
+   nxt[0] = src1[2];
+   nxt++;
+
+   /* Remainder */
+   rp_end = row_info->rowbytes - 3;
+
+   for (count = 0; count < rp_end; count += 12)
+   {
+      src2 = LD_UB(prev_nxt);
+      prev_nxt += 12;
+      src6 = LD_UB(prev_row);
+      prev_row += 12;
+      src10 = LD_UB(nxt);
+
+      SLDI_B3_0_UB(src2, src6, src10, src3, src7, src11, 3);
+      SLDI_B3_0_UB(src2, src6, src10, src4, src8, src12, 6);
+      SLDI_B3_0_UB(src2, src6, src10, src5, src9, src13, 9);
+      ILVR_B2_SH(src2, src6, src1, src6, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src1, src2, src6, src10);
+      ILVR_B2_SH(src3, src7, src10, src7, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src10, src3, src7, src11);
+      ILVR_B2_SH(src4, src8, src11, src8, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src11, src4, src8, src12);
+      ILVR_B2_SH(src5, src9, src12, src9, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src12, src5, src9, src13);
+      src1 = src13;
+      VSHF_B2_UB(src10, src11, src12, src13, mask0, mask0, dst0, dst1);
+      dst0 = (v16u8) __msa_vshf_b(mask1, (v16i8) dst1, (v16i8) dst0);
+      out0 = __msa_copy_s_d((v2i64) dst0, 0);
+      out1 = __msa_copy_s_w((v4i32) dst0, 2);
+
+      SD(out0, nxt);
+      nxt += 8;
+      SW(out1, nxt);
+      nxt += 4;
+   }
+}
+
+#endif /* PNG_MIPS_MSA_OPT > 0 */
+#endif /* PNG_MIPS_MSA_IMPLEMENTATION == 1 (intrinsics) */
+#endif /* READ */
@@ -0,0 +1,127 @@
+
+/* mips_init.c - MSA optimised filter functions
+ *
+ * Copyright (c) 2018 Cosmin Truta
+ * Copyright (c) 2016 Glenn Randers-Pehrson
+ * Written by Mandar Sahastrabuddhe, 2016.
+ *
+ * This code is released under the libpng license.
+ * For conditions of distribution and use, see the disclaimer
+ * and license in png.h
+ */
+
+/* Below, after checking __linux__, various non-C90 POSIX 1003.1 functions are
+ * called.
+ */
+#define _POSIX_SOURCE 1
+
+#include <stdio.h>
+#include "../pngpriv.h"
+
+#ifdef PNG_READ_SUPPORTED
+
+#if PNG_MIPS_MSA_OPT > 0
+#ifdef PNG_MIPS_MSA_CHECK_SUPPORTED /* Do run-time checks */
+/* WARNING: it is strongly recommended that you do not build libpng with
+ * run-time checks for CPU features if at all possible.  In the case of the MIPS
+ * MSA instructions there is no processor-specific way of detecting the
+ * presence of the required support, therefore run-time detection is extremely
+ * OS specific.
+ *
+ * You may set the macro PNG_MIPS_MSA_FILE to the file name of file containing
+ * a fragment of C source code which defines the png_have_msa function.  There
+ * are a number of implementations in contrib/mips-msa, but the only one that
+ * has partial support is contrib/mips-msa/linux.c - a generic Linux
+ * implementation which reads /proc/cpufino.
+ */
+#ifndef PNG_MIPS_MSA_FILE
+#  ifdef __linux__
+#     define PNG_MIPS_MSA_FILE "contrib/mips-msa/linux.c"
+#  endif
+#endif
+
+#ifdef PNG_MIPS_MSA_FILE
+
+#include <signal.h> /* for sig_atomic_t */
+static int png_have_msa(png_structp png_ptr);
+#include PNG_MIPS_MSA_FILE
+
+#else  /* PNG_MIPS_MSA_FILE */
+#  error "PNG_MIPS_MSA_FILE undefined: no support for run-time MIPS MSA checks"
+#endif /* PNG_MIPS_MSA_FILE */
+#endif /* PNG_MIPS_MSA_CHECK_SUPPORTED */
+
+#ifndef PNG_ALIGNED_MEMORY_SUPPORTED
+#  error "ALIGNED_MEMORY is required; set: -DPNG_ALIGNED_MEMORY_SUPPORTED"
+#endif
+
+void
+png_init_filter_functions_msa(png_structp pp, unsigned int bpp)
+{
+   /* The switch statement is compiled in for MIPS_MSA_API, the call to
+    * png_have_msa is compiled in for MIPS_MSA_CHECK. If both are defined
+    * the check is only performed if the API has not set the MSA option on
+    * or off explicitly. In this case the check controls what happens.
+    */
+
+#ifdef PNG_MIPS_MSA_API_SUPPORTED
+   switch ((pp->options >> PNG_MIPS_MSA) & 3)
+   {
+      case PNG_OPTION_UNSET:
+         /* Allow the run-time check to execute if it has been enabled -
+          * thus both API and CHECK can be turned on.  If it isn't supported
+          * this case will fall through to the 'default' below, which just
+          * returns.
+          */
+#ifdef PNG_MIPS_MSA_CHECK_SUPPORTED
+         {
+            static volatile sig_atomic_t no_msa = -1; /* not checked */
+
+            if (no_msa < 0)
+               no_msa = !png_have_msa(pp);
+
+            if (no_msa)
+               return;
+         }
+#endif /* PNG_MIPS_MSA_CHECK_SUPPORTED */
+         break;
+
+      default: /* OFF or INVALID */
+         return;
+
+      case PNG_OPTION_ON:
+         /* Option turned on */
+         break;
+   }
+   /* IMPORTANT: any new external functions used here must be declared using
+    * PNG_INTERNAL_FUNCTION in ../pngpriv.h.  This is required so that the
+    * 'prefix' option to configure works:
+    *
+    *    ./configure --with-libpng-prefix=foobar_
+    *
+    * Verify you have got this right by running the above command, doing a build
+    * and examining pngprefix.h; it must contain a #define for every external
+    * function you add.  (Notice that this happens automatically for the
+    * initialization function.)
+    */
+   pp->read_filter[PNG_FILTER_VALUE_UP-1] = png_read_filter_row_up_msa;
+
+   if (bpp == 3)
+   {
+      pp->read_filter[PNG_FILTER_VALUE_SUB-1] = png_read_filter_row_sub3_msa;
+      pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg3_msa;
+      pp->read_filter[PNG_FILTER_VALUE_PAETH-1] = png_read_filter_row_paeth3_msa;
+   }
+   else if (bpp == 4)
+   {
+      pp->read_filter[PNG_FILTER_VALUE_SUB-1] = png_read_filter_row_sub4_msa;
+      pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg4_msa;
+      pp->read_filter[PNG_FILTER_VALUE_PAETH-1] = png_read_filter_row_paeth4_msa;
+   }
+#else
+   (void)pp;
+   (void)bpp;
+#endif /* PNG_MIPS_MSA_API_SUPPORTED */
+}
+#endif /* PNG_MIPS_MSA_OPT > 0 */
+#endif /* READ */
@@ -0,0 +1,53 @@
+diff --git a/3rdparty/libpng/mips/mips_init.c b/3rdparty/libpng/mips/mips_init.c
+index 8dd283deef..6a061cccfa 100644
+--- a/3rdparty/libpng/mips/mips_init.c
+++ b/3rdparty/libpng/mips/mips_init.c
+@@ -73,7 +73,6 @@ png_init_filter_functions_msa(png_structp pp, unsigned int bpp)
+           * this case will fall through to the 'default' below, which just
+           * returns.
+           */
+-#endif /* PNG_MIPS_MSA_API_SUPPORTED */
+ #ifdef PNG_MIPS_MSA_CHECK_SUPPORTED
+          {
+             static volatile sig_atomic_t no_msa = -1; /* not checked */
+@@ -84,12 +83,9 @@ png_init_filter_functions_msa(png_structp pp, unsigned int bpp)
+             if (no_msa)
+                return;
+          }
+-#ifdef PNG_MIPS_MSA_API_SUPPORTED
+-         break;
+-#endif
+ #endif /* PNG_MIPS_MSA_CHECK_SUPPORTED */
+         break;
+ 
+-#ifdef PNG_MIPS_MSA_API_SUPPORTED
+       default: /* OFF or INVALID */
+          return;
+ 
+@@ -97,8 +93,6 @@ png_init_filter_functions_msa(png_structp pp, unsigned int bpp)
+          /* Option turned on */
+          break;
+    }
+-#endif
+-
+    /* IMPORTANT: any new external functions used here must be declared using
+     * PNG_INTERNAL_FUNCTION in ../pngpriv.h.  This is required so that the
+     * 'prefix' option to configure works:
+@@ -118,13 +112,16 @@ png_init_filter_functions_msa(png_structp pp, unsigned int bpp)
+       pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg3_msa;
+       pp->read_filter[PNG_FILTER_VALUE_PAETH-1] = png_read_filter_row_paeth3_msa;
+    }
+-
+    else if (bpp == 4)
+    {
+       pp->read_filter[PNG_FILTER_VALUE_SUB-1] = png_read_filter_row_sub4_msa;
+       pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg4_msa;
+       pp->read_filter[PNG_FILTER_VALUE_PAETH-1] = png_read_filter_row_paeth4_msa;
+    }
+#else
+   (void)pp;
+   (void)bpp;
+#endif /* PNG_MIPS_MSA_API_SUPPORTED */
+ }
+ #endif /* PNG_MIPS_MSA_OPT > 0 */
+ #endif /* READ */
@@ -0,0 +1,89 @@
+# COPYRIGHT
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# License); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# AS IS BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+#
+# Copyright (c) 2020, OPEN AI LAB
+# Author: qtang@openailab.com or https://github.com/BUG1989
+#         qli@openailab.com
+#         sqfu@openailab.com
+#
+
+SET(TENGINE_COMMIT_VERSION "8a4c58e0e05cd850f4bb0936a330edc86dc0e28c")
+SET(OCV_TENGINE_DIR "${OpenCV_BINARY_DIR}/3rdparty/libtengine")
+SET(OCV_TENGINE_SOURCE_PATH "${OCV_TENGINE_DIR}/Tengine-${TENGINE_COMMIT_VERSION}")
+
+IF(EXISTS "${OCV_TENGINE_SOURCE_PATH}")
+	MESSAGE(STATUS "Tengine is exist already at: ${OCV_TENGINE_SOURCE_PATH}")
+
+	SET(Tengine_FOUND ON)
+	SET(BUILD_TENGINE ON)
+ELSE()
+	SET(OCV_TENGINE_FILENAME "${TENGINE_COMMIT_VERSION}.zip")#name2
+	SET(OCV_TENGINE_URL "https://github.com/OAID/Tengine/archive/") #url2
+	SET(tengine_md5sum f51ca8f3963faeeff3f019a6f6edc206) #md5sum2
+
+	#MESSAGE(STATUS "**** TENGINE DOWNLOAD BEGIN ****")
+	ocv_download(FILENAME ${OCV_TENGINE_FILENAME}
+						HASH ${tengine_md5sum}
+						URL
+						"${OPENCV_TENGINE_URL}"
+						"$ENV{OPENCV_TENGINE_URL}"
+						"${OCV_TENGINE_URL}"
+						DESTINATION_DIR "${OCV_TENGINE_DIR}"
+						ID TENGINE
+						STATUS res
+						UNPACK RELATIVE_URL)
+
+	if (NOT res)
+		MESSAGE(STATUS "TENGINE DOWNLOAD FAILED. Turning Tengine_FOUND off.")
+		SET(Tengine_FOUND OFF)
+	else ()
+		MESSAGE(STATUS "TENGINE DOWNLOAD success . ")
+
+		SET(Tengine_FOUND ON)
+		SET(BUILD_TENGINE ON)
+	endif()
+ENDIF()
+
+if(BUILD_TENGINE)
+	SET(HAVE_TENGINE 1)
+
+	# android system
+	if(ANDROID)
+	   if(${ANDROID_ABI} STREQUAL "armeabi-v7a")
+			   SET(CONFIG_ARCH_ARM32 ON)
+	   elseif(${ANDROID_ABI} STREQUAL "arm64-v8a")
+			   SET(CONFIG_ARCH_ARM64 ON)
+	   endif()
+	else()
+		# linux system
+		if(CMAKE_SYSTEM_PROCESSOR STREQUAL arm)
+			   SET(CONFIG_ARCH_ARM32 ON)
+		elseif(CMAKE_SYSTEM_PROCESSOR STREQUAL aarch64) ## AARCH64
+			   SET(CONFIG_ARCH_ARM64 ON)
+		endif()
+	endif()
+
+	SET(BUILT_IN_OPENCV ON) ## set for tengine compile discern .
+	SET(Tengine_INCLUDE_DIR  "${OCV_TENGINE_SOURCE_PATH}/core/include" CACHE INTERNAL "")
+	if(EXISTS "${OCV_TENGINE_SOURCE_PATH}/CMakeLists.txt")
+		add_subdirectory("${OCV_TENGINE_SOURCE_PATH}" "${OCV_TENGINE_DIR}/build")
+	else()
+		message(WARNING "TENGINE: Missing 'CMakeLists.txt' in source code package: ${OCV_TENGINE_SOURCE_PATH}")
+	endif()
+	SET(Tengine_LIB "tengine" CACHE INTERNAL "")
+endif()
--- a/Show More
+++ b/Show More