release: OpenCV 3.4.15

backport of commit: e5841d3126

3rdparty/ffmpeg/ffmpeg.cmake

@@ -1,8 +1,8 @@
-# Binaries branch name: ffmpeg/3.4_20200907
-# Binaries were created for OpenCV: 03bee14372f5537daa56c62e771ec16181ca1f98
-ocv_update(FFMPEG_BINARIES_COMMIT "2a96257b743695a47f8012aab1ffb995a1dee8b4")
-ocv_update(FFMPEG_FILE_HASH_BIN32 "5e68a3ff82f43ac6524e50e448a34c9c")
-ocv_update(FFMPEG_FILE_HASH_BIN64 "205db629d893e7d4865fd1459807ff47")
+# Binaries branch name: ffmpeg/3.4_20210302
+# Binaries were created for OpenCV: 2ab1f3f166fccc3a01497209cc01c5cea44ff201
+ocv_update(FFMPEG_BINARIES_COMMIT "e99214251d9f3cde7c48abd46b2259bddc9885b6")
+ocv_update(FFMPEG_FILE_HASH_BIN32 "fad5ada9be36120bba8966709e7953a8")
+ocv_update(FFMPEG_FILE_HASH_BIN64 "650e2272728491923e566f784f79cfef")
 ocv_update(FFMPEG_FILE_HASH_CMAKE "3b90f67f4b429e77d3da36698cef700c")
 
 function(download_win_ffmpeg script_var)

3rdparty/libjpeg-turbo/CMakeLists.txt

@@ -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()
 
 

3rdparty/libjpeg-turbo/LICENSE.md

@@ -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

3rdparty/libjpeg-turbo/README.ijg

@@ -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
 ==========
 

3rdparty/libjpeg-turbo/README.md

@@ -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

3rdparty/libjpeg-turbo/jconfig.h.in

@@ -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 */
 

3rdparty/libjpeg-turbo/jconfig.h.win.in

@@ -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 */

3rdparty/libjpeg-turbo/src/jccolext.c

@@ -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;
     }
   }

3rdparty/libjpeg-turbo/src/jccolor.c

@@ -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;
         }
       }

3rdparty/libjpeg-turbo/src/jcdctmgr.c

@@ -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
   }

3rdparty/libjpeg-turbo/src/jchuff.c

@@ -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;
 }
 
 

3rdparty/libjpeg-turbo/src/jcphuff.c

@@ -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;

3rdparty/libjpeg-turbo/src/jcsample.c

@@ -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);

3rdparty/libjpeg-turbo/src/jdapistd.c

@@ -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);
       }
     }

3rdparty/libjpeg-turbo/src/jdarith.c

@@ -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++)

3rdparty/libjpeg-turbo/src/jdcoefct.c

@@ -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,

3rdparty/libjpeg-turbo/src/jdcoefct.h

@@ -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;
 

3rdparty/libjpeg-turbo/src/jdcol565.c

@@ -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;
     }

3rdparty/libjpeg-turbo/src/jdcolext.c

@@ -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];

3rdparty/libjpeg-turbo/src/jdcolor.c

@@ -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;
     }
   }

3rdparty/libjpeg-turbo/src/jdhuff.c

@@ -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;
 }

3rdparty/libjpeg-turbo/src/jdhuff.h

@@ -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 */

3rdparty/libjpeg-turbo/src/jdicc.c

@@ -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];

3rdparty/libjpeg-turbo/src/jdmarker.c

@@ -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;

3rdparty/libjpeg-turbo/src/jdmaster.c

@@ -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

3rdparty/libjpeg-turbo/src/jdmrg565.c

@@ -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)];

3rdparty/libjpeg-turbo/src/jdmrgext.c

@@ -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];

3rdparty/libjpeg-turbo/src/jdphuff.c

@@ -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++)

3rdparty/libjpeg-turbo/src/jdsample.c

@@ -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) {

3rdparty/libjpeg-turbo/src/jerror.h

@@ -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), \

3rdparty/libjpeg-turbo/src/jidctint.c

@@ -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] */

3rdparty/libjpeg-turbo/src/jmorecfg.h

@@ -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. */
 

3rdparty/libjpeg-turbo/src/jpegint.h

@@ -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 */

3rdparty/libjpeg-turbo/src/jquant1.c

@@ -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.

3rdparty/libjpeg-turbo/src/jquant2.c

@@ -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

3rdparty/libjpeg-turbo/src/jsimd.h

@@ -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);

3rdparty/libjpeg-turbo/src/jsimd_none.c

@@ -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)
 {

3rdparty/libjpeg-turbo/src/jversion.h

@@ -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"

CMakeLists.txt

@@ -1000,6 +1000,12 @@ if(COMMAND ocv_pylint_finalize)
   ocv_pylint_add_directory_recurse(${CMAKE_CURRENT_LIST_DIR}/samples/python/tutorial_code)
   ocv_pylint_finalize()
 endif()
+if(TARGET check_pylint)
+  message(STATUS "Registered 'check_pylint' target: using ${PYLINT_EXECUTABLE} (ver: ${PYLINT_VERSION}), checks: ${PYLINT_TOTAL_TARGETS}")
+endif()
+if(TARGET check_flake8)
+  message(STATUS "Registered 'check_flake8' target: using ${FLAKE8_EXECUTABLE} (ver: ${FLAKE8_VERSION})")
+endif()
 
 if(OPENCV_GENERATE_SETUPVARS)
   include(cmake/OpenCVGenSetupVars.cmake)
@@ -1633,12 +1639,6 @@ endif()
 
 status("")
 status("  Python (for build):"  PYTHON_DEFAULT_AVAILABLE THEN "${PYTHON_DEFAULT_EXECUTABLE}" ELSE NO)
-if(PYLINT_FOUND AND PYLINT_EXECUTABLE)
-  status("    Pylint:"  PYLINT_FOUND THEN "${PYLINT_EXECUTABLE} (ver: ${PYLINT_VERSION}, checks: ${PYLINT_TOTAL_TARGETS})" ELSE NO)
-endif()
-if(FLAKE8_FOUND AND FLAKE8_EXECUTABLE)
-  status("    Flake8:"  FLAKE8_FOUND THEN "${FLAKE8_EXECUTABLE} (ver: ${FLAKE8_VERSION})" ELSE NO)
-endif()
 
 # ========================== java ==========================
 if(BUILD_JAVA)

cmake/FindPylint.cmake

@@ -16,7 +16,7 @@ if(PYLINT_EXECUTABLE AND NOT DEFINED PYLINT_VERSION)
   execute_process(COMMAND ${PYLINT_EXECUTABLE} --version RESULT_VARIABLE _result OUTPUT_VARIABLE PYLINT_VERSION_RAW)
   if(NOT _result EQUAL 0)
     ocv_clear_vars(PYLINT_EXECUTABLE PYLINT_VERSION)
-  elseif(PYLINT_VERSION_RAW MATCHES "pylint([^,]*) ([0-9\\.]+[0-9])")
+  elseif(PYLINT_VERSION_RAW MATCHES "pylint([^,\n]*) ([0-9\\.]+[0-9])")
     set(PYLINT_VERSION "${CMAKE_MATCH_2}")
   else()
     set(PYLINT_VERSION "unknown")

cmake/OpenCVCompilerOptions.cmake

@@ -177,7 +177,13 @@ if(CV_GCC OR CV_CLANG)
   endif()
 
   # We need pthread's
-  if(UNIX AND NOT ANDROID AND NOT (APPLE AND CV_CLANG)) # TODO
+  if((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
+  )
     add_extra_compiler_option(-pthread)
   endif()
 
@@ -225,9 +231,11 @@ 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()
@@ -279,6 +287,7 @@ 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
@@ -346,6 +355,7 @@ 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} ${_option}")
     endif()
   endif()
 endif()
@@ -364,6 +374,9 @@ if(NOT OPENCV_SKIP_EXTRA_COMPILER_FLAGS)
   set(CMAKE_SHARED_LINKER_FLAGS         "${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)

cmake/OpenCVDetectHalide.cmake

@@ -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()

cmake/OpenCVDetectInferenceEngine.cmake

@@ -140,12 +140,21 @@ endif()
 # Add more features to the target
 
 if(INF_ENGINE_TARGET)
-  if(NOT INF_ENGINE_RELEASE)
-    message(WARNING "InferenceEngine version has not been set, 2021.2 will be used by default. Set INF_ENGINE_RELEASE variable if you experience build errors.")
+  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 + ${InferenceEngine_VERSION_PATCH} * 100")
+    endif()
   endif()
-  set(INF_ENGINE_RELEASE "2021020000" CACHE STRING "Force IE version, should be in form YYYYAABBCC (e.g. 2020.1.0.2 -> 2020010002)")
+  if(NOT INF_ENGINE_RELEASE AND NOT INF_ENGINE_RELEASE_INIT)
+    message(WARNING "InferenceEngine version has not been set, 2021.4 will be used by default. Set INF_ENGINE_RELEASE variable if you experience build errors.")
+    set(INF_ENGINE_RELEASE_INIT "2021040000")
+  elseif(DEFINED INF_ENGINE_RELEASE)
+    set(INF_ENGINE_RELEASE_INIT "${INF_ENGINE_RELEASE}")
+  endif()
+  set(INF_ENGINE_RELEASE "${INF_ENGINE_RELEASE_INIT}" CACHE STRING "Force IE version, should be in form YYYYAABBCC (e.g. 2020.1.0.2 -> 2020010002)")
   set_target_properties(${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()
 

cmake/OpenCVDownload.cmake

@@ -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

cmake/OpenCVFindIPP.cmake

@@ -143,10 +143,25 @@ macro(ipp_detect_version)
         list(APPEND IPP_LIBRARIES ${IPP_LIBRARY_DIR}/${IPP_LIB_PREFIX}${IPP_PREFIX}${name}${IPP_SUFFIX}${IPP_LIB_SUFFIX})
       else ()
         add_library(ipp${name} STATIC IMPORTED)
+        set(_filename "${IPP_LIB_PREFIX}${IPP_PREFIX}${name}${IPP_SUFFIX}${IPP_LIB_SUFFIX}")
         set_target_properties(ipp${name} PROPERTIES
           IMPORTED_LINK_INTERFACE_LIBRARIES ""
-          IMPORTED_LOCATION ${IPP_LIBRARY_DIR}/${IPP_LIB_PREFIX}${IPP_PREFIX}${name}${IPP_SUFFIX}${IPP_LIB_SUFFIX}
+          IMPORTED_LOCATION ${IPP_LIBRARY_DIR}/${_filename}
         )
+        if("${name}" STREQUAL "core")  # https://github.com/opencv/opencv/pull/19681
+          if(OPENCV_FORCE_IPP_EXCLUDE_LIBS OR OPENCV_FORCE_IPP_EXCLUDE_LIBS_CORE
+              OR (UNIX AND NOT ANDROID AND NOT APPLE
+                  AND CMAKE_CXX_COMPILER_ID MATCHES "GNU|Clang|Intel"
+              )
+              AND NOT OPENCV_SKIP_IPP_EXCLUDE_LIBS_CORE
+          )
+            if(CMAKE_VERSION VERSION_LESS "3.13.0")
+              set(CMAKE_SHARED_LINKER_FLAGS "-Wl,--exclude-libs,${_filename} ${CMAKE_SHARED_LINKER_FLAGS}")
+            else()
+              target_link_options(ipp${name} INTERFACE "LINKER:--exclude-libs,${_filename}")
+            endif()
+          endif()
+        endif()
         list(APPEND IPP_LIBRARIES ipp${name})
         if (NOT BUILD_SHARED_LIBS AND (HAVE_IPP_ICV OR ";${OPENCV_INSTALL_EXTERNAL_DEPENDENCIES};" MATCHES ";ipp;"))
           # CMake doesn't support "install(TARGETS ${IPP_PREFIX}${name} " command with imported targets

cmake/OpenCVFindLibsPerf.cmake

@@ -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
     )

cmake/OpenCVModule.cmake

@@ -876,7 +876,11 @@ endmacro()
 macro(_ocv_create_module)
 
   ocv_compiler_optimization_process_sources(OPENCV_MODULE_${the_module}_SOURCES OPENCV_MODULE_${the_module}_DEPS_EXT ${the_module})
-  set(OPENCV_MODULE_${the_module}_HEADERS ${OPENCV_MODULE_${the_module}_HEADERS} CACHE INTERNAL "List of header files for ${the_module}")
+  set(__module_headers ${OPENCV_MODULE_${the_module}_HEADERS})
+  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}")
 
   # The condition we ought to be testing here is whether ocv_add_precompiled_headers will

cmake/OpenCVPylint.cmake

@@ -122,7 +122,6 @@ function(ocv_pylint_finalize)
 
   list(LENGTH PYLINT_TARGET_ID __total)
   set(PYLINT_TOTAL_TARGETS "${__total}" CACHE INTERNAL "")
-  message(STATUS "Pylint: registered ${__total} targets. Build 'check_pylint' target to run checks (\"cmake --build . --target check_pylint\" or \"make check_pylint\")")
   configure_file("${OpenCV_SOURCE_DIR}/cmake/templates/pylint.cmake.in" "${CMAKE_BINARY_DIR}/pylint.cmake" @ONLY)
 
   add_custom_target(check_pylint

doc/Doxyfile.in

@@ -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"
@@ -249,6 +249,12 @@ PREDEFINED             = __cplusplus=1 \
                          CV_DEFAULT(x)=" = x" \
                          CV_NEON=1 \
                          CV_SSE2=1 \
+                         CV_SIMD128=1 \
+                         CV_SIMD256=1 \
+                         CV_SIMD512=1 \
+                         CV_SIMD128_64F=1 \
+                         CV_SIMD256_64F=1 \
+                         CV_SIMD512_64F=1 \
                          CV__DEBUG_NS_BEGIN= \
                          CV__DEBUG_NS_END= \
                          CV_DEPRECATED_EXTERNAL= \

doc/js_tutorials/js_setup/js_setup/js_setup.markdown

@@ -138,6 +138,7 @@ Building OpenCV.js from Source
     python ./platforms/js/build_js.py build_js --cmake_option="-DOPENCV_EXTRA_MODULES_PATH=opencv_contrib/modules"
     @endcode
 
+
 Running OpenCV.js Tests
 ---------------------------------------
 
@@ -303,6 +304,12 @@ The example uses latest version of emscripten. If the build fails you should try
 docker run --rm -v $(pwd):/src -u $(id -u):$(id -g) emscripten/emsdk:2.0.10 emcmake python3 ./platforms/js/build_js.py build_js
 @endcode
 
+In Windows use the following PowerShell command:
+
+@code{.bash}
+docker run --rm --workdir /src -v "$(get-location):/src" "emscripten/emsdk:2.0.10" emcmake python3 ./platforms/js/build_js.py build_js
+@endcode
+
 ### Building the documentation with Docker
 
 To build the documentation `doxygen` needs to be installed. Create a file named `Dockerfile` with the following content:

doc/opencv.bib

@@ -835,12 +835,12 @@
   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},
@@ -1028,12 +1028,12 @@
   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},

doc/py_tutorials/py_feature2d/py_features_harris/py_features_harris.markdown

@@ -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.
 

doc/py_tutorials/py_feature2d/py_shi_tomasi/py_shi_tomasi.markdown

@@ -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
 ----

doc/py_tutorials/py_feature2d/py_sift_intro/py_sift_intro.markdown

@@ -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.

doc/tools/html_functions.py

@@ -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)
 
 

doc/tutorials/calib3d/real_time_pose/real_time_pose.markdown

@@ -80,7 +80,7 @@ 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

doc/tutorials/introduction/cross_referencing/tutorial_cross_referencing.markdown

@@ -39,14 +39,14 @@ Open your Doxyfile using your favorite text editor and search for the key
 `TAGFILES`. Change it as follows:
 
 @code
-TAGFILES = ./docs/doxygen-tags/opencv.tag=http://docs.opencv.org/3.4.14
+TAGFILES = ./docs/doxygen-tags/opencv.tag=http://docs.opencv.org/3.4.15
 @endcode
 
 If you had other definitions already, you can append the line using a `\`:
 
 @code
 TAGFILES = ./docs/doxygen-tags/libstdc++.tag=https://gcc.gnu.org/onlinedocs/libstdc++/latest-doxygen \
-           ./docs/doxygen-tags/opencv.tag=http://docs.opencv.org/3.4.14
+           ./docs/doxygen-tags/opencv.tag=http://docs.opencv.org/3.4.15
 @endcode
 
 Doxygen can now use the information from the tag file to link to the OpenCV

doc/tutorials/videoio/video-write/video_write.markdown

@@ -59,7 +59,7 @@ extension, its first version. A direct limitation of this is that you cannot sav
 larger than 2 GB. Furthermore you can only create and expand a single video track inside the
 container. No audio or other track editing support here. Nevertheless, any video codec present on
 your system might work. If you encounter some of these limitations you will need to look into more
-specialized video writing libraries such as *FFMpeg* or codecs as *HuffYUV*, *CorePNG* and *LCL*. As
+specialized video writing libraries such as *FFmpeg* or codecs as *HuffYUV*, *CorePNG* and *LCL*. As
 an alternative, create the video track with OpenCV and expand it with sound tracks or convert it to
 other formats by using video manipulation programs such as *VirtualDub* or *AviSynth*.
 

modules/calib3d/include/opencv2/calib3d.hpp

@@ -1510,8 +1510,8 @@ concatenated together.
 @param imageSize Size of the image used only to initialize the camera intrinsic matrix.
 @param cameraMatrix Input/output 3x3 floating-point camera intrinsic matrix
 \f$\cameramatrix{A}\f$ . If @ref CALIB_USE_INTRINSIC_GUESS
-and/or @ref CALIB_FIX_ASPECT_RATIO are specified, some or all of fx, fy, cx, cy must be
-initialized before calling the function.
+and/or @ref CALIB_FIX_ASPECT_RATIO, @ref CALIB_FIX_PRINCIPAL_POINT or @ref CALIB_FIX_FOCAL_LENGTH
+are specified, some or all of fx, fy, cx, cy must be initialized before calling the function.
 @param distCoeffs Input/output vector of distortion coefficients
 \f$\distcoeffs\f$.
 @param rvecs Output vector of rotation vectors (@ref Rodrigues ) estimated for each pattern view
@@ -1537,7 +1537,7 @@ the number of pattern views. \f$R_i, T_i\f$ are concatenated 1x3 vectors.
 fx, fy, cx, cy that are optimized further. Otherwise, (cx, cy) is initially set to the image
 center ( imageSize is used), and focal distances are computed in a least-squares fashion.
 Note, that if intrinsic parameters are known, there is no need to use this function just to
-estimate extrinsic parameters. Use solvePnP instead.
+estimate extrinsic parameters. Use @ref solvePnP instead.
 -   @ref CALIB_FIX_PRINCIPAL_POINT The principal point is not changed during the global
 optimization. It stays at the center or at a different location specified when
  @ref CALIB_USE_INTRINSIC_GUESS is set too.
@@ -1547,24 +1547,23 @@ ratio fx/fy stays the same as in the input cameraMatrix . When
 ignored, only their ratio is computed and used further.
 -   @ref CALIB_ZERO_TANGENT_DIST Tangential distortion coefficients \f$(p_1, p_2)\f$ are set
 to zeros and stay zero.
+-   @ref CALIB_FIX_FOCAL_LENGTH The focal length is not changed during the global optimization if
+ @ref CALIB_USE_INTRINSIC_GUESS is set.
 -   @ref CALIB_FIX_K1,..., @ref CALIB_FIX_K6 The corresponding radial distortion
 coefficient is not changed during the optimization. If @ref CALIB_USE_INTRINSIC_GUESS is
 set, the coefficient from the supplied distCoeffs matrix is used. Otherwise, it is set to 0.
 -   @ref CALIB_RATIONAL_MODEL Coefficients k4, k5, and k6 are enabled. To provide the
 backward compatibility, this extra flag should be explicitly specified to make the
-calibration function use the rational model and return 8 coefficients. If the flag is not
-set, the function computes and returns only 5 distortion coefficients.
+calibration function use the rational model and return 8 coefficients or more.
 -   @ref CALIB_THIN_PRISM_MODEL Coefficients s1, s2, s3 and s4 are enabled. To provide the
 backward compatibility, this extra flag should be explicitly specified to make the
-calibration function use the thin prism model and return 12 coefficients. If the flag is not
-set, the function computes and returns only 5 distortion coefficients.
+calibration function use the thin prism model and return 12 coefficients or more.
 -   @ref CALIB_FIX_S1_S2_S3_S4 The thin prism distortion coefficients are not changed during
 the optimization. If @ref CALIB_USE_INTRINSIC_GUESS is set, the coefficient from the
 supplied distCoeffs matrix is used. Otherwise, it is set to 0.
 -   @ref CALIB_TILTED_MODEL Coefficients tauX and tauY are enabled. To provide the
 backward compatibility, this extra flag should be explicitly specified to make the
-calibration function use the tilted sensor model and return 14 coefficients. If the flag is not
-set, the function computes and returns only 5 distortion coefficients.
+calibration function use the tilted sensor model and return 14 coefficients.
 -   @ref CALIB_FIX_TAUX_TAUY The coefficients of the tilted sensor model are not changed during
 the optimization. If @ref CALIB_USE_INTRINSIC_GUESS is set, the coefficient from the
 supplied distCoeffs matrix is used. Otherwise, it is set to 0.
@@ -1589,12 +1588,12 @@ The algorithm performs the following steps:
     zeros initially unless some of CALIB_FIX_K? are specified.
 
 -   Estimate the initial camera pose as if the intrinsic parameters have been already known. This is
-    done using solvePnP .
+    done using @ref solvePnP .
 
 -   Run the global Levenberg-Marquardt optimization algorithm to minimize the reprojection error,
     that is, the total sum of squared distances between the observed feature points imagePoints and
     the projected (using the current estimates for camera parameters and the poses) object points
-    objectPoints. See projectPoints for details.
+    objectPoints. See @ref projectPoints for details.
 
 @note
     If you use a non-square (i.e. non-N-by-N) grid and @ref findChessboardCorners for calibration,
@@ -2241,6 +2240,7 @@ final fundamental matrix. It can be set to something like 1-3, depending on the
 point localization, image resolution, and the image noise.
 @param mask Output array of N elements, every element of which is set to 0 for outliers and to 1
 for the other points. The array is computed only in the RANSAC and LMedS methods.
+@param maxIters The maximum number of robust method iterations.
 
 This function estimates essential matrix based on the five-point algorithm solver in @cite Nister03 .
 @cite SteweniusCFS is also a related. The epipolar geometry is described by the following equation:
@@ -2251,6 +2251,12 @@ where \f$E\f$ is an essential matrix, \f$p_1\f$ and \f$p_2\f$ are corresponding
 second images, respectively. The result of this function may be passed further to
 decomposeEssentialMat or recoverPose to recover the relative pose between cameras.
  */
+CV_EXPORTS_W Mat findEssentialMat( InputArray points1, InputArray points2,
+                                 InputArray cameraMatrix, int method,
+                                 double prob, double threshold,
+                                 int maxIters, OutputArray mask = noArray() );
+
+/** @overload */
 CV_EXPORTS_W Mat findEssentialMat( InputArray points1, InputArray points2,
                                  InputArray cameraMatrix, int method = RANSAC,
                                  double prob = 0.999, double threshold = 1.0,
@@ -2274,6 +2280,7 @@ point localization, image resolution, and the image noise.
 confidence (probability) that the estimated matrix is correct.
 @param mask Output array of N elements, every element of which is set to 0 for outliers and to 1
 for the other points. The array is computed only in the RANSAC and LMedS methods.
+@param maxIters The maximum number of robust method iterations.
 
 This function differs from the one above that it computes camera intrinsic matrix from focal length and
 principal point:
@@ -2285,6 +2292,13 @@ f & 0 & x_{pp}  \\
 0 & 0 & 1
 \end{bmatrix}\f]
  */
+CV_EXPORTS_W Mat findEssentialMat( InputArray points1, InputArray points2,
+                                 double focal, Point2d pp,
+                                 int method, double prob,
+                                 double threshold, int maxIters,
+                                 OutputArray mask = noArray() );
+
+/** @overload */
 CV_EXPORTS_W Mat findEssentialMat( InputArray points1, InputArray points2,
                                  double focal = 1.0, Point2d pp = Point2d(0, 0),
                                  int method = RANSAC, double prob = 0.999,

modules/calib3d/src/five-point.cpp

@@ -403,7 +403,8 @@ protected:
 
 // Input should be a vector of n 2D points or a Nx2 matrix
 cv::Mat cv::findEssentialMat( InputArray _points1, InputArray _points2, InputArray _cameraMatrix,
-                              int method, double prob, double threshold, OutputArray _mask)
+                              int method, double prob, double threshold,
+                              int maxIters, OutputArray _mask)
 {
     CV_INSTRUMENT_REGION();
 
@@ -442,20 +443,36 @@ cv::Mat cv::findEssentialMat( InputArray _points1, InputArray _points2, InputArr
 
     Mat E;
     if( method == RANSAC )
-        createRANSACPointSetRegistrator(makePtr<EMEstimatorCallback>(), 5, threshold, prob)->run(points1, points2, E, _mask);
+        createRANSACPointSetRegistrator(makePtr<EMEstimatorCallback>(), 5, threshold, prob, maxIters)->run(points1, points2, E, _mask);
     else
-        createLMeDSPointSetRegistrator(makePtr<EMEstimatorCallback>(), 5, prob)->run(points1, points2, E, _mask);
+        createLMeDSPointSetRegistrator(makePtr<EMEstimatorCallback>(), 5, prob, maxIters)->run(points1, points2, E, _mask);
 
     return E;
 }
 
+cv::Mat cv::findEssentialMat( InputArray _points1, InputArray _points2, InputArray _cameraMatrix,
+                              int method, double prob, double threshold,
+                              OutputArray _mask)
+{
+    return cv::findEssentialMat(_points1, _points2, _cameraMatrix, method, prob, threshold, 1000, _mask);
+}
+
+cv::Mat cv::findEssentialMat( InputArray _points1, InputArray _points2, double focal, Point2d pp,
+                              int method, double prob, double threshold, int maxIters, OutputArray _mask)
+{
+    CV_INSTRUMENT_REGION();
+
+    Mat cameraMatrix = (Mat_<double>(3,3) << focal, 0, pp.x, 0, focal, pp.y, 0, 0, 1);
+    return cv::findEssentialMat(_points1, _points2, cameraMatrix, method, prob, threshold, maxIters, _mask);
+}
+
 cv::Mat cv::findEssentialMat( InputArray _points1, InputArray _points2, double focal, Point2d pp,
                               int method, double prob, double threshold, OutputArray _mask)
 {
     CV_INSTRUMENT_REGION();
 
     Mat cameraMatrix = (Mat_<double>(3,3) << focal, 0, pp.x, 0, focal, pp.y, 0, 0, 1);
-    return cv::findEssentialMat(_points1, _points2, cameraMatrix, method, prob, threshold, _mask);
+    return cv::findEssentialMat(_points1, _points2, cameraMatrix, method, prob, threshold, 1000, _mask);
 }
 
 int cv::recoverPose( InputArray E, InputArray _points1, InputArray _points2,

modules/calib3d/src/fundam.cpp

@@ -866,7 +866,7 @@ cv::Mat cv::findFundamentalMat( InputArray _points1, InputArray _points2,
         if( (method & ~3) == FM_RANSAC && npoints >= 15 )
             result = createRANSACPointSetRegistrator(cb, 7, ransacReprojThreshold, confidence, maxIters)->run(m1, m2, F, _mask);
         else
-            result = createLMeDSPointSetRegistrator(cb, 7, confidence)->run(m1, m2, F, _mask);
+            result = createLMeDSPointSetRegistrator(cb, 7, confidence, maxIters)->run(m1, m2, F, _mask);
     }
 
     if( result <= 0 )

modules/calib3d/src/polynom_solver.cpp

@@ -65,7 +65,8 @@ int solve_deg3(double a, double b, double c, double d,
       return 3;
     }
     else {
-      x0 = pow(2 * R, 1 / 3.0) - b_a_3;
+      double cube_root = cv::cubeRoot(2 * R);
+      x0 = cube_root - b_a_3;
       return 1;
     }
   }
@@ -82,8 +83,15 @@ int solve_deg3(double a, double b, double c, double d,
   }
 
   // D > 0, only one real root
-  double AD = pow(fabs(R) + sqrt(D), 1.0 / 3.0) * (R > 0 ? 1 : (R < 0 ? -1 : 0));
-  double BD = (AD == 0) ? 0 : -Q / AD;
+  double AD = 0.;
+  double BD = 0.;
+  double R_abs = fabs(R);
+  if (R_abs > DBL_EPSILON)
+  {
+    AD = cv::cubeRoot(R_abs + sqrt(D));
+    AD = (R >= 0) ? AD : -AD;
+    BD = -Q / AD;
+  }
 
   // Calculate the only real root
   x0 = AD + BD - b_a_3;

modules/calib3d/src/solvepnp.cpp

@@ -311,18 +311,42 @@ bool solvePnPRansac(InputArray _opoints, InputArray _ipoints,
 
     opoints_inliers.resize(npoints1);
     ipoints_inliers.resize(npoints1);
-    result = solvePnP(opoints_inliers, ipoints_inliers, cameraMatrix,
-                      distCoeffs, rvec, tvec, useExtrinsicGuess,
-                      (flags == SOLVEPNP_P3P || flags == SOLVEPNP_AP3P) ? SOLVEPNP_EPNP : flags) ? 1 : -1;
+    try
+    {
+        result = solvePnP(opoints_inliers, ipoints_inliers, cameraMatrix,
+                          distCoeffs, rvec, tvec, useExtrinsicGuess,
+                          (flags == SOLVEPNP_P3P || flags == SOLVEPNP_AP3P) ? SOLVEPNP_EPNP : flags) ? 1 : -1;
+    }
+    catch (const cv::Exception& e)
+    {
+        if (flags == SOLVEPNP_ITERATIVE &&
+            npoints1 == 5 &&
+            e.what() &&
+            std::string(e.what()).find("DLT algorithm needs at least 6 points") != std::string::npos
+        )
+        {
+            CV_LOG_INFO(NULL, "solvePnPRansac(): solvePnP stage to compute the final pose using points "
+                "in the consensus set raised DLT 6 points exception, use result from MSS (Minimal Sample Sets) stage instead.");
+            rvec = _local_model.col(0);    // output rotation vector
+            tvec = _local_model.col(1);    // output translation vector
+            result = 1;
+        }
+        else
+        {
+            // raise other exceptions
+            throw;
+        }
+    }
 
-    if( result <= 0 )
+    if (result <= 0)
     {
         _rvec.assign(_local_model.col(0));    // output rotation vector
         _tvec.assign(_local_model.col(1));    // output translation vector
 
-        if( _inliers.needed() )
+        if (_inliers.needed())
             _inliers.release();
 
+        CV_LOG_DEBUG(NULL, "solvePnPRansac(): solvePnP stage to compute the final pose using points in the consensus set failed. Return false");
         return false;
     }
     else

modules/calib3d/src/stereobm.cpp

@@ -1148,13 +1148,15 @@ class StereoBMImpl CV_FINAL : public StereoBM
 {
 public:
     StereoBMImpl()
+        : params()
     {
-        params = StereoBMParams();
+        // nothing
     }
 
     StereoBMImpl( int _numDisparities, int _SADWindowSize )
+        : params(_numDisparities, _SADWindowSize)
     {
-        params = StereoBMParams(_numDisparities, _SADWindowSize);
+        // nothing
     }
 
     void compute( InputArray leftarr, InputArray rightarr, OutputArray disparr ) CV_OVERRIDE

modules/calib3d/src/stereosgbm.cpp

@@ -2186,19 +2186,21 @@ class StereoSGBMImpl CV_FINAL : public StereoSGBM
 {
 public:
     StereoSGBMImpl()
+        : params()
     {
-        params = StereoSGBMParams();
+        // nothing
     }
 
     StereoSGBMImpl( int _minDisparity, int _numDisparities, int _SADWindowSize,
                     int _P1, int _P2, int _disp12MaxDiff, int _preFilterCap,
                     int _uniquenessRatio, int _speckleWindowSize, int _speckleRange,
                     int _mode )
+        : params(_minDisparity, _numDisparities, _SADWindowSize,
+                 _P1, _P2, _disp12MaxDiff, _preFilterCap,
+                 _uniquenessRatio, _speckleWindowSize, _speckleRange,
+                 _mode)
     {
-        params = StereoSGBMParams( _minDisparity, _numDisparities, _SADWindowSize,
-                                   _P1, _P2, _disp12MaxDiff, _preFilterCap,
-                                   _uniquenessRatio, _speckleWindowSize, _speckleRange,
-                                   _mode );
+        // nothing
     }
 
     void compute( InputArray leftarr, InputArray rightarr, OutputArray disparr ) CV_OVERRIDE

modules/calib3d/test/test_solvepnp_ransac.cpp

@@ -837,6 +837,43 @@ TEST(Calib3d_SolvePnPRansac, double_support)
     EXPECT_LE(cvtest::norm(t, Mat_<double>(tF), NORM_INF), 1e-3);
 }
 
+TEST(Calib3d_SolvePnPRansac, bad_input_points_19253)
+{
+    // with this specific data
+    // when computing the final pose using points in the consensus set with SOLVEPNP_ITERATIVE and solvePnP()
+    // an exception is thrown from solvePnP because there are 5 non-coplanar 3D points and the DLT algorithm needs at least 6 non-coplanar 3D points
+    // with PR #19253 we choose to return true, with the pose estimated from the MSS stage instead of throwing the exception
+
+    float pts2d_[] = {
+        -5.38358629e-01f, -5.09638414e-02f,
+        -5.07192254e-01f, -2.20743284e-01f,
+        -5.43107152e-01f, -4.90474701e-02f,
+        -5.54325163e-01f, -1.86715424e-01f,
+        -5.59334219e-01f, -4.01909500e-02f,
+        -5.43504596e-01f, -4.61776406e-02f
+    };
+    Mat pts2d(6, 2, CV_32FC1, pts2d_);
+
+    float pts3d_[] = {
+        -3.01153604e-02f, -1.55665115e-01f, 4.50000018e-01f,
+        4.27827090e-01f, 4.28645730e-01f, 1.08600008e+00f,
+        -3.14165242e-02f, -1.52656138e-01f, 4.50000018e-01f,
+        -1.46217480e-01f, 5.57961613e-02f, 7.17000008e-01f,
+        -4.89348806e-02f, -1.38795510e-01f, 4.47000027e-01f,
+        -3.13065052e-02f, -1.52636901e-01f, 4.51000035e-01f
+    };
+    Mat pts3d(6, 3, CV_32FC1, pts3d_);
+
+    Mat camera_mat = Mat::eye(3, 3, CV_64FC1);
+    Mat rvec, tvec;
+    vector<int> inliers;
+
+    // solvePnPRansac will return true with 5 inliers, which means the result is from MSS stage.
+    bool result = solvePnPRansac(pts3d, pts2d, camera_mat, noArray(), rvec, tvec, false, 100, 4.f / 460.f, 0.99, inliers);
+    EXPECT_EQ(inliers.size(), size_t(5));
+    EXPECT_TRUE(result);
+}
+
 TEST(Calib3d_SolvePnP, input_type)
 {
     Matx33d intrinsics(5.4794130238156129e+002, 0., 2.9835545700043139e+002, 0.,

modules/core/CMakeLists.txt

@@ -112,6 +112,10 @@ ocv_target_link_libraries(${the_module} PRIVATE
     "${OPENCV_HAL_LINKER_LIBS}"
 )
 
+if(OPENCV_CORE_EXCLUDE_C_API)
+  ocv_target_compile_definitions(${the_module} PRIVATE "OPENCV_EXCLUDE_C_API=1")
+endif()
+
 ocv_add_accuracy_tests()
 ocv_add_perf_tests()
 

modules/core/include/opencv2/core/base.hpp

@@ -587,6 +587,21 @@ _AccTp normInf(const _Tp* a, const _Tp* b, int n)
  */
 CV_EXPORTS_W float cubeRoot(float val);
 
+/** @overload
+
+cubeRoot with argument of `double` type calls `std::cbrt(double)` (C++11) or falls back on `pow()` for C++98 compilation mode.
+*/
+static inline
+double cubeRoot(double val)
+{
+#ifdef CV_CXX11
+    return std::cbrt(val);
+#else
+    double v = pow(abs(val), 1/3.);  // pow doesn't support negative inputs with fractional exponents
+    return val >= 0 ? v : -v;
+#endif
+}
+
 /** @brief Calculates the angle of a 2D vector in degrees.
 
  The function fastAtan2 calculates the full-range angle of an input 2D vector. The angle is measured

modules/core/include/opencv2/core/bindings_utils.hpp

@@ -7,6 +7,7 @@
 
 #include <opencv2/core/async.hpp>
 #include <opencv2/core/detail/async_promise.hpp>
+#include <opencv2/core/utils/logger.hpp>
 
 #include <stdexcept>
 
@@ -144,7 +145,26 @@ AsyncArray testAsyncException()
     return p.getArrayResult();
 }
 
-//! @}
-}} // namespace
+//! @}  // core_utils
+}  // namespace cv::utils
+
+//! @cond IGNORED
+
+CV_WRAP static inline
+int setLogLevel(int level)
+{
+    // NB: Binding generators doesn't work with enums properly yet, so we define separate overload here
+    return cv::utils::logging::setLogLevel((cv::utils::logging::LogLevel)level);
+}
+
+CV_WRAP static inline
+int getLogLevel()
+{
+    return cv::utils::logging::getLogLevel();
+}
+
+//! @endcond IGNORED
+
+} // namespaces cv /  utils
 
 #endif // OPENCV_CORE_BINDINGS_UTILS_HPP

modules/core/include/opencv2/core/cvdef.h

@@ -554,7 +554,7 @@ Cv64suf;
 #  define CV_XADD(addr, delta) (int)_InterlockedExchangeAdd((long volatile*)addr, delta)
 #else
   #ifdef OPENCV_FORCE_UNSAFE_XADD
-    CV_INLINE CV_XADD(int* addr, int delta) { int tmp = *addr; *addr += delta; return tmp; }
+    CV_INLINE int CV_XADD(int* addr, int delta) { int tmp = *addr; *addr += delta; return tmp; }
   #else
     #error "OpenCV: can't define safe CV_XADD macro for current platform (unsupported). Define CV_XADD macro through custom port header (see OPENCV_INCLUDE_PORT_FILE)"
   #endif

modules/core/include/opencv2/core/fast_math.hpp

@@ -76,6 +76,9 @@
   #if defined __PPC64__ && defined __GNUC__ && defined _ARCH_PWR8 \
       && !defined(OPENCV_SKIP_INCLUDE_ALTIVEC_H)
     #include <altivec.h>
+    #undef vector
+    #undef bool
+    #undef pixel
   #endif
 
   #if defined(CV_INLINE_ROUND_FLT)

modules/core/include/opencv2/core/hal/intrin.hpp

@@ -104,7 +104,7 @@ template<typename _Tp> struct V_TypeTraits
 {
 };
 
-#define CV_INTRIN_DEF_TYPE_TRAITS(type, int_type_, uint_type_, abs_type_, w_type_, q_type_, sum_type_, nlanes128_) \
+#define CV_INTRIN_DEF_TYPE_TRAITS(type, int_type_, uint_type_, abs_type_, w_type_, q_type_, sum_type_) \
     template<> struct V_TypeTraits<type> \
     { \
         typedef type value_type; \
@@ -114,7 +114,6 @@ template<typename _Tp> struct V_TypeTraits
         typedef w_type_ w_type; \
         typedef q_type_ q_type; \
         typedef sum_type_ sum_type; \
-        enum { nlanes128 = nlanes128_ }; \
     \
         static inline int_type reinterpret_int(type x) \
         { \
@@ -131,7 +130,7 @@ template<typename _Tp> struct V_TypeTraits
         } \
     }
 
-#define CV_INTRIN_DEF_TYPE_TRAITS_NO_Q_TYPE(type, int_type_, uint_type_, abs_type_, w_type_, sum_type_, nlanes128_) \
+#define CV_INTRIN_DEF_TYPE_TRAITS_NO_Q_TYPE(type, int_type_, uint_type_, abs_type_, w_type_, sum_type_) \
     template<> struct V_TypeTraits<type> \
     { \
         typedef type value_type; \
@@ -140,7 +139,6 @@ template<typename _Tp> struct V_TypeTraits
         typedef uint_type_ uint_type; \
         typedef w_type_ w_type; \
         typedef sum_type_ sum_type; \
-        enum { nlanes128 = nlanes128_ }; \
     \
         static inline int_type reinterpret_int(type x) \
         { \
@@ -157,16 +155,16 @@ template<typename _Tp> struct V_TypeTraits
         } \
     }
 
-CV_INTRIN_DEF_TYPE_TRAITS(uchar, schar, uchar, uchar, ushort, unsigned, unsigned, 16);
-CV_INTRIN_DEF_TYPE_TRAITS(schar, schar, uchar, uchar, short, int, int, 16);
-CV_INTRIN_DEF_TYPE_TRAITS(ushort, short, ushort, ushort, unsigned, uint64, unsigned, 8);
-CV_INTRIN_DEF_TYPE_TRAITS(short, short, ushort, ushort, int, int64, int, 8);
-CV_INTRIN_DEF_TYPE_TRAITS_NO_Q_TYPE(unsigned, int, unsigned, unsigned, uint64, unsigned, 4);
-CV_INTRIN_DEF_TYPE_TRAITS_NO_Q_TYPE(int, int, unsigned, unsigned, int64, int, 4);
-CV_INTRIN_DEF_TYPE_TRAITS_NO_Q_TYPE(float, int, unsigned, float, double, float, 4);
-CV_INTRIN_DEF_TYPE_TRAITS_NO_Q_TYPE(uint64, int64, uint64, uint64, void, uint64, 2);
-CV_INTRIN_DEF_TYPE_TRAITS_NO_Q_TYPE(int64, int64, uint64, uint64, void, int64, 2);
-CV_INTRIN_DEF_TYPE_TRAITS_NO_Q_TYPE(double, int64, uint64, double, void, double, 2);
+CV_INTRIN_DEF_TYPE_TRAITS(uchar, schar, uchar, uchar, ushort, unsigned, unsigned);
+CV_INTRIN_DEF_TYPE_TRAITS(schar, schar, uchar, uchar, short, int, int);
+CV_INTRIN_DEF_TYPE_TRAITS(ushort, short, ushort, ushort, unsigned, uint64, unsigned);
+CV_INTRIN_DEF_TYPE_TRAITS(short, short, ushort, ushort, int, int64, int);
+CV_INTRIN_DEF_TYPE_TRAITS_NO_Q_TYPE(unsigned, int, unsigned, unsigned, uint64, unsigned);
+CV_INTRIN_DEF_TYPE_TRAITS_NO_Q_TYPE(int, int, unsigned, unsigned, int64, int);
+CV_INTRIN_DEF_TYPE_TRAITS_NO_Q_TYPE(float, int, unsigned, float, double, float);
+CV_INTRIN_DEF_TYPE_TRAITS_NO_Q_TYPE(uint64, int64, uint64, uint64, void, uint64);
+CV_INTRIN_DEF_TYPE_TRAITS_NO_Q_TYPE(int64, int64, uint64, uint64, void, int64);
+CV_INTRIN_DEF_TYPE_TRAITS_NO_Q_TYPE(double, int64, uint64, double, void, double);
 
 #ifndef CV_DOXYGEN
 
@@ -310,54 +308,6 @@ CV_CPU_OPTIMIZATION_HAL_NAMESPACE_BEGIN
 
 //==================================================================================================
 
-#define CV_INTRIN_DEFINE_WIDE_INTRIN(typ, vtyp, short_typ, prefix, loadsfx) \
-    inline vtyp vx_setall_##short_typ(typ v) { return prefix##_setall_##short_typ(v); } \
-    inline vtyp vx_setzero_##short_typ() { return prefix##_setzero_##short_typ(); } \
-    inline vtyp vx_##loadsfx(const typ* ptr) { return prefix##_##loadsfx(ptr); } \
-    inline vtyp vx_##loadsfx##_aligned(const typ* ptr) { return prefix##_##loadsfx##_aligned(ptr); } \
-    inline vtyp vx_##loadsfx##_low(const typ* ptr) { return prefix##_##loadsfx##_low(ptr); } \
-    inline vtyp vx_##loadsfx##_halves(const typ* ptr0, const typ* ptr1) { return prefix##_##loadsfx##_halves(ptr0, ptr1); } \
-    inline void vx_store(typ* ptr, const vtyp& v) { return v_store(ptr, v); } \
-    inline void vx_store_aligned(typ* ptr, const vtyp& v) { return v_store_aligned(ptr, v); } \
-    inline vtyp vx_lut(const typ* ptr, const int* idx) { return prefix##_lut(ptr, idx); } \
-    inline vtyp vx_lut_pairs(const typ* ptr, const int* idx) { return prefix##_lut_pairs(ptr, idx); }
-
-#define CV_INTRIN_DEFINE_WIDE_LUT_QUAD(typ, vtyp, prefix) \
-    inline vtyp vx_lut_quads(const typ* ptr, const int* idx) { return prefix##_lut_quads(ptr, idx); }
-
-#define CV_INTRIN_DEFINE_WIDE_LOAD_EXPAND(typ, wtyp, prefix) \
-    inline wtyp vx_load_expand(const typ* ptr) { return prefix##_load_expand(ptr); }
-
-#define CV_INTRIN_DEFINE_WIDE_LOAD_EXPAND_Q(typ, qtyp, prefix) \
-    inline qtyp vx_load_expand_q(const typ* ptr) { return prefix##_load_expand_q(ptr); }
-
-#define CV_INTRIN_DEFINE_WIDE_INTRIN_WITH_EXPAND(typ, vtyp, short_typ, wtyp, qtyp, prefix, loadsfx) \
-    CV_INTRIN_DEFINE_WIDE_INTRIN(typ, vtyp, short_typ, prefix, loadsfx) \
-    CV_INTRIN_DEFINE_WIDE_LUT_QUAD(typ, vtyp, prefix) \
-    CV_INTRIN_DEFINE_WIDE_LOAD_EXPAND(typ, wtyp, prefix) \
-    CV_INTRIN_DEFINE_WIDE_LOAD_EXPAND_Q(typ, qtyp, prefix)
-
-#define CV_INTRIN_DEFINE_WIDE_INTRIN_ALL_TYPES(prefix) \
-    CV_INTRIN_DEFINE_WIDE_INTRIN_WITH_EXPAND(uchar, v_uint8, u8, v_uint16, v_uint32, prefix, load) \
-    CV_INTRIN_DEFINE_WIDE_INTRIN_WITH_EXPAND(schar, v_int8, s8, v_int16, v_int32, prefix, load) \
-    CV_INTRIN_DEFINE_WIDE_INTRIN(ushort, v_uint16, u16, prefix, load) \
-    CV_INTRIN_DEFINE_WIDE_LUT_QUAD(ushort, v_uint16, prefix) \
-    CV_INTRIN_DEFINE_WIDE_LOAD_EXPAND(ushort, v_uint32, prefix) \
-    CV_INTRIN_DEFINE_WIDE_INTRIN(short, v_int16, s16, prefix, load) \
-    CV_INTRIN_DEFINE_WIDE_LUT_QUAD(short, v_int16, prefix) \
-    CV_INTRIN_DEFINE_WIDE_LOAD_EXPAND(short, v_int32, prefix) \
-    CV_INTRIN_DEFINE_WIDE_INTRIN(int, v_int32, s32, prefix, load) \
-    CV_INTRIN_DEFINE_WIDE_LUT_QUAD(int, v_int32, prefix) \
-    CV_INTRIN_DEFINE_WIDE_LOAD_EXPAND(int, v_int64, prefix) \
-    CV_INTRIN_DEFINE_WIDE_INTRIN(unsigned, v_uint32, u32, prefix, load) \
-    CV_INTRIN_DEFINE_WIDE_LUT_QUAD(unsigned, v_uint32, prefix) \
-    CV_INTRIN_DEFINE_WIDE_LOAD_EXPAND(unsigned, v_uint64, prefix) \
-    CV_INTRIN_DEFINE_WIDE_INTRIN(float, v_float32, f32, prefix, load) \
-    CV_INTRIN_DEFINE_WIDE_LUT_QUAD(float, v_float32, prefix) \
-    CV_INTRIN_DEFINE_WIDE_INTRIN(int64, v_int64, s64, prefix, load) \
-    CV_INTRIN_DEFINE_WIDE_INTRIN(uint64, v_uint64, u64, prefix, load) \
-    CV_INTRIN_DEFINE_WIDE_LOAD_EXPAND(float16_t, v_float32, prefix)
-
 template<typename _Tp> struct V_RegTraits
 {
 };
@@ -417,6 +367,7 @@ template<typename _Tp> struct V_RegTraits
     CV_DEF_REG_TRAITS(v512, v_int64x8, int64, s64, v_uint64x8, void, void, v_int64x8, void);
     CV_DEF_REG_TRAITS(v512, v_float64x8, double, f64, v_float64x8, void, void, v_int64x8, v_int32x16);
 #endif
+//! @endcond
 
 #if CV_SIMD512 && (!defined(CV__SIMD_FORCE_WIDTH) || CV__SIMD_FORCE_WIDTH == 512)
 #define CV__SIMD_NAMESPACE simd512
@@ -425,21 +376,33 @@ namespace CV__SIMD_NAMESPACE {
     #define CV_SIMD_64F CV_SIMD512_64F
     #define CV_SIMD_FP16 CV_SIMD512_FP16
     #define CV_SIMD_WIDTH 64
+//! @addtogroup core_hal_intrin
+//! @{
+    //! @brief Maximum available vector register capacity 8-bit unsigned integer values
     typedef v_uint8x64    v_uint8;
+    //! @brief Maximum available vector register capacity 8-bit signed integer values
     typedef v_int8x64     v_int8;
+    //! @brief Maximum available vector register capacity 16-bit unsigned integer values
     typedef v_uint16x32   v_uint16;
+    //! @brief Maximum available vector register capacity 16-bit signed integer values
     typedef v_int16x32    v_int16;
+    //! @brief Maximum available vector register capacity 32-bit unsigned integer values
     typedef v_uint32x16   v_uint32;
+    //! @brief Maximum available vector register capacity 32-bit signed integer values
     typedef v_int32x16    v_int32;
+    //! @brief Maximum available vector register capacity 64-bit unsigned integer values
     typedef v_uint64x8    v_uint64;
+    //! @brief Maximum available vector register capacity 64-bit signed integer values
     typedef v_int64x8     v_int64;
+    //! @brief Maximum available vector register capacity 32-bit floating point values (single precision)
     typedef v_float32x16  v_float32;
-    CV_INTRIN_DEFINE_WIDE_INTRIN_ALL_TYPES(v512)
-#if CV_SIMD512_64F
+    #if CV_SIMD512_64F
+    //! @brief Maximum available vector register capacity 64-bit floating point values (double precision)
     typedef v_float64x8   v_float64;
-    CV_INTRIN_DEFINE_WIDE_INTRIN(double, v_float64, f64, v512, load)
-#endif
-        inline void vx_cleanup() { v512_cleanup(); }
+    #endif
+//! @}
+
+    #define VXPREFIX(func) v512##func
 } // namespace
 using namespace CV__SIMD_NAMESPACE;
 #elif CV_SIMD256 && (!defined(CV__SIMD_FORCE_WIDTH) || CV__SIMD_FORCE_WIDTH == 256)
@@ -449,21 +412,33 @@ namespace CV__SIMD_NAMESPACE {
     #define CV_SIMD_64F CV_SIMD256_64F
     #define CV_SIMD_FP16 CV_SIMD256_FP16
     #define CV_SIMD_WIDTH 32
+//! @addtogroup core_hal_intrin
+//! @{
+    //! @brief Maximum available vector register capacity 8-bit unsigned integer values
     typedef v_uint8x32   v_uint8;
+    //! @brief Maximum available vector register capacity 8-bit signed integer values
     typedef v_int8x32    v_int8;
+    //! @brief Maximum available vector register capacity 16-bit unsigned integer values
     typedef v_uint16x16  v_uint16;
+    //! @brief Maximum available vector register capacity 16-bit signed integer values
     typedef v_int16x16   v_int16;
+    //! @brief Maximum available vector register capacity 32-bit unsigned integer values
     typedef v_uint32x8   v_uint32;
+    //! @brief Maximum available vector register capacity 32-bit signed integer values
     typedef v_int32x8    v_int32;
+    //! @brief Maximum available vector register capacity 64-bit unsigned integer values
     typedef v_uint64x4   v_uint64;
+    //! @brief Maximum available vector register capacity 64-bit signed integer values
     typedef v_int64x4    v_int64;
+    //! @brief Maximum available vector register capacity 32-bit floating point values (single precision)
     typedef v_float32x8  v_float32;
-    CV_INTRIN_DEFINE_WIDE_INTRIN_ALL_TYPES(v256)
     #if CV_SIMD256_64F
+    //! @brief Maximum available vector register capacity 64-bit floating point values (double precision)
     typedef v_float64x4  v_float64;
-    CV_INTRIN_DEFINE_WIDE_INTRIN(double, v_float64, f64, v256, load)
     #endif
-    inline void vx_cleanup() { v256_cleanup(); }
+//! @}
+
+    #define VXPREFIX(func) v256##func
 } // namespace
 using namespace CV__SIMD_NAMESPACE;
 #elif (CV_SIMD128 || CV_SIMD128_CPP) && (!defined(CV__SIMD_FORCE_WIDTH) || CV__SIMD_FORCE_WIDTH == 128)
@@ -476,25 +451,228 @@ namespace CV__SIMD_NAMESPACE {
     #define CV_SIMD CV_SIMD128
     #define CV_SIMD_64F CV_SIMD128_64F
     #define CV_SIMD_WIDTH 16
+//! @addtogroup core_hal_intrin
+//! @{
+    //! @brief Maximum available vector register capacity 8-bit unsigned integer values
     typedef v_uint8x16  v_uint8;
+    //! @brief Maximum available vector register capacity 8-bit signed integer values
     typedef v_int8x16   v_int8;
+    //! @brief Maximum available vector register capacity 16-bit unsigned integer values
     typedef v_uint16x8  v_uint16;
+    //! @brief Maximum available vector register capacity 16-bit signed integer values
     typedef v_int16x8   v_int16;
+    //! @brief Maximum available vector register capacity 32-bit unsigned integer values
     typedef v_uint32x4  v_uint32;
+    //! @brief Maximum available vector register capacity 32-bit signed integer values
     typedef v_int32x4   v_int32;
+    //! @brief Maximum available vector register capacity 64-bit unsigned integer values
     typedef v_uint64x2  v_uint64;
+    //! @brief Maximum available vector register capacity 64-bit signed integer values
     typedef v_int64x2   v_int64;
+    //! @brief Maximum available vector register capacity 32-bit floating point values (single precision)
     typedef v_float32x4 v_float32;
-    CV_INTRIN_DEFINE_WIDE_INTRIN_ALL_TYPES(v)
     #if CV_SIMD128_64F
+    //! @brief Maximum available vector register capacity 64-bit floating point values (double precision)
     typedef v_float64x2 v_float64;
-    CV_INTRIN_DEFINE_WIDE_INTRIN(double, v_float64, f64, v, load)
     #endif
-    inline void vx_cleanup() { v_cleanup(); }
+//! @}
+
+    #define VXPREFIX(func) v##func
 } // namespace
 using namespace CV__SIMD_NAMESPACE;
 #endif
 
+namespace CV__SIMD_NAMESPACE {
+//! @addtogroup core_hal_intrin
+//! @{
+    //! @name Wide init with value
+    //! @{
+    //! @brief Create maximum available capacity vector with elements set to a specific value
+    inline v_uint8 vx_setall_u8(uchar v) { return VXPREFIX(_setall_u8)(v); }
+    inline v_int8 vx_setall_s8(schar v) { return VXPREFIX(_setall_s8)(v); }
+    inline v_uint16 vx_setall_u16(ushort v) { return VXPREFIX(_setall_u16)(v); }
+    inline v_int16 vx_setall_s16(short v) { return VXPREFIX(_setall_s16)(v); }
+    inline v_int32 vx_setall_s32(int v) { return VXPREFIX(_setall_s32)(v); }
+    inline v_uint32 vx_setall_u32(unsigned v) { return VXPREFIX(_setall_u32)(v); }
+    inline v_float32 vx_setall_f32(float v) { return VXPREFIX(_setall_f32)(v); }
+    inline v_int64 vx_setall_s64(int64 v) { return VXPREFIX(_setall_s64)(v); }
+    inline v_uint64 vx_setall_u64(uint64 v) { return VXPREFIX(_setall_u64)(v); }
+#if CV_SIMD_64F
+    inline v_float64 vx_setall_f64(double v) { return VXPREFIX(_setall_f64)(v); }
+#endif
+    //! @}
+
+    //! @name Wide init with zero
+    //! @{
+    //! @brief Create maximum available capacity vector with elements set to zero
+    inline v_uint8 vx_setzero_u8() { return VXPREFIX(_setzero_u8)(); }
+    inline v_int8 vx_setzero_s8() { return VXPREFIX(_setzero_s8)(); }
+    inline v_uint16 vx_setzero_u16() { return VXPREFIX(_setzero_u16)(); }
+    inline v_int16 vx_setzero_s16() { return VXPREFIX(_setzero_s16)(); }
+    inline v_int32 vx_setzero_s32() { return VXPREFIX(_setzero_s32)(); }
+    inline v_uint32 vx_setzero_u32() { return VXPREFIX(_setzero_u32)(); }
+    inline v_float32 vx_setzero_f32() { return VXPREFIX(_setzero_f32)(); }
+    inline v_int64 vx_setzero_s64() { return VXPREFIX(_setzero_s64)(); }
+    inline v_uint64 vx_setzero_u64() { return VXPREFIX(_setzero_u64)(); }
+#if CV_SIMD_64F
+    inline v_float64 vx_setzero_f64() { return VXPREFIX(_setzero_f64)(); }
+#endif
+    //! @}
+
+    //! @name Wide load from memory
+    //! @{
+    //! @brief Load maximum available capacity register contents from memory
+    inline v_uint8 vx_load(const uchar * ptr) { return VXPREFIX(_load)(ptr); }
+    inline v_int8 vx_load(const schar * ptr) { return VXPREFIX(_load)(ptr); }
+    inline v_uint16 vx_load(const ushort * ptr) { return VXPREFIX(_load)(ptr); }
+    inline v_int16 vx_load(const short * ptr) { return VXPREFIX(_load)(ptr); }
+    inline v_int32 vx_load(const int * ptr) { return VXPREFIX(_load)(ptr); }
+    inline v_uint32 vx_load(const unsigned * ptr) { return VXPREFIX(_load)(ptr); }
+    inline v_float32 vx_load(const float * ptr) { return VXPREFIX(_load)(ptr); }
+    inline v_int64 vx_load(const int64 * ptr) { return VXPREFIX(_load)(ptr); }
+    inline v_uint64 vx_load(const uint64 * ptr) { return VXPREFIX(_load)(ptr); }
+#if CV_SIMD_64F
+    inline v_float64 vx_load(const double * ptr) { return VXPREFIX(_load)(ptr); }
+#endif
+    //! @}
+
+    //! @name Wide load from memory(aligned)
+    //! @{
+    //! @brief Load maximum available capacity register contents from memory(aligned)
+    inline v_uint8 vx_load_aligned(const uchar * ptr) { return VXPREFIX(_load_aligned)(ptr); }
+    inline v_int8 vx_load_aligned(const schar * ptr) { return VXPREFIX(_load_aligned)(ptr); }
+    inline v_uint16 vx_load_aligned(const ushort * ptr) { return VXPREFIX(_load_aligned)(ptr); }
+    inline v_int16 vx_load_aligned(const short * ptr) { return VXPREFIX(_load_aligned)(ptr); }
+    inline v_int32 vx_load_aligned(const int * ptr) { return VXPREFIX(_load_aligned)(ptr); }
+    inline v_uint32 vx_load_aligned(const unsigned * ptr) { return VXPREFIX(_load_aligned)(ptr); }
+    inline v_float32 vx_load_aligned(const float * ptr) { return VXPREFIX(_load_aligned)(ptr); }
+    inline v_int64 vx_load_aligned(const int64 * ptr) { return VXPREFIX(_load_aligned)(ptr); }
+    inline v_uint64 vx_load_aligned(const uint64 * ptr) { return VXPREFIX(_load_aligned)(ptr); }
+#if CV_SIMD_64F
+    inline v_float64 vx_load_aligned(const double * ptr) { return VXPREFIX(_load_aligned)(ptr); }
+#endif
+    //! @}
+
+    //! @name Wide load lower half from memory
+    //! @{
+    //! @brief Load lower half of maximum available capacity register from memory
+    inline v_uint8 vx_load_low(const uchar * ptr) { return VXPREFIX(_load_low)(ptr); }
+    inline v_int8 vx_load_low(const schar * ptr) { return VXPREFIX(_load_low)(ptr); }
+    inline v_uint16 vx_load_low(const ushort * ptr) { return VXPREFIX(_load_low)(ptr); }
+    inline v_int16 vx_load_low(const short * ptr) { return VXPREFIX(_load_low)(ptr); }
+    inline v_int32 vx_load_low(const int * ptr) { return VXPREFIX(_load_low)(ptr); }
+    inline v_uint32 vx_load_low(const unsigned * ptr) { return VXPREFIX(_load_low)(ptr); }
+    inline v_float32 vx_load_low(const float * ptr) { return VXPREFIX(_load_low)(ptr); }
+    inline v_int64 vx_load_low(const int64 * ptr) { return VXPREFIX(_load_low)(ptr); }
+    inline v_uint64 vx_load_low(const uint64 * ptr) { return VXPREFIX(_load_low)(ptr); }
+#if CV_SIMD_64F
+    inline v_float64 vx_load_low(const double * ptr) { return VXPREFIX(_load_low)(ptr); }
+#endif
+    //! @}
+
+    //! @name Wide load halfs from memory
+    //! @{
+    //! @brief Load maximum available capacity register contents from two memory blocks
+    inline v_uint8 vx_load_halves(const uchar * ptr0, const uchar * ptr1) { return VXPREFIX(_load_halves)(ptr0, ptr1); }
+    inline v_int8 vx_load_halves(const schar * ptr0, const schar * ptr1) { return VXPREFIX(_load_halves)(ptr0, ptr1); }
+    inline v_uint16 vx_load_halves(const ushort * ptr0, const ushort * ptr1) { return VXPREFIX(_load_halves)(ptr0, ptr1); }
+    inline v_int16 vx_load_halves(const short * ptr0, const short * ptr1) { return VXPREFIX(_load_halves)(ptr0, ptr1); }
+    inline v_int32 vx_load_halves(const int * ptr0, const int * ptr1) { return VXPREFIX(_load_halves)(ptr0, ptr1); }
+    inline v_uint32 vx_load_halves(const unsigned * ptr0, const unsigned * ptr1) { return VXPREFIX(_load_halves)(ptr0, ptr1); }
+    inline v_float32 vx_load_halves(const float * ptr0, const float * ptr1) { return VXPREFIX(_load_halves)(ptr0, ptr1); }
+    inline v_int64 vx_load_halves(const int64 * ptr0, const int64 * ptr1) { return VXPREFIX(_load_halves)(ptr0, ptr1); }
+    inline v_uint64 vx_load_halves(const uint64 * ptr0, const uint64 * ptr1) { return VXPREFIX(_load_halves)(ptr0, ptr1); }
+#if CV_SIMD_64F
+    inline v_float64 vx_load_halves(const double * ptr0, const double * ptr1) { return VXPREFIX(_load_halves)(ptr0, ptr1); }
+#endif
+    //! @}
+
+    //! @name Wide LUT of elements
+    //! @{
+    //! @brief Load maximum available capacity register contents with array elements by provided indexes
+    inline v_uint8 vx_lut(const uchar * ptr, const int* idx) { return VXPREFIX(_lut)(ptr, idx); }
+    inline v_int8 vx_lut(const schar * ptr, const int* idx) { return VXPREFIX(_lut)(ptr, idx); }
+    inline v_uint16 vx_lut(const ushort * ptr, const int* idx) { return VXPREFIX(_lut)(ptr, idx); }
+    inline v_int16 vx_lut(const short* ptr, const int* idx) { return VXPREFIX(_lut)(ptr, idx); }
+    inline v_int32 vx_lut(const int* ptr, const int* idx) { return VXPREFIX(_lut)(ptr, idx); }
+    inline v_uint32 vx_lut(const unsigned* ptr, const int* idx) { return VXPREFIX(_lut)(ptr, idx); }
+    inline v_float32 vx_lut(const float* ptr, const int* idx) { return VXPREFIX(_lut)(ptr, idx); }
+    inline v_int64 vx_lut(const int64 * ptr, const int* idx) { return VXPREFIX(_lut)(ptr, idx); }
+    inline v_uint64 vx_lut(const uint64 * ptr, const int* idx) { return VXPREFIX(_lut)(ptr, idx); }
+#if CV_SIMD_64F
+    inline v_float64 vx_lut(const double* ptr, const int* idx) { return VXPREFIX(_lut)(ptr, idx); }
+#endif
+    //! @}
+
+    //! @name Wide LUT of element pairs
+    //! @{
+    //! @brief Load maximum available capacity register contents with array element pairs by provided indexes
+    inline v_uint8 vx_lut_pairs(const uchar * ptr, const int* idx) { return VXPREFIX(_lut_pairs)(ptr, idx); }
+    inline v_int8 vx_lut_pairs(const schar * ptr, const int* idx) { return VXPREFIX(_lut_pairs)(ptr, idx); }
+    inline v_uint16 vx_lut_pairs(const ushort * ptr, const int* idx) { return VXPREFIX(_lut_pairs)(ptr, idx); }
+    inline v_int16 vx_lut_pairs(const short* ptr, const int* idx) { return VXPREFIX(_lut_pairs)(ptr, idx); }
+    inline v_int32 vx_lut_pairs(const int* ptr, const int* idx) { return VXPREFIX(_lut_pairs)(ptr, idx); }
+    inline v_uint32 vx_lut_pairs(const unsigned* ptr, const int* idx) { return VXPREFIX(_lut_pairs)(ptr, idx); }
+    inline v_float32 vx_lut_pairs(const float* ptr, const int* idx) { return VXPREFIX(_lut_pairs)(ptr, idx); }
+    inline v_int64 vx_lut_pairs(const int64 * ptr, const int* idx) { return VXPREFIX(_lut_pairs)(ptr, idx); }
+    inline v_uint64 vx_lut_pairs(const uint64 * ptr, const int* idx) { return VXPREFIX(_lut_pairs)(ptr, idx); }
+#if CV_SIMD_64F
+    inline v_float64 vx_lut_pairs(const double* ptr, const int* idx) { return VXPREFIX(_lut_pairs)(ptr, idx); }
+#endif
+    //! @}
+
+    //! @name Wide LUT of element quads
+    //! @{
+    //! @brief Load maximum available capacity register contents with array element quads by provided indexes
+    inline v_uint8 vx_lut_quads(const uchar* ptr, const int* idx) { return VXPREFIX(_lut_quads)(ptr, idx); }
+    inline v_int8 vx_lut_quads(const schar* ptr, const int* idx) { return VXPREFIX(_lut_quads)(ptr, idx); }
+    inline v_uint16 vx_lut_quads(const ushort* ptr, const int* idx) { return VXPREFIX(_lut_quads)(ptr, idx); }
+    inline v_int16 vx_lut_quads(const short* ptr, const int* idx) { return VXPREFIX(_lut_quads)(ptr, idx); }
+    inline v_int32 vx_lut_quads(const int* ptr, const int* idx) { return VXPREFIX(_lut_quads)(ptr, idx); }
+    inline v_uint32 vx_lut_quads(const unsigned* ptr, const int* idx) { return VXPREFIX(_lut_quads)(ptr, idx); }
+    inline v_float32 vx_lut_quads(const float* ptr, const int* idx) { return VXPREFIX(_lut_quads)(ptr, idx); }
+    //! @}
+
+    //! @name Wide load with double expansion
+    //! @{
+    //! @brief Load maximum available capacity register contents from memory with double expand
+    inline v_uint16 vx_load_expand(const uchar * ptr) { return VXPREFIX(_load_expand)(ptr); }
+    inline v_int16 vx_load_expand(const schar * ptr) { return VXPREFIX(_load_expand)(ptr); }
+    inline v_uint32 vx_load_expand(const ushort * ptr) { return VXPREFIX(_load_expand)(ptr); }
+    inline v_int32 vx_load_expand(const short* ptr) { return VXPREFIX(_load_expand)(ptr); }
+    inline v_int64 vx_load_expand(const int* ptr) { return VXPREFIX(_load_expand)(ptr); }
+    inline v_uint64 vx_load_expand(const unsigned* ptr) { return VXPREFIX(_load_expand)(ptr); }
+    inline v_float32 vx_load_expand(const float16_t * ptr) { return VXPREFIX(_load_expand)(ptr); }
+    //! @}
+
+    //! @name Wide load with quad expansion
+    //! @{
+    //! @brief Load maximum available capacity register contents from memory with quad expand
+    inline v_uint32 vx_load_expand_q(const uchar * ptr) { return VXPREFIX(_load_expand_q)(ptr); }
+    inline v_int32 vx_load_expand_q(const schar * ptr) { return VXPREFIX(_load_expand_q)(ptr); }
+    //! @}
+
+    /** @brief SIMD processing state cleanup call */
+    inline void vx_cleanup() { VXPREFIX(_cleanup)(); }
+
+
+//! @cond IGNORED
+
+    // backward compatibility
+    template<typename _Tp, typename _Tvec> static inline
+    void vx_store(_Tp* dst, const _Tvec& v) { return v_store(dst, v); }
+    // backward compatibility
+    template<typename _Tp, typename _Tvec> static inline
+    void vx_store_aligned(_Tp* dst, const _Tvec& v) { return v_store_aligned(dst, v); }
+
+//! @endcond
+
+
+//! @}
+    #undef VXPREFIX
+} // namespace
+
+//! @cond IGNORED
 #ifndef CV_SIMD_64F
 #define CV_SIMD_64F 0
 #endif

modules/core/include/opencv2/core/hal/intrin_neon.hpp

@@ -538,49 +538,81 @@ inline void v_mul_expand(const v_int8x16& a, const v_int8x16& b,
                          v_int16x8& c, v_int16x8& d)
 {
     c.val = vmull_s8(vget_low_s8(a.val), vget_low_s8(b.val));
+#if CV_NEON_AARCH64
+    d.val = vmull_high_s8(a.val, b.val);
+#else // #if CV_NEON_AARCH64
     d.val = vmull_s8(vget_high_s8(a.val), vget_high_s8(b.val));
+#endif // #if CV_NEON_AARCH64
 }
 
 inline void v_mul_expand(const v_uint8x16& a, const v_uint8x16& b,
                          v_uint16x8& c, v_uint16x8& d)
 {
     c.val = vmull_u8(vget_low_u8(a.val), vget_low_u8(b.val));
+#if CV_NEON_AARCH64
+    d.val = vmull_high_u8(a.val, b.val);
+#else // #if CV_NEON_AARCH64
     d.val = vmull_u8(vget_high_u8(a.val), vget_high_u8(b.val));
+#endif // #if CV_NEON_AARCH64
 }
 
 inline void v_mul_expand(const v_int16x8& a, const v_int16x8& b,
                          v_int32x4& c, v_int32x4& d)
 {
     c.val = vmull_s16(vget_low_s16(a.val), vget_low_s16(b.val));
+#if CV_NEON_AARCH64
+    d.val = vmull_high_s16(a.val, b.val);
+#else // #if CV_NEON_AARCH64
     d.val = vmull_s16(vget_high_s16(a.val), vget_high_s16(b.val));
+#endif // #if CV_NEON_AARCH64
 }
 
 inline void v_mul_expand(const v_uint16x8& a, const v_uint16x8& b,
                          v_uint32x4& c, v_uint32x4& d)
 {
     c.val = vmull_u16(vget_low_u16(a.val), vget_low_u16(b.val));
+#if CV_NEON_AARCH64
+    d.val = vmull_high_u16(a.val, b.val);
+#else // #if CV_NEON_AARCH64
     d.val = vmull_u16(vget_high_u16(a.val), vget_high_u16(b.val));
+#endif // #if CV_NEON_AARCH64
 }
 
 inline void v_mul_expand(const v_uint32x4& a, const v_uint32x4& b,
                          v_uint64x2& c, v_uint64x2& d)
 {
     c.val = vmull_u32(vget_low_u32(a.val), vget_low_u32(b.val));
+#if CV_NEON_AARCH64
+    d.val = vmull_high_u32(a.val, b.val);
+#else // #if CV_NEON_AARCH64
     d.val = vmull_u32(vget_high_u32(a.val), vget_high_u32(b.val));
+#endif // #if CV_NEON_AARCH64
 }
 
 inline v_int16x8 v_mul_hi(const v_int16x8& a, const v_int16x8& b)
 {
     return v_int16x8(vcombine_s16(
                                   vshrn_n_s32(vmull_s16( vget_low_s16(a.val),  vget_low_s16(b.val)), 16),
-                                  vshrn_n_s32(vmull_s16(vget_high_s16(a.val), vget_high_s16(b.val)), 16)
+                                  vshrn_n_s32(
+#if CV_NEON_AARCH64
+                                    vmull_high_s16(a.val, b.val)
+#else // #if CV_NEON_AARCH64
+                                    vmull_s16(vget_high_s16(a.val), vget_high_s16(b.val))
+#endif // #if CV_NEON_AARCH64
+                                    , 16)
                                  ));
 }
 inline v_uint16x8 v_mul_hi(const v_uint16x8& a, const v_uint16x8& b)
 {
     return v_uint16x8(vcombine_u16(
                                    vshrn_n_u32(vmull_u16( vget_low_u16(a.val),  vget_low_u16(b.val)), 16),
-                                   vshrn_n_u32(vmull_u16(vget_high_u16(a.val), vget_high_u16(b.val)), 16)
+                                   vshrn_n_u32(
+#if CV_NEON_AARCH64
+                                    vmull_high_u16(a.val, b.val)
+#else // #if CV_NEON_AARCH64
+                                    vmull_u16(vget_high_u16(a.val), vget_high_u16(b.val))
+#endif // #if CV_NEON_AARCH64
+                                    , 16)
                                   ));
 }
 
@@ -1254,29 +1286,56 @@ OPENCV_HAL_IMPL_NEON_LOADSTORE_OP(v_float64x2, double, f64)
 
 inline unsigned v_reduce_sum(const v_uint8x16& a)
 {
+#if CV_NEON_AARCH64
+    uint16_t t0 = vaddlvq_u8(a.val);
+    return t0;
+#else // #if CV_NEON_AARCH64
     uint32x4_t t0 = vpaddlq_u16(vpaddlq_u8(a.val));
     uint32x2_t t1 = vpadd_u32(vget_low_u32(t0), vget_high_u32(t0));
     return vget_lane_u32(vpadd_u32(t1, t1), 0);
+#endif // #if CV_NEON_AARCH64
 }
 inline int v_reduce_sum(const v_int8x16& a)
 {
+#if CV_NEON_AARCH64
+    int16_t t0 = vaddlvq_s8(a.val);
+    return t0;
+#else // #if CV_NEON_AARCH64
     int32x4_t t0 = vpaddlq_s16(vpaddlq_s8(a.val));
     int32x2_t t1 = vpadd_s32(vget_low_s32(t0), vget_high_s32(t0));
     return vget_lane_s32(vpadd_s32(t1, t1), 0);
+#endif // #if CV_NEON_AARCH64
 }
 inline unsigned v_reduce_sum(const v_uint16x8& a)
 {
+#if CV_NEON_AARCH64
+    uint32_t t0 = vaddlvq_u16(a.val);
+    return t0;
+#else // #if CV_NEON_AARCH64
     uint32x4_t t0 = vpaddlq_u16(a.val);
     uint32x2_t t1 = vpadd_u32(vget_low_u32(t0), vget_high_u32(t0));
     return vget_lane_u32(vpadd_u32(t1, t1), 0);
+#endif // #if CV_NEON_AARCH64
 }
 inline int v_reduce_sum(const v_int16x8& a)
 {
+#if CV_NEON_AARCH64
+    int32_t t0 = vaddlvq_s16(a.val);
+    return t0;
+#else // #if CV_NEON_AARCH64
     int32x4_t t0 = vpaddlq_s16(a.val);
     int32x2_t t1 = vpadd_s32(vget_low_s32(t0), vget_high_s32(t0));
     return vget_lane_s32(vpadd_s32(t1, t1), 0);
+#endif // #if CV_NEON_AARCH64
 }
 
+#if CV_NEON_AARCH64
+#define OPENCV_HAL_IMPL_NEON_REDUCE_OP_16(_Tpvec, _Tpnvec, scalartype, func, vectorfunc, suffix) \
+inline scalartype v_reduce_##func(const _Tpvec& a) \
+{ \
+    return v##vectorfunc##vq_##suffix(a.val); \
+}
+#else // #if CV_NEON_AARCH64
 #define OPENCV_HAL_IMPL_NEON_REDUCE_OP_16(_Tpvec, _Tpnvec, scalartype, func, vectorfunc, suffix) \
 inline scalartype v_reduce_##func(const _Tpvec& a) \
 { \
@@ -1285,12 +1344,20 @@ inline scalartype v_reduce_##func(const _Tpvec& a) \
     a0 = vp##vectorfunc##_##suffix(a0, a0); \
     return (scalartype)vget_lane_##suffix(vp##vectorfunc##_##suffix(a0, a0),0); \
 }
+#endif // #if CV_NEON_AARCH64
 
 OPENCV_HAL_IMPL_NEON_REDUCE_OP_16(v_uint8x16, uint8x8, uchar, max, max, u8)
 OPENCV_HAL_IMPL_NEON_REDUCE_OP_16(v_uint8x16, uint8x8, uchar, min, min, u8)
 OPENCV_HAL_IMPL_NEON_REDUCE_OP_16(v_int8x16, int8x8, schar, max, max, s8)
 OPENCV_HAL_IMPL_NEON_REDUCE_OP_16(v_int8x16, int8x8, schar, min, min, s8)
 
+#if CV_NEON_AARCH64
+#define OPENCV_HAL_IMPL_NEON_REDUCE_OP_8(_Tpvec, _Tpnvec, scalartype, func, vectorfunc, suffix) \
+inline scalartype v_reduce_##func(const _Tpvec& a) \
+{ \
+    return v##vectorfunc##vq_##suffix(a.val); \
+}
+#else // #if CV_NEON_AARCH64
 #define OPENCV_HAL_IMPL_NEON_REDUCE_OP_8(_Tpvec, _Tpnvec, scalartype, func, vectorfunc, suffix) \
 inline scalartype v_reduce_##func(const _Tpvec& a) \
 { \
@@ -1298,18 +1365,27 @@ inline scalartype v_reduce_##func(const _Tpvec& a) \
     a0 = vp##vectorfunc##_##suffix(a0, a0); \
     return (scalartype)vget_lane_##suffix(vp##vectorfunc##_##suffix(a0, a0),0); \
 }
+#endif // #if CV_NEON_AARCH64
 
 OPENCV_HAL_IMPL_NEON_REDUCE_OP_8(v_uint16x8, uint16x4, ushort, max, max, u16)
 OPENCV_HAL_IMPL_NEON_REDUCE_OP_8(v_uint16x8, uint16x4, ushort, min, min, u16)
 OPENCV_HAL_IMPL_NEON_REDUCE_OP_8(v_int16x8, int16x4, short, max, max, s16)
 OPENCV_HAL_IMPL_NEON_REDUCE_OP_8(v_int16x8, int16x4, short, min, min, s16)
 
+#if CV_NEON_AARCH64
+#define OPENCV_HAL_IMPL_NEON_REDUCE_OP_4(_Tpvec, _Tpnvec, scalartype, func, vectorfunc, suffix) \
+inline scalartype v_reduce_##func(const _Tpvec& a) \
+{ \
+    return v##vectorfunc##vq_##suffix(a.val); \
+}
+#else // #if CV_NEON_AARCH64
 #define OPENCV_HAL_IMPL_NEON_REDUCE_OP_4(_Tpvec, _Tpnvec, scalartype, func, vectorfunc, suffix) \
 inline scalartype v_reduce_##func(const _Tpvec& a) \
 { \
     _Tpnvec##_t a0 = vp##vectorfunc##_##suffix(vget_low_##suffix(a.val), vget_high_##suffix(a.val)); \
     return (scalartype)vget_lane_##suffix(vp##vectorfunc##_##suffix(a0, vget_high_##suffix(a.val)),0); \
 }
+#endif // #if CV_NEON_AARCH64
 
 OPENCV_HAL_IMPL_NEON_REDUCE_OP_4(v_uint32x4, uint32x2, unsigned, sum, add, u32)
 OPENCV_HAL_IMPL_NEON_REDUCE_OP_4(v_uint32x4, uint32x2, unsigned, max, max, u32)
@@ -1322,9 +1398,21 @@ OPENCV_HAL_IMPL_NEON_REDUCE_OP_4(v_float32x4, float32x2, float, max, max, f32)
 OPENCV_HAL_IMPL_NEON_REDUCE_OP_4(v_float32x4, float32x2, float, min, min, f32)
 
 inline uint64 v_reduce_sum(const v_uint64x2& a)
-{ return vget_lane_u64(vadd_u64(vget_low_u64(a.val), vget_high_u64(a.val)),0); }
+{
+#if CV_NEON_AARCH64
+    return vaddvq_u64(a.val);
+#else // #if CV_NEON_AARCH64
+    return vget_lane_u64(vadd_u64(vget_low_u64(a.val), vget_high_u64(a.val)),0);
+#endif // #if CV_NEON_AARCH64
+}
 inline int64 v_reduce_sum(const v_int64x2& a)
-{ return vget_lane_s64(vadd_s64(vget_low_s64(a.val), vget_high_s64(a.val)),0); }
+{
+#if CV_NEON_AARCH64
+    return vaddvq_s64(a.val);
+#else // #if CV_NEON_AARCH64
+    return vget_lane_s64(vadd_s64(vget_low_s64(a.val), vget_high_s64(a.val)),0);
+#endif // #if CV_NEON_AARCH64
+}
 #if CV_SIMD128_64F
 inline double v_reduce_sum(const v_float64x2& a)
 {
@@ -1335,6 +1423,11 @@ inline double v_reduce_sum(const v_float64x2& a)
 inline v_float32x4 v_reduce_sum4(const v_float32x4& a, const v_float32x4& b,
                                  const v_float32x4& c, const v_float32x4& d)
 {
+#if CV_NEON_AARCH64
+    float32x4_t ab = vpaddq_f32(a.val, b.val); // a0+a1 a2+a3 b0+b1 b2+b3
+    float32x4_t cd = vpaddq_f32(c.val, d.val); // c0+c1 d0+d1 c2+c3 d2+d3
+    return v_float32x4(vpaddq_f32(ab, cd));  // sumA sumB sumC sumD
+#else // #if CV_NEON_AARCH64
     float32x4x2_t ab = vtrnq_f32(a.val, b.val);
     float32x4x2_t cd = vtrnq_f32(c.val, d.val);
 
@@ -1345,49 +1438,91 @@ inline v_float32x4 v_reduce_sum4(const v_float32x4& a, const v_float32x4& b,
     float32x4_t v1 = vcombine_f32(vget_high_f32(u0), vget_high_f32(u1));
 
     return v_float32x4(vaddq_f32(v0, v1));
+#endif // #if CV_NEON_AARCH64
 }
 
 inline unsigned v_reduce_sad(const v_uint8x16& a, const v_uint8x16& b)
 {
+#if CV_NEON_AARCH64
+    uint8x16_t t0 = vabdq_u8(a.val, b.val);
+    uint16_t t1 = vaddlvq_u8(t0);
+    return t1;
+#else // #if CV_NEON_AARCH64
     uint32x4_t t0 = vpaddlq_u16(vpaddlq_u8(vabdq_u8(a.val, b.val)));
     uint32x2_t t1 = vpadd_u32(vget_low_u32(t0), vget_high_u32(t0));
     return vget_lane_u32(vpadd_u32(t1, t1), 0);
+#endif // #if CV_NEON_AARCH64
 }
 inline unsigned v_reduce_sad(const v_int8x16& a, const v_int8x16& b)
 {
+#if CV_NEON_AARCH64
+    uint8x16_t t0 = vreinterpretq_u8_s8(vabdq_s8(a.val, b.val));
+    uint16_t t1 = vaddlvq_u8(t0);
+    return t1;
+#else // #if CV_NEON_AARCH64
     uint32x4_t t0 = vpaddlq_u16(vpaddlq_u8(vreinterpretq_u8_s8(vabdq_s8(a.val, b.val))));
     uint32x2_t t1 = vpadd_u32(vget_low_u32(t0), vget_high_u32(t0));
     return vget_lane_u32(vpadd_u32(t1, t1), 0);
+#endif // #if CV_NEON_AARCH64
 }
 inline unsigned v_reduce_sad(const v_uint16x8& a, const v_uint16x8& b)
 {
+#if CV_NEON_AARCH64
+    uint16x8_t t0 = vabdq_u16(a.val, b.val);
+    uint32_t t1 = vaddlvq_u16(t0);
+    return t1;
+#else // #if CV_NEON_AARCH64
     uint32x4_t t0 = vpaddlq_u16(vabdq_u16(a.val, b.val));
     uint32x2_t t1 = vpadd_u32(vget_low_u32(t0), vget_high_u32(t0));
     return vget_lane_u32(vpadd_u32(t1, t1), 0);
+#endif // #if CV_NEON_AARCH64
 }
 inline unsigned v_reduce_sad(const v_int16x8& a, const v_int16x8& b)
 {
+#if CV_NEON_AARCH64
+    uint16x8_t t0 = vreinterpretq_u16_s16(vabdq_s16(a.val, b.val));
+    uint32_t t1 = vaddlvq_u16(t0);
+    return t1;
+#else // #if CV_NEON_AARCH64
     uint32x4_t t0 = vpaddlq_u16(vreinterpretq_u16_s16(vabdq_s16(a.val, b.val)));
     uint32x2_t t1 = vpadd_u32(vget_low_u32(t0), vget_high_u32(t0));
     return vget_lane_u32(vpadd_u32(t1, t1), 0);
+#endif // #if CV_NEON_AARCH64
 }
 inline unsigned v_reduce_sad(const v_uint32x4& a, const v_uint32x4& b)
 {
+#if CV_NEON_AARCH64
+    uint32x4_t t0 = vabdq_u32(a.val, b.val);
+    uint32_t t1 = vaddvq_u32(t0);
+    return t1;
+#else // #if CV_NEON_AARCH64
     uint32x4_t t0 = vabdq_u32(a.val, b.val);
     uint32x2_t t1 = vpadd_u32(vget_low_u32(t0), vget_high_u32(t0));
     return vget_lane_u32(vpadd_u32(t1, t1), 0);
+#endif // #if CV_NEON_AARCH64
 }
 inline unsigned v_reduce_sad(const v_int32x4& a, const v_int32x4& b)
 {
+#if CV_NEON_AARCH64
+    uint32x4_t t0 = vreinterpretq_u32_s32(vabdq_s32(a.val, b.val));
+    uint32_t t1 = vaddvq_u32(t0);
+    return t1;
+#else // #if CV_NEON_AARCH64
     uint32x4_t t0 = vreinterpretq_u32_s32(vabdq_s32(a.val, b.val));
     uint32x2_t t1 = vpadd_u32(vget_low_u32(t0), vget_high_u32(t0));
     return vget_lane_u32(vpadd_u32(t1, t1), 0);
+#endif // #if CV_NEON_AARCH64
 }
 inline float v_reduce_sad(const v_float32x4& a, const v_float32x4& b)
 {
+#if CV_NEON_AARCH64
+    float32x4_t t0 = vabdq_f32(a.val, b.val);
+    return vaddvq_f32(t0);
+#else // #if CV_NEON_AARCH64
     float32x4_t t0 = vabdq_f32(a.val, b.val);
     float32x2_t t1 = vpadd_f32(vget_low_f32(t0), vget_high_f32(t0));
     return vget_lane_f32(vpadd_f32(t1, t1), 0);
+#endif // #if CV_NEON_AARCH64
 }
 
 inline v_uint8x16 v_popcount(const v_uint8x16& a)
@@ -1409,30 +1544,54 @@ inline v_uint64x2 v_popcount(const v_int64x2& a)
 
 inline int v_signmask(const v_uint8x16& a)
 {
+#if CV_NEON_AARCH64
+    const int8x16_t signPosition = {0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7};
+    const uint8x16_t byteOrder = {0,8,1,9,2,10,3,11,4,12,5,13,6,14,7,15};
+    uint8x16_t v0 = vshlq_u8(vshrq_n_u8(a.val, 7), signPosition);
+    uint8x16_t v1 = vqtbl1q_u8(v0, byteOrder);
+    uint32_t t0 = vaddlvq_u16(vreinterpretq_u16_u8(v1));
+    return t0;
+#else // #if CV_NEON_AARCH64
     int8x8_t m0 = vcreate_s8(CV_BIG_UINT(0x0706050403020100));
     uint8x16_t v0 = vshlq_u8(vshrq_n_u8(a.val, 7), vcombine_s8(m0, m0));
     uint64x2_t v1 = vpaddlq_u32(vpaddlq_u16(vpaddlq_u8(v0)));
     return (int)vgetq_lane_u64(v1, 0) + ((int)vgetq_lane_u64(v1, 1) << 8);
+#endif // #if CV_NEON_AARCH64
 }
+
 inline int v_signmask(const v_int8x16& a)
 { return v_signmask(v_reinterpret_as_u8(a)); }
 
 inline int v_signmask(const v_uint16x8& a)
 {
+#if CV_NEON_AARCH64
+    const int16x8_t signPosition = {0,1,2,3,4,5,6,7};
+    uint16x8_t v0 = vshlq_u16(vshrq_n_u16(a.val, 15), signPosition);
+    uint32_t t0 = vaddlvq_u16(v0);
+    return t0;
+#else // #if CV_NEON_AARCH64
     int16x4_t m0 = vcreate_s16(CV_BIG_UINT(0x0003000200010000));
     uint16x8_t v0 = vshlq_u16(vshrq_n_u16(a.val, 15), vcombine_s16(m0, m0));
     uint64x2_t v1 = vpaddlq_u32(vpaddlq_u16(v0));
     return (int)vgetq_lane_u64(v1, 0) + ((int)vgetq_lane_u64(v1, 1) << 4);
+#endif // #if CV_NEON_AARCH64
 }
 inline int v_signmask(const v_int16x8& a)
 { return v_signmask(v_reinterpret_as_u16(a)); }
 
 inline int v_signmask(const v_uint32x4& a)
 {
+#if CV_NEON_AARCH64
+    const int32x4_t signPosition = {0,1,2,3};
+    uint32x4_t v0 = vshlq_u32(vshrq_n_u32(a.val, 31), signPosition);
+    uint32_t t0 = vaddvq_u32(v0);
+    return t0;
+#else // #if CV_NEON_AARCH64
     int32x2_t m0 = vcreate_s32(CV_BIG_UINT(0x0000000100000000));
     uint32x4_t v0 = vshlq_u32(vshrq_n_u32(a.val, 31), vcombine_s32(m0, m0));
     uint64x2_t v1 = vpaddlq_u32(v0);
     return (int)vgetq_lane_u64(v1, 0) + ((int)vgetq_lane_u64(v1, 1) << 2);
+#endif // #if CV_NEON_AARCH64
 }
 inline int v_signmask(const v_int32x4& a)
 { return v_signmask(v_reinterpret_as_u32(a)); }
@@ -1440,9 +1599,16 @@ inline int v_signmask(const v_float32x4& a)
 { return v_signmask(v_reinterpret_as_u32(a)); }
 inline int v_signmask(const v_uint64x2& a)
 {
+#if CV_NEON_AARCH64
+    const int64x2_t signPosition = {0,1};
+    uint64x2_t v0 = vshlq_u64(vshrq_n_u64(a.val, 63), signPosition);
+    uint64_t t0 = vaddvq_u64(v0);
+    return t0;
+#else // #if CV_NEON_AARCH64
     int64x1_t m0 = vdup_n_s64(0);
     uint64x2_t v0 = vshlq_u64(vshrq_n_u64(a.val, 63), vcombine_s64(m0, m0));
     return (int)vgetq_lane_u64(v0, 0) + ((int)vgetq_lane_u64(v0, 1) << 1);
+#endif // #if CV_NEON_AARCH64
 }
 inline int v_signmask(const v_int64x2& a)
 { return v_signmask(v_reinterpret_as_u64(a)); }
@@ -1464,19 +1630,31 @@ inline int v_scan_forward(const v_uint64x2& a) { return trailingZeros32(v_signma
 inline int v_scan_forward(const v_float64x2& a) { return trailingZeros32(v_signmask(a)); }
 #endif
 
-#define OPENCV_HAL_IMPL_NEON_CHECK_ALLANY(_Tpvec, suffix, shift) \
-inline bool v_check_all(const v_##_Tpvec& a) \
-{ \
-    _Tpvec##_t v0 = vshrq_n_##suffix(vmvnq_##suffix(a.val), shift); \
-    uint64x2_t v1 = vreinterpretq_u64_##suffix(v0); \
-    return (vgetq_lane_u64(v1, 0) | vgetq_lane_u64(v1, 1)) == 0; \
-} \
-inline bool v_check_any(const v_##_Tpvec& a) \
-{ \
-    _Tpvec##_t v0 = vshrq_n_##suffix(a.val, shift); \
-    uint64x2_t v1 = vreinterpretq_u64_##suffix(v0); \
-    return (vgetq_lane_u64(v1, 0) | vgetq_lane_u64(v1, 1)) != 0; \
-}
+#if CV_NEON_AARCH64
+    #define OPENCV_HAL_IMPL_NEON_CHECK_ALLANY(_Tpvec, suffix, shift) \
+    inline bool v_check_all(const v_##_Tpvec& a) \
+    { \
+        return (vminvq_##suffix(a.val) >> shift) != 0; \
+    } \
+    inline bool v_check_any(const v_##_Tpvec& a) \
+    { \
+        return (vmaxvq_##suffix(a.val) >> shift) != 0; \
+    }
+#else // #if CV_NEON_AARCH64
+    #define OPENCV_HAL_IMPL_NEON_CHECK_ALLANY(_Tpvec, suffix, shift) \
+    inline bool v_check_all(const v_##_Tpvec& a) \
+    { \
+        _Tpvec##_t v0 = vshrq_n_##suffix(vmvnq_##suffix(a.val), shift); \
+        uint64x2_t v1 = vreinterpretq_u64_##suffix(v0); \
+        return (vgetq_lane_u64(v1, 0) | vgetq_lane_u64(v1, 1)) == 0; \
+    } \
+    inline bool v_check_any(const v_##_Tpvec& a) \
+    { \
+        _Tpvec##_t v0 = vshrq_n_##suffix(a.val, shift); \
+        uint64x2_t v1 = vreinterpretq_u64_##suffix(v0); \
+        return (vgetq_lane_u64(v1, 0) | vgetq_lane_u64(v1, 1)) != 0; \
+    }
+#endif // #if CV_NEON_AARCH64
 
 OPENCV_HAL_IMPL_NEON_CHECK_ALLANY(uint8x16, u8, 7)
 OPENCV_HAL_IMPL_NEON_CHECK_ALLANY(uint16x8, u16, 15)
@@ -1539,6 +1717,26 @@ OPENCV_HAL_IMPL_NEON_SELECT(v_float32x4, f32, u32)
 OPENCV_HAL_IMPL_NEON_SELECT(v_float64x2, f64, u64)
 #endif
 
+#if CV_NEON_AARCH64
+#define OPENCV_HAL_IMPL_NEON_EXPAND(_Tpvec, _Tpwvec, _Tp, suffix) \
+inline void v_expand(const _Tpvec& a, _Tpwvec& b0, _Tpwvec& b1) \
+{ \
+    b0.val = vmovl_##suffix(vget_low_##suffix(a.val)); \
+    b1.val = vmovl_high_##suffix(a.val); \
+} \
+inline _Tpwvec v_expand_low(const _Tpvec& a) \
+{ \
+    return _Tpwvec(vmovl_##suffix(vget_low_##suffix(a.val))); \
+} \
+inline _Tpwvec v_expand_high(const _Tpvec& a) \
+{ \
+    return _Tpwvec(vmovl_high_##suffix(a.val)); \
+} \
+inline _Tpwvec v_load_expand(const _Tp* ptr) \
+{ \
+    return _Tpwvec(vmovl_##suffix(vld1_##suffix(ptr))); \
+}
+#else
 #define OPENCV_HAL_IMPL_NEON_EXPAND(_Tpvec, _Tpwvec, _Tp, suffix) \
 inline void v_expand(const _Tpvec& a, _Tpwvec& b0, _Tpwvec& b1) \
 { \
@@ -1557,6 +1755,7 @@ inline _Tpwvec v_load_expand(const _Tp* ptr) \
 { \
     return _Tpwvec(vmovl_##suffix(vld1_##suffix(ptr))); \
 }
+#endif
 
 OPENCV_HAL_IMPL_NEON_EXPAND(v_uint8x16, v_uint16x8, uchar, u8)
 OPENCV_HAL_IMPL_NEON_EXPAND(v_int8x16, v_int16x8, schar, s8)
@@ -1808,6 +2007,37 @@ inline v_int32x4 v_trunc(const v_float64x2& a)
 }
 #endif
 
+#if CV_NEON_AARCH64
+#define OPENCV_HAL_IMPL_NEON_TRANSPOSE4x4(_Tpvec, suffix) \
+inline void v_transpose4x4(const v_##_Tpvec& a0, const v_##_Tpvec& a1, \
+                         const v_##_Tpvec& a2, const v_##_Tpvec& a3, \
+                         v_##_Tpvec& b0, v_##_Tpvec& b1, \
+                         v_##_Tpvec& b2, v_##_Tpvec& b3) \
+{ \
+    /* -- Pass 1: 64b transpose */ \
+    _Tpvec##_t t0 = vreinterpretq_##suffix##32_##suffix##64( \
+                        vtrn1q_##suffix##64(vreinterpretq_##suffix##64_##suffix##32(a0.val), \
+                                            vreinterpretq_##suffix##64_##suffix##32(a2.val))); \
+    _Tpvec##_t t1 = vreinterpretq_##suffix##32_##suffix##64( \
+                        vtrn1q_##suffix##64(vreinterpretq_##suffix##64_##suffix##32(a1.val), \
+                                            vreinterpretq_##suffix##64_##suffix##32(a3.val))); \
+    _Tpvec##_t t2 = vreinterpretq_##suffix##32_##suffix##64( \
+                        vtrn2q_##suffix##64(vreinterpretq_##suffix##64_##suffix##32(a0.val), \
+                                            vreinterpretq_##suffix##64_##suffix##32(a2.val))); \
+    _Tpvec##_t t3 = vreinterpretq_##suffix##32_##suffix##64( \
+                        vtrn2q_##suffix##64(vreinterpretq_##suffix##64_##suffix##32(a1.val), \
+                                            vreinterpretq_##suffix##64_##suffix##32(a3.val))); \
+    /* -- Pass 2: 32b transpose */ \
+    b0.val = vtrn1q_##suffix##32(t0, t1); \
+    b1.val = vtrn2q_##suffix##32(t0, t1); \
+    b2.val = vtrn1q_##suffix##32(t2, t3); \
+    b3.val = vtrn2q_##suffix##32(t2, t3); \
+}
+
+OPENCV_HAL_IMPL_NEON_TRANSPOSE4x4(uint32x4, u)
+OPENCV_HAL_IMPL_NEON_TRANSPOSE4x4(int32x4, s)
+OPENCV_HAL_IMPL_NEON_TRANSPOSE4x4(float32x4, f)
+#else // #if CV_NEON_AARCH64
 #define OPENCV_HAL_IMPL_NEON_TRANSPOSE4x4(_Tpvec, suffix) \
 inline void v_transpose4x4(const v_##_Tpvec& a0, const v_##_Tpvec& a1, \
                          const v_##_Tpvec& a2, const v_##_Tpvec& a3, \
@@ -1833,6 +2063,7 @@ inline void v_transpose4x4(const v_##_Tpvec& a0, const v_##_Tpvec& a1, \
 OPENCV_HAL_IMPL_NEON_TRANSPOSE4x4(uint32x4, u32)
 OPENCV_HAL_IMPL_NEON_TRANSPOSE4x4(int32x4, s32)
 OPENCV_HAL_IMPL_NEON_TRANSPOSE4x4(float32x4, f32)
+#endif // #if CV_NEON_AARCH64
 
 #define OPENCV_HAL_IMPL_NEON_INTERLEAVED(_Tpvec, _Tp, suffix) \
 inline void v_load_deinterleave(const _Tp* ptr, v_##_Tpvec& a, v_##_Tpvec& b) \

modules/core/include/opencv2/core/mat.hpp

@@ -583,24 +583,24 @@ struct CV_EXPORTS UMatData
 
 struct CV_EXPORTS MatSize
 {
-    explicit MatSize(int* _p);
-    int dims() const;
+    explicit MatSize(int* _p) CV_NOEXCEPT;
+    int dims() const CV_NOEXCEPT;
     Size operator()() const;
     const int& operator[](int i) const;
     int& operator[](int i);
-    operator const int*() const;  // TODO OpenCV 4.0: drop this
-    bool operator == (const MatSize& sz) const;
-    bool operator != (const MatSize& sz) const;
+    operator const int*() const CV_NOEXCEPT;  // TODO OpenCV 4.0: drop this
+    bool operator == (const MatSize& sz) const CV_NOEXCEPT;
+    bool operator != (const MatSize& sz) const CV_NOEXCEPT;
 
     int* p;
 };
 
 struct CV_EXPORTS MatStep
 {
-    MatStep();
-    explicit MatStep(size_t s);
-    const size_t& operator[](int i) const;
-    size_t& operator[](int i);
+    MatStep() CV_NOEXCEPT;
+    explicit MatStep(size_t s) CV_NOEXCEPT;
+    const size_t& operator[](int i) const CV_NOEXCEPT;
+    size_t& operator[](int i) CV_NOEXCEPT;
     operator size_t() const;
     MatStep& operator = (size_t s);
 
@@ -819,7 +819,7 @@ public:
     The constructed matrix can further be assigned to another matrix or matrix expression or can be
     allocated with Mat::create . In the former case, the old content is de-referenced.
      */
-    Mat();
+    Mat() CV_NOEXCEPT;
 
     /** @overload
     @param rows Number of rows in a 2D array.
@@ -2220,7 +2220,7 @@ public:
     typedef MatConstIterator_<_Tp> const_iterator;
 
     //! default constructor
-    Mat_();
+    Mat_() CV_NOEXCEPT;
     //! equivalent to Mat(_rows, _cols, DataType<_Tp>::type)
     Mat_(int _rows, int _cols);
     //! constructor that sets each matrix element to specified value
@@ -2420,7 +2420,7 @@ class CV_EXPORTS UMat
 {
 public:
     //! default constructor
-    UMat(UMatUsageFlags usageFlags = USAGE_DEFAULT);
+    UMat(UMatUsageFlags usageFlags = USAGE_DEFAULT) CV_NOEXCEPT;
     //! constructs 2D matrix of the specified size and type
     // (_type is CV_8UC1, CV_64FC3, CV_32SC(12) etc.)
     UMat(int rows, int cols, int type, UMatUsageFlags usageFlags = USAGE_DEFAULT);

modules/core/include/opencv2/core/mat.inl.hpp

@@ -898,6 +898,33 @@ const _Tp* Mat::ptr(const int* idx) const
     return (const _Tp*)p;
 }
 
+template<int n> inline
+uchar* Mat::ptr(const Vec<int, n>& idx)
+{
+    return Mat::ptr(idx.val);
+}
+
+template<int n> inline
+const uchar* Mat::ptr(const Vec<int, n>& idx) const
+{
+    return Mat::ptr(idx.val);
+}
+
+template<typename _Tp, int n> inline
+_Tp* Mat::ptr(const Vec<int, n>& idx)
+{
+    CV_DbgAssert( elemSize() == sizeof(_Tp) );
+    return Mat::ptr<_Tp>(idx.val);
+}
+
+template<typename _Tp, int n> inline
+const _Tp* Mat::ptr(const Vec<int, n>& idx) const
+{
+    CV_DbgAssert( elemSize() == sizeof(_Tp) );
+    return Mat::ptr<_Tp>(idx.val);
+}
+
+
 template<typename _Tp> inline
 _Tp& Mat::at(int i0, int i1)
 {
@@ -1218,11 +1245,11 @@ Mat& Mat::operator = (Mat&& m)
 ///////////////////////////// MatSize ////////////////////////////
 
 inline
-MatSize::MatSize(int* _p)
+MatSize::MatSize(int* _p) CV_NOEXCEPT
     : p(_p) {}
 
 inline
-int MatSize::dims() const
+int MatSize::dims() const CV_NOEXCEPT
 {
     return (p - 1)[0];
 }
@@ -1255,13 +1282,13 @@ int& MatSize::operator[](int i)
 }
 
 inline
-MatSize::operator const int*() const
+MatSize::operator const int*() const CV_NOEXCEPT
 {
     return p;
 }
 
 inline
-bool MatSize::operator != (const MatSize& sz) const
+bool MatSize::operator != (const MatSize& sz) const CV_NOEXCEPT
 {
     return !(*this == sz);
 }
@@ -1271,25 +1298,25 @@ bool MatSize::operator != (const MatSize& sz) const
 ///////////////////////////// MatStep ////////////////////////////
 
 inline
-MatStep::MatStep()
+MatStep::MatStep() CV_NOEXCEPT
 {
     p = buf; p[0] = p[1] = 0;
 }
 
 inline
-MatStep::MatStep(size_t s)
+MatStep::MatStep(size_t s) CV_NOEXCEPT
 {
     p = buf; p[0] = s; p[1] = 0;
 }
 
 inline
-const size_t& MatStep::operator[](int i) const
+const size_t& MatStep::operator[](int i) const CV_NOEXCEPT
 {
     return p[i];
 }
 
 inline
-size_t& MatStep::operator[](int i)
+size_t& MatStep::operator[](int i) CV_NOEXCEPT
 {
     return p[i];
 }
@@ -1312,7 +1339,7 @@ inline MatStep& MatStep::operator = (size_t s)
 ////////////////////////////// Mat_<_Tp> ////////////////////////////
 
 template<typename _Tp> inline
-Mat_<_Tp>::Mat_()
+Mat_<_Tp>::Mat_() CV_NOEXCEPT
     : Mat()
 {
     flags = (flags & ~CV_MAT_TYPE_MASK) | traits::Type<_Tp>::value;

modules/core/include/opencv2/core/types.hpp

@@ -702,24 +702,24 @@ public:
     //! the default constructor
     CV_WRAP KeyPoint();
     /**
-    @param _pt x & y coordinates of the keypoint
-    @param _size keypoint diameter
-    @param _angle keypoint orientation
-    @param _response keypoint detector response on the keypoint (that is, strength of the keypoint)
-    @param _octave pyramid octave in which the keypoint has been detected
-    @param _class_id object id
+    @param pt x & y coordinates of the keypoint
+    @param size keypoint diameter
+    @param angle keypoint orientation
+    @param response keypoint detector response on the keypoint (that is, strength of the keypoint)
+    @param octave pyramid octave in which the keypoint has been detected
+    @param class_id object id
      */
-    KeyPoint(Point2f _pt, float _size, float _angle=-1, float _response=0, int _octave=0, int _class_id=-1);
+    KeyPoint(Point2f pt, float size, float angle=-1, float response=0, int octave=0, int class_id=-1);
     /**
     @param x x-coordinate of the keypoint
     @param y y-coordinate of the keypoint
-    @param _size keypoint diameter
-    @param _angle keypoint orientation
-    @param _response keypoint detector response on the keypoint (that is, strength of the keypoint)
-    @param _octave pyramid octave in which the keypoint has been detected
-    @param _class_id object id
+    @param size keypoint diameter
+    @param angle keypoint orientation
+    @param response keypoint detector response on the keypoint (that is, strength of the keypoint)
+    @param octave pyramid octave in which the keypoint has been detected
+    @param class_id object id
      */
-    CV_WRAP KeyPoint(float x, float y, float _size, float _angle=-1, float _response=0, int _octave=0, int _class_id=-1);
+    CV_WRAP KeyPoint(float x, float y, float size, float angle=-1, float response=0, int octave=0, int class_id=-1);
 
     size_t hash() const;
 

modules/core/include/opencv2/core/types_c.h

@@ -358,7 +358,11 @@ _IplImage
                                needed for correct deallocation */
 
 #if defined(CV__ENABLE_C_API_CTORS) && defined(__cplusplus)
-    _IplImage() {}
+    _IplImage()
+    {
+        memset(this, 0, sizeof(*this));  // valid for POD structure
+        nSize = sizeof(IplImage);
+    }
     _IplImage(const cv::Mat& m) { *this = cvIplImage(m); }
 #endif
 }

modules/core/include/opencv2/core/utils/filesystem.private.hpp

@@ -16,8 +16,8 @@
 #      define OPENCV_HAVE_FILESYSTEM_SUPPORT 1
 #  elif defined(__APPLE__)
 #    include <TargetConditionals.h>
-#    if (defined(TARGET_OS_OSX) && TARGET_OS_OSX) || (!defined(TARGET_OS_OSX) && !TARGET_OS_IPHONE)
-#      define OPENCV_HAVE_FILESYSTEM_SUPPORT 1 // OSX only
+#    if (defined(TARGET_OS_OSX) && TARGET_OS_OSX) || (defined(TARGET_OS_IOS) && TARGET_OS_IOS)
+#      define OPENCV_HAVE_FILESYSTEM_SUPPORT 1 // OSX, iOS only
 #    endif
 #  else
      /* unknown */

modules/core/include/opencv2/core/version.hpp

@@ -7,8 +7,8 @@
 
 #define CV_VERSION_MAJOR    3
 #define CV_VERSION_MINOR    4
-#define CV_VERSION_REVISION 14
-#define CV_VERSION_STATUS   "-pre"
+#define CV_VERSION_REVISION 15
+#define CV_VERSION_STATUS   ""
 
 #define CVAUX_STR_EXP(__A)  #__A
 #define CVAUX_STR(__A)      CVAUX_STR_EXP(__A)

modules/core/include/opencv2/core/vsx_utils.hpp

@@ -497,11 +497,13 @@ VSX_IMPL_CONV_EVEN_2_4(vec_uint4,  vec_double2, vec_ctu, vec_ctuo)
     VSX_FINLINE(rt) fnm(const rg& a, int only_truncate) \
     {                                                   \
         assert(only_truncate == 0);                     \
-        CV_UNUSED(only_truncate);                            \
+        CV_UNUSED(only_truncate);                       \
         return fn2(a);                                  \
     }
     VSX_IMPL_CONV_2VARIANT(vec_int4,   vec_float4,  vec_cts, vec_cts)
+    VSX_IMPL_CONV_2VARIANT(vec_uint4,  vec_float4,  vec_ctu, vec_ctu)
     VSX_IMPL_CONV_2VARIANT(vec_float4, vec_int4,    vec_ctf, vec_ctf)
+    VSX_IMPL_CONV_2VARIANT(vec_float4, vec_uint4,   vec_ctf, vec_ctf)
     // define vec_cts for converting double precision to signed doubleword
     // which isn't compatible with xlc but its okay since Eigen only uses it for gcc
     VSX_IMPL_CONV_2VARIANT(vec_dword2, vec_double2, vec_cts, vec_ctsl)

modules/core/perf/opencl/perf_arithm.cpp

@@ -678,7 +678,12 @@ OCL_PERF_TEST_P(SqrtFixture, Sqrt, ::testing::Combine(
 
     OCL_TEST_CYCLE() cv::sqrt(src, dst);
 
-    if (CV_MAT_DEPTH(type) >= CV_32F)
+    // To square root 32 bit floats we use native_sqrt, which has implementation
+    // defined accuracy. We know intel devices have accurate native_sqrt, but
+    // otherwise stick to a relaxed sanity check. For types larger than 32 bits
+    // we can do the accuracy check for all devices as normal.
+    if (CV_MAT_DEPTH(type) > CV_32F || !ocl::useOpenCL() ||
+        ocl::Device::getDefault().isIntel())
         SANITY_CHECK(dst, 1e-5, ERROR_RELATIVE);
     else
         SANITY_CHECK(dst, 1);

modules/core/src/arithm.cpp

@@ -57,24 +57,6 @@ namespace cv
 *                                   logical operations                                   *
 \****************************************************************************************/
 
-void convertAndUnrollScalar( const Mat& sc, int buftype, uchar* scbuf, size_t blocksize )
-{
-    int scn = (int)sc.total(), cn = CV_MAT_CN(buftype);
-    size_t esz = CV_ELEM_SIZE(buftype);
-    getConvertFunc(sc.depth(), buftype)(sc.ptr(), 1, 0, 1, scbuf, 1, Size(std::min(cn, scn), 1), 0);
-    // unroll the scalar
-    if( scn < cn )
-    {
-        CV_Assert( scn == 1 );
-        size_t esz1 = CV_ELEM_SIZE1(buftype);
-        for( size_t i = esz1; i < esz; i++ )
-            scbuf[i] = scbuf[i - esz1];
-    }
-    for( size_t i = esz; i < blocksize*esz; i++ )
-        scbuf[i] = scbuf[i - esz];
-}
-
-
 enum { OCL_OP_ADD=0, OCL_OP_SUB=1, OCL_OP_RSUB=2, OCL_OP_ABSDIFF=3, OCL_OP_MUL=4,
        OCL_OP_MUL_SCALE=5, OCL_OP_DIV_SCALE=6, OCL_OP_RECIP_SCALE=7, OCL_OP_ADDW=8,
        OCL_OP_AND=9, OCL_OP_OR=10, OCL_OP_XOR=11, OCL_OP_NOT=12, OCL_OP_MIN=13, OCL_OP_MAX=14,
@@ -641,7 +623,8 @@ static void arithm_op(InputArray _src1, InputArray _src2, OutputArray _dst,
         (kind1 == _InputArray::MATX && (sz1 == Size(1,4) || sz1 == Size(1,1))) ||
         (kind2 == _InputArray::MATX && (sz2 == Size(1,4) || sz2 == Size(1,1))) )
     {
-        if( checkScalar(*psrc1, type2, kind1, kind2) )
+        if ((type1 == CV_64F && (sz1.height == 1 || sz1.height == 4)) &&
+            checkScalar(*psrc1, type2, kind1, kind2))
         {
             // src1 is a scalar; swap it with src2
             swap(psrc1, psrc2);
@@ -1041,9 +1024,7 @@ static BinaryFuncC* getRecipTab()
     return recipTab;
 }
 
-}
-
-void cv::multiply(InputArray src1, InputArray src2,
+void multiply(InputArray src1, InputArray src2,
                   OutputArray dst, double scale, int dtype)
 {
     CV_INSTRUMENT_REGION();
@@ -1052,7 +1033,7 @@ void cv::multiply(InputArray src1, InputArray src2,
               true, &scale, std::abs(scale - 1.0) < DBL_EPSILON ? OCL_OP_MUL : OCL_OP_MUL_SCALE);
 }
 
-void cv::divide(InputArray src1, InputArray src2,
+void divide(InputArray src1, InputArray src2,
                 OutputArray dst, double scale, int dtype)
 {
     CV_INSTRUMENT_REGION();
@@ -1060,7 +1041,7 @@ void cv::divide(InputArray src1, InputArray src2,
     arithm_op(src1, src2, dst, noArray(), dtype, getDivTab(), true, &scale, OCL_OP_DIV_SCALE);
 }
 
-void cv::divide(double scale, InputArray src2,
+void divide(double scale, InputArray src2,
                 OutputArray dst, int dtype)
 {
     CV_INSTRUMENT_REGION();
@@ -1068,13 +1049,17 @@ void cv::divide(double scale, InputArray src2,
     arithm_op(src2, src2, dst, noArray(), dtype, getRecipTab(), true, &scale, OCL_OP_RECIP_SCALE);
 }
 
+UMat UMat::mul(InputArray m, double scale) const
+{
+    UMat dst;
+    multiply(*this, m, dst, scale);
+    return dst;
+}
+
 /****************************************************************************************\
 *                                      addWeighted                                       *
 \****************************************************************************************/
 
-namespace cv
-{
-
 static BinaryFuncC* getAddWeightedTab()
 {
     static BinaryFuncC addWeightedTab[] =
@@ -1879,6 +1864,9 @@ void cv::inRange(InputArray _src, InputArray _lowerb,
     }
 }
 
+
+#ifndef OPENCV_EXCLUDE_C_API
+
 /****************************************************************************************\
 *                                Earlier API: cvAdd etc.                                 *
 \****************************************************************************************/
@@ -2141,4 +2129,5 @@ cvMaxS( const void* srcarr1, double value, void* dstarr )
     cv::max( src1, value, dst );
 }
 
+#endif  // OPENCV_EXCLUDE_C_API
 /* End of file. */

modules/core/src/array.cpp

@@ -48,6 +48,8 @@
 
 #include "precomp.hpp"
 
+#ifndef OPENCV_EXCLUDE_C_API
+
 #define  CV_ORIGIN_TL  0
 #define  CV_ORIGIN_BL  1
 
@@ -3223,51 +3225,50 @@ template<> void DefaultDeleter<CvMemStorage>::operator ()(CvMemStorage* obj) con
 template<> void DefaultDeleter<CvFileStorage>::operator ()(CvFileStorage* obj) const
 { cvReleaseFileStorage(&obj); }
 
-template <typename T> static inline
-void scalarToRawData_(const Scalar& s, T * const buf, const int cn, const int unroll_to)
-{
-    int i = 0;
-    for(; i < cn; i++)
-        buf[i] = saturate_cast<T>(s.val[i]);
-    for(; i < unroll_to; i++)
-        buf[i] = buf[i-cn];
-}
-
-void scalarToRawData(const Scalar& s, void* _buf, int type, int unroll_to)
-{
-    CV_INSTRUMENT_REGION();
-
-    const int depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
-    CV_Assert(cn <= 4);
-    switch(depth)
-    {
-    case CV_8U:
-        scalarToRawData_<uchar>(s, (uchar*)_buf, cn, unroll_to);
-        break;
-    case CV_8S:
-        scalarToRawData_<schar>(s, (schar*)_buf, cn, unroll_to);
-        break;
-    case CV_16U:
-        scalarToRawData_<ushort>(s, (ushort*)_buf, cn, unroll_to);
-        break;
-    case CV_16S:
-        scalarToRawData_<short>(s, (short*)_buf, cn, unroll_to);
-        break;
-    case CV_32S:
-        scalarToRawData_<int>(s, (int*)_buf, cn, unroll_to);
-        break;
-    case CV_32F:
-        scalarToRawData_<float>(s, (float*)_buf, cn, unroll_to);
-        break;
-    case CV_64F:
-        scalarToRawData_<double>(s, (double*)_buf, cn, unroll_to);
-        break;
-    default:
-        CV_Error(CV_StsUnsupportedFormat,"");
-    }
-}
-
 } // cv::
 
 
+/* universal functions */
+CV_IMPL void
+cvRelease( void** struct_ptr )
+{
+    CvTypeInfo* info;
+
+    if( !struct_ptr )
+        CV_Error( CV_StsNullPtr, "NULL double pointer" );
+
+    if( *struct_ptr )
+    {
+        info = cvTypeOf( *struct_ptr );
+        if( !info )
+            CV_Error( CV_StsError, "Unknown object type" );
+        if( !info->release )
+            CV_Error( CV_StsError, "release function pointer is NULL" );
+
+        info->release( struct_ptr );
+        *struct_ptr = 0;
+    }
+}
+
+
+void* cvClone( const void* struct_ptr )
+{
+    void* struct_copy = 0;
+    CvTypeInfo* info;
+
+    if( !struct_ptr )
+        CV_Error( CV_StsNullPtr, "NULL structure pointer" );
+
+    info = cvTypeOf( struct_ptr );
+    if( !info )
+        CV_Error( CV_StsError, "Unknown object type" );
+    if( !info->clone )
+        CV_Error( CV_StsError, "clone function pointer is NULL" );
+
+    struct_copy = info->clone( struct_ptr );
+    return struct_copy;
+}
+
+
+#endif  // OPENCV_EXCLUDE_C_API
 /* End of file. */

modules/core/src/convert_c.cpp

@@ -5,6 +5,7 @@
 
 #include "precomp.hpp"
 
+#ifndef OPENCV_EXCLUDE_C_API
 
 CV_IMPL void
 cvSplit( const void* srcarr, void* dstarr0, void* dstarr1, void* dstarr2, void* dstarr3 )
@@ -132,3 +133,5 @@ CV_IMPL void cvNormalize( const CvArr* srcarr, CvArr* dstarr,
     CV_Assert( dst.size() == src.size() && src.channels() == dst.channels() );
     cv::normalize( src, dst, a, b, norm_type, dst.type(), mask );
 }
+
+#endif  // OPENCV_EXCLUDE_C_API

modules/core/src/convert_scale.dispatch.cpp

@@ -9,7 +9,6 @@
 #include "convert_scale.simd.hpp"
 #include "convert_scale.simd_declarations.hpp" // defines CV_CPU_DISPATCH_MODES_ALL=AVX2,...,BASELINE based on CMakeLists.txt content
 
-
 namespace cv
 {
 
@@ -117,143 +116,4 @@ void convertScaleAbs(InputArray _src, OutputArray _dst, double alpha, double bet
     }
 }
 
-//==================================================================================================
-
-#ifdef HAVE_OPENCL
-
-static bool ocl_normalize( InputArray _src, InputOutputArray _dst, InputArray _mask, int dtype,
-                           double scale, double delta )
-{
-    UMat src = _src.getUMat();
-
-    if( _mask.empty() )
-        src.convertTo( _dst, dtype, scale, delta );
-    else if (src.channels() <= 4)
-    {
-        const ocl::Device & dev = ocl::Device::getDefault();
-
-        int stype = _src.type(), sdepth = CV_MAT_DEPTH(stype), cn = CV_MAT_CN(stype),
-                ddepth = CV_MAT_DEPTH(dtype), wdepth = std::max(CV_32F, std::max(sdepth, ddepth)),
-                rowsPerWI = dev.isIntel() ? 4 : 1;
-
-        float fscale = static_cast<float>(scale), fdelta = static_cast<float>(delta);
-        bool haveScale = std::fabs(scale - 1) > DBL_EPSILON,
-                haveZeroScale = !(std::fabs(scale) > DBL_EPSILON),
-                haveDelta = std::fabs(delta) > DBL_EPSILON,
-                doubleSupport = dev.doubleFPConfig() > 0;
-
-        if (!haveScale && !haveDelta && stype == dtype)
-        {
-            _src.copyTo(_dst, _mask);
-            return true;
-        }
-        if (haveZeroScale)
-        {
-            _dst.setTo(Scalar(delta), _mask);
-            return true;
-        }
-
-        if ((sdepth == CV_64F || ddepth == CV_64F) && !doubleSupport)
-            return false;
-
-        char cvt[2][40];
-        String opts = format("-D srcT=%s -D dstT=%s -D convertToWT=%s -D cn=%d -D rowsPerWI=%d"
-                             " -D convertToDT=%s -D workT=%s%s%s%s -D srcT1=%s -D dstT1=%s",
-                             ocl::typeToStr(stype), ocl::typeToStr(dtype),
-                             ocl::convertTypeStr(sdepth, wdepth, cn, cvt[0]), cn,
-                             rowsPerWI, ocl::convertTypeStr(wdepth, ddepth, cn, cvt[1]),
-                             ocl::typeToStr(CV_MAKE_TYPE(wdepth, cn)),
-                             doubleSupport ? " -D DOUBLE_SUPPORT" : "",
-                             haveScale ? " -D HAVE_SCALE" : "",
-                             haveDelta ? " -D HAVE_DELTA" : "",
-                             ocl::typeToStr(sdepth), ocl::typeToStr(ddepth));
-
-        ocl::Kernel k("normalizek", ocl::core::normalize_oclsrc, opts);
-        if (k.empty())
-            return false;
-
-        UMat mask = _mask.getUMat(), dst = _dst.getUMat();
-
-        ocl::KernelArg srcarg = ocl::KernelArg::ReadOnlyNoSize(src),
-                maskarg = ocl::KernelArg::ReadOnlyNoSize(mask),
-                dstarg = ocl::KernelArg::ReadWrite(dst);
-
-        if (haveScale)
-        {
-            if (haveDelta)
-                k.args(srcarg, maskarg, dstarg, fscale, fdelta);
-            else
-                k.args(srcarg, maskarg, dstarg, fscale);
-        }
-        else
-        {
-            if (haveDelta)
-                k.args(srcarg, maskarg, dstarg, fdelta);
-            else
-                k.args(srcarg, maskarg, dstarg);
-        }
-
-        size_t globalsize[2] = { (size_t)src.cols, ((size_t)src.rows + rowsPerWI - 1) / rowsPerWI };
-        return k.run(2, globalsize, NULL, false);
-    }
-    else
-    {
-        UMat temp;
-        src.convertTo( temp, dtype, scale, delta );
-        temp.copyTo( _dst, _mask );
-    }
-
-    return true;
-}
-
-#endif
-
-void normalize(InputArray _src, InputOutputArray _dst, double a, double b,
-               int norm_type, int rtype, InputArray _mask)
-{
-    CV_INSTRUMENT_REGION();
-
-    double scale = 1, shift = 0;
-    int type = _src.type(), depth = CV_MAT_DEPTH(type);
-
-    if( rtype < 0 )
-        rtype = _dst.fixedType() ? _dst.depth() : depth;
-
-    if( norm_type == CV_MINMAX )
-    {
-        double smin = 0, smax = 0;
-        double dmin = MIN( a, b ), dmax = MAX( a, b );
-        minMaxIdx( _src, &smin, &smax, 0, 0, _mask );
-        scale = (dmax - dmin)*(smax - smin > DBL_EPSILON ? 1./(smax - smin) : 0);
-        if( rtype == CV_32F )
-        {
-            scale = (float)scale;
-            shift = (float)dmin - (float)(smin*scale);
-        }
-        else
-            shift = dmin - smin*scale;
-    }
-    else if( norm_type == CV_L2 || norm_type == CV_L1 || norm_type == CV_C )
-    {
-        scale = norm( _src, norm_type, _mask );
-        scale = scale > DBL_EPSILON ? a/scale : 0.;
-        shift = 0;
-    }
-    else
-        CV_Error( CV_StsBadArg, "Unknown/unsupported norm type" );
-
-    CV_OCL_RUN(_dst.isUMat(),
-               ocl_normalize(_src, _dst, _mask, rtype, scale, shift))
-
-    Mat src = _src.getMat();
-    if( _mask.empty() )
-        src.convertTo( _dst, rtype, scale, shift );
-    else
-    {
-        Mat temp;
-        src.convertTo( temp, rtype, scale, shift );
-        temp.copyTo( _dst, _mask );
-    }
-}
-
 } // namespace

modules/core/src/copy.cpp

@@ -53,6 +53,75 @@
 namespace cv
 {
 
+template <typename T> static inline
+void scalarToRawData_(const Scalar& s, T * const buf, const int cn, const int unroll_to)
+{
+    int i = 0;
+    for(; i < cn; i++)
+        buf[i] = saturate_cast<T>(s.val[i]);
+    for(; i < unroll_to; i++)
+        buf[i] = buf[i-cn];
+}
+
+void scalarToRawData(const Scalar& s, void* _buf, int type, int unroll_to)
+{
+    CV_INSTRUMENT_REGION();
+
+    const int depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
+    CV_Assert(cn <= 4);
+    switch(depth)
+    {
+    case CV_8U:
+        scalarToRawData_<uchar>(s, (uchar*)_buf, cn, unroll_to);
+        break;
+    case CV_8S:
+        scalarToRawData_<schar>(s, (schar*)_buf, cn, unroll_to);
+        break;
+    case CV_16U:
+        scalarToRawData_<ushort>(s, (ushort*)_buf, cn, unroll_to);
+        break;
+    case CV_16S:
+        scalarToRawData_<short>(s, (short*)_buf, cn, unroll_to);
+        break;
+    case CV_32S:
+        scalarToRawData_<int>(s, (int*)_buf, cn, unroll_to);
+        break;
+    case CV_32F:
+        scalarToRawData_<float>(s, (float*)_buf, cn, unroll_to);
+        break;
+    case CV_64F:
+        scalarToRawData_<double>(s, (double*)_buf, cn, unroll_to);
+        break;
+#if CV_VERSION_MAJOR >= 4
+    case CV_16F:
+        scalarToRawData_<float16_t>(s, (float16_t*)_buf, cn, unroll_to);
+        break;
+#endif
+    default:
+        CV_Error(CV_StsUnsupportedFormat,"");
+    }
+}
+
+void convertAndUnrollScalar( const Mat& sc, int buftype, uchar* scbuf, size_t blocksize )
+{
+    int scn = (int)sc.total(), cn = CV_MAT_CN(buftype);
+    size_t esz = CV_ELEM_SIZE(buftype);
+    BinaryFunc cvtFn = getConvertFunc(sc.depth(), buftype);
+    CV_Assert(cvtFn);
+    cvtFn(sc.ptr(), 1, 0, 1, scbuf, 1, Size(std::min(cn, scn), 1), 0);
+    // unroll the scalar
+    if( scn < cn )
+    {
+        CV_Assert( scn == 1 );
+        size_t esz1 = CV_ELEM_SIZE1(buftype);
+        for( size_t i = esz1; i < esz; i++ )
+            scbuf[i] = scbuf[i - esz1];
+    }
+    for( size_t i = esz; i < blocksize*esz; i++ )
+        scbuf[i] = scbuf[i - esz];
+}
+
+
 template<typename T> static void
 copyMask_(const uchar* _src, size_t sstep, const uchar* mask, size_t mstep, uchar* _dst, size_t dstep, Size size)
 {
@@ -594,490 +663,6 @@ Mat& Mat::setTo(InputArray _value, InputArray _mask)
     return *this;
 }
 
-#if CV_SIMD128
-template<typename V> CV_ALWAYS_INLINE void flipHoriz_single( const uchar* src, size_t sstep, uchar* dst, size_t dstep, Size size, size_t esz )
-{
-    typedef typename V::lane_type T;
-    int end = (int)(size.width*esz);
-    int width = (end + 1)/2;
-    int width_1 = width & -v_uint8x16::nlanes;
-    int i, j;
-
-#if CV_STRONG_ALIGNMENT
-    CV_Assert(isAligned<sizeof(T)>(src, dst));
-#endif
-
-    for( ; size.height--; src += sstep, dst += dstep )
-    {
-        for( i = 0, j = end; i < width_1; i += v_uint8x16::nlanes, j -= v_uint8x16::nlanes )
-        {
-            V t0, t1;
-
-            t0 = v_load((T*)((uchar*)src + i));
-            t1 = v_load((T*)((uchar*)src + j - v_uint8x16::nlanes));
-            t0 = v_reverse(t0);
-            t1 = v_reverse(t1);
-            v_store((T*)(dst + j - v_uint8x16::nlanes), t0);
-            v_store((T*)(dst + i), t1);
-        }
-        if (isAligned<sizeof(T)>(src, dst))
-        {
-            for ( ; i < width; i += sizeof(T), j -= sizeof(T) )
-            {
-                T t0, t1;
-
-                t0 = *((T*)((uchar*)src + i));
-                t1 = *((T*)((uchar*)src + j - sizeof(T)));
-                *((T*)(dst + j - sizeof(T))) = t0;
-                *((T*)(dst + i)) = t1;
-            }
-        }
-        else
-        {
-            for ( ; i < width; i += sizeof(T), j -= sizeof(T) )
-            {
-                for (int k = 0; k < (int)sizeof(T); k++)
-                {
-                    uchar t0, t1;
-
-                    t0 = *((uchar*)src + i + k);
-                    t1 = *((uchar*)src + j + k - sizeof(T));
-                    *(dst + j + k - sizeof(T)) = t0;
-                    *(dst + i + k) = t1;
-                }
-            }
-        }
-    }
-}
-
-template<typename T1, typename T2> CV_ALWAYS_INLINE void flipHoriz_double( const uchar* src, size_t sstep, uchar* dst, size_t dstep, Size size, size_t esz )
-{
-    int end = (int)(size.width*esz);
-    int width = (end + 1)/2;
-
-#if CV_STRONG_ALIGNMENT
-    CV_Assert(isAligned<sizeof(T1)>(src, dst));
-    CV_Assert(isAligned<sizeof(T2)>(src, dst));
-#endif
-
-    for( ; size.height--; src += sstep, dst += dstep )
-    {
-        for ( int i = 0, j = end; i < width; i += sizeof(T1) + sizeof(T2), j -= sizeof(T1) + sizeof(T2) )
-        {
-            T1 t0, t1;
-            T2 t2, t3;
-
-            t0 = *((T1*)((uchar*)src + i));
-            t2 = *((T2*)((uchar*)src + i + sizeof(T1)));
-            t1 = *((T1*)((uchar*)src + j - sizeof(T1) - sizeof(T2)));
-            t3 = *((T2*)((uchar*)src + j - sizeof(T2)));
-            *((T1*)(dst + j - sizeof(T1) - sizeof(T2))) = t0;
-            *((T2*)(dst + j - sizeof(T2))) = t2;
-            *((T1*)(dst + i)) = t1;
-            *((T2*)(dst + i + sizeof(T1))) = t3;
-        }
-    }
-}
-#endif
-
-static void
-flipHoriz( const uchar* src, size_t sstep, uchar* dst, size_t dstep, Size size, size_t esz )
-{
-#if CV_SIMD
-#if CV_STRONG_ALIGNMENT
-    size_t alignmentMark = ((size_t)src)|((size_t)dst)|sstep|dstep;
-#endif
-    if (esz == 2 * v_uint8x16::nlanes)
-    {
-        int end = (int)(size.width*esz);
-        int width = end/2;
-
-        for( ; size.height--; src += sstep, dst += dstep )
-        {
-            for( int i = 0, j = end - 2 * v_uint8x16::nlanes; i < width; i += 2 * v_uint8x16::nlanes, j -= 2 * v_uint8x16::nlanes )
-            {
-#if CV_SIMD256
-                v_uint8x32 t0, t1;
-
-                t0 = v256_load((uchar*)src + i);
-                t1 = v256_load((uchar*)src + j);
-                v_store(dst + j, t0);
-                v_store(dst + i, t1);
-#else
-                v_uint8x16 t0, t1, t2, t3;
-
-                t0 = v_load((uchar*)src + i);
-                t1 = v_load((uchar*)src + i + v_uint8x16::nlanes);
-                t2 = v_load((uchar*)src + j);
-                t3 = v_load((uchar*)src + j + v_uint8x16::nlanes);
-                v_store(dst + j, t0);
-                v_store(dst + j + v_uint8x16::nlanes, t1);
-                v_store(dst + i, t2);
-                v_store(dst + i + v_uint8x16::nlanes, t3);
-#endif
-            }
-        }
-    }
-    else if (esz == v_uint8x16::nlanes)
-    {
-        int end = (int)(size.width*esz);
-        int width = end/2;
-
-        for( ; size.height--; src += sstep, dst += dstep )
-        {
-            for( int i = 0, j = end - v_uint8x16::nlanes; i < width; i += v_uint8x16::nlanes, j -= v_uint8x16::nlanes )
-            {
-                v_uint8x16 t0, t1;
-
-                t0 = v_load((uchar*)src + i);
-                t1 = v_load((uchar*)src + j);
-                v_store(dst + j, t0);
-                v_store(dst + i, t1);
-            }
-        }
-    }
-    else if (esz == 8
-#if CV_STRONG_ALIGNMENT
-            && isAligned<sizeof(uint64)>(alignmentMark)
-#endif
-    )
-    {
-        flipHoriz_single<v_uint64x2>(src, sstep, dst, dstep, size, esz);
-    }
-    else if (esz == 4
-#if CV_STRONG_ALIGNMENT
-            && isAligned<sizeof(unsigned)>(alignmentMark)
-#endif
-    )
-    {
-        flipHoriz_single<v_uint32x4>(src, sstep, dst, dstep, size, esz);
-    }
-    else if (esz == 2
-#if CV_STRONG_ALIGNMENT
-            && isAligned<sizeof(ushort)>(alignmentMark)
-#endif
-    )
-    {
-        flipHoriz_single<v_uint16x8>(src, sstep, dst, dstep, size, esz);
-    }
-    else if (esz == 1)
-    {
-        flipHoriz_single<v_uint8x16>(src, sstep, dst, dstep, size, esz);
-    }
-    else if (esz == 24
-#if CV_STRONG_ALIGNMENT
-            && isAligned<sizeof(uint64_t)>(alignmentMark)
-#endif
-    )
-    {
-        int end = (int)(size.width*esz);
-        int width = (end + 1)/2;
-
-        for( ; size.height--; src += sstep, dst += dstep )
-        {
-            for ( int i = 0, j = end; i < width; i += v_uint8x16::nlanes + sizeof(uint64_t), j -= v_uint8x16::nlanes + sizeof(uint64_t) )
-            {
-                v_uint8x16 t0, t1;
-                uint64_t t2, t3;
-
-                t0 = v_load((uchar*)src + i);
-                t2 = *((uint64_t*)((uchar*)src + i + v_uint8x16::nlanes));
-                t1 = v_load((uchar*)src + j - v_uint8x16::nlanes - sizeof(uint64_t));
-                t3 = *((uint64_t*)((uchar*)src + j - sizeof(uint64_t)));
-                v_store(dst + j - v_uint8x16::nlanes - sizeof(uint64_t), t0);
-                *((uint64_t*)(dst + j - sizeof(uint64_t))) = t2;
-                v_store(dst + i, t1);
-                *((uint64_t*)(dst + i + v_uint8x16::nlanes)) = t3;
-            }
-        }
-    }
-#if !CV_STRONG_ALIGNMENT
-    else if (esz == 12)
-    {
-        flipHoriz_double<uint64_t,uint>(src, sstep, dst, dstep, size, esz);
-    }
-    else if (esz == 6)
-    {
-        flipHoriz_double<uint,ushort>(src, sstep, dst, dstep, size, esz);
-    }
-    else if (esz == 3)
-    {
-        flipHoriz_double<ushort,uchar>(src, sstep, dst, dstep, size, esz);
-    }
-#endif
-    else
-#endif // CV_SIMD
-    {
-        int i, j, limit = (int)(((size.width + 1)/2)*esz);
-        AutoBuffer<int> _tab(size.width*esz);
-        int* tab = _tab.data();
-
-        for( i = 0; i < size.width; i++ )
-            for( size_t k = 0; k < esz; k++ )
-                tab[i*esz + k] = (int)((size.width - i - 1)*esz + k);
-
-        for( ; size.height--; src += sstep, dst += dstep )
-        {
-            for( i = 0; i < limit; i++ )
-            {
-                j = tab[i];
-                uchar t0 = src[i], t1 = src[j];
-                dst[i] = t1; dst[j] = t0;
-            }
-        }
-    }
-}
-
-static void
-flipVert( const uchar* src0, size_t sstep, uchar* dst0, size_t dstep, Size size, size_t esz )
-{
-    const uchar* src1 = src0 + (size.height - 1)*sstep;
-    uchar* dst1 = dst0 + (size.height - 1)*dstep;
-    size.width *= (int)esz;
-
-    for( int y = 0; y < (size.height + 1)/2; y++, src0 += sstep, src1 -= sstep,
-                                                  dst0 += dstep, dst1 -= dstep )
-    {
-        int i = 0;
-#if CV_SIMD
-#if CV_STRONG_ALIGNMENT
-        if (isAligned<sizeof(int)>(src0, src1, dst0, dst1))
-#endif
-        {
-            for (; i <= size.width - CV_SIMD_WIDTH; i += CV_SIMD_WIDTH)
-            {
-                v_int32 t0 = vx_load((int*)(src0 + i));
-                v_int32 t1 = vx_load((int*)(src1 + i));
-                vx_store((int*)(dst0 + i), t1);
-                vx_store((int*)(dst1 + i), t0);
-            }
-        }
-#if CV_STRONG_ALIGNMENT
-        else
-        {
-            for (; i <= size.width - CV_SIMD_WIDTH; i += CV_SIMD_WIDTH)
-            {
-                v_uint8 t0 = vx_load(src0 + i);
-                v_uint8 t1 = vx_load(src1 + i);
-                vx_store(dst0 + i, t1);
-                vx_store(dst1 + i, t0);
-            }
-        }
-#endif
-#endif
-
-        if (isAligned<sizeof(int)>(src0, src1, dst0, dst1))
-        {
-            for( ; i <= size.width - 16; i += 16 )
-            {
-                int t0 = ((int*)(src0 + i))[0];
-                int t1 = ((int*)(src1 + i))[0];
-
-                ((int*)(dst0 + i))[0] = t1;
-                ((int*)(dst1 + i))[0] = t0;
-
-                t0 = ((int*)(src0 + i))[1];
-                t1 = ((int*)(src1 + i))[1];
-
-                ((int*)(dst0 + i))[1] = t1;
-                ((int*)(dst1 + i))[1] = t0;
-
-                t0 = ((int*)(src0 + i))[2];
-                t1 = ((int*)(src1 + i))[2];
-
-                ((int*)(dst0 + i))[2] = t1;
-                ((int*)(dst1 + i))[2] = t0;
-
-                t0 = ((int*)(src0 + i))[3];
-                t1 = ((int*)(src1 + i))[3];
-
-                ((int*)(dst0 + i))[3] = t1;
-                ((int*)(dst1 + i))[3] = t0;
-            }
-
-            for( ; i <= size.width - 4; i += 4 )
-            {
-                int t0 = ((int*)(src0 + i))[0];
-                int t1 = ((int*)(src1 + i))[0];
-
-                ((int*)(dst0 + i))[0] = t1;
-                ((int*)(dst1 + i))[0] = t0;
-            }
-        }
-
-        for( ; i < size.width; i++ )
-        {
-            uchar t0 = src0[i];
-            uchar t1 = src1[i];
-
-            dst0[i] = t1;
-            dst1[i] = t0;
-        }
-    }
-}
-
-#ifdef HAVE_OPENCL
-
-enum { FLIP_COLS = 1 << 0, FLIP_ROWS = 1 << 1, FLIP_BOTH = FLIP_ROWS | FLIP_COLS };
-
-static bool ocl_flip(InputArray _src, OutputArray _dst, int flipCode )
-{
-    CV_Assert(flipCode >= -1 && flipCode <= 1);
-
-    const ocl::Device & dev = ocl::Device::getDefault();
-    int type = _src.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type),
-            flipType, kercn = std::min(ocl::predictOptimalVectorWidth(_src, _dst), 4);
-
-    bool doubleSupport = dev.doubleFPConfig() > 0;
-    if (!doubleSupport && depth == CV_64F)
-        kercn = cn;
-
-    if (cn > 4)
-        return false;
-
-    const char * kernelName;
-    if (flipCode == 0)
-        kernelName = "arithm_flip_rows", flipType = FLIP_ROWS;
-    else if (flipCode > 0)
-        kernelName = "arithm_flip_cols", flipType = FLIP_COLS;
-    else
-        kernelName = "arithm_flip_rows_cols", flipType = FLIP_BOTH;
-
-    int pxPerWIy = (dev.isIntel() && (dev.type() & ocl::Device::TYPE_GPU)) ? 4 : 1;
-    kercn = (cn!=3 || flipType == FLIP_ROWS) ? std::max(kercn, cn) : cn;
-
-    ocl::Kernel k(kernelName, ocl::core::flip_oclsrc,
-        format( "-D T=%s -D T1=%s -D DEPTH=%d -D cn=%d -D PIX_PER_WI_Y=%d -D kercn=%d",
-                kercn != cn ? ocl::typeToStr(CV_MAKE_TYPE(depth, kercn)) : ocl::vecopTypeToStr(CV_MAKE_TYPE(depth, kercn)),
-                kercn != cn ? ocl::typeToStr(depth) : ocl::vecopTypeToStr(depth), depth, cn, pxPerWIy, kercn));
-    if (k.empty())
-        return false;
-
-    Size size = _src.size();
-    _dst.create(size, type);
-    UMat src = _src.getUMat(), dst = _dst.getUMat();
-
-    int cols = size.width * cn / kercn, rows = size.height;
-    cols = flipType == FLIP_COLS ? (cols + 1) >> 1 : cols;
-    rows = flipType & FLIP_ROWS ? (rows + 1) >> 1 : rows;
-
-    k.args(ocl::KernelArg::ReadOnlyNoSize(src),
-           ocl::KernelArg::WriteOnly(dst, cn, kercn), rows, cols);
-
-    size_t maxWorkGroupSize = dev.maxWorkGroupSize();
-    CV_Assert(maxWorkGroupSize % 4 == 0);
-
-    size_t globalsize[2] = { (size_t)cols, ((size_t)rows + pxPerWIy - 1) / pxPerWIy },
-            localsize[2] = { maxWorkGroupSize / 4, 4 };
-    return k.run(2, globalsize, (flipType == FLIP_COLS) && !dev.isIntel() ? localsize : NULL, false);
-}
-
-#endif
-
-#if defined HAVE_IPP
-static bool ipp_flip(Mat &src, Mat &dst, int flip_mode)
-{
-#ifdef HAVE_IPP_IW
-    CV_INSTRUMENT_REGION_IPP();
-
-    // Details: https://github.com/opencv/opencv/issues/12943
-    if (flip_mode <= 0 /* swap rows */
-        && cv::ipp::getIppTopFeatures() != ippCPUID_SSE42
-        && (int64_t)(src.total()) * src.elemSize() >= CV_BIG_INT(0x80000000)/*2Gb*/
-    )
-        return false;
-
-    IppiAxis ippMode;
-    if(flip_mode < 0)
-        ippMode = ippAxsBoth;
-    else if(flip_mode == 0)
-        ippMode = ippAxsHorizontal;
-    else
-        ippMode = ippAxsVertical;
-
-    try
-    {
-        ::ipp::IwiImage iwSrc = ippiGetImage(src);
-        ::ipp::IwiImage iwDst = ippiGetImage(dst);
-
-        CV_INSTRUMENT_FUN_IPP(::ipp::iwiMirror, iwSrc, iwDst, ippMode);
-    }
-    catch(const ::ipp::IwException &)
-    {
-        return false;
-    }
-
-    return true;
-#else
-    CV_UNUSED(src); CV_UNUSED(dst); CV_UNUSED(flip_mode);
-    return false;
-#endif
-}
-#endif
-
-
-void flip( InputArray _src, OutputArray _dst, int flip_mode )
-{
-    CV_INSTRUMENT_REGION();
-
-    CV_Assert( _src.dims() <= 2 );
-    Size size = _src.size();
-
-    if (flip_mode < 0)
-    {
-        if (size.width == 1)
-            flip_mode = 0;
-        if (size.height == 1)
-            flip_mode = 1;
-    }
-
-    if ((size.width == 1 && flip_mode > 0) ||
-        (size.height == 1 && flip_mode == 0))
-    {
-        return _src.copyTo(_dst);
-    }
-
-    CV_OCL_RUN( _dst.isUMat(), ocl_flip(_src, _dst, flip_mode))
-
-    Mat src = _src.getMat();
-    int type = src.type();
-    _dst.create( size, type );
-    Mat dst = _dst.getMat();
-
-    CV_IPP_RUN_FAST(ipp_flip(src, dst, flip_mode));
-
-    size_t esz = CV_ELEM_SIZE(type);
-
-    if( flip_mode <= 0 )
-        flipVert( src.ptr(), src.step, dst.ptr(), dst.step, src.size(), esz );
-    else
-        flipHoriz( src.ptr(), src.step, dst.ptr(), dst.step, src.size(), esz );
-
-    if( flip_mode < 0 )
-        flipHoriz( dst.ptr(), dst.step, dst.ptr(), dst.step, dst.size(), esz );
-}
-
-void rotate(InputArray _src, OutputArray _dst, int rotateMode)
-{
-    CV_Assert(_src.dims() <= 2);
-
-    switch (rotateMode)
-    {
-    case ROTATE_90_CLOCKWISE:
-        transpose(_src, _dst);
-        flip(_dst, _dst, 1);
-        break;
-    case ROTATE_180:
-        flip(_src, _dst, -1);
-        break;
-    case ROTATE_90_COUNTERCLOCKWISE:
-        transpose(_src, _dst);
-        flip(_dst, _dst, 0);
-        break;
-    default:
-        break;
-    }
-}
 
 #if defined HAVE_OPENCL && !defined __APPLE__
 
@@ -1499,6 +1084,9 @@ void cv::copyMakeBorder( InputArray _src, OutputArray _dst, int top, int bottom,
     }
 }
 
+
+#ifndef OPENCV_EXCLUDE_C_API
+
 /* dst = src */
 CV_IMPL void
 cvCopy( const void* srcarr, void* dstarr, const void* maskarr )
@@ -1614,4 +1202,5 @@ cvRepeat( const CvArr* srcarr, CvArr* dstarr )
     cv::repeat(src, dst.rows/src.rows, dst.cols/src.cols, dst);
 }
 
+#endif  // OPENCV_EXCLUDE_C_API
 /* End of file. */

modules/core/src/datastructs.cpp

@@ -40,6 +40,8 @@
 //M*/
 #include "precomp.hpp"
 
+#ifndef OPENCV_EXCLUDE_C_API
+
 /* default alignment for dynamic data strucutures, resided in storages. */
 #define  CV_STRUCT_ALIGN    ((int)sizeof(double))
 
@@ -3585,4 +3587,5 @@ void  seqInsertSlice( CvSeq* seq, int before_index, const CvArr* from_arr )
 
 }
 
+#endif  // OPENCV_EXCLUDE_C_API
 /* End of file. */

modules/core/src/directx.cpp

@@ -901,7 +901,7 @@ bool ocl_convert_nv12_to_bgr(
 
     k.args(clImageY, clImageUV, clBuffer, step, cols, rows);
 
-    size_t globalsize[] = { (size_t)cols, (size_t)rows };
+    size_t globalsize[] = { (size_t)cols/2, (size_t)rows/2 };
     return k.run(2, globalsize, 0, false);
 }
 
@@ -922,7 +922,7 @@ bool ocl_convert_bgr_to_nv12(
 
     k.args(clBuffer, step, cols, rows, clImageY, clImageUV);
 
-    size_t globalsize[] = { (size_t)cols, (size_t)rows };
+    size_t globalsize[] = { (size_t)cols/2, (size_t)rows/2 };
     return k.run(2, globalsize, 0, false);
 }
 

modules/core/src/dxt.cpp

@@ -4640,6 +4640,9 @@ int cv::getOptimalDFTSize( int size0 )
     return optimalDFTSizeTab[b];
 }
 
+
+#ifndef OPENCV_EXCLUDE_C_API
+
 CV_IMPL void
 cvDFT( const CvArr* srcarr, CvArr* dstarr, int flags, int nonzero_rows )
 {
@@ -4695,4 +4698,5 @@ cvGetOptimalDFTSize( int size0 )
     return cv::getOptimalDFTSize(size0);
 }
 
+#endif  // OPENCV_EXCLUDE_C_API
 /* End of file. */

modules/core/src/lapack.cpp

@@ -753,8 +753,6 @@ SVBkSb( int m, int n, const double* w, size_t wstep,
                 (double*)alignPtr(buffer, sizeof(double)), DBL_EPSILON*2 );
 }
 
-}
-
 /****************************************************************************************\
 *                                 Determinant of the matrix                              *
 \****************************************************************************************/
@@ -764,7 +762,7 @@ SVBkSb( int m, int n, const double* w, size_t wstep,
                    m(0,1)*((double)m(1,0)*m(2,2) - (double)m(1,2)*m(2,0)) +  \
                    m(0,2)*((double)m(1,0)*m(2,1) - (double)m(1,1)*m(2,0)))
 
-double cv::determinant( InputArray _mat )
+double determinant( InputArray _mat )
 {
     CV_INSTRUMENT_REGION();
 
@@ -842,7 +840,7 @@ double cv::determinant( InputArray _mat )
 #define Df( y, x ) ((float*)(dstdata + y*dststep))[x]
 #define Dd( y, x ) ((double*)(dstdata + y*dststep))[x]
 
-double cv::invert( InputArray _src, OutputArray _dst, int method )
+double invert( InputArray _src, OutputArray _dst, int method )
 {
     CV_INSTRUMENT_REGION();
 
@@ -1069,13 +1067,19 @@ double cv::invert( InputArray _src, OutputArray _dst, int method )
     return result;
 }
 
+UMat UMat::inv(int method) const
+{
+    UMat m;
+    invert(*this, m, method);
+    return m;
+}
 
 
 /****************************************************************************************\
 *                              Solving a linear system                                   *
 \****************************************************************************************/
 
-bool cv::solve( InputArray _src, InputArray _src2arg, OutputArray _dst, int method )
+bool solve( InputArray _src, InputArray _src2arg, OutputArray _dst, int method )
 {
     CV_INSTRUMENT_REGION();
 
@@ -1374,7 +1378,7 @@ bool cv::solve( InputArray _src, InputArray _src2arg, OutputArray _dst, int meth
 
 /////////////////// finding eigenvalues and eigenvectors of a symmetric matrix ///////////////
 
-bool cv::eigen( InputArray _src, OutputArray _evals, OutputArray _evects )
+bool eigen( InputArray _src, OutputArray _evals, OutputArray _evects )
 {
     CV_INSTRUMENT_REGION();
 
@@ -1396,7 +1400,7 @@ bool cv::eigen( InputArray _src, OutputArray _evals, OutputArray _evects )
     const bool evecNeeded = _evects.needed();
     const int esOptions = evecNeeded ? Eigen::ComputeEigenvectors : Eigen::EigenvaluesOnly;
     _evals.create(n, 1, type);
-    cv::Mat evals = _evals.getMat();
+    Mat evals = _evals.getMat();
     if ( type == CV_64F )
     {
         Eigen::MatrixXd src_eig, zeros_eig;
@@ -1448,9 +1452,6 @@ bool cv::eigen( InputArray _src, OutputArray _evals, OutputArray _evects )
 #endif
 }
 
-namespace cv
-{
-
 static void _SVDcompute( InputArray _aarr, OutputArray _w,
                          OutputArray _u, OutputArray _vt, int flags )
 {
@@ -1598,6 +1599,9 @@ void cv::SVBackSubst(InputArray w, InputArray u, InputArray vt, InputArray rhs,
 }
 
 
+
+#ifndef OPENCV_EXCLUDE_C_API
+
 CV_IMPL double
 cvDet( const CvArr* arr )
 {
@@ -1789,3 +1793,4 @@ cvSVBkSb( const CvArr* warr, const CvArr* uarr,
     cv::SVD::backSubst(w, u, v, rhs, dst);
     CV_Assert( dst.data == dst0.data );
 }
+#endif  // OPENCV_EXCLUDE_C_API

modules/core/src/mathfuncs.cpp

@@ -1637,6 +1637,9 @@ void patchNaNs( InputOutputArray _a, double _val )
 
 }
 
+
+#ifndef OPENCV_EXCLUDE_C_API
+
 CV_IMPL float cvCbrt(float value) { return cv::cubeRoot(value); }
 CV_IMPL float cvFastArctan(float y, float x) { return cv::fastAtan2(y, x); }
 
@@ -1720,6 +1723,7 @@ CV_IMPL int cvCheckArr( const CvArr* arr, int flags,
     return cv::checkRange(cv::cvarrToMat(arr), (flags & CV_CHECK_QUIET) != 0, 0, minVal, maxVal );
 }
 
+#endif  // OPENCV_EXCLUDE_C_API
 
 /*
   Finds real roots of cubic, quadratic or linear equation.
@@ -2015,6 +2019,8 @@ double cv::solvePoly( InputArray _coeffs0, OutputArray _roots0, int maxIters )
 }
 
 
+#ifndef OPENCV_EXCLUDE_C_API
+
 CV_IMPL int
 cvSolveCubic( const CvMat* coeffs, CvMat* roots )
 {
@@ -2034,6 +2040,7 @@ void cvSolvePoly(const CvMat* a, CvMat *r, int maxiter, int)
     CV_Assert( _r.data == _r0.data ); // check that the array of roots was not reallocated
 }
 
+#endif  // OPENCV_EXCLUDE_C_API
 
 
 // Common constants for dispatched code

modules/core/src/mathfuncs_core.dispatch.cpp

@@ -7,6 +7,10 @@
 #include "mathfuncs_core.simd.hpp"
 #include "mathfuncs_core.simd_declarations.hpp" // defines CV_CPU_DISPATCH_MODES_ALL=AVX2,...,BASELINE based on CMakeLists.txt content
 
+
+#define IPP_DISABLE_MAGNITUDE_32F 1  // accuracy: https://github.com/opencv/opencv/issues/19506
+
+
 namespace cv { namespace hal {
 
 ///////////////////////////////////// ATAN2 ////////////////////////////////////
@@ -44,8 +48,25 @@ void magnitude32f(const float* x, const float* y, float* mag, int len)
     CV_INSTRUMENT_REGION();
 
     CALL_HAL(magnitude32f, cv_hal_magnitude32f, x, y, mag, len);
+
+#ifdef HAVE_IPP
+    bool allowIPP = true;
+#ifdef IPP_DISABLE_MAGNITUDE_32F
+    if (cv::ipp::getIppTopFeatures() & (
+#if IPP_VERSION_X100 >= 201700
+            ippCPUID_AVX512F |
+#endif
+            ippCPUID_AVX2)
+    )
+    {
+        allowIPP = (len & 7) == 0;
+    }
+#endif
+
     // SSE42 performance issues
-    CV_IPP_RUN(IPP_VERSION_X100 > 201800 || cv::ipp::getIppTopFeatures() != ippCPUID_SSE42, CV_INSTRUMENT_FUN_IPP(ippsMagnitude_32f, x, y, mag, len) >= 0);
+    CV_IPP_RUN((IPP_VERSION_X100 > 201800 || cv::ipp::getIppTopFeatures() != ippCPUID_SSE42) && allowIPP,
+        CV_INSTRUMENT_FUN_IPP(ippsMagnitude_32f, x, y, mag, len) >= 0);
+#endif
 
     CV_CPU_DISPATCH(magnitude32f, (x, y, mag, len),
         CV_CPU_DISPATCH_MODES_ALL);

modules/core/src/matmul.dispatch.cpp

@@ -999,8 +999,79 @@ double Mat::dot(InputArray _mat) const
     return r;
 }
 
+
+#ifdef HAVE_OPENCL
+
+static bool ocl_dot( InputArray _src1, InputArray _src2, double & res )
+{
+    UMat src1 = _src1.getUMat().reshape(1), src2 = _src2.getUMat().reshape(1);
+
+    int type = src1.type(), depth = CV_MAT_DEPTH(type),
+            kercn = ocl::predictOptimalVectorWidth(src1, src2);
+    bool doubleSupport = ocl::Device::getDefault().doubleFPConfig() > 0;
+
+    if ( !doubleSupport && depth == CV_64F )
+        return false;
+
+    int dbsize = ocl::Device::getDefault().maxComputeUnits();
+    size_t wgs = ocl::Device::getDefault().maxWorkGroupSize();
+    int ddepth = std::max(CV_32F, depth);
+
+    int wgs2_aligned = 1;
+    while (wgs2_aligned < (int)wgs)
+        wgs2_aligned <<= 1;
+    wgs2_aligned >>= 1;
+
+    char cvt[40];
+    ocl::Kernel k("reduce", ocl::core::reduce_oclsrc,
+                  format("-D srcT=%s -D srcT1=%s -D dstT=%s -D dstTK=%s -D ddepth=%d -D convertToDT=%s -D OP_DOT "
+                         "-D WGS=%d -D WGS2_ALIGNED=%d%s%s%s -D kercn=%d",
+                         ocl::typeToStr(CV_MAKE_TYPE(depth, kercn)), ocl::typeToStr(depth),
+                         ocl::typeToStr(ddepth), ocl::typeToStr(CV_MAKE_TYPE(ddepth, kercn)),
+                         ddepth, ocl::convertTypeStr(depth, ddepth, kercn, cvt),
+                         (int)wgs, wgs2_aligned, doubleSupport ? " -D DOUBLE_SUPPORT" : "",
+                         _src1.isContinuous() ? " -D HAVE_SRC_CONT" : "",
+                         _src2.isContinuous() ? " -D HAVE_SRC2_CONT" : "", kercn));
+    if (k.empty())
+        return false;
+
+    UMat db(1, dbsize, ddepth);
+
+    ocl::KernelArg src1arg = ocl::KernelArg::ReadOnlyNoSize(src1),
+            src2arg = ocl::KernelArg::ReadOnlyNoSize(src2),
+            dbarg = ocl::KernelArg::PtrWriteOnly(db);
+
+    k.args(src1arg, src1.cols, (int)src1.total(), dbsize, dbarg, src2arg);
+
+    size_t globalsize = dbsize * wgs;
+    if (k.run(1, &globalsize, &wgs, false))
+    {
+        res = sum(db.getMat(ACCESS_READ))[0];
+        return true;
+    }
+    return false;
+}
+
+#endif
+
+double UMat::dot(InputArray m) const
+{
+    CV_INSTRUMENT_REGION();
+
+    CV_Assert(m.sameSize(*this) && m.type() == type());
+
+#ifdef HAVE_OPENCL
+    double r = 0;
+    CV_OCL_RUN_(dims <= 2, ocl_dot(*this, m, r), r)
+#endif
+
+    return getMat(ACCESS_READ).dot(m);
+}
+
 }  // namespace cv::
 
+
+#ifndef OPENCV_EXCLUDE_C_API
 /****************************************************************************************\
 *                                    Earlier API                                         *
 \****************************************************************************************/
@@ -1225,4 +1296,6 @@ cvBackProjectPCA( const CvArr* proj_arr, const CvArr* avg_arr,
     CV_Assert(dst0.data == dst.data);
 }
 
+#endif  // OPENCV_EXCLUDE_C_API
+
 /* End of file. */

modules/core/src/matrix.cpp

@@ -204,7 +204,7 @@ MatAllocator* Mat::getStdAllocator()
 
 //==================================================================================================
 
-bool MatSize::operator==(const MatSize& sz) const
+bool MatSize::operator==(const MatSize& sz) const CV_NOEXCEPT
 {
     int d = dims();
     int dsz = sz.dims();
@@ -337,7 +337,7 @@ void finalizeHdr(Mat& m)
 
 //======================================= Mat ======================================================
 
-Mat::Mat()
+Mat::Mat() CV_NOEXCEPT
     : flags(MAGIC_VAL), dims(0), rows(0), cols(0), data(0), datastart(0), dataend(0),
       datalimit(0), allocator(0), u(0), size(&rows), step(0)
 {}

modules/core/src/matrix_c.cpp

@@ -6,6 +6,7 @@
 #include "opencv2/core/mat.hpp"
 #include "opencv2/core/types_c.h"
 
+#ifndef OPENCV_EXCLUDE_C_API
 // glue
 
 CvMatND cvMatND(const cv::Mat& m)
@@ -360,7 +361,6 @@ cvSort( const CvArr* _src, CvArr* _dst, CvArr* _idx, int flags )
     }
 }
 
-
 CV_IMPL int
 cvKMeans2( const CvArr* _samples, int cluster_count, CvArr* _labels,
            CvTermCriteria termcrit, int attempts, CvRNG*,
@@ -389,3 +389,5 @@ cvKMeans2( const CvArr* _samples, int cluster_count, CvArr* _labels,
         *_compactness = compactness;
     return 1;
 }
+
+#endif  // OPENCV_EXCLUDE_C_API

modules/core/src/matrix_operations.cpp

@@ -226,6 +226,23 @@ void cv::setIdentity( InputOutputArray _m, const Scalar& s )
     }
 }
 
+
+namespace cv {
+
+UMat UMat::eye(int rows, int cols, int type)
+{
+    return UMat::eye(Size(cols, rows), type);
+}
+
+UMat UMat::eye(Size size, int type)
+{
+    UMat m(size, type);
+    setIdentity(m);
+    return m;
+}
+
+}  // namespace
+
 //////////////////////////////////////////// trace ///////////////////////////////////////////
 
 cv::Scalar cv::trace( InputArray _m )
@@ -260,285 +277,6 @@ cv::Scalar cv::trace( InputArray _m )
     return cv::sum(m.diag());
 }
 
-////////////////////////////////////// transpose /////////////////////////////////////////
-
-namespace cv
-{
-
-template<typename T> static void
-transpose_( const uchar* src, size_t sstep, uchar* dst, size_t dstep, Size sz )
-{
-    int i=0, j, m = sz.width, n = sz.height;
-
-    #if CV_ENABLE_UNROLLED
-    for(; i <= m - 4; i += 4 )
-    {
-        T* d0 = (T*)(dst + dstep*i);
-        T* d1 = (T*)(dst + dstep*(i+1));
-        T* d2 = (T*)(dst + dstep*(i+2));
-        T* d3 = (T*)(dst + dstep*(i+3));
-
-        for( j = 0; j <= n - 4; j += 4 )
-        {
-            const T* s0 = (const T*)(src + i*sizeof(T) + sstep*j);
-            const T* s1 = (const T*)(src + i*sizeof(T) + sstep*(j+1));
-            const T* s2 = (const T*)(src + i*sizeof(T) + sstep*(j+2));
-            const T* s3 = (const T*)(src + i*sizeof(T) + sstep*(j+3));
-
-            d0[j] = s0[0]; d0[j+1] = s1[0]; d0[j+2] = s2[0]; d0[j+3] = s3[0];
-            d1[j] = s0[1]; d1[j+1] = s1[1]; d1[j+2] = s2[1]; d1[j+3] = s3[1];
-            d2[j] = s0[2]; d2[j+1] = s1[2]; d2[j+2] = s2[2]; d2[j+3] = s3[2];
-            d3[j] = s0[3]; d3[j+1] = s1[3]; d3[j+2] = s2[3]; d3[j+3] = s3[3];
-        }
-
-        for( ; j < n; j++ )
-        {
-            const T* s0 = (const T*)(src + i*sizeof(T) + j*sstep);
-            d0[j] = s0[0]; d1[j] = s0[1]; d2[j] = s0[2]; d3[j] = s0[3];
-        }
-    }
-    #endif
-    for( ; i < m; i++ )
-    {
-        T* d0 = (T*)(dst + dstep*i);
-        j = 0;
-        #if CV_ENABLE_UNROLLED
-        for(; j <= n - 4; j += 4 )
-        {
-            const T* s0 = (const T*)(src + i*sizeof(T) + sstep*j);
-            const T* s1 = (const T*)(src + i*sizeof(T) + sstep*(j+1));
-            const T* s2 = (const T*)(src + i*sizeof(T) + sstep*(j+2));
-            const T* s3 = (const T*)(src + i*sizeof(T) + sstep*(j+3));
-
-            d0[j] = s0[0]; d0[j+1] = s1[0]; d0[j+2] = s2[0]; d0[j+3] = s3[0];
-        }
-        #endif
-        for( ; j < n; j++ )
-        {
-            const T* s0 = (const T*)(src + i*sizeof(T) + j*sstep);
-            d0[j] = s0[0];
-        }
-    }
-}
-
-template<typename T> static void
-transposeI_( uchar* data, size_t step, int n )
-{
-    for( int i = 0; i < n; i++ )
-    {
-        T* row = (T*)(data + step*i);
-        uchar* data1 = data + i*sizeof(T);
-        for( int j = i+1; j < n; j++ )
-            std::swap( row[j], *(T*)(data1 + step*j) );
-    }
-}
-
-typedef void (*TransposeFunc)( const uchar* src, size_t sstep, uchar* dst, size_t dstep, Size sz );
-typedef void (*TransposeInplaceFunc)( uchar* data, size_t step, int n );
-
-#define DEF_TRANSPOSE_FUNC(suffix, type) \
-static void transpose_##suffix( const uchar* src, size_t sstep, uchar* dst, size_t dstep, Size sz ) \
-{ transpose_<type>(src, sstep, dst, dstep, sz); } \
-\
-static void transposeI_##suffix( uchar* data, size_t step, int n ) \
-{ transposeI_<type>(data, step, n); }
-
-DEF_TRANSPOSE_FUNC(8u, uchar)
-DEF_TRANSPOSE_FUNC(16u, ushort)
-DEF_TRANSPOSE_FUNC(8uC3, Vec3b)
-DEF_TRANSPOSE_FUNC(32s, int)
-DEF_TRANSPOSE_FUNC(16uC3, Vec3s)
-DEF_TRANSPOSE_FUNC(32sC2, Vec2i)
-DEF_TRANSPOSE_FUNC(32sC3, Vec3i)
-DEF_TRANSPOSE_FUNC(32sC4, Vec4i)
-DEF_TRANSPOSE_FUNC(32sC6, Vec6i)
-DEF_TRANSPOSE_FUNC(32sC8, Vec8i)
-
-static TransposeFunc transposeTab[] =
-{
-    0, transpose_8u, transpose_16u, transpose_8uC3, transpose_32s, 0, transpose_16uC3, 0,
-    transpose_32sC2, 0, 0, 0, transpose_32sC3, 0, 0, 0, transpose_32sC4,
-    0, 0, 0, 0, 0, 0, 0, transpose_32sC6, 0, 0, 0, 0, 0, 0, 0, transpose_32sC8
-};
-
-static TransposeInplaceFunc transposeInplaceTab[] =
-{
-    0, transposeI_8u, transposeI_16u, transposeI_8uC3, transposeI_32s, 0, transposeI_16uC3, 0,
-    transposeI_32sC2, 0, 0, 0, transposeI_32sC3, 0, 0, 0, transposeI_32sC4,
-    0, 0, 0, 0, 0, 0, 0, transposeI_32sC6, 0, 0, 0, 0, 0, 0, 0, transposeI_32sC8
-};
-
-#ifdef HAVE_OPENCL
-
-static bool ocl_transpose( InputArray _src, OutputArray _dst )
-{
-    const ocl::Device & dev = ocl::Device::getDefault();
-    const int TILE_DIM = 32, BLOCK_ROWS = 8;
-    int type = _src.type(), cn = CV_MAT_CN(type), depth = CV_MAT_DEPTH(type),
-        rowsPerWI = dev.isIntel() ? 4 : 1;
-
-    UMat src = _src.getUMat();
-    _dst.create(src.cols, src.rows, type);
-    UMat dst = _dst.getUMat();
-
-    String kernelName("transpose");
-    bool inplace = dst.u == src.u;
-
-    if (inplace)
-    {
-        CV_Assert(dst.cols == dst.rows);
-        kernelName += "_inplace";
-    }
-    else
-    {
-        // check required local memory size
-        size_t required_local_memory = (size_t) TILE_DIM*(TILE_DIM+1)*CV_ELEM_SIZE(type);
-        if (required_local_memory > ocl::Device::getDefault().localMemSize())
-            return false;
-    }
-
-    ocl::Kernel k(kernelName.c_str(), ocl::core::transpose_oclsrc,
-                  format("-D T=%s -D T1=%s -D cn=%d -D TILE_DIM=%d -D BLOCK_ROWS=%d -D rowsPerWI=%d%s",
-                         ocl::memopTypeToStr(type), ocl::memopTypeToStr(depth),
-                         cn, TILE_DIM, BLOCK_ROWS, rowsPerWI, inplace ? " -D INPLACE" : ""));
-    if (k.empty())
-        return false;
-
-    if (inplace)
-        k.args(ocl::KernelArg::ReadWriteNoSize(dst), dst.rows);
-    else
-        k.args(ocl::KernelArg::ReadOnly(src),
-               ocl::KernelArg::WriteOnlyNoSize(dst));
-
-    size_t localsize[2]  = { TILE_DIM, BLOCK_ROWS };
-    size_t globalsize[2] = { (size_t)src.cols, inplace ? ((size_t)src.rows + rowsPerWI - 1) / rowsPerWI : (divUp((size_t)src.rows, TILE_DIM) * BLOCK_ROWS) };
-
-    if (inplace && dev.isIntel())
-    {
-        localsize[0] = 16;
-        localsize[1] = dev.maxWorkGroupSize() / localsize[0];
-    }
-
-    return k.run(2, globalsize, localsize, false);
-}
-
-#endif
-
-#ifdef HAVE_IPP
-static bool ipp_transpose( Mat &src, Mat &dst )
-{
-    CV_INSTRUMENT_REGION_IPP();
-
-    int type = src.type();
-    typedef IppStatus (CV_STDCALL * IppiTranspose)(const void * pSrc, int srcStep, void * pDst, int dstStep, IppiSize roiSize);
-    typedef IppStatus (CV_STDCALL * IppiTransposeI)(const void * pSrcDst, int srcDstStep, IppiSize roiSize);
-    IppiTranspose ippiTranspose = 0;
-    IppiTransposeI ippiTranspose_I = 0;
-
-    if (dst.data == src.data && dst.cols == dst.rows)
-    {
-        CV_SUPPRESS_DEPRECATED_START
-        ippiTranspose_I =
-            type == CV_8UC1 ? (IppiTransposeI)ippiTranspose_8u_C1IR :
-            type == CV_8UC3 ? (IppiTransposeI)ippiTranspose_8u_C3IR :
-            type == CV_8UC4 ? (IppiTransposeI)ippiTranspose_8u_C4IR :
-            type == CV_16UC1 ? (IppiTransposeI)ippiTranspose_16u_C1IR :
-            type == CV_16UC3 ? (IppiTransposeI)ippiTranspose_16u_C3IR :
-            type == CV_16UC4 ? (IppiTransposeI)ippiTranspose_16u_C4IR :
-            type == CV_16SC1 ? (IppiTransposeI)ippiTranspose_16s_C1IR :
-            type == CV_16SC3 ? (IppiTransposeI)ippiTranspose_16s_C3IR :
-            type == CV_16SC4 ? (IppiTransposeI)ippiTranspose_16s_C4IR :
-            type == CV_32SC1 ? (IppiTransposeI)ippiTranspose_32s_C1IR :
-            type == CV_32SC3 ? (IppiTransposeI)ippiTranspose_32s_C3IR :
-            type == CV_32SC4 ? (IppiTransposeI)ippiTranspose_32s_C4IR :
-            type == CV_32FC1 ? (IppiTransposeI)ippiTranspose_32f_C1IR :
-            type == CV_32FC3 ? (IppiTransposeI)ippiTranspose_32f_C3IR :
-            type == CV_32FC4 ? (IppiTransposeI)ippiTranspose_32f_C4IR : 0;
-        CV_SUPPRESS_DEPRECATED_END
-    }
-    else
-    {
-        ippiTranspose =
-            type == CV_8UC1 ? (IppiTranspose)ippiTranspose_8u_C1R :
-            type == CV_8UC3 ? (IppiTranspose)ippiTranspose_8u_C3R :
-            type == CV_8UC4 ? (IppiTranspose)ippiTranspose_8u_C4R :
-            type == CV_16UC1 ? (IppiTranspose)ippiTranspose_16u_C1R :
-            type == CV_16UC3 ? (IppiTranspose)ippiTranspose_16u_C3R :
-            type == CV_16UC4 ? (IppiTranspose)ippiTranspose_16u_C4R :
-            type == CV_16SC1 ? (IppiTranspose)ippiTranspose_16s_C1R :
-            type == CV_16SC3 ? (IppiTranspose)ippiTranspose_16s_C3R :
-            type == CV_16SC4 ? (IppiTranspose)ippiTranspose_16s_C4R :
-            type == CV_32SC1 ? (IppiTranspose)ippiTranspose_32s_C1R :
-            type == CV_32SC3 ? (IppiTranspose)ippiTranspose_32s_C3R :
-            type == CV_32SC4 ? (IppiTranspose)ippiTranspose_32s_C4R :
-            type == CV_32FC1 ? (IppiTranspose)ippiTranspose_32f_C1R :
-            type == CV_32FC3 ? (IppiTranspose)ippiTranspose_32f_C3R :
-            type == CV_32FC4 ? (IppiTranspose)ippiTranspose_32f_C4R : 0;
-    }
-
-    IppiSize roiSize = { src.cols, src.rows };
-    if (ippiTranspose != 0)
-    {
-        if (CV_INSTRUMENT_FUN_IPP(ippiTranspose, src.ptr(), (int)src.step, dst.ptr(), (int)dst.step, roiSize) >= 0)
-            return true;
-    }
-    else if (ippiTranspose_I != 0)
-    {
-        if (CV_INSTRUMENT_FUN_IPP(ippiTranspose_I, dst.ptr(), (int)dst.step, roiSize) >= 0)
-            return true;
-    }
-    return false;
-}
-#endif
-
-}
-
-
-void cv::transpose( InputArray _src, OutputArray _dst )
-{
-    CV_INSTRUMENT_REGION();
-
-    int type = _src.type(), esz = CV_ELEM_SIZE(type);
-    CV_Assert( _src.dims() <= 2 && esz <= 32 );
-
-    CV_OCL_RUN(_dst.isUMat(),
-               ocl_transpose(_src, _dst))
-
-    Mat src = _src.getMat();
-    if( src.empty() )
-    {
-        _dst.release();
-        return;
-    }
-
-    _dst.create(src.cols, src.rows, src.type());
-    Mat dst = _dst.getMat();
-
-    // handle the case of single-column/single-row matrices, stored in STL vectors.
-    if( src.rows != dst.cols || src.cols != dst.rows )
-    {
-        CV_Assert( src.size() == dst.size() && (src.cols == 1 || src.rows == 1) );
-        src.copyTo(dst);
-        return;
-    }
-
-    CV_IPP_RUN_FAST(ipp_transpose(src, dst))
-
-    if( dst.data == src.data )
-    {
-        TransposeInplaceFunc func = transposeInplaceTab[esz];
-        CV_Assert( func != 0 );
-        CV_Assert( dst.cols == dst.rows );
-        func( dst.ptr(), dst.step, dst.rows );
-    }
-    else
-    {
-        TransposeFunc func = transposeTab[esz];
-        CV_Assert( func != 0 );
-        func( src.ptr(), src.step, dst.ptr(), dst.step, src.size() );
-    }
-}
-
 
 ////////////////////////////////////// completeSymm /////////////////////////////////////////
 

modules/core/src/matrix_transform.cpp

@@ -0,0 +1,770 @@
+// This file is part of OpenCV project.
+// It is subject to the license terms in the LICENSE file found in the top-level directory
+// of this distribution and at http://opencv.org/license.html
+
+#include "precomp.hpp"
+#include "opencl_kernels_core.hpp"
+
+namespace cv {
+
+////////////////////////////////////// transpose /////////////////////////////////////////
+
+template<typename T> static void
+transpose_( const uchar* src, size_t sstep, uchar* dst, size_t dstep, Size sz )
+{
+    int i=0, j, m = sz.width, n = sz.height;
+
+    #if CV_ENABLE_UNROLLED
+    for(; i <= m - 4; i += 4 )
+    {
+        T* d0 = (T*)(dst + dstep*i);
+        T* d1 = (T*)(dst + dstep*(i+1));
+        T* d2 = (T*)(dst + dstep*(i+2));
+        T* d3 = (T*)(dst + dstep*(i+3));
+
+        for( j = 0; j <= n - 4; j += 4 )
+        {
+            const T* s0 = (const T*)(src + i*sizeof(T) + sstep*j);
+            const T* s1 = (const T*)(src + i*sizeof(T) + sstep*(j+1));
+            const T* s2 = (const T*)(src + i*sizeof(T) + sstep*(j+2));
+            const T* s3 = (const T*)(src + i*sizeof(T) + sstep*(j+3));
+
+            d0[j] = s0[0]; d0[j+1] = s1[0]; d0[j+2] = s2[0]; d0[j+3] = s3[0];
+            d1[j] = s0[1]; d1[j+1] = s1[1]; d1[j+2] = s2[1]; d1[j+3] = s3[1];
+            d2[j] = s0[2]; d2[j+1] = s1[2]; d2[j+2] = s2[2]; d2[j+3] = s3[2];
+            d3[j] = s0[3]; d3[j+1] = s1[3]; d3[j+2] = s2[3]; d3[j+3] = s3[3];
+        }
+
+        for( ; j < n; j++ )
+        {
+            const T* s0 = (const T*)(src + i*sizeof(T) + j*sstep);
+            d0[j] = s0[0]; d1[j] = s0[1]; d2[j] = s0[2]; d3[j] = s0[3];
+        }
+    }
+    #endif
+    for( ; i < m; i++ )
+    {
+        T* d0 = (T*)(dst + dstep*i);
+        j = 0;
+        #if CV_ENABLE_UNROLLED
+        for(; j <= n - 4; j += 4 )
+        {
+            const T* s0 = (const T*)(src + i*sizeof(T) + sstep*j);
+            const T* s1 = (const T*)(src + i*sizeof(T) + sstep*(j+1));
+            const T* s2 = (const T*)(src + i*sizeof(T) + sstep*(j+2));
+            const T* s3 = (const T*)(src + i*sizeof(T) + sstep*(j+3));
+
+            d0[j] = s0[0]; d0[j+1] = s1[0]; d0[j+2] = s2[0]; d0[j+3] = s3[0];
+        }
+        #endif
+        for( ; j < n; j++ )
+        {
+            const T* s0 = (const T*)(src + i*sizeof(T) + j*sstep);
+            d0[j] = s0[0];
+        }
+    }
+}
+
+template<typename T> static void
+transposeI_( uchar* data, size_t step, int n )
+{
+    for( int i = 0; i < n; i++ )
+    {
+        T* row = (T*)(data + step*i);
+        uchar* data1 = data + i*sizeof(T);
+        for( int j = i+1; j < n; j++ )
+            std::swap( row[j], *(T*)(data1 + step*j) );
+    }
+}
+
+typedef void (*TransposeFunc)( const uchar* src, size_t sstep, uchar* dst, size_t dstep, Size sz );
+typedef void (*TransposeInplaceFunc)( uchar* data, size_t step, int n );
+
+#define DEF_TRANSPOSE_FUNC(suffix, type) \
+static void transpose_##suffix( const uchar* src, size_t sstep, uchar* dst, size_t dstep, Size sz ) \
+{ transpose_<type>(src, sstep, dst, dstep, sz); } \
+\
+static void transposeI_##suffix( uchar* data, size_t step, int n ) \
+{ transposeI_<type>(data, step, n); }
+
+DEF_TRANSPOSE_FUNC(8u, uchar)
+DEF_TRANSPOSE_FUNC(16u, ushort)
+DEF_TRANSPOSE_FUNC(8uC3, Vec3b)
+DEF_TRANSPOSE_FUNC(32s, int)
+DEF_TRANSPOSE_FUNC(16uC3, Vec3s)
+DEF_TRANSPOSE_FUNC(32sC2, Vec2i)
+DEF_TRANSPOSE_FUNC(32sC3, Vec3i)
+DEF_TRANSPOSE_FUNC(32sC4, Vec4i)
+DEF_TRANSPOSE_FUNC(32sC6, Vec6i)
+DEF_TRANSPOSE_FUNC(32sC8, Vec8i)
+
+static TransposeFunc transposeTab[] =
+{
+    0, transpose_8u, transpose_16u, transpose_8uC3, transpose_32s, 0, transpose_16uC3, 0,
+    transpose_32sC2, 0, 0, 0, transpose_32sC3, 0, 0, 0, transpose_32sC4,
+    0, 0, 0, 0, 0, 0, 0, transpose_32sC6, 0, 0, 0, 0, 0, 0, 0, transpose_32sC8
+};
+
+static TransposeInplaceFunc transposeInplaceTab[] =
+{
+    0, transposeI_8u, transposeI_16u, transposeI_8uC3, transposeI_32s, 0, transposeI_16uC3, 0,
+    transposeI_32sC2, 0, 0, 0, transposeI_32sC3, 0, 0, 0, transposeI_32sC4,
+    0, 0, 0, 0, 0, 0, 0, transposeI_32sC6, 0, 0, 0, 0, 0, 0, 0, transposeI_32sC8
+};
+
+#ifdef HAVE_OPENCL
+
+static bool ocl_transpose( InputArray _src, OutputArray _dst )
+{
+    const ocl::Device & dev = ocl::Device::getDefault();
+    const int TILE_DIM = 32, BLOCK_ROWS = 8;
+    int type = _src.type(), cn = CV_MAT_CN(type), depth = CV_MAT_DEPTH(type),
+        rowsPerWI = dev.isIntel() ? 4 : 1;
+
+    UMat src = _src.getUMat();
+    _dst.create(src.cols, src.rows, type);
+    UMat dst = _dst.getUMat();
+
+    String kernelName("transpose");
+    bool inplace = dst.u == src.u;
+
+    if (inplace)
+    {
+        CV_Assert(dst.cols == dst.rows);
+        kernelName += "_inplace";
+    }
+    else
+    {
+        // check required local memory size
+        size_t required_local_memory = (size_t) TILE_DIM*(TILE_DIM+1)*CV_ELEM_SIZE(type);
+        if (required_local_memory > ocl::Device::getDefault().localMemSize())
+            return false;
+    }
+
+    ocl::Kernel k(kernelName.c_str(), ocl::core::transpose_oclsrc,
+                  format("-D T=%s -D T1=%s -D cn=%d -D TILE_DIM=%d -D BLOCK_ROWS=%d -D rowsPerWI=%d%s",
+                         ocl::memopTypeToStr(type), ocl::memopTypeToStr(depth),
+                         cn, TILE_DIM, BLOCK_ROWS, rowsPerWI, inplace ? " -D INPLACE" : ""));
+    if (k.empty())
+        return false;
+
+    if (inplace)
+        k.args(ocl::KernelArg::ReadWriteNoSize(dst), dst.rows);
+    else
+        k.args(ocl::KernelArg::ReadOnly(src),
+               ocl::KernelArg::WriteOnlyNoSize(dst));
+
+    size_t localsize[2]  = { TILE_DIM, BLOCK_ROWS };
+    size_t globalsize[2] = { (size_t)src.cols, inplace ? ((size_t)src.rows + rowsPerWI - 1) / rowsPerWI : (divUp((size_t)src.rows, TILE_DIM) * BLOCK_ROWS) };
+
+    if (inplace && dev.isIntel())
+    {
+        localsize[0] = 16;
+        localsize[1] = dev.maxWorkGroupSize() / localsize[0];
+    }
+
+    return k.run(2, globalsize, localsize, false);
+}
+
+#endif
+
+#ifdef HAVE_IPP
+static bool ipp_transpose( Mat &src, Mat &dst )
+{
+    CV_INSTRUMENT_REGION_IPP();
+
+    int type = src.type();
+    typedef IppStatus (CV_STDCALL * IppiTranspose)(const void * pSrc, int srcStep, void * pDst, int dstStep, IppiSize roiSize);
+    typedef IppStatus (CV_STDCALL * IppiTransposeI)(const void * pSrcDst, int srcDstStep, IppiSize roiSize);
+    IppiTranspose ippiTranspose = 0;
+    IppiTransposeI ippiTranspose_I = 0;
+
+    if (dst.data == src.data && dst.cols == dst.rows)
+    {
+        CV_SUPPRESS_DEPRECATED_START
+        ippiTranspose_I =
+            type == CV_8UC1 ? (IppiTransposeI)ippiTranspose_8u_C1IR :
+            type == CV_8UC3 ? (IppiTransposeI)ippiTranspose_8u_C3IR :
+            type == CV_8UC4 ? (IppiTransposeI)ippiTranspose_8u_C4IR :
+            type == CV_16UC1 ? (IppiTransposeI)ippiTranspose_16u_C1IR :
+            type == CV_16UC3 ? (IppiTransposeI)ippiTranspose_16u_C3IR :
+            type == CV_16UC4 ? (IppiTransposeI)ippiTranspose_16u_C4IR :
+            type == CV_16SC1 ? (IppiTransposeI)ippiTranspose_16s_C1IR :
+            type == CV_16SC3 ? (IppiTransposeI)ippiTranspose_16s_C3IR :
+            type == CV_16SC4 ? (IppiTransposeI)ippiTranspose_16s_C4IR :
+            type == CV_32SC1 ? (IppiTransposeI)ippiTranspose_32s_C1IR :
+            type == CV_32SC3 ? (IppiTransposeI)ippiTranspose_32s_C3IR :
+            type == CV_32SC4 ? (IppiTransposeI)ippiTranspose_32s_C4IR :
+            type == CV_32FC1 ? (IppiTransposeI)ippiTranspose_32f_C1IR :
+            type == CV_32FC3 ? (IppiTransposeI)ippiTranspose_32f_C3IR :
+            type == CV_32FC4 ? (IppiTransposeI)ippiTranspose_32f_C4IR : 0;
+        CV_SUPPRESS_DEPRECATED_END
+    }
+    else
+    {
+        ippiTranspose =
+            type == CV_8UC1 ? (IppiTranspose)ippiTranspose_8u_C1R :
+            type == CV_8UC3 ? (IppiTranspose)ippiTranspose_8u_C3R :
+            type == CV_8UC4 ? (IppiTranspose)ippiTranspose_8u_C4R :
+            type == CV_16UC1 ? (IppiTranspose)ippiTranspose_16u_C1R :
+            type == CV_16UC3 ? (IppiTranspose)ippiTranspose_16u_C3R :
+            type == CV_16UC4 ? (IppiTranspose)ippiTranspose_16u_C4R :
+            type == CV_16SC1 ? (IppiTranspose)ippiTranspose_16s_C1R :
+            type == CV_16SC3 ? (IppiTranspose)ippiTranspose_16s_C3R :
+            type == CV_16SC4 ? (IppiTranspose)ippiTranspose_16s_C4R :
+            type == CV_32SC1 ? (IppiTranspose)ippiTranspose_32s_C1R :
+            type == CV_32SC3 ? (IppiTranspose)ippiTranspose_32s_C3R :
+            type == CV_32SC4 ? (IppiTranspose)ippiTranspose_32s_C4R :
+            type == CV_32FC1 ? (IppiTranspose)ippiTranspose_32f_C1R :
+            type == CV_32FC3 ? (IppiTranspose)ippiTranspose_32f_C3R :
+            type == CV_32FC4 ? (IppiTranspose)ippiTranspose_32f_C4R : 0;
+    }
+
+    IppiSize roiSize = { src.cols, src.rows };
+    if (ippiTranspose != 0)
+    {
+        if (CV_INSTRUMENT_FUN_IPP(ippiTranspose, src.ptr(), (int)src.step, dst.ptr(), (int)dst.step, roiSize) >= 0)
+            return true;
+    }
+    else if (ippiTranspose_I != 0)
+    {
+        if (CV_INSTRUMENT_FUN_IPP(ippiTranspose_I, dst.ptr(), (int)dst.step, roiSize) >= 0)
+            return true;
+    }
+    return false;
+}
+#endif
+
+
+void transpose( InputArray _src, OutputArray _dst )
+{
+    CV_INSTRUMENT_REGION();
+
+    int type = _src.type(), esz = CV_ELEM_SIZE(type);
+    CV_Assert( _src.dims() <= 2 && esz <= 32 );
+
+    CV_OCL_RUN(_dst.isUMat(),
+               ocl_transpose(_src, _dst))
+
+    Mat src = _src.getMat();
+    if( src.empty() )
+    {
+        _dst.release();
+        return;
+    }
+
+    _dst.create(src.cols, src.rows, src.type());
+    Mat dst = _dst.getMat();
+
+    // handle the case of single-column/single-row matrices, stored in STL vectors.
+    if( src.rows != dst.cols || src.cols != dst.rows )
+    {
+        CV_Assert( src.size() == dst.size() && (src.cols == 1 || src.rows == 1) );
+        src.copyTo(dst);
+        return;
+    }
+
+    CV_IPP_RUN_FAST(ipp_transpose(src, dst))
+
+    if( dst.data == src.data )
+    {
+        TransposeInplaceFunc func = transposeInplaceTab[esz];
+        CV_Assert( func != 0 );
+        CV_Assert( dst.cols == dst.rows );
+        func( dst.ptr(), dst.step, dst.rows );
+    }
+    else
+    {
+        TransposeFunc func = transposeTab[esz];
+        CV_Assert( func != 0 );
+        func( src.ptr(), src.step, dst.ptr(), dst.step, src.size() );
+    }
+}
+
+
+#if CV_SIMD128
+template<typename V> CV_ALWAYS_INLINE void flipHoriz_single( const uchar* src, size_t sstep, uchar* dst, size_t dstep, Size size, size_t esz )
+{
+    typedef typename V::lane_type T;
+    int end = (int)(size.width*esz);
+    int width = (end + 1)/2;
+    int width_1 = width & -v_uint8x16::nlanes;
+    int i, j;
+
+#if CV_STRONG_ALIGNMENT
+    CV_Assert(isAligned<sizeof(T)>(src, dst));
+#endif
+
+    for( ; size.height--; src += sstep, dst += dstep )
+    {
+        for( i = 0, j = end; i < width_1; i += v_uint8x16::nlanes, j -= v_uint8x16::nlanes )
+        {
+            V t0, t1;
+
+            t0 = v_load((T*)((uchar*)src + i));
+            t1 = v_load((T*)((uchar*)src + j - v_uint8x16::nlanes));
+            t0 = v_reverse(t0);
+            t1 = v_reverse(t1);
+            v_store((T*)(dst + j - v_uint8x16::nlanes), t0);
+            v_store((T*)(dst + i), t1);
+        }
+        if (isAligned<sizeof(T)>(src, dst))
+        {
+            for ( ; i < width; i += sizeof(T), j -= sizeof(T) )
+            {
+                T t0, t1;
+
+                t0 = *((T*)((uchar*)src + i));
+                t1 = *((T*)((uchar*)src + j - sizeof(T)));
+                *((T*)(dst + j - sizeof(T))) = t0;
+                *((T*)(dst + i)) = t1;
+            }
+        }
+        else
+        {
+            for ( ; i < width; i += sizeof(T), j -= sizeof(T) )
+            {
+                for (int k = 0; k < (int)sizeof(T); k++)
+                {
+                    uchar t0, t1;
+
+                    t0 = *((uchar*)src + i + k);
+                    t1 = *((uchar*)src + j + k - sizeof(T));
+                    *(dst + j + k - sizeof(T)) = t0;
+                    *(dst + i + k) = t1;
+                }
+            }
+        }
+    }
+}
+
+template<typename T1, typename T2> CV_ALWAYS_INLINE void flipHoriz_double( const uchar* src, size_t sstep, uchar* dst, size_t dstep, Size size, size_t esz )
+{
+    int end = (int)(size.width*esz);
+    int width = (end + 1)/2;
+
+#if CV_STRONG_ALIGNMENT
+    CV_Assert(isAligned<sizeof(T1)>(src, dst));
+    CV_Assert(isAligned<sizeof(T2)>(src, dst));
+#endif
+
+    for( ; size.height--; src += sstep, dst += dstep )
+    {
+        for ( int i = 0, j = end; i < width; i += sizeof(T1) + sizeof(T2), j -= sizeof(T1) + sizeof(T2) )
+        {
+            T1 t0, t1;
+            T2 t2, t3;
+
+            t0 = *((T1*)((uchar*)src + i));
+            t2 = *((T2*)((uchar*)src + i + sizeof(T1)));
+            t1 = *((T1*)((uchar*)src + j - sizeof(T1) - sizeof(T2)));
+            t3 = *((T2*)((uchar*)src + j - sizeof(T2)));
+            *((T1*)(dst + j - sizeof(T1) - sizeof(T2))) = t0;
+            *((T2*)(dst + j - sizeof(T2))) = t2;
+            *((T1*)(dst + i)) = t1;
+            *((T2*)(dst + i + sizeof(T1))) = t3;
+        }
+    }
+}
+#endif
+
+static void
+flipHoriz( const uchar* src, size_t sstep, uchar* dst, size_t dstep, Size size, size_t esz )
+{
+#if CV_SIMD
+#if CV_STRONG_ALIGNMENT
+    size_t alignmentMark = ((size_t)src)|((size_t)dst)|sstep|dstep;
+#endif
+    if (esz == 2 * v_uint8x16::nlanes)
+    {
+        int end = (int)(size.width*esz);
+        int width = end/2;
+
+        for( ; size.height--; src += sstep, dst += dstep )
+        {
+            for( int i = 0, j = end - 2 * v_uint8x16::nlanes; i < width; i += 2 * v_uint8x16::nlanes, j -= 2 * v_uint8x16::nlanes )
+            {
+#if CV_SIMD256
+                v_uint8x32 t0, t1;
+
+                t0 = v256_load((uchar*)src + i);
+                t1 = v256_load((uchar*)src + j);
+                v_store(dst + j, t0);
+                v_store(dst + i, t1);
+#else
+                v_uint8x16 t0, t1, t2, t3;
+
+                t0 = v_load((uchar*)src + i);
+                t1 = v_load((uchar*)src + i + v_uint8x16::nlanes);
+                t2 = v_load((uchar*)src + j);
+                t3 = v_load((uchar*)src + j + v_uint8x16::nlanes);
+                v_store(dst + j, t0);
+                v_store(dst + j + v_uint8x16::nlanes, t1);
+                v_store(dst + i, t2);
+                v_store(dst + i + v_uint8x16::nlanes, t3);
+#endif
+            }
+        }
+    }
+    else if (esz == v_uint8x16::nlanes)
+    {
+        int end = (int)(size.width*esz);
+        int width = end/2;
+
+        for( ; size.height--; src += sstep, dst += dstep )
+        {
+            for( int i = 0, j = end - v_uint8x16::nlanes; i < width; i += v_uint8x16::nlanes, j -= v_uint8x16::nlanes )
+            {
+                v_uint8x16 t0, t1;
+
+                t0 = v_load((uchar*)src + i);
+                t1 = v_load((uchar*)src + j);
+                v_store(dst + j, t0);
+                v_store(dst + i, t1);
+            }
+        }
+    }
+    else if (esz == 8
+#if CV_STRONG_ALIGNMENT
+            && isAligned<sizeof(uint64)>(alignmentMark)
+#endif
+    )
+    {
+        flipHoriz_single<v_uint64x2>(src, sstep, dst, dstep, size, esz);
+    }
+    else if (esz == 4
+#if CV_STRONG_ALIGNMENT
+            && isAligned<sizeof(unsigned)>(alignmentMark)
+#endif
+    )
+    {
+        flipHoriz_single<v_uint32x4>(src, sstep, dst, dstep, size, esz);
+    }
+    else if (esz == 2
+#if CV_STRONG_ALIGNMENT
+            && isAligned<sizeof(ushort)>(alignmentMark)
+#endif
+    )
+    {
+        flipHoriz_single<v_uint16x8>(src, sstep, dst, dstep, size, esz);
+    }
+    else if (esz == 1)
+    {
+        flipHoriz_single<v_uint8x16>(src, sstep, dst, dstep, size, esz);
+    }
+    else if (esz == 24
+#if CV_STRONG_ALIGNMENT
+            && isAligned<sizeof(uint64_t)>(alignmentMark)
+#endif
+    )
+    {
+        int end = (int)(size.width*esz);
+        int width = (end + 1)/2;
+
+        for( ; size.height--; src += sstep, dst += dstep )
+        {
+            for ( int i = 0, j = end; i < width; i += v_uint8x16::nlanes + sizeof(uint64_t), j -= v_uint8x16::nlanes + sizeof(uint64_t) )
+            {
+                v_uint8x16 t0, t1;
+                uint64_t t2, t3;
+
+                t0 = v_load((uchar*)src + i);
+                t2 = *((uint64_t*)((uchar*)src + i + v_uint8x16::nlanes));
+                t1 = v_load((uchar*)src + j - v_uint8x16::nlanes - sizeof(uint64_t));
+                t3 = *((uint64_t*)((uchar*)src + j - sizeof(uint64_t)));
+                v_store(dst + j - v_uint8x16::nlanes - sizeof(uint64_t), t0);
+                *((uint64_t*)(dst + j - sizeof(uint64_t))) = t2;
+                v_store(dst + i, t1);
+                *((uint64_t*)(dst + i + v_uint8x16::nlanes)) = t3;
+            }
+        }
+    }
+#if !CV_STRONG_ALIGNMENT
+    else if (esz == 12)
+    {
+        flipHoriz_double<uint64_t,uint>(src, sstep, dst, dstep, size, esz);
+    }
+    else if (esz == 6)
+    {
+        flipHoriz_double<uint,ushort>(src, sstep, dst, dstep, size, esz);
+    }
+    else if (esz == 3)
+    {
+        flipHoriz_double<ushort,uchar>(src, sstep, dst, dstep, size, esz);
+    }
+#endif
+    else
+#endif // CV_SIMD
+    {
+        int i, j, limit = (int)(((size.width + 1)/2)*esz);
+        AutoBuffer<int> _tab(size.width*esz);
+        int* tab = _tab.data();
+
+        for( i = 0; i < size.width; i++ )
+            for( size_t k = 0; k < esz; k++ )
+                tab[i*esz + k] = (int)((size.width - i - 1)*esz + k);
+
+        for( ; size.height--; src += sstep, dst += dstep )
+        {
+            for( i = 0; i < limit; i++ )
+            {
+                j = tab[i];
+                uchar t0 = src[i], t1 = src[j];
+                dst[i] = t1; dst[j] = t0;
+            }
+        }
+    }
+}
+
+static void
+flipVert( const uchar* src0, size_t sstep, uchar* dst0, size_t dstep, Size size, size_t esz )
+{
+    const uchar* src1 = src0 + (size.height - 1)*sstep;
+    uchar* dst1 = dst0 + (size.height - 1)*dstep;
+    size.width *= (int)esz;
+
+    for( int y = 0; y < (size.height + 1)/2; y++, src0 += sstep, src1 -= sstep,
+                                                  dst0 += dstep, dst1 -= dstep )
+    {
+        int i = 0;
+#if CV_SIMD
+#if CV_STRONG_ALIGNMENT
+        if (isAligned<sizeof(int)>(src0, src1, dst0, dst1))
+#endif
+        {
+            for (; i <= size.width - CV_SIMD_WIDTH; i += CV_SIMD_WIDTH)
+            {
+                v_int32 t0 = vx_load((int*)(src0 + i));
+                v_int32 t1 = vx_load((int*)(src1 + i));
+                v_store((int*)(dst0 + i), t1);
+                v_store((int*)(dst1 + i), t0);
+            }
+        }
+#if CV_STRONG_ALIGNMENT
+        else
+        {
+            for (; i <= size.width - CV_SIMD_WIDTH; i += CV_SIMD_WIDTH)
+            {
+                v_uint8 t0 = vx_load(src0 + i);
+                v_uint8 t1 = vx_load(src1 + i);
+                v_store(dst0 + i, t1);
+                v_store(dst1 + i, t0);
+            }
+        }
+#endif
+#endif
+
+        if (isAligned<sizeof(int)>(src0, src1, dst0, dst1))
+        {
+            for( ; i <= size.width - 16; i += 16 )
+            {
+                int t0 = ((int*)(src0 + i))[0];
+                int t1 = ((int*)(src1 + i))[0];
+
+                ((int*)(dst0 + i))[0] = t1;
+                ((int*)(dst1 + i))[0] = t0;
+
+                t0 = ((int*)(src0 + i))[1];
+                t1 = ((int*)(src1 + i))[1];
+
+                ((int*)(dst0 + i))[1] = t1;
+                ((int*)(dst1 + i))[1] = t0;
+
+                t0 = ((int*)(src0 + i))[2];
+                t1 = ((int*)(src1 + i))[2];
+
+                ((int*)(dst0 + i))[2] = t1;
+                ((int*)(dst1 + i))[2] = t0;
+
+                t0 = ((int*)(src0 + i))[3];
+                t1 = ((int*)(src1 + i))[3];
+
+                ((int*)(dst0 + i))[3] = t1;
+                ((int*)(dst1 + i))[3] = t0;
+            }
+
+            for( ; i <= size.width - 4; i += 4 )
+            {
+                int t0 = ((int*)(src0 + i))[0];
+                int t1 = ((int*)(src1 + i))[0];
+
+                ((int*)(dst0 + i))[0] = t1;
+                ((int*)(dst1 + i))[0] = t0;
+            }
+        }
+
+        for( ; i < size.width; i++ )
+        {
+            uchar t0 = src0[i];
+            uchar t1 = src1[i];
+
+            dst0[i] = t1;
+            dst1[i] = t0;
+        }
+    }
+}
+
+#ifdef HAVE_OPENCL
+
+enum { FLIP_COLS = 1 << 0, FLIP_ROWS = 1 << 1, FLIP_BOTH = FLIP_ROWS | FLIP_COLS };
+
+static bool ocl_flip(InputArray _src, OutputArray _dst, int flipCode )
+{
+    CV_Assert(flipCode >= -1 && flipCode <= 1);
+
+    const ocl::Device & dev = ocl::Device::getDefault();
+    int type = _src.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type),
+            flipType, kercn = std::min(ocl::predictOptimalVectorWidth(_src, _dst), 4);
+
+    bool doubleSupport = dev.doubleFPConfig() > 0;
+    if (!doubleSupport && depth == CV_64F)
+        kercn = cn;
+
+    if (cn > 4)
+        return false;
+
+    const char * kernelName;
+    if (flipCode == 0)
+        kernelName = "arithm_flip_rows", flipType = FLIP_ROWS;
+    else if (flipCode > 0)
+        kernelName = "arithm_flip_cols", flipType = FLIP_COLS;
+    else
+        kernelName = "arithm_flip_rows_cols", flipType = FLIP_BOTH;
+
+    int pxPerWIy = (dev.isIntel() && (dev.type() & ocl::Device::TYPE_GPU)) ? 4 : 1;
+    kercn = (cn!=3 || flipType == FLIP_ROWS) ? std::max(kercn, cn) : cn;
+
+    ocl::Kernel k(kernelName, ocl::core::flip_oclsrc,
+        format( "-D T=%s -D T1=%s -D DEPTH=%d -D cn=%d -D PIX_PER_WI_Y=%d -D kercn=%d",
+                kercn != cn ? ocl::typeToStr(CV_MAKE_TYPE(depth, kercn)) : ocl::vecopTypeToStr(CV_MAKE_TYPE(depth, kercn)),
+                kercn != cn ? ocl::typeToStr(depth) : ocl::vecopTypeToStr(depth), depth, cn, pxPerWIy, kercn));
+    if (k.empty())
+        return false;
+
+    Size size = _src.size();
+    _dst.create(size, type);
+    UMat src = _src.getUMat(), dst = _dst.getUMat();
+
+    int cols = size.width * cn / kercn, rows = size.height;
+    cols = flipType == FLIP_COLS ? (cols + 1) >> 1 : cols;
+    rows = flipType & FLIP_ROWS ? (rows + 1) >> 1 : rows;
+
+    k.args(ocl::KernelArg::ReadOnlyNoSize(src),
+           ocl::KernelArg::WriteOnly(dst, cn, kercn), rows, cols);
+
+    size_t maxWorkGroupSize = dev.maxWorkGroupSize();
+    CV_Assert(maxWorkGroupSize % 4 == 0);
+
+    size_t globalsize[2] = { (size_t)cols, ((size_t)rows + pxPerWIy - 1) / pxPerWIy },
+            localsize[2] = { maxWorkGroupSize / 4, 4 };
+    return k.run(2, globalsize, (flipType == FLIP_COLS) && !dev.isIntel() ? localsize : NULL, false);
+}
+
+#endif
+
+#if defined HAVE_IPP
+static bool ipp_flip(Mat &src, Mat &dst, int flip_mode)
+{
+#ifdef HAVE_IPP_IW
+    CV_INSTRUMENT_REGION_IPP();
+
+    // Details: https://github.com/opencv/opencv/issues/12943
+    if (flip_mode <= 0 /* swap rows */
+        && cv::ipp::getIppTopFeatures() != ippCPUID_SSE42
+        && (int64_t)(src.total()) * src.elemSize() >= CV_BIG_INT(0x80000000)/*2Gb*/
+    )
+        return false;
+
+    IppiAxis ippMode;
+    if(flip_mode < 0)
+        ippMode = ippAxsBoth;
+    else if(flip_mode == 0)
+        ippMode = ippAxsHorizontal;
+    else
+        ippMode = ippAxsVertical;
+
+    try
+    {
+        ::ipp::IwiImage iwSrc = ippiGetImage(src);
+        ::ipp::IwiImage iwDst = ippiGetImage(dst);
+
+        CV_INSTRUMENT_FUN_IPP(::ipp::iwiMirror, iwSrc, iwDst, ippMode);
+    }
+    catch(const ::ipp::IwException &)
+    {
+        return false;
+    }
+
+    return true;
+#else
+    CV_UNUSED(src); CV_UNUSED(dst); CV_UNUSED(flip_mode);
+    return false;
+#endif
+}
+#endif
+
+
+void flip( InputArray _src, OutputArray _dst, int flip_mode )
+{
+    CV_INSTRUMENT_REGION();
+
+    CV_Assert( _src.dims() <= 2 );
+    Size size = _src.size();
+
+    if (flip_mode < 0)
+    {
+        if (size.width == 1)
+            flip_mode = 0;
+        if (size.height == 1)
+            flip_mode = 1;
+    }
+
+    if ((size.width == 1 && flip_mode > 0) ||
+        (size.height == 1 && flip_mode == 0))
+    {
+        return _src.copyTo(_dst);
+    }
+
+    CV_OCL_RUN( _dst.isUMat(), ocl_flip(_src, _dst, flip_mode))
+
+    Mat src = _src.getMat();
+    int type = src.type();
+    _dst.create( size, type );
+    Mat dst = _dst.getMat();
+
+    CV_IPP_RUN_FAST(ipp_flip(src, dst, flip_mode));
+
+    size_t esz = CV_ELEM_SIZE(type);
+
+    if( flip_mode <= 0 )
+        flipVert( src.ptr(), src.step, dst.ptr(), dst.step, src.size(), esz );
+    else
+        flipHoriz( src.ptr(), src.step, dst.ptr(), dst.step, src.size(), esz );
+
+    if( flip_mode < 0 )
+        flipHoriz( dst.ptr(), dst.step, dst.ptr(), dst.step, dst.size(), esz );
+}
+
+void rotate(InputArray _src, OutputArray _dst, int rotateMode)
+{
+    CV_Assert(_src.dims() <= 2);
+
+    switch (rotateMode)
+    {
+    case ROTATE_90_CLOCKWISE:
+        transpose(_src, _dst);
+        flip(_dst, _dst, 1);
+        break;
+    case ROTATE_180:
+        flip(_src, _dst, -1);
+        break;
+    case ROTATE_90_COUNTERCLOCKWISE:
+        transpose(_src, _dst);
+        flip(_dst, _dst, 0);
+        break;
+    default:
+        break;
+    }
+}
+
+}  // namespace

modules/core/src/matrix_wrap.cpp

@@ -316,6 +316,7 @@ void _InputArray::getUMatVector(std::vector<UMat>& umv) const
 
 cuda::GpuMat _InputArray::getGpuMat() const
 {
+#ifdef HAVE_CUDA
     int k = kind();
 
     if (k == CUDA_GPU_MAT)
@@ -339,14 +340,22 @@ cuda::GpuMat _InputArray::getGpuMat() const
         return cuda::GpuMat();
 
     CV_Error(cv::Error::StsNotImplemented, "getGpuMat is available only for cuda::GpuMat and cuda::HostMem");
+#else
+    CV_Error(Error::StsNotImplemented, "CUDA support is not enabled in this OpenCV build (missing HAVE_CUDA)");
+#endif
 }
 void _InputArray::getGpuMatVector(std::vector<cuda::GpuMat>& gpumv) const
 {
+#ifdef HAVE_CUDA
     int k = kind();
     if (k == STD_VECTOR_CUDA_GPU_MAT)
     {
         gpumv = *(std::vector<cuda::GpuMat>*)obj;
     }
+#else
+    CV_UNUSED(gpumv);
+    CV_Error(Error::StsNotImplemented, "CUDA support is not enabled in this OpenCV build (missing HAVE_CUDA)");
+#endif
 }
 ogl::Buffer _InputArray::getOGlBuffer() const
 {
@@ -457,11 +466,15 @@ Size _InputArray::size(int i) const
 
     if (k == STD_VECTOR_CUDA_GPU_MAT)
     {
+#ifdef HAVE_CUDA
         const std::vector<cuda::GpuMat>& vv = *(const std::vector<cuda::GpuMat>*)obj;
         if (i < 0)
             return vv.empty() ? Size() : Size((int)vv.size(), 1);
         CV_Assert(i < (int)vv.size());
         return vv[i].size();
+#else
+        CV_Error(Error::StsNotImplemented, "CUDA support is not enabled in this OpenCV build (missing HAVE_CUDA)");
+#endif
     }
 
     if( k == STD_VECTOR_UMAT )
@@ -795,6 +808,7 @@ int _InputArray::type(int i) const
 
     if (k == STD_VECTOR_CUDA_GPU_MAT)
     {
+#ifdef HAVE_CUDA
         const std::vector<cuda::GpuMat>& vv = *(const std::vector<cuda::GpuMat>*)obj;
         if (vv.empty())
         {
@@ -803,6 +817,9 @@ int _InputArray::type(int i) const
         }
         CV_Assert(i < (int)vv.size());
         return vv[i >= 0 ? i : 0].type();
+#else
+        CV_Error(Error::StsNotImplemented, "CUDA support is not enabled in this OpenCV build (missing HAVE_CUDA)");
+#endif
     }
 
     if( k == OPENGL_BUFFER )
@@ -1164,22 +1181,34 @@ void _OutputArray::create(Size _sz, int mtype, int i, bool allowTransposed, int
     {
         CV_Assert(!fixedSize() || ((cuda::GpuMat*)obj)->size() == _sz);
         CV_Assert(!fixedType() || ((cuda::GpuMat*)obj)->type() == mtype);
+#ifdef HAVE_CUDA
         ((cuda::GpuMat*)obj)->create(_sz, mtype);
         return;
+#else
+        CV_Error(Error::StsNotImplemented, "CUDA support is not enabled in this OpenCV build (missing HAVE_CUDA)");
+#endif
     }
     if( k == OPENGL_BUFFER && i < 0 && !allowTransposed && fixedDepthMask == 0 )
     {
         CV_Assert(!fixedSize() || ((ogl::Buffer*)obj)->size() == _sz);
         CV_Assert(!fixedType() || ((ogl::Buffer*)obj)->type() == mtype);
+#ifdef HAVE_OPENGL
         ((ogl::Buffer*)obj)->create(_sz, mtype);
         return;
+#else
+        CV_Error(Error::StsNotImplemented, "OpenGL support is not enabled in this OpenCV build (missing HAVE_OPENGL)");
+#endif
     }
     if( k == CUDA_HOST_MEM && i < 0 && !allowTransposed && fixedDepthMask == 0 )
     {
         CV_Assert(!fixedSize() || ((cuda::HostMem*)obj)->size() == _sz);
         CV_Assert(!fixedType() || ((cuda::HostMem*)obj)->type() == mtype);
+#ifdef HAVE_CUDA
         ((cuda::HostMem*)obj)->create(_sz, mtype);
         return;
+#else
+        CV_Error(Error::StsNotImplemented, "CUDA support is not enabled in this OpenCV build (missing HAVE_CUDA)");
+#endif
     }
     int sizes[] = {_sz.height, _sz.width};
     create(2, sizes, mtype, i, allowTransposed, fixedDepthMask);
@@ -1206,22 +1235,34 @@ void _OutputArray::create(int _rows, int _cols, int mtype, int i, bool allowTran
     {
         CV_Assert(!fixedSize() || ((cuda::GpuMat*)obj)->size() == Size(_cols, _rows));
         CV_Assert(!fixedType() || ((cuda::GpuMat*)obj)->type() == mtype);
+#ifdef HAVE_CUDA
         ((cuda::GpuMat*)obj)->create(_rows, _cols, mtype);
         return;
+#else
+        CV_Error(Error::StsNotImplemented, "CUDA support is not enabled in this OpenCV build (missing HAVE_CUDA)");
+#endif
     }
     if( k == OPENGL_BUFFER && i < 0 && !allowTransposed && fixedDepthMask == 0 )
     {
         CV_Assert(!fixedSize() || ((ogl::Buffer*)obj)->size() == Size(_cols, _rows));
         CV_Assert(!fixedType() || ((ogl::Buffer*)obj)->type() == mtype);
+#ifdef HAVE_OPENGL
         ((ogl::Buffer*)obj)->create(_rows, _cols, mtype);
         return;
+#else
+        CV_Error(Error::StsNotImplemented, "OpenGL support is not enabled in this OpenCV build (missing HAVE_OPENGL)");
+#endif
     }
     if( k == CUDA_HOST_MEM && i < 0 && !allowTransposed && fixedDepthMask == 0 )
     {
         CV_Assert(!fixedSize() || ((cuda::HostMem*)obj)->size() == Size(_cols, _rows));
         CV_Assert(!fixedType() || ((cuda::HostMem*)obj)->type() == mtype);
+#ifdef HAVE_CUDA
         ((cuda::HostMem*)obj)->create(_rows, _cols, mtype);
         return;
+#else
+        CV_Error(Error::StsNotImplemented, "CUDA support is not enabled in this OpenCV build (missing HAVE_CUDA)");
+#endif
     }
     int sizes[] = {_rows, _cols};
     create(2, sizes, mtype, i, allowTransposed, fixedDepthMask);
@@ -1644,20 +1685,32 @@ void _OutputArray::release() const
 
     if( k == CUDA_GPU_MAT )
     {
+#ifdef HAVE_CUDA
         ((cuda::GpuMat*)obj)->release();
         return;
+#else
+        CV_Error(Error::StsNotImplemented, "CUDA support is not enabled in this OpenCV build (missing HAVE_CUDA)");
+#endif
     }
 
     if( k == CUDA_HOST_MEM )
     {
+#ifdef HAVE_CUDA
         ((cuda::HostMem*)obj)->release();
         return;
+#else
+        CV_Error(Error::StsNotImplemented, "CUDA support is not enabled in this OpenCV build (missing HAVE_CUDA)");
+#endif
     }
 
     if( k == OPENGL_BUFFER )
     {
+#ifdef HAVE_OPENGL
         ((ogl::Buffer*)obj)->release();
         return;
+#else
+        CV_Error(Error::StsNotImplemented, "OpenGL support is not enabled in this OpenCV build (missing HAVE_OPENGL)");
+#endif
     }
 
     if( k == NONE )
@@ -1688,8 +1741,12 @@ void _OutputArray::release() const
     }
     if (k == STD_VECTOR_CUDA_GPU_MAT)
     {
+#ifdef HAVE_CUDA
         ((std::vector<cuda::GpuMat>*)obj)->clear();
         return;
+#else
+        CV_Error(Error::StsNotImplemented, "CUDA support is not enabled in this OpenCV build (missing HAVE_CUDA)");
+#endif
     }
     CV_Error(Error::StsNotImplemented, "Unknown/unsupported array type");
 }
@@ -1797,9 +1854,13 @@ void _OutputArray::setTo(const _InputArray& arr, const _InputArray & mask) const
         ((UMat*)obj)->setTo(arr, mask);
     else if( k == CUDA_GPU_MAT )
     {
+#ifdef HAVE_CUDA
         Mat value = arr.getMat();
         CV_Assert( checkScalar(value, type(), arr.kind(), _InputArray::CUDA_GPU_MAT) );
         ((cuda::GpuMat*)obj)->setTo(Scalar(Vec<double, 4>(value.ptr<double>())), mask);
+#else
+        CV_Error(Error::StsNotImplemented, "CUDA support is not enabled in this OpenCV build (missing HAVE_CUDA)");
+#endif
     }
     else
         CV_Error(Error::StsNotImplemented, "");

modules/core/src/norm.cpp

@@ -205,13 +205,10 @@ int normL1_(const uchar* a, const uchar* b, int n)
     return d;
 }
 
-}} //cv::hal
+} //cv::hal
 
 //==================================================================================================
 
-namespace cv
-{
-
 template<typename T, typename ST> int
 normInf_(const T* src, const uchar* mask, ST* _result, int len, int cn)
 {
@@ -591,12 +588,10 @@ static bool ipp_norm(Mat &src, int normType, Mat &mask, double &result)
     CV_UNUSED(src); CV_UNUSED(normType); CV_UNUSED(mask); CV_UNUSED(result);
 #endif
     return false;
-}
-#endif
+}  // ipp_norm()
+#endif  // HAVE_IPP
 
-} // cv::
-
-double cv::norm( InputArray _src, int normType, InputArray _mask )
+double norm( InputArray _src, int normType, InputArray _mask )
 {
     CV_INSTRUMENT_REGION();
 
@@ -769,9 +764,6 @@ double cv::norm( InputArray _src, int normType, InputArray _mask )
 //==================================================================================================
 
 #ifdef HAVE_OPENCL
-
-namespace cv {
-
 static bool ocl_norm( InputArray _src1, InputArray _src2, int normType, InputArray _mask, double & result )
 {
 #ifdef __ANDROID__
@@ -826,15 +818,10 @@ static bool ocl_norm( InputArray _src1, InputArray _src2, int normType, InputArr
         result /= (s2 + DBL_EPSILON);
 
     return true;
-}
-
-}
-
-#endif
+}  // ocl_norm()
+#endif  // HAVE_OPENCL
 
 #ifdef HAVE_IPP
-namespace cv
-{
 static bool ipp_norm(InputArray _src1, InputArray _src2, int normType, InputArray _mask, double &result)
 {
     CV_INSTRUMENT_REGION_IPP();
@@ -1060,12 +1047,11 @@ static bool ipp_norm(InputArray _src1, InputArray _src2, int normType, InputArra
     CV_UNUSED(_src1); CV_UNUSED(_src2); CV_UNUSED(normType); CV_UNUSED(_mask); CV_UNUSED(result);
 #endif
     return false;
-}
-}
-#endif
+}  // ipp_norm
+#endif  // HAVE_IPP
 
 
-double cv::norm( InputArray _src1, InputArray _src2, int normType, InputArray _mask )
+double norm( InputArray _src1, InputArray _src2, int normType, InputArray _mask )
 {
     CV_INSTRUMENT_REGION();
 
@@ -1185,7 +1171,7 @@ double cv::norm( InputArray _src1, InputArray _src2, int normType, InputArray _m
         // special case to handle "integer" overflow in accumulator
         const size_t esz = src1.elemSize();
         const int total = (int)it.size;
-        const int intSumBlockSize = normType == NORM_L1 && depth <= CV_8S ? (1 << 23) : (1 << 15);
+        const int intSumBlockSize = (normType == NORM_L1 && depth <= CV_8S ? (1 << 23) : (1 << 15))/cn;
         const int blockSize = std::min(total, intSumBlockSize);
         int isum = 0;
         int count = 0;
@@ -1234,12 +1220,12 @@ double cv::norm( InputArray _src1, InputArray _src2, int normType, InputArray _m
     return result.d;
 }
 
-cv::Hamming::ResultType cv::Hamming::operator()( const unsigned char* a, const unsigned char* b, int size ) const
+cv::Hamming::ResultType Hamming::operator()( const unsigned char* a, const unsigned char* b, int size ) const
 {
     return cv::hal::normHamming(a, b, size);
 }
 
-double cv::PSNR(InputArray _src1, InputArray _src2)
+double PSNR(InputArray _src1, InputArray _src2)
 {
     CV_INSTRUMENT_REGION();
 
@@ -1249,3 +1235,141 @@ double cv::PSNR(InputArray _src1, InputArray _src2)
     double diff = std::sqrt(norm(_src1, _src2, NORM_L2SQR)/(_src1.total()*_src1.channels()));
     return 20*log10(255./(diff+DBL_EPSILON));
 }
+
+
+#ifdef HAVE_OPENCL
+static bool ocl_normalize( InputArray _src, InputOutputArray _dst, InputArray _mask, int dtype,
+                           double scale, double delta )
+{
+    UMat src = _src.getUMat();
+
+    if( _mask.empty() )
+        src.convertTo( _dst, dtype, scale, delta );
+    else if (src.channels() <= 4)
+    {
+        const ocl::Device & dev = ocl::Device::getDefault();
+
+        int stype = _src.type(), sdepth = CV_MAT_DEPTH(stype), cn = CV_MAT_CN(stype),
+                ddepth = CV_MAT_DEPTH(dtype), wdepth = std::max(CV_32F, std::max(sdepth, ddepth)),
+                rowsPerWI = dev.isIntel() ? 4 : 1;
+
+        float fscale = static_cast<float>(scale), fdelta = static_cast<float>(delta);
+        bool haveScale = std::fabs(scale - 1) > DBL_EPSILON,
+                haveZeroScale = !(std::fabs(scale) > DBL_EPSILON),
+                haveDelta = std::fabs(delta) > DBL_EPSILON,
+                doubleSupport = dev.doubleFPConfig() > 0;
+
+        if (!haveScale && !haveDelta && stype == dtype)
+        {
+            _src.copyTo(_dst, _mask);
+            return true;
+        }
+        if (haveZeroScale)
+        {
+            _dst.setTo(Scalar(delta), _mask);
+            return true;
+        }
+
+        if ((sdepth == CV_64F || ddepth == CV_64F) && !doubleSupport)
+            return false;
+
+        char cvt[2][40];
+        String opts = format("-D srcT=%s -D dstT=%s -D convertToWT=%s -D cn=%d -D rowsPerWI=%d"
+                             " -D convertToDT=%s -D workT=%s%s%s%s -D srcT1=%s -D dstT1=%s",
+                             ocl::typeToStr(stype), ocl::typeToStr(dtype),
+                             ocl::convertTypeStr(sdepth, wdepth, cn, cvt[0]), cn,
+                             rowsPerWI, ocl::convertTypeStr(wdepth, ddepth, cn, cvt[1]),
+                             ocl::typeToStr(CV_MAKE_TYPE(wdepth, cn)),
+                             doubleSupport ? " -D DOUBLE_SUPPORT" : "",
+                             haveScale ? " -D HAVE_SCALE" : "",
+                             haveDelta ? " -D HAVE_DELTA" : "",
+                             ocl::typeToStr(sdepth), ocl::typeToStr(ddepth));
+
+        ocl::Kernel k("normalizek", ocl::core::normalize_oclsrc, opts);
+        if (k.empty())
+            return false;
+
+        UMat mask = _mask.getUMat(), dst = _dst.getUMat();
+
+        ocl::KernelArg srcarg = ocl::KernelArg::ReadOnlyNoSize(src),
+                maskarg = ocl::KernelArg::ReadOnlyNoSize(mask),
+                dstarg = ocl::KernelArg::ReadWrite(dst);
+
+        if (haveScale)
+        {
+            if (haveDelta)
+                k.args(srcarg, maskarg, dstarg, fscale, fdelta);
+            else
+                k.args(srcarg, maskarg, dstarg, fscale);
+        }
+        else
+        {
+            if (haveDelta)
+                k.args(srcarg, maskarg, dstarg, fdelta);
+            else
+                k.args(srcarg, maskarg, dstarg);
+        }
+
+        size_t globalsize[2] = { (size_t)src.cols, ((size_t)src.rows + rowsPerWI - 1) / rowsPerWI };
+        return k.run(2, globalsize, NULL, false);
+    }
+    else
+    {
+        UMat temp;
+        src.convertTo( temp, dtype, scale, delta );
+        temp.copyTo( _dst, _mask );
+    }
+
+    return true;
+}  // ocl_normalize
+#endif  // HAVE_OPENCL
+
+void normalize(InputArray _src, InputOutputArray _dst, double a, double b,
+               int norm_type, int rtype, InputArray _mask)
+{
+    CV_INSTRUMENT_REGION();
+
+    double scale = 1, shift = 0;
+    int type = _src.type(), depth = CV_MAT_DEPTH(type);
+
+    if( rtype < 0 )
+        rtype = _dst.fixedType() ? _dst.depth() : depth;
+
+    if( norm_type == CV_MINMAX )
+    {
+        double smin = 0, smax = 0;
+        double dmin = MIN( a, b ), dmax = MAX( a, b );
+        minMaxIdx( _src, &smin, &smax, 0, 0, _mask );
+        scale = (dmax - dmin)*(smax - smin > DBL_EPSILON ? 1./(smax - smin) : 0);
+        if( rtype == CV_32F )
+        {
+            scale = (float)scale;
+            shift = (float)dmin - (float)(smin*scale);
+        }
+        else
+            shift = dmin - smin*scale;
+    }
+    else if( norm_type == CV_L2 || norm_type == CV_L1 || norm_type == CV_C )
+    {
+        scale = norm( _src, norm_type, _mask );
+        scale = scale > DBL_EPSILON ? a/scale : 0.;
+        shift = 0;
+    }
+    else
+        CV_Error( CV_StsBadArg, "Unknown/unsupported norm type" );
+
+    CV_OCL_RUN(_dst.isUMat(),
+               ocl_normalize(_src, _dst, _mask, rtype, scale, shift))
+
+    Mat src = _src.getMat();
+    if( _mask.empty() )
+        src.convertTo( _dst, rtype, scale, shift );
+    else
+    {
+        Mat temp;
+        src.convertTo( temp, rtype, scale, shift );
+        temp.copyTo( _dst, _mask );
+    }
+}
+
+}  // namespace