release: OpenCV 3.4.10

Merge pull request #16959 from alalek:winpack_dldt_update
Merge pull request #16956 from alalek:ocl_control_buffer_mapping
2026-07-24 12:53:04 +04:00 · 2020-04-02 19:59:58 +00:00 · 2020-04-02 18:13:58 +00:00 · 2020-04-02 17:56:50 +00:00 · 2020-04-02 16:52:46 +00:00 · 2020-04-02 13:06:34 +00:00
1529 changed files with 123516 additions and 39559 deletions
@@ -22,3 +22,5 @@ bin/
 *.log
 *.tlog
 build
+node_modules
+CMakeSettings.json
@@ -20,8 +20,11 @@ if(CMAKE_COMPILER_IS_GNUCC)
    # - matchTemplate about 5-10%
    # - goodFeaturesToTrack 10-20%
    # - cornerHarris 30% for some cases
-
-    set_source_files_properties(${carotene_sources} COMPILE_FLAGS "--param ipcp-unit-growth=100000 --param inline-unit-growth=100000 --param large-stack-frame-growth=5000")
+    if(CMAKE_CXX_COMPILER_VERSION VERSION_LESS "10.0.0")
+        set_source_files_properties(${carotene_sources} COMPILE_FLAGS "--param ipcp-unit-growth=100000 --param inline-unit-growth=100000 --param large-stack-frame-growth=5000")
+    else()
+        set_source_files_properties(${carotene_sources} COMPILE_FLAGS "--param ipa-cp-unit-growth=100000 --param inline-unit-growth=100000 --param large-stack-frame-growth=5000")
+    endif()
 endif()

 add_library(carotene_objs OBJECT
@@ -58,22 +58,6 @@ function(compile_carotene)
  endif()

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

 compile_carotene()
@@ -88,7 +72,11 @@ set_property(DIRECTORY APPEND PROPERTY COMPILE_DEFINITIONS ${carotene_defs})
    #   matchTemplate about 5-10%
    #   goodFeaturesToTrack 10-20%
    #   cornerHarris 30% for some cases
-    set_source_files_properties(impl.cpp $<TARGET_OBJECTS:carotene_objs> COMPILE_FLAGS "--param ipcp-unit-growth=100000 --param inline-unit-growth=100000 --param large-stack-frame-growth=5000")
+    if(CMAKE_CXX_COMPILER_VERSION VERSION_LESS "10.0.0")
+      set_source_files_properties(impl.cpp $<TARGET_OBJECTS:carotene_objs> COMPILE_FLAGS "--param ipcp-unit-growth=100000 --param inline-unit-growth=100000 --param large-stack-frame-growth=5000")
+    else()
+      set_source_files_properties(impl.cpp $<TARGET_OBJECTS:carotene_objs> COMPILE_FLAGS "--param ipa-cp-unit-growth=100000 --param inline-unit-growth=100000 --param large-stack-frame-growth=5000")
+    endif()
 #    set_source_files_properties(impl.cpp $<TARGET_OBJECTS:carotene_objs> COMPILE_FLAGS "--param ipcp-unit-growth=100000 --param inline-unit-growth=100000 --param large-stack-frame-growth=5000")
  endif()

@@ -1127,6 +1127,7 @@ inline int TEGRA_FILTERINIT(cvhalFilter2D **context, uchar *kernel_data, size_t
        {
            std::memcpy(ctx->kernel_data + kernel_width * j, kernel_data + kernel_step * j, kernel_width * sizeof(int16_t));
        }
+        break;
    default:
        delete[] ctx->kernel_data;
        delete ctx;
@@ -1243,6 +1244,7 @@ inline int TEGRA_SEPFILTERINIT(cvhalFilter2D **context, int src_type, int dst_ty
        ctx->kernely_data[0]=((int16_t*)kernely_data)[0];
        ctx->kernely_data[1]=((int16_t*)kernely_data)[1];
        ctx->kernely_data[2]=((int16_t*)kernely_data)[2];
+        break;
    default:
        delete ctx;
        return CV_HAL_ERROR_NOT_IMPLEMENTED;
@@ -1,8 +1,8 @@
-# Binaries branch name: ffmpeg/3.4_20181103
-# Binaries were created for OpenCV: c6d2f0399a706028f476f3d08358ea787503020d
-ocv_update(FFMPEG_BINARIES_COMMIT "fe71c0ad807fdc33c2178e48e488f1e9b177c39a")
-ocv_update(FFMPEG_FILE_HASH_BIN32 "41b81bb9a50cabd4bea385f7b50a069a")
-ocv_update(FFMPEG_FILE_HASH_BIN64 "a9ea7dbbc8e5afd08e00e223a831b578")
+# Binaries branch name: ffmpeg/3.4_20200310
+# Binaries were created for OpenCV: 4966186e10e2a940514d8c20447ca4a828af5f46
+ocv_update(FFMPEG_BINARIES_COMMIT "e81ccda615672833b578c6cefdb859ad69c560ba")
+ocv_update(FFMPEG_FILE_HASH_BIN32 "301ae2000e25f800ab8e0065f277ad28")
+ocv_update(FFMPEG_FILE_HASH_BIN64 "d87ce032289c3f811d02f0c3d8dbe366")
 ocv_update(FFMPEG_FILE_HASH_CMAKE "3b90f67f4b429e77d3da36698cef700c")

 function(download_win_ffmpeg script_var)
@@ -2,37 +2,32 @@ function(download_ippicv root_var)
  set(${root_var} "" PARENT_SCOPE)

  # Commit SHA in the opencv_3rdparty repo
-  set(IPPICV_COMMIT "32e315a5b106a7b89dbed51c28f8120a48b368b4")
+  set(IPPICV_COMMIT "a56b6ac6f030c312b2dce17430eef13aed9af274")
  # Define actual ICV versions
  if(APPLE)
    set(OPENCV_ICV_PLATFORM "macosx")
    set(OPENCV_ICV_PACKAGE_SUBDIR "ippicv_mac")
-    if(X86_64)
-      set(OPENCV_ICV_NAME "ippicv_2019_mac_intel64_general_20180723.tgz")
-      set(OPENCV_ICV_HASH "fe6b2bb75ae0e3f19ad3ae1a31dfa4a2")
-    else()
-      set(OPENCV_ICV_NAME "ippicv_2019_mac_ia32_general_20180723.tgz")
-      set(OPENCV_ICV_HASH "b5dfa78c87eb75c64470cbe5ec876f4f")
-    endif()
+    set(OPENCV_ICV_NAME "ippicv_2020_mac_intel64_20191018_general.tgz")
+    set(OPENCV_ICV_HASH "1c3d675c2a2395d094d523024896e01b")
  elseif((UNIX AND NOT ANDROID) OR (UNIX AND ANDROID_ABI MATCHES "x86"))
    set(OPENCV_ICV_PLATFORM "linux")
    set(OPENCV_ICV_PACKAGE_SUBDIR "ippicv_lnx")
    if(X86_64)
-      set(OPENCV_ICV_NAME "ippicv_2019_lnx_intel64_general_20180723.tgz")
-      set(OPENCV_ICV_HASH "c0bd78adb4156bbf552c1dfe90599607")
+      set(OPENCV_ICV_NAME "ippicv_2020_lnx_intel64_20191018_general.tgz")
+      set(OPENCV_ICV_HASH "7421de0095c7a39162ae13a6098782f9")
    else()
-      set(OPENCV_ICV_NAME "ippicv_2019_lnx_ia32_general_20180723.tgz")
-      set(OPENCV_ICV_HASH "4f38432c30bfd6423164b7a24bbc98a0")
+      set(OPENCV_ICV_NAME "ippicv_2020_lnx_ia32_20191018_general.tgz")
+      set(OPENCV_ICV_HASH "ad189a940fb60eb71f291321322fe3e8")
    endif()
  elseif(WIN32 AND NOT ARM)
    set(OPENCV_ICV_PLATFORM "windows")
    set(OPENCV_ICV_PACKAGE_SUBDIR "ippicv_win")
    if(X86_64)
-      set(OPENCV_ICV_NAME "ippicv_2019_win_intel64_20180723_general.zip")
-      set(OPENCV_ICV_HASH "1d222685246896fe089f88b8858e4b2f")
+      set(OPENCV_ICV_NAME "ippicv_2020_win_intel64_20191018_general.zip")
+      set(OPENCV_ICV_HASH "879741a7946b814455eee6c6ffde2984")
    else()
-      set(OPENCV_ICV_NAME "ippicv_2019_win_ia32_20180723_general.zip")
-      set(OPENCV_ICV_HASH "0157251a2eb9cd63a3ebc7eed0f3e59e")
+      set(OPENCV_ICV_NAME "ippicv_2020_win_ia32_20191018_general.zip")
+      set(OPENCV_ICV_HASH "cd39bdf0c2e1cac9a61101dad7a2413e")
    endif()
  else()
    return()
@@ -196,6 +196,10 @@
 #  define ITT_ARCH_PPC64  5
 #endif /* ITT_ARCH_PPC64 */

+#ifndef ITT_ARCH_AARCH64  /* 64-bit ARM */
+#  define ITT_ARCH_AARCH64  6
+#endif /* ITT_ARCH_AARCH64 */
+
 #ifndef ITT_ARCH
 #  if defined _M_IX86 || defined __i386__
 #    define ITT_ARCH ITT_ARCH_IA32
@@ -205,6 +209,8 @@
 #    define ITT_ARCH ITT_ARCH_IA64
 #  elif defined _M_ARM || defined __arm__
 #    define ITT_ARCH ITT_ARCH_ARM
+#  elif defined __aarch64__
+#    define ITT_ARCH ITT_ARCH_AARCH64
 #  elif defined __powerpc64__
 #    define ITT_ARCH ITT_ARCH_PPC64
 #  endif
@@ -359,7 +365,7 @@ ITT_INLINE long __TBB_machine_fetchadd4(volatile void* ptr, long addend)
                          : "memory");
    return result;
 }
-#elif ITT_ARCH==ITT_ARCH_ARM || ITT_ARCH==ITT_ARCH_PPC64
+#elif ITT_ARCH==ITT_ARCH_ARM || ITT_ARCH==ITT_ARCH_AARCH64 || ITT_ARCH==ITT_ARCH_PPC64
 #define __TBB_machine_fetchadd4(addr, val) __sync_fetch_and_add(addr, val)
 #endif /* ITT_ARCH==ITT_ARCH_IA64 */
 #ifndef ITT_SIMPLE_INIT
@@ -17,6 +17,12 @@
 #if !defined(JAS_WIN_MSVC_BUILD)
 /* A configure-based build is being used. */

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


 /* Extra debugging support */
@@ -4,9 +4,9 @@ ocv_warnings_disable(CMAKE_C_FLAGS -Wunused-parameter -Wsign-compare -Wshorten-6

 set(VERSION_MAJOR 2)
 set(VERSION_MINOR 0)
-set(VERSION_REVISION 2)
+set(VERSION_REVISION 4)
 set(VERSION ${VERSION_MAJOR}.${VERSION_MINOR}.${VERSION_REVISION})
-set(LIBJPEG_TURBO_VERSION_NUMBER 2000002)
+set(LIBJPEG_TURBO_VERSION_NUMBER 2000004)

 string(TIMESTAMP BUILD "opencv-${OPENCV_VERSION}-libjpeg-turbo")
 if(CMAKE_BUILD_TYPE STREQUAL "Debug")
@@ -78,10 +78,11 @@ configure_file(jconfigint.h.in jconfigint.h)

 include_directories(${CMAKE_CURRENT_BINARY_DIR} ${CMAKE_CURRENT_SOURCE_DIR}/src)

-set(JPEG_SOURCES jcapimin.c jcapistd.c jccoefct.c jccolor.c jcdctmgr.c jchuff.c
+set(JPEG_SOURCES
+  jcapimin.c jcapistd.c jccoefct.c jccolor.c jcdctmgr.c jchuff.c jcicc.c
  jcinit.c jcmainct.c jcmarker.c jcmaster.c jcomapi.c jcparam.c jcphuff.c
  jcprepct.c jcsample.c jctrans.c jdapimin.c jdapistd.c jdatadst.c jdatasrc.c
-  jdcoefct.c jdcolor.c jddctmgr.c jdhuff.c jdinput.c jdmainct.c jdmarker.c
+  jdcoefct.c jdcolor.c jddctmgr.c jdhuff.c jdicc.c jdinput.c jdmainct.c jdmarker.c
  jdmaster.c jdmerge.c jdphuff.c jdpostct.c jdsample.c jdtrans.c jerror.c
  jfdctflt.c jfdctfst.c jfdctint.c jidctflt.c jidctfst.c jidctint.c jidctred.c
  jquant1.c jquant2.c jutils.c jmemmgr.c jmemnobs.c)
@@ -1,14 +1,14 @@
 Background
 ==========

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

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

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

 ### Support for libjpeg v7 and v8 Features

 #### Fully supported

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

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

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

 - **cjpeg: Separate quality settings for luminance and chrominance**<br>
@@ -185,14 +184,14 @@ means of quality improvement.  The reader is invited to peruse the research at
 but it is the general belief of our project that these features have not
 demonstrated sufficient usefulness to justify inclusion in libjpeg-turbo.

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

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

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

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

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

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

 Note that, on most Un*x systems, the dynamic linker will not look for a
 function in a library until that function is actually used.  Thus, if a program
@@ -331,7 +329,7 @@ in a way that makes the rest of the libjpeg infrastructure happy, so it is
 necessary to use the slow Huffman decoder when decompressing a JPEG image that
 has restart markers.  This can cause the decompression performance to drop by
 as much as 20%, but the performance will still be much greater than that of
-libjpeg.  Many consumer packages, such as PhotoShop, use restart markers when
+libjpeg.  Many consumer packages, such as Photoshop, use restart markers when
 generating JPEG images, so images generated by those programs will experience
 this issue.

@@ -344,3 +342,15 @@ quality of 98-100.  Thus, libjpeg-turbo must use the non-SIMD quantization
 function in those cases.  This causes performance to drop by as much as 40%.
 It is therefore strongly advised that you use the slow integer forward DCT
 whenever encoding images with a JPEG quality of 98 or higher.
+
+
+Memory Debugger Pitfalls
+========================
+
+Valgrind and Memory Sanitizer (MSan) can generate false positives
+(specifically, incorrect reports of uninitialized memory accesses) when used
+with libjpeg-turbo's SIMD extensions.  It is generally recommended that the
+SIMD extensions be disabled, either by passing an argument of `-DWITH_SIMD=0`
+to `cmake` when configuring the build or by setting the environment variable
+`JSIMD_FORCENONE` to `1` at run time, when testing libjpeg-turbo with Valgrind,
+MSan, or other memory debuggers.
@@ -4,7 +4,7 @@
 * This file was part of the Independent JPEG Group's software:
 * Copyright (C) 1991-1997, Thomas G. Lane.
 * libjpeg-turbo Modifications:
- * Copyright (C) 2009-2011, 2014-2016, 2018, D. R. Commander.
+ * Copyright (C) 2009-2011, 2014-2016, 2018-2019, D. R. Commander.
 * Copyright (C) 2015, Matthieu Darbois.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
@@ -43,8 +43,8 @@
 */

 /* NOTE: Both GCC and Clang define __GNUC__ */
-#if defined __GNUC__ && (defined __arm__ || defined __aarch64__)
-#if !defined __thumb__ || defined __thumb2__
+#if defined(__GNUC__) && (defined(__arm__) || defined(__aarch64__))
+#if !defined(__thumb__) || defined(__thumb2__)
 #define USE_CLZ_INTRINSIC
 #endif
 #endif
@@ -356,6 +356,8 @@ dump_buffer(working_state *state)
  put_buffer = (put_buffer << size) | code; \
 }

+#if SIZEOF_SIZE_T != 8 && !defined(_WIN64)
+
 #define CHECKBUF15() { \
  if (put_bits > 15) { \
    EMIT_BYTE() \
@@ -363,6 +365,8 @@ dump_buffer(working_state *state)
  } \
 }

+#endif
+
 #define CHECKBUF31() { \
  if (put_bits > 31) { \
    EMIT_BYTE() \
@@ -428,7 +432,7 @@ dump_buffer(working_state *state)
 * scanning order-- 1, 8, 16, etc.), then this will produce an encoded block
 * larger than 200 bytes.
 */
-#define BUFSIZE  (DCTSIZE2 * 4)
+#define BUFSIZE  (DCTSIZE2 * 8)

 #define LOAD_BUFFER() { \
  if (state->free_in_buffer < BUFSIZE) { \
@@ -0,0 +1,105 @@
+/*
+ * jcicc.c
+ *
+ * Copyright (C) 1997-1998, Thomas G. Lane, Todd Newman.
+ * Copyright (C) 2017, D. R. Commander.
+ * For conditions of distribution and use, see the accompanying README.ijg
+ * file.
+ *
+ * This file provides code to write International Color Consortium (ICC) device
+ * profiles embedded in JFIF JPEG image files.  The ICC has defined a standard
+ * for including such data in JPEG "APP2" markers.  The code given here does
+ * not know anything about the internal structure of the ICC profile data; it
+ * just knows how to embed the profile data in a JPEG file while writing it.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jerror.h"
+
+
+/*
+ * Since an ICC profile can be larger than the maximum size of a JPEG marker
+ * (64K), we need provisions to split it into multiple markers.  The format
+ * defined by the ICC specifies one or more APP2 markers containing the
+ * following data:
+ *      Identifying string      ASCII "ICC_PROFILE\0"  (12 bytes)
+ *      Marker sequence number  1 for first APP2, 2 for next, etc (1 byte)
+ *      Number of markers       Total number of APP2's used (1 byte)
+ *      Profile data            (remainder of APP2 data)
+ * Decoders should use the marker sequence numbers to reassemble the profile,
+ * rather than assuming that the APP2 markers appear in the correct sequence.
+ */
+
+#define ICC_MARKER  (JPEG_APP0 + 2)     /* JPEG marker code for ICC */
+#define ICC_OVERHEAD_LEN  14            /* size of non-profile data in APP2 */
+#define MAX_BYTES_IN_MARKER  65533      /* maximum data len of a JPEG marker */
+#define MAX_DATA_BYTES_IN_MARKER  (MAX_BYTES_IN_MARKER - ICC_OVERHEAD_LEN)
+
+
+/*
+ * This routine writes the given ICC profile data into a JPEG file.  It *must*
+ * be called AFTER calling jpeg_start_compress() and BEFORE the first call to
+ * jpeg_write_scanlines().  (This ordering ensures that the APP2 marker(s) will
+ * appear after the SOI and JFIF or Adobe markers, but before all else.)
+ */
+
+GLOBAL(void)
+jpeg_write_icc_profile(j_compress_ptr cinfo, const JOCTET *icc_data_ptr,
+                       unsigned int icc_data_len)
+{
+  unsigned int num_markers;     /* total number of markers we'll write */
+  int cur_marker = 1;           /* per spec, counting starts at 1 */
+  unsigned int length;          /* number of bytes to write in this marker */
+
+  if (icc_data_ptr == NULL || icc_data_len == 0)
+    ERREXIT(cinfo, JERR_BUFFER_SIZE);
+  if (cinfo->global_state < CSTATE_SCANNING)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+
+  /* Calculate the number of markers we'll need, rounding up of course */
+  num_markers = icc_data_len / MAX_DATA_BYTES_IN_MARKER;
+  if (num_markers * MAX_DATA_BYTES_IN_MARKER != icc_data_len)
+    num_markers++;
+
+  while (icc_data_len > 0) {
+    /* length of profile to put in this marker */
+    length = icc_data_len;
+    if (length > MAX_DATA_BYTES_IN_MARKER)
+      length = MAX_DATA_BYTES_IN_MARKER;
+    icc_data_len -= length;
+
+    /* Write the JPEG marker header (APP2 code and marker length) */
+    jpeg_write_m_header(cinfo, ICC_MARKER,
+                        (unsigned int)(length + ICC_OVERHEAD_LEN));
+
+    /* Write the marker identifying string "ICC_PROFILE" (null-terminated).  We
+     * code it in this less-than-transparent way so that the code works even if
+     * the local character set is not ASCII.
+     */
+    jpeg_write_m_byte(cinfo, 0x49);
+    jpeg_write_m_byte(cinfo, 0x43);
+    jpeg_write_m_byte(cinfo, 0x43);
+    jpeg_write_m_byte(cinfo, 0x5F);
+    jpeg_write_m_byte(cinfo, 0x50);
+    jpeg_write_m_byte(cinfo, 0x52);
+    jpeg_write_m_byte(cinfo, 0x4F);
+    jpeg_write_m_byte(cinfo, 0x46);
+    jpeg_write_m_byte(cinfo, 0x49);
+    jpeg_write_m_byte(cinfo, 0x4C);
+    jpeg_write_m_byte(cinfo, 0x45);
+    jpeg_write_m_byte(cinfo, 0x0);
+
+    /* Add the sequencing info */
+    jpeg_write_m_byte(cinfo, cur_marker);
+    jpeg_write_m_byte(cinfo, (int)num_markers);
+
+    /* Add the profile data */
+    while (length--) {
+      jpeg_write_m_byte(cinfo, *icc_data_ptr);
+      icc_data_ptr++;
+    }
+    cur_marker++;
+  }
+}
@@ -492,8 +492,8 @@ prepare_for_pass(j_compress_ptr cinfo)
     */
    master->pass_type = output_pass;
    master->pass_number++;
-    /*FALLTHROUGH*/
 #endif
+    /*FALLTHROUGH*/
  case output_pass:
    /* Do a data-output pass. */
    /* We need not repeat per-scan setup if prior optimization pass did it. */
@@ -52,8 +52,8 @@
 */

 /* NOTE: Both GCC and Clang define __GNUC__ */
-#if defined __GNUC__ && (defined __arm__ || defined __aarch64__)
-#if !defined __thumb__ || defined __thumb2__
+#if defined(__GNUC__) && (defined(__arm__) || defined(__aarch64__))
+#if !defined(__thumb__) || defined(__thumb2__)
 #define USE_CLZ_INTRINSIC
 #endif
 #endif
@@ -4,7 +4,7 @@
 * This file was part of the Independent JPEG Group's software:
 * Copyright (C) 1991-1997, Thomas G. Lane.
 * libjpeg-turbo Modifications:
- * Copyright (C) 2009-2011, 2016, 2018, D. R. Commander.
+ * Copyright (C) 2009-2011, 2016, 2018-2019, D. R. Commander.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -589,7 +589,11 @@ 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];
-      s += state.last_dc_val[ci];
+      /* This is really just
+       *   s += state.last_dc_val[ci];
+       * It is written this way in order to shut up UBSan.
+       */
+      s = (int)((unsigned int)s + (unsigned int)state.last_dc_val[ci]);
      state.last_dc_val[ci] = s;
      if (block) {
        /* Output the DC coefficient (assumes jpeg_natural_order[0] = 0) */
@@ -684,7 +688,7 @@ decode_mcu_fast(j_decompress_ptr cinfo, JBLOCKROW *MCU_data)

    if (entropy->dc_needed[blkn]) {
      int ci = cinfo->MCU_membership[blkn];
-      s += state.last_dc_val[ci];
+      s = (int)((unsigned int)s + (unsigned int)state.last_dc_val[ci]);
      state.last_dc_val[ci] = s;
      if (block)
        (*block)[0] = (JCOEF)s;
@@ -0,0 +1,171 @@
+/*
+ * jdicc.c
+ *
+ * Copyright (C) 1997-1998, Thomas G. Lane, Todd Newman.
+ * Copyright (C) 2017, D. R. Commander.
+ * For conditions of distribution and use, see the accompanying README.ijg
+ * file.
+ *
+ * This file provides code to read International Color Consortium (ICC) device
+ * profiles embedded in JFIF JPEG image files.  The ICC has defined a standard
+ * for including such data in JPEG "APP2" markers.  The code given here does
+ * not know anything about the internal structure of the ICC profile data; it
+ * just knows how to get the profile data from a JPEG file while reading it.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jerror.h"
+
+#ifndef HAVE_STDLIB_H           /* <stdlib.h> should declare malloc() */
+extern void *malloc(size_t size);
+#endif
+
+
+#define ICC_MARKER  (JPEG_APP0 + 2)     /* JPEG marker code for ICC */
+#define ICC_OVERHEAD_LEN  14            /* size of non-profile data in APP2 */
+
+
+/*
+ * Handy subroutine to test whether a saved marker is an ICC profile marker.
+ */
+
+LOCAL(boolean)
+marker_is_icc(jpeg_saved_marker_ptr marker)
+{
+  return
+    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;
+}
+
+
+/*
+ * See if there was an ICC profile in the JPEG file being read; if so,
+ * reassemble and return the profile data.
+ *
+ * TRUE is returned if an ICC profile was found, FALSE if not.  If TRUE is
+ * returned, *icc_data_ptr is set to point to the returned data, and
+ * *icc_data_len is set to its length.
+ *
+ * IMPORTANT: the data at *icc_data_ptr is allocated with malloc() and must be
+ * freed by the caller with free() when the caller no longer needs it.
+ * (Alternatively, we could write this routine to use the IJG library's memory
+ * allocator, so that the data would be freed implicitly when
+ * jpeg_finish_decompress() is called.  But it seems likely that many
+ * applications will prefer to have the data stick around after decompression
+ * finishes.)
+ */
+
+GLOBAL(boolean)
+jpeg_read_icc_profile(j_decompress_ptr cinfo, JOCTET **icc_data_ptr,
+                      unsigned int *icc_data_len)
+{
+  jpeg_saved_marker_ptr marker;
+  int num_markers = 0;
+  int seq_no;
+  JOCTET *icc_data;
+  unsigned int total_length;
+#define MAX_SEQ_NO  255         /* sufficient since marker numbers are bytes */
+  char marker_present[MAX_SEQ_NO + 1];      /* 1 if marker found */
+  unsigned int data_length[MAX_SEQ_NO + 1]; /* size of profile data in marker */
+  unsigned int data_offset[MAX_SEQ_NO + 1]; /* offset for data in marker */
+
+  if (icc_data_ptr == NULL || icc_data_len == NULL)
+    ERREXIT(cinfo, JERR_BUFFER_SIZE);
+  if (cinfo->global_state < DSTATE_READY)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+
+  *icc_data_ptr = NULL;         /* avoid confusion if FALSE return */
+  *icc_data_len = 0;
+
+  /* This first pass over the saved markers discovers whether there are
+   * any ICC markers and verifies the consistency of the marker numbering.
+   */
+
+  for (seq_no = 1; seq_no <= MAX_SEQ_NO; seq_no++)
+    marker_present[seq_no] = 0;
+
+  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])) {
+        WARNMS(cinfo, JWRN_BOGUS_ICC);  /* inconsistent num_markers fields */
+        return FALSE;
+      }
+      seq_no = GETJOCTET(marker->data[12]);
+      if (seq_no <= 0 || seq_no > num_markers) {
+        WARNMS(cinfo, JWRN_BOGUS_ICC);  /* bogus sequence number */
+        return FALSE;
+      }
+      if (marker_present[seq_no]) {
+        WARNMS(cinfo, JWRN_BOGUS_ICC);  /* duplicate sequence numbers */
+        return FALSE;
+      }
+      marker_present[seq_no] = 1;
+      data_length[seq_no] = marker->data_length - ICC_OVERHEAD_LEN;
+    }
+  }
+
+  if (num_markers == 0)
+    return FALSE;
+
+  /* Check for missing markers, count total space needed,
+   * compute offset of each marker's part of the data.
+   */
+
+  total_length = 0;
+  for (seq_no = 1; seq_no <= num_markers; seq_no++) {
+    if (marker_present[seq_no] == 0) {
+      WARNMS(cinfo, JWRN_BOGUS_ICC);  /* missing sequence number */
+      return FALSE;
+    }
+    data_offset[seq_no] = total_length;
+    total_length += data_length[seq_no];
+  }
+
+  if (total_length == 0) {
+    WARNMS(cinfo, JWRN_BOGUS_ICC);  /* found only empty markers? */
+    return FALSE;
+  }
+
+  /* Allocate space for assembled data */
+  icc_data = (JOCTET *)malloc(total_length * sizeof(JOCTET));
+  if (icc_data == NULL)
+    ERREXIT1(cinfo, JERR_OUT_OF_MEMORY, 11);  /* oops, out of memory */
+
+  /* and fill it in */
+  for (marker = cinfo->marker_list; marker != NULL; marker = marker->next) {
+    if (marker_is_icc(marker)) {
+      JOCTET FAR *src_ptr;
+      JOCTET *dst_ptr;
+      unsigned int length;
+      seq_no = GETJOCTET(marker->data[12]);
+      dst_ptr = icc_data + data_offset[seq_no];
+      src_ptr = marker->data + ICC_OVERHEAD_LEN;
+      length = data_length[seq_no];
+      while (length--) {
+        *dst_ptr++ = *src_ptr++;
+      }
+    }
+  }
+
+  *icc_data_ptr = icc_data;
+  *icc_data_len = total_length;
+
+  return TRUE;
+}
@@ -429,8 +429,6 @@ h2v2_merged_upsample(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
 #define PACK_TWO_PIXELS_LE(l, r)    ((r << 16) | l)
 #define PACK_TWO_PIXELS_BE(l, r)    ((l << 16) | r)

-#define PACK_NEED_ALIGNMENT(ptr)    (((size_t)(ptr)) & 3)
-
 #define WRITE_TWO_PIXELS_LE(addr, pixels) { \
  ((INT16 *)(addr))[0] = (INT16)(pixels); \
  ((INT16 *)(addr))[1] = (INT16)((pixels) >> 16); \
@@ -8,6 +8,7 @@
 * Copyright (C) 2010, 2015-2016, D. R. Commander.
 * Copyright (C) 2014, MIPS Technologies, Inc., California.
 * Copyright (C) 2015, Google, Inc.
+ * Copyright (C) 2019, Arm Limited.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
@@ -315,9 +316,9 @@ h1v2_fancy_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr,
  JSAMPARRAY output_data = *output_data_ptr;
  JSAMPROW inptr0, inptr1, outptr;
 #if BITS_IN_JSAMPLE == 8
-  int thiscolsum;
+  int thiscolsum, bias;
 #else
-  JLONG thiscolsum;
+  JLONG thiscolsum, bias;
 #endif
  JDIMENSION colctr;
  int inrow, outrow, v;
@@ -327,15 +328,18 @@ h1v2_fancy_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr,
    for (v = 0; v < 2; v++) {
      /* inptr0 points to nearest input row, inptr1 points to next nearest */
      inptr0 = input_data[inrow];
-      if (v == 0)               /* next nearest is row above */
+      if (v == 0) {             /* next nearest is row above */
        inptr1 = input_data[inrow - 1];
-      else                      /* next nearest is row below */
+        bias = 1;
+      } else {                  /* next nearest is row below */
        inptr1 = input_data[inrow + 1];
+        bias = 2;
+      }
      outptr = output_data[outrow++];

      for (colctr = 0; colctr < compptr->downsampled_width; colctr++) {
        thiscolsum = GETJSAMPLE(*inptr0++) * 3 + GETJSAMPLE(*inptr1++);
-        *outptr++ = (JSAMPLE)((thiscolsum + 1) >> 2);
+        *outptr++ = (JSAMPLE)((thiscolsum + bias) >> 2);
      }
    }
    inrow++;
@@ -1,7 +1,7 @@
 /*
 * jfdctint.c
 *
- * This file was part of the Independent JPEG Group's software.
+ * This file was part of the Independent JPEG Group's software:
 * Copyright (C) 1991-1996, Thomas G. Lane.
 * libjpeg-turbo Modifications:
 * Copyright (C) 2015, D. R. Commander.
@@ -1,7 +1,7 @@
 /*
 * jidctint.c
 *
- * This file was part of the Independent JPEG Group's software.
+ * This file was part of the Independent JPEG Group's software:
 * Copyright (C) 1991-1998, Thomas G. Lane.
 * Modification developed 2002-2009 by Guido Vollbeding.
 * libjpeg-turbo Modifications:
@@ -1,7 +1,7 @@
 /*
 * jidctred.c
 *
- * This file was part of the Independent JPEG Group's software.
+ * This file was part of the Independent JPEG Group's software:
 * Copyright (C) 1994-1998, Thomas G. Lane.
 * libjpeg-turbo Modifications:
 * Copyright (C) 2015, D. R. Commander.
@@ -15,13 +15,6 @@ else()
  ocv_list_filterout(lib_srcs jmemnobs.c)
 endif()

-if(WINRT)
-    add_definitions(-DNO_GETENV)
-    get_directory_property( DirDefs COMPILE_DEFINITIONS )
-    message(STATUS "Adding NO_GETENV to compiler definitions for WINRT:")
-    message(STATUS "   COMPILE_DEFINITIONS = ${DirDefs}")
-endif()
-
 # ----------------------------------------------------------------------------------
 #         Define the library target:
 # ----------------------------------------------------------------------------------
@@ -1,7 +1,7 @@
 The Independent JPEG Group's JPEG software
 ==========================================

-README for release 9b of 17-Jan-2016
+README for release 9c of 14-Jan-2018
 ====================================

 This distribution contains the ninth public release of the Independent JPEG
@@ -115,7 +115,7 @@ with respect to this software, its quality, accuracy, merchantability, or
 fitness for a particular purpose.  This software is provided "AS IS", and you,
 its user, assume the entire risk as to its quality and accuracy.

-This software is copyright (C) 1991-2016, Thomas G. Lane, Guido Vollbeding.
+This software is copyright (C) 1991-2018, Thomas G. Lane, Guido Vollbeding.
 All Rights Reserved except as specified below.

 Permission is hereby granted to use, copy, modify, and distribute this
@@ -246,8 +246,8 @@ ARCHIVE LOCATIONS
 The "official" archive site for this software is www.ijg.org.
 The most recent released version can always be found there in
 directory "files".  This particular version will be archived as
-http://www.ijg.org/files/jpegsrc.v9b.tar.gz, and in Windows-compatible
-"zip" archive format as http://www.ijg.org/files/jpegsr9b.zip.
+http://www.ijg.org/files/jpegsrc.v9c.tar.gz, and in Windows-compatible
+"zip" archive format as http://www.ijg.org/files/jpegsr9c.zip.

 The JPEG FAQ (Frequently Asked Questions) article is a source of some
 general information about JPEG.
@@ -293,8 +293,11 @@ communication about JPEG configuration in Sigma Photo Pro software.

 Thank to Andrew Finkenstadt for hosting the ijg.org site.

-Last but not least special thank to Thomas G. Lane for the original
-design and development of this singular software package.
+Thank to Thomas G. Lane for the original design and development of
+this singular software package.
+
+Thank to Lars Goehler, Andreas Heinecke, Sebastian Fuss, Yvonne Roebert,
+Andrej Werner, and Ulf-Dietrich Braumann for support and public relations.


 FILE FORMAT WARS
@@ -1,6 +1,27 @@
 CHANGE LOG for Independent JPEG Group's JPEG software


+Version 9c  14-Jan-2018
+-----------------------
+
+jpegtran: add an option to the -wipe switch to fill the region
+with the average of adjacent blocks, instead of gray out.
+Thank to Caitlyn Feddock and Maddie Ziegler for inspiration.
+
+Make range extension bits adjustable (in jpegint.h).
+Thank to Robin Watts for suggestion.
+
+Provide macros for fflush() and ferror() in jinclude.h in order
+to facilitate adaption by applications using an own FILE class.
+Thank to Gerhard Huber for suggestion.
+
+Add libjpeg pkg-config file.  Thank to Mark Lavi, Vincent Torri,
+Patrick McMunn, and Huw Davies for suggestion.
+
+Add sanity checks in cjpeg image reader modules.
+Thank to Bingchang, Liu for reports.
+
+
 Version 9b  17-Jan-2016
 -----------------------

@@ -2,7 +2,7 @@
 * jcinit.c
 *
 * Copyright (C) 1991-1997, Thomas G. Lane.
- * Modified 2003-2013 by Guido Vollbeding.
+ * Modified 2003-2017 by Guido Vollbeding.
 * This file is part of the Independent JPEG Group's software.
 * For conditions of distribution and use, see the accompanying README file.
 *
@@ -21,6 +21,168 @@
 #include "jpeglib.h"


+/*
+ * Compute JPEG image dimensions and related values.
+ * NOTE: this is exported for possible use by application.
+ * Hence it mustn't do anything that can't be done twice.
+ */
+
+GLOBAL(void)
+jpeg_calc_jpeg_dimensions (j_compress_ptr cinfo)
+/* Do computations that are needed before master selection phase */
+{
+  /* Sanity check on input image dimensions to prevent overflow in
+   * following calculations.
+   * We do check jpeg_width and jpeg_height in initial_setup in jcmaster.c,
+   * but image_width and image_height can come from arbitrary data,
+   * and we need some space for multiplication by block_size.
+   */
+  if (((long) cinfo->image_width >> 24) || ((long) cinfo->image_height >> 24))
+    ERREXIT1(cinfo, JERR_IMAGE_TOO_BIG, (unsigned int) JPEG_MAX_DIMENSION);
+
+#ifdef DCT_SCALING_SUPPORTED
+
+  /* Compute actual JPEG image dimensions and DCT scaling choices. */
+  if (cinfo->scale_num >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/1 scaling */
+    cinfo->jpeg_width = cinfo->image_width * cinfo->block_size;
+    cinfo->jpeg_height = cinfo->image_height * cinfo->block_size;
+    cinfo->min_DCT_h_scaled_size = 1;
+    cinfo->min_DCT_v_scaled_size = 1;
+  } else if (cinfo->scale_num * 2 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/2 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 2L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 2L);
+    cinfo->min_DCT_h_scaled_size = 2;
+    cinfo->min_DCT_v_scaled_size = 2;
+  } else if (cinfo->scale_num * 3 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/3 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 3L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 3L);
+    cinfo->min_DCT_h_scaled_size = 3;
+    cinfo->min_DCT_v_scaled_size = 3;
+  } else if (cinfo->scale_num * 4 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/4 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 4L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 4L);
+    cinfo->min_DCT_h_scaled_size = 4;
+    cinfo->min_DCT_v_scaled_size = 4;
+  } else if (cinfo->scale_num * 5 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/5 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 5L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 5L);
+    cinfo->min_DCT_h_scaled_size = 5;
+    cinfo->min_DCT_v_scaled_size = 5;
+  } else if (cinfo->scale_num * 6 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/6 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 6L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 6L);
+    cinfo->min_DCT_h_scaled_size = 6;
+    cinfo->min_DCT_v_scaled_size = 6;
+  } else if (cinfo->scale_num * 7 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/7 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 7L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 7L);
+    cinfo->min_DCT_h_scaled_size = 7;
+    cinfo->min_DCT_v_scaled_size = 7;
+  } else if (cinfo->scale_num * 8 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/8 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 8L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 8L);
+    cinfo->min_DCT_h_scaled_size = 8;
+    cinfo->min_DCT_v_scaled_size = 8;
+  } else if (cinfo->scale_num * 9 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/9 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 9L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 9L);
+    cinfo->min_DCT_h_scaled_size = 9;
+    cinfo->min_DCT_v_scaled_size = 9;
+  } else if (cinfo->scale_num * 10 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/10 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 10L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 10L);
+    cinfo->min_DCT_h_scaled_size = 10;
+    cinfo->min_DCT_v_scaled_size = 10;
+  } else if (cinfo->scale_num * 11 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/11 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 11L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 11L);
+    cinfo->min_DCT_h_scaled_size = 11;
+    cinfo->min_DCT_v_scaled_size = 11;
+  } else if (cinfo->scale_num * 12 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/12 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 12L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 12L);
+    cinfo->min_DCT_h_scaled_size = 12;
+    cinfo->min_DCT_v_scaled_size = 12;
+  } else if (cinfo->scale_num * 13 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/13 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 13L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 13L);
+    cinfo->min_DCT_h_scaled_size = 13;
+    cinfo->min_DCT_v_scaled_size = 13;
+  } else if (cinfo->scale_num * 14 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/14 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 14L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 14L);
+    cinfo->min_DCT_h_scaled_size = 14;
+    cinfo->min_DCT_v_scaled_size = 14;
+  } else if (cinfo->scale_num * 15 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/15 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 15L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 15L);
+    cinfo->min_DCT_h_scaled_size = 15;
+    cinfo->min_DCT_v_scaled_size = 15;
+  } else {
+    /* Provide block_size/16 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 16L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 16L);
+    cinfo->min_DCT_h_scaled_size = 16;
+    cinfo->min_DCT_v_scaled_size = 16;
+  }
+
+#else /* !DCT_SCALING_SUPPORTED */
+
+  /* Hardwire it to "no scaling" */
+  cinfo->jpeg_width = cinfo->image_width;
+  cinfo->jpeg_height = cinfo->image_height;
+  cinfo->min_DCT_h_scaled_size = DCTSIZE;
+  cinfo->min_DCT_v_scaled_size = DCTSIZE;
+
+#endif /* DCT_SCALING_SUPPORTED */
+}
+
+
 /*
 * Master selection of compression modules.
 * This is done once at the start of processing an image.  We determine
@@ -37,7 +199,7 @@ jinit_compress_master (j_compress_ptr cinfo)
  if (cinfo->data_precision != BITS_IN_JSAMPLE)
    ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision);

-  /* Sanity check on image dimensions */
+  /* Sanity check on input image dimensions */
  if (cinfo->image_height <= 0 || cinfo->image_width <= 0 ||
      cinfo->input_components <= 0)
    ERREXIT(cinfo, JERR_EMPTY_IMAGE);
@@ -48,6 +210,9 @@ jinit_compress_master (j_compress_ptr cinfo)
  if ((long) jd_samplesperrow != samplesperrow)
    ERREXIT(cinfo, JERR_WIDTH_OVERFLOW);

+  /* Compute JPEG image dimensions and related values. */
+  jpeg_calc_jpeg_dimensions(cinfo);
+
  /* Initialize master control (includes parameter checking/processing) */
  jinit_c_master_control(cinfo, FALSE /* full compression */);

@@ -2,7 +2,7 @@
 * jcmaster.c
 *
 * Copyright (C) 1991-1997, Thomas G. Lane.
- * Modified 2003-2013 by Guido Vollbeding.
+ * Modified 2003-2017 by Guido Vollbeding.
 * This file is part of the Independent JPEG Group's software.
 * For conditions of distribution and use, see the accompanying README file.
 *
@@ -43,191 +43,13 @@ typedef my_comp_master * my_master_ptr;
 * Support routines that do various essential calculations.
 */

-/*
- * Compute JPEG image dimensions and related values.
- * NOTE: this is exported for possible use by application.
- * Hence it mustn't do anything that can't be done twice.
- */
-
-GLOBAL(void)
-jpeg_calc_jpeg_dimensions (j_compress_ptr cinfo)
-/* Do computations that are needed before master selection phase */
-{
-#ifdef DCT_SCALING_SUPPORTED
-
-  /* Sanity check on input image dimensions to prevent overflow in
-   * following calculation.
-   * We do check jpeg_width and jpeg_height in initial_setup below,
-   * but image_width and image_height can come from arbitrary data,
-   * and we need some space for multiplication by block_size.
-   */
-  if (((long) cinfo->image_width >> 24) || ((long) cinfo->image_height >> 24))
-    ERREXIT1(cinfo, JERR_IMAGE_TOO_BIG, (unsigned int) JPEG_MAX_DIMENSION);
-
-  /* Compute actual JPEG image dimensions and DCT scaling choices. */
-  if (cinfo->scale_num >= cinfo->scale_denom * cinfo->block_size) {
-    /* Provide block_size/1 scaling */
-    cinfo->jpeg_width = cinfo->image_width * cinfo->block_size;
-    cinfo->jpeg_height = cinfo->image_height * cinfo->block_size;
-    cinfo->min_DCT_h_scaled_size = 1;
-    cinfo->min_DCT_v_scaled_size = 1;
-  } else if (cinfo->scale_num * 2 >= cinfo->scale_denom * cinfo->block_size) {
-    /* Provide block_size/2 scaling */
-    cinfo->jpeg_width = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 2L);
-    cinfo->jpeg_height = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 2L);
-    cinfo->min_DCT_h_scaled_size = 2;
-    cinfo->min_DCT_v_scaled_size = 2;
-  } else if (cinfo->scale_num * 3 >= cinfo->scale_denom * cinfo->block_size) {
-    /* Provide block_size/3 scaling */
-    cinfo->jpeg_width = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 3L);
-    cinfo->jpeg_height = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 3L);
-    cinfo->min_DCT_h_scaled_size = 3;
-    cinfo->min_DCT_v_scaled_size = 3;
-  } else if (cinfo->scale_num * 4 >= cinfo->scale_denom * cinfo->block_size) {
-    /* Provide block_size/4 scaling */
-    cinfo->jpeg_width = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 4L);
-    cinfo->jpeg_height = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 4L);
-    cinfo->min_DCT_h_scaled_size = 4;
-    cinfo->min_DCT_v_scaled_size = 4;
-  } else if (cinfo->scale_num * 5 >= cinfo->scale_denom * cinfo->block_size) {
-    /* Provide block_size/5 scaling */
-    cinfo->jpeg_width = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 5L);
-    cinfo->jpeg_height = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 5L);
-    cinfo->min_DCT_h_scaled_size = 5;
-    cinfo->min_DCT_v_scaled_size = 5;
-  } else if (cinfo->scale_num * 6 >= cinfo->scale_denom * cinfo->block_size) {
-    /* Provide block_size/6 scaling */
-    cinfo->jpeg_width = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 6L);
-    cinfo->jpeg_height = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 6L);
-    cinfo->min_DCT_h_scaled_size = 6;
-    cinfo->min_DCT_v_scaled_size = 6;
-  } else if (cinfo->scale_num * 7 >= cinfo->scale_denom * cinfo->block_size) {
-    /* Provide block_size/7 scaling */
-    cinfo->jpeg_width = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 7L);
-    cinfo->jpeg_height = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 7L);
-    cinfo->min_DCT_h_scaled_size = 7;
-    cinfo->min_DCT_v_scaled_size = 7;
-  } else if (cinfo->scale_num * 8 >= cinfo->scale_denom * cinfo->block_size) {
-    /* Provide block_size/8 scaling */
-    cinfo->jpeg_width = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 8L);
-    cinfo->jpeg_height = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 8L);
-    cinfo->min_DCT_h_scaled_size = 8;
-    cinfo->min_DCT_v_scaled_size = 8;
-  } else if (cinfo->scale_num * 9 >= cinfo->scale_denom * cinfo->block_size) {
-    /* Provide block_size/9 scaling */
-    cinfo->jpeg_width = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 9L);
-    cinfo->jpeg_height = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 9L);
-    cinfo->min_DCT_h_scaled_size = 9;
-    cinfo->min_DCT_v_scaled_size = 9;
-  } else if (cinfo->scale_num * 10 >= cinfo->scale_denom * cinfo->block_size) {
-    /* Provide block_size/10 scaling */
-    cinfo->jpeg_width = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 10L);
-    cinfo->jpeg_height = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 10L);
-    cinfo->min_DCT_h_scaled_size = 10;
-    cinfo->min_DCT_v_scaled_size = 10;
-  } else if (cinfo->scale_num * 11 >= cinfo->scale_denom * cinfo->block_size) {
-    /* Provide block_size/11 scaling */
-    cinfo->jpeg_width = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 11L);
-    cinfo->jpeg_height = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 11L);
-    cinfo->min_DCT_h_scaled_size = 11;
-    cinfo->min_DCT_v_scaled_size = 11;
-  } else if (cinfo->scale_num * 12 >= cinfo->scale_denom * cinfo->block_size) {
-    /* Provide block_size/12 scaling */
-    cinfo->jpeg_width = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 12L);
-    cinfo->jpeg_height = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 12L);
-    cinfo->min_DCT_h_scaled_size = 12;
-    cinfo->min_DCT_v_scaled_size = 12;
-  } else if (cinfo->scale_num * 13 >= cinfo->scale_denom * cinfo->block_size) {
-    /* Provide block_size/13 scaling */
-    cinfo->jpeg_width = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 13L);
-    cinfo->jpeg_height = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 13L);
-    cinfo->min_DCT_h_scaled_size = 13;
-    cinfo->min_DCT_v_scaled_size = 13;
-  } else if (cinfo->scale_num * 14 >= cinfo->scale_denom * cinfo->block_size) {
-    /* Provide block_size/14 scaling */
-    cinfo->jpeg_width = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 14L);
-    cinfo->jpeg_height = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 14L);
-    cinfo->min_DCT_h_scaled_size = 14;
-    cinfo->min_DCT_v_scaled_size = 14;
-  } else if (cinfo->scale_num * 15 >= cinfo->scale_denom * cinfo->block_size) {
-    /* Provide block_size/15 scaling */
-    cinfo->jpeg_width = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 15L);
-    cinfo->jpeg_height = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 15L);
-    cinfo->min_DCT_h_scaled_size = 15;
-    cinfo->min_DCT_v_scaled_size = 15;
-  } else {
-    /* Provide block_size/16 scaling */
-    cinfo->jpeg_width = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 16L);
-    cinfo->jpeg_height = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 16L);
-    cinfo->min_DCT_h_scaled_size = 16;
-    cinfo->min_DCT_v_scaled_size = 16;
-  }
-
-#else /* !DCT_SCALING_SUPPORTED */
-
-  /* Hardwire it to "no scaling" */
-  cinfo->jpeg_width = cinfo->image_width;
-  cinfo->jpeg_height = cinfo->image_height;
-  cinfo->min_DCT_h_scaled_size = DCTSIZE;
-  cinfo->min_DCT_v_scaled_size = DCTSIZE;
-
-#endif /* DCT_SCALING_SUPPORTED */
-}
-
-
 LOCAL(void)
-jpeg_calc_trans_dimensions (j_compress_ptr cinfo)
-{
-  if (cinfo->min_DCT_h_scaled_size != cinfo->min_DCT_v_scaled_size)
-    ERREXIT2(cinfo, JERR_BAD_DCTSIZE,
-	     cinfo->min_DCT_h_scaled_size, cinfo->min_DCT_v_scaled_size);
-
-  cinfo->block_size = cinfo->min_DCT_h_scaled_size;
-}
-
-
-LOCAL(void)
-initial_setup (j_compress_ptr cinfo, boolean transcode_only)
+initial_setup (j_compress_ptr cinfo)
 /* Do computations that are needed before master selection phase */
 {
  int ci, ssize;
  jpeg_component_info *compptr;

-  if (transcode_only)
-    jpeg_calc_trans_dimensions(cinfo);
-  else
-    jpeg_calc_jpeg_dimensions(cinfo);
-
  /* Sanity check on block_size */
  if (cinfo->block_size < 1 || cinfo->block_size > 16)
    ERREXIT2(cinfo, JERR_BAD_DCTSIZE, cinfo->block_size, cinfo->block_size);
@@ -414,13 +236,9 @@ validate_script (j_compress_ptr cinfo)
       * out-of-range reconstructed DC values during the first DC scan,
       * which might cause problems for some decoders.
       */
-#if BITS_IN_JSAMPLE == 8
-#define MAX_AH_AL 10
-#else
-#define MAX_AH_AL 13
-#endif
      if (Ss < 0 || Ss >= DCTSIZE2 || Se < Ss || Se >= DCTSIZE2 ||
-	  Ah < 0 || Ah > MAX_AH_AL || Al < 0 || Al > MAX_AH_AL)
+	  Ah < 0 || Ah > (cinfo->data_precision > 8 ? 13 : 10) ||
+	  Al < 0 || Al > (cinfo->data_precision > 8 ? 13 : 10))
 	ERREXIT1(cinfo, JERR_BAD_PROG_SCRIPT, scanno);
      if (Ss == 0) {
 	if (Se != 0)		/* DC and AC together not OK */
@@ -812,7 +630,7 @@ jinit_c_master_control (j_compress_ptr cinfo, boolean transcode_only)
  master->pub.is_last_pass = FALSE;

  /* Validate parameters, determine derived values */
-  initial_setup(cinfo, transcode_only);
+  initial_setup(cinfo);

  if (cinfo->scan_info != NULL) {
 #ifdef C_MULTISCAN_FILES_SUPPORTED
@@ -2,7 +2,7 @@
 * jctrans.c
 *
 * Copyright (C) 1995-1998, Thomas G. Lane.
- * Modified 2000-2013 by Guido Vollbeding.
+ * Modified 2000-2017 by Guido Vollbeding.
 * This file is part of the Independent JPEG Group's software.
 * For conditions of distribution and use, see the accompanying README file.
 *
@@ -85,12 +85,15 @@ jpeg_copy_critical_parameters (j_decompress_ptr srcinfo,
  jpeg_set_defaults(dstinfo);
  /* jpeg_set_defaults may choose wrong colorspace, eg YCbCr if input is RGB.
   * Fix it to get the right header markers for the image colorspace.
-   * Note: Entropy table assignment in jpeg_set_colorspace depends
-   * on color_transform.
+   * Note: Entropy table assignment in jpeg_set_colorspace
+   * depends on color_transform.
+   * Adaption is also required for setting the appropriate
+   * entropy coding mode dependent on image data precision.
   */
  dstinfo->color_transform = srcinfo->color_transform;
  jpeg_set_colorspace(dstinfo, srcinfo->jpeg_color_space);
  dstinfo->data_precision = srcinfo->data_precision;
+  dstinfo->arith_code = srcinfo->data_precision > 8 ? TRUE : FALSE;
  dstinfo->CCIR601_sampling = srcinfo->CCIR601_sampling;
  /* Copy the source's quantization tables. */
  for (tblno = 0; tblno < NUM_QUANT_TBLS; tblno++) {
@@ -157,6 +160,18 @@ jpeg_copy_critical_parameters (j_decompress_ptr srcinfo,
 }


+LOCAL(void)
+jpeg_calc_trans_dimensions (j_compress_ptr cinfo)
+/* Do computations that are needed before master selection phase */
+{
+  if (cinfo->min_DCT_h_scaled_size != cinfo->min_DCT_v_scaled_size)
+    ERREXIT2(cinfo, JERR_BAD_DCTSIZE,
+	     cinfo->min_DCT_h_scaled_size, cinfo->min_DCT_v_scaled_size);
+
+  cinfo->block_size = cinfo->min_DCT_h_scaled_size;
+}
+
+
 /*
 * Master selection of compression modules for transcoding.
 * This substitutes for jcinit.c's initialization of the full compressor.
@@ -166,6 +181,9 @@ LOCAL(void)
 transencode_master_selection (j_compress_ptr cinfo,
 			      jvirt_barray_ptr * coef_arrays)
 {
+  /* Do computations that are needed before master selection phase */
+  jpeg_calc_trans_dimensions(cinfo);
+
  /* Initialize master control (includes parameter checking/processing) */
  jinit_c_master_control(cinfo, TRUE /* transcode only */);

@@ -2,7 +2,7 @@
 * jdatadst.c
 *
 * Copyright (C) 1994-1996, Thomas G. Lane.
- * Modified 2009-2012 by Guido Vollbeding.
+ * Modified 2009-2017 by Guido Vollbeding.
 * This file is part of the Independent JPEG Group's software.
 * For conditions of distribution and use, see the accompanying README file.
 *
@@ -170,9 +170,9 @@ term_destination (j_compress_ptr cinfo)
    if (JFWRITE(dest->outfile, dest->buffer, datacount) != datacount)
      ERREXIT(cinfo, JERR_FILE_WRITE);
  }
-  fflush(dest->outfile);
+  JFFLUSH(dest->outfile);
  /* Make sure we wrote the output file OK */
-  if (ferror(dest->outfile))
+  if (JFERROR(dest->outfile))
    ERREXIT(cinfo, JERR_FILE_WRITE);
 }

@@ -2,7 +2,7 @@
 * jdcolor.c
 *
 * Copyright (C) 1991-1997, Thomas G. Lane.
- * Modified 2011-2015 by Guido Vollbeding.
+ * Modified 2011-2017 by Guido Vollbeding.
 * This file is part of the Independent JPEG Group's software.
 * For conditions of distribution and use, see the accompanying README file.
 *
@@ -14,6 +14,12 @@
 #include "jpeglib.h"


+#if RANGE_BITS < 2
+  /* Deliberate syntax err */
+  Sorry, this code requires 2 or more range extension bits.
+#endif
+
+
 /* Private subobject */

 typedef struct {
@@ -2,7 +2,7 @@
 * jdct.h
 *
 * Copyright (C) 1994-1996, Thomas G. Lane.
- * Modified 2002-2015 by Guido Vollbeding.
+ * Modified 2002-2017 by Guido Vollbeding.
 * This file is part of the Independent JPEG Group's software.
 * For conditions of distribution and use, see the accompanying README file.
 *
@@ -79,13 +79,12 @@ typedef FAST_FLOAT FLOAT_MULT_TYPE; /* preferred floating type */
 * converting them to unsigned form (0..MAXJSAMPLE).  The raw outputs could
 * be quite far out of range if the input data is corrupt, so a bulletproof
 * range-limiting step is required.  We use a mask-and-table-lookup method
- * to do the combined operations quickly, assuming that MAXJSAMPLE+1
- * is a power of 2.  See the comments with prepare_range_limit_table
- * (in jdmaster.c) for more info.
+ * to do the combined operations quickly, assuming that RANGE_CENTER
+ * (defined in jpegint.h) is a power of 2.  See the comments with
+ * prepare_range_limit_table (in jdmaster.c) for more info.
 */

-#define RANGE_MASK  (MAXJSAMPLE * 4 + 3) /* 2 bits wider than legal samples */
-#define RANGE_CENTER  (MAXJSAMPLE * 2 + 2)
+#define RANGE_MASK  (RANGE_CENTER * 2 - 1)
 #define RANGE_SUBSET  (RANGE_CENTER - CENTERJSAMPLE)

 #define IDCT_range_limit(cinfo)  ((cinfo)->sample_range_limit - RANGE_SUBSET)
@@ -2,7 +2,7 @@
 * jdhuff.c
 *
 * Copyright (C) 1991-1997, Thomas G. Lane.
- * Modified 2006-2013 by Guido Vollbeding.
+ * Modified 2006-2016 by Guido Vollbeding.
 * This file is part of the Independent JPEG Group's software.
 * For conditions of distribution and use, see the accompanying README file.
 *
@@ -799,10 +799,6 @@ decode_mcu_AC_first (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
   */
  if (! entropy->insufficient_data) {

-    Se = cinfo->Se;
-    Al = cinfo->Al;
-    natural_order = cinfo->natural_order;
-
    /* Load up working state.
     * We can avoid loading/saving bitread state if in an EOB run.
     */
@@ -814,6 +810,9 @@ decode_mcu_AC_first (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
      EOBRUN--;			/* ...process it now (we do nothing) */
    else {
      BITREAD_LOAD_STATE(cinfo,entropy->bitstate);
+      Se = cinfo->Se;
+      Al = cinfo->Al;
+      natural_order = cinfo->natural_order;
      block = MCU_data[0];
      tbl = entropy->ac_derived_tbl;

@@ -2,7 +2,7 @@
 * jdmainct.c
 *
 * Copyright (C) 1994-1996, Thomas G. Lane.
- * Modified 2002-2012 by Guido Vollbeding.
+ * Modified 2002-2016 by Guido Vollbeding.
 * This file is part of the Independent JPEG Group's software.
 * For conditions of distribution and use, see the accompanying README file.
 *
@@ -26,8 +26,8 @@
 * trivial.  Its responsibility is to provide context rows for upsampling/
 * rescaling, and doing this in an efficient fashion is a bit tricky.
 *
- * Postprocessor input data is counted in "row groups".  A row group
- * is defined to be (v_samp_factor * DCT_scaled_size / min_DCT_scaled_size)
+ * Postprocessor input data is counted in "row groups".  A row group is
+ * defined to be (v_samp_factor * DCT_v_scaled_size / min_DCT_v_scaled_size)
 * sample rows of each component.  (We require DCT_scaled_size values to be
 * chosen such that these numbers are integers.  In practice DCT_scaled_size
 * values will likely be powers of two, so we actually have the stronger
@@ -37,8 +37,8 @@
 * applying).
 *
 * The coefficient controller will deliver data to us one iMCU row at a time;
- * each iMCU row contains v_samp_factor * DCT_scaled_size sample rows, or
- * exactly min_DCT_scaled_size row groups.  (This amount of data corresponds
+ * each iMCU row contains v_samp_factor * DCT_v_scaled_size sample rows, or
+ * exactly min_DCT_v_scaled_size row groups.  (This amount of data corresponds
 * to one row of MCUs when the image is fully interleaved.)  Note that the
 * number of sample rows varies across components, but the number of row
 * groups does not.  Some garbage sample rows may be included in the last iMCU
@@ -75,7 +75,7 @@
 * We could do this most simply by copying data around in our buffer, but
 * that'd be very slow.  We can avoid copying any data by creating a rather
 * strange pointer structure.  Here's how it works.  We allocate a workspace
- * consisting of M+2 row groups (where M = min_DCT_scaled_size is the number
+ * consisting of M+2 row groups (where M = min_DCT_v_scaled_size is the number
 * of row groups per iMCU row).  We create two sets of redundant pointers to
 * the workspace.  Labeling the physical row groups 0 to M+1, the synthesized
 * pointer lists look like this:
@@ -100,11 +100,11 @@
 * the first or last sample row as necessary (this is cheaper than copying
 * sample rows around).
 *
- * This scheme breaks down if M < 2, ie, min_DCT_scaled_size is 1.  In that
+ * This scheme breaks down if M < 2, ie, min_DCT_v_scaled_size is 1.  In that
 * situation each iMCU row provides only one row group so the buffering logic
 * must be different (eg, we must read two iMCU rows before we can emit the
 * first row group).  For now, we simply do not support providing context
- * rows when min_DCT_scaled_size is 1.  That combination seems unlikely to
+ * rows when min_DCT_v_scaled_size is 1.  That combination seems unlikely to
 * be worth providing --- if someone wants a 1/8th-size preview, they probably
 * want it quick and dirty, so a context-free upsampler is sufficient.
 */
@@ -118,17 +118,18 @@ typedef struct {
  /* Pointer to allocated workspace (M or M+2 row groups). */
  JSAMPARRAY buffer[MAX_COMPONENTS];

-  boolean buffer_full;		/* Have we gotten an iMCU row from decoder? */
  JDIMENSION rowgroup_ctr;	/* counts row groups output to postprocessor */
+  JDIMENSION rowgroups_avail;	/* row groups available to postprocessor */

  /* Remaining fields are only used in the context case. */

+  boolean buffer_full;		/* Have we gotten an iMCU row from decoder? */
+
  /* These are the master pointers to the funny-order pointer lists. */
  JSAMPIMAGE xbuffer[2];	/* pointers to weird pointer lists */

  int whichptr;			/* indicates which pointer set is now in use */
  int context_state;		/* process_data state machine status */
-  JDIMENSION rowgroups_avail;	/* row groups available to postprocessor */
  JDIMENSION iMCU_row_ctr;	/* counts iMCU rows to detect image top/bot */
 } my_main_controller;

@@ -195,7 +196,7 @@ alloc_funny_pointers (j_decompress_ptr cinfo)
 LOCAL(void)
 make_funny_pointers (j_decompress_ptr cinfo)
 /* Create the funny pointer lists discussed in the comments above.
- * The actual workspace is already allocated (in main->buffer),
+ * The actual workspace is already allocated (in mainp->buffer),
 * and the space for the pointer lists is allocated too.
 * This routine just fills in the curiously ordered lists.
 * This will be repeated at the beginning of each pass.
@@ -317,12 +318,12 @@ start_pass_main (j_decompress_ptr cinfo, J_BUF_MODE pass_mode)
      mainp->whichptr = 0;	/* Read first iMCU row into xbuffer[0] */
      mainp->context_state = CTX_PREPARE_FOR_IMCU;
      mainp->iMCU_row_ctr = 0;
+      mainp->buffer_full = FALSE; /* Mark buffer empty */
    } else {
      /* Simple case with no context needed */
      mainp->pub.process_data = process_data_simple_main;
+      mainp->rowgroup_ctr = mainp->rowgroups_avail; /* Mark buffer empty */
    }
-    mainp->buffer_full = FALSE;	/* Mark buffer empty */
-    mainp->rowgroup_ctr = 0;
    break;
 #ifdef QUANT_2PASS_SUPPORTED
  case JBUF_CRANK_DEST:
@@ -348,17 +349,14 @@ process_data_simple_main (j_decompress_ptr cinfo,
 			  JDIMENSION out_rows_avail)
 {
  my_main_ptr mainp = (my_main_ptr) cinfo->main;
-  JDIMENSION rowgroups_avail;

  /* Read input data if we haven't filled the main buffer yet */
-  if (! mainp->buffer_full) {
+  if (mainp->rowgroup_ctr >= mainp->rowgroups_avail) {
    if (! (*cinfo->coef->decompress_data) (cinfo, mainp->buffer))
      return;			/* suspension forced, can do nothing more */
-    mainp->buffer_full = TRUE;	/* OK, we have an iMCU row to work with */
+    mainp->rowgroup_ctr = 0;	/* OK, we have an iMCU row to work with */
  }

-  /* There are always min_DCT_scaled_size row groups in an iMCU row. */
-  rowgroups_avail = (JDIMENSION) cinfo->min_DCT_v_scaled_size;
  /* Note: at the bottom of the image, we may pass extra garbage row groups
   * to the postprocessor.  The postprocessor has to check for bottom
   * of image anyway (at row resolution), so no point in us doing it too.
@@ -366,14 +364,8 @@ process_data_simple_main (j_decompress_ptr cinfo,

  /* Feed the postprocessor */
  (*cinfo->post->post_process_data) (cinfo, mainp->buffer,
-				     &mainp->rowgroup_ctr, rowgroups_avail,
-				     output_buf, out_row_ctr, out_rows_avail);
-
-  /* Has postprocessor consumed all the data yet? If so, mark buffer empty */
-  if (mainp->rowgroup_ctr >= rowgroups_avail) {
-    mainp->buffer_full = FALSE;
-    mainp->rowgroup_ctr = 0;
-  }
+			&mainp->rowgroup_ctr, mainp->rowgroups_avail,
+			output_buf, out_row_ctr, out_rows_avail);
 }


@@ -498,7 +490,9 @@ jinit_d_main_controller (j_decompress_ptr cinfo, boolean need_full_buffer)
    alloc_funny_pointers(cinfo); /* Alloc space for xbuffer[] lists */
    ngroups = cinfo->min_DCT_v_scaled_size + 2;
  } else {
+    /* There are always min_DCT_v_scaled_size row groups in an iMCU row. */
    ngroups = cinfo->min_DCT_v_scaled_size;
+    mainp->rowgroups_avail = (JDIMENSION) ngroups;
  }

  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
@@ -2,7 +2,7 @@
 * jdmaster.c
 *
 * Copyright (C) 1991-1997, Thomas G. Lane.
- * Modified 2002-2015 by Guido Vollbeding.
+ * Modified 2002-2017 by Guido Vollbeding.
 * This file is part of the Independent JPEG Group's software.
 * For conditions of distribution and use, see the accompanying README file.
 *
@@ -237,18 +237,17 @@ prepare_range_limit_table (j_decompress_ptr cinfo)
  JSAMPLE * table;
  int i;

-  table = (JSAMPLE *)
-    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-				5 * (MAXJSAMPLE+1) * SIZEOF(JSAMPLE));
+  table = (JSAMPLE *) (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo,
+    JPOOL_IMAGE, (RANGE_CENTER * 2 + MAXJSAMPLE + 1) * SIZEOF(JSAMPLE));
  /* First segment of range limit table: limit[x] = 0 for x < 0 */
-  MEMZERO(table, 2 * (MAXJSAMPLE+1) * SIZEOF(JSAMPLE));
-  table += 2 * (MAXJSAMPLE+1);	/* allow negative subscripts of table */
+  MEMZERO(table, RANGE_CENTER * SIZEOF(JSAMPLE));
+  table += RANGE_CENTER;	/* allow negative subscripts of table */
  cinfo->sample_range_limit = table;
  /* Main part of range limit table: limit[x] = x */
  for (i = 0; i <= MAXJSAMPLE; i++)
    table[i] = (JSAMPLE) i;
  /* End of range limit table: limit[x] = MAXJSAMPLE for x > MAXJSAMPLE */
-  for (; i < 3 * (MAXJSAMPLE+1); i++)
+  for (; i <=  MAXJSAMPLE + RANGE_CENTER; i++)
    table[i] = MAXJSAMPLE;
 }

@@ -2,7 +2,7 @@
 * jdmerge.c
 *
 * Copyright (C) 1994-1996, Thomas G. Lane.
- * Modified 2013-2015 by Guido Vollbeding.
+ * Modified 2013-2017 by Guido Vollbeding.
 * This file is part of the Independent JPEG Group's software.
 * For conditions of distribution and use, see the accompanying README file.
 *
@@ -40,6 +40,12 @@
 #ifdef UPSAMPLE_MERGING_SUPPORTED


+#if RANGE_BITS < 2
+  /* Deliberate syntax err */
+  Sorry, this code requires 2 or more range extension bits.
+#endif
+
+
 /* Private subobject */

 typedef struct {
@@ -2,7 +2,7 @@
 * jfdctflt.c
 *
 * Copyright (C) 1994-1996, Thomas G. Lane.
- * Modified 2003-2015 by Guido Vollbeding.
+ * Modified 2003-2017 by Guido Vollbeding.
 * This file is part of the Independent JPEG Group's software.
 * For conditions of distribution and use, see the accompanying README file.
 *
@@ -48,7 +48,7 @@
 */

 #if DCTSIZE != 8
-  Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
+  Sorry, this code only copes with 8x8 DCT blocks. /* deliberate syntax err */
 #endif


@@ -2,7 +2,7 @@
 * jfdctfst.c
 *
 * Copyright (C) 1994-1996, Thomas G. Lane.
- * Modified 2003-2015 by Guido Vollbeding.
+ * Modified 2003-2017 by Guido Vollbeding.
 * This file is part of the Independent JPEG Group's software.
 * For conditions of distribution and use, see the accompanying README file.
 *
@@ -44,7 +44,7 @@
 */

 #if DCTSIZE != 8
-  Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
+  Sorry, this code only copes with 8x8 DCT blocks. /* deliberate syntax err */
 #endif


@@ -2,7 +2,7 @@
 * jidctflt.c
 *
 * Copyright (C) 1994-1998, Thomas G. Lane.
- * Modified 2010-2015 by Guido Vollbeding.
+ * Modified 2010-2017 by Guido Vollbeding.
 * This file is part of the Independent JPEG Group's software.
 * For conditions of distribution and use, see the accompanying README file.
 *
@@ -50,7 +50,7 @@
 */

 #if DCTSIZE != 8
-  Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
+  Sorry, this code only copes with 8x8 DCT blocks. /* deliberate syntax err */
 #endif


@@ -2,7 +2,7 @@
 * jidctfst.c
 *
 * Copyright (C) 1994-1998, Thomas G. Lane.
- * Modified 2015 by Guido Vollbeding.
+ * Modified 2015-2017 by Guido Vollbeding.
 * This file is part of the Independent JPEG Group's software.
 * For conditions of distribution and use, see the accompanying README file.
 *
@@ -46,7 +46,7 @@
 */

 #if DCTSIZE != 8
-  Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
+  Sorry, this code only copes with 8x8 DCT blocks. /* deliberate syntax err */
 #endif


@@ -2,7 +2,7 @@
 * jidctint.c
 *
 * Copyright (C) 1991-1998, Thomas G. Lane.
- * Modification developed 2002-2015 by Guido Vollbeding.
+ * Modification developed 2002-2016 by Guido Vollbeding.
 * This file is part of the Independent JPEG Group's software.
 * For conditions of distribution and use, see the accompanying README file.
 *
@@ -166,6 +166,7 @@
 /*
 * Perform dequantization and inverse DCT on one block of coefficients.
 *
+ * Optimized algorithm with 12 multiplications in the 1-D kernel.
 * cK represents sqrt(2) * cos(K*pi/16).
 */

@@ -428,7 +429,7 @@ 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).
@@ -2623,7 +2624,7 @@ jpeg_idct_16x16 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    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] */
@@ -2920,13 +2921,6 @@ jpeg_idct_16x8 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
     * The rotator is c(-6).
     */

-    z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
-    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
-
-    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);       /* c6 */
-    tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865);     /* c2-c6 */
-    tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065);     /* c2+c6 */
-
    z2 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    z2 <<= CONST_BITS;
@@ -2937,6 +2931,13 @@ jpeg_idct_16x8 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
    tmp0 = z2 + z3;
    tmp1 = z2 - z3;

+    z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
+
+    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);       /* c6 */
+    tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865);     /* c2-c6 */
+    tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065);     /* c2+c6 */
+
    tmp10 = tmp0 + tmp2;
    tmp13 = tmp0 - tmp2;
    tmp11 = tmp1 + tmp3;
@@ -4883,13 +4884,6 @@ jpeg_idct_4x8 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
     * The rotator is c(-6).
     */

-    z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
-    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
-
-    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);       /* c6 */
-    tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865);     /* c2-c6 */
-    tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065);     /* c2+c6 */
-
    z2 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    z2 <<= CONST_BITS;
@@ -4900,6 +4894,13 @@ jpeg_idct_4x8 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
    tmp0 = z2 + z3;
    tmp1 = z2 - z3;

+    z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
+
+    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);       /* c6 */
+    tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865);     /* c2-c6 */
+    tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065);     /* c2+c6 */
+
    tmp10 = tmp0 + tmp2;
    tmp13 = tmp0 - tmp2;
    tmp11 = tmp1 + tmp3;
@@ -2,6 +2,7 @@
 * jinclude.h
 *
 * Copyright (C) 1991-1994, Thomas G. Lane.
+ * Modified 2017 by Guido Vollbeding.
 * This file is part of the Independent JPEG Group's software.
 * For conditions of distribution and use, see the accompanying README file.
 *
@@ -83,9 +84,14 @@
 * The modules that use fread() and fwrite() always invoke them through
 * these macros.  On some systems you may need to twiddle the argument casts.
 * CAUTION: argument order is different from underlying functions!
+ *
+ * Furthermore, macros are provided for fflush() and ferror() in order
+ * to facilitate adaption by applications using an own FILE class.
 */

 #define JFREAD(file,buf,sizeofbuf)  \
  ((size_t) fread((void *) (buf), (size_t) 1, (size_t) (sizeofbuf), (file)))
 #define JFWRITE(file,buf,sizeofbuf)  \
  ((size_t) fwrite((const void *) (buf), (size_t) 1, (size_t) (sizeofbuf), (file)))
+#define JFFLUSH(file)	fflush(file)
+#define JFERROR(file)	ferror(file)
@@ -2,7 +2,7 @@
 * jpegint.h
 *
 * Copyright (C) 1991-1997, Thomas G. Lane.
- * Modified 1997-2013 by Guido Vollbeding.
+ * Modified 1997-2017 by Guido Vollbeding.
 * This file is part of the Independent JPEG Group's software.
 * For conditions of distribution and use, see the accompanying README file.
 *
@@ -260,6 +260,19 @@ struct jpeg_color_quantizer {
 };


+/* Definition of range extension bits for decompression processes.
+ * See the comments with prepare_range_limit_table (in jdmaster.c)
+ * for more info.
+ * The recommended default value for normal applications is 2.
+ * Applications with special requirements may use a different value.
+ * For example, Ghostscript wants to use 3 for proper handling of
+ * wacky images with oversize coefficient values.
+ */
+
+#define RANGE_BITS	2
+#define RANGE_CENTER	(CENTERJSAMPLE << RANGE_BITS)
+
+
 /* Miscellaneous useful macros */

 #undef MAX
@@ -2,7 +2,7 @@
 * jpeglib.h
 *
 * Copyright (C) 1991-1998, Thomas G. Lane.
- * Modified 2002-2015 by Guido Vollbeding.
+ * Modified 2002-2017 by Guido Vollbeding.
 * This file is part of the Independent JPEG Group's software.
 * For conditions of distribution and use, see the accompanying README file.
 *
@@ -39,7 +39,7 @@ extern "C" {

 #define JPEG_LIB_VERSION        90	/* Compatibility version 9.0 */
 #define JPEG_LIB_VERSION_MAJOR  9
-#define JPEG_LIB_VERSION_MINOR  2
+#define JPEG_LIB_VERSION_MINOR  3


 /* Various constants determining the sizes of things.
@@ -137,9 +137,9 @@ typedef struct {
  /* The decompressor output side may not use these variables. */
  int dc_tbl_no;		/* DC entropy table selector (0..3) */
  int ac_tbl_no;		/* AC entropy table selector (0..3) */
-  
+
  /* Remaining fields should be treated as private by applications. */
-  
+
  /* These values are computed during compression or decompression startup: */
  /* Component's size in DCT blocks.
   * Any dummy blocks added to complete an MCU are not counted; therefore
@@ -411,10 +411,10 @@ struct jpeg_compress_struct {
  JDIMENSION total_iMCU_rows;	/* # of iMCU rows to be input to coef ctlr */
  /* The coefficient controller receives data in units of MCU rows as defined
   * for fully interleaved scans (whether the JPEG file is interleaved or not).
-   * There are v_samp_factor * DCTSIZE sample rows of each component in an
-   * "iMCU" (interleaved MCU) row.
+   * There are v_samp_factor * DCT_v_scaled_size sample rows of each component
+   * in an "iMCU" (interleaved MCU) row.
   */
-  
+
  /*
   * These fields are valid during any one scan.
   * They describe the components and MCUs actually appearing in the scan.
@@ -422,10 +422,10 @@ struct jpeg_compress_struct {
  int comps_in_scan;		/* # of JPEG components in this scan */
  jpeg_component_info * cur_comp_info[MAX_COMPS_IN_SCAN];
  /* *cur_comp_info[i] describes component that appears i'th in SOS */
-  
+
  JDIMENSION MCUs_per_row;	/* # of MCUs across the image */
  JDIMENSION MCU_rows_in_scan;	/* # of MCU rows in the image */
-  
+
  int blocks_in_MCU;		/* # of DCT blocks per MCU */
  int MCU_membership[C_MAX_BLOCKS_IN_MCU];
  /* MCU_membership[i] is index in cur_comp_info of component owning */
@@ -636,7 +636,7 @@ struct jpeg_decompress_struct {
   * in fully interleaved JPEG scans, but are used whether the scan is
   * interleaved or not.  We define an iMCU row as v_samp_factor DCT block
   * rows of each component.  Therefore, the IDCT output contains
-   * v_samp_factor*DCT_v_scaled_size sample rows of a component per iMCU row.
+   * v_samp_factor * DCT_v_scaled_size sample rows of a component per iMCU row.
   */

  JSAMPLE * sample_range_limit; /* table for fast range-limiting */
@@ -711,7 +711,7 @@ struct jpeg_error_mgr {
 #define JMSG_LENGTH_MAX  200	/* recommended size of format_message buffer */
  /* Reset error state variables at start of a new image */
  JMETHOD(void, reset_error_mgr, (j_common_ptr cinfo));
-  
+
  /* The message ID code and any parameters are saved here.
   * A message can have one string parameter or up to 8 int parameters.
   */
@@ -721,11 +721,11 @@ struct jpeg_error_mgr {
    int i[8];
    char s[JMSG_STR_PARM_MAX];
  } msg_parm;
-  
+
  /* Standard state variables for error facility */
-  
+
  int trace_level;		/* max msg_level that will be displayed */
-  
+
  /* For recoverable corrupt-data errors, we emit a warning message,
   * but keep going unless emit_message chooses to abort.  emit_message
   * should count warnings in num_warnings.  The surrounding application
@@ -1,7 +1,7 @@
 /*
 * jversion.h
 *
- * Copyright (C) 1991-2016, Thomas G. Lane, Guido Vollbeding.
+ * Copyright (C) 1991-2018, Thomas G. Lane, Guido Vollbeding.
 * This file is part of the Independent JPEG Group's software.
 * For conditions of distribution and use, see the accompanying README file.
 *
@@ -9,6 +9,6 @@
 */


-#define JVERSION	"9b  17-Jan-2016"
+#define JVERSION	"9c  14-Jan-2018"

-#define JCOPYRIGHT	"Copyright (C) 2016, Thomas G. Lane, Guido Vollbeding"
+#define JCOPYRIGHT	"Copyright (C) 2018, Thomas G. Lane, Guido Vollbeding"
@@ -6066,31 +6066,44 @@ Version 1.6.35 [July 15, 2018]
 Version 1.6.36 [December 1, 2018]
  Optimized png_do_expand_palette for ARM processors.
  Improved performance by around 10-22% on a recent ARM Chromebook.
-  (Contributed by Richard Townsend, ARM Holdings)
+    (Contributed by Richard Townsend, ARM Holdings)
  Fixed manipulation of machine-specific optimization options.
-  (Contributed by Vicki Pfau)
+    (Contributed by Vicki Pfau)
  Used memcpy instead of manual pointer arithmetic on Intel SSE2.
-  (Contributed by Samuel Williams)
+    (Contributed by Samuel Williams)
  Fixed build errors with MSVC on ARM64.
-  (Contributed by Zhijie Liang)
+    (Contributed by Zhijie Liang)
  Fixed detection of libm in CMakeLists.
-  (Contributed by Cameron Cawley)
+    (Contributed by Cameron Cawley)
  Fixed incorrect creation of pkg-config file in CMakeLists.
-  (Contributed by Kyle Bentley)
+    (Contributed by Kyle Bentley)
  Fixed the CMake build on Windows MSYS by avoiding symlinks.
  Fixed a build warning on OpenBSD.
-  (Contributed by Theo Buehler)
+    (Contributed by Theo Buehler)
  Fixed various typos in comments.
-  (Contributed by "luz.paz")
+    (Contributed by "luz.paz")
  Raised the minimum required CMake version from 3.0.2 to 3.1.
  Removed yet more of the vestigial support for pre-ANSI C compilers.
  Removed ancient makefiles for ancient systems that have been broken
-  across all previous libpng-1.6.x versions.
+    across all previous libpng-1.6.x versions.
  Removed the Y2K compliance statement and the export control
-  information.
+    information.
  Applied various code style and documentation fixes.

-Send comments/corrections/commendations to png-mng-implement at lists.sf.net
-(subscription required; visit
+Version 1.6.37 [April 14, 2019]
+  Fixed a use-after-free vulnerability (CVE-2019-7317) in png_image_free.
+  Fixed a memory leak in the ARM NEON implementation of png_do_expand_palette.
+  Fixed a memory leak in pngtest.c.
+  Fixed two vulnerabilities (CVE-2018-14048, CVE-2018-14550) in
+    contrib/pngminus; refactor.
+  Changed the license of contrib/pngminus to MIT; refresh makefile and docs.
+    (Contributed by Willem van Schaik)
+  Fixed a typo in the libpng license v2.
+    (Contributed by Miguel Ojeda)
+  Added makefiles for AddressSanitizer-enabled builds.
+  Cleaned up various makefiles.
+
+Send comments/corrections/commendations to png-mng-implement at lists.sf.net.
+Subscription is required; visit
 https://lists.sourceforge.net/lists/listinfo/png-mng-implement
-to subscribe).
+to subscribe.
@@ -46,8 +46,19 @@ if(";${CPU_BASELINE_FINAL};" MATCHES "SSE2"
  add_definitions(-DPNG_INTEL_SSE)
 endif()

+# set definitions and sources for MIPS
+if(";${CPU_BASELINE_FINAL};" MATCHES "MSA")
+    list(APPEND lib_srcs mips/mips_init.c mips/filter_msa_intrinsics.c)
+    add_definitions(-DPNG_MIPS_MSA_OPT=2)
+    ocv_warnings_disable(CMAKE_C_FLAGS -Wshadow)
+else()
+    add_definitions(-DPNG_MIPS_MSA_OPT=0)
+endif()
+
 if(PPC64LE OR PPC64)
-  if(ENABLE_VSX AND NOT PPC64)
+  # VSX3 features are backwards compatible
+  if(";${CPU_BASELINE_FINAL};" MATCHES "VSX.*"
+      AND NOT PPC64)
    list(APPEND lib_srcs powerpc/powerpc_init.c powerpc/filter_vsx_intrinsics.c)
    add_definitions(-DPNG_POWERPC_VSX_OPT=2)
  else()
@@ -4,8 +4,8 @@ COPYRIGHT NOTICE, DISCLAIMER, and LICENSE
 PNG Reference Library License version 2
 ---------------------------------------

- * Copyright (c) 1995-2018 The PNG Reference Library Authors.
- * Copyright (c) 2018 Cosmin Truta.
+ * Copyright (c) 1995-2019 The PNG Reference Library Authors.
+ * Copyright (c) 2018-2019 Cosmin Truta.
 * Copyright (c) 2000-2002, 2004, 2006-2018 Glenn Randers-Pehrson.
 * Copyright (c) 1996-1997 Andreas Dilger.
 * Copyright (c) 1995-1996 Guy Eric Schalnat, Group 42, Inc.
@@ -13,7 +13,7 @@ PNG Reference Library License version 2
 The software is supplied "as is", without warranty of any kind,
 express or implied, including, without limitation, the warranties
 of merchantability, fitness for a particular purpose, title, and
-non-infringement.  In no even shall the Copyright owners, or
+non-infringement.  In no event shall the Copyright owners, or
 anyone distributing the software, be liable for any damages or
 other liability, whether in contract, tort or otherwise, arising
 from, out of, or in connection with the software, or the use or
@@ -39,7 +39,7 @@ subject to the following restrictions:
 PNG Reference Library License version 1 (for libpng 0.5 through 1.6.35)
 -----------------------------------------------------------------------

-libpng versions 1.0.7, July 1, 2000 through 1.6.35, July 15, 2018 are
+libpng versions 1.0.7, July 1, 2000, through 1.6.35, July 15, 2018 are
 Copyright (c) 2000-2002, 2004, 2006-2018 Glenn Randers-Pehrson, are
 derived from libpng-1.0.6, and are distributed according to the same
 disclaimer and license as libpng-1.0.6 with the following individuals
@@ -1,5 +1,5 @@
-README for libpng version 1.6.36 - December 1, 2018
-===================================================
+README for libpng version 1.6.37 - April 14, 2019
+=================================================

 See the note about version numbers near the top of png.h.
 See INSTALL for instructions on how to install libpng.
@@ -1,7 +1,7 @@

 /* palette_neon_intrinsics.c - NEON optimised palette expansion functions
 *
- * Copyright (c) 2018 Cosmin Truta
+ * Copyright (c) 2018-2019 Cosmin Truta
 * Copyright (c) 2017-2018 Arm Holdings. All rights reserved.
 * Written by Richard Townsend <Richard.Townsend@arm.com>, February 2017.
 *
@@ -20,9 +20,9 @@
 #  include <arm_neon.h>
 #endif

-/* Build an RGBA palette from the RGB and separate alpha palettes. */
+/* Build an RGBA8 palette from the separate RGB and alpha palettes. */
 void
-png_riffle_palette_rgba(png_structrp png_ptr, png_row_infop row_info)
+png_riffle_palette_neon(png_structrp png_ptr)
 {
   png_const_colorp palette = png_ptr->palette;
   png_bytep riffled_palette = png_ptr->riffled_palette;
@@ -30,6 +30,8 @@ png_riffle_palette_rgba(png_structrp png_ptr, png_row_infop row_info)
   int num_trans = png_ptr->num_trans;
   int i;

+   png_debug(1, "in png_riffle_palette_neon");
+
   /* Initially black, opaque. */
   uint8x16x4_t w = {{
      vdupq_n_u8(0x00),
@@ -38,16 +40,10 @@ png_riffle_palette_rgba(png_structrp png_ptr, png_row_infop row_info)
      vdupq_n_u8(0xff),
   }};

-   if (row_info->bit_depth != 8)
-   {
-      png_error(png_ptr, "bit_depth must be 8 for png_riffle_palette_rgba");
-      return;
-   }
-
-   /* First, riffle the RGB colours into a RGBA palette, the A value is
-    * set to opaque for now.
+   /* First, riffle the RGB colours into an RGBA8 palette.
+    * The alpha component is set to opaque for now.
    */
-   for (i = 0; i < (1 << row_info->bit_depth); i += 16)
+   for (i = 0; i < 256; i += 16)
   {
      uint8x16x3_t v = vld3q_u8((png_const_bytep)(palette + i));
      w.val[0] = v.val[0];
@@ -61,9 +57,9 @@ png_riffle_palette_rgba(png_structrp png_ptr, png_row_infop row_info)
      riffled_palette[(i << 2) + 3] = trans_alpha[i];
 }

-/* Expands a palettized row into RGBA. */
+/* Expands a palettized row into RGBA8. */
 int
-png_do_expand_palette_neon_rgba(png_structrp png_ptr, png_row_infop row_info,
+png_do_expand_palette_rgba8_neon(png_structrp png_ptr, png_row_infop row_info,
    png_const_bytep row, png_bytepp ssp, png_bytepp ddp)
 {
   png_uint_32 row_width = row_info->width;
@@ -72,6 +68,8 @@ png_do_expand_palette_neon_rgba(png_structrp png_ptr, png_row_infop row_info,
   const png_int_32 pixels_per_chunk = 4;
   int i;

+   png_debug(1, "in png_do_expand_palette_rgba8_neon");
+
   if (row_width < pixels_per_chunk)
      return 0;

@@ -103,9 +101,9 @@ png_do_expand_palette_neon_rgba(png_structrp png_ptr, png_row_infop row_info,
   return i;
 }

-/* Expands a palettized row into RGB format. */
+/* Expands a palettized row into RGB8. */
 int
-png_do_expand_palette_neon_rgb(png_structrp png_ptr, png_row_infop row_info,
+png_do_expand_palette_rgb8_neon(png_structrp png_ptr, png_row_infop row_info,
    png_const_bytep row, png_bytepp ssp, png_bytepp ddp)
 {
   png_uint_32 row_width = row_info->width;
@@ -113,6 +111,8 @@ png_do_expand_palette_neon_rgb(png_structrp png_ptr, png_row_infop row_info,
   const png_uint_32 pixels_per_chunk = 8;
   int i;

+   png_debug(1, "in png_do_expand_palette_rgb8_neon");
+
   if (row_width <= pixels_per_chunk)
      return 0;

@@ -0,0 +1,808 @@
+
+/* filter_msa_intrinsics.c - MSA optimised filter functions
+ *
+ * Copyright (c) 2018 Cosmin Truta
+ * Copyright (c) 2016 Glenn Randers-Pehrson
+ * Written by Mandar Sahastrabuddhe, August 2016.
+ *
+ * This code is released under the libpng license.
+ * For conditions of distribution and use, see the disclaimer
+ * and license in png.h
+ */
+
+#include <stdio.h>
+#include <stdint.h>
+#include "../pngpriv.h"
+
+#ifdef PNG_READ_SUPPORTED
+
+/* This code requires -mfpu=msa on the command line: */
+#if PNG_MIPS_MSA_IMPLEMENTATION == 1 /* intrinsics code from pngpriv.h */
+
+#include <msa.h>
+
+/* libpng row pointers are not necessarily aligned to any particular boundary,
+ * however this code will only work with appropriate alignment. mips/mips_init.c
+ * checks for this (and will not compile unless it is done). This code uses
+ * variants of png_aligncast to avoid compiler warnings.
+ */
+#define png_ptr(type,pointer) png_aligncast(type *,pointer)
+#define png_ptrc(type,pointer) png_aligncastconst(const type *,pointer)
+
+/* The following relies on a variable 'temp_pointer' being declared with type
+ * 'type'.  This is written this way just to hide the GCC strict aliasing
+ * warning; note that the code is safe because there never is an alias between
+ * the input and output pointers.
+ */
+#define png_ldr(type,pointer)\
+   (temp_pointer = png_ptr(type,pointer), *temp_pointer)
+
+#if PNG_MIPS_MSA_OPT > 0
+
+#ifdef CLANG_BUILD
+   #define MSA_SRLI_B(a, b)   __msa_srli_b((v16i8) a, b)
+
+   #define LW(psrc)                              \
+   ( {                                           \
+       uint8_t *psrc_lw_m = (uint8_t *) (psrc);  \
+       uint32_t val_m;                           \
+                                                 \
+       asm volatile (                            \
+           "lw  %[val_m],  %[psrc_lw_m]  \n\t"   \
+                                                 \
+           : [val_m] "=r" (val_m)                \
+           : [psrc_lw_m] "m" (*psrc_lw_m)        \
+       );                                        \
+                                                 \
+       val_m;                                    \
+   } )
+
+   #define SH(val, pdst)                         \
+   {                                             \
+       uint8_t *pdst_sh_m = (uint8_t *) (pdst);  \
+       uint16_t val_m = (val);                   \
+                                                 \
+       asm volatile (                            \
+           "sh  %[val_m],  %[pdst_sh_m]  \n\t"   \
+                                                 \
+           : [pdst_sh_m] "=m" (*pdst_sh_m)       \
+           : [val_m] "r" (val_m)                 \
+       );                                        \
+   }
+
+   #define SW(val, pdst)                         \
+   {                                             \
+       uint8_t *pdst_sw_m = (uint8_t *) (pdst);  \
+       uint32_t val_m = (val);                   \
+                                                 \
+       asm volatile (                            \
+           "sw  %[val_m],  %[pdst_sw_m]  \n\t"   \
+                                                 \
+           : [pdst_sw_m] "=m" (*pdst_sw_m)       \
+           : [val_m] "r" (val_m)                 \
+       );                                        \
+   }
+
+       #if (__mips == 64)
+        #define SD(val, pdst)                         \
+        {                                             \
+            uint8_t *pdst_sd_m = (uint8_t *) (pdst);  \
+            uint64_t val_m = (val);                   \
+                                                      \
+            asm volatile (                            \
+                "sd  %[val_m],  %[pdst_sd_m]  \n\t"   \
+                                                      \
+                : [pdst_sd_m] "=m" (*pdst_sd_m)       \
+                : [val_m] "r" (val_m)                 \
+            );                                        \
+        }
+    #else
+        #define SD(val, pdst)                                          \
+        {                                                              \
+            uint8_t *pdst_sd_m = (uint8_t *) (pdst);                   \
+            uint32_t val0_m, val1_m;                                   \
+                                                                       \
+            val0_m = (uint32_t) ((val) & 0x00000000FFFFFFFF);          \
+            val1_m = (uint32_t) (((val) >> 32) & 0x00000000FFFFFFFF);  \
+                                                                       \
+            SW(val0_m, pdst_sd_m);                                     \
+            SW(val1_m, pdst_sd_m + 4);                                 \
+        }
+    #endif
+#else
+   #define MSA_SRLI_B(a, b)   (a >> b)
+
+#if (__mips_isa_rev >= 6)
+   #define LW(psrc)                              \
+   ( {                                           \
+       uint8_t *psrc_lw_m = (uint8_t *) (psrc);  \
+       uint32_t val_m;                           \
+                                                 \
+       asm volatile (                            \
+           "lw  %[val_m],  %[psrc_lw_m]  \n\t"   \
+                                                 \
+           : [val_m] "=r" (val_m)                \
+           : [psrc_lw_m] "m" (*psrc_lw_m)        \
+       );                                        \
+                                                 \
+       val_m;                                    \
+   } )
+
+   #define SH(val, pdst)                         \
+   {                                             \
+       uint8_t *pdst_sh_m = (uint8_t *) (pdst);  \
+       uint16_t val_m = (val);                   \
+                                                 \
+       asm volatile (                            \
+           "sh  %[val_m],  %[pdst_sh_m]  \n\t"   \
+                                                 \
+           : [pdst_sh_m] "=m" (*pdst_sh_m)       \
+           : [val_m] "r" (val_m)                 \
+       );                                        \
+   }
+
+   #define SW(val, pdst)                         \
+   {                                             \
+       uint8_t *pdst_sw_m = (uint8_t *) (pdst);  \
+       uint32_t val_m = (val);                   \
+                                                 \
+       asm volatile (                            \
+           "sw  %[val_m],  %[pdst_sw_m]  \n\t"   \
+                                                 \
+           : [pdst_sw_m] "=m" (*pdst_sw_m)       \
+           : [val_m] "r" (val_m)                 \
+       );                                        \
+   }
+
+   #if (__mips == 64)
+        #define SD(val, pdst)                         \
+        {                                             \
+            uint8_t *pdst_sd_m = (uint8_t *) (pdst);  \
+            uint64_t val_m = (val);                   \
+                                                      \
+            asm volatile (                            \
+                "sd  %[val_m],  %[pdst_sd_m]  \n\t"   \
+                                                      \
+                : [pdst_sd_m] "=m" (*pdst_sd_m)       \
+                : [val_m] "r" (val_m)                 \
+            );                                        \
+        }
+    #else
+        #define SD(val, pdst)                                          \
+        {                                                              \
+            uint8_t *pdst_sd_m = (uint8_t *) (pdst);                   \
+            uint32_t val0_m, val1_m;                                   \
+                                                                       \
+            val0_m = (uint32_t) ((val) & 0x00000000FFFFFFFF);          \
+            val1_m = (uint32_t) (((val) >> 32) & 0x00000000FFFFFFFF);  \
+                                                                       \
+            SW(val0_m, pdst_sd_m);                                     \
+            SW(val1_m, pdst_sd_m + 4);                                 \
+        }
+    #endif
+#else  // !(__mips_isa_rev >= 6)
+   #define LW(psrc)                              \
+   ( {                                           \
+       uint8_t *psrc_lw_m = (uint8_t *) (psrc);  \
+       uint32_t val_m;                           \
+                                                 \
+       asm volatile (                            \
+           "ulw  %[val_m],  %[psrc_lw_m]  \n\t"  \
+                                                 \
+           : [val_m] "=r" (val_m)                \
+           : [psrc_lw_m] "m" (*psrc_lw_m)        \
+       );                                        \
+                                                 \
+       val_m;                                    \
+   } )
+
+   #define SH(val, pdst)                         \
+   {                                             \
+       uint8_t *pdst_sh_m = (uint8_t *) (pdst);  \
+       uint16_t val_m = (val);                   \
+                                                 \
+       asm volatile (                            \
+           "ush  %[val_m],  %[pdst_sh_m]  \n\t"  \
+                                                 \
+           : [pdst_sh_m] "=m" (*pdst_sh_m)       \
+           : [val_m] "r" (val_m)                 \
+       );                                        \
+   }
+
+   #define SW(val, pdst)                         \
+   {                                             \
+       uint8_t *pdst_sw_m = (uint8_t *) (pdst);  \
+       uint32_t val_m = (val);                   \
+                                                 \
+       asm volatile (                            \
+           "usw  %[val_m],  %[pdst_sw_m]  \n\t"  \
+                                                 \
+           : [pdst_sw_m] "=m" (*pdst_sw_m)       \
+           : [val_m] "r" (val_m)                 \
+       );                                        \
+   }
+
+   #define SD(val, pdst)                                          \
+    {                                                              \
+        uint8_t *pdst_sd_m = (uint8_t *) (pdst);                   \
+        uint32_t val0_m, val1_m;                                   \
+                                                                   \
+        val0_m = (uint32_t) ((val) & 0x00000000FFFFFFFF);          \
+        val1_m = (uint32_t) (((val) >> 32) & 0x00000000FFFFFFFF);  \
+                                                                   \
+        SW(val0_m, pdst_sd_m);                                     \
+        SW(val1_m, pdst_sd_m + 4);                                 \
+    }
+
+    #define SW_ZERO(pdst)                      \
+    {                                          \
+        uint8_t *pdst_m = (uint8_t *) (pdst);  \
+                                               \
+        asm volatile (                         \
+            "usw  $0,  %[pdst_m]  \n\t"        \
+                                               \
+            : [pdst_m] "=m" (*pdst_m)          \
+            :                                  \
+        );                                     \
+    }
+#endif  // (__mips_isa_rev >= 6)
+#endif
+
+#define LD_B(RTYPE, psrc) *((RTYPE *) (psrc))
+#define LD_UB(...) LD_B(v16u8, __VA_ARGS__)
+#define LD_B2(RTYPE, psrc, stride, out0, out1)  \
+{                                               \
+    out0 = LD_B(RTYPE, (psrc));                 \
+    out1 = LD_B(RTYPE, (psrc) + stride);        \
+}
+#define LD_UB2(...) LD_B2(v16u8, __VA_ARGS__)
+#define LD_B4(RTYPE, psrc, stride, out0, out1, out2, out3)   \
+{                                                            \
+    LD_B2(RTYPE, (psrc), stride, out0, out1);                \
+    LD_B2(RTYPE, (psrc) + 2 * stride , stride, out2, out3);  \
+}
+#define LD_UB4(...) LD_B4(v16u8, __VA_ARGS__)
+
+#define ST_B(RTYPE, in, pdst) *((RTYPE *) (pdst)) = (in)
+#define ST_UB(...) ST_B(v16u8, __VA_ARGS__)
+#define ST_B2(RTYPE, in0, in1, pdst, stride)  \
+{                                             \
+    ST_B(RTYPE, in0, (pdst));                 \
+    ST_B(RTYPE, in1, (pdst) + stride);        \
+}
+#define ST_UB2(...) ST_B2(v16u8, __VA_ARGS__)
+#define ST_B4(RTYPE, in0, in1, in2, in3, pdst, stride)    \
+{                                                         \
+    ST_B2(RTYPE, in0, in1, (pdst), stride);               \
+    ST_B2(RTYPE, in2, in3, (pdst) + 2 * stride, stride);  \
+}
+#define ST_UB4(...) ST_B4(v16u8, __VA_ARGS__)
+
+#define ADD2(in0, in1, in2, in3, out0, out1)  \
+{                                             \
+    out0 = in0 + in1;                         \
+    out1 = in2 + in3;                         \
+}
+#define ADD3(in0, in1, in2, in3, in4, in5,  \
+             out0, out1, out2)              \
+{                                           \
+    ADD2(in0, in1, in2, in3, out0, out1);   \
+    out2 = in4 + in5;                       \
+}
+#define ADD4(in0, in1, in2, in3, in4, in5, in6, in7,  \
+             out0, out1, out2, out3)                  \
+{                                                     \
+    ADD2(in0, in1, in2, in3, out0, out1);             \
+    ADD2(in4, in5, in6, in7, out2, out3);             \
+}
+
+#define ILVR_B2(RTYPE, in0, in1, in2, in3, out0, out1)      \
+{                                                           \
+    out0 = (RTYPE) __msa_ilvr_b((v16i8) in0, (v16i8) in1);  \
+    out1 = (RTYPE) __msa_ilvr_b((v16i8) in2, (v16i8) in3);  \
+}
+#define ILVR_B2_SH(...) ILVR_B2(v8i16, __VA_ARGS__)
+
+#define HSUB_UB2(RTYPE, in0, in1, out0, out1)                 \
+{                                                             \
+    out0 = (RTYPE) __msa_hsub_u_h((v16u8) in0, (v16u8) in0);  \
+    out1 = (RTYPE) __msa_hsub_u_h((v16u8) in1, (v16u8) in1);  \
+}
+#define HSUB_UB2_SH(...) HSUB_UB2(v8i16, __VA_ARGS__)
+
+#define SLDI_B2_0(RTYPE, in0, in1, out0, out1, slide_val)                 \
+{                                                                         \
+    v16i8 zero_m = { 0 };                                                 \
+    out0 = (RTYPE) __msa_sldi_b((v16i8) zero_m, (v16i8) in0, slide_val);  \
+    out1 = (RTYPE) __msa_sldi_b((v16i8) zero_m, (v16i8) in1, slide_val);  \
+}
+#define SLDI_B2_0_UB(...) SLDI_B2_0(v16u8, __VA_ARGS__)
+
+#define SLDI_B3_0(RTYPE, in0, in1, in2, out0, out1, out2,  slide_val)     \
+{                                                                         \
+    v16i8 zero_m = { 0 };                                                 \
+    SLDI_B2_0(RTYPE, in0, in1, out0, out1, slide_val);                    \
+    out2 = (RTYPE) __msa_sldi_b((v16i8) zero_m, (v16i8) in2, slide_val);  \
+}
+#define SLDI_B3_0_UB(...) SLDI_B3_0(v16u8, __VA_ARGS__)
+
+#define ILVEV_W2(RTYPE, in0, in1, in2, in3, out0, out1)      \
+{                                                            \
+    out0 = (RTYPE) __msa_ilvev_w((v4i32) in1, (v4i32) in0);  \
+    out1 = (RTYPE) __msa_ilvev_w((v4i32) in3, (v4i32) in2);  \
+}
+#define ILVEV_W2_UB(...) ILVEV_W2(v16u8, __VA_ARGS__)
+
+#define ADD_ABS_H3(RTYPE, in0, in1, in2, out0, out1, out2)  \
+{                                                           \
+    RTYPE zero = {0};                                       \
+                                                            \
+    out0 = __msa_add_a_h((v8i16) zero, in0);                \
+    out1 = __msa_add_a_h((v8i16) zero, in1);                \
+    out2 = __msa_add_a_h((v8i16) zero, in2);                \
+}
+#define ADD_ABS_H3_SH(...) ADD_ABS_H3(v8i16, __VA_ARGS__)
+
+#define VSHF_B2(RTYPE, in0, in1, in2, in3, mask0, mask1, out0, out1)       \
+{                                                                          \
+    out0 = (RTYPE) __msa_vshf_b((v16i8) mask0, (v16i8) in1, (v16i8) in0);  \
+    out1 = (RTYPE) __msa_vshf_b((v16i8) mask1, (v16i8) in3, (v16i8) in2);  \
+}
+#define VSHF_B2_UB(...) VSHF_B2(v16u8, __VA_ARGS__)
+
+#define CMP_AND_SELECT(inp0, inp1, inp2, inp3, inp4, inp5, out0)              \
+{                                                                             \
+   v8i16 _sel_h0, _sel_h1;                                                    \
+   v16u8 _sel_b0, _sel_b1;                                                    \
+   _sel_h0 = (v8i16) __msa_clt_u_h((v8u16) inp1, (v8u16) inp0);               \
+   _sel_b0 = (v16u8) __msa_pckev_b((v16i8) _sel_h0, (v16i8) _sel_h0);         \
+   inp0 = (v8i16) __msa_bmnz_v((v16u8) inp0, (v16u8) inp1, (v16u8) _sel_h0);  \
+   inp4 = (v16u8) __msa_bmnz_v(inp3, inp4, _sel_b0);                          \
+   _sel_h1 = (v8i16) __msa_clt_u_h((v8u16) inp2, (v8u16) inp0);               \
+   _sel_b1 = (v16u8) __msa_pckev_b((v16i8) _sel_h1, (v16i8) _sel_h1);         \
+   inp4 = (v16u8) __msa_bmnz_v(inp4, inp5, _sel_b1);                          \
+   out0 += inp4;                                                              \
+}
+
+void png_read_filter_row_up_msa(png_row_infop row_info, png_bytep row,
+                                png_const_bytep prev_row)
+{
+   size_t i, cnt, cnt16, cnt32;
+   size_t istop = row_info->rowbytes;
+   png_bytep rp = row;
+   png_const_bytep pp = prev_row;
+   v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
+
+   for (i = 0; i < (istop >> 6); i++)
+   {
+      LD_UB4(rp, 16, src0, src1, src2, src3);
+      LD_UB4(pp, 16, src4, src5, src6, src7);
+      pp += 64;
+
+	  ADD4(src0, src4, src1, src5, src2, src6, src3, src7,
+	       src0, src1, src2, src3);
+
+      ST_UB4(src0, src1, src2, src3, rp, 16);
+      rp += 64;
+   }
+
+   if (istop & 0x3F)
+   {
+      cnt32 = istop & 0x20;
+      cnt16 = istop & 0x10;
+      cnt = istop & 0xF;
+
+      if(cnt32)
+      {
+         if (cnt16 && cnt)
+         {
+            LD_UB4(rp, 16, src0, src1, src2, src3);
+            LD_UB4(pp, 16, src4, src5, src6, src7);
+
+            ADD4(src0, src4, src1, src5, src2, src6, src3, src7,
+	             src0, src1, src2, src3);
+
+            ST_UB4(src0, src1, src2, src3, rp, 16);
+            rp += 64;
+         }
+         else if (cnt16 || cnt)
+         {
+            LD_UB2(rp, 16, src0, src1);
+            LD_UB2(pp, 16, src4, src5);
+            pp += 32;
+            src2 = LD_UB(rp + 32);
+            src6 = LD_UB(pp);
+
+            ADD3(src0, src4, src1, src5, src2, src6, src0, src1, src2);
+
+            ST_UB2(src0, src1, rp, 16);
+            rp += 32;
+            ST_UB(src2, rp);
+            rp += 16;
+         }
+         else
+         {
+            LD_UB2(rp, 16, src0, src1);
+            LD_UB2(pp, 16, src4, src5);
+
+			ADD2(src0, src4, src1, src5, src0, src1);
+
+            ST_UB2(src0, src1, rp, 16);
+            rp += 32;
+         }
+      }
+      else if (cnt16 && cnt)
+      {
+         LD_UB2(rp, 16, src0, src1);
+         LD_UB2(pp, 16, src4, src5);
+
+         ADD2(src0, src4, src1, src5, src0, src1);
+
+         ST_UB2(src0, src1, rp, 16);
+         rp += 32;
+      }
+      else if (cnt16 || cnt)
+      {
+         src0 = LD_UB(rp);
+         src4 = LD_UB(pp);
+         pp += 16;
+
+         src0 += src4;
+
+         ST_UB(src0, rp);
+         rp += 16;
+      }
+   }
+}
+
+void png_read_filter_row_sub4_msa(png_row_infop row_info, png_bytep row,
+                                  png_const_bytep prev_row)
+{
+   size_t count;
+   size_t istop = row_info->rowbytes;
+   png_bytep src = row;
+   png_bytep nxt = row + 4;
+   int32_t inp0;
+   v16u8 src0, src1, src2, src3, src4;
+   v16u8 dst0, dst1;
+   v16u8 zero = { 0 };
+
+   istop -= 4;
+
+   inp0 = LW(src);
+   src += 4;
+   src0 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp0);
+
+   for (count = 0; count < istop; count += 16)
+   {
+      src1 = LD_UB(src);
+      src += 16;
+
+      src2 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) src1, 4);
+      src3 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) src1, 8);
+      src4 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) src1, 12);
+      src1 += src0;
+      src2 += src1;
+      src3 += src2;
+      src4 += src3;
+      src0 = src4;
+      ILVEV_W2_UB(src1, src2, src3, src4, dst0, dst1);
+      dst0 = (v16u8) __msa_pckev_d((v2i64) dst1, (v2i64) dst0);
+
+      ST_UB(dst0, nxt);
+      nxt += 16;
+   }
+}
+
+void png_read_filter_row_sub3_msa(png_row_infop row_info, png_bytep row,
+                                  png_const_bytep prev_row)
+{
+   size_t count;
+   size_t istop = row_info->rowbytes;
+   png_bytep src = row;
+   png_bytep nxt = row + 3;
+   int64_t out0;
+   int32_t inp0, out1;
+   v16u8 src0, src1, src2, src3, src4, dst0, dst1;
+   v16u8 zero = { 0 };
+   v16i8 mask0 = { 0, 1, 2, 16, 17, 18, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+   v16i8 mask1 = { 0, 1, 2, 3, 4, 5, 16, 17, 18, 19, 20, 21, 0, 0, 0, 0 };
+
+   istop -= 3;
+
+   inp0 = LW(src);
+   src += 3;
+   src0 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp0);
+
+   for (count = 0; count < istop; count += 12)
+   {
+      src1 = LD_UB(src);
+      src += 12;
+
+      src2 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) src1, 3);
+      src3 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) src1, 6);
+      src4 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) src1, 9);
+      src1 += src0;
+      src2 += src1;
+      src3 += src2;
+      src4 += src3;
+      src0 = src4;
+      VSHF_B2_UB(src1, src2, src3, src4, mask0, mask0, dst0, dst1);
+      dst0 = (v16u8) __msa_vshf_b(mask1, (v16i8) dst1, (v16i8) dst0);
+      out0 = __msa_copy_s_d((v2i64) dst0, 0);
+      out1 = __msa_copy_s_w((v4i32) dst0, 2);
+
+      SD(out0, nxt);
+      nxt += 8;
+      SW(out1, nxt);
+      nxt += 4;
+   }
+}
+
+void png_read_filter_row_avg4_msa(png_row_infop row_info, png_bytep row,
+                                  png_const_bytep prev_row)
+{
+   size_t i;
+   png_bytep src = row;
+   png_bytep nxt = row;
+   png_const_bytep pp = prev_row;
+   size_t istop = row_info->rowbytes - 4;
+   int32_t inp0, inp1, out0;
+   v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, dst0, dst1;
+   v16u8 zero = { 0 };
+
+   inp0 = LW(pp);
+   pp += 4;
+   inp1 = LW(src);
+   src += 4;
+   src0 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp0);
+   src1 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp1);
+   src0 = (v16u8) MSA_SRLI_B(src0, 1);
+   src1 += src0;
+   out0 = __msa_copy_s_w((v4i32) src1, 0);
+   SW(out0, nxt);
+   nxt += 4;
+
+   for (i = 0; i < istop; i += 16)
+   {
+      src2 = LD_UB(pp);
+      pp += 16;
+      src6 = LD_UB(src);
+      src += 16;
+
+      SLDI_B2_0_UB(src2, src6, src3, src7, 4);
+      SLDI_B2_0_UB(src2, src6, src4, src8, 8);
+      SLDI_B2_0_UB(src2, src6, src5, src9, 12);
+      src2 = __msa_ave_u_b(src2, src1);
+      src6 += src2;
+      src3 = __msa_ave_u_b(src3, src6);
+      src7 += src3;
+      src4 = __msa_ave_u_b(src4, src7);
+      src8 += src4;
+      src5 = __msa_ave_u_b(src5, src8);
+      src9 += src5;
+      src1 = src9;
+      ILVEV_W2_UB(src6, src7, src8, src9, dst0, dst1);
+      dst0 = (v16u8) __msa_pckev_d((v2i64) dst1, (v2i64) dst0);
+
+      ST_UB(dst0, nxt);
+      nxt += 16;
+   }
+}
+
+void png_read_filter_row_avg3_msa(png_row_infop row_info, png_bytep row,
+                                  png_const_bytep prev_row)
+{
+   size_t i;
+   png_bytep src = row;
+   png_bytep nxt = row;
+   png_const_bytep pp = prev_row;
+   size_t istop = row_info->rowbytes - 3;
+   int64_t out0;
+   int32_t inp0, inp1, out1;
+   int16_t out2;
+   v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, dst0, dst1;
+   v16u8 zero = { 0 };
+   v16i8 mask0 = { 0, 1, 2, 16, 17, 18, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+   v16i8 mask1 = { 0, 1, 2, 3, 4, 5, 16, 17, 18, 19, 20, 21, 0, 0, 0, 0 };
+
+   inp0 = LW(pp);
+   pp += 3;
+   inp1 = LW(src);
+   src += 3;
+   src0 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp0);
+   src1 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp1);
+   src0 = (v16u8) MSA_SRLI_B(src0, 1);
+   src1 += src0;
+   out2 = __msa_copy_s_h((v8i16) src1, 0);
+   SH(out2, nxt);
+   nxt += 2;
+   nxt[0] = src1[2];
+   nxt++;
+
+   for (i = 0; i < istop; i += 12)
+   {
+      src2 = LD_UB(pp);
+      pp += 12;
+      src6 = LD_UB(src);
+      src += 12;
+
+      SLDI_B2_0_UB(src2, src6, src3, src7, 3);
+      SLDI_B2_0_UB(src2, src6, src4, src8, 6);
+      SLDI_B2_0_UB(src2, src6, src5, src9, 9);
+      src2 = __msa_ave_u_b(src2, src1);
+      src6 += src2;
+      src3 = __msa_ave_u_b(src3, src6);
+      src7 += src3;
+      src4 = __msa_ave_u_b(src4, src7);
+      src8 += src4;
+      src5 = __msa_ave_u_b(src5, src8);
+      src9 += src5;
+      src1 = src9;
+      VSHF_B2_UB(src6, src7, src8, src9, mask0, mask0, dst0, dst1);
+      dst0 = (v16u8) __msa_vshf_b(mask1, (v16i8) dst1, (v16i8) dst0);
+      out0 = __msa_copy_s_d((v2i64) dst0, 0);
+      out1 = __msa_copy_s_w((v4i32) dst0, 2);
+
+      SD(out0, nxt);
+      nxt += 8;
+      SW(out1, nxt);
+      nxt += 4;
+   }
+}
+
+void png_read_filter_row_paeth4_msa(png_row_infop row_info,
+                                    png_bytep row,
+                                    png_const_bytep prev_row)
+{
+   int32_t count, rp_end;
+   png_bytep nxt;
+   png_const_bytep prev_nxt;
+   int32_t inp0, inp1, res0;
+   v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9;
+   v16u8 src10, src11, src12, src13, dst0, dst1;
+   v8i16 vec0, vec1, vec2;
+   v16u8 zero = { 0 };
+
+   nxt = row;
+   prev_nxt = prev_row;
+
+   inp0 = LW(nxt);
+   inp1 = LW(prev_nxt);
+   prev_nxt += 4;
+   src0 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp0);
+   src1 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp1);
+
+   src1 += src0;
+   res0 = __msa_copy_s_w((v4i32) src1, 0);
+
+   SW(res0, nxt);
+   nxt += 4;
+
+   /* Remainder */
+   rp_end = row_info->rowbytes - 4;
+
+   for (count = 0; count < rp_end; count += 16)
+   {
+      src2 = LD_UB(prev_nxt);
+      prev_nxt += 16;
+      src6 = LD_UB(prev_row);
+      prev_row += 16;
+      src10 = LD_UB(nxt);
+
+      SLDI_B3_0_UB(src2, src6, src10, src3, src7, src11, 4);
+      SLDI_B3_0_UB(src2, src6, src10, src4, src8, src12, 8);
+      SLDI_B3_0_UB(src2, src6, src10, src5, src9, src13, 12);
+      ILVR_B2_SH(src2, src6, src1, src6, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src1, src2, src6, src10);
+      ILVR_B2_SH(src3, src7, src10, src7, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src10, src3, src7, src11);
+      ILVR_B2_SH(src4, src8, src11, src8, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src11, src4, src8, src12);
+      ILVR_B2_SH(src5, src9, src12, src9, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src12, src5, src9, src13);
+      src1 = src13;
+      ILVEV_W2_UB(src10, src11, src12, src1, dst0, dst1);
+      dst0 = (v16u8) __msa_pckev_d((v2i64) dst1, (v2i64) dst0);
+
+      ST_UB(dst0, nxt);
+      nxt += 16;
+   }
+}
+
+void png_read_filter_row_paeth3_msa(png_row_infop row_info,
+                                    png_bytep row,
+                                    png_const_bytep prev_row)
+{
+   int32_t count, rp_end;
+   png_bytep nxt;
+   png_const_bytep prev_nxt;
+   int64_t out0;
+   int32_t inp0, inp1, out1;
+   int16_t out2;
+   v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, dst0, dst1;
+   v16u8 src10, src11, src12, src13;
+   v8i16 vec0, vec1, vec2;
+   v16u8 zero = { 0 };
+   v16i8 mask0 = { 0, 1, 2, 16, 17, 18, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+   v16i8 mask1 = { 0, 1, 2, 3, 4, 5, 16, 17, 18, 19, 20, 21, 0, 0, 0, 0 };
+
+   nxt = row;
+   prev_nxt = prev_row;
+
+   inp0 = LW(nxt);
+   inp1 = LW(prev_nxt);
+   prev_nxt += 3;
+   src0 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp0);
+   src1 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp1);
+
+   src1 += src0;
+   out2 = __msa_copy_s_h((v8i16) src1, 0);
+
+   SH(out2, nxt);
+   nxt += 2;
+   nxt[0] = src1[2];
+   nxt++;
+
+   /* Remainder */
+   rp_end = row_info->rowbytes - 3;
+
+   for (count = 0; count < rp_end; count += 12)
+   {
+      src2 = LD_UB(prev_nxt);
+      prev_nxt += 12;
+      src6 = LD_UB(prev_row);
+      prev_row += 12;
+      src10 = LD_UB(nxt);
+
+      SLDI_B3_0_UB(src2, src6, src10, src3, src7, src11, 3);
+      SLDI_B3_0_UB(src2, src6, src10, src4, src8, src12, 6);
+      SLDI_B3_0_UB(src2, src6, src10, src5, src9, src13, 9);
+      ILVR_B2_SH(src2, src6, src1, src6, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src1, src2, src6, src10);
+      ILVR_B2_SH(src3, src7, src10, src7, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src10, src3, src7, src11);
+      ILVR_B2_SH(src4, src8, src11, src8, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src11, src4, src8, src12);
+      ILVR_B2_SH(src5, src9, src12, src9, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src12, src5, src9, src13);
+      src1 = src13;
+      VSHF_B2_UB(src10, src11, src12, src13, mask0, mask0, dst0, dst1);
+      dst0 = (v16u8) __msa_vshf_b(mask1, (v16i8) dst1, (v16i8) dst0);
+      out0 = __msa_copy_s_d((v2i64) dst0, 0);
+      out1 = __msa_copy_s_w((v4i32) dst0, 2);
+
+      SD(out0, nxt);
+      nxt += 8;
+      SW(out1, nxt);
+      nxt += 4;
+   }
+}
+
+#endif /* PNG_MIPS_MSA_OPT > 0 */
+#endif /* PNG_MIPS_MSA_IMPLEMENTATION == 1 (intrinsics) */
+#endif /* READ */
@@ -0,0 +1,127 @@
+
+/* mips_init.c - MSA optimised filter functions
+ *
+ * Copyright (c) 2018 Cosmin Truta
+ * Copyright (c) 2016 Glenn Randers-Pehrson
+ * Written by Mandar Sahastrabuddhe, 2016.
+ *
+ * This code is released under the libpng license.
+ * For conditions of distribution and use, see the disclaimer
+ * and license in png.h
+ */
+
+/* Below, after checking __linux__, various non-C90 POSIX 1003.1 functions are
+ * called.
+ */
+#define _POSIX_SOURCE 1
+
+#include <stdio.h>
+#include "../pngpriv.h"
+
+#ifdef PNG_READ_SUPPORTED
+
+#if PNG_MIPS_MSA_OPT > 0
+#ifdef PNG_MIPS_MSA_CHECK_SUPPORTED /* Do run-time checks */
+/* WARNING: it is strongly recommended that you do not build libpng with
+ * run-time checks for CPU features if at all possible.  In the case of the MIPS
+ * MSA instructions there is no processor-specific way of detecting the
+ * presence of the required support, therefore run-time detection is extremely
+ * OS specific.
+ *
+ * You may set the macro PNG_MIPS_MSA_FILE to the file name of file containing
+ * a fragment of C source code which defines the png_have_msa function.  There
+ * are a number of implementations in contrib/mips-msa, but the only one that
+ * has partial support is contrib/mips-msa/linux.c - a generic Linux
+ * implementation which reads /proc/cpufino.
+ */
+#ifndef PNG_MIPS_MSA_FILE
+#  ifdef __linux__
+#     define PNG_MIPS_MSA_FILE "contrib/mips-msa/linux.c"
+#  endif
+#endif
+
+#ifdef PNG_MIPS_MSA_FILE
+
+#include <signal.h> /* for sig_atomic_t */
+static int png_have_msa(png_structp png_ptr);
+#include PNG_MIPS_MSA_FILE
+
+#else  /* PNG_MIPS_MSA_FILE */
+#  error "PNG_MIPS_MSA_FILE undefined: no support for run-time MIPS MSA checks"
+#endif /* PNG_MIPS_MSA_FILE */
+#endif /* PNG_MIPS_MSA_CHECK_SUPPORTED */
+
+#ifndef PNG_ALIGNED_MEMORY_SUPPORTED
+#  error "ALIGNED_MEMORY is required; set: -DPNG_ALIGNED_MEMORY_SUPPORTED"
+#endif
+
+void
+png_init_filter_functions_msa(png_structp pp, unsigned int bpp)
+{
+   /* The switch statement is compiled in for MIPS_MSA_API, the call to
+    * png_have_msa is compiled in for MIPS_MSA_CHECK. If both are defined
+    * the check is only performed if the API has not set the MSA option on
+    * or off explicitly. In this case the check controls what happens.
+    */
+
+#ifdef PNG_MIPS_MSA_API_SUPPORTED
+   switch ((pp->options >> PNG_MIPS_MSA) & 3)
+   {
+      case PNG_OPTION_UNSET:
+         /* Allow the run-time check to execute if it has been enabled -
+          * thus both API and CHECK can be turned on.  If it isn't supported
+          * this case will fall through to the 'default' below, which just
+          * returns.
+          */
+#ifdef PNG_MIPS_MSA_CHECK_SUPPORTED
+         {
+            static volatile sig_atomic_t no_msa = -1; /* not checked */
+
+            if (no_msa < 0)
+               no_msa = !png_have_msa(pp);
+
+            if (no_msa)
+               return;
+         }
+#endif /* PNG_MIPS_MSA_CHECK_SUPPORTED */
+         break;
+
+      default: /* OFF or INVALID */
+         return;
+
+      case PNG_OPTION_ON:
+         /* Option turned on */
+         break;
+   }
+   /* IMPORTANT: any new external functions used here must be declared using
+    * PNG_INTERNAL_FUNCTION in ../pngpriv.h.  This is required so that the
+    * 'prefix' option to configure works:
+    *
+    *    ./configure --with-libpng-prefix=foobar_
+    *
+    * Verify you have got this right by running the above command, doing a build
+    * and examining pngprefix.h; it must contain a #define for every external
+    * function you add.  (Notice that this happens automatically for the
+    * initialization function.)
+    */
+   pp->read_filter[PNG_FILTER_VALUE_UP-1] = png_read_filter_row_up_msa;
+
+   if (bpp == 3)
+   {
+      pp->read_filter[PNG_FILTER_VALUE_SUB-1] = png_read_filter_row_sub3_msa;
+      pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg3_msa;
+      pp->read_filter[PNG_FILTER_VALUE_PAETH-1] = png_read_filter_row_paeth3_msa;
+   }
+   else if (bpp == 4)
+   {
+      pp->read_filter[PNG_FILTER_VALUE_SUB-1] = png_read_filter_row_sub4_msa;
+      pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg4_msa;
+      pp->read_filter[PNG_FILTER_VALUE_PAETH-1] = png_read_filter_row_paeth4_msa;
+   }
+#else
+   (void)pp;
+   (void)bpp;
+#endif /* PNG_MIPS_MSA_API_SUPPORTED */
+}
+#endif /* PNG_MIPS_MSA_OPT > 0 */
+#endif /* READ */
@@ -0,0 +1,17 @@
+diff --git a/3rdparty/libpng/pngrutil.c b/3rdparty/libpng/pngrutil.c
+index d5fa08c397..4db3de990b 100644
+--- a/3rdparty/libpng/pngrutil.c
+++ b/3rdparty/libpng/pngrutil.c
+@@ -2087,10 +2087,8 @@ png_handle_eXIf(png_structrp png_ptr, png_inforp info_ptr, png_uint_32 length)
+       }
+    }
+ 
+-   if (png_crc_finish(png_ptr, 0) != 0)
+-      return;
+-
+-   png_set_eXIf_1(png_ptr, info_ptr, length, info_ptr->eXIf_buf);
+   if (png_crc_finish(png_ptr, 0) == 0)
+      png_set_eXIf_1(png_ptr, info_ptr, length, info_ptr->eXIf_buf);
+ 
+    png_free(png_ptr, info_ptr->eXIf_buf);
+    info_ptr->eXIf_buf = NULL;
@@ -0,0 +1,53 @@
+diff --git a/3rdparty/libpng/mips/mips_init.c b/3rdparty/libpng/mips/mips_init.c
+index 8dd283deef..6a061cccfa 100644
+--- a/3rdparty/libpng/mips/mips_init.c
+++ b/3rdparty/libpng/mips/mips_init.c
+@@ -73,7 +73,6 @@ png_init_filter_functions_msa(png_structp pp, unsigned int bpp)
+           * this case will fall through to the 'default' below, which just
+           * returns.
+           */
+-#endif /* PNG_MIPS_MSA_API_SUPPORTED */
+ #ifdef PNG_MIPS_MSA_CHECK_SUPPORTED
+          {
+             static volatile sig_atomic_t no_msa = -1; /* not checked */
+@@ -84,12 +83,9 @@ png_init_filter_functions_msa(png_structp pp, unsigned int bpp)
+             if (no_msa)
+                return;
+          }
+-#ifdef PNG_MIPS_MSA_API_SUPPORTED
+-         break;
+-#endif
+ #endif /* PNG_MIPS_MSA_CHECK_SUPPORTED */
+         break;
+ 
+-#ifdef PNG_MIPS_MSA_API_SUPPORTED
+       default: /* OFF or INVALID */
+          return;
+ 
+@@ -97,8 +93,6 @@ png_init_filter_functions_msa(png_structp pp, unsigned int bpp)
+          /* Option turned on */
+          break;
+    }
+-#endif
+-
+    /* IMPORTANT: any new external functions used here must be declared using
+     * PNG_INTERNAL_FUNCTION in ../pngpriv.h.  This is required so that the
+     * 'prefix' option to configure works:
+@@ -118,13 +112,16 @@ png_init_filter_functions_msa(png_structp pp, unsigned int bpp)
+       pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg3_msa;
+       pp->read_filter[PNG_FILTER_VALUE_PAETH-1] = png_read_filter_row_paeth3_msa;
+    }
+-
+    else if (bpp == 4)
+    {
+       pp->read_filter[PNG_FILTER_VALUE_SUB-1] = png_read_filter_row_sub4_msa;
+       pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg4_msa;
+       pp->read_filter[PNG_FILTER_VALUE_PAETH-1] = png_read_filter_row_paeth4_msa;
+    }
+#else
+   (void)pp;
+   (void)bpp;
+#endif /* PNG_MIPS_MSA_API_SUPPORTED */
+ }
+ #endif /* PNG_MIPS_MSA_OPT > 0 */
+ #endif /* READ */
@@ -1,7 +1,7 @@

 /* png.c - location for general purpose libpng functions
 *
- * Copyright (c) 2018 Cosmin Truta
+ * Copyright (c) 2018-2019 Cosmin Truta
 * Copyright (c) 1998-2002,2004,2006-2018 Glenn Randers-Pehrson
 * Copyright (c) 1996-1997 Andreas Dilger
 * Copyright (c) 1995-1996 Guy Eric Schalnat, Group 42, Inc.
@@ -14,7 +14,7 @@
 #include "pngpriv.h"

 /* Generate a compiler error if there is an old png.h in the search path. */
-typedef png_libpng_version_1_6_36 Your_png_h_is_not_version_1_6_36;
+typedef png_libpng_version_1_6_37 Your_png_h_is_not_version_1_6_37;

 #ifdef __GNUC__
 /* The version tests may need to be added to, but the problem warning has
@@ -815,8 +815,8 @@ png_get_copyright(png_const_structrp png_ptr)
   return PNG_STRING_COPYRIGHT
 #else
   return PNG_STRING_NEWLINE \
-      "libpng version 1.6.36" PNG_STRING_NEWLINE \
-      "Copyright (c) 2018 Cosmin Truta" PNG_STRING_NEWLINE \
+      "libpng version 1.6.37" PNG_STRING_NEWLINE \
+      "Copyright (c) 2018-2019 Cosmin Truta" PNG_STRING_NEWLINE \
      "Copyright (c) 1998-2002,2004,2006-2018 Glenn Randers-Pehrson" \
      PNG_STRING_NEWLINE \
      "Copyright (c) 1996-1997 Andreas Dilger" PNG_STRING_NEWLINE \
@@ -4588,8 +4588,7 @@ png_image_free(png_imagep image)
   if (image != NULL && image->opaque != NULL &&
      image->opaque->error_buf == NULL)
   {
-      /* Ignore errors here: */
-      (void)png_safe_execute(image, png_image_free_function, image);
+      png_image_free_function(image);
      image->opaque = NULL;
   }
 }
@@ -1,9 +1,9 @@

 /* png.h - header file for PNG reference library
 *
- * libpng version 1.6.36 - December 1, 2018
+ * libpng version 1.6.37 - April 14, 2019
 *
- * Copyright (c) 2018 Cosmin Truta
+ * Copyright (c) 2018-2019 Cosmin Truta
 * Copyright (c) 1998-2002,2004,2006-2018 Glenn Randers-Pehrson
 * Copyright (c) 1996-1997 Andreas Dilger
 * Copyright (c) 1995-1996 Guy Eric Schalnat, Group 42, Inc.
@@ -14,8 +14,9 @@
 *   libpng versions 0.71, May 1995, through 0.88, January 1996: Guy Schalnat
 *   libpng versions 0.89, June 1996, through 0.96, May 1997: Andreas Dilger
 *   libpng versions 0.97, January 1998, through 1.6.35, July 2018:
- *     Glenn Randers-Pehrson.
- *   libpng version 1.6.36, December 1, 2018: Cosmin Truta
+ *     Glenn Randers-Pehrson
+ *   libpng versions 1.6.36, December 2018, through 1.6.37, April 2019:
+ *     Cosmin Truta
 *   See also "Contributing Authors", below.
 */

@@ -26,8 +27,8 @@
 * PNG Reference Library License version 2
 * ---------------------------------------
 *
- *  * Copyright (c) 1995-2018 The PNG Reference Library Authors.
- *  * Copyright (c) 2018 Cosmin Truta.
+ *  * Copyright (c) 1995-2019 The PNG Reference Library Authors.
+ *  * Copyright (c) 2018-2019 Cosmin Truta.
 *  * Copyright (c) 2000-2002, 2004, 2006-2018 Glenn Randers-Pehrson.
 *  * Copyright (c) 1996-1997 Andreas Dilger.
 *  * Copyright (c) 1995-1996 Guy Eric Schalnat, Group 42, Inc.
@@ -35,7 +36,7 @@
 * The software is supplied "as is", without warranty of any kind,
 * express or implied, including, without limitation, the warranties
 * of merchantability, fitness for a particular purpose, title, and
- * non-infringement.  In no even shall the Copyright owners, or
+ * non-infringement.  In no event shall the Copyright owners, or
 * anyone distributing the software, be liable for any damages or
 * other liability, whether in contract, tort or otherwise, arising
 * from, out of, or in connection with the software, or the use or
@@ -61,7 +62,7 @@
 * PNG Reference Library License version 1 (for libpng 0.5 through 1.6.35)
 * -----------------------------------------------------------------------
 *
- * libpng versions 1.0.7, July 1, 2000 through 1.6.35, July 15, 2018 are
+ * libpng versions 1.0.7, July 1, 2000, through 1.6.35, July 15, 2018 are
 * Copyright (c) 2000-2002, 2004, 2006-2018 Glenn Randers-Pehrson, are
 * derived from libpng-1.0.6, and are distributed according to the same
 * disclaimer and license as libpng-1.0.6 with the following individuals
@@ -238,7 +239,7 @@
 *    ...
 *    1.5.30                  15    10530  15.so.15.30[.0]
 *    ...
- *    1.6.36                  16    10636  16.so.16.36[.0]
+ *    1.6.37                  16    10637  16.so.16.37[.0]
 *
 *    Henceforth the source version will match the shared-library major and
 *    minor numbers; the shared-library major version number will be used for
@@ -277,8 +278,8 @@
 */

 /* Version information for png.h - this should match the version in png.c */
-#define PNG_LIBPNG_VER_STRING "1.6.36"
-#define PNG_HEADER_VERSION_STRING " libpng version 1.6.36 - December 1, 2018\n"
+#define PNG_LIBPNG_VER_STRING "1.6.37"
+#define PNG_HEADER_VERSION_STRING " libpng version 1.6.37 - April 14, 2019\n"

 #define PNG_LIBPNG_VER_SONUM   16
 #define PNG_LIBPNG_VER_DLLNUM  16
@@ -286,12 +287,11 @@
 /* These should match the first 3 components of PNG_LIBPNG_VER_STRING: */
 #define PNG_LIBPNG_VER_MAJOR   1
 #define PNG_LIBPNG_VER_MINOR   6
-#define PNG_LIBPNG_VER_RELEASE 36
+#define PNG_LIBPNG_VER_RELEASE 37

-/* This should match the numeric part of the final component of
- * PNG_LIBPNG_VER_STRING, omitting any leading zero:
+/* This should be zero for a public release, or non-zero for a
+ * development version.  [Deprecated]
 */
-
 #define PNG_LIBPNG_VER_BUILD  0

 /* Release Status */
@@ -318,7 +318,7 @@
 * From version 1.0.1 it is:
 * XXYYZZ, where XX=major, YY=minor, ZZ=release
 */
-#define PNG_LIBPNG_VER 10636 /* 1.6.36 */
+#define PNG_LIBPNG_VER 10637 /* 1.6.37 */

 /* Library configuration: these options cannot be changed after
 * the library has been built.
@@ -428,7 +428,7 @@ extern "C" {
 /* This triggers a compiler error in png.c, if png.c and png.h
 * do not agree upon the version number.
 */
-typedef char* png_libpng_version_1_6_36;
+typedef char* png_libpng_version_1_6_37;

 /* Basic control structions.  Read libpng-manual.txt or libpng.3 for more info.
 *
@@ -1,9 +1,9 @@

-/* pngconf.h - machine configurable file for libpng
+/* pngconf.h - machine-configurable file for libpng
 *
- * libpng version 1.6.36
+ * libpng version 1.6.37
 *
- * Copyright (c) 2018 Cosmin Truta
+ * Copyright (c) 2018-2019 Cosmin Truta
 * Copyright (c) 1998-2002,2004,2006-2016,2018 Glenn Randers-Pehrson
 * Copyright (c) 1996-1997 Andreas Dilger
 * Copyright (c) 1995-1996 Guy Eric Schalnat, Group 42, Inc.
@@ -1,8 +1,8 @@
 /* pnglibconf.h - library build configuration */

-/* libpng version 1.6.36 */
+/* libpng version 1.6.37 */

-/* Copyright (c) 2018 Cosmin Truta */
+/* Copyright (c) 2018-2019 Cosmin Truta */
 /* Copyright (c) 1998-2002,2004,2006-2018 Glenn Randers-Pehrson */

 /* This code is released under the libpng license. */
@@ -1,7 +1,7 @@

 /* pngpriv.h - private declarations for use inside libpng
 *
- * Copyright (c) 2018 Cosmin Truta
+ * Copyright (c) 2018-2019 Cosmin Truta
 * Copyright (c) 1998-2002,2004,2006-2018 Glenn Randers-Pehrson
 * Copyright (c) 1996-1997 Andreas Dilger
 * Copyright (c) 1995-1996 Guy Eric Schalnat, Group 42, Inc.
@@ -2119,11 +2119,11 @@ PNG_INTERNAL_FUNCTION(png_uint_32, png_check_keyword, (png_structrp png_ptr,

 #if PNG_ARM_NEON_IMPLEMENTATION == 1
 PNG_INTERNAL_FUNCTION(void,
-                      png_riffle_palette_rgba,
-                      (png_structrp, png_row_infop),
+                      png_riffle_palette_neon,
+                      (png_structrp),
                      PNG_EMPTY);
 PNG_INTERNAL_FUNCTION(int,
-                      png_do_expand_palette_neon_rgba,
+                      png_do_expand_palette_rgba8_neon,
                      (png_structrp,
                       png_row_infop,
                       png_const_bytep,
@@ -2131,7 +2131,7 @@ PNG_INTERNAL_FUNCTION(int,
                       const png_bytepp),
                      PNG_EMPTY);
 PNG_INTERNAL_FUNCTION(int,
-                      png_do_expand_palette_neon_rgb,
+                      png_do_expand_palette_rgb8_neon,
                      (png_structrp,
                       png_row_infop,
                       png_const_bytep,
@@ -1,7 +1,7 @@

 /* pngread.c - read a PNG file
 *
- * Copyright (c) 2018 Cosmin Truta
+ * Copyright (c) 2018-2019 Cosmin Truta
 * Copyright (c) 1998-2002,2004,2006-2018 Glenn Randers-Pehrson
 * Copyright (c) 1996-1997 Andreas Dilger
 * Copyright (c) 1995-1996 Guy Eric Schalnat, Group 42, Inc.
@@ -994,6 +994,12 @@ png_read_destroy(png_structrp png_ptr)
   png_ptr->chunk_list = NULL;
 #endif

+#if defined(PNG_READ_EXPAND_SUPPORTED) && \
+    defined(PNG_ARM_NEON_IMPLEMENTATION)
+   png_free(png_ptr, png_ptr->riffled_palette);
+   png_ptr->riffled_palette = NULL;
+#endif
+
   /* NOTE: the 'setjmp' buffer may still be allocated and the memory and error
    * callbacks are still set at this point.  They are required to complete the
    * destruction of the png_struct itself.
@@ -1,7 +1,7 @@

 /* pngrtran.c - transforms the data in a row for PNG readers
 *
- * Copyright (c) 2018 Cosmin Truta
+ * Copyright (c) 2018-2019 Cosmin Truta
 * Copyright (c) 1998-2002,2004,2006-2018 Glenn Randers-Pehrson
 * Copyright (c) 1996-1997 Andreas Dilger
 * Copyright (c) 1995-1996 Guy Eric Schalnat, Group 42, Inc.
@@ -1182,20 +1182,20 @@ png_init_palette_transformations(png_structrp png_ptr)
             png_ptr->palette[png_ptr->background.index].blue;

 #ifdef PNG_READ_INVERT_ALPHA_SUPPORTED
-        if ((png_ptr->transformations & PNG_INVERT_ALPHA) != 0)
-        {
-           if ((png_ptr->transformations & PNG_EXPAND_tRNS) == 0)
-           {
-              /* Invert the alpha channel (in tRNS) unless the pixels are
-               * going to be expanded, in which case leave it for later
-               */
-              int i, istop = png_ptr->num_trans;
+         if ((png_ptr->transformations & PNG_INVERT_ALPHA) != 0)
+         {
+            if ((png_ptr->transformations & PNG_EXPAND_tRNS) == 0)
+            {
+               /* Invert the alpha channel (in tRNS) unless the pixels are
+                * going to be expanded, in which case leave it for later
+                */
+               int i, istop = png_ptr->num_trans;

-              for (i=0; i<istop; i++)
-                 png_ptr->trans_alpha[i] = (png_byte)(255 -
-                    png_ptr->trans_alpha[i]);
-           }
-        }
+               for (i = 0; i < istop; i++)
+                  png_ptr->trans_alpha[i] =
+                      (png_byte)(255 - png_ptr->trans_alpha[i]);
+            }
+         }
 #endif /* READ_INVERT_ALPHA */
      }
   } /* background expand and (therefore) no alpha association. */
@@ -4320,9 +4320,11 @@ png_do_expand_palette(png_structrp png_ptr, png_row_infop row_info,
                   * but sometimes row_info->bit_depth has been changed to 8.
                   * In these cases, the palette hasn't been riffled.
                   */
-                  i = png_do_expand_palette_neon_rgba(png_ptr, row_info, row,
+                  i = png_do_expand_palette_rgba8_neon(png_ptr, row_info, row,
                      &sp, &dp);
               }
+#else
+               PNG_UNUSED(png_ptr)
 #endif

               for (; i < row_width; i++)
@@ -4349,8 +4351,10 @@ png_do_expand_palette(png_structrp png_ptr, png_row_infop row_info,
               dp = row + (size_t)(row_width * 3) - 1;
               i = 0;
 #ifdef PNG_ARM_NEON_INTRINSICS_AVAILABLE
-               i = png_do_expand_palette_neon_rgb(png_ptr, row_info, row,
+               i = png_do_expand_palette_rgb8_neon(png_ptr, row_info, row,
                   &sp, &dp);
+#else
+               PNG_UNUSED(png_ptr)
 #endif

               for (; i < row_width; i++)
@@ -4770,19 +4774,17 @@ png_do_read_transformations(png_structrp png_ptr, png_row_infop row_info)
 #ifdef PNG_ARM_NEON_INTRINSICS_AVAILABLE
         if ((png_ptr->num_trans > 0) && (png_ptr->bit_depth == 8))
         {
-            /* Allocate space for the decompressed full palette. */
            if (png_ptr->riffled_palette == NULL)
            {
-               png_ptr->riffled_palette = png_malloc(png_ptr, 256*4);
-               if (png_ptr->riffled_palette == NULL)
-                  png_error(png_ptr, "NULL row buffer");
-               /* Build the RGBA palette. */
-               png_riffle_palette_rgba(png_ptr, row_info);
+               /* Initialize the accelerated palette expansion. */
+               png_ptr->riffled_palette =
+                   (png_bytep)png_malloc(png_ptr, 256 * 4);
+               png_riffle_palette_neon(png_ptr);
            }
         }
 #endif
         png_do_expand_palette(png_ptr, row_info, png_ptr->row_buf + 1,
-            png_ptr->palette, png_ptr->trans_alpha, png_ptr->num_trans);
+             png_ptr->palette, png_ptr->trans_alpha, png_ptr->num_trans);
      }

      else
@@ -2087,10 +2087,8 @@ png_handle_eXIf(png_structrp png_ptr, png_inforp info_ptr, png_uint_32 length)
      }
   }

-   if (png_crc_finish(png_ptr, 0) != 0)
-      return;
-
-   png_set_eXIf_1(png_ptr, info_ptr, length, info_ptr->eXIf_buf);
+   if (png_crc_finish(png_ptr, 0) == 0)
+      png_set_eXIf_1(png_ptr, info_ptr, length, info_ptr->eXIf_buf);

   png_free(png_ptr, info_ptr->eXIf_buf);
   info_ptr->eXIf_buf = NULL;
@@ -1,7 +1,7 @@

 /* pngstruct.h - header file for PNG reference library
 *
- * Copyright (c) 2018 Cosmin Truta
+ * Copyright (c) 2018-2019 Cosmin Truta
 * Copyright (c) 1998-2002,2004,2006-2018 Glenn Randers-Pehrson
 * Copyright (c) 1996-1997 Andreas Dilger
 * Copyright (c) 1995-1996 Guy Eric Schalnat, Group 42, Inc.
@@ -228,10 +228,6 @@ struct png_struct_def
                               * big_row_buf; while writing it is separately
                               * allocated.
                               */
-#ifdef PNG_READ_EXPAND_SUPPORTED
-   /* Buffer to accelerate palette transformations. */
-   png_bytep riffled_palette;
-#endif
 #ifdef PNG_WRITE_FILTER_SUPPORTED
   png_bytep try_row;    /* buffer to save trial row when filtering */
   png_bytep tst_row;    /* buffer to save best trial row when filtering */
@@ -396,6 +392,12 @@ struct png_struct_def
   /* deleted in 1.5.5: rgb_to_gray_blue_coeff; */
 #endif

+/* New member added in libpng-1.6.36 */
+#if defined(PNG_READ_EXPAND_SUPPORTED) && \
+    defined(PNG_ARM_NEON_IMPLEMENTATION)
+   png_bytep riffled_palette; /* buffer for accelerated palette expansion */
+#endif
+
 /* New member added in libpng-1.0.4 (renamed in 1.0.9) */
 #if defined(PNG_MNG_FEATURES_SUPPORTED)
 /* Changed from png_byte to png_uint_32 at version 1.2.0 */
@@ -1,7 +1,7 @@

 /* pngwrite.c - general routines to write a PNG file
 *
- * Copyright (c) 2018 Cosmin Truta
+ * Copyright (c) 2018-2019 Cosmin Truta
 * Copyright (c) 1998-2002,2004,2006-2018 Glenn Randers-Pehrson
 * Copyright (c) 1996-1997 Andreas Dilger
 * Copyright (c) 1995-1996 Guy Eric Schalnat, Group 42, Inc.
@@ -948,10 +948,6 @@ png_write_destroy(png_structrp png_ptr)
   png_free_buffer_list(png_ptr, &png_ptr->zbuffer_list);
   png_free(png_ptr, png_ptr->row_buf);
   png_ptr->row_buf = NULL;
-#ifdef PNG_READ_EXPANDED_SUPPORTED
-   png_free(png_ptr, png_ptr->riffled_palette);
-   png_ptr->riffled_palette = NULL;
-#endif
 #ifdef PNG_WRITE_FILTER_SUPPORTED
   png_free(png_ptr, png_ptr->prev_row);
   png_free(png_ptr, png_ptr->try_row);
@@ -0,0 +1,91 @@
+# COPYRIGHT
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# License); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# AS IS BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+#
+# Copyright (c) 2020, OPEN AI LAB
+# Author: qtang@openailab.com or https://github.com/BUG1989
+#         qli@openailab.com
+#         sqfu@openailab.com
+#
+
+SET(TENGINE_COMMIT_VERSION "2f3cd86217f3530c8e4a82f3ed5af14c7a4e3943")
+SET(OCV_TENGINE_DIR "${OpenCV_BINARY_DIR}/3rdparty/libtengine")
+SET(OCV_TENGINE_SOURCE_PATH "${OCV_TENGINE_DIR}/Tengine-${TENGINE_COMMIT_VERSION}")
+
+IF(EXISTS "${OCV_TENGINE_SOURCE_PATH}")
+	MESSAGE(STATUS "Tengine is exist already at: ${OCV_TENGINE_SOURCE_PATH}")
+
+	SET(Tengine_FOUND ON)
+	SET(BUILD_TENGINE ON)
+ELSE()
+	SET(OCV_TENGINE_FILENAME "${TENGINE_COMMIT_VERSION}.zip")#name2
+	SET(OCV_TENGINE_URL "https://github.com/OAID/Tengine/archive/") #url2
+	SET(tengine_md5sum 9124324b6e2b350012e46ae1db4bad7d) #md5sum2
+
+	#MESSAGE(STATUS "**** TENGINE DOWNLOAD BEGIN ****")
+	ocv_download(FILENAME ${OCV_TENGINE_FILENAME}
+						HASH ${tengine_md5sum}
+						URL
+						"${OPENCV_TENGINE_URL}"
+						"$ENV{OPENCV_TENGINE_URL}"
+						"${OCV_TENGINE_URL}"
+						DESTINATION_DIR "${OCV_TENGINE_DIR}"
+						ID TENGINE
+						STATUS res
+						UNPACK RELATIVE_URL)
+
+	if (NOT res)
+		MESSAGE(STATUS "TENGINE DOWNLOAD FAILED. Turning Tengine_FOUND off.")
+		SET(Tengine_FOUND OFF)
+	else ()
+		MESSAGE(STATUS "TENGINE DOWNLOAD success . ")
+
+		SET(Tengine_FOUND ON)
+		SET(BUILD_TENGINE ON)
+	endif()
+ENDIF()
+
+if(BUILD_TENGINE)
+	SET(HAVE_TENGINE 1)
+
+	# android system
+	if(ANDROID)
+	   if(${ANDROID_ABI} STREQUAL "armeabi-v7a")
+			   SET(CONFIG_ARCH_ARM32 ON)
+	   elseif(${ANDROID_ABI} STREQUAL "arm64-v8a")
+			   SET(CONFIG_ARCH_ARM64 ON)
+	   endif()
+	   SET(Tengine_LIB "tengine" CACHE INTERNAL "")
+	else()
+		# linux system
+		if(CMAKE_SYSTEM_PROCESSOR STREQUAL arm)
+			   SET(CONFIG_ARCH_ARM32 ON)
+		elseif(CMAKE_SYSTEM_PROCESSOR STREQUAL aarch64) ## AARCH64
+			   SET(CONFIG_ARCH_ARM64 ON)
+		endif()
+		SET(Tengine_LIB "tengine" CACHE INTERNAL "")
+	endif()
+
+	SET(BUILT_IN_OPENCV ON) ## set for tengine compile discern .
+	SET(Tengine_INCLUDE_DIR  "${OCV_TENGINE_SOURCE_PATH}/core/include" CACHE INTERNAL "")
+	if(EXISTS "${OCV_TENGINE_SOURCE_PATH}/CMakeLists.txt")
+		add_subdirectory("${OCV_TENGINE_SOURCE_PATH}" "${OCV_TENGINE_DIR}/build")
+	else()
+		message(WARNING "TENGINE: Missing 'CMakeLists.txt' in source code package: ${OCV_TENGINE_SOURCE_PATH}")
+		SET(HAVE_TENGINE 1)
+	endif()
+endif()
@@ -126,30 +126,30 @@ endif()
 if(SIZEOF_UNSIGNED_INT EQUAL SIZEOF_SIZE_T)
  set(TIFF_SIZE_T "unsigned int")
  set(TIFF_SIZE_FORMAT "%u")
+  set(TIFF_SSIZE_T "signed int")
+  set(TIFF_SSIZE_FORMAT "%d")
 elseif(SIZEOF_UNSIGNED_LONG EQUAL SIZEOF_SIZE_T)
  set(TIFF_SIZE_T "unsigned long")
  set(TIFF_SIZE_FORMAT "%lu")
+  set(TIFF_SSIZE_T "signed long")
+  set(TIFF_SSIZE_FORMAT "%ld")
 elseif(SIZEOF_UNSIGNED_LONG_LONG EQUAL SIZEOF_SIZE_T)
  set(TIFF_SIZE_T "unsigned long")
  if(MINGW)
    set(TIFF_SIZE_FORMAT "%I64u")
+    set(TIFF_SSIZE_FORMAT "%I64d")
  else()
    set(TIFF_SIZE_FORMAT "%llu")
+    set(TIFF_SSIZE_FORMAT "%lld")
  endif()
 endif()

 if(SIZEOF_SIGNED_INT EQUAL SIZEOF_UNSIGNED_CHAR_P)
-  set(TIFF_SSIZE_T "signed int")
-  set(TIFF_SSIZE_FORMAT "%d")
 elseif(SIZEOF_SIGNED_LONG EQUAL SIZEOF_UNSIGNED_CHAR_P)
-  set(TIFF_SSIZE_T "signed long")
-  set(TIFF_SSIZE_FORMAT "%ld")
 elseif(SIZEOF_SIGNED_LONG_LONG EQUAL SIZEOF_UNSIGNED_CHAR_P)
  set(TIFF_SSIZE_T "signed long long")
  if(MINGW)
-    set(TIFF_SSIZE_FORMAT "%I64d")
  else()
-    set(TIFF_SSIZE_FORMAT "%lld")
  endif()
 endif()

@@ -216,6 +216,8 @@ endif()
 set(fillorder FILLORDER_MSB2LSB)
 if(CMAKE_HOST_SYSTEM_PROCESSOR MATCHES "i.*86.*" OR
   CMAKE_HOST_SYSTEM_PROCESSOR MATCHES "amd64.*" OR
+   # AMD64 on Windows
+   CMAKE_HOST_SYSTEM_PROCESSOR MATCHES "AMD64" OR
   CMAKE_HOST_SYSTEM_PROCESSOR MATCHES "x86_64.*")
  set(fillorder FILLORDER_LSB2MSB)
 endif()
@@ -16,7 +16,11 @@ int _TIFF_vsnprintf_f(char* str, size_t size, const char* format, va_list ap)
  int count = -1;

  if (size != 0)
+#if _MSC_VER <=	1310
+    count = _vsnprintf(str, size, format, ap);
+#else
    count = _vsnprintf_s(str, size, _TRUNCATE, format, ap);
+#endif
  if (count == -1)
    count = _vscprintf(format, ap);

@@ -30,6 +30,7 @@
 #include "tiffiop.h"
 #include "tif_predict.h"
 #include <math.h>
+#include <float.h>

 uint32
 _TIFFMultiply32(TIFF* tif, uint32 first, uint32 second, const char* where)
@@ -357,6 +358,15 @@ _TIFFUInt64ToDouble(uint64 ui64)
 	}
 }

+float _TIFFClampDoubleToFloat( double val )
+{
+    if( val > FLT_MAX )
+        return FLT_MAX;
+    if( val < -FLT_MAX )
+        return -FLT_MAX;
+    return (float)val;
+}
+
 int _TIFFSeekOK(TIFF* tif, toff_t off)
 {
    /* Huge offsets, especially -1 / UINT64_MAX, can cause issues */
@@ -216,7 +216,7 @@
 #endif

 /* Number of bits in a file offset, on hosts where this is settable. */
-//disabled for OpenCV CMakeLists.txt: #define _FILE_OFFSET_BITS @FILE_OFFSET_BITS@
+#define _FILE_OFFSET_BITS @FILE_OFFSET_BITS@

 /* Define to `__inline__' or `__inline' if that's what the C compiler
   calls it, or to nothing if 'inline' is not supported under any name.  */
@@ -29,7 +29,6 @@
 * (and also some miscellaneous stuff)
 */
 #include "tiffiop.h"
-#include <float.h>

 /*
 * These are used in the backwards compatibility code...
@@ -88,13 +87,15 @@ setDoubleArrayOneValue(double** vpp, double value, size_t nmemb)
 * Install extra samples information.
 */
 static int
-setExtraSamples(TIFFDirectory* td, va_list ap, uint32* v)
+setExtraSamples(TIFF* tif, va_list ap, uint32* v)
 {
 /* XXX: Unassociated alpha data == 999 is a known Corel Draw bug, see below */
 #define EXTRASAMPLE_COREL_UNASSALPHA 999 

 	uint16* va;
 	uint32 i;
+        TIFFDirectory* td = &tif->tif_dir;
+        static const char module[] = "setExtraSamples";

 	*v = (uint16) va_arg(ap, uint16_vap);
 	if ((uint16) *v > td->td_samplesperpixel)
@@ -116,6 +117,18 @@ setExtraSamples(TIFFDirectory* td, va_list ap, uint32* v)
 				return 0;
 		}
 	}
+
+        if ( td->td_transferfunction[0] != NULL && (td->td_samplesperpixel - *v > 1) &&
+                !(td->td_samplesperpixel - td->td_extrasamples > 1))
+        {
+                TIFFWarningExt(tif->tif_clientdata,module,
+                    "ExtraSamples tag value is changing, "
+                    "but TransferFunction was read with a different value. Cancelling it");
+                TIFFClrFieldBit(tif,FIELD_TRANSFERFUNCTION);
+                _TIFFfree(td->td_transferfunction[0]);
+                td->td_transferfunction[0] = NULL;
+        }
+
 	td->td_extrasamples = (uint16) *v;
 	_TIFFsetShortArray(&td->td_sampleinfo, va, td->td_extrasamples);
 	return 1;
@@ -153,15 +166,6 @@ bad:
 	return (0);
 }

-static float TIFFClampDoubleToFloat( double val )
-{
-    if( val > FLT_MAX )
-        return FLT_MAX;
-    if( val < -FLT_MAX )
-        return -FLT_MAX;
-    return (float)val;
-}
-
 static int
 _TIFFVSetField(TIFF* tif, uint32 tag, va_list ap)
 {
@@ -285,6 +289,18 @@ _TIFFVSetField(TIFF* tif, uint32 tag, va_list ap)
                _TIFFfree(td->td_smaxsamplevalue);
                td->td_smaxsamplevalue = NULL;
            }
+            /* Test if 3 transfer functions instead of just one are now needed
+               See http://bugzilla.maptools.org/show_bug.cgi?id=2820 */
+            if( td->td_transferfunction[0] != NULL && (v - td->td_extrasamples > 1) &&
+                !(td->td_samplesperpixel - td->td_extrasamples > 1))
+            {
+                    TIFFWarningExt(tif->tif_clientdata,module,
+                        "SamplesPerPixel tag value is changing, "
+                        "but TransferFunction was read with a different value. Cancelling it");
+                    TIFFClrFieldBit(tif,FIELD_TRANSFERFUNCTION);
+                    _TIFFfree(td->td_transferfunction[0]);
+                    td->td_transferfunction[0] = NULL;
+            }
        }
 		td->td_samplesperpixel = (uint16) v;
 		break;
@@ -320,13 +336,13 @@ _TIFFVSetField(TIFF* tif, uint32 tag, va_list ap)
        dblval = va_arg(ap, double);
        if( dblval < 0 )
            goto badvaluedouble;
-		td->td_xresolution = TIFFClampDoubleToFloat( dblval );
+		td->td_xresolution = _TIFFClampDoubleToFloat( dblval );
 		break;
 	case TIFFTAG_YRESOLUTION:
        dblval = va_arg(ap, double);
        if( dblval < 0 )
            goto badvaluedouble;
-		td->td_yresolution = TIFFClampDoubleToFloat( dblval );
+		td->td_yresolution = _TIFFClampDoubleToFloat( dblval );
 		break;
 	case TIFFTAG_PLANARCONFIG:
 		v = (uint16) va_arg(ap, uint16_vap);
@@ -335,10 +351,10 @@ _TIFFVSetField(TIFF* tif, uint32 tag, va_list ap)
 		td->td_planarconfig = (uint16) v;
 		break;
 	case TIFFTAG_XPOSITION:
-		td->td_xposition = TIFFClampDoubleToFloat( va_arg(ap, double) );
+		td->td_xposition = _TIFFClampDoubleToFloat( va_arg(ap, double) );
 		break;
 	case TIFFTAG_YPOSITION:
-		td->td_yposition = TIFFClampDoubleToFloat( va_arg(ap, double) );
+		td->td_yposition = _TIFFClampDoubleToFloat( va_arg(ap, double) );
 		break;
 	case TIFFTAG_RESOLUTIONUNIT:
 		v = (uint16) va_arg(ap, uint16_vap);
@@ -361,7 +377,7 @@ _TIFFVSetField(TIFF* tif, uint32 tag, va_list ap)
 		_TIFFsetShortArray(&td->td_colormap[2], va_arg(ap, uint16*), v32);
 		break;
 	case TIFFTAG_EXTRASAMPLES:
-		if (!setExtraSamples(td, ap, &v))
+		if (!setExtraSamples(tif, ap, &v))
 			goto badvalue;
 		break;
 	case TIFFTAG_MATTEING:
@@ -684,7 +700,7 @@ _TIFFVSetField(TIFF* tif, uint32 tag, va_list ap)
 				case TIFF_SRATIONAL:
 				case TIFF_FLOAT:
 					{
-						float v2 = TIFFClampDoubleToFloat(va_arg(ap, double));
+						float v2 = _TIFFClampDoubleToFloat(va_arg(ap, double));
 						_TIFFmemcpy(val, &v2, tv_size);
 					}
 					break;
@@ -164,6 +164,7 @@ static int TIFFFetchNormalTag(TIFF*, TIFFDirEntry*, int recover);
 static int TIFFFetchStripThing(TIFF* tif, TIFFDirEntry* dir, uint32 nstrips, uint64** lpp);
 static int TIFFFetchSubjectDistance(TIFF*, TIFFDirEntry*);
 static void ChopUpSingleUncompressedStrip(TIFF*);
+static void TryChopUpUncompressedBigTiff(TIFF*);
 static uint64 TIFFReadUInt64(const uint8 *value);
 static int _TIFFGetMaxColorChannels(uint16 photometric);

@@ -4246,6 +4247,19 @@ TIFFReadDirectory(TIFF* tif)
 		ChopUpSingleUncompressedStrip(tif);
    }

+        /* There are also uncompressed stripped files with strips larger than */
+        /* 2 GB, which make them unfriendly with a lot of code. If possible, */
+        /* try to expose smaller "virtual" strips. */
+        if( tif->tif_dir.td_planarconfig == PLANARCONFIG_CONTIG &&
+            tif->tif_dir.td_compression == COMPRESSION_NONE &&
+            (tif->tif_flags&(TIFF_STRIPCHOP|TIFF_ISTILED)) == TIFF_STRIPCHOP &&
+            TIFFStripSize64(tif) > 0x7FFFFFFFUL )
+        {
+            if ( !_TIFFFillStriles(tif) || !tif->tif_dir.td_stripbytecount )
+                return 0;
+            TryChopUpUncompressedBigTiff(tif);
+        }
+
        /*
         * Clear the dirty directory flag. 
         */
@@ -5699,6 +5713,63 @@ TIFFFetchSubjectDistance(TIFF* tif, TIFFDirEntry* dir)
 	}
 }

+static void allocChoppedUpStripArrays(TIFF* tif, uint32 nstrips,
+                                      uint64 stripbytes, uint32 rowsperstrip)
+{
+    TIFFDirectory *td = &tif->tif_dir;
+    uint64 bytecount;
+    uint64 offset;
+    uint32 i;
+    uint64 *newcounts;
+    uint64 *newoffsets;
+
+    newcounts = (uint64*) _TIFFCheckMalloc(tif, nstrips, sizeof (uint64),
+                            "for chopped \"StripByteCounts\" array");
+    newoffsets = (uint64*) _TIFFCheckMalloc(tif, nstrips, sizeof (uint64),
+                            "for chopped \"StripOffsets\" array");
+    if (newcounts == NULL || newoffsets == NULL) {
+        /*
+        * Unable to allocate new strip information, give up and use
+        * the original one strip information.
+        */
+        if (newcounts != NULL)
+            _TIFFfree(newcounts);
+        if (newoffsets != NULL)
+            _TIFFfree(newoffsets);
+        return;
+    }
+
+    /*
+     * Fill the strip information arrays with new bytecounts and offsets
+     * that reflect the broken-up format.
+     */
+    offset = td->td_stripoffset[0];
+    bytecount = td->td_stripoffset[td->td_nstrips-1] +
+                td->td_stripbytecount[td->td_nstrips-1] - offset;
+    for (i = 0; i < nstrips; i++)
+    {
+        if (stripbytes > bytecount)
+            stripbytes = bytecount;
+        newcounts[i] = stripbytes;
+        newoffsets[i] = stripbytes ? offset : 0;
+        offset += stripbytes;
+        bytecount -= stripbytes;
+    }
+
+    /*
+     * Replace old single strip info with multi-strip info.
+     */
+    td->td_stripsperimage = td->td_nstrips = nstrips;
+    TIFFSetField(tif, TIFFTAG_ROWSPERSTRIP, rowsperstrip);
+
+    _TIFFfree(td->td_stripbytecount);
+    _TIFFfree(td->td_stripoffset);
+    td->td_stripbytecount = newcounts;
+    td->td_stripoffset = newoffsets;
+    td->td_stripbytecountsorted = 1;
+}
+
+
 /*
 * Replace a single strip (tile) of uncompressed data by multiple strips
 * (tiles), each approximately STRIP_SIZE_DEFAULT bytes. This is useful for
@@ -5714,11 +5785,8 @@ ChopUpSingleUncompressedStrip(TIFF* tif)
 	uint32 rowblock;
 	uint64 rowblockbytes;
 	uint64 stripbytes;
-	uint32 strip;
 	uint32 nstrips;
 	uint32 rowsperstrip;
-	uint64* newcounts;
-	uint64* newoffsets;

 	bytecount = td->td_stripbytecount[0];
        /* On a newly created file, just re-opened to be filled, we */
@@ -5769,46 +5837,106 @@ ChopUpSingleUncompressedStrip(TIFF* tif)
            return;
        }

-	newcounts = (uint64*) _TIFFCheckMalloc(tif, nstrips, sizeof (uint64),
-				"for chopped \"StripByteCounts\" array");
-	newoffsets = (uint64*) _TIFFCheckMalloc(tif, nstrips, sizeof (uint64),
-				"for chopped \"StripOffsets\" array");
-	if (newcounts == NULL || newoffsets == NULL) {
-		/*
-		 * Unable to allocate new strip information, give up and use
-		 * the original one strip information.
-		 */
-		if (newcounts != NULL)
-			_TIFFfree(newcounts);
-		if (newoffsets != NULL)
-			_TIFFfree(newoffsets);
-		return;
-	}
-	/*
-	 * Fill the strip information arrays with new bytecounts and offsets
-	 * that reflect the broken-up format.
-	 */
-	for (strip = 0; strip < nstrips; strip++) {
-		if (stripbytes > bytecount)
-			stripbytes = bytecount;
-		newcounts[strip] = stripbytes;
-		newoffsets[strip] = stripbytes ? offset : 0;
-		offset += stripbytes;
-		bytecount -= stripbytes;
-	}
-	/*
-	 * Replace old single strip info with multi-strip info.
-	 */
-	td->td_stripsperimage = td->td_nstrips = nstrips;
-	TIFFSetField(tif, TIFFTAG_ROWSPERSTRIP, rowsperstrip);
-
-	_TIFFfree(td->td_stripbytecount);
-	_TIFFfree(td->td_stripoffset);
-	td->td_stripbytecount = newcounts;
-	td->td_stripoffset = newoffsets;
-	td->td_stripbytecountsorted = 1;
+        allocChoppedUpStripArrays(tif, nstrips, stripbytes, rowsperstrip);
 }

+
+/*
+ * Replace a file with contiguous strips > 2 GB of uncompressed data by
+ * multiple smaller strips. This is useful for
+ * dealing with large images or for dealing with machines with a limited
+ * amount memory.
+ */
+static void TryChopUpUncompressedBigTiff( TIFF* tif )
+{
+    TIFFDirectory *td = &tif->tif_dir;
+    uint32 rowblock;
+    uint64 rowblockbytes;
+    uint32 i;
+    uint64 stripsize;
+    uint32 rowblocksperstrip;
+    uint32 rowsperstrip;
+    uint64 stripbytes;
+    uint32 nstrips;
+
+    stripsize = TIFFStripSize64(tif);
+
+    assert( tif->tif_dir.td_planarconfig == PLANARCONFIG_CONTIG );
+    assert( tif->tif_dir.td_compression == COMPRESSION_NONE );
+    assert( (tif->tif_flags&(TIFF_STRIPCHOP|TIFF_ISTILED)) == TIFF_STRIPCHOP );
+    assert( stripsize > 0x7FFFFFFFUL );
+
+    /* On a newly created file, just re-opened to be filled, we */
+    /* don't want strip chop to trigger as it is going to cause issues */
+    /* later ( StripOffsets and StripByteCounts improperly filled) . */
+    if( td->td_stripbytecount[0] == 0 && tif->tif_mode != O_RDONLY )
+        return;
+
+    if ((td->td_photometric == PHOTOMETRIC_YCBCR)&&
+        (!isUpSampled(tif)))
+        rowblock = td->td_ycbcrsubsampling[1];
+    else
+        rowblock = 1;
+    rowblockbytes = TIFFVStripSize64(tif, rowblock);
+    if( rowblockbytes == 0 || rowblockbytes > 0x7FFFFFFFUL )
+    {
+        /* In case of file with gigantic width */
+        return;
+    }
+
+    /* Check that the strips are contiguous and of the expected size */
+    for( i = 0; i < td->td_nstrips; i++ )
+    {
+        if( i == td->td_nstrips - 1 )
+        {
+            if( td->td_stripbytecount[i] < TIFFVStripSize64(
+                    tif, td->td_imagelength - i * td->td_rowsperstrip ) )
+            {
+                return;
+            }
+        }
+        else
+        {
+            if( td->td_stripbytecount[i] != stripsize )
+            {
+                return;
+            }
+            if( i > 0 && td->td_stripoffset[i] !=
+                    td->td_stripoffset[i-1] + td->td_stripbytecount[i - 1] )
+            {
+                return;
+            }
+        }
+    }
+
+    /* Aim for 512 MB strips (that will still be manageable by 32 bit builds */
+    rowblocksperstrip = (uint32) (512 * 1024 * 1024 / rowblockbytes);
+    if( rowblocksperstrip == 0 )
+        rowblocksperstrip = 1;
+    rowsperstrip = rowblocksperstrip * rowblock;
+    stripbytes = rowblocksperstrip * rowblockbytes;
+    assert( stripbytes <= 0x7FFFFFFFUL );
+
+    nstrips = TIFFhowmany_32(td->td_imagelength, rowsperstrip);
+    if( nstrips == 0 )
+        return;
+
+    /* If we are going to allocate a lot of memory, make sure that the */
+    /* file is as big as needed */
+    if( tif->tif_mode == O_RDONLY &&
+        nstrips > 1000000 &&
+        (td->td_stripoffset[td->td_nstrips-1] > TIFFGetFileSize(tif) ||
+         td->td_stripoffset[td->td_nstrips-1] +
+         td->td_stripbytecount[td->td_nstrips-1] > TIFFGetFileSize(tif)) )
+    {
+        return;
+    }
+
+    allocChoppedUpStripArrays(tif, nstrips, stripbytes, rowsperstrip);
+}
+
+
+
 int _TIFFFillStriles( TIFF *tif )
 {
    return _TIFFFillStrilesInternal( tif, 1 );
@@ -28,7 +28,6 @@
 * Directory Write Support Routines.
 */
 #include "tiffiop.h"
-#include <float.h>

 #ifdef HAVE_IEEEFP
 #define TIFFCvtNativeToIEEEFloat(tif, n, fp)
@@ -946,15 +945,6 @@ bad:
 	return(0);
 }

-static float TIFFClampDoubleToFloat( double val )
-{
-    if( val > FLT_MAX )
-        return FLT_MAX;
-    if( val < -FLT_MAX )
-        return -FLT_MAX;
-    return (float)val;
-}
-
 static int8 TIFFClampDoubleToInt8( double val )
 {
    if( val > 127 )
@@ -1029,7 +1019,7 @@ TIFFWriteDirectoryTagSampleformatArray(TIFF* tif, uint32* ndir, TIFFDirEntry* di
 			if (tif->tif_dir.td_bitspersample<=32)
 			{
 				for (i = 0; i < count; ++i)
-					((float*)conv)[i] = TIFFClampDoubleToFloat(value[i]);
+					((float*)conv)[i] = _TIFFClampDoubleToFloat(value[i]);
 				ok = TIFFWriteDirectoryTagFloatArray(tif,ndir,dir,tag,count,(float*)conv);
 			}
 			else
@@ -1893,12 +1883,14 @@ TIFFWriteDirectoryTagTransferfunction(TIFF* tif, uint32* ndir, TIFFDirEntry* dir
 		n=3;
 	if (n==3)
 	{
-		if (!_TIFFmemcmp(tif->tif_dir.td_transferfunction[0],tif->tif_dir.td_transferfunction[2],m*sizeof(uint16)))
+		if (tif->tif_dir.td_transferfunction[2] == NULL ||
+		    !_TIFFmemcmp(tif->tif_dir.td_transferfunction[0],tif->tif_dir.td_transferfunction[2],m*sizeof(uint16)))
 			n=2;
 	}
 	if (n==2)
 	{
-		if (!_TIFFmemcmp(tif->tif_dir.td_transferfunction[0],tif->tif_dir.td_transferfunction[1],m*sizeof(uint16)))
+		if (tif->tif_dir.td_transferfunction[1] == NULL ||
+		    !_TIFFmemcmp(tif->tif_dir.td_transferfunction[0],tif->tif_dir.td_transferfunction[1],m*sizeof(uint16)))
 			n=1;
 	}
 	if (n==0)
@@ -742,9 +742,14 @@ LogLuvEncodeTile(TIFF* tif, uint8* bp, tmsize_t cc, uint16 s)
 #undef exp2  /* Conflict with C'99 function */
 #define exp2(x)		exp(M_LN2*(x))

-#define itrunc(x,m)	((m)==SGILOGENCODE_NODITHER ? \
-				(int)(x) : \
-				(int)((x) + rand()*(1./RAND_MAX) - .5))
+static int itrunc(double x, int m)
+{
+    if( m == SGILOGENCODE_NODITHER )
+        return (int)x;
+    /* Silence CoverityScan warning about bad crypto function */
+    /* coverity[dont_call] */
+    return (int)(x + rand()*(1./RAND_MAX) - .5);
+}

 #if !LOGLUV_PUBLIC
 static
@@ -247,6 +247,8 @@ LZWSetupDecode(TIFF* tif)
 		/*
 		 * Zero-out the unused entries
                 */
+                /* Silence false positive */
+                /* coverity[overrun-buffer-arg] */
                 _TIFFmemset(&sp->dec_codetab[CODE_CLEAR], 0,
 			     (CODE_FIRST - CODE_CLEAR) * sizeof (code_t));
 	}
@@ -640,6 +640,7 @@ PixarLogGuessDataFmt(TIFFDirectory *td)
 static tmsize_t
 multiply_ms(tmsize_t m1, tmsize_t m2)
 {
+        assert(m1 >= 0 && m2 >= 0);
        if( m1 == 0 || m2 > TIFF_TMSIZE_T_MAX / m1 )
            return 0;
        return m1 * m2;
@@ -648,6 +649,7 @@ multiply_ms(tmsize_t m1, tmsize_t m2)
 static tmsize_t
 add_ms(tmsize_t m1, tmsize_t m2)
 {
+        assert(m1 >= 0 && m2 >= 0);
 	/* if either input is zero, assume overflow already occurred */
 	if (m1 == 0 || m2 == 0)
 		return 0;
@@ -817,9 +819,7 @@ PixarLogDecode(TIFF* tif, uint8* op, tmsize_t occ, uint16 s)
 			TIFFErrorExt(tif->tif_clientdata, module,
 			    "Decoding error at scanline %lu, %s",
 			    (unsigned long) tif->tif_row, sp->stream.msg ? sp->stream.msg : "(null)");
-			if (inflateSync(&sp->stream) != Z_OK)
-				return (0);
-			continue;
+			return (0);
 		}
 		if (state != Z_OK) {
 			TIFFErrorExt(tif->tif_clientdata, module, "ZLib error: %s",
@@ -103,6 +103,11 @@ static int TIFFReadAndRealloc( TIFF* tif, tmsize_t size,
                }
                tif->tif_rawdata = new_rawdata;
            }
+            if( tif->tif_rawdata == NULL )
+            {
+                /* should not happen in practice but helps CoverityScan */
+                return 0;
+            }

            bytes_read = TIFFReadFile(tif,
                tif->tif_rawdata + rawdata_offset + already_read, to_read);
@@ -1367,7 +1372,8 @@ TIFFFillTile(TIFF* tif, uint32 tile)
                        tif->tif_rawdataoff = 0;
                        tif->tif_rawdataloaded = bytecountm;
                        
-			if (!isFillOrder(tif, td->td_fillorder) &&
+			if (tif->tif_rawdata != NULL &&
+                            !isFillOrder(tif, td->td_fillorder) &&
 			    (tif->tif_flags & TIFF_NOBITREV) == 0)
 				TIFFReverseBits(tif->tif_rawdata,
                                                tif->tif_rawdataloaded);
@@ -349,6 +349,12 @@ TWebPSetupEncode(TIFF* tif)

  sp->state |= LSTATE_INIT_ENCODE;

+  if (!WebPPictureInit(&sp->sPicture)) {
+    TIFFErrorExt(tif->tif_clientdata, module,
+        "Error initializing WebP picture.");
+    return 0;
+  }
+
  if (!WebPConfigInitInternal(&sp->sEncoderConfig, WEBP_PRESET_DEFAULT,
                              sp->quality_level,
                              WEBP_ENCODER_ABI_VERSION)) {
@@ -357,9 +363,13 @@ TWebPSetupEncode(TIFF* tif)
    return 0;
  }

-#if WEBP_ENCODER_ABI_VERSION >= 0x0100
-  sp->sEncoderConfig.lossless = sp->lossless;
-#endif
+  // WebPConfigInitInternal above sets lossless to false
+  #if WEBP_ENCODER_ABI_VERSION >= 0x0100
+    sp->sEncoderConfig.lossless = sp->lossless;
+    if (sp->lossless) {
+      sp->sPicture.use_argb = 1;
+    }
+  #endif

  if (!WebPValidateConfig(&sp->sEncoderConfig)) {
    TIFFErrorExt(tif->tif_clientdata, module,
@@ -367,12 +377,6 @@ TWebPSetupEncode(TIFF* tif)
    return 0;
  }

-  if (!WebPPictureInit(&sp->sPicture)) {
-    TIFFErrorExt(tif->tif_clientdata, module,
-        "Error initializing WebP picture.");
-    return 0;
-  }
-
  return 1;
 }

@@ -415,6 +419,12 @@ TWebPPreEncode(TIFF* tif, uint16 s)
  /* set up buffer for raw data */
  /* given above check and that nSamples <= 4, buffer_size is <= 1 GB */
  sp->buffer_size = segment_width * segment_height * sp->nSamples;
+  
+  if (sp->pBuffer != NULL) {
+      _TIFFfree(sp->pBuffer);
+      sp->pBuffer = NULL;    
+  }
+  
  sp->pBuffer = _TIFFmalloc(sp->buffer_size);
  if( !sp->pBuffer) {
      TIFFErrorExt(tif->tif_clientdata, module, "Cannot allocate buffer");
@@ -460,7 +470,7 @@ TWebPPostEncode(TIFF* tif)
                    "WebPPictureImportRGB() failed");
      return 0;
  }
-
+  
  if (!WebPEncode(&sp->sEncoderConfig, &sp->sPicture)) {

 #if WEBP_ENCODER_ABI_VERSION >= 0x0100
@@ -540,15 +550,13 @@ TWebPCleanup(TIFF* tif)
  }
  
  if (sp->pBuffer != NULL) {
-    _TIFFfree(sp->pBuffer);
-    sp->pBuffer = NULL;
+      _TIFFfree(sp->pBuffer);
+      sp->pBuffer = NULL;    
  }

-  if (tif->tif_data) {
-    _TIFFfree(tif->tif_data);
-    tif->tif_data = NULL;
-  }
-  
+  _TIFFfree(tif->tif_data);
+  tif->tif_data = NULL;
+
  _TIFFSetDefaultCompressionState(tif);
 }

@@ -570,6 +578,9 @@ TWebPVSetField(TIFF* tif, uint32 tag, va_list ap)
  case TIFFTAG_WEBP_LOSSLESS:
    #if WEBP_ENCODER_ABI_VERSION >= 0x0100
    sp->lossless = va_arg(ap, int);
+    if (sp->lossless){
+      sp->quality_level = 100.0f;      
+    }
    return 1;
    #else
      TIFFErrorExt(tif->tif_clientdata, module,
@@ -124,7 +124,6 @@ ZIPSetupDecode(TIFF* tif)
 static int
 ZIPPreDecode(TIFF* tif, uint16 s)
 {
-	static const char module[] = "ZIPPreDecode";
 	ZIPState* sp = DecoderState(tif);

 	(void) s;
@@ -138,12 +137,7 @@ ZIPPreDecode(TIFF* tif, uint16 s)
 	    we need to simplify this code to reflect a ZLib that is likely updated
 	    to deal with 8byte memory sizes, though this code will respond
 	    appropriately even before we simplify it */
-	sp->stream.avail_in = (uInt) tif->tif_rawcc;
-	if ((tmsize_t)sp->stream.avail_in != tif->tif_rawcc)
-	{
-		TIFFErrorExt(tif->tif_clientdata, module, "ZLib cannot deal with buffers this size");
-		return (0);
-	}
+	sp->stream.avail_in = (uint64)tif->tif_rawcc < 0xFFFFFFFFU ? (uInt) tif->tif_rawcc : 0xFFFFFFFFU;
 	return (inflateReset(&sp->stream) == Z_OK);
 }

@@ -158,46 +152,43 @@ ZIPDecode(TIFF* tif, uint8* op, tmsize_t occ, uint16 s)
 	assert(sp->state == ZSTATE_INIT_DECODE);

        sp->stream.next_in = tif->tif_rawcp;
-	sp->stream.avail_in = (uInt) tif->tif_rawcc;
        
 	sp->stream.next_out = op;
 	assert(sizeof(sp->stream.avail_out)==4);  /* if this assert gets raised,
 	    we need to simplify this code to reflect a ZLib that is likely updated
 	    to deal with 8byte memory sizes, though this code will respond
 	    appropriately even before we simplify it */
-	sp->stream.avail_out = (uInt) occ;
-	if ((tmsize_t)sp->stream.avail_out != occ)
-	{
-		TIFFErrorExt(tif->tif_clientdata, module, "ZLib cannot deal with buffers this size");
-		return (0);
-	}
 	do {
-		int state = inflate(&sp->stream, Z_PARTIAL_FLUSH);
+                int state;
+                uInt avail_in_before = (uint64)tif->tif_rawcc <= 0xFFFFFFFFU ? (uInt)tif->tif_rawcc : 0xFFFFFFFFU;
+                uInt avail_out_before = (uint64)occ < 0xFFFFFFFFU ? (uInt) occ : 0xFFFFFFFFU;
+                sp->stream.avail_in = avail_in_before;
+                sp->stream.avail_out = avail_out_before;
+		state = inflate(&sp->stream, Z_PARTIAL_FLUSH);
+		tif->tif_rawcc -= (avail_in_before - sp->stream.avail_in);
+                occ -= (avail_out_before - sp->stream.avail_out);
 		if (state == Z_STREAM_END)
 			break;
 		if (state == Z_DATA_ERROR) {
 			TIFFErrorExt(tif->tif_clientdata, module,
 			    "Decoding error at scanline %lu, %s",
 			     (unsigned long) tif->tif_row, SAFE_MSG(sp));
-			if (inflateSync(&sp->stream) != Z_OK)
-				return (0);
-			continue;
+			return (0);
 		}
 		if (state != Z_OK) {
 			TIFFErrorExt(tif->tif_clientdata, module, 
 				     "ZLib error: %s", SAFE_MSG(sp));
 			return (0);
 		}
-	} while (sp->stream.avail_out > 0);
-	if (sp->stream.avail_out != 0) {
+	} while (occ > 0);
+	if (occ != 0) {
 		TIFFErrorExt(tif->tif_clientdata, module,
 		    "Not enough data at scanline %lu (short " TIFF_UINT64_FORMAT " bytes)",
-		    (unsigned long) tif->tif_row, (TIFF_UINT64_T) sp->stream.avail_out);
+		    (unsigned long) tif->tif_row, (TIFF_UINT64_T) occ);
 		return (0);
 	}

        tif->tif_rawcp = sp->stream.next_in;
-        tif->tif_rawcc = sp->stream.avail_in;

 	return (1);
 }
@@ -229,7 +220,6 @@ ZIPSetupEncode(TIFF* tif)
 static int
 ZIPPreEncode(TIFF* tif, uint16 s)
 {
-	static const char module[] = "ZIPPreEncode";
 	ZIPState *sp = EncoderState(tif);

 	(void) s;
@@ -242,12 +232,7 @@ ZIPPreEncode(TIFF* tif, uint16 s)
 	    we need to simplify this code to reflect a ZLib that is likely updated
 	    to deal with 8byte memory sizes, though this code will respond
 	    appropriately even before we simplify it */
-	sp->stream.avail_out = (uInt)tif->tif_rawdatasize;
-	if ((tmsize_t)sp->stream.avail_out != tif->tif_rawdatasize)
-	{
-		TIFFErrorExt(tif->tif_clientdata, module, "ZLib cannot deal with buffers this size");
-		return (0);
-	}
+	sp->stream.avail_out = (uint64)tif->tif_rawdatasize <= 0xFFFFFFFFU ? (uInt)tif->tif_rawdatasize : 0xFFFFFFFFU;
 	return (deflateReset(&sp->stream) == Z_OK);
 }

@@ -269,13 +254,9 @@ ZIPEncode(TIFF* tif, uint8* bp, tmsize_t cc, uint16 s)
 	    we need to simplify this code to reflect a ZLib that is likely updated
 	    to deal with 8byte memory sizes, though this code will respond
 	    appropriately even before we simplify it */
-	sp->stream.avail_in = (uInt) cc;
-	if ((tmsize_t)sp->stream.avail_in != cc)
-	{
-		TIFFErrorExt(tif->tif_clientdata, module, "ZLib cannot deal with buffers this size");
-		return (0);
-	}
 	do {
+                uInt avail_in_before = (uint64)cc <= 0xFFFFFFFFU ? (uInt)cc : 0xFFFFFFFFU;
+                sp->stream.avail_in = avail_in_before;
 		if (deflate(&sp->stream, Z_NO_FLUSH) != Z_OK) {
 			TIFFErrorExt(tif->tif_clientdata, module, 
 				     "Encoder error: %s",
@@ -286,9 +267,10 @@ ZIPEncode(TIFF* tif, uint8* bp, tmsize_t cc, uint16 s)
 			tif->tif_rawcc = tif->tif_rawdatasize;
 			TIFFFlushData1(tif);
 			sp->stream.next_out = tif->tif_rawdata;
-			sp->stream.avail_out = (uInt) tif->tif_rawdatasize;  /* this is a safe typecast, as check is made already in ZIPPreEncode */
+			sp->stream.avail_out = (uint64)tif->tif_rawdatasize <= 0xFFFFFFFFU ? (uInt)tif->tif_rawdatasize : 0xFFFFFFFFU;
 		}
-	} while (sp->stream.avail_in > 0);
+		cc -= (avail_in_before - sp->stream.avail_in);
+	} while (cc > 0);
 	return (1);
 }

@@ -314,7 +296,7 @@ ZIPPostEncode(TIFF* tif)
 				tif->tif_rawcc =  tif->tif_rawdatasize - sp->stream.avail_out;
 				TIFFFlushData1(tif);
 				sp->stream.next_out = tif->tif_rawdata;
-				sp->stream.avail_out = (uInt) tif->tif_rawdatasize;  /* this is a safe typecast, as check is made already in ZIPPreEncode */
+				sp->stream.avail_out = (uint64)tif->tif_rawdatasize <= 0xFFFFFFFFU ? (uInt)tif->tif_rawdatasize : 0xFFFFFFFFU;
 			}
 			break;
 		default:
@@ -374,6 +374,8 @@ extern void* _TIFFCheckRealloc(TIFF*, void*, tmsize_t, tmsize_t, const char*);
 extern double _TIFFUInt64ToDouble(uint64);
 extern float _TIFFUInt64ToFloat(uint64);

+extern float _TIFFClampDoubleToFloat(double);
+
 extern tmsize_t
 _TIFFReadEncodedStripAndAllocBuffer(TIFF* tif, uint32 strip,
                                    void **buf, tmsize_t bufsizetoalloc,
@@ -21,6 +21,11 @@ if(ANDROID AND ARMEABI_V7A AND NOT NEON)
  endforeach()
 endif()

+# FIX for quant.h - requires C99 for() loops
+ocv_check_flag_support(C "-std=c99" _varname "${CMAKE_C_FLAGS}")
+if(${_varname})
+  set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -std=c99")
+endif()


 # ----------------------------------------------------------------------------------
@@ -0,0 +1,30 @@
+Copyright (c) 2010, Google Inc. All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+
+  * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+
+  * Redistributions in binary form must reproduce the above copyright
+    notice, this list of conditions and the following disclaimer in
+    the documentation and/or other materials provided with the
+    distribution.
+
+  * Neither the name of Google nor the names of its contributors may
+    be used to endorse or promote products derived from this software
+    without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
@@ -0,0 +1,22 @@
+diff --git a/3rdparty/libwebp/src/dsp/msa_macro.h b/3rdparty/libwebp/src/dsp/msa_macro.h
+index de026a1d9e..a16c0bb300 100644
+--- a/3rdparty/libwebp/src/dsp/msa_macro.h
+++ b/3rdparty/libwebp/src/dsp/msa_macro.h
+@@ -73,7 +73,7 @@
+   static inline TYPE FUNC_NAME(const void* const psrc) {  \
+     const uint8_t* const psrc_m = (const uint8_t*)psrc;   \
+     TYPE val_m;                                           \
+-    asm volatile (                                        \
+    __asm__ volatile (                                        \
+       "" #INSTR " %[val_m], %[psrc_m]  \n\t"              \
+       : [val_m] "=r" (val_m)                              \
+       : [psrc_m] "m" (*psrc_m));                          \
+@@ -86,7 +86,7 @@
+   static inline void FUNC_NAME(TYPE val, void* const pdst) { \
+     uint8_t* const pdst_m = (uint8_t*)pdst;                  \
+     TYPE val_m = val;                                        \
+-    asm volatile (                                           \
+    __asm__ volatile (                                           \
+       " " #INSTR "  %[val_m],  %[pdst_m]  \n\t"              \
+       : [pdst_m] "=m" (*pdst_m)                              \
+       : [val_m] "r" (val_m));                                \
@@ -51,4 +51,4 @@ void WebPDeallocateAlphaMemory(VP8Decoder* const dec);
 }    // extern "C"
 #endif

-#endif  /* WEBP_DEC_ALPHAI_DEC_H_ */
+#endif  // WEBP_DEC_ALPHAI_DEC_H_
@@ -74,7 +74,8 @@ static VP8StatusCode CheckDecBuffer(const WebPDecBuffer* const buffer) {
  } else {    // RGB checks
    const WebPRGBABuffer* const buf = &buffer->u.RGBA;
    const int stride = abs(buf->stride);
-    const uint64_t size = MIN_BUFFER_SIZE(width, height, stride);
+    const uint64_t size =
+        MIN_BUFFER_SIZE(width * kModeBpp[mode], height, stride);
    ok &= (size <= buf->size);
    ok &= (stride >= width * kModeBpp[mode]);
    ok &= (buf->rgba != NULL);
@@ -51,4 +51,4 @@ enum { MB_FEATURE_TREE_PROBS = 3,
       NUM_PROBAS = 11
     };

-#endif    // WEBP_DEC_COMMON_DEC_H_
+#endif  // WEBP_DEC_COMMON_DEC_H_
@@ -338,7 +338,6 @@ void VP8InitDithering(const WebPDecoderOptions* const options,
      for (s = 0; s < NUM_MB_SEGMENTS; ++s) {
        VP8QuantMatrix* const dqm = &dec->dqm_[s];
        if (dqm->uv_quant_ < DITHER_AMP_TAB_SIZE) {
-          // TODO(skal): should we specially dither more for uv_quant_ < 0?
          const int idx = (dqm->uv_quant_ < 0) ? 0 : dqm->uv_quant_;
          dqm->dither_ = (f * kQuantToDitherAmp[idx]) >> 3;
        }
@@ -669,15 +668,9 @@ int VP8GetThreadMethod(const WebPDecoderOptions* const options,
  (void)height;
  assert(headers == NULL || !headers->is_lossless);
 #if defined(WEBP_USE_THREAD)
-  if (width < MIN_WIDTH_FOR_THREADS) return 0;
-  // TODO(skal): tune the heuristic further
-#if 0
-  if (height < 2 * width) return 2;
+  if (width >= MIN_WIDTH_FOR_THREADS) return 2;
 #endif
-  return 2;
-#else   // !WEBP_USE_THREAD
  return 0;
-#endif
 }

 #undef MT_CACHE_LINES
@@ -739,7 +732,7 @@ static int AllocateMemory(VP8Decoder* const dec) {
  mem += f_info_size;
  dec->thread_ctx_.id_ = 0;
  dec->thread_ctx_.f_info_ = dec->f_info_;
-  if (dec->mt_method_ > 0) {
+  if (dec->filter_type_ > 0 && dec->mt_method_ > 0) {
    // secondary cache line. The deblocking process need to make use of the
    // filtering strength from previous macroblock row, while the new ones
    // are being decoded in parallel. We'll just swap the pointers.
@@ -140,10 +140,9 @@ static void DoRemap(WebPIDecoder* const idec, ptrdiff_t offset) {
      if (NeedCompressedAlpha(idec)) {
        ALPHDecoder* const alph_dec = dec->alph_dec_;
        dec->alpha_data_ += offset;
-        if (alph_dec != NULL) {
+        if (alph_dec != NULL && alph_dec->vp8l_dec_ != NULL) {
          if (alph_dec->method_ == ALPHA_LOSSLESS_COMPRESSION) {
            VP8LDecoder* const alph_vp8l_dec = alph_dec->vp8l_dec_;
-            assert(alph_vp8l_dec != NULL);
            assert(dec->alpha_data_size_ >= ALPHA_HEADER_LEN);
            VP8LBitReaderSetBuffer(&alph_vp8l_dec->br_,
                                   dec->alpha_data_ + ALPHA_HEADER_LEN,
@@ -167,9 +166,11 @@ static int AppendToMemBuffer(WebPIDecoder* const idec,
  VP8Decoder* const dec = (VP8Decoder*)idec->dec_;
  MemBuffer* const mem = &idec->mem_;
  const int need_compressed_alpha = NeedCompressedAlpha(idec);
-  const uint8_t* const old_start = mem->buf_ + mem->start_;
+  const uint8_t* const old_start =
+      (mem->buf_ == NULL) ? NULL : mem->buf_ + mem->start_;
  const uint8_t* const old_base =
      need_compressed_alpha ? dec->alpha_data_ : old_start;
+  assert(mem->buf_ != NULL || mem->start_ == 0);
  assert(mem->mode_ == MEM_MODE_APPEND);
  if (data_size > MAX_CHUNK_PAYLOAD) {
    // security safeguard: trying to allocate more than what the format
@@ -185,7 +186,7 @@ static int AppendToMemBuffer(WebPIDecoder* const idec,
    uint8_t* const new_buf =
        (uint8_t*)WebPSafeMalloc(extra_size, sizeof(*new_buf));
    if (new_buf == NULL) return 0;
-    memcpy(new_buf, old_base, current_size);
+    if (old_base != NULL) memcpy(new_buf, old_base, current_size);
    WebPSafeFree(mem->buf_);
    mem->buf_ = new_buf;
    mem->buf_size_ = (size_t)extra_size;
@@ -193,6 +194,7 @@ static int AppendToMemBuffer(WebPIDecoder* const idec,
    mem->end_ = current_size;
  }

+  assert(mem->buf_ != NULL);
  memcpy(mem->buf_ + mem->end_, data, data_size);
  mem->end_ += data_size;
  assert(mem->end_ <= mem->buf_size_);
@@ -205,7 +207,9 @@ static int RemapMemBuffer(WebPIDecoder* const idec,
                          const uint8_t* const data, size_t data_size) {
  MemBuffer* const mem = &idec->mem_;
  const uint8_t* const old_buf = mem->buf_;
-  const uint8_t* const old_start = old_buf + mem->start_;
+  const uint8_t* const old_start =
+      (old_buf == NULL) ? NULL : old_buf + mem->start_;
+  assert(old_buf != NULL || mem->start_ == 0);
  assert(mem->mode_ == MEM_MODE_MAP);

  if (data_size < mem->buf_size_) return 0;  // can't remap to a shorter buffer!
@@ -283,10 +287,8 @@ static void RestoreContext(const MBContext* context, VP8Decoder* const dec,

 static VP8StatusCode IDecError(WebPIDecoder* const idec, VP8StatusCode error) {
  if (idec->state_ == STATE_VP8_DATA) {
-    VP8Io* const io = &idec->io_;
-    if (io->teardown != NULL) {
-      io->teardown(io);
-    }
+    // Synchronize the thread, clean-up and check for errors.
+    VP8ExitCritical((VP8Decoder*)idec->dec_, &idec->io_);
  }
  idec->state_ = STATE_ERROR;
  return error;
@@ -451,7 +453,10 @@ static VP8StatusCode DecodeRemaining(WebPIDecoder* const idec) {
  VP8Decoder* const dec = (VP8Decoder*)idec->dec_;
  VP8Io* const io = &idec->io_;

-  assert(dec->ready_);
+  // Make sure partition #0 has been read before, to set dec to ready_.
+  if (!dec->ready_) {
+    return IDecError(idec, VP8_STATUS_BITSTREAM_ERROR);
+  }
  for (; dec->mb_y_ < dec->mb_h_; ++dec->mb_y_) {
    if (idec->last_mb_y_ != dec->mb_y_) {
      if (!VP8ParseIntraModeRow(&dec->br_, dec)) {
@@ -473,6 +478,12 @@ static VP8StatusCode DecodeRemaining(WebPIDecoder* const idec) {
            MemDataSize(&idec->mem_) > MAX_MB_SIZE) {
          return IDecError(idec, VP8_STATUS_BITSTREAM_ERROR);
        }
+        // Synchronize the threads.
+        if (dec->mt_method_ > 0) {
+          if (!WebPGetWorkerInterface()->Sync(&dec->worker_)) {
+            return IDecError(idec, VP8_STATUS_BITSTREAM_ERROR);
+          }
+        }
        RestoreContext(&context, dec, token_br);
        return VP8_STATUS_SUSPENDED;
      }
@@ -491,6 +502,7 @@ static VP8StatusCode DecodeRemaining(WebPIDecoder* const idec) {
  }
  // Synchronize the thread and check for errors.
  if (!VP8ExitCritical(dec, io)) {
+    idec->state_ = STATE_ERROR;  // prevent re-entry in IDecError
    return IDecError(idec, VP8_STATUS_USER_ABORT);
  }
  dec->ready_ = 0;
@@ -571,6 +583,10 @@ static VP8StatusCode IDecode(WebPIDecoder* idec) {
    status = DecodePartition0(idec);
  }
  if (idec->state_ == STATE_VP8_DATA) {
+    const VP8Decoder* const dec = (VP8Decoder*)idec->dec_;
+    if (dec == NULL) {
+      return VP8_STATUS_SUSPENDED;  // can't continue if we have no decoder.
+    }
    status = DecodeRemaining(idec);
  }
  if (idec->state_ == STATE_VP8L_HEADER) {
@@ -61,12 +61,17 @@ static const uint16_t kAcTable[128] = {

 void VP8ParseQuant(VP8Decoder* const dec) {
  VP8BitReader* const br = &dec->br_;
-  const int base_q0 = VP8GetValue(br, 7);
-  const int dqy1_dc = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0;
-  const int dqy2_dc = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0;
-  const int dqy2_ac = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0;
-  const int dquv_dc = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0;
-  const int dquv_ac = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0;
+  const int base_q0 = VP8GetValue(br, 7, "global-header");
+  const int dqy1_dc = VP8Get(br, "global-header") ?
+       VP8GetSignedValue(br, 4, "global-header") : 0;
+  const int dqy2_dc = VP8Get(br, "global-header") ?
+       VP8GetSignedValue(br, 4, "global-header") : 0;
+  const int dqy2_ac = VP8Get(br, "global-header") ?
+       VP8GetSignedValue(br, 4, "global-header") : 0;
+  const int dquv_dc = VP8Get(br, "global-header") ?
+       VP8GetSignedValue(br, 4, "global-header") : 0;
+  const int dquv_ac = VP8Get(br, "global-header") ?
+       VP8GetSignedValue(br, 4, "global-header") : 0;

  const VP8SegmentHeader* const hdr = &dec->segment_hdr_;
  int i;
@@ -296,20 +296,21 @@ static void ParseIntraMode(VP8BitReader* const br,
  // to decode more than 1 keyframe.
  if (dec->segment_hdr_.update_map_) {
    // Hardcoded tree parsing
-    block->segment_ = !VP8GetBit(br, dec->proba_.segments_[0])
-                    ? VP8GetBit(br, dec->proba_.segments_[1])
-                    : 2 + VP8GetBit(br, dec->proba_.segments_[2]);
+    block->segment_ = !VP8GetBit(br, dec->proba_.segments_[0], "segments")
+                    ?  VP8GetBit(br, dec->proba_.segments_[1], "segments")
+                    :  VP8GetBit(br, dec->proba_.segments_[2], "segments") + 2;
  } else {
    block->segment_ = 0;  // default for intra
  }
-  if (dec->use_skip_proba_) block->skip_ = VP8GetBit(br, dec->skip_p_);
+  if (dec->use_skip_proba_) block->skip_ = VP8GetBit(br, dec->skip_p_, "skip");

-  block->is_i4x4_ = !VP8GetBit(br, 145);   // decide for B_PRED first
+  block->is_i4x4_ = !VP8GetBit(br, 145, "block-size");
  if (!block->is_i4x4_) {
    // Hardcoded 16x16 intra-mode decision tree.
    const int ymode =
-        VP8GetBit(br, 156) ? (VP8GetBit(br, 128) ? TM_PRED : H_PRED)
-                           : (VP8GetBit(br, 163) ? V_PRED : DC_PRED);
+        VP8GetBit(br, 156, "pred-modes") ?
+            (VP8GetBit(br, 128, "pred-modes") ? TM_PRED : H_PRED) :
+            (VP8GetBit(br, 163, "pred-modes") ? V_PRED : DC_PRED);
    block->imodes_[0] = ymode;
    memset(top, ymode, 4 * sizeof(*top));
    memset(left, ymode, 4 * sizeof(*left));
@@ -323,22 +324,25 @@ static void ParseIntraMode(VP8BitReader* const br,
        const uint8_t* const prob = kBModesProba[top[x]][ymode];
 #if (USE_GENERIC_TREE == 1)
        // Generic tree-parsing
-        int i = kYModesIntra4[VP8GetBit(br, prob[0])];
+        int i = kYModesIntra4[VP8GetBit(br, prob[0], "pred-modes")];
        while (i > 0) {
-          i = kYModesIntra4[2 * i + VP8GetBit(br, prob[i])];
+          i = kYModesIntra4[2 * i + VP8GetBit(br, prob[i], "pred-modes")];
        }
        ymode = -i;
 #else
        // Hardcoded tree parsing
-        ymode = !VP8GetBit(br, prob[0]) ? B_DC_PRED :
-                  !VP8GetBit(br, prob[1]) ? B_TM_PRED :
-                    !VP8GetBit(br, prob[2]) ? B_VE_PRED :
-                      !VP8GetBit(br, prob[3]) ?
-                        (!VP8GetBit(br, prob[4]) ? B_HE_PRED :
-                          (!VP8GetBit(br, prob[5]) ? B_RD_PRED : B_VR_PRED)) :
-                        (!VP8GetBit(br, prob[6]) ? B_LD_PRED :
-                          (!VP8GetBit(br, prob[7]) ? B_VL_PRED :
-                            (!VP8GetBit(br, prob[8]) ? B_HD_PRED : B_HU_PRED)));
+        ymode = !VP8GetBit(br, prob[0], "pred-modes") ? B_DC_PRED :
+                  !VP8GetBit(br, prob[1], "pred-modes") ? B_TM_PRED :
+                    !VP8GetBit(br, prob[2], "pred-modes") ? B_VE_PRED :
+                      !VP8GetBit(br, prob[3], "pred-modes") ?
+                        (!VP8GetBit(br, prob[4], "pred-modes") ? B_HE_PRED :
+                          (!VP8GetBit(br, prob[5], "pred-modes") ? B_RD_PRED
+                                                                 : B_VR_PRED)) :
+                        (!VP8GetBit(br, prob[6], "pred-modes") ? B_LD_PRED :
+                          (!VP8GetBit(br, prob[7], "pred-modes") ? B_VL_PRED :
+                            (!VP8GetBit(br, prob[8], "pred-modes") ? B_HD_PRED
+                                                                   : B_HU_PRED))
+                        );
 #endif  // USE_GENERIC_TREE
        top[x] = ymode;
      }
@@ -348,9 +352,9 @@ static void ParseIntraMode(VP8BitReader* const br,
    }
  }
  // Hardcoded UVMode decision tree
-  block->uvmode_ = !VP8GetBit(br, 142) ? DC_PRED
-                 : !VP8GetBit(br, 114) ? V_PRED
-                 : VP8GetBit(br, 183) ? TM_PRED : H_PRED;
+  block->uvmode_ = !VP8GetBit(br, 142, "pred-modes-uv") ? DC_PRED
+                 : !VP8GetBit(br, 114, "pred-modes-uv") ? V_PRED
+                 : VP8GetBit(br, 183, "pred-modes-uv") ? TM_PRED : H_PRED;
 }

 int VP8ParseIntraModeRow(VP8BitReader* const br, VP8Decoder* const dec) {
@@ -514,8 +518,10 @@ void VP8ParseProba(VP8BitReader* const br, VP8Decoder* const dec) {
    for (b = 0; b < NUM_BANDS; ++b) {
      for (c = 0; c < NUM_CTX; ++c) {
        for (p = 0; p < NUM_PROBAS; ++p) {
-          const int v = VP8GetBit(br, CoeffsUpdateProba[t][b][c][p]) ?
-                        VP8GetValue(br, 8) : CoeffsProba0[t][b][c][p];
+          const int v =
+              VP8GetBit(br, CoeffsUpdateProba[t][b][c][p], "global-header") ?
+                        VP8GetValue(br, 8, "global-header") :
+                        CoeffsProba0[t][b][c][p];
          proba->bands_[t][b].probas_[c][p] = v;
        }
      }
@@ -524,9 +530,8 @@ void VP8ParseProba(VP8BitReader* const br, VP8Decoder* const dec) {
      proba->bands_ptr_[t][b] = &proba->bands_[t][kBands[b]];
    }
  }
-  dec->use_skip_proba_ = VP8Get(br);
+  dec->use_skip_proba_ = VP8Get(br, "global-header");
  if (dec->use_skip_proba_) {
-    dec->skip_p_ = VP8GetValue(br, 8);
+    dec->skip_p_ = VP8GetValue(br, 8, "global-header");
  }
 }
-
@@ -161,23 +161,26 @@ static int ParseSegmentHeader(VP8BitReader* br,
                              VP8SegmentHeader* hdr, VP8Proba* proba) {
  assert(br != NULL);
  assert(hdr != NULL);
-  hdr->use_segment_ = VP8Get(br);
+  hdr->use_segment_ = VP8Get(br, "global-header");
  if (hdr->use_segment_) {
-    hdr->update_map_ = VP8Get(br);
-    if (VP8Get(br)) {   // update data
+    hdr->update_map_ = VP8Get(br, "global-header");
+    if (VP8Get(br, "global-header")) {   // update data
      int s;
-      hdr->absolute_delta_ = VP8Get(br);
+      hdr->absolute_delta_ = VP8Get(br, "global-header");
      for (s = 0; s < NUM_MB_SEGMENTS; ++s) {
-        hdr->quantizer_[s] = VP8Get(br) ? VP8GetSignedValue(br, 7) : 0;
+        hdr->quantizer_[s] = VP8Get(br, "global-header") ?
+            VP8GetSignedValue(br, 7, "global-header") : 0;
      }
      for (s = 0; s < NUM_MB_SEGMENTS; ++s) {
-        hdr->filter_strength_[s] = VP8Get(br) ? VP8GetSignedValue(br, 6) : 0;
+        hdr->filter_strength_[s] = VP8Get(br, "global-header") ?
+            VP8GetSignedValue(br, 6, "global-header") : 0;
      }
    }
    if (hdr->update_map_) {
      int s;
      for (s = 0; s < MB_FEATURE_TREE_PROBS; ++s) {
-        proba->segments_[s] = VP8Get(br) ? VP8GetValue(br, 8) : 255u;
+        proba->segments_[s] = VP8Get(br, "global-header") ?
+            VP8GetValue(br, 8, "global-header") : 255u;
      }
    }
  } else {
@@ -205,7 +208,7 @@ static VP8StatusCode ParsePartitions(VP8Decoder* const dec,
  size_t last_part;
  size_t p;

-  dec->num_parts_minus_one_ = (1 << VP8GetValue(br, 2)) - 1;
+  dec->num_parts_minus_one_ = (1 << VP8GetValue(br, 2, "global-header")) - 1;
  last_part = dec->num_parts_minus_one_;
  if (size < 3 * last_part) {
    // we can't even read the sizes with sz[]! That's a failure.
@@ -229,21 +232,21 @@ static VP8StatusCode ParsePartitions(VP8Decoder* const dec,
 // Paragraph 9.4
 static int ParseFilterHeader(VP8BitReader* br, VP8Decoder* const dec) {
  VP8FilterHeader* const hdr = &dec->filter_hdr_;
-  hdr->simple_    = VP8Get(br);
-  hdr->level_     = VP8GetValue(br, 6);
-  hdr->sharpness_ = VP8GetValue(br, 3);
-  hdr->use_lf_delta_ = VP8Get(br);
+  hdr->simple_    = VP8Get(br, "global-header");
+  hdr->level_     = VP8GetValue(br, 6, "global-header");
+  hdr->sharpness_ = VP8GetValue(br, 3, "global-header");
+  hdr->use_lf_delta_ = VP8Get(br, "global-header");
  if (hdr->use_lf_delta_) {
-    if (VP8Get(br)) {   // update lf-delta?
+    if (VP8Get(br, "global-header")) {   // update lf-delta?
      int i;
      for (i = 0; i < NUM_REF_LF_DELTAS; ++i) {
-        if (VP8Get(br)) {
-          hdr->ref_lf_delta_[i] = VP8GetSignedValue(br, 6);
+        if (VP8Get(br, "global-header")) {
+          hdr->ref_lf_delta_[i] = VP8GetSignedValue(br, 6, "global-header");
        }
      }
      for (i = 0; i < NUM_MODE_LF_DELTAS; ++i) {
-        if (VP8Get(br)) {
-          hdr->mode_lf_delta_[i] = VP8GetSignedValue(br, 6);
+        if (VP8Get(br, "global-header")) {
+          hdr->mode_lf_delta_[i] = VP8GetSignedValue(br, 6, "global-header");
        }
      }
    }
@@ -352,8 +355,8 @@ int VP8GetHeaders(VP8Decoder* const dec, VP8Io* const io) {
  buf_size -= frm_hdr->partition_length_;

  if (frm_hdr->key_frame_) {
-    pic_hdr->colorspace_ = VP8Get(br);
-    pic_hdr->clamp_type_ = VP8Get(br);
+    pic_hdr->colorspace_ = VP8Get(br, "global-header");
+    pic_hdr->clamp_type_ = VP8Get(br, "global-header");
  }
  if (!ParseSegmentHeader(br, &dec->segment_hdr_, &dec->proba_)) {
    return VP8SetError(dec, VP8_STATUS_BITSTREAM_ERROR,
@@ -378,7 +381,7 @@ int VP8GetHeaders(VP8Decoder* const dec, VP8Io* const io) {
                       "Not a key frame.");
  }

-  VP8Get(br);   // ignore the value of update_proba_
+  VP8Get(br, "global-header");   // ignore the value of update_proba_

  VP8ParseProba(br, dec);

@@ -403,28 +406,28 @@ static const uint8_t kZigzag[16] = {
 // See section 13-2: http://tools.ietf.org/html/rfc6386#section-13.2
 static int GetLargeValue(VP8BitReader* const br, const uint8_t* const p) {
  int v;
-  if (!VP8GetBit(br, p[3])) {
-    if (!VP8GetBit(br, p[4])) {
+  if (!VP8GetBit(br, p[3], "coeffs")) {
+    if (!VP8GetBit(br, p[4], "coeffs")) {
      v = 2;
    } else {
-      v = 3 + VP8GetBit(br, p[5]);
+      v = 3 + VP8GetBit(br, p[5], "coeffs");
    }
  } else {
-    if (!VP8GetBit(br, p[6])) {
-      if (!VP8GetBit(br, p[7])) {
-        v = 5 + VP8GetBit(br, 159);
+    if (!VP8GetBit(br, p[6], "coeffs")) {
+      if (!VP8GetBit(br, p[7], "coeffs")) {
+        v = 5 + VP8GetBit(br, 159, "coeffs");
      } else {
-        v = 7 + 2 * VP8GetBit(br, 165);
-        v += VP8GetBit(br, 145);
+        v = 7 + 2 * VP8GetBit(br, 165, "coeffs");
+        v += VP8GetBit(br, 145, "coeffs");
      }
    } else {
      const uint8_t* tab;
-      const int bit1 = VP8GetBit(br, p[8]);
-      const int bit0 = VP8GetBit(br, p[9 + bit1]);
+      const int bit1 = VP8GetBit(br, p[8], "coeffs");
+      const int bit0 = VP8GetBit(br, p[9 + bit1], "coeffs");
      const int cat = 2 * bit1 + bit0;
      v = 0;
      for (tab = kCat3456[cat]; *tab; ++tab) {
-        v += v + VP8GetBit(br, *tab);
+        v += v + VP8GetBit(br, *tab, "coeffs");
      }
      v += 3 + (8 << cat);
    }
@@ -438,24 +441,24 @@ static int GetCoeffsFast(VP8BitReader* const br,
                         int ctx, const quant_t dq, int n, int16_t* out) {
  const uint8_t* p = prob[n]->probas_[ctx];
  for (; n < 16; ++n) {
-    if (!VP8GetBit(br, p[0])) {
+    if (!VP8GetBit(br, p[0], "coeffs")) {
      return n;  // previous coeff was last non-zero coeff
    }
-    while (!VP8GetBit(br, p[1])) {       // sequence of zero coeffs
+    while (!VP8GetBit(br, p[1], "coeffs")) {       // sequence of zero coeffs
      p = prob[++n]->probas_[0];
      if (n == 16) return 16;
    }
    {        // non zero coeff
      const VP8ProbaArray* const p_ctx = &prob[n + 1]->probas_[0];
      int v;
-      if (!VP8GetBit(br, p[2])) {
+      if (!VP8GetBit(br, p[2], "coeffs")) {
        v = 1;
        p = p_ctx[1];
      } else {
        v = GetLargeValue(br, p);
        p = p_ctx[2];
      }
-      out[kZigzag[n]] = VP8GetSigned(br, v) * dq[n > 0];
+      out[kZigzag[n]] = VP8GetSigned(br, v, "coeffs") * dq[n > 0];
    }
  }
  return 16;
@@ -468,24 +471,24 @@ static int GetCoeffsAlt(VP8BitReader* const br,
                        int ctx, const quant_t dq, int n, int16_t* out) {
  const uint8_t* p = prob[n]->probas_[ctx];
  for (; n < 16; ++n) {
-    if (!VP8GetBitAlt(br, p[0])) {
+    if (!VP8GetBitAlt(br, p[0], "coeffs")) {
      return n;  // previous coeff was last non-zero coeff
    }
-    while (!VP8GetBitAlt(br, p[1])) {       // sequence of zero coeffs
+    while (!VP8GetBitAlt(br, p[1], "coeffs")) {       // sequence of zero coeffs
      p = prob[++n]->probas_[0];
      if (n == 16) return 16;
    }
    {        // non zero coeff
      const VP8ProbaArray* const p_ctx = &prob[n + 1]->probas_[0];
      int v;
-      if (!VP8GetBitAlt(br, p[2])) {
+      if (!VP8GetBitAlt(br, p[2], "coeffs")) {
        v = 1;
        p = p_ctx[1];
      } else {
        v = GetLargeValue(br, p);
        p = p_ctx[2];
      }
-      out[kZigzag[n]] = VP8GetSigned(br, v) * dq[n > 0];
+      out[kZigzag[n]] = VP8GetSigned(br, v, "coeffs") * dq[n > 0];
    }
  }
  return 16;
@@ -182,4 +182,4 @@ WEBP_EXTERN int VP8LGetInfo(
 }    // extern "C"
 #endif

-#endif  /* WEBP_DEC_VP8_DEC_H_ */
+#endif  // WEBP_DEC_VP8_DEC_H_
@@ -31,7 +31,7 @@ extern "C" {

 // version numbers
 #define DEC_MAJ_VERSION 1
-#define DEC_MIN_VERSION 0
+#define DEC_MIN_VERSION 1
 #define DEC_REV_VERSION 0

 // YUV-cache parameters. Cache is 32-bytes wide (= one cacheline).
@@ -316,4 +316,4 @@ const uint8_t* VP8DecompressAlphaRows(VP8Decoder* const dec,
 }    // extern "C"
 #endif

-#endif  /* WEBP_DEC_VP8I_DEC_H_ */
+#endif  // WEBP_DEC_VP8I_DEC_H_
@@ -363,11 +363,14 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
  uint32_t* huffman_image = NULL;
  HTreeGroup* htree_groups = NULL;
  HuffmanCode* huffman_tables = NULL;
-  HuffmanCode* next = NULL;
+  HuffmanCode* huffman_table = NULL;
  int num_htree_groups = 1;
+  int num_htree_groups_max = 1;
  int max_alphabet_size = 0;
  int* code_lengths = NULL;
  const int table_size = kTableSize[color_cache_bits];
+  int* mapping = NULL;
+  int ok = 0;

  if (allow_recursion && VP8LReadBits(br, 1)) {
    // use meta Huffman codes.
@@ -384,10 +387,36 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
      // The huffman data is stored in red and green bytes.
      const int group = (huffman_image[i] >> 8) & 0xffff;
      huffman_image[i] = group;
-      if (group >= num_htree_groups) {
-        num_htree_groups = group + 1;
+      if (group >= num_htree_groups_max) {
+        num_htree_groups_max = group + 1;
      }
    }
+    // Check the validity of num_htree_groups_max. If it seems too big, use a
+    // smaller value for later. This will prevent big memory allocations to end
+    // up with a bad bitstream anyway.
+    // The value of 1000 is totally arbitrary. We know that num_htree_groups_max
+    // is smaller than (1 << 16) and should be smaller than the number of pixels
+    // (though the format allows it to be bigger).
+    if (num_htree_groups_max > 1000 || num_htree_groups_max > xsize * ysize) {
+      // Create a mapping from the used indices to the minimal set of used
+      // values [0, num_htree_groups)
+      mapping = (int*)WebPSafeMalloc(num_htree_groups_max, sizeof(*mapping));
+      if (mapping == NULL) {
+        dec->status_ = VP8_STATUS_OUT_OF_MEMORY;
+        goto Error;
+      }
+      // -1 means a value is unmapped, and therefore unused in the Huffman
+      // image.
+      memset(mapping, 0xff, num_htree_groups_max * sizeof(*mapping));
+      for (num_htree_groups = 0, i = 0; i < huffman_pixs; ++i) {
+        // Get the current mapping for the group and remap the Huffman image.
+        int* const mapped_group = &mapping[huffman_image[i]];
+        if (*mapped_group == -1) *mapped_group = num_htree_groups++;
+        huffman_image[i] = *mapped_group;
+      }
+    } else {
+      num_htree_groups = num_htree_groups_max;
+    }
  }

  if (br->eos_) goto Error;
@@ -403,83 +432,100 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
    }
  }

+  code_lengths = (int*)WebPSafeCalloc((uint64_t)max_alphabet_size,
+                                      sizeof(*code_lengths));
  huffman_tables = (HuffmanCode*)WebPSafeMalloc(num_htree_groups * table_size,
                                                sizeof(*huffman_tables));
  htree_groups = VP8LHtreeGroupsNew(num_htree_groups);
-  code_lengths = (int*)WebPSafeCalloc((uint64_t)max_alphabet_size,
-                                      sizeof(*code_lengths));

  if (htree_groups == NULL || code_lengths == NULL || huffman_tables == NULL) {
    dec->status_ = VP8_STATUS_OUT_OF_MEMORY;
    goto Error;
  }

-  next = huffman_tables;
-  for (i = 0; i < num_htree_groups; ++i) {
-    HTreeGroup* const htree_group = &htree_groups[i];
-    HuffmanCode** const htrees = htree_group->htrees;
-    int size;
-    int total_size = 0;
-    int is_trivial_literal = 1;
-    int max_bits = 0;
-    for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; ++j) {
-      int alphabet_size = kAlphabetSize[j];
-      htrees[j] = next;
-      if (j == 0 && color_cache_bits > 0) {
-        alphabet_size += 1 << color_cache_bits;
-      }
-      size = ReadHuffmanCode(alphabet_size, dec, code_lengths, next);
-      if (size == 0) {
-        goto Error;
-      }
-      if (is_trivial_literal && kLiteralMap[j] == 1) {
-        is_trivial_literal = (next->bits == 0);
-      }
-      total_size += next->bits;
-      next += size;
-      if (j <= ALPHA) {
-        int local_max_bits = code_lengths[0];
-        int k;
-        for (k = 1; k < alphabet_size; ++k) {
-          if (code_lengths[k] > local_max_bits) {
-            local_max_bits = code_lengths[k];
-          }
+  huffman_table = huffman_tables;
+  for (i = 0; i < num_htree_groups_max; ++i) {
+    // If the index "i" is unused in the Huffman image, just make sure the
+    // coefficients are valid but do not store them.
+    if (mapping != NULL && mapping[i] == -1) {
+      for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; ++j) {
+        int alphabet_size = kAlphabetSize[j];
+        if (j == 0 && color_cache_bits > 0) {
+          alphabet_size += (1 << color_cache_bits);
+        }
+        // Passing in NULL so that nothing gets filled.
+        if (!ReadHuffmanCode(alphabet_size, dec, code_lengths, NULL)) {
+          goto Error;
        }
-        max_bits += local_max_bits;
      }
-    }
-    htree_group->is_trivial_literal = is_trivial_literal;
-    htree_group->is_trivial_code = 0;
-    if (is_trivial_literal) {
-      const int red = htrees[RED][0].value;
-      const int blue = htrees[BLUE][0].value;
-      const int alpha = htrees[ALPHA][0].value;
-      htree_group->literal_arb =
-          ((uint32_t)alpha << 24) | (red << 16) | blue;
-      if (total_size == 0 && htrees[GREEN][0].value < NUM_LITERAL_CODES) {
-        htree_group->is_trivial_code = 1;
-        htree_group->literal_arb |= htrees[GREEN][0].value << 8;
+    } else {
+      HTreeGroup* const htree_group =
+          &htree_groups[(mapping == NULL) ? i : mapping[i]];
+      HuffmanCode** const htrees = htree_group->htrees;
+      int size;
+      int total_size = 0;
+      int is_trivial_literal = 1;
+      int max_bits = 0;
+      for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; ++j) {
+        int alphabet_size = kAlphabetSize[j];
+        htrees[j] = huffman_table;
+        if (j == 0 && color_cache_bits > 0) {
+          alphabet_size += (1 << color_cache_bits);
+        }
+        size = ReadHuffmanCode(alphabet_size, dec, code_lengths, huffman_table);
+        if (size == 0) {
+          goto Error;
+        }
+        if (is_trivial_literal && kLiteralMap[j] == 1) {
+          is_trivial_literal = (huffman_table->bits == 0);
+        }
+        total_size += huffman_table->bits;
+        huffman_table += size;
+        if (j <= ALPHA) {
+          int local_max_bits = code_lengths[0];
+          int k;
+          for (k = 1; k < alphabet_size; ++k) {
+            if (code_lengths[k] > local_max_bits) {
+              local_max_bits = code_lengths[k];
+            }
+          }
+          max_bits += local_max_bits;
+        }
      }
+      htree_group->is_trivial_literal = is_trivial_literal;
+      htree_group->is_trivial_code = 0;
+      if (is_trivial_literal) {
+        const int red = htrees[RED][0].value;
+        const int blue = htrees[BLUE][0].value;
+        const int alpha = htrees[ALPHA][0].value;
+        htree_group->literal_arb = ((uint32_t)alpha << 24) | (red << 16) | blue;
+        if (total_size == 0 && htrees[GREEN][0].value < NUM_LITERAL_CODES) {
+          htree_group->is_trivial_code = 1;
+          htree_group->literal_arb |= htrees[GREEN][0].value << 8;
+        }
+      }
+      htree_group->use_packed_table =
+          !htree_group->is_trivial_code && (max_bits < HUFFMAN_PACKED_BITS);
+      if (htree_group->use_packed_table) BuildPackedTable(htree_group);
    }
-    htree_group->use_packed_table = !htree_group->is_trivial_code &&
-                                    (max_bits < HUFFMAN_PACKED_BITS);
-    if (htree_group->use_packed_table) BuildPackedTable(htree_group);
  }
-  WebPSafeFree(code_lengths);
+  ok = 1;

-  // All OK. Finalize pointers and return.
+  // All OK. Finalize pointers.
  hdr->huffman_image_ = huffman_image;
  hdr->num_htree_groups_ = num_htree_groups;
  hdr->htree_groups_ = htree_groups;
  hdr->huffman_tables_ = huffman_tables;
-  return 1;

 Error:
  WebPSafeFree(code_lengths);
-  WebPSafeFree(huffman_image);
-  WebPSafeFree(huffman_tables);
-  VP8LHtreeGroupsFree(htree_groups);
-  return 0;
+  WebPSafeFree(mapping);
+  if (!ok) {
+    WebPSafeFree(huffman_image);
+    WebPSafeFree(huffman_tables);
+    VP8LHtreeGroupsFree(htree_groups);
+  }
+  return ok;
 }

 //------------------------------------------------------------------------------
@@ -708,11 +754,11 @@ static WEBP_INLINE HTreeGroup* GetHtreeGroupForPos(VP8LMetadata* const hdr,

 typedef void (*ProcessRowsFunc)(VP8LDecoder* const dec, int row);

-static void ApplyInverseTransforms(VP8LDecoder* const dec, int num_rows,
+static void ApplyInverseTransforms(VP8LDecoder* const dec,
+                                   int start_row, int num_rows,
                                   const uint32_t* const rows) {
  int n = dec->next_transform_;
  const int cache_pixs = dec->width_ * num_rows;
-  const int start_row = dec->last_row_;
  const int end_row = start_row + num_rows;
  const uint32_t* rows_in = rows;
  uint32_t* const rows_out = dec->argb_cache_;
@@ -743,8 +789,7 @@ static void ProcessRows(VP8LDecoder* const dec, int row) {
    VP8Io* const io = dec->io_;
    uint8_t* rows_data = (uint8_t*)dec->argb_cache_;
    const int in_stride = io->width * sizeof(uint32_t);  // in unit of RGBA
-
-    ApplyInverseTransforms(dec, num_rows, rows);
+    ApplyInverseTransforms(dec, dec->last_row_, num_rows, rows);
    if (!SetCropWindow(io, dec->last_row_, row, &rows_data, in_stride)) {
      // Nothing to output (this time).
    } else {
@@ -884,7 +929,11 @@ static WEBP_INLINE void CopyBlock8b(uint8_t* const dst, int dist, int length) {
 #endif
        break;
      case 2:
+#if !defined(WORDS_BIGENDIAN)
        memcpy(&pattern, src, sizeof(uint16_t));
+#else
+        pattern = ((uint32_t)src[0] << 8) | src[1];
+#endif
 #if defined(__arm__) || defined(_M_ARM)
        pattern |= pattern << 16;
 #elif defined(WEBP_USE_MIPS_DSP_R2)
@@ -1143,6 +1192,7 @@ static int DecodeImageData(VP8LDecoder* const dec, uint32_t* const data,
      VP8LFillBitWindow(br);
      dist_code = GetCopyDistance(dist_symbol, br);
      dist = PlaneCodeToDistance(width, dist_code);
+
      if (VP8LIsEndOfStream(br)) break;
      if (src - data < (ptrdiff_t)dist || src_end - src < (ptrdiff_t)length) {
        goto Error;
@@ -1503,7 +1553,7 @@ static void ExtractAlphaRows(VP8LDecoder* const dec, int last_row) {
    const int cache_pixs = width * num_rows_to_process;
    uint8_t* const dst = output + width * cur_row;
    const uint32_t* const src = dec->argb_cache_;
-    ApplyInverseTransforms(dec, num_rows_to_process, in);
+    ApplyInverseTransforms(dec, cur_row, num_rows_to_process, in);
    WebPExtractGreen(src, dst, cache_pixs);
    AlphaApplyFilter(alph_dec,
                     cur_row, cur_row + num_rows_to_process, dst, width);
@@ -1523,7 +1573,6 @@ int VP8LDecodeAlphaHeader(ALPHDecoder* const alph_dec,
  if (dec == NULL) return 0;

  assert(alph_dec != NULL);
-  alph_dec->vp8l_dec_ = dec;

  dec->width_ = alph_dec->width_;
  dec->height_ = alph_dec->height_;
@@ -1555,11 +1604,12 @@ int VP8LDecodeAlphaHeader(ALPHDecoder* const alph_dec,

  if (!ok) goto Err;

+  // Only set here, once we are sure it is valid (to avoid thread races).
+  alph_dec->vp8l_dec_ = dec;
  return 1;

 Err:
-  VP8LDelete(alph_dec->vp8l_dec_);
-  alph_dec->vp8l_dec_ = NULL;
+  VP8LDelete(dec);
  return 0;
 }

@@ -37,7 +37,7 @@ struct VP8LTransform {
  int                    bits_;   // subsampling bits defining transform window.
  int                    xsize_;  // transform window X index.
  int                    ysize_;  // transform window Y index.
-  uint32_t              *data_;   // transform data.
+  uint32_t*              data_;   // transform data.
 };

 typedef struct {
@@ -48,23 +48,23 @@ typedef struct {
  int             huffman_mask_;
  int             huffman_subsample_bits_;
  int             huffman_xsize_;
-  uint32_t       *huffman_image_;
+  uint32_t*       huffman_image_;
  int             num_htree_groups_;
-  HTreeGroup     *htree_groups_;
-  HuffmanCode    *huffman_tables_;
+  HTreeGroup*     htree_groups_;
+  HuffmanCode*    huffman_tables_;
 } VP8LMetadata;

 typedef struct VP8LDecoder VP8LDecoder;
 struct VP8LDecoder {
  VP8StatusCode    status_;
  VP8LDecodeState  state_;
-  VP8Io           *io_;
+  VP8Io*           io_;

-  const WebPDecBuffer *output_;    // shortcut to io->opaque->output
+  const WebPDecBuffer* output_;    // shortcut to io->opaque->output

-  uint32_t        *pixels_;        // Internal data: either uint8_t* for alpha
+  uint32_t*        pixels_;        // Internal data: either uint8_t* for alpha
                                   // or uint32_t* for BGRA.
-  uint32_t        *argb_cache_;    // Scratch buffer for temporary BGRA storage.
+  uint32_t*        argb_cache_;    // Scratch buffer for temporary BGRA storage.

  VP8LBitReader    br_;
  int              incremental_;   // if true, incremental decoding is expected
@@ -86,8 +86,8 @@ struct VP8LDecoder {
  // or'd bitset storing the transforms types.
  uint32_t         transforms_seen_;

-  uint8_t         *rescaler_memory;  // Working memory for rescaling work.
-  WebPRescaler    *rescaler;         // Common rescaler for all channels.
+  uint8_t*         rescaler_memory;  // Working memory for rescaling work.
+  WebPRescaler*    rescaler;         // Common rescaler for all channels.
 };

 //------------------------------------------------------------------------------
@@ -132,4 +132,4 @@ void VP8LDelete(VP8LDecoder* const dec);
 }    // extern "C"
 #endif

-#endif  /* WEBP_DEC_VP8LI_DEC_H_ */
+#endif  // WEBP_DEC_VP8LI_DEC_H_
@@ -130,4 +130,4 @@ int WebPAvoidSlowMemory(const WebPDecBuffer* const output,
 }    // extern "C"
 #endif

-#endif  /* WEBP_DEC_WEBPI_DEC_H_ */
+#endif  // WEBP_DEC_WEBPI_DEC_H_
@@ -24,7 +24,7 @@
 #include "src/webp/format_constants.h"

 #define DMUX_MAJ_VERSION 1
-#define DMUX_MIN_VERSION 0
+#define DMUX_MIN_VERSION 1
 #define DMUX_REV_VERSION 0

 typedef struct {
@@ -214,7 +214,7 @@ static void ApplyAlphaMultiply_SSE2(uint8_t* rgba, int alpha_first,
 // Alpha detection

 static int HasAlpha8b_SSE2(const uint8_t* src, int length) {
-  const __m128i all_0xff = _mm_set1_epi8(0xff);
+  const __m128i all_0xff = _mm_set1_epi8((char)0xff);
  int i = 0;
  for (; i + 16 <= length; i += 16) {
    const __m128i v = _mm_loadu_si128((const __m128i*)(src + i));
@@ -228,7 +228,7 @@ static int HasAlpha8b_SSE2(const uint8_t* src, int length) {

 static int HasAlpha32b_SSE2(const uint8_t* src, int length) {
  const __m128i alpha_mask = _mm_set1_epi32(0xff);
-  const __m128i all_0xff = _mm_set1_epi8(0xff);
+  const __m128i all_0xff = _mm_set1_epi8((char)0xff);
  int i = 0;
  // We don't know if we can access the last 3 bytes after the last alpha
  // value 'src[4 * length - 4]' (because we don't know if alpha is the first
@@ -377,6 +377,7 @@ VP8SetResidualCoeffsFunc VP8SetResidualCoeffs;
 extern void VP8EncDspCostInitMIPS32(void);
 extern void VP8EncDspCostInitMIPSdspR2(void);
 extern void VP8EncDspCostInitSSE2(void);
+extern void VP8EncDspCostInitNEON(void);

 WEBP_DSP_INIT_FUNC(VP8EncDspCostInit) {
  VP8GetResidualCost = GetResidualCost_C;
@@ -398,6 +399,11 @@ WEBP_DSP_INIT_FUNC(VP8EncDspCostInit) {
    if (VP8GetCPUInfo(kSSE2)) {
      VP8EncDspCostInitSSE2();
    }
+#endif
+#if defined(WEBP_USE_NEON)
+    if (VP8GetCPUInfo(kNEON)) {
+      VP8EncDspCostInitNEON();
+    }
 #endif
  }
 }
@@ -0,0 +1,122 @@
+// Copyright 2018 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// ARM NEON version of cost functions
+
+#include "src/dsp/dsp.h"
+
+#if defined(WEBP_USE_NEON)
+
+#include "src/dsp/neon.h"
+#include "src/enc/cost_enc.h"
+
+static const uint8_t position[16] = { 1, 2,  3,  4,  5,  6,  7,  8,
+                                      9, 10, 11, 12, 13, 14, 15, 16 };
+
+static void SetResidualCoeffs_NEON(const int16_t* const coeffs,
+                                   VP8Residual* const res) {
+  const int16x8_t minus_one = vdupq_n_s16(-1);
+  const int16x8_t coeffs_0 = vld1q_s16(coeffs);
+  const int16x8_t coeffs_1 = vld1q_s16(coeffs + 8);
+  const uint16x8_t eob_0 = vtstq_s16(coeffs_0, minus_one);
+  const uint16x8_t eob_1 = vtstq_s16(coeffs_1, minus_one);
+  const uint8x16_t eob = vcombine_u8(vqmovn_u16(eob_0), vqmovn_u16(eob_1));
+  const uint8x16_t masked = vandq_u8(eob, vld1q_u8(position));
+
+#ifdef __aarch64__
+  res->last = vmaxvq_u8(masked) - 1;
+#else
+  const uint8x8_t eob_8x8 = vmax_u8(vget_low_u8(masked), vget_high_u8(masked));
+  const uint16x8_t eob_16x8 = vmovl_u8(eob_8x8);
+  const uint16x4_t eob_16x4 =
+      vmax_u16(vget_low_u16(eob_16x8), vget_high_u16(eob_16x8));
+  const uint32x4_t eob_32x4 = vmovl_u16(eob_16x4);
+  uint32x2_t eob_32x2 =
+      vmax_u32(vget_low_u32(eob_32x4), vget_high_u32(eob_32x4));
+  eob_32x2 = vpmax_u32(eob_32x2, eob_32x2);
+
+  vst1_lane_s32(&res->last, vreinterpret_s32_u32(eob_32x2), 0);
+  --res->last;
+#endif  // __aarch64__
+
+  res->coeffs = coeffs;
+}
+
+static int GetResidualCost_NEON(int ctx0, const VP8Residual* const res) {
+  uint8_t levels[16], ctxs[16];
+  uint16_t abs_levels[16];
+  int n = res->first;
+  // should be prob[VP8EncBands[n]], but it's equivalent for n=0 or 1
+  const int p0 = res->prob[n][ctx0][0];
+  CostArrayPtr const costs = res->costs;
+  const uint16_t* t = costs[n][ctx0];
+  // bit_cost(1, p0) is already incorporated in t[] tables, but only if ctx != 0
+  // (as required by the syntax). For ctx0 == 0, we need to add it here or it'll
+  // be missing during the loop.
+  int cost = (ctx0 == 0) ? VP8BitCost(1, p0) : 0;
+
+  if (res->last < 0) {
+    return VP8BitCost(0, p0);
+  }
+
+  {   // precompute clamped levels and contexts, packed to 8b.
+    const uint8x16_t kCst2 = vdupq_n_u8(2);
+    const uint8x16_t kCst67 = vdupq_n_u8(MAX_VARIABLE_LEVEL);
+    const int16x8_t c0 = vld1q_s16(res->coeffs);
+    const int16x8_t c1 = vld1q_s16(res->coeffs + 8);
+    const uint16x8_t E0 = vreinterpretq_u16_s16(vabsq_s16(c0));
+    const uint16x8_t E1 = vreinterpretq_u16_s16(vabsq_s16(c1));
+    const uint8x16_t F = vcombine_u8(vqmovn_u16(E0), vqmovn_u16(E1));
+    const uint8x16_t G = vminq_u8(F, kCst2);   // context = 0,1,2
+    const uint8x16_t H = vminq_u8(F, kCst67);  // clamp_level in [0..67]
+
+    vst1q_u8(ctxs, G);
+    vst1q_u8(levels, H);
+
+    vst1q_u16(abs_levels, E0);
+    vst1q_u16(abs_levels + 8, E1);
+  }
+  for (; n < res->last; ++n) {
+    const int ctx = ctxs[n];
+    const int level = levels[n];
+    const int flevel = abs_levels[n];   // full level
+    cost += VP8LevelFixedCosts[flevel] + t[level];  // simplified VP8LevelCost()
+    t = costs[n + 1][ctx];
+  }
+  // Last coefficient is always non-zero
+  {
+    const int level = levels[n];
+    const int flevel = abs_levels[n];
+    assert(flevel != 0);
+    cost += VP8LevelFixedCosts[flevel] + t[level];
+    if (n < 15) {
+      const int b = VP8EncBands[n + 1];
+      const int ctx = ctxs[n];
+      const int last_p0 = res->prob[b][ctx][0];
+      cost += VP8BitCost(0, last_p0);
+    }
+  }
+  return cost;
+}
+
+//------------------------------------------------------------------------------
+// Entry point
+
+extern void VP8EncDspCostInitNEON(void);
+
+WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspCostInitNEON(void) {
+  VP8SetResidualCoeffs = SetResidualCoeffs_NEON;
+  VP8GetResidualCost = GetResidualCost_NEON;
+}
+
+#else  // !WEBP_USE_NEON
+
+WEBP_DSP_INIT_STUB(VP8EncDspCostInitNEON)
+
+#endif  // WEBP_USE_NEON
@@ -173,8 +173,8 @@ static int AndroidCPUInfo(CPUFeature feature) {
  const AndroidCpuFamily cpu_family = android_getCpuFamily();
  const uint64_t cpu_features = android_getCpuFeatures();
  if (feature == kNEON) {
-    return (cpu_family == ANDROID_CPU_FAMILY_ARM &&
-            0 != (cpu_features & ANDROID_CPU_ARM_FEATURE_NEON));
+    return cpu_family == ANDROID_CPU_FAMILY_ARM &&
+           (cpu_features & ANDROID_CPU_ARM_FEATURE_NEON) != 0;
  }
  return 0;
 }
@@ -1361,7 +1361,8 @@ static void RD4_NEON(uint8_t* dst) {   // Down-right
  const uint32_t J = dst[-1 + 1 * BPS];
  const uint32_t K = dst[-1 + 2 * BPS];
  const uint32_t L = dst[-1 + 3 * BPS];
-  const uint64x1_t LKJI____ = vcreate_u64(L | (K << 8) | (J << 16) | (I << 24));
+  const uint64x1_t LKJI____ =
+      vcreate_u64((uint64_t)L | (K << 8) | (J << 16) | (I << 24));
  const uint64x1_t LKJIXABC = vorr_u64(LKJI____, ____XABC);
  const uint8x8_t KJIXABC_ = vreinterpret_u8_u64(vshr_n_u64(LKJIXABC, 8));
  const uint8x8_t JIXABC__ = vreinterpret_u8_u64(vshr_n_u64(LKJIXABC, 16));
@@ -1427,10 +1428,16 @@ static WEBP_INLINE void DC8_NEON(uint8_t* dst, int do_top, int do_left) {

  if (do_top) {
    const uint8x8_t A = vld1_u8(dst - BPS);  // top row
+#if defined(__aarch64__)
+    const uint16x8_t B = vmovl_u8(A);
+    const uint16_t p2 = vaddvq_u16(B);
+    sum_top = vdupq_n_u16(p2);
+#else
    const uint16x4_t p0 = vpaddl_u8(A);  // cascading summation of the top
    const uint16x4_t p1 = vpadd_u16(p0, p0);
    const uint16x4_t p2 = vpadd_u16(p1, p1);
    sum_top = vcombine_u16(p2, p2);
+#endif
  }

  if (do_left) {
@@ -326,7 +326,7 @@ static WEBP_INLINE void Update2Pixels_SSE2(__m128i* const pi, __m128i* const qi,
  const __m128i a1_lo = _mm_srai_epi16(*a0_lo, 7);
  const __m128i a1_hi = _mm_srai_epi16(*a0_hi, 7);
  const __m128i delta = _mm_packs_epi16(a1_lo, a1_hi);
-  const __m128i sign_bit = _mm_set1_epi8(0x80);
+  const __m128i sign_bit = _mm_set1_epi8((char)0x80);
  *pi = _mm_adds_epi8(*pi, delta);
  *qi = _mm_subs_epi8(*qi, delta);
  FLIP_SIGN_BIT2(*pi, *qi);
@@ -338,9 +338,9 @@ static WEBP_INLINE void NeedsFilter_SSE2(const __m128i* const p1,
                                         const __m128i* const q0,
                                         const __m128i* const q1,
                                         int thresh, __m128i* const mask) {
-  const __m128i m_thresh = _mm_set1_epi8(thresh);
+  const __m128i m_thresh = _mm_set1_epi8((char)thresh);
  const __m128i t1 = MM_ABS(*p1, *q1);        // abs(p1 - q1)
-  const __m128i kFE = _mm_set1_epi8(0xFE);
+  const __m128i kFE = _mm_set1_epi8((char)0xFE);
  const __m128i t2 = _mm_and_si128(t1, kFE);  // set lsb of each byte to zero
  const __m128i t3 = _mm_srli_epi16(t2, 1);   // abs(p1 - q1) / 2

@@ -360,7 +360,7 @@ static WEBP_INLINE void DoFilter2_SSE2(__m128i* const p1, __m128i* const p0,
                                       __m128i* const q0, __m128i* const q1,
                                       int thresh) {
  __m128i a, mask;
-  const __m128i sign_bit = _mm_set1_epi8(0x80);
+  const __m128i sign_bit = _mm_set1_epi8((char)0x80);
  // convert p1/q1 to int8_t (for GetBaseDelta_SSE2)
  const __m128i p1s = _mm_xor_si128(*p1, sign_bit);
  const __m128i q1s = _mm_xor_si128(*q1, sign_bit);
@@ -380,7 +380,7 @@ static WEBP_INLINE void DoFilter4_SSE2(__m128i* const p1, __m128i* const p0,
                                       const __m128i* const mask,
                                       int hev_thresh) {
  const __m128i zero = _mm_setzero_si128();
-  const __m128i sign_bit = _mm_set1_epi8(0x80);
+  const __m128i sign_bit = _mm_set1_epi8((char)0x80);
  const __m128i k64 = _mm_set1_epi8(64);
  const __m128i k3 = _mm_set1_epi8(3);
  const __m128i k4 = _mm_set1_epi8(4);
@@ -427,7 +427,7 @@ static WEBP_INLINE void DoFilter6_SSE2(__m128i* const p2, __m128i* const p1,
                                       const __m128i* const mask,
                                       int hev_thresh) {
  const __m128i zero = _mm_setzero_si128();
-  const __m128i sign_bit = _mm_set1_epi8(0x80);
+  const __m128i sign_bit = _mm_set1_epi8((char)0x80);
  __m128i a, not_hev;

  // compute hev mask
@@ -941,7 +941,7 @@ static void VR4_SSE2(uint8_t* dst) {   // Vertical-Right
  const __m128i ABCD0 = _mm_srli_si128(XABCD, 1);
  const __m128i abcd = _mm_avg_epu8(XABCD, ABCD0);
  const __m128i _XABCD = _mm_slli_si128(XABCD, 1);
-  const __m128i IXABCD = _mm_insert_epi16(_XABCD, I | (X << 8), 0);
+  const __m128i IXABCD = _mm_insert_epi16(_XABCD, (short)(I | (X << 8)), 0);
  const __m128i avg1 = _mm_avg_epu8(IXABCD, ABCD0);
  const __m128i lsb = _mm_and_si128(_mm_xor_si128(IXABCD, ABCD0), one);
  const __m128i avg2 = _mm_subs_epu8(avg1, lsb);
--- a/Show More
+++ b/Show More