release: OpenCV 4.5.4

Merge pull request #20848 from alalek:reverse_plugins_candidates_order
Merge pull request #20721 from YanivHollander:DocFixes
2026-07-26 13:53:03 +04:00 · 2021-10-09 15:48:26 +00:00 · 2021-10-09 15:45:17 +00:00 · 2021-10-09 15:43:50 +00:00 · 2021-10-09 06:22:02 +00:00 · 2021-10-09 05:59:24 +00:00
771 changed files with 58070 additions and 16919 deletions
@@ -1778,30 +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 : \
@@ -1809,29 +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
@@ -1847,6 +1850,8 @@ TegraCvtColor_Invoker(bgrx2hsvf, bgrx2hsv, src_data + static_cast<size_t>(range.
 #define cv_hal_cvtBGRtoHSV TEGRA_CVTBGRTOHSV
 #undef cv_hal_cvtTwoPlaneYUVtoBGR
 #define cv_hal_cvtTwoPlaneYUVtoBGR TEGRA_CVT2PYUVTOBGR
+#undef cv_hal_cvtTwoPlaneYUVtoBGREx
+#define cv_hal_cvtTwoPlaneYUVtoBGREx TEGRA_CVT2PYUVTOBGR_EX

 #endif // OPENCV_IMGPROC_HAL_INTERFACE_H

@@ -1,9 +1,9 @@
-# Binaries branch name: ffmpeg/master_20210303
-# Binaries were created for OpenCV: 7ac6abe02a33bef445a5b77214ad31964e2c5cc1
-ocv_update(FFMPEG_BINARIES_COMMIT "629590c3ba09fb0c8eaa9ab858ff13d3a84ca1aa")
-ocv_update(FFMPEG_FILE_HASH_BIN32 "638065d5a0dab8a828879942375dcac4")
-ocv_update(FFMPEG_FILE_HASH_BIN64 "7f10ae2e6a080ba3714f7a38ee03ae15")
-ocv_update(FFMPEG_FILE_HASH_CMAKE "f8e65dbe4a3b4eedc0d2997e07c3f3fd")
+# Binaries branch name: ffmpeg/master_20211005
+# Binaries were created for OpenCV: 672399c751c431bbe52818b33fd3ca17b51e0e16
+ocv_update(FFMPEG_BINARIES_COMMIT "40b4666d1aa374205fd61373496e15d92ecd5313")
+ocv_update(FFMPEG_FILE_HASH_BIN32 "c2f9a897d464a2dce2286f8067ad9d90")
+ocv_update(FFMPEG_FILE_HASH_BIN64 "878a4e8fe5a4d68f18c9cdde543b9ead")
+ocv_update(FFMPEG_FILE_HASH_CMAKE "8862c87496e2e8c375965e1277dee1c7")

 function(download_win_ffmpeg script_var)
  set(${script_var} "" PARENT_SCOPE)
@@ -3,10 +3,10 @@ project(${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_REVISION 6)
+set(VERSION_MINOR 1)
+set(VERSION_REVISION 0)
 set(VERSION ${VERSION_MAJOR}.${VERSION_MINOR}.${VERSION_REVISION})
-set(LIBJPEG_TURBO_VERSION_NUMBER 2000006)
+set(LIBJPEG_TURBO_VERSION_NUMBER 2001000)

 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()


@@ -91,7 +91,7 @@ best of our understanding.
 The Modified (3-clause) BSD License
 ===================================

-Copyright (C)2009-2020 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
 ==========

@@ -3,7 +3,7 @@ Background

 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 and x86-64
+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
@@ -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 */
@@ -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, 2018-2019, 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.
 *
@@ -42,15 +44,19 @@
 * flags (this defines __thumb__).
 */

-/* NOTE: Both GCC and Clang define __GNUC__ */
-#if defined(__GNUC__) && (defined(__arm__) || defined(__aarch64__))
+#if 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 +71,43 @@
 * 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 +140,7 @@ 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 +219,17 @@ 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 +314,7 @@ 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
@@ -334,94 +362,94 @@ 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

-#if SIZEOF_SIZE_T != 8 && !defined(_WIN64)
-
-#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; \
  } \
 }

-#endif
-
-#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
@@ -444,6 +472,7 @@ dump_buffer(working_state *state)

 #define STORE_BUFFER() { \
  if (localbuf) { \
+    size_t bytes, bytestocopy; \
    bytes = buffer - _buffer; \
    buffer = _buffer; \
    while (bytes > 0) { \
@@ -466,20 +495,46 @@ 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;
@@ -493,7 +548,7 @@ 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()

@@ -509,53 +564,41 @@ 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
@@ -563,51 +606,46 @@ 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;
@@ -654,8 +692,9 @@ 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) {
@@ -694,7 +733,7 @@ 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) {
@@ -723,8 +762,9 @@ 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))
@@ -733,7 +773,7 @@ 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;
 }


@@ -4,8 +4,9 @@
 * 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.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -51,15 +52,19 @@
 * flags (this defines __thumb__).
 */

-/* NOTE: Both GCC and Clang define __GNUC__ */
-#if defined(__GNUC__) && (defined(__arm__) || defined(__aarch64__))
+#if 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 +174,26 @@ 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;
 }


@@ -860,7 +867,7 @@ 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 +924,7 @@ 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;
@@ -968,7 +975,7 @@ 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);
@@ -4,7 +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, 2020, 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.
@@ -319,6 +319,8 @@ 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,
@@ -329,6 +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) {
@@ -532,6 +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);
      }
    }
@@ -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++)
@@ -5,7 +5,7 @@
 * 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) 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];
@@ -487,27 +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] +
-                         cinfo->master->first_MCU_col[ci];
-      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] +
-                         cinfo->master->first_MCU_col[ci];
+        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];
@@ -515,18 +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))
@@ -541,7 +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))
@@ -556,7 +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))
@@ -571,7 +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))
@@ -586,7 +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))
@@ -599,14 +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;
@@ -655,7 +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;
    }
  }
@@ -5,6 +5,7 @@
 * Copyright (C) 1991-1997, Thomas G. Lane.
 * libjpeg-turbo Modifications:
 * 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;
@@ -589,11 +578,15 @@ 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];
-      /* This is really just
-       *   s += state.last_dc_val[ci];
-       * It is written this way in order to shut up UBSan.
+      /* 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.
       */
-      s = (int)((unsigned int)s + (unsigned int)state.last_dc_val[ci]);
+      s += state.last_dc_val[ci];
      state.last_dc_val[ci] = s;
      if (block) {
        /* Output the DC coefficient (assumes jpeg_natural_order[0] = 0) */
@@ -653,7 +646,7 @@ 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;
 }

@@ -671,7 +664,7 @@ 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;
@@ -688,7 +681,7 @@ decode_mcu_fast(j_decompress_ptr cinfo, JBLOCKROW *MCU_data)

    if (entropy->dc_needed[blkn]) {
      int ci = cinfo->MCU_membership[blkn];
-      s = (int)((unsigned int)s + (unsigned int)state.last_dc_val[ci]);
+      s += state.last_dc_val[ci];
      state.last_dc_val[ci] = s;
      if (block)
        (*block)[0] = (JCOEF)s;
@@ -740,7 +733,7 @@ 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;
 }

@@ -795,7 +788,8 @@ 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];
@@ -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
@@ -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];
@@ -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)];
@@ -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];
@@ -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)];
@@ -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];
@@ -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++)
@@ -8,7 +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, Arm Limited.
+ * Copyright (C) 2019-2020, Arm Limited.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -177,7 +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;
      }
@@ -213,7 +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;
    }
@@ -242,7 +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;
    }
@@ -283,20 +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;
  }
 }
@@ -338,7 +338,7 @@ h1v2_fancy_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr,
      outptr = output_data[outrow++];

      for (colctr = 0; colctr < compptr->downsampled_width; colctr++) {
-        thiscolsum = GETJSAMPLE(*inptr0++) * 3 + GETJSAMPLE(*inptr1++);
+        thiscolsum = (*inptr0++) * 3 + (*inptr1++);
        *outptr++ = (JSAMPLE)((thiscolsum + bias) >> 2);
      }
    }
@@ -381,8 +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;
@@ -390,7 +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;
@@ -477,7 +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) {
@@ -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), \
@@ -3,7 +3,7 @@
 *
 * 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, 2020, D. R. Commander.
 * For conditions of distribution and use, see the accompanying README.ijg
@@ -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] */
@@ -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. */

@@ -5,7 +5,7 @@
 * 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, D. R. Commander.
 * Copyright (C) 2015, Google, Inc.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
@@ -158,6 +158,9 @@ 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 */
@@ -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.
@@ -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
@@ -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, 2012-2020, D. R. Commander.
+ * Copyright (C) 2010, 2012-2021, D. R. Commander.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -37,9 +37,9 @@
 */

 #define JCOPYRIGHT \
-  "Copyright (C) 2009-2020 D. R. Commander\n" \
+  "Copyright (C) 2009-2021 D. R. Commander\n" \
  "Copyright (C) 2015, 2020 Google, Inc.\n" \
-  "Copyright (C) 2019 Arm Limited\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" \
@@ -48,7 +48,7 @@
  "Copyright (C) 2009, 2012 Pierre Ossman for Cendio AB\n" \
  "Copyright (C) 2009-2011 Nokia Corporation and/or its subsidiary(-ies)\n" \
  "Copyright (C) 1999-2006 MIYASAKA Masaru\n" \
-  "Copyright (C) 1991-2017 Thomas G. Lane, Guido Vollbeding"
+  "Copyright (C) 1991-2020 Thomas G. Lane, Guido Vollbeding"

 #define JCOPYRIGHT_SHORT \
-  "Copyright (C) 1991-2020 The libjpeg-turbo Project and many others"
+  "Copyright (C) 1991-2021 The libjpeg-turbo Project and many others"
@@ -32,7 +32,9 @@ endif()
 #         Define the library target:
 # ----------------------------------------------------------------------------------

-add_definitions(-DWEBP_USE_THREAD)
+if(NOT OPENCV_DISABLE_THREAD_SUPPORT)
+  add_definitions(-DWEBP_USE_THREAD)
+endif()

 add_library(${WEBP_LIBRARY} STATIC ${OPENCV_3RDPARTY_EXCLUDE_FROM_ALL} ${lib_srcs} ${lib_hdrs})
 if(ANDROID)
@@ -923,6 +923,11 @@ int ovx_hal_cvtGraytoBGR(const uchar * a, size_t astep, uchar * b, size_t bstep,
 }

 int ovx_hal_cvtTwoPlaneYUVtoBGR(const uchar * a, size_t astep, uchar * b, size_t bstep, int w, int h, int bcn, bool swapBlue, int uIdx)
+{
+    return ovx_hal_cvtTwoPlaneYUVtoBGREx(a, astep, a + h * astep, astep, b, bstep, w, h, bcn, swapBlue, uIdx);
+}
+
+int ovx_hal_cvtTwoPlaneYUVtoBGREx(const uchar * a, size_t astep, const uchar * b, size_t bstep, uchar * c, size_t cstep, int w, int h, int bcn, bool swapBlue, int uIdx)
 {
    if (skipSmallImages<VX_KERNEL_COLOR_CONVERT>(w, h))
        return CV_HAL_ERROR_NOT_IMPLEMENTED;
@@ -933,8 +938,7 @@ int ovx_hal_cvtTwoPlaneYUVtoBGR(const uchar * a, size_t astep, uchar * b, size_t

    if (w & 1 || h & 1) // It's not described in spec but sample implementation unable to convert odd sized images
        return CV_HAL_ERROR_NOT_IMPLEMENTED;
-    refineStep(w, h, uIdx ? VX_DF_IMAGE_NV21 : VX_DF_IMAGE_NV12, astep);
-    refineStep(w, h, bcn == 3 ? VX_DF_IMAGE_RGB : VX_DF_IMAGE_RGBX, bstep);
+
    try
    {
        ivx::Context ctx = getOpenVXHALContext();
@@ -943,8 +947,8 @@ int ovx_hal_cvtTwoPlaneYUVtoBGR(const uchar * a, size_t astep, uchar * b, size_t
        std::vector<void *> ptrs;
            addr.push_back(ivx::Image::createAddressing(w, h, 1, (vx_int32)astep));
            ptrs.push_back((void*)a);
-            addr.push_back(ivx::Image::createAddressing(w / 2, h / 2, 2, (vx_int32)astep));
-            ptrs.push_back((void*)(a + h * astep));
+            addr.push_back(ivx::Image::createAddressing(w / 2, h / 2, 2, (vx_int32)bstep));
+            ptrs.push_back((void*)b);

        vxImage
            ia = ivx::Image::createFromHandle(ctx, uIdx ? VX_DF_IMAGE_NV21 : VX_DF_IMAGE_NV12, addr, ptrs);
@@ -952,7 +956,7 @@ int ovx_hal_cvtTwoPlaneYUVtoBGR(const uchar * a, size_t astep, uchar * b, size_t
            return CV_HAL_ERROR_NOT_IMPLEMENTED; // OpenCV store NV12/NV21 as RANGE_RESTRICTED while OpenVX expect RANGE_FULL
        vxImage
            ib = ivx::Image::createFromHandle(ctx, bcn == 3 ? VX_DF_IMAGE_RGB : VX_DF_IMAGE_RGBX,
-                ivx::Image::createAddressing(w, h, bcn, (vx_int32)bstep), b);
+                ivx::Image::createAddressing(w, h, bcn, (vx_int32)cstep), c);
        ivx::IVX_CHECK_STATUS(vxuColorConvert(ctx, ia, ib));
    }
    catch (ivx::RuntimeError & e)
@@ -49,6 +49,7 @@ int ovx_hal_morph(cvhalFilter2D *filter_context, uchar *a, size_t astep, uchar *
 int ovx_hal_cvtBGRtoBGR(const uchar * a, size_t astep, uchar * b, size_t bstep, int w, int h, int depth, int acn, int bcn, bool swapBlue);
 int ovx_hal_cvtGraytoBGR(const uchar * a, size_t astep, uchar * b, size_t bstep, int w, int h, int depth, int bcn);
 int ovx_hal_cvtTwoPlaneYUVtoBGR(const uchar * a, size_t astep, uchar * b, size_t bstep, int w, int h, int bcn, bool swapBlue, int uIdx);
+int ovx_hal_cvtTwoPlaneYUVtoBGREx(const uchar * a, size_t astep, const uchar * b, size_t bstep, uchar * c, size_t cstep, int w, int h, int bcn, bool swapBlue, int uIdx);
 int ovx_hal_cvtThreePlaneYUVtoBGR(const uchar * a, size_t astep, uchar * b, size_t bstep, int w, int h, int bcn, bool swapBlue, int uIdx);
 int ovx_hal_cvtBGRtoThreePlaneYUV(const uchar * a, size_t astep, uchar * b, size_t bstep, int w, int h, int acn, bool swapBlue, int uIdx);
 int ovx_hal_cvtOnePlaneYUVtoBGR(const uchar * a, size_t astep, uchar * b, size_t bstep, int w, int h, int bcn, bool swapBlue, int uIdx, int ycn);
@@ -130,6 +131,8 @@ int ovx_hal_integral(int depth, int sdepth, int, const uchar * a, size_t astep,
 #define cv_hal_cvtGraytoBGR ovx_hal_cvtGraytoBGR
 #undef cv_hal_cvtTwoPlaneYUVtoBGR
 #define cv_hal_cvtTwoPlaneYUVtoBGR ovx_hal_cvtTwoPlaneYUVtoBGR
+#undef cv_hal_cvtTwoPlaneYUVtoBGREx
+#define cv_hal_cvtTwoPlaneYUVtoBGREx ovx_hal_cvtTwoPlaneYUVtoBGREx
 #undef cv_hal_cvtThreePlaneYUVtoBGR
 #define cv_hal_cvtThreePlaneYUVtoBGR ovx_hal_cvtThreePlaneYUVtoBGR
 #undef cv_hal_cvtBGRtoThreePlaneYUV
@@ -31,7 +31,7 @@ libpng                Portable Network Graphics library.
 libtiff               Tag Image File Format (TIFF) Software
                      Copyright (c) 1988-1997 Sam Leffler
                      Copyright (c) 1991-1997 Silicon Graphics, Inc.
-                      See libtiff home page http://www.remotesensing.org/libtiff/
+                      See libtiff home page http://www.libtiff.org/
                      for details and links to the source code

                      WITH_TIFF CMake option must be ON to add libtiff & zlib support to imgcodecs.
@@ -51,7 +51,9 @@ jasper                JasPer is a collection of software
                      Copyright (c) 1999-2000 The University of British Columbia
                      Copyright (c) 2001-2003 Michael David Adams

-                      The JasPer license can be found in libjasper.
+                      See JasPer official GitHub repository
+                      https://github.com/jasper-software/jasper.git
+                      for details and links to source code
 ------------------------------------------------------------------------------------
 openexr               OpenEXR is a high dynamic-range (HDR) image file format developed
                      by Industrial Light & Magic for use in computer imaging applications.
@@ -238,7 +238,7 @@ OCV_OPTION(WITH_CAP_IOS "Enable iOS video capture" ON
  VISIBLE_IF IOS
  VERIFY HAVE_CAP_IOS)
 OCV_OPTION(WITH_CAROTENE "Use NVidia carotene acceleration library for ARM platform" ON
-  VISIBLE_IF (ARM OR AARCH64) AND NOT IOS AND NOT (CMAKE_VERSION VERSION_LESS "2.8.11"))
+  VISIBLE_IF (ARM OR AARCH64) AND NOT IOS)
 OCV_OPTION(WITH_CPUFEATURES "Use cpufeatures Android library" ON
  VISIBLE_IF ANDROID
  VERIFY HAVE_CPUFEATURES)
@@ -467,6 +467,7 @@ OCV_OPTION(BUILD_ANDROID_SERVICE    "Build OpenCV Manager for Google Play" OFF I
 OCV_OPTION(BUILD_CUDA_STUBS         "Build CUDA modules stubs when no CUDA SDK" OFF  IF (NOT APPLE_FRAMEWORK) )
 OCV_OPTION(BUILD_JAVA               "Enable Java support"                         (ANDROID OR NOT CMAKE_CROSSCOMPILING)  IF (ANDROID OR (NOT APPLE_FRAMEWORK AND NOT WINRT)) )
 OCV_OPTION(BUILD_OBJC               "Enable Objective-C support"                  ON  IF APPLE_FRAMEWORK )
+OCV_OPTION(BUILD_KOTLIN_EXTENSIONS  "Build Kotlin extensions (Android)"           ON  IF ANDROID )

 # OpenCV installation options
 # ===================================================
@@ -498,7 +499,7 @@ OCV_OPTION(OPENCV_WARNINGS_ARE_ERRORS "Treat warnings as errors"
 OCV_OPTION(ANDROID_EXAMPLES_WITH_LIBS "Build binaries of Android examples with native libraries" OFF  IF ANDROID )
 OCV_OPTION(ENABLE_IMPL_COLLECTION     "Collect implementation data on function call"             OFF )
 OCV_OPTION(ENABLE_INSTRUMENTATION     "Instrument functions to collect calls trace and performance" OFF )
-OCV_OPTION(ENABLE_GNU_STL_DEBUG       "Enable GNU STL Debug mode (defines _GLIBCXX_DEBUG)"       OFF IF ((NOT CMAKE_VERSION VERSION_LESS "2.8.11") AND CV_GCC) )
+OCV_OPTION(ENABLE_GNU_STL_DEBUG       "Enable GNU STL Debug mode (defines _GLIBCXX_DEBUG)"       OFF IF CV_GCC )
 OCV_OPTION(ENABLE_BUILD_HARDENING     "Enable hardening of the resulting binaries (against security attacks, detects memory corruption, etc)" OFF)
 OCV_OPTION(ENABLE_LTO                 "Enable Link Time Optimization" OFF IF CV_GCC OR MSVC)
 OCV_OPTION(ENABLE_THIN_LTO            "Enable Thin LTO" OFF IF CV_CLANG)
@@ -510,6 +511,9 @@ OCV_OPTION(CV_TRACE                   "Enable OpenCV code trace" ON)
 OCV_OPTION(OPENCV_GENERATE_SETUPVARS  "Generate setup_vars* scripts" ON IF (NOT ANDROID AND NOT APPLE_FRAMEWORK) )
 OCV_OPTION(ENABLE_CONFIG_VERIFICATION "Fail build if actual configuration doesn't match requested (WITH_XXX != HAVE_XXX)" OFF)
 OCV_OPTION(OPENCV_ENABLE_MEMALIGN     "Enable posix_memalign or memalign usage" ON)
+OCV_OPTION(OPENCV_DISABLE_FILESYSTEM_SUPPORT "Disable filesystem support" OFF)
+OCV_OPTION(OPENCV_DISABLE_THREAD_SUPPORT "Build the library without multi-threaded code." OFF)
+OCV_OPTION(OPENCV_SEMIHOSTING         "Build the library for semihosting target (Arm). See https://developer.arm.com/documentation/100863/latest." OFF)

 OCV_OPTION(ENABLE_PYLINT              "Add target with Pylint checks"                            (BUILD_DOCS OR BUILD_EXAMPLES) IF (NOT CMAKE_CROSSCOMPILING AND NOT APPLE_FRAMEWORK) )
 OCV_OPTION(ENABLE_FLAKE8              "Add target with Python flake8 checker"                    (BUILD_DOCS OR BUILD_EXAMPLES) IF (NOT CMAKE_CROSSCOMPILING AND NOT APPLE_FRAMEWORK) )
@@ -518,6 +522,10 @@ if(ENABLE_IMPL_COLLECTION)
  add_definitions(-DCV_COLLECT_IMPL_DATA)
 endif()

+if(OPENCV_DISABLE_FILESYSTEM_SUPPORT)
+  add_definitions(-DOPENCV_HAVE_FILESYSTEM_SUPPORT=0)
+endif()
+
 set(OPENCV_MATHJAX_RELPATH "https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.0" CACHE STRING "URI to a MathJax installation")

 # ----------------------------------------------------------------------------
@@ -651,6 +659,8 @@ if(UNIX)
      set(OPENCV_LINKER_LIBS ${OPENCV_LINKER_LIBS} m pthread)
    elseif(EMSCRIPTEN)
      # no need to link to system libs with emscripten
+    elseif(QNXNTO)
+      set(OPENCV_LINKER_LIBS ${OPENCV_LINKER_LIBS} m)
    else()
      set(OPENCV_LINKER_LIBS ${OPENCV_LINKER_LIBS} dl m pthread rt)
    endif()
@@ -658,6 +668,11 @@ if(UNIX)
    set(HAVE_PTHREAD 1)
  endif()

+  # Ensure that libpthread is not listed as one of the libraries to pass to the linker.
+  if (OPENCV_DISABLE_THREAD_SUPPORT)
+    list(REMOVE_ITEM OPENCV_LINKER_LIBS pthread)
+  endif()
+
  if(OPENCV_ENABLE_MEMALIGN)
    CHECK_SYMBOL_EXISTS(posix_memalign stdlib.h HAVE_POSIX_MEMALIGN)
    CHECK_INCLUDE_FILE(malloc.h HAVE_MALLOC_H)
@@ -899,7 +914,7 @@ add_subdirectory(include)
 ocv_add_modules_compiler_options()

 # OpenCV modules
-add_subdirectory(modules)
+ocv_register_modules()

 # Generate targets for documentation
 add_subdirectory(doc)
@@ -1041,7 +1056,6 @@ endif()
 status("")
 status("  Platform:")
 if(NOT DEFINED OPENCV_TIMESTAMP
-    AND NOT CMAKE_VERSION VERSION_LESS 2.8.11
    AND NOT BUILD_INFO_SKIP_TIMESTAMP
 )
  string(TIMESTAMP OPENCV_TIMESTAMP "" UTC)
@@ -1126,6 +1140,10 @@ endif()
 status("    ccache:"                  OPENCV_COMPILER_IS_CCACHE THEN YES ELSE NO)
 status("    Precompiled headers:"     PCHSupport_FOUND AND ENABLE_PRECOMPILED_HEADERS THEN YES ELSE NO)

+if(OPENCV_DISABLE_FILESYSTEM_SUPPORT)
+  status("    Filesystem support is disabled")
+endif()
+
 # ========================== Dependencies ============================
 ocv_get_all_libs(deps_modules deps_extra deps_3rdparty)
 status("    Extra dependencies:" ${deps_extra})
@@ -1226,15 +1244,20 @@ endif(WIN32)

 # ========================== GUI ==========================
 status("")
-status("  GUI: ")
+status("  GUI: " "${OPENCV_HIGHGUI_BUILTIN_BACKEND}")

 if(WITH_QT OR HAVE_QT)
-  if(HAVE_QT5)
-    status("    QT:" "YES (ver ${Qt5Core_VERSION_STRING})")
-    status("      QT OpenGL support:" HAVE_QT_OPENGL THEN "YES (${Qt5OpenGL_LIBRARIES} ${Qt5OpenGL_VERSION_STRING})" ELSE NO)
-  elseif(HAVE_QT)
+  if(HAVE_QT)
    status("    QT:" "YES (ver ${QT_VERSION_MAJOR}.${QT_VERSION_MINOR}.${QT_VERSION_PATCH} ${QT_EDITION})")
-    status("      QT OpenGL support:" HAVE_QT_OPENGL THEN "YES (${QT_QTOPENGL_LIBRARY})" ELSE NO)
+    if(HAVE_QT_OPENGL)
+      if(Qt${QT_VERSION_MAJOR}OpenGL_LIBRARIES)
+        status("      QT OpenGL support:" HAVE_QT_OPENGL THEN "YES (${Qt${QT_VERSION_MAJOR}OpenGL_LIBRARIES} ${Qt${QT_VERSION_MAJOR}OpenGL_VERSION_STRING})" ELSE NO)
+      else()
+        status("      QT OpenGL support:" HAVE_QT_OPENGL THEN "YES (${QT_QTOPENGL_LIBRARY})" ELSE NO)
+      endif()
+    else()
+      status("      QT OpenGL support:" "NO")
+    endif()
  else()
    status("    QT:" "NO")
  endif()
@@ -1421,7 +1444,16 @@ if(WITH_LIBREALSENSE OR HAVE_LIBREALSENSE)
 endif()

 if(WITH_MFX OR HAVE_MFX)
-  status("    Intel Media SDK:" HAVE_MFX      THEN "YES (${MFX_LIBRARY})" ELSE NO)
+  if(HAVE_MFX)
+    if(MFX_LIBRARY)
+      set(__details " (${MFX_LIBRARY})")
+    elseif(MFX_LIBRARIES)
+      set(__details " (${MFX_LIBRARIES})")
+    else()
+      set(__details " (unknown)")
+    endif()
+  endif()
+  status("    Intel Media SDK:" HAVE_MFX      THEN "YES${__details}" ELSE NO)
 endif()

 if(WITH_GPHOTO2 OR HAVE_GPHOTO2)
@@ -1439,6 +1471,15 @@ ocv_build_features_string(parallel_status EXCLUSIVE
  ELSE "none")
 status("")
 status("  Parallel framework:" "${parallel_status}")
+if (OPENCV_DISABLE_THREAD_SUPPORT)
+  status("" "Multi thread code explicitly disabled with OPENCV_DISABLE_THREAD_SUPPORT.")
+  if(HAVE_PTHREADS_PF OR HAVE_HPX OR HAVE_OPENMP OR HAVE_GCD OR HAVE_CONCURRENCY)
+    message(FATAL_ERROR "Not all parallel frameworks have been disabled (using ${parallel_status}).")
+  endif()
+  if(HAVE_PTHREAD)
+    message(FATAL_ERROR "Thread execution might be in use in some component.")
+  endif()
+endif()

 if(CV_TRACE OR OPENCV_TRACE)
  ocv_build_features_string(trace_status EXCLUSIVE
@@ -1490,6 +1531,7 @@ if(WITH_INF_ENGINE OR INF_ENGINE_TARGET)
  if(INF_ENGINE_TARGET)
    list(GET INF_ENGINE_TARGET 0 ie_target)
    set(__msg "YES (${INF_ENGINE_RELEASE} / ${INF_ENGINE_VERSION})")
+    ocv_get_imported_target(ie_target "${ie_target}")
    get_target_property(_lib ${ie_target} IMPORTED_LOCATION)
    get_target_property(_lib_imp_rel ${ie_target} IMPORTED_IMPLIB_RELEASE)
    get_target_property(_lib_imp_dbg ${ie_target} IMPORTED_IMPLIB_DEBUG)
@@ -1513,7 +1555,7 @@ if(WITH_INF_ENGINE OR INF_ENGINE_TARGET)
 endif()
 if(WITH_NGRAPH OR HAVE_NGRAPH)
  if(HAVE_NGRAPH)
-    set(__target ngraph::ngraph)
+    ocv_get_imported_target(__target ngraph::ngraph)
    set(__msg "YES (${ngraph_VERSION})")
    get_target_property(_lib ${__target} IMPORTED_LOCATION)
    get_target_property(_lib_imp_rel ${__target} IMPORTED_IMPLIB_RELEASE)
@@ -1,9 +1,10 @@
 /*************************************************
 USAGE:
-./model_diagnostics -m <onnx file location>
+./model_diagnostics -m <model file location>
 **************************************************/
 #include <opencv2/dnn.hpp>
 #include <opencv2/core/utils/filesystem.hpp>
+#include <opencv2/dnn/utils/debug_utils.hpp>

 #include <iostream>

@@ -32,7 +33,7 @@ static std::string checkFileExists(const std::string& fileName)
 }

 std::string diagnosticKeys =
-        "{ model m     | | Path to the model .onnx file. }"
+        "{ model m     | | Path to the model file. }"
        "{ config c    | | Path to the model configuration file. }"
        "{ framework f | | [Optional] Name of the model framework. }";

@@ -41,7 +42,7 @@ std::string diagnosticKeys =
 int main( int argc, const char** argv )
 {
    CommandLineParser argParser(argc, argv, diagnosticKeys);
-    argParser.about("Use this tool to run the diagnostics of provided ONNX model"
+    argParser.about("Use this tool to run the diagnostics of provided ONNX/TF model"
                    "to obtain the information about its support (supported layers).");

    if (argc == 1)
@@ -57,6 +58,7 @@ int main( int argc, const char** argv )
    CV_Assert(!model.empty());

    enableModelDiagnostics(true);
+    skipModelImport(true);
    redirectError(diagnosticsErrorCallback, NULL);

    Net ocvNet = readNet(model, config, frameworkId);
@@ -178,8 +178,17 @@ if(CV_GCC OR CV_CLANG)
    add_extra_compiler_option(-Wno-long-long)
  endif()

-  # We need pthread's
-  if(UNIX AND NOT ANDROID AND NOT (APPLE AND CV_CLANG)) # TODO
+  # We need pthread's, unless we have explicitly disabled multi-thread execution.
+  if(NOT OPENCV_DISABLE_THREAD_SUPPORT
+      AND (
+        (UNIX
+          AND NOT ANDROID
+          AND NOT (APPLE AND CV_CLANG)
+          AND NOT EMSCRIPTEN
+        )
+        OR (EMSCRIPTEN AND WITH_PTHREADS_PF)  # https://github.com/opencv/opencv/issues/20285
+      )
+  ) # TODO
    add_extra_compiler_option(-pthread)
  endif()

@@ -227,9 +236,11 @@ if(CV_GCC OR CV_CLANG)
        if(APPLE)
          set(OPENCV_EXTRA_EXE_LINKER_FLAGS "${OPENCV_EXTRA_EXE_LINKER_FLAGS} -Wl,-dead_strip")
          set(OPENCV_EXTRA_SHARED_LINKER_FLAGS "${OPENCV_EXTRA_SHARED_LINKER_FLAGS} -Wl,-dead_strip")
+          set(OPENCV_EXTRA_MODULE_LINKER_FLAGS "${OPENCV_EXTRA_MODULE_LINKER_FLAGS} -Wl,-dead_strip")
        else()
          set(OPENCV_EXTRA_EXE_LINKER_FLAGS "${OPENCV_EXTRA_EXE_LINKER_FLAGS} -Wl,--gc-sections")
          set(OPENCV_EXTRA_SHARED_LINKER_FLAGS "${OPENCV_EXTRA_SHARED_LINKER_FLAGS} -Wl,--gc-sections")
+          set(OPENCV_EXTRA_MODULE_LINKER_FLAGS "${OPENCV_EXTRA_MODULE_LINKER_FLAGS} -Wl,--gc-sections")
        endif()
      endif()
    endif()
@@ -281,6 +292,7 @@ if(MSVC)
    set(OPENCV_EXTRA_FLAGS_RELEASE "${OPENCV_EXTRA_FLAGS_RELEASE} /Zi")
    set(OPENCV_EXTRA_EXE_LINKER_FLAGS_RELEASE "${OPENCV_EXTRA_EXE_LINKER_FLAGS_RELEASE} /debug")
    set(OPENCV_EXTRA_SHARED_LINKER_FLAGS_RELEASE "${OPENCV_EXTRA_SHARED_LINKER_FLAGS_RELEASE} /debug")
+    set(OPENCV_EXTRA_MODULE_LINKER_FLAGS_RELEASE "${OPENCV_EXTRA_MODULE_LINKER_FLAGS_RELEASE} /debug")
  endif()

  # Remove unreferenced functions: function level linking
@@ -350,6 +362,7 @@ if(NOT OPENCV_SKIP_LINK_AS_NEEDED)
    if(HAVE_LINK_AS_NEEDED)
      set(OPENCV_EXTRA_EXE_LINKER_FLAGS "${OPENCV_EXTRA_EXE_LINKER_FLAGS} ${_option}")
      set(OPENCV_EXTRA_SHARED_LINKER_FLAGS "${OPENCV_EXTRA_SHARED_LINKER_FLAGS} ${_option}")
+      set(OPENCV_EXTRA_MODULE_LINKER_FLAGS "${OPENCV_EXTRA_MODULE_LINKER_FLAGS} ${_option}")
    endif()
  endif()
 endif()
@@ -368,6 +381,9 @@ if(NOT OPENCV_SKIP_EXTRA_COMPILER_FLAGS)
  set(CMAKE_SHARED_LINKER_FLAGS         "${CMAKE_SHARED_LINKER_FLAGS} ${OPENCV_EXTRA_SHARED_LINKER_FLAGS}")
  set(CMAKE_SHARED_LINKER_FLAGS_RELEASE "${CMAKE_SHARED_LINKER_FLAGS_RELEASE} ${OPENCV_EXTRA_SHARED_LINKER_FLAGS_RELEASE}")
  set(CMAKE_SHARED_LINKER_FLAGS_DEBUG   "${CMAKE_SHARED_LINKER_FLAGS_DEBUG} ${OPENCV_EXTRA_SHARED_LINKER_FLAGS_DEBUG}")
+  set(CMAKE_MODULE_LINKER_FLAGS         "${CMAKE_MODULE_LINKER_FLAGS} ${OPENCV_EXTRA_MODULE_LINKER_FLAGS}")
+  set(CMAKE_MODULE_LINKER_FLAGS_RELEASE "${CMAKE_MODULE_LINKER_FLAGS_RELEASE} ${OPENCV_EXTRA_MODULE_LINKER_FLAGS_RELEASE}")
+  set(CMAKE_MODULE_LINKER_FLAGS_DEBUG   "${CMAKE_MODULE_LINKER_FLAGS_DEBUG} ${OPENCV_EXTRA_MODULE_LINKER_FLAGS_DEBUG}")
 endif()

 if(MSVC)
@@ -391,6 +407,9 @@ if(MSVC)
    endif()
  endif()

+  # Enable [[attribute]] syntax checking to prevent silent failure: "attribute is ignored in this syntactic position"
+  add_extra_compiler_option("/w15240")
+
  if(NOT ENABLE_NOISY_WARNINGS)
    ocv_warnings_disable(CMAKE_CXX_FLAGS /wd4127) # conditional expression is constant
    ocv_warnings_disable(CMAKE_CXX_FLAGS /wd4251) # class 'std::XXX' needs to have dll-interface to be used by clients of YYY
@@ -112,7 +112,7 @@ if(CUDA_FOUND)
  if(CUDA_GENERATION)
    if(NOT ";${_generations};" MATCHES ";${CUDA_GENERATION};")
      string(REPLACE ";" ", " _generations "${_generations}")
-      message(FATAL_ERROR "ERROR: ${_generations} Generations are suppered.")
+      message(FATAL_ERROR "ERROR: ${_generations} Generations are supported.")
    endif()
    unset(CUDA_ARCH_BIN CACHE)
    unset(CUDA_ARCH_PTX CACHE)
@@ -171,6 +171,8 @@ elseif(MSVC)
    set(OpenCV_RUNTIME vc15)
  elseif(MSVC_VERSION MATCHES "^192[0-9]$")
    set(OpenCV_RUNTIME vc16)
+  elseif(MSVC_VERSION MATCHES "^193[0-9]$")
+    set(OpenCV_RUNTIME vc17)
  else()
    message(WARNING "OpenCV does not recognize MSVC_VERSION \"${MSVC_VERSION}\". Cannot set OpenCV_RUNTIME")
  endif()
@@ -9,9 +9,14 @@ set(HALIDE_ROOT_DIR "${HALIDE_ROOT_DIR}" CACHE PATH "Halide root directory")
 if(NOT HAVE_HALIDE)
  find_package(Halide QUIET) # Try CMake-based config files
  if(Halide_FOUND)
-    set(HALIDE_INCLUDE_DIRS "${Halide_INCLUDE_DIRS}" CACHE PATH "Halide include directories" FORCE)
-    set(HALIDE_LIBRARIES "${Halide_LIBRARIES}" CACHE PATH "Halide libraries" FORCE)
-    set(HAVE_HALIDE TRUE)
+    if(TARGET Halide::Halide)  # modern Halide scripts defines imported target
+      set(HALIDE_INCLUDE_DIRS "")
+      set(HALIDE_LIBRARIES "Halide::Halide")
+      set(HAVE_HALIDE TRUE)
+    else()
+      # using HALIDE_INCLUDE_DIRS / Halide_LIBRARIES
+      set(HAVE_HALIDE TRUE)
+    endif()
  endif()
 endif()

@@ -28,18 +33,15 @@ if(NOT HAVE_HALIDE AND HALIDE_ROOT_DIR)
  )
  if(HALIDE_LIBRARY AND HALIDE_INCLUDE_DIR)
    # TODO try_compile
-    set(HALIDE_INCLUDE_DIRS "${HALIDE_INCLUDE_DIR}" CACHE PATH "Halide include directories" FORCE)
-    set(HALIDE_LIBRARIES "${HALIDE_LIBRARY}" CACHE PATH "Halide libraries" FORCE)
+    set(HALIDE_INCLUDE_DIRS "${HALIDE_INCLUDE_DIR}")
+    set(HALIDE_LIBRARIES "${HALIDE_LIBRARY}")
    set(HAVE_HALIDE TRUE)
  endif()
-  if(NOT HAVE_HALIDE)
-    ocv_clear_vars(HALIDE_LIBRARIES HALIDE_INCLUDE_DIRS CACHE)
-  endif()
 endif()

 if(HAVE_HALIDE)
-  include_directories(${HALIDE_INCLUDE_DIRS})
+  if(HALIDE_INCLUDE_DIRS)
+    include_directories(${HALIDE_INCLUDE_DIRS})
+  endif()
  list(APPEND OPENCV_LINKER_LIBS ${HALIDE_LIBRARIES})
-else()
-  ocv_clear_vars(HALIDE_INCLUDE_DIRS HALIDE_LIBRARIES)
 endif()
@@ -99,6 +99,20 @@ if(InferenceEngine_FOUND)
  message(STATUS "Detected InferenceEngine: cmake package (${InferenceEngine_VERSION})")
 endif()

+if(DEFINED InferenceEngine_VERSION)
+  message(STATUS "InferenceEngine: ${InferenceEngine_VERSION}")
+  if(NOT INF_ENGINE_RELEASE AND NOT (InferenceEngine_VERSION VERSION_LESS "2021.4"))
+    math(EXPR INF_ENGINE_RELEASE_INIT "${InferenceEngine_VERSION_MAJOR} * 1000000 + ${InferenceEngine_VERSION_MINOR} * 10000 + ${InferenceEngine_VERSION_PATCH} * 100")
+  endif()
+endif()
+if(NOT INF_ENGINE_RELEASE AND NOT INF_ENGINE_RELEASE_INIT)
+  message(STATUS "WARNING: InferenceEngine version has not been set, 2021.4.1 will be used by default. Set INF_ENGINE_RELEASE variable if you experience build errors.")
+  set(INF_ENGINE_RELEASE_INIT "2021040100")
+elseif(DEFINED INF_ENGINE_RELEASE)
+  set(INF_ENGINE_RELEASE_INIT "${INF_ENGINE_RELEASE}")
+endif()
+set(INF_ENGINE_RELEASE "${INF_ENGINE_RELEASE_INIT}" CACHE STRING "Force IE version, should be in form YYYYAABBCC (e.g. 2020.1.0.2 -> 2020010002)")
+
 if(NOT INF_ENGINE_TARGET AND INF_ENGINE_LIB_DIRS AND INF_ENGINE_INCLUDE_DIRS)
  find_path(ie_custom_inc "inference_engine.hpp" PATHS "${INF_ENGINE_INCLUDE_DIRS}" NO_DEFAULT_PATH)
  if(CMAKE_BUILD_TYPE STREQUAL "Debug")
@@ -134,12 +148,8 @@ endif()
 # Add more features to the target

 if(INF_ENGINE_TARGET)
-  if(NOT INF_ENGINE_RELEASE)
-    message(WARNING "InferenceEngine version has not been set, 2021.3 will be used by default. Set INF_ENGINE_RELEASE variable if you experience build errors.")
-  endif()
-  set(INF_ENGINE_RELEASE "2021030000" CACHE STRING "Force IE version, should be in form YYYYAABBCC (e.g. 2020.1.0.2 -> 2020010002)")
  set_target_properties(${INF_ENGINE_TARGET} PROPERTIES
-    INTERFACE_COMPILE_DEFINITIONS "HAVE_INF_ENGINE=1;INF_ENGINE_RELEASE=${INF_ENGINE_RELEASE}"
+      INTERFACE_COMPILE_DEFINITIONS "HAVE_INF_ENGINE=1;INF_ENGINE_RELEASE=${INF_ENGINE_RELEASE}"
  )
 endif()

@@ -1,34 +1,34 @@
-# VTK 9.0
 if(NOT VTK_FOUND)
-  find_package(VTK 9 QUIET NAMES vtk COMPONENTS
-    FiltersExtraction
-    FiltersSources
-    FiltersTexture
-    IOExport
-    IOGeometry
-    IOPLY
-    InteractionStyle
-    RenderingCore
-    RenderingLOD
-    RenderingOpenGL2
-    NO_MODULE)
-endif()
-
-# VTK 6.x components
-if(NOT VTK_FOUND)
-  find_package(VTK QUIET COMPONENTS vtkInteractionStyle vtkRenderingLOD vtkIOPLY vtkFiltersTexture vtkRenderingFreeType vtkIOExport NO_MODULE)
-  IF(VTK_FOUND)
-    IF(VTK_RENDERING_BACKEND) #in vtk 7, the rendering backend is exported as a var.
-      find_package(VTK QUIET COMPONENTS vtkRendering${VTK_RENDERING_BACKEND} vtkInteractionStyle vtkRenderingLOD vtkIOPLY vtkFiltersTexture vtkRenderingFreeType vtkIOExport vtkIOGeometry NO_MODULE)
-    ELSE(VTK_RENDERING_BACKEND)
-      find_package(VTK QUIET COMPONENTS vtkRenderingOpenGL vtkInteractionStyle vtkRenderingLOD vtkIOPLY vtkFiltersTexture vtkRenderingFreeType vtkIOExport NO_MODULE)
-    ENDIF(VTK_RENDERING_BACKEND)
-  ENDIF(VTK_FOUND)
-endif()
-
-# VTK 5.x components
-if(NOT VTK_FOUND)
-  find_package(VTK QUIET COMPONENTS vtkCommon NO_MODULE)
+  find_package(VTK QUIET NAMES vtk VTK)
+  if(VTK_FOUND)
+    if(NOT (VTK_VERSION VERSION_LESS "9.0.0") AND (VTK_VERSION VERSION_LESS "10.0.0")) # VTK 9.x
+      find_package(VTK 9 QUIET NAMES vtk COMPONENTS
+              FiltersExtraction
+              FiltersSources
+              FiltersTexture
+              IOExport
+              IOGeometry
+              IOPLY
+              InteractionStyle
+              RenderingCore
+              RenderingLOD
+              RenderingOpenGL2
+              NO_MODULE)
+    elseif(VTK_VERSION VERSION_GREATER "5") # VTK 6.x components
+      find_package(VTK QUIET COMPONENTS vtkInteractionStyle vtkRenderingLOD vtkIOPLY vtkFiltersTexture vtkRenderingFreeType vtkIOExport NO_MODULE)
+      IF(VTK_FOUND)
+        IF(VTK_RENDERING_BACKEND) #in vtk 7, the rendering backend is exported as a var.
+          find_package(VTK QUIET COMPONENTS vtkRendering${VTK_RENDERING_BACKEND} vtkInteractionStyle vtkRenderingLOD vtkIOPLY vtkFiltersTexture vtkRenderingFreeType vtkIOExport vtkIOGeometry NO_MODULE)
+        ELSE(VTK_RENDERING_BACKEND)
+          find_package(VTK QUIET COMPONENTS vtkRenderingOpenGL vtkInteractionStyle vtkRenderingLOD vtkIOPLY vtkFiltersTexture vtkRenderingFreeType vtkIOExport NO_MODULE)
+        ENDIF(VTK_RENDERING_BACKEND)
+      ENDIF(VTK_FOUND)
+    elseif(VTK_VERSION VERSION_EQUAL "5") # VTK 5.x components
+      find_package(VTK QUIET COMPONENTS vtkCommon NO_MODULE)
+    else()
+      set(VTK_FOUND FALSE)
+    endif()
+  endif()
 endif()

 if(NOT VTK_FOUND)
@@ -44,32 +44,15 @@ if(VTK_VERSION VERSION_LESS "5.8.0")
 endif()

 # Different Qt versions can't be linked together
-if(HAVE_QT5 AND VTK_VERSION VERSION_LESS "6.0.0")
-  if(VTK_USE_QT)
-    message(STATUS "VTK support is disabled. Incompatible combination: OpenCV + Qt5 and VTK ver.${VTK_VERSION} + Qt4")
-  endif()
-endif()
-
-# Different Qt versions can't be linked together. VTK 6.0.0 doesn't provide a way to get Qt version it was linked with
-if(HAVE_QT5 AND VTK_VERSION VERSION_EQUAL "6.0.0" AND NOT DEFINED FORCE_VTK)
-  message(STATUS "VTK support is disabled. Possible incompatible combination: OpenCV+Qt5, and VTK ver.${VTK_VERSION} with Qt4")
-  message(STATUS "If it is known that VTK was compiled without Qt4, please define '-DFORCE_VTK=TRUE' flag in CMake")
+if((HAVE_QT AND VTK_USE_QT)
+    AND NOT DEFINED FORCE_VTK  # deprecated
+    AND NOT DEFINED OPENCV_FORCE_VTK
+)
+  message(STATUS "VTK support is disabled. Possible incompatible combination: OpenCV+Qt, and VTK ver.${VTK_VERSION} with Qt")
+  message(STATUS "If it is known that VTK was compiled without Qt4, please define '-DOPENCV_FORCE_VTK=TRUE' flag in CMake")
  return()
 endif()

-# Different Qt versions can't be linked together
-if(HAVE_QT AND VTK_VERSION VERSION_GREATER "6.0.0" AND NOT ${VTK_QT_VERSION} STREQUAL "")
-  if(HAVE_QT5 AND ${VTK_QT_VERSION} EQUAL "4")
-    message(STATUS "VTK support is disabled. Incompatible combination: OpenCV + Qt5 and VTK ver.${VTK_VERSION} + Qt4")
-    return()
-  endif()
-
-  if(NOT HAVE_QT5 AND ${VTK_QT_VERSION} EQUAL "5")
-    message(STATUS "VTK support is disabled. Incompatible combination: OpenCV + Qt4 and VTK ver.${VTK_VERSION} + Qt5")
-    return()
-  endif()
-endif()
-
 try_compile(VTK_COMPILE_STATUS
    "${OpenCV_BINARY_DIR}"
    "${OpenCV_SOURCE_DIR}/cmake/checks/vtk_test.cpp"
@@ -23,7 +23,7 @@ set(OPENCV_DOWNLOAD_LOG "${OpenCV_BINARY_DIR}/CMakeDownloadLog.txt")
 set(OPENCV_DOWNLOAD_WITH_CURL "${OpenCV_BINARY_DIR}/download_with_curl.sh")
 set(OPENCV_DOWNLOAD_WITH_WGET "${OpenCV_BINARY_DIR}/download_with_wget.sh")
 set(OPENCV_DOWNLOAD_TRIES_LIST 1 CACHE STRING "List of download tries") # a list
-set(OPENCV_DOWNLOAD_PARAMS INACTIVITY_TIMEOUT 60 TIMEOUT 600 CACHE STRING "Download parameters to be passed to file(DOWNLAOD ...)")
+set(OPENCV_DOWNLOAD_PARAMS INACTIVITY_TIMEOUT 60 TIMEOUT 600 CACHE STRING "Download parameters to be passed to file(DOWNLOAD ...)")
 mark_as_advanced(OPENCV_DOWNLOAD_TRIES_LIST OPENCV_DOWNLOAD_PARAMS)

 # Init download cache directory and log file and helper scripts
@@ -17,7 +17,7 @@ else()
 endif()

 # --- Concurrency ---
-if(MSVC AND NOT HAVE_TBB)
+if(MSVC AND NOT HAVE_TBB AND NOT OPENCV_DISABLE_THREAD_SUPPORT)
  set(_fname "${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeTmp/concurrencytest.cpp")
  file(WRITE "${_fname}" "#if _MSC_VER < 1600\n#error\n#endif\nint main() { return 0; }\n")
  try_compile(HAVE_CONCURRENCY "${CMAKE_BINARY_DIR}" "${_fname}")
@@ -151,7 +151,7 @@ macro(ipp_detect_version)
        if("${name}" STREQUAL "core")  # https://github.com/opencv/opencv/pull/19681
          if(OPENCV_FORCE_IPP_EXCLUDE_LIBS OR OPENCV_FORCE_IPP_EXCLUDE_LIBS_CORE
              OR (UNIX AND NOT ANDROID AND NOT APPLE
-                  AND (CMAKE_CXX_COMPILER_ID MATCHES "GNU" OR CMAKE_CXX_COMPILER_ID MATCHES "Clang")
+                  AND CMAKE_CXX_COMPILER_ID MATCHES "GNU|Clang|Intel"
              )
              AND NOT OPENCV_SKIP_IPP_EXCLUDE_LIBS_CORE
          )
@@ -2,71 +2,54 @@
 #  Detect 3rd-party GUI libraries
 # ----------------------------------------------------------------------------

-#--- Win32 UI ---
-ocv_clear_vars(HAVE_WIN32UI)
-if(WITH_WIN32UI)
-  try_compile(HAVE_WIN32UI
-    "${OpenCV_BINARY_DIR}"
-    "${OpenCV_SOURCE_DIR}/cmake/checks/win32uitest.cpp"
-    CMAKE_FLAGS "-DLINK_LIBRARIES:STRING=user32;gdi32")
-endif()
-
-# --- QT4 ---
+# --- QT4/5 ---
 ocv_clear_vars(HAVE_QT HAVE_QT5)
+
+macro(ocv_find_package_Qt4)
+  find_package(Qt4 COMPONENTS QtCore QtGui QtTest ${ARGN})
+  if(QT4_FOUND)
+    set(QT_FOUND 1)
+    ocv_assert(QT_VERSION_MAJOR EQUAL 4)
+  endif()
+endmacro()
+
+macro(ocv_find_package_Qt OCV_QT_VER)
+  find_package(Qt${OCV_QT_VER} COMPONENTS Core Gui Widgets Test Concurrent ${ARGN} NO_MODULE)
+  if(Qt${OCV_QT_VER}_FOUND)
+    set(QT_FOUND 1)
+    set(QT_VERSION "${Qt${OCV_QT_VER}_VERSION}")
+    set(QT_VERSION_MAJOR "${Qt${OCV_QT_VER}_VERSION_MAJOR}")
+    set(QT_VERSION_MINOR "${Qt${OCV_QT_VER}_VERSION_MINOR}")
+    set(QT_VERSION_PATCH "${Qt${OCV_QT_VER}_VERSION_PATCH}")
+    set(QT_VERSION_TWEAK "${Qt${OCV_QT_VER}_VERSION_TWEAK}")
+    set(QT_VERSION_COUNT "${Qt${OCV_QT_VER}_VERSION_COUNT}")
+  endif()
+endmacro()
+
 if(WITH_QT)
-  if(NOT WITH_QT EQUAL 4)
-    find_package(Qt5 COMPONENTS Core Gui Widgets Test Concurrent REQUIRED NO_MODULE)
-    if(Qt5_FOUND)
-      set(HAVE_QT5 ON)
-      set(HAVE_QT  ON)
-      find_package(Qt5 COMPONENTS OpenGL QUIET)
-      if(Qt5OpenGL_FOUND)
-        set(QT_QTOPENGL_FOUND ON)
+  if(NOT WITH_QT GREATER 0)
+    # BUG: Qt5Config.cmake script can't handle components properly: find_package(QT NAMES Qt6 Qt5 REQUIRED NO_MODULE COMPONENTS Core Gui Widgets Test Concurrent)
+    ocv_find_package_Qt(6 QUIET)
+    if(NOT QT_FOUND)
+      ocv_find_package_Qt(5 QUIET)
+    endif()
+    if(NOT QT_FOUND)
+      ocv_find_package_Qt4(QUIET)
+    endif()
+  elseif(WITH_QT EQUAL 4)
+    ocv_find_package_Qt4(REQUIRED)
+  else()  # WITH_QT=<major version>
+    ocv_find_package_Qt("${WITH_QT}" REQUIRED)
+  endif()
+  if(QT_FOUND)
+    set(HAVE_QT ON)
+    if(QT_VERSION_MAJOR GREATER 4)
+      find_package(Qt${QT_VERSION_MAJOR} COMPONENTS OpenGL QUIET)
+      if(Qt${QT_VERSION_MAJOR}OpenGL_FOUND)
+        set(QT_QTOPENGL_FOUND ON)  # HAVE_QT_OPENGL is defined below
      endif()
    endif()
  endif()
-
-  if(NOT HAVE_QT)
-    find_package(Qt4 REQUIRED QtCore QtGui QtTest)
-    if(QT4_FOUND)
-      set(HAVE_QT TRUE)
-    endif()
-  endif()
-endif()
-
-# --- GTK ---
-ocv_clear_vars(HAVE_GTK HAVE_GTK3 HAVE_GTHREAD HAVE_GTKGLEXT)
-if(WITH_GTK AND NOT HAVE_QT)
-  if(NOT WITH_GTK_2_X)
-    ocv_check_modules(GTK3 gtk+-3.0)
-    if(HAVE_GTK3)
-      ocv_append_build_options(HIGHGUI GTK3)
-      set(HAVE_GTK TRUE)
-    endif()
-  endif()
-  if(NOT HAVE_GTK)
-    ocv_check_modules(GTK2 gtk+-2.0)
-    if(HAVE_GTK2)
-      if (GTK2_VERSION VERSION_LESS MIN_VER_GTK)
-        message (FATAL_ERROR "GTK support requires a minimum version of ${MIN_VER_GTK} (${GTK2_VERSION} found)")
-      else()
-        ocv_append_build_options(HIGHGUI GTK2)
-        set(HAVE_GTK TRUE)
-      endif()
-    endif()
-  endif()
-  ocv_check_modules(GTHREAD gthread-2.0)
-  if(HAVE_GTK AND NOT HAVE_GTHREAD)
-    message(FATAL_ERROR "gthread not found. This library is required when building with GTK support")
-  else()
-    ocv_append_build_options(HIGHGUI GTHREAD)
-  endif()
-  if(WITH_OPENGL AND NOT HAVE_GTK3)
-    ocv_check_modules(GTKGLEXT gtkglext-1.0)
-    if(HAVE_GTKGLEXT)
-      ocv_append_build_options(HIGHGUI GTKGLEXT)
-    endif()
-  endif()
 endif()

 # --- OpenGl ---
@@ -29,7 +29,7 @@ if(WITH_IPP)
    if(OPENCV_FORCE_IPP_EXCLUDE_LIBS
        OR (HAVE_IPP_ICV
            AND UNIX AND NOT ANDROID AND NOT APPLE
-            AND (CMAKE_CXX_COMPILER_ID MATCHES "GNU" OR CMAKE_CXX_COMPILER_ID MATCHES "Clang")
+            AND CMAKE_CXX_COMPILER_ID MATCHES "GNU|Clang|Intel"
        )
        AND NOT OPENCV_SKIP_IPP_EXCLUDE_LIBS
    )
@@ -9,6 +9,14 @@
 # OPENEXR_LIBRARIES = libraries that are needed to use OpenEXR.
 #

+find_package(OpenEXR 3.0 CONFIG QUIET)
+if(TARGET OpenEXR::OpenEXR)
+    SET(OPENEXR_FOUND TRUE)
+    SET(OPENEXR_LIBRARIES OpenEXR::OpenEXR)
+    SET(OPENEXR_VERSION ${OpenEXR_VERSION})
+    return()
+endif()
+
 SET(OPENEXR_LIBRARIES "")
 SET(OPENEXR_LIBSEARCH_SUFFIXES "")
 file(TO_CMAKE_PATH "$ENV{ProgramFiles}" ProgramFiles_ENV_PATH)
@@ -6,4 +6,3 @@ set(MIN_VER_CUDNN 7.5)
 set(MIN_VER_PYTHON2 2.7)
 set(MIN_VER_PYTHON3 3.2)
 set(MIN_VER_ZLIB 1.2.3)
-set(MIN_VER_GTK 2.18.0)
@@ -254,7 +254,7 @@ function(_glob_locations out_paths out_names)
    list(LENGTH paths before)
    get_filename_component(path "${path}" ABSOLUTE)
    # Either module itself
-    if(NOT path STREQUAL CMAKE_CURRENT_SOURCE_DIR AND EXISTS "${path}/CMakeLists.txt")
+    if(NOT path STREQUAL "${OpenCV_SOURCE_DIR}/modules" AND EXISTS "${path}/CMakeLists.txt")
      get_filename_component(name "${path}" NAME)
      list(APPEND paths "${path}")
      list(APPEND names "${name}")
@@ -296,7 +296,7 @@ macro(_add_modules_1 paths names)
      list(GET ${names} ${i} __name)
      #message(STATUS "First pass: ${__name} => ${__path}")
      include("${__path}/cmake/init.cmake" OPTIONAL)
-      add_subdirectory("${__path}" "${CMAKE_CURRENT_BINARY_DIR}/.firstpass/${__name}")
+      add_subdirectory("${__path}" "${OpenCV_BINARY_DIR}/modules/.firstpass/${__name}")
    endforeach()
  endif()
 endmacro()
@@ -316,7 +316,7 @@ macro(_add_modules_2)
      endif()
      string(REGEX REPLACE "^opencv_" "" name "${m}")
      #message(STATUS "Second pass: ${name} => ${OPENCV_MODULE_${m}_LOCATION}")
-      add_subdirectory("${OPENCV_MODULE_${m}_LOCATION}" "${CMAKE_CURRENT_BINARY_DIR}/${name}")
+      add_subdirectory("${OPENCV_MODULE_${m}_LOCATION}" "${OpenCV_BINARY_DIR}/modules/${name}")
    endif()
    ocv_cmake_hook(POST_MODULES_CREATE_${the_module})
  endforeach()
@@ -369,7 +369,6 @@ macro(ocv_glob_modules main_root)
  __ocv_resolve_dependencies()

  # create modules
-  set(OPENCV_INITIAL_PASS OFF PARENT_SCOPE)
  set(OPENCV_INITIAL_PASS OFF)
  ocv_cmake_hook(PRE_MODULES_CREATE)
  _add_modules_2(${OPENCV_MODULES_BUILD})
@@ -377,6 +376,37 @@ macro(ocv_glob_modules main_root)
 endmacro()


+# called by root CMakeLists.txt
+macro(ocv_register_modules)
+  if(NOT OPENCV_MODULES_PATH)
+    set(OPENCV_MODULES_PATH "${OpenCV_SOURCE_DIR}/modules")
+  endif()
+
+  ocv_glob_modules(${OPENCV_MODULES_PATH} ${OPENCV_EXTRA_MODULES_PATH})
+
+  # build lists of modules to be documented
+  set(OPENCV_MODULES_MAIN "")
+  set(OPENCV_MODULES_EXTRA "")
+
+  foreach(mod ${OPENCV_MODULES_BUILD} ${OPENCV_MODULES_DISABLED_USER} ${OPENCV_MODULES_DISABLED_AUTO} ${OPENCV_MODULES_DISABLED_FORCE})
+    string(REGEX REPLACE "^opencv_" "" mod "${mod}")
+    if("${OPENCV_MODULE_opencv_${mod}_LOCATION}" STREQUAL "${OpenCV_SOURCE_DIR}/modules/${mod}")
+      list(APPEND OPENCV_MODULES_MAIN ${mod})
+    else()
+      list(APPEND OPENCV_MODULES_EXTRA ${mod})
+    endif()
+  endforeach()
+  ocv_list_sort(OPENCV_MODULES_MAIN)
+  ocv_list_sort(OPENCV_MODULES_EXTRA)
+  set(FIXED_ORDER_MODULES core imgproc imgcodecs videoio highgui video calib3d features2d objdetect dnn ml flann photo stitching)
+  list(REMOVE_ITEM OPENCV_MODULES_MAIN ${FIXED_ORDER_MODULES})
+  set(OPENCV_MODULES_MAIN ${FIXED_ORDER_MODULES} ${OPENCV_MODULES_MAIN})
+
+  set(OPENCV_MODULES_MAIN ${OPENCV_MODULES_MAIN} CACHE INTERNAL "List of main modules" FORCE)
+  set(OPENCV_MODULES_EXTRA ${OPENCV_MODULES_EXTRA} CACHE INTERNAL "List of extra modules" FORCE)
+endmacro()
+
+
 # disables OpenCV module with missing dependencies
 function(__ocv_module_turn_off the_module)
  list(REMOVE_ITEM OPENCV_MODULES_DISABLED_AUTO "${the_module}")
@@ -877,10 +907,13 @@ macro(ocv_create_module)
 endmacro()

 macro(_ocv_create_module)
+  add_definitions(-D__OPENCV_BUILD=1)

  ocv_compiler_optimization_process_sources(OPENCV_MODULE_${the_module}_SOURCES OPENCV_MODULE_${the_module}_DEPS_EXT ${the_module})
  set(__module_headers ${OPENCV_MODULE_${the_module}_HEADERS})
-  list(SORT __module_headers)  # fix headers order, useful for bindings
+  if(__module_headers)
+    list(SORT __module_headers)  # fix headers order, useful for bindings
+  endif()
  set(OPENCV_MODULE_${the_module}_HEADERS ${__module_headers} CACHE INTERNAL "List of header files for ${the_module}")
  set(OPENCV_MODULE_${the_module}_SOURCES ${OPENCV_MODULE_${the_module}_SOURCES} CACHE INTERNAL "List of source files for ${the_module}")

@@ -1181,6 +1214,9 @@ function(ocv_add_perf_tests)
      if(TARGET opencv_videoio_plugins)
        add_dependencies(${the_target} opencv_videoio_plugins)
      endif()
+      if(TARGET opencv_highgui_plugins)
+        add_dependencies(${the_target} opencv_highgui_plugins)
+      endif()

      if(HAVE_HPX)
        message("Linking HPX to Perf test of module ${name}")
@@ -1276,6 +1312,9 @@ function(ocv_add_accuracy_tests)
      if(TARGET opencv_videoio_plugins)
        add_dependencies(${the_target} opencv_videoio_plugins)
      endif()
+      if(TARGET opencv_highgui_plugins)
+        add_dependencies(${the_target} opencv_highgui_plugins)
+      endif()

      if(HAVE_HPX)
        message("Linking HPX to Perf test of module ${name}")
@@ -1366,6 +1405,9 @@ function(ocv_add_samples)
        if(TARGET opencv_videoio_plugins)
          add_dependencies(${the_target} opencv_videoio_plugins)
        endif()
+        if(TARGET opencv_highgui_plugins)
+          add_dependencies(${the_target} opencv_highgui_plugins)
+        endif()

        if(INSTALL_BIN_EXAMPLES)
          install(TARGETS ${the_target} RUNTIME DESTINATION "${OPENCV_SAMPLES_BIN_INSTALL_PATH}/${module_id}" COMPONENT samples)
@@ -866,7 +866,9 @@ macro(ocv_check_modules define)
      foreach(flag ${${define}_LDFLAGS})
        if(flag MATCHES "^-L(.*)")
          list(APPEND _libs_paths ${CMAKE_MATCH_1})
-        elseif(IS_ABSOLUTE "${flag}")
+        elseif(IS_ABSOLUTE "${flag}"
+            OR flag STREQUAL "-lstdc++"
+        )
          list(APPEND _libs "${flag}")
        elseif(flag MATCHES "^-l(.*)")
          set(_lib "${CMAKE_MATCH_1}")
@@ -1479,8 +1481,8 @@ function(ocv_target_link_libraries target)
      if(NOT LINK_PENDING STREQUAL "")
        __ocv_push_target_link_libraries(${LINK_MODE} ${LINK_PENDING})
        set(LINK_PENDING "")
-        set(LINK_MODE "${dep}")
      endif()
+      set(LINK_MODE "${dep}")
    else()
      if(BUILD_opencv_world)
        if(OPENCV_MODULE_${dep}_IS_PART_OF_WORLD)
@@ -1578,28 +1580,57 @@ endfunction()


 function(ocv_add_external_target name inc link def)
-  if(BUILD_SHARED_LIBS)
+  if(BUILD_SHARED_LIBS AND link)
    set(imp IMPORTED)
  endif()
  add_library(ocv.3rdparty.${name} INTERFACE ${imp})
-  set_target_properties(ocv.3rdparty.${name} PROPERTIES
-    INTERFACE_INCLUDE_DIRECTORIES "${inc}"
-    INTERFACE_SYSTEM_INCLUDE_DIRECTORIES "${inc}"
-    INTERFACE_COMPILE_DEFINITIONS "${def}")
-  # When cmake version is greater than or equal to 3.11, INTERFACE_LINK_LIBRARIES no longer applies to interface library
-  # See https://github.com/opencv/opencv/pull/18658
-  if (CMAKE_VERSION VERSION_LESS 3.11)
-    set_target_properties(ocv.3rdparty.${name} PROPERTIES
-      INTERFACE_LINK_LIBRARIES "${link}")
-  else()
-    target_link_libraries(ocv.3rdparty.${name} INTERFACE ${link})
+  if(def)
+    if(NOT (CMAKE_VERSION VERSION_LESS "3.11.0"))  # https://gitlab.kitware.com/cmake/cmake/-/merge_requests/1264 : eliminates "Cannot specify compile definitions for imported target" error message
+      target_compile_definitions(ocv.3rdparty.${name} INTERFACE "${def}")
+    else()
+      set_target_properties(ocv.3rdparty.${name} PROPERTIES INTERFACE_COMPILE_DEFINITIONS "${def}")
+    endif()
  endif()
-  #
-  if(NOT BUILD_SHARED_LIBS)
+  if(inc)
+    if(NOT (CMAKE_VERSION VERSION_LESS "3.11.0"))  # https://gitlab.kitware.com/cmake/cmake/-/merge_requests/1264 : eliminates "Cannot specify compile definitions for imported target" error message
+      target_include_directories(ocv.3rdparty.${name} SYSTEM INTERFACE "$<BUILD_INTERFACE:${inc}>")
+    else()
+      set_target_properties(ocv.3rdparty.${name} PROPERTIES
+          INTERFACE_INCLUDE_DIRECTORIES "$<BUILD_INTERFACE:${inc}>"
+          INTERFACE_SYSTEM_INCLUDE_DIRECTORIES "$<BUILD_INTERFACE:${inc}>"
+      )
+    endif()
+  endif()
+  if(link)
+    # When cmake version is greater than or equal to 3.11, INTERFACE_LINK_LIBRARIES no longer applies to interface library
+    # See https://github.com/opencv/opencv/pull/18658
+    if(CMAKE_VERSION VERSION_LESS 3.11)
+      set_target_properties(ocv.3rdparty.${name} PROPERTIES
+        INTERFACE_LINK_LIBRARIES "${link}")
+    else()
+      target_link_libraries(ocv.3rdparty.${name} INTERFACE ${link})
+    endif()
+  endif()
+  # to install used target only upgrade CMake
+  if(NOT BUILD_SHARED_LIBS
+      AND CMAKE_VERSION VERSION_LESS "3.13.0"  # https://gitlab.kitware.com/cmake/cmake/-/merge_requests/2152
+  )
    install(TARGETS ocv.3rdparty.${name} EXPORT OpenCVModules)
  endif()
 endfunction()

+# Returns the first non-interface target
+function(ocv_get_imported_target imported interface)
+  set(__result "${interface}")
+  get_target_property(__type "${__result}" TYPE)
+  if(__type STREQUAL "INTERFACE_LIBRARY")
+    get_target_property(__libs "${__result}" INTERFACE_LINK_LIBRARIES)
+    list(GET __libs 0 __interface)
+    ocv_get_imported_target(__result "${__interface}")
+  endif()
+  set(${imported} "${__result}" PARENT_SCOPE)
+endfunction()
+

 macro(ocv_get_libname var_name)
  get_filename_component(__libname "${ARGN}" NAME)
@@ -1954,3 +1985,9 @@ if(NOT BUILD_SHARED_LIBS AND (CMAKE_VERSION VERSION_LESS "3.14.0"))
 else()
  ocv_update(OPENCV_3RDPARTY_EXCLUDE_FROM_ALL "EXCLUDE_FROM_ALL")
 endif()
+
+
+#
+# Include configuration override settings
+#
+include("${CMAKE_CURRENT_LIST_DIR}/vars/EnableModeVars.cmake")
@@ -2,6 +2,17 @@
 set(ANDROID_GRADLE_PLUGIN_VERSION "3.2.1" CACHE STRING "Android Gradle Plugin version")
 message(STATUS "Android Gradle Plugin version: ${ANDROID_GRADLE_PLUGIN_VERSION}")

+set(KOTLIN_PLUGIN_VERSION "1.4.10" CACHE STRING "Kotlin Plugin version")
+message(STATUS "kotlin Plugin version: ${KOTLIN_GRADLE_PLUGIN_VERSION}")
+
+if(BUILD_KOTLIN_EXTENSIONS)
+  set(KOTLIN_PLUGIN_DECLARATION "apply plugin: 'kotlin-android'" CACHE STRING "Kotlin Plugin version")
+  set(KOTLIN_STD_LIB "implementation 'org.jetbrains.kotlin:kotlin-stdlib:${KOTLIN_PLUGIN_VERSION}'" CACHE STRING "Kotlin Standard Library dependency")
+else()
+  set(KOTLIN_PLUGIN_DECLARATION "" CACHE STRING "Kotlin Plugin version")
+  set(KOTLIN_STD_LIB "" CACHE STRING "Kotlin Standard Library dependency")
+endif()
+
 set(GRADLE_VERSION "5.6.4" CACHE STRING "Gradle version")
 message(STATUS "Gradle version: ${GRADLE_VERSION}")

@@ -9,7 +9,7 @@
 int test()
 {
    const float src[] = { 0.0f, 0.0f, 0.0f, 0.0f };
-    vfloat32m1_t val = vle32_v_f32m1((const float*)(src));
+    vfloat32m1_t val = vle32_v_f32m1((const float*)(src), 4);
    return (int)vfmv_f_s_f32m1_f32(val);
 }
 #else
@@ -137,6 +137,20 @@ elseif(MSVC)
        set(OpenCV_RUNTIME vc14) # selecting previous compatible runtime version
      endif()
    endif()
+  elseif(MSVC_VERSION MATCHES "^193[0-9]$")
+    set(OpenCV_RUNTIME vc17)
+    check_one_config(has_VS2022)
+    if(NOT has_VS2022)
+     set(OpenCV_RUNTIME vc16)
+     check_one_config(has_VS2019)
+     if(NOT has_VS2019)
+       set(OpenCV_RUNTIME vc15) # selecting previous compatible runtime version
+       check_one_config(has_VS2017)
+       if(NOT has_VS2017)
+         set(OpenCV_RUNTIME vc14) # selecting previous compatible runtime version
+       endif()
+     endif()
+    endif()
  endif()
 elseif(MINGW)
  set(OpenCV_RUNTIME mingw)
@@ -28,9 +28,6 @@
 /* Clp support */
 #cmakedefine HAVE_CLP

-/* Cocoa API */
-#cmakedefine HAVE_COCOA
-
 /* NVIDIA CUDA Runtime API*/
 #cmakedefine HAVE_CUDA

@@ -56,12 +53,6 @@
 /* Geospatial Data Abstraction Library */
 #cmakedefine HAVE_GDAL

-/* GTK+ 2.0 Thread support */
-#cmakedefine HAVE_GTHREAD
-
-/* GTK+ 2.x toolkit */
-#cmakedefine HAVE_GTK
-
 /* Halide support */
 #cmakedefine HAVE_HALIDE

@@ -121,12 +112,6 @@
 /* parallel_for with pthreads */
 #cmakedefine HAVE_PTHREADS_PF

-/* Qt support */
-#cmakedefine HAVE_QT
-
-/* Qt OpenGL support */
-#cmakedefine HAVE_QT_OPENGL
-
 /* Intel Threading Building Blocks */
 #cmakedefine HAVE_TBB

@@ -136,9 +121,6 @@
 /* TIFF codec */
 #cmakedefine HAVE_TIFF

-/* Win32 UI */
-#cmakedefine HAVE_WIN32UI
-
 /* Define if your processor stores words with the most significant byte
   first (like Motorola and SPARC, unlike Intel and VAX). */
 #cmakedefine WORDS_BIGENDIAN
@@ -0,0 +1,21 @@
+set(__OCV_MODE_VARS_DIR "${CMAKE_CURRENT_LIST_DIR}")
+
+macro(ocv_change_mode_var)
+  set(__var "${ARGV0}")
+  set(__mode "${ARGV1}")
+  set(__value "${ARGV2}")
+  if(__mode STREQUAL "MODIFIED_ACCESS" AND __value)
+    if(NOT __applied_mode_${__var})
+      include("${__OCV_MODE_VARS_DIR}/${__var}.cmake")
+      set(__applied_mode_${__var} 1)
+    else()
+      #message("Mode is already applied: ${__var}")
+    endif()
+  endif()
+endmacro()
+
+variable_watch(OPENCV_DISABLE_THREAD_SUPPORT ocv_change_mode_var)
+set(OPENCV_DISABLE_THREAD_SUPPORT "${OPENCV_DISABLE_THREAD_SUPPORT}")
+
+variable_watch(OPENCV_SEMIHOSTING ocv_change_mode_var)
+set(OPENCV_SEMIHOSTING "${OPENCV_SEMIHOSTING}")
@@ -0,0 +1,28 @@
+# Force removal of code conditionally compiled with `#if
+# HAVE_PTHREAD`.
+ocv_update(HAVE_PTHREAD 0)
+
+# There components are disabled because they require
+# multi-threaded execution.
+ocv_update(WITH_PROTOBUF OFF)
+ocv_update(WITH_GSTREAMER OFF)
+ocv_update(WITH_IPP OFF)
+ocv_update(WITH_ITT OFF)
+ocv_update(WITH_OPENCL OFF)
+ocv_update(WITH_VA OFF)
+ocv_update(WITH_VA_INTEL OFF)
+
+# Disable bindings
+ocv_update(BUILD_opencv_python2 OFF)
+ocv_update(BUILD_opencv_python3 OFF)
+ocv_update(BUILD_JAVA OFF)
+ocv_update(BUILD_opencv_java OFF)
+
+# These modules require `#include
+# <[thread|mutex|condition_variable|future]>` and linkage into
+# `libpthread` to work.
+ocv_update(BUILD_opencv_objdetect OFF)
+ocv_update(BUILD_opencv_gapi OFF)
+ocv_update(BUILD_opencv_dnn OFF)
+
+set(OPJ_USE_THREAD "OFF" CACHE INTERNAL "")
@@ -0,0 +1,10 @@
+set(CV_TRACE OFF)
+
+# These third parties libraries are incompatible with the semihosting
+# toolchain.
+set(WITH_JPEG OFF)
+set(WITH_OPENEXR OFF)
+set(WITH_TIFF OFF)
+
+# Turn off `libpng` for some linking issues.
+set(WITH_PNG OFF)
@@ -31,7 +31,7 @@ MULTILINE_CPP_IS_BRIEF = NO
 INHERIT_DOCS           = YES
 SEPARATE_MEMBER_PAGES  = NO
 TAB_SIZE               = 4
-ALIASES               += add_toggle{1}="@htmlonly[block] <div class='newInnerHTML'>\1</div><div> <script type="text/javascript"> addToggle(); </script>@endhtmlonly"
+ALIASES               += add_toggle{1}="@htmlonly[block] <div class='newInnerHTML'>\1</div><div> <script type='text/javascript'> addToggle(); </script>@endhtmlonly"
 ALIASES               += add_toggle_cpp="@htmlonly[block] <div class='newInnerHTML' title='cpp' style='display: none;'>C++</div><div class='toggleable_div label_cpp' style='display: none;'>@endhtmlonly"
 ALIASES               += add_toggle_java="@htmlonly[block] <div class='newInnerHTML' title='java' style='display: none;'>Java</div><div class='toggleable_div label_java' style='display: none;'>@endhtmlonly"
 ALIASES               += add_toggle_python="@htmlonly[block] <div class='newInnerHTML' title='python' style='display: none;'>Python</div><div class='toggleable_div label_python' style='display: none;'>@endhtmlonly"
@@ -106,7 +106,7 @@ RECURSIVE              = YES
 EXCLUDE                = @CMAKE_DOXYGEN_EXCLUDE_LIST@
 EXCLUDE_SYMLINKS       = NO
 EXCLUDE_PATTERNS       = *.inl.hpp *.impl.hpp *_detail.hpp */cudev/**/detail/*.hpp *.m */opencl/runtime/* */legacy/* *_c.h @DOXYGEN_EXCLUDE_PATTERNS@
-EXCLUDE_SYMBOLS        = cv::DataType<*> cv::traits::* int void CV__* T __CV*
+EXCLUDE_SYMBOLS        = cv::DataType<*> cv::traits::* int void CV__* T __CV* cv::gapi::detail*
 EXAMPLE_PATH           = @CMAKE_DOXYGEN_EXAMPLE_PATH@
 EXAMPLE_PATTERNS       = *
 EXAMPLE_RECURSIVE      = YES
@@ -1,6 +1,9 @@
 Contour Features {#tutorial_js_contour_features}
 ================

+@prev_tutorial{tutorial_js_contours_begin}
+@next_tutorial{tutorial_js_contour_properties}
+
 Goal
 ----

@@ -1,6 +1,9 @@
 Contour Properties {#tutorial_js_contour_properties}
 ==================

+@prev_tutorial{tutorial_js_contour_features}
+@next_tutorial{tutorial_js_contours_more_functions}
+
 Goal
 ----

@@ -1,6 +1,8 @@
 Contours : Getting Started {#tutorial_js_contours_begin}
 ==========================

+@next_tutorial{tutorial_js_contour_features}
+
 Goal
 ----

@@ -1,6 +1,8 @@
 Contours Hierarchy {#tutorial_js_contours_hierarchy}
 ==================

+@prev_tutorial{tutorial_js_contours_more_functions}
+
 Goal
 ----

@@ -1,6 +1,9 @@
 Contours : More Functions {#tutorial_js_contours_more_functions}
 =========================

+@prev_tutorial{tutorial_js_contour_properties}
+@next_tutorial{tutorial_js_contours_hierarchy}
+
 Goal
 ----

@@ -850,12 +850,12 @@
  journal = {IEEE Transactions on Robotics and Automation},
  title = {Robot sensor calibration: solving AX=XB on the Euclidean group},
  year = {1994},
+  month = oct,
  volume = {10},
  number = {5},
  pages = {717-721},
  doi = {10.1109/70.326576},
-  ISSN = {1042-296X},
-  month = {Oct}
+  issn = {1042-296X}
 }
@inproceedings{PM03,
  author = {P{\'e}rez, Patrick and Gangnet, Michel and Blake, Andrew},
@@ -1051,12 +1051,12 @@
  journal = {IEEE Transactions on Robotics and Automation},
  title = {A new technique for fully autonomous and efficient 3D robotics hand/eye calibration},
  year = {1989},
+  month = jun,
  volume = {5},
  number = {3},
  pages = {345-358},
  doi = {10.1109/70.34770},
-  ISSN = {1042-296X},
-  month = {June}
+  issn = {1042-296X}
 }
@inproceedings{UES01,
  author = {Uyttendaele, Matthew and Eden, Ashley and Skeliski, R},
@@ -1324,3 +1324,13 @@
  pages={5551--5560},
  year={2017}
 }
+@article{umeyama1991least,
+  title={Least-squares estimation of transformation parameters between two point patterns},
+  author={Umeyama, Shinji},
+  journal={IEEE Computer Architecture Letters},
+  volume={13},
+  number={04},
+  pages={376--380},
+  year={1991},
+  publisher={IEEE Computer Society}
+}
@@ -6,13 +6,14 @@ python gen_pattern.py -o out.svg -r 11 -c 8 -T circles -s 20.0 -R 5.0 -u mm -w 2
 -o, --output - output file (default out.svg)
 -r, --rows - pattern rows (default 11)
 -c, --columns - pattern columns (default 8)
-T, --type - type of pattern, circles, acircles, checkerboard (default circles)
+-T, --type - type of pattern, circles, acircles, checkerboard, radon_checkerboard (default circles)
 -s, --square_size - size of squares in pattern (default 20.0)
 -R, --radius_rate - circles_radius = square_size/radius_rate (default 5.0)
 -u, --units - mm, inches, px, m (default mm)
 -w, --page_width - page width in units (default 216)
 -h, --page_height - page height in units (default 279)
 -a, --page_size - page size (default A4), supersedes -h -w arguments
+-m, --markers - list of cells with markers for the radon checkerboard
 -H, --help - show help
 """

@@ -22,7 +23,7 @@ from svgfig import *


 class PatternMaker:
-    def __init__(self, cols, rows, output, units, square_size, radius_rate, page_width, page_height):
+    def __init__(self, cols, rows, output, units, square_size, radius_rate, page_width, page_height, markers):
        self.cols = cols
        self.rows = rows
        self.output = output
@@ -31,6 +32,7 @@ class PatternMaker:
        self.radius_rate = radius_rate
        self.width = page_width
        self.height = page_height
+        self.markers = markers
        self.g = SVG("g")  # the svg group container

    def make_circles_pattern(self):
@@ -70,6 +72,74 @@ class PatternMaker:
                                 height=spacing, fill="black", stroke="none")
                    self.g.append(square)

+    @staticmethod
+    def _make_round_rect(x, y, diam, corners=("right", "right", "right", "right")):
+        rad = diam / 2
+        cw_point = ((0, 0), (diam, 0), (diam, diam), (0, diam))
+        mid_cw_point = ((0, rad), (rad, 0), (diam, rad), (rad, diam))
+        res_str = "M{},{} ".format(x + mid_cw_point[0][0], y + mid_cw_point[0][1])
+        n = len(cw_point)
+        for i in range(n):
+            if corners[i] == "right":
+                res_str += "L{},{} L{},{} ".format(x + cw_point[i][0], y + cw_point[i][1],
+                                                   x + mid_cw_point[(i + 1) % n][0], y + mid_cw_point[(i + 1) % n][1])
+            elif corners[i] == "round":
+                res_str += "A{},{} 0,0,1 {},{} ".format(rad, rad, x + mid_cw_point[(i + 1) % n][0],
+                                                        y + mid_cw_point[(i + 1) % n][1])
+            else:
+                raise TypeError("unknown corner type")
+        return res_str
+
+    def _get_type(self, x, y):
+        corners = ["right", "right", "right", "right"]
+        is_inside = True
+        if x == 0:
+            corners[0] = "round"
+            corners[3] = "round"
+            is_inside = False
+        if y == 0:
+            corners[0] = "round"
+            corners[1] = "round"
+            is_inside = False
+        if x == self.cols - 1:
+            corners[1] = "round"
+            corners[2] = "round"
+            is_inside = False
+        if y == self.rows - 1:
+            corners[2] = "round"
+            corners[3] = "round"
+            is_inside = False
+        return corners, is_inside
+
+    def make_radon_checkerboard_pattern(self):
+        spacing = self.square_size
+        xspacing = (self.width - self.cols * self.square_size) / 2.0
+        yspacing = (self.height - self.rows * self.square_size) / 2.0
+        for x in range(0, self.cols):
+            for y in range(0, self.rows):
+                if x % 2 == y % 2:
+                    corner_types, is_inside = self._get_type(x, y)
+                    if is_inside:
+                        square = SVG("rect", x=x * spacing + xspacing, y=y * spacing + yspacing, width=spacing,
+                                     height=spacing, fill="black", stroke="none")
+                    else:
+                        square = SVG("path", d=self._make_round_rect(x * spacing + xspacing, y * spacing + yspacing,
+                                                                     spacing, corner_types), fill="black", stroke="none")
+                    self.g.append(square)
+        if self.markers is not None:
+            r = self.square_size * 0.17
+            pattern_width = ((self.cols - 1.0) * spacing) + (2.0 * r)
+            pattern_height = ((self.rows - 1.0) * spacing) + (2.0 * r)
+            x_spacing = (self.width - pattern_width) / 2.0
+            y_spacing = (self.height - pattern_height) / 2.0
+            for x, y in self.markers:
+                color = "black"
+                if x % 2 == y % 2:
+                    color = "white"
+                dot = SVG("circle", cx=(x * spacing) + x_spacing + r,
+                          cy=(y * spacing) + y_spacing + r, r=r, fill=color, stroke="none")
+                self.g.append(dot)
+
    def save(self):
        c = canvas(self.g, width="%d%s" % (self.width, self.units), height="%d%s" % (self.height, self.units),
                   viewBox="0 0 %d %d" % (self.width, self.height))
@@ -85,7 +155,7 @@ def main():
                        type=int)
    parser.add_argument("-r", "--rows", help="pattern rows", default="11", action="store", dest="rows", type=int)
    parser.add_argument("-T", "--type", help="type of pattern", default="circles", action="store", dest="p_type",
-                        choices=["circles", "acircles", "checkerboard"])
+                        choices=["circles", "acircles", "checkerboard", "radon_checkerboard"])
    parser.add_argument("-u", "--units", help="length unit", default="mm", action="store", dest="units",
                        choices=["mm", "inches", "px", "m"])
    parser.add_argument("-s", "--square_size", help="size of squares in pattern", default="20.0", action="store",
@@ -96,8 +166,12 @@ def main():
                        dest="page_width", type=float)
    parser.add_argument("-h", "--page_height", help="page height in units", default=argparse.SUPPRESS, action="store",
                        dest="page_height", type=float)
-    parser.add_argument("-a", "--page_size", help="page size, superseded if -h and -w are set", default="A4", action="store",
-                        dest="page_size", choices=["A0", "A1", "A2", "A3", "A4", "A5"])
+    parser.add_argument("-a", "--page_size", help="page size, superseded if -h and -w are set", default="A4",
+                        action="store", dest="page_size", choices=["A0", "A1", "A2", "A3", "A4", "A5"])
+    parser.add_argument("-m", "--markers", help="list of cells with markers for the radon checkerboard. Marker "
+                                                "coordinates as list of numbers: -m 1 2 3 4 means markers in cells "
+                                                "[1, 2] and [3, 4]",
+                        action="store", dest="markers", nargs="+", type=int)
    args = parser.parse_args()

    show_help = args.show_help
@@ -121,10 +195,19 @@ def main():
                      "A5": [148, 210]}
        page_width = page_sizes[page_size][0]
        page_height = page_sizes[page_size][1]
-    pm = PatternMaker(columns, rows, output, units, square_size, radius_rate, page_width, page_height)
+    if len(args.markers) % 2 == 1:
+        raise ValueError("The length of the markers array={} must be even".format(len(args.markers)))
+    markers = set()
+    for x, y in zip(args.markers[::2], args.markers[1::2]):
+        if x in range(0, columns) and y in range(0, rows):
+            markers.add((x, y))
+        else:
+            raise ValueError("The marker {},{} is outside the checkerboard".format(x, y))
+
+    pm = PatternMaker(columns, rows, output, units, square_size, radius_rate, page_width, page_height, markers)
    # dict for easy lookup of pattern type
    mp = {"circles": pm.make_circles_pattern, "acircles": pm.make_acircles_pattern,
-          "checkerboard": pm.make_checkerboard_pattern}
+          "checkerboard": pm.make_checkerboard_pattern, "radon_checkerboard": pm.make_radon_checkerboard_pattern}
    mp[p_type]()
    # this should save pattern to output
    pm.save()
@@ -60,6 +60,14 @@ of C++.

 So this is the basic version of how OpenCV-Python bindings are generated.

+@note There is no 1:1 mapping of numpy.ndarray on cv::Mat. For example, cv::Mat has channels field,
+which is emulated as last dimension of numpy.ndarray and implicitly converted.
+However, such implicit conversion has problem with passing of 3D numpy arrays into C++ code
+(the last dimension is implicitly reinterpreted as number of channels).
+Refer to the [issue](https://github.com/opencv/opencv/issues/19091) for workarounds if you need to process 3D arrays or ND-arrays with channels.
+OpenCV 4.5.4+ has `cv.Mat` wrapper derived from `numpy.ndarray` to explicitly handle the channels behavior.
+
+
 How to extend new modules to Python?
 ------------------------------------

@@ -98,7 +98,7 @@ import numpy as np
 import cv2 as cv
 from matplotlib import pyplot as plt

-img = cv.imread('simple.jpg',0)
+img = cv.imread('blox.jpg',0) # `<opencv_root>/samples/data/blox.jpg`

 # Initiate FAST object with default values
 fast = cv.FastFeatureDetector_create()
@@ -113,17 +113,17 @@ print( "nonmaxSuppression:{}".format(fast.getNonmaxSuppression()) )
 print( "neighborhood: {}".format(fast.getType()) )
 print( "Total Keypoints with nonmaxSuppression: {}".format(len(kp)) )

-cv.imwrite('fast_true.png',img2)
+cv.imwrite('fast_true.png', img2)

 # Disable nonmaxSuppression
 fast.setNonmaxSuppression(0)
-kp = fast.detect(img,None)
+kp = fast.detect(img, None)

 print( "Total Keypoints without nonmaxSuppression: {}".format(len(kp)) )

 img3 = cv.drawKeypoints(img, kp, None, color=(255,0,0))

-cv.imwrite('fast_false.png',img3)
+cv.imwrite('fast_false.png', img3)
@endcode
 See the results. First image shows FAST with nonmaxSuppression and second one without
 nonmaxSuppression:
@@ -40,12 +40,12 @@ using **cv.Sobel()**).
 Then comes the main part. After this, they created a score, basically an equation, which
 determines if a window can contain a corner or not.

-\f[R = det(M) - k(trace(M))^2\f]
+\f[R = \det(M) - k(\operatorname{trace}(M))^2\f]

 where
-    -   \f$det(M) = \lambda_1 \lambda_2\f$
-    -   \f$trace(M) = \lambda_1 + \lambda_2\f$
-    -   \f$\lambda_1\f$ and \f$\lambda_2\f$ are the eigenvalues of M
+    -   \f$\det(M) = \lambda_1 \lambda_2\f$
+    -   \f$\operatorname{trace}(M) = \lambda_1 + \lambda_2\f$
+    -   \f$\lambda_1\f$ and \f$\lambda_2\f$ are the eigenvalues of \f$M\f$

 So the magnitudes of these eigenvalues decide whether a region is a corner, an edge, or flat.

@@ -20,7 +20,7 @@ Harris Corner Detector. The scoring function in Harris Corner Detector was given

 Instead of this, Shi-Tomasi proposed:

-\f[R = min(\lambda_1, \lambda_2)\f]
+\f[R = \min(\lambda_1, \lambda_2)\f]

 If it is a greater than a threshold value, it is considered as a corner. If we plot it in
 \f$\lambda_1 - \lambda_2\f$ space as we did in Harris Corner Detector, we get an image as below:
@@ -28,7 +28,7 @@ If it is a greater than a threshold value, it is considered as a corner. If we p
 ![image](images/shitomasi_space.png)

 From the figure, you can see that only when \f$\lambda_1\f$ and \f$\lambda_2\f$ are above a minimum value,
-\f$\lambda_{min}\f$, it is considered as a corner(green region).
+\f$\lambda_{\min}\f$, it is considered as a corner(green region).

 Code
 ----
@@ -156,7 +156,7 @@ sift = cv.SIFT_create()
 kp, des = sift.detectAndCompute(gray,None)
@endcode
 Here kp will be a list of keypoints and des is a numpy array of shape
-\f$Number\_of\_Keypoints \times 128\f$.
+\f$\text{(Number of Keypoints)} \times 128\f$.

 So we got keypoints, descriptors etc. Now we want to see how to match keypoints in different images.
 That we will learn in coming chapters.
@@ -74,7 +74,7 @@ Canny Edge Detection in OpenCV

 OpenCV puts all the above in single function, **cv.Canny()**. We will see how to use it. First
 argument is our input image. Second and third arguments are our minVal and maxVal respectively.
-Third argument is aperture_size. It is the size of Sobel kernel used for find image gradients. By
+Fourth argument is aperture_size. It is the size of Sobel kernel used for find image gradients. By
 default it is 3. Last argument is L2gradient which specifies the equation for finding gradient
 magnitude. If it is True, it uses the equation mentioned above which is more accurate, otherwise it
 uses this function: \f$Edge\_Gradient \; (G) = |G_x| + |G_y|\f$. By default, it is False.
@@ -1,6 +1,9 @@
 Contour Features {#tutorial_py_contour_features}
 ================

+@prev_tutorial{tutorial_py_contours_begin}
+@next_tutorial{tutorial_py_contour_properties}
+
 Goal
 ----

@@ -91,7 +94,7 @@ convexity defects, which are the local maximum deviations of hull from contours.

 There is a little bit things to discuss about it its syntax:
@code{.py}
-hull = cv.convexHull(points[, hull[, clockwise[, returnPoints]]
+hull = cv.convexHull(points[, hull[, clockwise[, returnPoints]]])
@endcode
 Arguments details:

@@ -1,6 +1,9 @@
 Contour Properties {#tutorial_py_contour_properties}
 ==================

+@prev_tutorial{tutorial_py_contour_features}
+@next_tutorial{tutorial_py_contours_more_functions}
+
 Here we will learn to extract some frequently used properties of objects like Solidity, Equivalent
 Diameter, Mask image, Mean Intensity etc. More features can be found at [Matlab regionprops
 documentation](http://www.mathworks.in/help/images/ref/regionprops.html).
@@ -1,6 +1,8 @@
 Contours : Getting Started {#tutorial_py_contours_begin}
 ==========================

+@next_tutorial{tutorial_py_contour_features}
+
 Goal
 ----

@@ -1,6 +1,8 @@
 Contours Hierarchy {#tutorial_py_contours_hierarchy}
 ==================

+@prev_tutorial{tutorial_py_contours_more_functions}
+
 Goal
 ----

@@ -1,6 +1,10 @@
 Contours : More Functions {#tutorial_py_contours_more_functions}
 =========================

+@prev_tutorial{tutorial_py_contour_properties}
+@next_tutorial{tutorial_py_contours_hierarchy}
+
+
 Goal
 ----

@@ -107,17 +107,10 @@ def add_signature_to_table(soup, table, signature, language, type):
    """ Add a signature to an html table"""
    row = soup.new_tag('tr')
    row.append(soup.new_tag('td', style='width: 20px;'))
-
-    if 'ret' in signature:
-        row.append(append(soup.new_tag('td'), signature['ret']))
-        row.append(append(soup.new_tag('td'), '='))
-    else:
-        row.append(soup.new_tag('td')) # return values
-        row.append(soup.new_tag('td')) # '='
-
    row.append(append(soup.new_tag('td'), signature['name'] + '('))
    row.append(append(soup.new_tag('td', **{'class': 'paramname'}), signature['arg']))
-    row.append(append(soup.new_tag('td'), ')'))
+    row.append(append(soup.new_tag('td'), ') -> '))
+    row.append(append(soup.new_tag('td'), signature['ret']))
    table.append(row)


@@ -188,7 +188,7 @@ implementation below.

 This will return a similarity index for each channel of the image. This value is between zero and
 one, where one corresponds to perfect fit. Unfortunately, the many Gaussian blurring is quite
-costly, so while the PSNR may work in a real time like environment (24 frame per second) this will
+costly, so while the PSNR may work in a real time like environment (24 frames per second) this will
 take significantly more than to accomplish similar performance results.

 Therefore, the source code presented at the start of the tutorial will perform the PSNR measurement
@@ -36,6 +36,10 @@ create a circle board pattern in file acircleboard.svg with 7 rows, 5 columns an

        python gen_pattern.py -o acircleboard.svg --rows 7 --columns 5 --type acircles --square_size 10 --radius_rate 2

+create a radon checkerboard for findChessboardCornersSB() with markers in (7 4), (7 5), (8 5) cells:
+
+        python gen_pattern.py -o radon_checkerboard.svg --rows 10 --columns 15 --type radon_checkerboard -s 12.1 -m 7 4 7 5 8 5
+
 If you want to change unit use -u option (mm inches, px, m)

 If you want to change page size use -w and -h options
@@ -87,7 +87,7 @@ The tutorial consists of two main programs:

    The application starts up extracting the ORB features and descriptors from the input image and
    then uses the mesh along with the [Möller–Trumbore intersection
-    algorithm](http://http://en.wikipedia.org/wiki/M%C3%B6ller%E2%80%93Trumbore_intersection_algorithm/)
+    algorithm](http://en.wikipedia.org/wiki/M%C3%B6ller%E2%80%93Trumbore_intersection_algorithm/)
    to compute the 3D coordinates of the found features. Finally, the 3D points and the descriptors
    are stored in different lists in a file with YAML format which each row is a different point. The
    technical background on how to store the files can be found in the @ref tutorial_file_input_output_with_xml_yml
@@ -4,7 +4,7 @@ File Input and Output using XML and YAML files {#tutorial_file_input_output_with
@tableofcontents

@prev_tutorial{tutorial_discrete_fourier_transform}
-@next_tutorial{tutorial_how_to_use_OpenCV_parallel_for_}
+@next_tutorial{tutorial_how_to_use_OpenCV_parallel_for_new}

 |    |    |
 | -: | :- |
@@ -0,0 +1,166 @@
+How to use the OpenCV parallel_for_ to parallelize your code {#tutorial_how_to_use_OpenCV_parallel_for_new}
+==================================================================
+
+@tableofcontents
+
+@prev_tutorial{tutorial_file_input_output_with_xml_yml}
+@next_tutorial{tutorial_univ_intrin}
+
+|    |    |
+| -: | :- |
+| Compatibility | OpenCV >= 3.0 |
+
+Goal
+----
+
+The goal of this tutorial is to demonstrate the use of the OpenCV `parallel_for_` framework to easily parallelize your code. To illustrate the concept, we will write a program to perform convolution operation over an image.
+The full tutorial code is [here](https://github.com/opencv/opencv/tree/master/samples/cpp/tutorial_code/how_to_use_OpenCV_parallel_for_/how_to_use_OpenCV_parallel_for_new.cpp).
+
+Precondition
+----
+
+### Parallel Frameworks
+The first precondition is to have OpenCV built with a parallel framework.
+In OpenCV 4.5, the following parallel frameworks are available in that order:
+
+*   Intel Threading Building Blocks (3rdparty library, should be explicitly enabled)
+*   OpenMP (integrated to compiler, should be explicitly enabled)
+*   APPLE GCD (system wide, used automatically (APPLE only))
+*   Windows RT concurrency (system wide, used automatically (Windows RT only))
+*   Windows concurrency (part of runtime, used automatically (Windows only - MSVC++ >= 10))
+*   Pthreads
+
+As you can see, several parallel frameworks can be used in the OpenCV library. Some parallel libraries are third party libraries and have to be explicitly enabled in CMake before building, while others are automatically available with the platform (e.g. APPLE GCD).
+
+### Race Conditions
+Race conditions occur when more than one thread try to write *or* read and write to a particular memory location simultaneously.
+Based on that, we can broadly classify algorithms into two categories:-
+1. Algorithms in which only a single thread writes data to a particular memory location.
+    * In *convolution*, for example, even though multiple  threads may read from a pixel at a particular time, only a single thread *writes* to a particular pixel.
+
+2. Algorithms in which multiple threads may write to a single memory location.
+    * Finding contours, features, etc. Such algorithms may require each thread to add data to a global variable simultaneously. For example, when detecting features, each thread will add features of their respective parts of the image to a common vector, thus creating a race condition.
+
+Convolution
+-----------
+
+We will use the example of performing a convolution to demonstrate the use of `parallel_for_` to parallelize the computation. This is an example of an algorithm which does not lead to a race condition.
+
+Theory
+------
+Convolution is a simple mathematical operation widely used in image processing. Here, we slide a smaller matrix, called the *kernel*, over an image and a sum of the product of pixel values and corresponding values in the kernel gives us the value of the particular pixel in the output (called the anchor point of the kernel).  Based on the values in the kernel, we get different results.
+In the example below, we use a 3x3 kernel (anchored at its center) and convolve over a 5x5 matrix to produce a 3x3 matrix. The size of the output can be altered by padding the input with suitable values.
+![Convolution Animation](images/convolution-example-matrix.gif)
+
+For more information about different kernels and what they do, look [here](https://en.wikipedia.org/wiki/Kernel_(image_processing))
+
+For the purpose of this tutorial, we will implement the simplest form of the function which takes a grayscale image (1 channel) and an odd length square kernel and produces an output image.
+The operation will not be performed in-place.
+@note We can store a few of the relevant pixels temporarily to make sure we use the original values during the convolution and then do it in-place. However, the purpose of this tutorial is to introduce parallel_for_ function and an inplace implementation may be too complicated.
+
+Pseudocode
+-----------
+
+    InputImage src, OutputImage dst, kernel(size n)
+    makeborder(src, n/2)
+    for each pixel (i, j) strictly inside borders, do:
+    {
+        value := 0
+        for k := -n/2 to n/2, do:
+            for l := -n/2 to n/2, do:
+                value += kernel[n/2 + k][n/2 + l]*src[i + k][j + l]
+
+        dst[i][j] := value
+    }
+
+For an *n-sized kernel*, we will add a border of size *n/2* to handle edge cases.
+We then run two loops to move along the kernel and add the products to sum
+
+Implementation
+--------------
+
+### Sequential implementation
+
+@snippet how_to_use_OpenCV_parallel_for_new.cpp convolution-sequential
+
+We first make an output matrix(dst) with the same size as src and add borders to the src image(to handle edge cases).
+@snippet how_to_use_OpenCV_parallel_for_new.cpp convolution-make-borders
+
+We then sequentially iterate over the pixels in the src image and compute the value over the kernel and the neighbouring pixel values.
+We then fill value to the corresponding pixel in the dst image.
+@snippet how_to_use_OpenCV_parallel_for_new.cpp convolution-kernel-loop
+
+### Parallel implementation
+
+When looking at the sequential implementation, we can notice that each pixel depends on multiple neighbouring pixels but only one pixel is edited at a time. Thus, to optimize the computation, we can split the image into stripes and parallely perform convolution on each, by exploiting the multi-core architecture of modern processor. The OpenCV @ref cv::parallel_for_ framework automatically decides how to split the computation efficiently and does most of the work for us.
+
+@note Although values of a pixel in a particular stripe may depend on pixel values outside the stripe, these are only read only operations and hence will not cause undefined behaviour.
+
+
+We first declare a custom class that inherits from @ref cv::ParallelLoopBody and override the `virtual void operator ()(const cv::Range& range) const`.
+@snippet how_to_use_OpenCV_parallel_for_new.cpp convolution-parallel
+
+The range in the `operator ()` represents the subset of values that will be treated by an individual thread. Based on the requirement, there may be different ways of splitting the range which in turn changes the computation.
+
+For example, we can either
+1. Split the entire traversal of the image and obtain the [row, col] coordinate in the following way (as shown in the above code):
+
+    @snippet how_to_use_OpenCV_parallel_for_new.cpp overload-full
+
+    We would then call the parallel_for_ function in the following way:
+    @snippet how_to_use_OpenCV_parallel_for_new.cpp convolution-parallel-function
+<br>
+
+2. Split the rows and compute for each row:
+
+    @snippet how_to_use_OpenCV_parallel_for_new.cpp overload-row-split
+
+    In this case, we call the parallel_for_ function with a different range:
+    @snippet how_to_use_OpenCV_parallel_for_new.cpp convolution-parallel-function-row
+
+@note In our case, both implementations perform similarly. Some cases may allow better memory access patterns or other performance benefits.
+
+To set the number of threads, you can use: @ref cv::setNumThreads. You can also specify the number of splitting using the nstripes parameter in @ref cv::parallel_for_. For instance, if your processor has 4 threads, setting `cv::setNumThreads(2)` or setting `nstripes=2` should be the same as by default it will use all the processor threads available but will split the workload only on two threads.
+
+@note C++ 11 standard allows to simplify the parallel implementation by get rid of the `parallelConvolution` class and replacing it with lambda expression:
+
+@snippet how_to_use_OpenCV_parallel_for_new.cpp convolution-parallel-cxx11
+
+Results
+-----------
+
+The resulting time taken for execution of the two implementations on a
+* *512x512 input* with a *5x5 kernel*:
+
+        This program shows how to use the OpenCV parallel_for_ function and
+        compares the performance of the sequential and parallel implementations for a
+        convolution operation
+        Usage:
+        ./a.out [image_path -- default lena.jpg]
+
+        Sequential Implementation: 0.0953564s
+        Parallel Implementation: 0.0246762s
+        Parallel Implementation(Row Split): 0.0248722s
+
+<br>
+
+* *512x512 input with a 3x3 kernel*
+
+        This program shows how to use the OpenCV parallel_for_ function and
+        compares the performance of the sequential and parallel implementations for a
+        convolution operation
+        Usage:
+        ./a.out [image_path -- default lena.jpg]
+
+        Sequential Implementation: 0.0301325s
+        Parallel Implementation: 0.0117053s
+        Parallel Implementation(Row Split): 0.0117894s
+
+The performance of the parallel implementation depends on the type of CPU you have. For instance, on 4 cores - 8 threads CPU, runtime may be 6x to 7x faster than a sequential implementation. There are many factors to explain why we do not achieve a speed-up of 8x:
+    * the overhead to create and manage the threads,
+    * background processes running in parallel,
+    * the difference between 4 hardware cores with 2 logical threads for each core and 8 hardware cores.
+
+In the tutorial, we used a horizontal gradient filter(as shown in the animation above), which produces an image highlighting the vertical edges.
+
+![result image](images/resimg.jpg)
@@ -91,8 +91,8 @@ a new header with the new boundaries:
 Mat D (A, Rect(10, 10, 100, 100) ); // using a rectangle
 Mat E = A(Range::all(), Range(1,3)); // using row and column boundaries
@endcode
-Now you may ask -- if the matrix itself may belong to multiple *Mat* objects who takes responsibility
-for cleaning it up when it's no longer needed. The short answer is: the last object that used it.
+Now you may ask -- if the matrix itself may belong to multiple *Mat* objects, who takes responsibility
+for cleaning it up when it's no longer needed? The short answer is: the last object that used it.
 This is handled by using a reference counting mechanism. Whenever somebody copies a header of a
 *Mat* object, a counter is increased for the matrix. Whenever a header is cleaned, this counter
 is decreased. When the counter reaches zero the matrix is freed. Sometimes you will want to copy
@@ -102,12 +102,12 @@ Mat F = A.clone();
 Mat G;
 A.copyTo(G);
@endcode
-Now modifying *F* or *G* will not affect the matrix pointed by the *A*'s header. What you need to
+Now modifying *F* or *G* will not affect the matrix pointed to by the *A*'s header. What you need to
 remember from all this is that:

 -   Output image allocation for OpenCV functions is automatic (unless specified otherwise).
 -   You do not need to think about memory management with OpenCV's C++ interface.
-   The assignment operator and the copy constructor only copies the header.
+-   The assignment operator and the copy constructor only copy the header.
 -   The underlying matrix of an image may be copied using the @ref cv::Mat::clone() and @ref cv::Mat::copyTo()
    functions.

@@ -122,10 +122,10 @@ of these allows us to create many shades of gray.
 For *colorful* ways we have a lot more methods to choose from. Each of them breaks it down to three
 or four basic components and we can use the combination of these to create the others. The most
 popular one is RGB, mainly because this is also how our eye builds up colors. Its base colors are
-red, green and blue. To code the transparency of a color sometimes a fourth element: alpha (A) is
+red, green and blue. To code the transparency of a color sometimes a fourth element, alpha (A), is
 added.

-There are, however, many other color systems each with their own advantages:
+There are, however, many other color systems, each with their own advantages:

 -   RGB is the most common as our eyes use something similar, however keep in mind that OpenCV standard display
    system composes colors using the BGR color space (red and blue channels are swapped places).
@@ -139,11 +139,11 @@ There are, however, many other color systems each with their own advantages:
 Each of the building components has its own valid domains. This leads to the data type used. How
 we store a component defines the control we have over its domain. The smallest data type possible is
 *char*, which means one byte or 8 bits. This may be unsigned (so can store values from 0 to 255) or
-signed (values from -127 to +127). Although in case of three components this already gives 16
-million possible colors to represent (like in case of RGB) we may acquire an even finer control by
+signed (values from -127 to +127). Although this width, in the case of three components (like RGB), already gives 16
+million possible colors to represent, we may acquire an even finer control by
 using the float (4 byte = 32 bit) or double (8 byte = 64 bit) data types for each component.
 Nevertheless, remember that increasing the size of a component also increases the size of the whole
-picture in the memory.
+picture in memory.

 Creating a Mat object explicitly
 ----------------------------------
@@ -9,4 +9,5 @@ The Core Functionality (core module) {#tutorial_table_of_content_core}
 -   @subpage tutorial_basic_linear_transform
 -   @subpage tutorial_discrete_fourier_transform
 -   @subpage tutorial_file_input_output_with_xml_yml
-   @subpage tutorial_how_to_use_OpenCV_parallel_for_
+-   @subpage tutorial_how_to_use_OpenCV_parallel_for_new
+-   @subpage tutorial_univ_intrin
@@ -0,0 +1,334 @@
+Vectorizing your code using Universal Intrinsics {#tutorial_univ_intrin}
+==================================================================
+
+@tableofcontents
+
+@prev_tutorial{tutorial_how_to_use_OpenCV_parallel_for_new}
+
+|    |    |
+| -: | :- |
+| Compatibility | OpenCV >= 3.0 |
+
+Goal
+----
+
+The goal of this tutorial is to provide a guide to using the @ref core_hal_intrin feature to vectorize your C++ code for a faster runtime.
+We'll briefly look into _SIMD intrinsics_ and how to work with wide _registers_, followed by a tutorial on the basic operations using wide registers.
+
+Theory
+------
+
+In this section, we will briefly look into a few concepts to better help understand the functionality.
+
+### Intrinsics
+Intrinsics are functions which are separately handled by the compiler. These functions are often optimized to perform in the most efficient ways possible and hence run faster than normal implementations. However, since these functions depend on the compiler, it makes it difficult to write portable applications.
+
+### SIMD
+SIMD stands for **Single Instruction, Multiple Data**. SIMD Intrinsics allow the processor to vectorize calculations. The data is stored in what are known as *registers*. A *register* may be *128-bits*, *256-bits* or *512-bits* wide. Each *register* stores **multiple values** of the **same data type**. The size of the register and the size of each value determines the number of values stored in total.
+
+Depending on what *Instruction Sets* your CPU supports, you may be able to use the different registers. To learn more, look [here](https://en.wikipedia.org/wiki/Instruction_set_architecture)
+
+Universal Intrinsics
+--------------------
+
+OpenCVs universal intrinsics provides an abstraction to SIMD vectorization methods and allows the user to use intrinsics without the need to write system specific code.
+
+OpenCV Universal Intrinsics support the following instruction sets:
+* *128 bit* registers of various types support is implemented for a wide range of architectures including
+    * x86(SSE/SSE2/SSE4.2),
+    * ARM(NEON),
+    * PowerPC(VSX),
+    * MIPS(MSA).
+* *256 bit* registers are supported on x86(AVX2) and
+* *512 bit* registers are supported on x86(AVX512)
+
+**We will now introduce the available structures and functions:**
+* Register structures
+* Load and store
+* Mathematical Operations
+* Reduce and Mask
+
+### Register Structures
+
+The Universal Intrinsics set implements every register as a structure based on the particular SIMD register.
+All types contain the `nlanes` enumeration which gives the exact number of values that the type can hold. This eliminates the need to hardcode the number of values during implementations.
+
+@note Each register structure is under the `cv` namespace.
+
+There are **two types** of registers:
+
+* **Variable sized registers**: These structures do not have a fixed size and their exact bit length is deduced during compilation, based on the available SIMD capabilities. Consequently, the value of the `nlanes` enum is determined in compile time.
+    <br>
+
+    Each structure follows the following convention:
+
+        v_[type of value][size of each value in bits]
+
+    For instance, **v_uint8 holds 8-bit unsigned integers** and **v_float32 holds 32-bit floating point values**. We then declare a register like we would declare any object in C++
+
+    Based on the available SIMD instruction set, a particular register will hold different number of values.
+    For example: If your computer supports a maximum of 256bit registers,
+    * *v_uint8* will hold 32 8-bit unsigned integers
+    * *v_float64* will hold 4 64-bit floats (doubles)
+
+            v_uint8 a;                            // a is a register supporting uint8(char) data
+            int n = a.nlanes;                     // n holds 32
+
+    Available data type and sizes:
+    |Type|Size in bits|
+    |-:|:-|
+    |uint| 8, 16, 32, 64|
+    |int | 8, 16, 32, 64|
+    |float | 32, 64|
+
+* **Constant sized registers**: These structures have a fixed bit size and hold a constant number of values. We need to know what SIMD instruction set is supported by the system and select compatible registers. Use these only if exact bit length is necessary.
+    <br>
+
+    Each structure follows the convention:
+
+        v_[type of value][size of each value in bits]x[number of values]
+
+    Suppose we want to store
+    * 32-bit(*size in bits*) signed integers in a **128 bit register**. Since the register size is already known, we can find out the *number of data points in register* (*128/32 = 4*):
+
+            v_int32x8 reg1                       // holds 8 32-bit signed integers.
+
+    * 64-bit floats in 512 bit register:
+
+            v_float64x8 reg2                     // reg2.nlanes = 8
+
+### Load and Store operations
+
+Now that we know how registers work, let us look at the functions used for filling these registers with values.
+
+* **Load**: Load functions allow you to *load* values into a register.
+    * *Constructors* - When declaring a register structure, we can either provide a memory address from where the register will pick up contiguous values, or provide the values explicitly as multiple arguments (Explicit multiple arguments is available only for Constant Sized Registers):
+
+            float ptr[32] = {1, 2, 3 ..., 32};   // ptr is a pointer to a contiguous memory block of 32 floats
+
+            // Variable Sized Registers //
+            int x = v_float32().nlanes;          // set x as the number of values the register can hold
+
+            v_float32 reg1(ptr);                 // reg1 stores first x values according to the maximum register size available.
+            v_float32 reg2(ptr + x);             // reg stores the next x values
+
+            // Constant Sized Registers //
+            v_float32x4 reg1(ptr);               // reg1 stores the first 4 floats (1, 2, 3, 4)
+            v_float32x4 reg2(ptr + 4);           // reg2 stores the next 4 floats (5, 6, 7, 8)
+
+            // Or we can explicitly write down the values.
+            v_float32x4(1, 2, 3, 4);
+
+    <br>
+    * *Load Function* - We can use the load method and provide the memory address of the data:
+
+            float ptr[32] = {1, 2, 3, ..., 32};
+            v_float32 reg_var;
+            reg_var = vx_load(ptr);              // loads values from ptr[0] upto ptr[reg_var.nlanes - 1]
+
+            v_float32x4 reg_128;
+            reg_128 = v_load(ptr);               // loads values from ptr[0] upto ptr[3]
+
+            v_float32x8 reg_256;
+            reg_256 = v256_load(ptr);            // loads values from ptr[0] upto ptr[7]
+
+            v_float32x16 reg_512;
+            reg_512 = v512_load(ptr);            // loads values from ptr[0] upto ptr[15]
+
+        @note The load function assumes data is unaligned. If your data is aligned, you may use the `vx_load_aligned()` function.
+    <br>
+
+* **Store**: Store functions allow you to *store* the values from a register into a particular memory location.
+    * To store values from a register into a memory location, you may use the *v_store()* function:
+
+            float ptr[4];
+            v_store(ptr, reg); // store the first 128 bits(interpreted as 4x32-bit floats) of reg into ptr.
+<br>
+@note Ensure **ptr** has the same type as register. You can also cast the register into the proper type before carrying out operations. Simply typecasting the pointer to a particular type will lead wrong interpretation of data.
+
+### Binary and Unary Operators
+
+The universal intrinsics set provides element wise binary and unary operations.
+
+* **Arithmetics**: We can add, subtract, multiply and divide two registers element-wise. The registers must be of the same width and hold the same type. To multiply two registers, for example:
+
+        v_float32 a, b;                          // {a1, ..., an}, {b1, ..., bn}
+        v_float32 c;
+        c = a + b                                // {a1 + b1, ..., an + bn}
+        c = a * b;                               // {a1 * b1, ..., an * bn}
+
+<br>
+
+* **Bitwise Logic and Shifts**: We can left shift or right shift the bits of each element of the register. We can also apply bitwise &, |, ^ and ~ operators between two registers element-wise:
+
+        v_int32 as;                              // {a1, ..., an}
+        v_int32 al = as << 2;                    // {a1 << 2, ..., an << 2}
+        v_int32 bl = as >> 2;                    // {a1 >> 2, ..., an >> 2}
+
+        v_int32 a, b;
+        v_int32 a_and_b = a & b;                 // {a1 & b1, ..., an & bn}
+
+<br>
+
+* **Comparison Operators**: We can compare values between two registers using the <, >, <= , >=, == and != operators. Since each register contains multiple values, we don't get a single bool for these operations. Instead, for true values, all bits are converted to one (0xff for 8 bits, 0xffff for 16 bits, etc), while false values return bits converted to zero.
+
+        // let us consider the following code is run in a 128-bit register
+        v_uint8 a;                               // a = {0, 1, 2, ..., 15}
+        v_uint8 b;                               // b = {15, 14, 13, ..., 0}
+
+        v_uint8 c = a < b;
+
+        /*
+            let us look at the first 4 values in binary
+
+            a = |00000000|00000001|00000010|00000011|
+            b = |00001111|00001110|00001101|00001100|
+            c = |11111111|11111111|11111111|11111111|
+
+            If we store the values of c and print them as integers, we will get 255 for true values and 0 for false values.
+        */
+        ---
+        // In a computer supporting 256-bit registers
+        v_int32 a;                               // a = {1, 2, 3, 4, 5, 6, 7, 8}
+        v_int32 b;                               // b = {8, 7, 6, 5, 4, 3, 2, 1}
+
+        v_int32 c = (a < b);                     // c = {-1, -1, -1, -1, 0, 0, 0, 0}
+
+        /*
+            The true values are 0xffffffff, which in signed 32-bit integer representation is equal to -1.
+        */
+<br>
+
+* **Min/Max operations**: We can use the *v_min()* and *v_max()* functions to return registers containing element-wise min, or max, of the two registers:
+
+        v_int32 a;                               // {a1, ..., an}
+        v_int32 b;                               // {b1, ..., bn}
+
+        v_int32 mn = v_min(a, b);                // {min(a1, b1), ..., min(an, bn)}
+        v_int32 mx = v_max(a, b);                // {max(a1, b1), ..., max(an, bn)}
+<br>
+
+@note Comparison and Min/Max operators are not available for 64 bit integers. Bitwise shift and logic operators are available only for integer values. Bitwise shift is available only for 16, 32 and 64 bit registers.
+
+### Reduce and Mask
+
+* **Reduce Operations**: The *v_reduce_min()*, *v_reduce_max()* and *v_reduce_sum()* return a single value denoting the min, max or sum of the entire register:
+
+        v_int32 a;                                //  a = {a1, ..., a4}
+        int mn = v_reduce_min(a);                 // mn = min(a1, ..., an)
+        int sum = v_reduce_sum(a);                // sum = a1 + ... + an
+<br>
+
+* **Mask Operations**: Mask operations allow us to replicate conditionals in wide registers. These include:
+    * *v_check_all()* - Returns a bool, which is true if all the values in the register are less than zero.
+    * *v_check_any()* - Returns a bool, which is true if any value in the register is less than zero.
+    * *v_select()* - Returns a register, which blends two registers, based on a mask.
+
+            v_uint8 a;                           // {a1, .., an}
+            v_uint8 b;                           // {b1, ..., bn}
+
+            v_int32x4 mask:                      // {0xff, 0, 0, 0xff, ..., 0xff, 0}
+
+            v_uint8 Res = v_select(mask, a, b)   // {a1, b2, b3, a4, ..., an-1, bn}
+
+            /*
+                "Res" will contain the value from "a" if mask is true (all bits set to 1),
+                and value from "b" if mask is false (all bits set to 0)
+
+                We can use comparison operators to generate mask and v_select to obtain results based on conditionals.
+                It is common to set all values of b to 0. Thus, v_select will give values of "a" or 0 based on the mask.
+            */
+
+## Demonstration
+In the following section, we will vectorize a simple convolution function for single channel and compare the results to a scalar implementation.
+@note Not all algorithms are improved by manual vectorization. In fact, in certain cases, the compiler may *autovectorize* the code, thus producing faster results for scalar implementations.
+
+You may learn more about convolution from the previous tutorial. We use the same naive implementation from the previous tutorial and compare it to the vectorized version.
+
+The full tutorial code is [here](https://github.com/opencv/opencv/tree/master/samples/cpp/tutorial_code/univ_intrin/univ_intrin.cpp).
+
+### Vectorizing Convolution
+
+We will first implement a 1-D convolution and then vectorize it. The 2-D vectorized convolution will perform 1-D convolution across the rows to produce the correct results.
+
+#### 1-D Convolution: Scalar
+@snippet univ_intrin.cpp convolution-1D-scalar
+
+1. We first set up variables and make a border on both sides of the src matrix, to take care of edge cases.
+    @snippet univ_intrin.cpp convolution-1D-border
+
+2. For the main loop, we select an index *i* and offset it on both sides along with the kernel, using the k variable. We store the value in *value* and add it to the *dst* matrix.
+    @snippet univ_intrin.cpp convolution-1D-scalar-main
+
+#### 1-D Convolution: Vector
+
+We will now look at the vectorized version of 1-D convolution.
+@snippet univ_intrin.cpp convolution-1D-vector
+
+1. In our case, the kernel is a float. Since the kernel's datatype is the largest, we convert src to float32, forming *src_32*. We also make a border like we did for the naive case.
+    @snippet univ_intrin.cpp convolution-1D-convert
+
+2. Now, for each column in the *kernel*, we calculate the scalar product of the value with all *window* vectors of length `step`. We add these values to the already stored values in ans
+    @snippet univ_intrin.cpp convolution-1D-main
+
+    * We declare a pointer to the src_32 and kernel and run a loop for each kernel element
+        @snippet univ_intrin.cpp convolution-1D-main-h1
+
+    * We load a register with the current kernel element. A window is shifted from *0* to *len - step* and its product with the kernel_wide array is added to the values stored in *ans*. We store the values back into *ans*
+        @snippet univ_intrin.cpp convolution-1D-main-h2
+
+    * Since the length might not be divisible by steps, we take care of the remaining values directly. The number of *tail* values will always be less than *step* and will not affect the performance significantly. We store all the values to *ans* which is a float pointer. We can also directly store them in a `Mat` object
+        @snippet univ_intrin.cpp convolution-1D-main-h3
+
+    * Here is an iterative example:
+
+            For example:
+            kernel: {k1, k2, k3}
+            src:           ...|a1|a2|a3|a4|...
+
+
+            iter1:
+            for each idx i in (0, len), 'step' idx at a time
+                kernel_wide:          |k1|k1|k1|k1|
+                window:               |a0|a1|a2|a3|
+                ans:               ...| 0| 0| 0| 0|...
+                sum =  ans + window * kernel_wide
+                    =  |a0 * k1|a1 * k1|a2 * k1|a3 * k1|
+
+            iter2:
+                kernel_wide:          |k2|k2|k2|k2|
+                window:               |a1|a2|a3|a4|
+                ans:               ...|a0 * k1|a1 * k1|a2 * k1|a3 * k1|...
+                sum =  ans + window * kernel_wide
+                    =  |a0 * k1 + a1 * k2|a1 * k1 + a2 * k2|a2 * k1 + a3 * k2|a3 * k1 + a4 * k2|
+
+            iter3:
+                kernel_wide:          |k3|k3|k3|k3|
+                window:               |a2|a3|a4|a5|
+                ans:               ...|a0 * k1 + a1 * k2|a1 * k1 + a2 * k2|a2 * k1 + a3 * k2|a3 * k1 + a4 * k2|...
+                sum =  sum + window * kernel_wide
+                    =  |a0*k1 + a1*k2 + a2*k3|a1*k1 + a2*k2 + a3*k3|a2*k1 + a3*k2 + a4*k3|a3*k1 + a4*k2 + a5*k3|
+
+
+@note The function parameters also include *row*, *rowk* and *len*. These values are used when using the function as an intermediate step of 2-D convolution
+
+#### 2-D Convolution
+
+Suppose our kernel has *ksize* rows. To compute the values for a particular row, we compute the 1-D convolution of the previous *ksize/2* and the next *ksize/2* rows, with the corresponding kernel row. The final values is simply the sum of the individual 1-D convolutions
+@snippet univ_intrin.cpp convolution-2D
+
+1. We first initialize variables and make a border above and below the *src* matrix. The left and right sides are handled by the 1-D convolution function.
+    @snippet univ_intrin.cpp convolution-2D-init
+
+2. For each row, we calculate the 1-D convolution of the rows above and below it. we then add the values to the *dst* matrix.
+    @snippet univ_intrin.cpp convolution-2D-main
+
+3. We finally convert the *dst* matrix to a *8-bit* `unsigned char` matrix
+    @snippet univ_intrin.cpp convolution-2D-conv
+
+Results
+-------
+
+In the tutorial, we used a horizontal gradient kernel. We obtain the same output image for both methods.
+
+Improvement in runtime varies and will depend on the SIMD capabilities available in your CPU.
@@ -0,0 +1,95 @@
+# DNN-based Face Detection And Recognition {#tutorial_dnn_face}
+
+@tableofcontents
+
+@prev_tutorial{tutorial_dnn_text_spotting}
+@next_tutorial{pytorch_cls_tutorial_dnn_conversion}
+
+| | |
+| -: | :- |
+| Original Author | Chengrui Wang, Yuantao Feng |
+| Compatibility | OpenCV >= 4.5.1 |
+
+## Introduction
+
+In this section, we introduce the DNN-based module for face detection and face recognition. Models can be obtained in [Models](#Models). The usage of `FaceDetectorYN` and `FaceRecognizer` are presented in [Usage](#Usage).
+
+## Models
+
+There are two models (ONNX format) pre-trained and required for this module:
+- [Face Detection](https://github.com/ShiqiYu/libfacedetection.train/tree/master/tasks/task1/onnx):
+    - Size: 337KB
+    - Results on WIDER Face Val set: 0.830(easy), 0.824(medium), 0.708(hard)
+- [Face Recognition](https://drive.google.com/file/d/1ClK9WiB492c5OZFKveF3XiHCejoOxINW/view?usp=sharing)
+    - Size: 36.9MB
+    - Results:
+
+    | Database | Accuracy | Threshold (normL2) | Threshold (cosine) |
+    | -------- | -------- | ------------------ | ------------------ |
+    | LFW      | 99.60%   | 1.128              | 0.363              |
+    | CALFW    | 93.95%   | 1.149              | 0.340              |
+    | CPLFW    | 91.05%   | 1.204              | 0.275              |
+    | AgeDB-30 | 94.90%   | 1.202              | 0.277              |
+    | CFP-FP   | 94.80%   | 1.253              | 0.212              |
+
+## Usage
+
+### DNNFaceDetector
+
+```cpp
+// Initialize FaceDetectorYN
+Ptr<FaceDetectorYN> faceDetector = FaceDetectorYN::create(onnx_path, "", image.size(), score_thresh, nms_thresh, top_k);
+
+// Forward
+Mat faces;
+faceDetector->detect(image, faces);
+```
+
+The detection output `faces` is a two-dimension array of type CV_32F, whose rows are the detected face instances, columns are the location of a face and 5 facial landmarks. The format of each row is as follows:
+
+```
+x1, y1, w, h, x_re, y_re, x_le, y_le, x_nt, y_nt, x_rcm, y_rcm, x_lcm, y_lcm
+```
+, where `x1, y1, w, h` are the top-left coordinates, width and height of the face bounding box, `{x, y}_{re, le, nt, rcm, lcm}` stands for the coordinates of right eye, left eye, nose tip, the right corner and left corner of the mouth respectively.
+
+
+### Face Recognition
+
+Following Face Detection, run codes below to extract face feature from facial image.
+
+```cpp
+// Initialize FaceRecognizer with model path (cv::String)
+Ptr<FaceRecognizer> faceRecognizer = FaceRecognizer::create(model_path, "");
+
+// Aligning and cropping facial image through the first face of faces detected by dnn_face::DNNFaceDetector
+Mat aligned_face;
+faceRecognizer->alignCrop(image, faces.row(0), aligned_face);
+
+// Run feature extraction with given aligned_face (cv::Mat)
+Mat feature;
+faceRecognizer->feature(aligned_face, feature);
+feature = feature.clone();
+```
+
+After obtaining face features *feature1* and *feature2* of two facial images, run codes below to calculate the identity discrepancy between the two faces.
+
+```cpp
+// Calculating the discrepancy between two face features by using cosine distance.
+double cos_score = faceRecognizer->match(feature1, feature2, FaceRecognizer::DisType::COSINE);
+// Calculating the discrepancy between two face features by using normL2 distance.
+double L2_score = faceRecognizer->match(feature1, feature2, FaceRecognizer::DisType::NORM_L2);
+```
+
+For example, two faces have same identity if the cosine distance is greater than or equal to 0.363, or the normL2 distance is less than or equal to 1.128.
+
+## Reference:
+
+- https://github.com/ShiqiYu/libfacedetection
+- https://github.com/ShiqiYu/libfacedetection.train
+- https://github.com/zhongyy/SFace
+
+## Acknowledgement
+
+Thanks [Professor Shiqi Yu](https://github.com/ShiqiYu/) and [Yuantao Feng](https://github.com/fengyuentau) for training and providing the face detection model.
+
+Thanks [Professor Deng](http://www.whdeng.cn/), [PhD Candidate Zhong](https://github.com/zhongyy/) and [Master Candidate Wang](https://github.com/crywang/) for training and providing the face recognition model.
@@ -3,7 +3,7 @@
@tableofcontents

@prev_tutorial{tutorial_dnn_OCR}
-@next_tutorial{pytorch_cls_tutorial_dnn_conversion}
+@next_tutorial{tutorial_dnn_face}

 |    |    |
 | -: | :- |
@@ -26,6 +26,11 @@ Before recognition, you should `setVocabulary` and `setDecodeType`.
    - `T` is the sequence length
    - `B` is the batch size (only support `B=1` in inference)
    - and `Dim` is the length of vocabulary +1('Blank' of CTC is at the index=0 of Dim).
+- "CTC-prefix-beam-search", the output of the text recognition model should be a probability matrix same with "CTC-greedy".
+    - The algorithm is proposed at Hannun's [paper](https://arxiv.org/abs/1408.2873).
+    - `setDecodeOptsCTCPrefixBeamSearch` could be used to control the beam size in search step.
+    - To futher optimize for big vocabulary, a new option `vocPruneSize` is introduced to avoid iterate the whole vocbulary
+       but only the number of `vocPruneSize` tokens with top probabilty.

@ref cv::dnn::TextRecognitionModel::recognize() is the main function for text recognition.
 - The input image should be a cropped text image or an image with `roiRects`
@@ -10,6 +10,7 @@ Deep Neural Networks (dnn module) {#tutorial_table_of_content_dnn}
 -   @subpage tutorial_dnn_custom_layers
 -   @subpage tutorial_dnn_OCR
 -   @subpage tutorial_dnn_text_spotting
+-   @subpage tutorial_dnn_face

 #### PyTorch models with OpenCV
 In this section you will find the guides, which describe how to run classification, segmentation and detection PyTorch DNN models with OpenCV.
--- a/Show More
+++ b/Show More