core(logger): complete initialization of logger structures

- for using of logging functions from global destructors original commit: 99bc35a883
Disabling dafault NMS in yolo layer
2026-07-22 03:43:03 +04:00 · 2020-06-22 10:49:59 +00:00 · 2020-06-22 10:49:34 +00:00 · 2020-06-17 11:49:26 +03:00 · 2020-06-09 14:13:26 +03:00 · 2020-06-09 14:13:21 +03:00
1850 changed files with 127891 additions and 21589 deletions
@@ -1,23 +1,8 @@
-<!--
-If you have a question rather than reporting a bug please go to http://answers.opencv.org where you get much faster responses.
-If you need further assistance please read [How To Contribute](https://github.com/opencv/opencv/wiki/How_to_contribute).
-
-Please:
-
-* Read the documentation to test with the latest developer build.
-* Check if other person has already created the same issue to avoid duplicates. You can comment on it if there already is an issue.
-* Try to be as detailed as possible in your report.
-* Report only one problem per created issue.
-
-
-This is a template helping you to create an issue which can be processed as quickly as possible. This is the bug reporting section for the OpenCV library.
-->
-
 ##### System information (version)
 <!-- Example
- OpenCV => 3.1
+- OpenCV => 4.2
 - Operating System / Platform => Windows 64 Bit
- Compiler => Visual Studio 2015
+- Compiler => Visual Studio 2017
 -->

 - OpenCV => :grey_question:
@@ -36,3 +21,42 @@ This is a template helping you to create an issue which can be processed as quic
    ```
 or attach as .txt or .zip file
 -->
+
+##### Issue submission checklist
+
+ - [ ] I report the issue, it's not a question
+   <!--
+   OpenCV team works with answers.opencv.org, Stack Overflow and other communities
+   to discuss problems. Tickets with question without real issue statement will be
+   closed.
+   -->
+ - [ ] I checked the problem with documentation, FAQ, open issues,
+       answers.opencv.org, Stack Overflow, etc and have not found solution
+   <!--
+   Places to check:
+   * OpenCV documentation: https://docs.opencv.org
+   * FAQ page: https://github.com/opencv/opencv/wiki/FAQ
+   * OpenCV forum: https://answers.opencv.org
+   * OpenCV issue tracker: https://github.com/opencv/opencv/issues?q=is%3Aissue
+   * Stack Overflow branch: https://stackoverflow.com/questions/tagged/opencv
+   -->
+ - [ ] I updated to latest OpenCV version and the issue is still there
+   <!--
+   master branch for OpenCV 4.x and 3.4 branch for OpenCV 3.x releases.
+   OpenCV team supports only latest release for each branch.
+   The ticket is closed, if the problem is not reproduced with modern version.
+   -->
+ - [ ] There is reproducer code and related data files: videos, images, onnx, etc
+   <!--
+   The best reproducer -- test case for OpenCV that we can add to the library.
+   Recommendations for media files and binary files:
+   * Try to reproduce the issue with images and videos in opencv_extra repository
+     to reduce attachment size
+   * Use PNG for images, if you report some CV related bug, but not image reader
+     issue
+   * Attach the image as archite to the ticket, if you report some reader issue.
+     Image hosting services compress images and it breaks the repro code.
+   * Provide ONNX file for some public model or ONNX file with with random weights,
+     if you report ONNX parsing or handling issue. Architecture details diagram
+     from netron tool can be very useful too. See https://lutzroeder.github.io/netron/
+   -->
@@ -1,9 +1,11 @@
-<!-- Please use this line to close one or multiple issues when this pullrequest gets merged
-You can add another line right under the first one:
-resolves #1234
-resolves #1235
-->
+### Pull Request Readiness Checklist

-### This pullrequest changes
+See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request

-<!-- Please describe what your pullrequest is changing -->
+- [ ] I agree to contribute to the project under OpenCV (BSD) License.
+- [ ] To the best of my knowledge, the proposed patch is not based on a code under GPL or other license that is incompatible with OpenCV
+- [ ] The PR is proposed to proper branch
+- [ ] There is reference to original bug report and related work
+- [ ] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
+      Patch to opencv_extra has the same branch name.
+- [ ] The feature is well documented and sample code can be built with the project CMake
@@ -22,3 +22,6 @@ bin/
 *.log
 *.tlog
 build
+node_modules
+CMakeSettings.json
+xcuserdata/
@@ -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
@@ -60,22 +60,6 @@ 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()
@@ -90,7 +74,11 @@ 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;
@@ -1,9 +1,9 @@
-# Binaries branch name: ffmpeg/master_20190616
-# Binaries were created for OpenCV: 7b099e0fe2d929e55d6705b6ad510c2c9081606b
-ocv_update(FFMPEG_BINARIES_COMMIT "998718df34e35ea0fa429724875fc3900faa266f")
-ocv_update(FFMPEG_FILE_HASH_BIN32 "f03b47fb809edd2e06b6db135cbd3e49")
-ocv_update(FFMPEG_FILE_HASH_BIN64 "5c4571459570c288d874704244c428b5")
-ocv_update(FFMPEG_FILE_HASH_CMAKE "f710891525a04586d565d0e700e62a9c")
+# Binaries branch name: ffmpeg/master_20200609
+# Binaries were created for OpenCV: 5f3012fc9afbffbf53a38f1468523d3454b3e2f6
+ocv_update(FFMPEG_BINARIES_COMMIT "1df9bf0c0c6c2cf225bd3d8e4cf5985198352454")
+ocv_update(FFMPEG_FILE_HASH_BIN32 "854b3460c435d04277e1f1ecc06cb809")
+ocv_update(FFMPEG_FILE_HASH_BIN64 "3a46d6356220796e044817ae3a21cc31")
+ocv_update(FFMPEG_FILE_HASH_CMAKE "ad57c038ba34b868277ccbe6dd0f9602")

 function(download_win_ffmpeg script_var)
  set(${script_var} "" PARENT_SCOPE)
@@ -2,37 +2,32 @@ function(download_ippicv root_var)
  set(${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()
@@ -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")
@@ -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) { \
@@ -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;
@@ -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 = ${DirDefs}")
-endif()
-
 # ----------------------------------------------------------------------------------
 #         Define the library target:
 # ----------------------------------------------------------------------------------
@@ -46,8 +46,19 @@ if(";${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()
@@ -0,0 +1,808 @@
+
+/* filter_msa_intrinsics.c - MSA optimised filter functions
+ *
+ * Copyright (c) 2018 Cosmin Truta
+ * Copyright (c) 2016 Glenn Randers-Pehrson
+ * Written by Mandar Sahastrabuddhe, August 2016.
+ *
+ * This code is released under the libpng license.
+ * For conditions of distribution and use, see the disclaimer
+ * and license in png.h
+ */
+
+#include <stdio.h>
+#include <stdint.h>
+#include "../pngpriv.h"
+
+#ifdef PNG_READ_SUPPORTED
+
+/* This code requires -mfpu=msa on the command line: */
+#if PNG_MIPS_MSA_IMPLEMENTATION == 1 /* intrinsics code from pngpriv.h */
+
+#include <msa.h>
+
+/* libpng row pointers are not necessarily aligned to any particular boundary,
+ * however this code will only work with appropriate alignment. mips/mips_init.c
+ * checks for this (and will not compile unless it is done). This code uses
+ * variants of png_aligncast to avoid compiler warnings.
+ */
+#define png_ptr(type,pointer) png_aligncast(type *,pointer)
+#define png_ptrc(type,pointer) png_aligncastconst(const type *,pointer)
+
+/* The following relies on a variable 'temp_pointer' being declared with type
+ * 'type'.  This is written this way just to hide the GCC strict aliasing
+ * warning; note that the code is safe because there never is an alias between
+ * the input and output pointers.
+ */
+#define png_ldr(type,pointer)\
+   (temp_pointer = png_ptr(type,pointer), *temp_pointer)
+
+#if PNG_MIPS_MSA_OPT > 0
+
+#ifdef CLANG_BUILD
+   #define MSA_SRLI_B(a, b)   __msa_srli_b((v16i8) a, b)
+
+   #define LW(psrc)                              \
+   ( {                                           \
+       uint8_t *psrc_lw_m = (uint8_t *) (psrc);  \
+       uint32_t val_m;                           \
+                                                 \
+       asm volatile (                            \
+           "lw  %[val_m],  %[psrc_lw_m]  \n\t"   \
+                                                 \
+           : [val_m] "=r" (val_m)                \
+           : [psrc_lw_m] "m" (*psrc_lw_m)        \
+       );                                        \
+                                                 \
+       val_m;                                    \
+   } )
+
+   #define SH(val, pdst)                         \
+   {                                             \
+       uint8_t *pdst_sh_m = (uint8_t *) (pdst);  \
+       uint16_t val_m = (val);                   \
+                                                 \
+       asm volatile (                            \
+           "sh  %[val_m],  %[pdst_sh_m]  \n\t"   \
+                                                 \
+           : [pdst_sh_m] "=m" (*pdst_sh_m)       \
+           : [val_m] "r" (val_m)                 \
+       );                                        \
+   }
+
+   #define SW(val, pdst)                         \
+   {                                             \
+       uint8_t *pdst_sw_m = (uint8_t *) (pdst);  \
+       uint32_t val_m = (val);                   \
+                                                 \
+       asm volatile (                            \
+           "sw  %[val_m],  %[pdst_sw_m]  \n\t"   \
+                                                 \
+           : [pdst_sw_m] "=m" (*pdst_sw_m)       \
+           : [val_m] "r" (val_m)                 \
+       );                                        \
+   }
+
+       #if (__mips == 64)
+        #define SD(val, pdst)                         \
+        {                                             \
+            uint8_t *pdst_sd_m = (uint8_t *) (pdst);  \
+            uint64_t val_m = (val);                   \
+                                                      \
+            asm volatile (                            \
+                "sd  %[val_m],  %[pdst_sd_m]  \n\t"   \
+                                                      \
+                : [pdst_sd_m] "=m" (*pdst_sd_m)       \
+                : [val_m] "r" (val_m)                 \
+            );                                        \
+        }
+    #else
+        #define SD(val, pdst)                                          \
+        {                                                              \
+            uint8_t *pdst_sd_m = (uint8_t *) (pdst);                   \
+            uint32_t val0_m, val1_m;                                   \
+                                                                       \
+            val0_m = (uint32_t) ((val) & 0x00000000FFFFFFFF);          \
+            val1_m = (uint32_t) (((val) >> 32) & 0x00000000FFFFFFFF);  \
+                                                                       \
+            SW(val0_m, pdst_sd_m);                                     \
+            SW(val1_m, pdst_sd_m + 4);                                 \
+        }
+    #endif
+#else
+   #define MSA_SRLI_B(a, b)   (a >> b)
+
+#if (__mips_isa_rev >= 6)
+   #define LW(psrc)                              \
+   ( {                                           \
+       uint8_t *psrc_lw_m = (uint8_t *) (psrc);  \
+       uint32_t val_m;                           \
+                                                 \
+       asm volatile (                            \
+           "lw  %[val_m],  %[psrc_lw_m]  \n\t"   \
+                                                 \
+           : [val_m] "=r" (val_m)                \
+           : [psrc_lw_m] "m" (*psrc_lw_m)        \
+       );                                        \
+                                                 \
+       val_m;                                    \
+   } )
+
+   #define SH(val, pdst)                         \
+   {                                             \
+       uint8_t *pdst_sh_m = (uint8_t *) (pdst);  \
+       uint16_t val_m = (val);                   \
+                                                 \
+       asm volatile (                            \
+           "sh  %[val_m],  %[pdst_sh_m]  \n\t"   \
+                                                 \
+           : [pdst_sh_m] "=m" (*pdst_sh_m)       \
+           : [val_m] "r" (val_m)                 \
+       );                                        \
+   }
+
+   #define SW(val, pdst)                         \
+   {                                             \
+       uint8_t *pdst_sw_m = (uint8_t *) (pdst);  \
+       uint32_t val_m = (val);                   \
+                                                 \
+       asm volatile (                            \
+           "sw  %[val_m],  %[pdst_sw_m]  \n\t"   \
+                                                 \
+           : [pdst_sw_m] "=m" (*pdst_sw_m)       \
+           : [val_m] "r" (val_m)                 \
+       );                                        \
+   }
+
+   #if (__mips == 64)
+        #define SD(val, pdst)                         \
+        {                                             \
+            uint8_t *pdst_sd_m = (uint8_t *) (pdst);  \
+            uint64_t val_m = (val);                   \
+                                                      \
+            asm volatile (                            \
+                "sd  %[val_m],  %[pdst_sd_m]  \n\t"   \
+                                                      \
+                : [pdst_sd_m] "=m" (*pdst_sd_m)       \
+                : [val_m] "r" (val_m)                 \
+            );                                        \
+        }
+    #else
+        #define SD(val, pdst)                                          \
+        {                                                              \
+            uint8_t *pdst_sd_m = (uint8_t *) (pdst);                   \
+            uint32_t val0_m, val1_m;                                   \
+                                                                       \
+            val0_m = (uint32_t) ((val) & 0x00000000FFFFFFFF);          \
+            val1_m = (uint32_t) (((val) >> 32) & 0x00000000FFFFFFFF);  \
+                                                                       \
+            SW(val0_m, pdst_sd_m);                                     \
+            SW(val1_m, pdst_sd_m + 4);                                 \
+        }
+    #endif
+#else  // !(__mips_isa_rev >= 6)
+   #define LW(psrc)                              \
+   ( {                                           \
+       uint8_t *psrc_lw_m = (uint8_t *) (psrc);  \
+       uint32_t val_m;                           \
+                                                 \
+       asm volatile (                            \
+           "ulw  %[val_m],  %[psrc_lw_m]  \n\t"  \
+                                                 \
+           : [val_m] "=r" (val_m)                \
+           : [psrc_lw_m] "m" (*psrc_lw_m)        \
+       );                                        \
+                                                 \
+       val_m;                                    \
+   } )
+
+   #define SH(val, pdst)                         \
+   {                                             \
+       uint8_t *pdst_sh_m = (uint8_t *) (pdst);  \
+       uint16_t val_m = (val);                   \
+                                                 \
+       asm volatile (                            \
+           "ush  %[val_m],  %[pdst_sh_m]  \n\t"  \
+                                                 \
+           : [pdst_sh_m] "=m" (*pdst_sh_m)       \
+           : [val_m] "r" (val_m)                 \
+       );                                        \
+   }
+
+   #define SW(val, pdst)                         \
+   {                                             \
+       uint8_t *pdst_sw_m = (uint8_t *) (pdst);  \
+       uint32_t val_m = (val);                   \
+                                                 \
+       asm volatile (                            \
+           "usw  %[val_m],  %[pdst_sw_m]  \n\t"  \
+                                                 \
+           : [pdst_sw_m] "=m" (*pdst_sw_m)       \
+           : [val_m] "r" (val_m)                 \
+       );                                        \
+   }
+
+   #define SD(val, pdst)                                          \
+    {                                                              \
+        uint8_t *pdst_sd_m = (uint8_t *) (pdst);                   \
+        uint32_t val0_m, val1_m;                                   \
+                                                                   \
+        val0_m = (uint32_t) ((val) & 0x00000000FFFFFFFF);          \
+        val1_m = (uint32_t) (((val) >> 32) & 0x00000000FFFFFFFF);  \
+                                                                   \
+        SW(val0_m, pdst_sd_m);                                     \
+        SW(val1_m, pdst_sd_m + 4);                                 \
+    }
+
+    #define SW_ZERO(pdst)                      \
+    {                                          \
+        uint8_t *pdst_m = (uint8_t *) (pdst);  \
+                                               \
+        asm volatile (                         \
+            "usw  $0,  %[pdst_m]  \n\t"        \
+                                               \
+            : [pdst_m] "=m" (*pdst_m)          \
+            :                                  \
+        );                                     \
+    }
+#endif  // (__mips_isa_rev >= 6)
+#endif
+
+#define LD_B(RTYPE, psrc) *((RTYPE *) (psrc))
+#define LD_UB(...) LD_B(v16u8, __VA_ARGS__)
+#define LD_B2(RTYPE, psrc, stride, out0, out1)  \
+{                                               \
+    out0 = LD_B(RTYPE, (psrc));                 \
+    out1 = LD_B(RTYPE, (psrc) + stride);        \
+}
+#define LD_UB2(...) LD_B2(v16u8, __VA_ARGS__)
+#define LD_B4(RTYPE, psrc, stride, out0, out1, out2, out3)   \
+{                                                            \
+    LD_B2(RTYPE, (psrc), stride, out0, out1);                \
+    LD_B2(RTYPE, (psrc) + 2 * stride , stride, out2, out3);  \
+}
+#define LD_UB4(...) LD_B4(v16u8, __VA_ARGS__)
+
+#define ST_B(RTYPE, in, pdst) *((RTYPE *) (pdst)) = (in)
+#define ST_UB(...) ST_B(v16u8, __VA_ARGS__)
+#define ST_B2(RTYPE, in0, in1, pdst, stride)  \
+{                                             \
+    ST_B(RTYPE, in0, (pdst));                 \
+    ST_B(RTYPE, in1, (pdst) + stride);        \
+}
+#define ST_UB2(...) ST_B2(v16u8, __VA_ARGS__)
+#define ST_B4(RTYPE, in0, in1, in2, in3, pdst, stride)    \
+{                                                         \
+    ST_B2(RTYPE, in0, in1, (pdst), stride);               \
+    ST_B2(RTYPE, in2, in3, (pdst) + 2 * stride, stride);  \
+}
+#define ST_UB4(...) ST_B4(v16u8, __VA_ARGS__)
+
+#define ADD2(in0, in1, in2, in3, out0, out1)  \
+{                                             \
+    out0 = in0 + in1;                         \
+    out1 = in2 + in3;                         \
+}
+#define ADD3(in0, in1, in2, in3, in4, in5,  \
+             out0, out1, out2)              \
+{                                           \
+    ADD2(in0, in1, in2, in3, out0, out1);   \
+    out2 = in4 + in5;                       \
+}
+#define ADD4(in0, in1, in2, in3, in4, in5, in6, in7,  \
+             out0, out1, out2, out3)                  \
+{                                                     \
+    ADD2(in0, in1, in2, in3, out0, out1);             \
+    ADD2(in4, in5, in6, in7, out2, out3);             \
+}
+
+#define ILVR_B2(RTYPE, in0, in1, in2, in3, out0, out1)      \
+{                                                           \
+    out0 = (RTYPE) __msa_ilvr_b((v16i8) in0, (v16i8) in1);  \
+    out1 = (RTYPE) __msa_ilvr_b((v16i8) in2, (v16i8) in3);  \
+}
+#define ILVR_B2_SH(...) ILVR_B2(v8i16, __VA_ARGS__)
+
+#define HSUB_UB2(RTYPE, in0, in1, out0, out1)                 \
+{                                                             \
+    out0 = (RTYPE) __msa_hsub_u_h((v16u8) in0, (v16u8) in0);  \
+    out1 = (RTYPE) __msa_hsub_u_h((v16u8) in1, (v16u8) in1);  \
+}
+#define HSUB_UB2_SH(...) HSUB_UB2(v8i16, __VA_ARGS__)
+
+#define SLDI_B2_0(RTYPE, in0, in1, out0, out1, slide_val)                 \
+{                                                                         \
+    v16i8 zero_m = { 0 };                                                 \
+    out0 = (RTYPE) __msa_sldi_b((v16i8) zero_m, (v16i8) in0, slide_val);  \
+    out1 = (RTYPE) __msa_sldi_b((v16i8) zero_m, (v16i8) in1, slide_val);  \
+}
+#define SLDI_B2_0_UB(...) SLDI_B2_0(v16u8, __VA_ARGS__)
+
+#define SLDI_B3_0(RTYPE, in0, in1, in2, out0, out1, out2,  slide_val)     \
+{                                                                         \
+    v16i8 zero_m = { 0 };                                                 \
+    SLDI_B2_0(RTYPE, in0, in1, out0, out1, slide_val);                    \
+    out2 = (RTYPE) __msa_sldi_b((v16i8) zero_m, (v16i8) in2, slide_val);  \
+}
+#define SLDI_B3_0_UB(...) SLDI_B3_0(v16u8, __VA_ARGS__)
+
+#define ILVEV_W2(RTYPE, in0, in1, in2, in3, out0, out1)      \
+{                                                            \
+    out0 = (RTYPE) __msa_ilvev_w((v4i32) in1, (v4i32) in0);  \
+    out1 = (RTYPE) __msa_ilvev_w((v4i32) in3, (v4i32) in2);  \
+}
+#define ILVEV_W2_UB(...) ILVEV_W2(v16u8, __VA_ARGS__)
+
+#define ADD_ABS_H3(RTYPE, in0, in1, in2, out0, out1, out2)  \
+{                                                           \
+    RTYPE zero = {0};                                       \
+                                                            \
+    out0 = __msa_add_a_h((v8i16) zero, in0);                \
+    out1 = __msa_add_a_h((v8i16) zero, in1);                \
+    out2 = __msa_add_a_h((v8i16) zero, in2);                \
+}
+#define ADD_ABS_H3_SH(...) ADD_ABS_H3(v8i16, __VA_ARGS__)
+
+#define VSHF_B2(RTYPE, in0, in1, in2, in3, mask0, mask1, out0, out1)       \
+{                                                                          \
+    out0 = (RTYPE) __msa_vshf_b((v16i8) mask0, (v16i8) in1, (v16i8) in0);  \
+    out1 = (RTYPE) __msa_vshf_b((v16i8) mask1, (v16i8) in3, (v16i8) in2);  \
+}
+#define VSHF_B2_UB(...) VSHF_B2(v16u8, __VA_ARGS__)
+
+#define CMP_AND_SELECT(inp0, inp1, inp2, inp3, inp4, inp5, out0)              \
+{                                                                             \
+   v8i16 _sel_h0, _sel_h1;                                                    \
+   v16u8 _sel_b0, _sel_b1;                                                    \
+   _sel_h0 = (v8i16) __msa_clt_u_h((v8u16) inp1, (v8u16) inp0);               \
+   _sel_b0 = (v16u8) __msa_pckev_b((v16i8) _sel_h0, (v16i8) _sel_h0);         \
+   inp0 = (v8i16) __msa_bmnz_v((v16u8) inp0, (v16u8) inp1, (v16u8) _sel_h0);  \
+   inp4 = (v16u8) __msa_bmnz_v(inp3, inp4, _sel_b0);                          \
+   _sel_h1 = (v8i16) __msa_clt_u_h((v8u16) inp2, (v8u16) inp0);               \
+   _sel_b1 = (v16u8) __msa_pckev_b((v16i8) _sel_h1, (v16i8) _sel_h1);         \
+   inp4 = (v16u8) __msa_bmnz_v(inp4, inp5, _sel_b1);                          \
+   out0 += inp4;                                                              \
+}
+
+void png_read_filter_row_up_msa(png_row_infop row_info, png_bytep row,
+                                png_const_bytep prev_row)
+{
+   size_t i, cnt, cnt16, cnt32;
+   size_t istop = row_info->rowbytes;
+   png_bytep rp = row;
+   png_const_bytep pp = prev_row;
+   v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
+
+   for (i = 0; i < (istop >> 6); i++)
+   {
+      LD_UB4(rp, 16, src0, src1, src2, src3);
+      LD_UB4(pp, 16, src4, src5, src6, src7);
+      pp += 64;
+
+	  ADD4(src0, src4, src1, src5, src2, src6, src3, src7,
+	       src0, src1, src2, src3);
+
+      ST_UB4(src0, src1, src2, src3, rp, 16);
+      rp += 64;
+   }
+
+   if (istop & 0x3F)
+   {
+      cnt32 = istop & 0x20;
+      cnt16 = istop & 0x10;
+      cnt = istop & 0xF;
+
+      if(cnt32)
+      {
+         if (cnt16 && cnt)
+         {
+            LD_UB4(rp, 16, src0, src1, src2, src3);
+            LD_UB4(pp, 16, src4, src5, src6, src7);
+
+            ADD4(src0, src4, src1, src5, src2, src6, src3, src7,
+	             src0, src1, src2, src3);
+
+            ST_UB4(src0, src1, src2, src3, rp, 16);
+            rp += 64;
+         }
+         else if (cnt16 || cnt)
+         {
+            LD_UB2(rp, 16, src0, src1);
+            LD_UB2(pp, 16, src4, src5);
+            pp += 32;
+            src2 = LD_UB(rp + 32);
+            src6 = LD_UB(pp);
+
+            ADD3(src0, src4, src1, src5, src2, src6, src0, src1, src2);
+
+            ST_UB2(src0, src1, rp, 16);
+            rp += 32;
+            ST_UB(src2, rp);
+            rp += 16;
+         }
+         else
+         {
+            LD_UB2(rp, 16, src0, src1);
+            LD_UB2(pp, 16, src4, src5);
+
+			ADD2(src0, src4, src1, src5, src0, src1);
+
+            ST_UB2(src0, src1, rp, 16);
+            rp += 32;
+         }
+      }
+      else if (cnt16 && cnt)
+      {
+         LD_UB2(rp, 16, src0, src1);
+         LD_UB2(pp, 16, src4, src5);
+
+         ADD2(src0, src4, src1, src5, src0, src1);
+
+         ST_UB2(src0, src1, rp, 16);
+         rp += 32;
+      }
+      else if (cnt16 || cnt)
+      {
+         src0 = LD_UB(rp);
+         src4 = LD_UB(pp);
+         pp += 16;
+
+         src0 += src4;
+
+         ST_UB(src0, rp);
+         rp += 16;
+      }
+   }
+}
+
+void png_read_filter_row_sub4_msa(png_row_infop row_info, png_bytep row,
+                                  png_const_bytep prev_row)
+{
+   size_t count;
+   size_t istop = row_info->rowbytes;
+   png_bytep src = row;
+   png_bytep nxt = row + 4;
+   int32_t inp0;
+   v16u8 src0, src1, src2, src3, src4;
+   v16u8 dst0, dst1;
+   v16u8 zero = { 0 };
+
+   istop -= 4;
+
+   inp0 = LW(src);
+   src += 4;
+   src0 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp0);
+
+   for (count = 0; count < istop; count += 16)
+   {
+      src1 = LD_UB(src);
+      src += 16;
+
+      src2 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) src1, 4);
+      src3 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) src1, 8);
+      src4 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) src1, 12);
+      src1 += src0;
+      src2 += src1;
+      src3 += src2;
+      src4 += src3;
+      src0 = src4;
+      ILVEV_W2_UB(src1, src2, src3, src4, dst0, dst1);
+      dst0 = (v16u8) __msa_pckev_d((v2i64) dst1, (v2i64) dst0);
+
+      ST_UB(dst0, nxt);
+      nxt += 16;
+   }
+}
+
+void png_read_filter_row_sub3_msa(png_row_infop row_info, png_bytep row,
+                                  png_const_bytep prev_row)
+{
+   size_t count;
+   size_t istop = row_info->rowbytes;
+   png_bytep src = row;
+   png_bytep nxt = row + 3;
+   int64_t out0;
+   int32_t inp0, out1;
+   v16u8 src0, src1, src2, src3, src4, dst0, dst1;
+   v16u8 zero = { 0 };
+   v16i8 mask0 = { 0, 1, 2, 16, 17, 18, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+   v16i8 mask1 = { 0, 1, 2, 3, 4, 5, 16, 17, 18, 19, 20, 21, 0, 0, 0, 0 };
+
+   istop -= 3;
+
+   inp0 = LW(src);
+   src += 3;
+   src0 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp0);
+
+   for (count = 0; count < istop; count += 12)
+   {
+      src1 = LD_UB(src);
+      src += 12;
+
+      src2 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) src1, 3);
+      src3 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) src1, 6);
+      src4 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) src1, 9);
+      src1 += src0;
+      src2 += src1;
+      src3 += src2;
+      src4 += src3;
+      src0 = src4;
+      VSHF_B2_UB(src1, src2, src3, src4, mask0, mask0, dst0, dst1);
+      dst0 = (v16u8) __msa_vshf_b(mask1, (v16i8) dst1, (v16i8) dst0);
+      out0 = __msa_copy_s_d((v2i64) dst0, 0);
+      out1 = __msa_copy_s_w((v4i32) dst0, 2);
+
+      SD(out0, nxt);
+      nxt += 8;
+      SW(out1, nxt);
+      nxt += 4;
+   }
+}
+
+void png_read_filter_row_avg4_msa(png_row_infop row_info, png_bytep row,
+                                  png_const_bytep prev_row)
+{
+   size_t i;
+   png_bytep src = row;
+   png_bytep nxt = row;
+   png_const_bytep pp = prev_row;
+   size_t istop = row_info->rowbytes - 4;
+   int32_t inp0, inp1, out0;
+   v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, dst0, dst1;
+   v16u8 zero = { 0 };
+
+   inp0 = LW(pp);
+   pp += 4;
+   inp1 = LW(src);
+   src += 4;
+   src0 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp0);
+   src1 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp1);
+   src0 = (v16u8) MSA_SRLI_B(src0, 1);
+   src1 += src0;
+   out0 = __msa_copy_s_w((v4i32) src1, 0);
+   SW(out0, nxt);
+   nxt += 4;
+
+   for (i = 0; i < istop; i += 16)
+   {
+      src2 = LD_UB(pp);
+      pp += 16;
+      src6 = LD_UB(src);
+      src += 16;
+
+      SLDI_B2_0_UB(src2, src6, src3, src7, 4);
+      SLDI_B2_0_UB(src2, src6, src4, src8, 8);
+      SLDI_B2_0_UB(src2, src6, src5, src9, 12);
+      src2 = __msa_ave_u_b(src2, src1);
+      src6 += src2;
+      src3 = __msa_ave_u_b(src3, src6);
+      src7 += src3;
+      src4 = __msa_ave_u_b(src4, src7);
+      src8 += src4;
+      src5 = __msa_ave_u_b(src5, src8);
+      src9 += src5;
+      src1 = src9;
+      ILVEV_W2_UB(src6, src7, src8, src9, dst0, dst1);
+      dst0 = (v16u8) __msa_pckev_d((v2i64) dst1, (v2i64) dst0);
+
+      ST_UB(dst0, nxt);
+      nxt += 16;
+   }
+}
+
+void png_read_filter_row_avg3_msa(png_row_infop row_info, png_bytep row,
+                                  png_const_bytep prev_row)
+{
+   size_t i;
+   png_bytep src = row;
+   png_bytep nxt = row;
+   png_const_bytep pp = prev_row;
+   size_t istop = row_info->rowbytes - 3;
+   int64_t out0;
+   int32_t inp0, inp1, out1;
+   int16_t out2;
+   v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, dst0, dst1;
+   v16u8 zero = { 0 };
+   v16i8 mask0 = { 0, 1, 2, 16, 17, 18, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+   v16i8 mask1 = { 0, 1, 2, 3, 4, 5, 16, 17, 18, 19, 20, 21, 0, 0, 0, 0 };
+
+   inp0 = LW(pp);
+   pp += 3;
+   inp1 = LW(src);
+   src += 3;
+   src0 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp0);
+   src1 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp1);
+   src0 = (v16u8) MSA_SRLI_B(src0, 1);
+   src1 += src0;
+   out2 = __msa_copy_s_h((v8i16) src1, 0);
+   SH(out2, nxt);
+   nxt += 2;
+   nxt[0] = src1[2];
+   nxt++;
+
+   for (i = 0; i < istop; i += 12)
+   {
+      src2 = LD_UB(pp);
+      pp += 12;
+      src6 = LD_UB(src);
+      src += 12;
+
+      SLDI_B2_0_UB(src2, src6, src3, src7, 3);
+      SLDI_B2_0_UB(src2, src6, src4, src8, 6);
+      SLDI_B2_0_UB(src2, src6, src5, src9, 9);
+      src2 = __msa_ave_u_b(src2, src1);
+      src6 += src2;
+      src3 = __msa_ave_u_b(src3, src6);
+      src7 += src3;
+      src4 = __msa_ave_u_b(src4, src7);
+      src8 += src4;
+      src5 = __msa_ave_u_b(src5, src8);
+      src9 += src5;
+      src1 = src9;
+      VSHF_B2_UB(src6, src7, src8, src9, mask0, mask0, dst0, dst1);
+      dst0 = (v16u8) __msa_vshf_b(mask1, (v16i8) dst1, (v16i8) dst0);
+      out0 = __msa_copy_s_d((v2i64) dst0, 0);
+      out1 = __msa_copy_s_w((v4i32) dst0, 2);
+
+      SD(out0, nxt);
+      nxt += 8;
+      SW(out1, nxt);
+      nxt += 4;
+   }
+}
+
+void png_read_filter_row_paeth4_msa(png_row_infop row_info,
+                                    png_bytep row,
+                                    png_const_bytep prev_row)
+{
+   int32_t count, rp_end;
+   png_bytep nxt;
+   png_const_bytep prev_nxt;
+   int32_t inp0, inp1, res0;
+   v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9;
+   v16u8 src10, src11, src12, src13, dst0, dst1;
+   v8i16 vec0, vec1, vec2;
+   v16u8 zero = { 0 };
+
+   nxt = row;
+   prev_nxt = prev_row;
+
+   inp0 = LW(nxt);
+   inp1 = LW(prev_nxt);
+   prev_nxt += 4;
+   src0 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp0);
+   src1 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp1);
+
+   src1 += src0;
+   res0 = __msa_copy_s_w((v4i32) src1, 0);
+
+   SW(res0, nxt);
+   nxt += 4;
+
+   /* Remainder */
+   rp_end = row_info->rowbytes - 4;
+
+   for (count = 0; count < rp_end; count += 16)
+   {
+      src2 = LD_UB(prev_nxt);
+      prev_nxt += 16;
+      src6 = LD_UB(prev_row);
+      prev_row += 16;
+      src10 = LD_UB(nxt);
+
+      SLDI_B3_0_UB(src2, src6, src10, src3, src7, src11, 4);
+      SLDI_B3_0_UB(src2, src6, src10, src4, src8, src12, 8);
+      SLDI_B3_0_UB(src2, src6, src10, src5, src9, src13, 12);
+      ILVR_B2_SH(src2, src6, src1, src6, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src1, src2, src6, src10);
+      ILVR_B2_SH(src3, src7, src10, src7, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src10, src3, src7, src11);
+      ILVR_B2_SH(src4, src8, src11, src8, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src11, src4, src8, src12);
+      ILVR_B2_SH(src5, src9, src12, src9, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src12, src5, src9, src13);
+      src1 = src13;
+      ILVEV_W2_UB(src10, src11, src12, src1, dst0, dst1);
+      dst0 = (v16u8) __msa_pckev_d((v2i64) dst1, (v2i64) dst0);
+
+      ST_UB(dst0, nxt);
+      nxt += 16;
+   }
+}
+
+void png_read_filter_row_paeth3_msa(png_row_infop row_info,
+                                    png_bytep row,
+                                    png_const_bytep prev_row)
+{
+   int32_t count, rp_end;
+   png_bytep nxt;
+   png_const_bytep prev_nxt;
+   int64_t out0;
+   int32_t inp0, inp1, out1;
+   int16_t out2;
+   v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, dst0, dst1;
+   v16u8 src10, src11, src12, src13;
+   v8i16 vec0, vec1, vec2;
+   v16u8 zero = { 0 };
+   v16i8 mask0 = { 0, 1, 2, 16, 17, 18, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+   v16i8 mask1 = { 0, 1, 2, 3, 4, 5, 16, 17, 18, 19, 20, 21, 0, 0, 0, 0 };
+
+   nxt = row;
+   prev_nxt = prev_row;
+
+   inp0 = LW(nxt);
+   inp1 = LW(prev_nxt);
+   prev_nxt += 3;
+   src0 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp0);
+   src1 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp1);
+
+   src1 += src0;
+   out2 = __msa_copy_s_h((v8i16) src1, 0);
+
+   SH(out2, nxt);
+   nxt += 2;
+   nxt[0] = src1[2];
+   nxt++;
+
+   /* Remainder */
+   rp_end = row_info->rowbytes - 3;
+
+   for (count = 0; count < rp_end; count += 12)
+   {
+      src2 = LD_UB(prev_nxt);
+      prev_nxt += 12;
+      src6 = LD_UB(prev_row);
+      prev_row += 12;
+      src10 = LD_UB(nxt);
+
+      SLDI_B3_0_UB(src2, src6, src10, src3, src7, src11, 3);
+      SLDI_B3_0_UB(src2, src6, src10, src4, src8, src12, 6);
+      SLDI_B3_0_UB(src2, src6, src10, src5, src9, src13, 9);
+      ILVR_B2_SH(src2, src6, src1, src6, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src1, src2, src6, src10);
+      ILVR_B2_SH(src3, src7, src10, src7, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src10, src3, src7, src11);
+      ILVR_B2_SH(src4, src8, src11, src8, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src11, src4, src8, src12);
+      ILVR_B2_SH(src5, src9, src12, src9, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src12, src5, src9, src13);
+      src1 = src13;
+      VSHF_B2_UB(src10, src11, src12, src13, mask0, mask0, dst0, dst1);
+      dst0 = (v16u8) __msa_vshf_b(mask1, (v16i8) dst1, (v16i8) dst0);
+      out0 = __msa_copy_s_d((v2i64) dst0, 0);
+      out1 = __msa_copy_s_w((v4i32) dst0, 2);
+
+      SD(out0, nxt);
+      nxt += 8;
+      SW(out1, nxt);
+      nxt += 4;
+   }
+}
+
+#endif /* PNG_MIPS_MSA_OPT > 0 */
+#endif /* PNG_MIPS_MSA_IMPLEMENTATION == 1 (intrinsics) */
+#endif /* READ */
@@ -0,0 +1,127 @@
+
+/* mips_init.c - MSA optimised filter functions
+ *
+ * Copyright (c) 2018 Cosmin Truta
+ * Copyright (c) 2016 Glenn Randers-Pehrson
+ * Written by Mandar Sahastrabuddhe, 2016.
+ *
+ * This code is released under the libpng license.
+ * For conditions of distribution and use, see the disclaimer
+ * and license in png.h
+ */
+
+/* Below, after checking __linux__, various non-C90 POSIX 1003.1 functions are
+ * called.
+ */
+#define _POSIX_SOURCE 1
+
+#include <stdio.h>
+#include "../pngpriv.h"
+
+#ifdef PNG_READ_SUPPORTED
+
+#if PNG_MIPS_MSA_OPT > 0
+#ifdef PNG_MIPS_MSA_CHECK_SUPPORTED /* Do run-time checks */
+/* WARNING: it is strongly recommended that you do not build libpng with
+ * run-time checks for CPU features if at all possible.  In the case of the MIPS
+ * MSA instructions there is no processor-specific way of detecting the
+ * presence of the required support, therefore run-time detection is extremely
+ * OS specific.
+ *
+ * You may set the macro PNG_MIPS_MSA_FILE to the file name of file containing
+ * a fragment of C source code which defines the png_have_msa function.  There
+ * are a number of implementations in contrib/mips-msa, but the only one that
+ * has partial support is contrib/mips-msa/linux.c - a generic Linux
+ * implementation which reads /proc/cpufino.
+ */
+#ifndef PNG_MIPS_MSA_FILE
+#  ifdef __linux__
+#     define PNG_MIPS_MSA_FILE "contrib/mips-msa/linux.c"
+#  endif
+#endif
+
+#ifdef PNG_MIPS_MSA_FILE
+
+#include <signal.h> /* for sig_atomic_t */
+static int png_have_msa(png_structp png_ptr);
+#include PNG_MIPS_MSA_FILE
+
+#else  /* PNG_MIPS_MSA_FILE */
+#  error "PNG_MIPS_MSA_FILE undefined: no support for run-time MIPS MSA checks"
+#endif /* PNG_MIPS_MSA_FILE */
+#endif /* PNG_MIPS_MSA_CHECK_SUPPORTED */
+
+#ifndef PNG_ALIGNED_MEMORY_SUPPORTED
+#  error "ALIGNED_MEMORY is required; set: -DPNG_ALIGNED_MEMORY_SUPPORTED"
+#endif
+
+void
+png_init_filter_functions_msa(png_structp pp, unsigned int bpp)
+{
+   /* The switch statement is compiled in for MIPS_MSA_API, the call to
+    * png_have_msa is compiled in for MIPS_MSA_CHECK. If both are defined
+    * the check is only performed if the API has not set the MSA option on
+    * or off explicitly. In this case the check controls what happens.
+    */
+
+#ifdef PNG_MIPS_MSA_API_SUPPORTED
+   switch ((pp->options >> PNG_MIPS_MSA) & 3)
+   {
+      case PNG_OPTION_UNSET:
+         /* Allow the run-time check to execute if it has been enabled -
+          * thus both API and CHECK can be turned on.  If it isn't supported
+          * this case will fall through to the 'default' below, which just
+          * returns.
+          */
+#ifdef PNG_MIPS_MSA_CHECK_SUPPORTED
+         {
+            static volatile sig_atomic_t no_msa = -1; /* not checked */
+
+            if (no_msa < 0)
+               no_msa = !png_have_msa(pp);
+
+            if (no_msa)
+               return;
+         }
+#endif /* PNG_MIPS_MSA_CHECK_SUPPORTED */
+         break;
+
+      default: /* OFF or INVALID */
+         return;
+
+      case PNG_OPTION_ON:
+         /* Option turned on */
+         break;
+   }
+   /* IMPORTANT: any new external functions used here must be declared using
+    * PNG_INTERNAL_FUNCTION in ../pngpriv.h.  This is required so that the
+    * 'prefix' option to configure works:
+    *
+    *    ./configure --with-libpng-prefix=foobar_
+    *
+    * Verify you have got this right by running the above command, doing a build
+    * and examining pngprefix.h; it must contain a #define for every external
+    * function you add.  (Notice that this happens automatically for the
+    * initialization function.)
+    */
+   pp->read_filter[PNG_FILTER_VALUE_UP-1] = png_read_filter_row_up_msa;
+
+   if (bpp == 3)
+   {
+      pp->read_filter[PNG_FILTER_VALUE_SUB-1] = png_read_filter_row_sub3_msa;
+      pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg3_msa;
+      pp->read_filter[PNG_FILTER_VALUE_PAETH-1] = png_read_filter_row_paeth3_msa;
+   }
+   else if (bpp == 4)
+   {
+      pp->read_filter[PNG_FILTER_VALUE_SUB-1] = png_read_filter_row_sub4_msa;
+      pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg4_msa;
+      pp->read_filter[PNG_FILTER_VALUE_PAETH-1] = png_read_filter_row_paeth4_msa;
+   }
+#else
+   (void)pp;
+   (void)bpp;
+#endif /* PNG_MIPS_MSA_API_SUPPORTED */
+}
+#endif /* PNG_MIPS_MSA_OPT > 0 */
+#endif /* READ */
@@ -0,0 +1,53 @@
+diff --git a/3rdparty/libpng/mips/mips_init.c b/3rdparty/libpng/mips/mips_init.c
+index 8dd283deef..6a061cccfa 100644
+--- a/3rdparty/libpng/mips/mips_init.c
+++ b/3rdparty/libpng/mips/mips_init.c
+@@ -73,7 +73,6 @@ png_init_filter_functions_msa(png_structp pp, unsigned int bpp)
+           * this case will fall through to the 'default' below, which just
+           * returns.
+           */
+-#endif /* PNG_MIPS_MSA_API_SUPPORTED */
+ #ifdef PNG_MIPS_MSA_CHECK_SUPPORTED
+          {
+             static volatile sig_atomic_t no_msa = -1; /* not checked */
+@@ -84,12 +83,9 @@ png_init_filter_functions_msa(png_structp pp, unsigned int bpp)
+             if (no_msa)
+                return;
+          }
+-#ifdef PNG_MIPS_MSA_API_SUPPORTED
+-         break;
+-#endif
+ #endif /* PNG_MIPS_MSA_CHECK_SUPPORTED */
+         break;
+ 
+-#ifdef PNG_MIPS_MSA_API_SUPPORTED
+       default: /* OFF or INVALID */
+          return;
+ 
+@@ -97,8 +93,6 @@ png_init_filter_functions_msa(png_structp pp, unsigned int bpp)
+          /* Option turned on */
+          break;
+    }
+-#endif
+-
+    /* IMPORTANT: any new external functions used here must be declared using
+     * PNG_INTERNAL_FUNCTION in ../pngpriv.h.  This is required so that the
+     * 'prefix' option to configure works:
+@@ -118,13 +112,16 @@ png_init_filter_functions_msa(png_structp pp, unsigned int bpp)
+       pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg3_msa;
+       pp->read_filter[PNG_FILTER_VALUE_PAETH-1] = png_read_filter_row_paeth3_msa;
+    }
+-
+    else if (bpp == 4)
+    {
+       pp->read_filter[PNG_FILTER_VALUE_SUB-1] = png_read_filter_row_sub4_msa;
+       pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg4_msa;
+       pp->read_filter[PNG_FILTER_VALUE_PAETH-1] = png_read_filter_row_paeth4_msa;
+    }
+#else
+   (void)pp;
+   (void)bpp;
+#endif /* PNG_MIPS_MSA_API_SUPPORTED */
+ }
+ #endif /* PNG_MIPS_MSA_OPT > 0 */
+ #endif /* READ */
@@ -0,0 +1,91 @@
+# 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,
@@ -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));                                \
@@ -732,7 +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.
@@ -166,9 +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
@@ -184,7 +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;
@@ -192,6 +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_);
@@ -204,7 +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!
@@ -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;
@@ -31,8 +31,8 @@ extern "C" {

 // version numbers
 #define DEC_MAJ_VERSION 1
-#define DEC_MIN_VERSION 0
-#define DEC_REV_VERSION 2
+#define DEC_MIN_VERSION 1
+#define DEC_REV_VERSION 0

 // YUV-cache parameters. Cache is 32-bytes wide (= one cacheline).
 // Constraints are: We need to store one 16x16 block of luma samples (y),
@@ -362,12 +362,8 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
  VP8LMetadata* const hdr = &dec->hdr_;
  uint32_t* huffman_image = NULL;
  HTreeGroup* htree_groups = NULL;
-  // When reading htrees, some might be unused, as the format allows it.
-  // We will still read them but put them in this htree_group_bogus.
-  HTreeGroup htree_group_bogus;
  HuffmanCode* huffman_tables = NULL;
-  HuffmanCode* huffman_tables_bogus = NULL;
-  HuffmanCode* next = NULL;
+  HuffmanCode* huffman_table = NULL;
  int num_htree_groups = 1;
  int num_htree_groups_max = 1;
  int max_alphabet_size = 0;
@@ -418,12 +414,6 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
        if (*mapped_group == -1) *mapped_group = num_htree_groups++;
        huffman_image[i] = *mapped_group;
      }
-      huffman_tables_bogus = (HuffmanCode*)WebPSafeMalloc(
-          table_size, sizeof(*huffman_tables_bogus));
-      if (huffman_tables_bogus == NULL) {
-        dec->status_ = VP8_STATUS_OUT_OF_MEMORY;
-        goto Error;
-      }
    } else {
      num_htree_groups = num_htree_groups_max;
    }
@@ -453,63 +443,71 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
    goto Error;
  }

-  next = huffman_tables;
+  huffman_table = huffman_tables;
  for (i = 0; i < num_htree_groups_max; ++i) {
-    // If the index "i" is unused in the Huffman image, read the coefficients
-    // but store them to a bogus htree_group.
-    const int is_bogus = (mapping != NULL && mapping[i] == -1);
-    HTreeGroup* const htree_group =
-        is_bogus ? &htree_group_bogus :
-        &htree_groups[(mapping == NULL) ? i : mapping[i]];
-    HuffmanCode** const htrees = htree_group->htrees;
-    HuffmanCode* huffman_tables_i = is_bogus ? huffman_tables_bogus : next;
-    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_tables_i;
-      if (j == 0 && color_cache_bits > 0) {
-        alphabet_size += 1 << color_cache_bits;
-      }
-      size =
-          ReadHuffmanCode(alphabet_size, dec, code_lengths, huffman_tables_i);
-      if (size == 0) {
-        goto Error;
-      }
-      if (is_trivial_literal && kLiteralMap[j] == 1) {
-        is_trivial_literal = (huffman_tables_i->bits == 0);
-      }
-      total_size += huffman_tables_i->bits;
-      huffman_tables_i += 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];
-          }
+    // 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;
      }
-    }
-    if (!is_bogus) next = huffman_tables_i;
-    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);
  }
  ok = 1;

@@ -521,7 +519,6 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,

 Error:
  WebPSafeFree(code_lengths);
-  WebPSafeFree(huffman_tables_bogus);
  WebPSafeFree(mapping);
  if (!ok) {
    WebPSafeFree(huffman_image);
@@ -757,11 +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_;
@@ -792,8 +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 {
@@ -1196,6 +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;
@@ -1556,7 +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);
@@ -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.
 };

 //------------------------------------------------------------------------------
@@ -24,8 +24,8 @@
 #include "src/webp/format_constants.h"

 #define DMUX_MAJ_VERSION 1
-#define DMUX_MIN_VERSION 0
-#define DMUX_REV_VERSION 2
+#define DMUX_MIN_VERSION 1
+#define DMUX_REV_VERSION 0

 typedef struct {
  size_t start_;        // start location of the data
@@ -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
@@ -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);
@@ -246,9 +246,9 @@ extern VP8Fdct VP8FTransform2;   // performs two transforms at a time
 extern VP8WHT VP8FTransformWHT;
 // Predictions
 // *dst is the destination block. *top and *left can be NULL.
-typedef void (*VP8IntraPreds)(uint8_t *dst, const uint8_t* left,
+typedef void (*VP8IntraPreds)(uint8_t* dst, const uint8_t* left,
                              const uint8_t* top);
-typedef void (*VP8Intra4Preds)(uint8_t *dst, const uint8_t* top);
+typedef void (*VP8Intra4Preds)(uint8_t* dst, const uint8_t* top);
 extern VP8Intra4Preds VP8EncPredLuma4;
 extern VP8IntraPreds VP8EncPredLuma16;
 extern VP8IntraPreds VP8EncPredChroma8;
@@ -777,7 +777,7 @@ static WEBP_INLINE void VR4_SSE2(uint8_t* dst,
  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);
@@ -33,9 +33,9 @@ static WEBP_INLINE void PredictLine_C(const uint8_t* src, const uint8_t* pred,
                                      uint8_t* dst, int length, int inverse) {
  int i;
  if (inverse) {
-    for (i = 0; i < length; ++i) dst[i] = src[i] + pred[i];
+    for (i = 0; i < length; ++i) dst[i] = (uint8_t)(src[i] + pred[i]);
  } else {
-    for (i = 0; i < length; ++i) dst[i] = src[i] - pred[i];
+    for (i = 0; i < length; ++i) dst[i] = (uint8_t)(src[i] - pred[i]);
  }
 }

@@ -155,7 +155,7 @@ static WEBP_INLINE void DoGradientFilter_C(const uint8_t* in,
      const int pred = GradientPredictor_C(preds[w - 1],
                                           preds[w - stride],
                                           preds[w - stride - 1]);
-      out[w] = in[w] + (inverse ? pred : -pred);
+      out[w] = (uint8_t)(in[w] + (inverse ? pred : -pred));
    }
    ++row;
    preds += stride;
@@ -194,7 +194,7 @@ static void HorizontalUnfilter_C(const uint8_t* prev, const uint8_t* in,
  uint8_t pred = (prev == NULL) ? 0 : prev[0];
  int i;
  for (i = 0; i < width; ++i) {
-    out[i] = pred + in[i];
+    out[i] = (uint8_t)(pred + in[i]);
    pred = out[i];
  }
 }
@@ -206,7 +206,7 @@ static void VerticalUnfilter_C(const uint8_t* prev, const uint8_t* in,
    HorizontalUnfilter_C(NULL, in, out, width);
  } else {
    int i;
-    for (i = 0; i < width; ++i) out[i] = prev[i] + in[i];
+    for (i = 0; i < width; ++i) out[i] = (uint8_t)(prev[i] + in[i]);
  }
 }
 #endif  // !WEBP_NEON_OMIT_C_CODE
@@ -220,7 +220,7 @@ static void GradientUnfilter_C(const uint8_t* prev, const uint8_t* in,
    int i;
    for (i = 0; i < width; ++i) {
      top = prev[i];  // need to read this first, in case prev==out
-      left = in[i] + GradientPredictor_C(left, top, top_left);
+      left = (uint8_t)(in[i] + GradientPredictor_C(left, top, top_left));
      top_left = top;
      out[i] = left;
    }
@@ -163,7 +163,8 @@ static void GradientPredictDirect_SSE2(const uint8_t* const row,
    _mm_storel_epi64((__m128i*)(out + i), H);
  }
  for (; i < length; ++i) {
-    out[i] = row[i] - GradientPredictor_SSE2(row[i - 1], top[i], top[i - 1]);
+    const int delta = GradientPredictor_SSE2(row[i - 1], top[i], top[i - 1]);
+    out[i] = (uint8_t)(row[i] - delta);
  }
 }

@@ -188,7 +189,7 @@ static WEBP_INLINE void DoGradientFilter_SSE2(const uint8_t* in,

  // Filter line-by-line.
  while (row < last_row) {
-    out[0] = in[0] - in[-stride];
+    out[0] = (uint8_t)(in[0] - in[-stride]);
    GradientPredictDirect_SSE2(in + 1, in + 1 - stride, out + 1, width - 1);
    ++row;
    in += stride;
@@ -223,7 +224,7 @@ static void HorizontalUnfilter_SSE2(const uint8_t* prev, const uint8_t* in,
                                    uint8_t* out, int width) {
  int i;
  __m128i last;
-  out[0] = in[0] + (prev == NULL ? 0 : prev[0]);
+  out[0] = (uint8_t)(in[0] + (prev == NULL ? 0 : prev[0]));
  if (width <= 1) return;
  last = _mm_set_epi32(0, 0, 0, out[0]);
  for (i = 1; i + 8 <= width; i += 8) {
@@ -238,7 +239,7 @@ static void HorizontalUnfilter_SSE2(const uint8_t* prev, const uint8_t* in,
    _mm_storel_epi64((__m128i*)(out + i), A7);
    last = _mm_srli_epi64(A7, 56);
  }
-  for (; i < width; ++i) out[i] = in[i] + out[i - 1];
+  for (; i < width; ++i) out[i] = (uint8_t)(in[i] + out[i - 1]);
 }

 static void VerticalUnfilter_SSE2(const uint8_t* prev, const uint8_t* in,
@@ -259,7 +260,7 @@ static void VerticalUnfilter_SSE2(const uint8_t* prev, const uint8_t* in,
      _mm_storeu_si128((__m128i*)&out[i +  0], C0);
      _mm_storeu_si128((__m128i*)&out[i + 16], C1);
    }
-    for (; i < width; ++i) out[i] = in[i] + prev[i];
+    for (; i < width; ++i) out[i] = (uint8_t)(in[i] + prev[i]);
  }
 }

@@ -296,7 +297,8 @@ static void GradientPredictInverse_SSE2(const uint8_t* const in,
      _mm_storel_epi64((__m128i*)&row[i], out);
    }
    for (; i < length; ++i) {
-      row[i] = in[i] + GradientPredictor_SSE2(row[i - 1], top[i], top[i - 1]);
+      const int delta = GradientPredictor_SSE2(row[i - 1], top[i], top[i - 1]);
+      row[i] = (uint8_t)(in[i] + delta);
    }
  }
 }
@@ -306,7 +308,7 @@ static void GradientUnfilter_SSE2(const uint8_t* prev, const uint8_t* in,
  if (prev == NULL) {
    HorizontalUnfilter_SSE2(NULL, in, out, width);
  } else {
-    out[0] = in[0] + prev[0];  // predict from above
+    out[0] = (uint8_t)(in[0] + prev[0]);  // predict from above
    GradientPredictInverse_SSE2(in + 1, prev + 1, out + 1, width - 1);
  }
 }
@@ -81,7 +81,7 @@ static WEBP_INLINE uint32_t ClampedAddSubtractHalf(uint32_t c0, uint32_t c1,

 // gcc <= 4.9 on ARM generates incorrect code in Select() when Sub3() is
 // inlined.
-#if defined(__arm__) && LOCAL_GCC_VERSION <= 0x409
+#if defined(__arm__) && defined(__GNUC__) && LOCAL_GCC_VERSION <= 0x409
 # define LOCAL_INLINE __attribute__ ((noinline))
 #else
 # define LOCAL_INLINE WEBP_INLINE
@@ -167,15 +167,20 @@ static uint32_t Predictor13_C(uint32_t left, const uint32_t* const top) {
  return pred;
 }

-GENERATE_PREDICTOR_ADD(Predictor0_C, PredictorAdd0_C)
+static void PredictorAdd0_C(const uint32_t* in, const uint32_t* upper,
+                            int num_pixels, uint32_t* out) {
+  int x;
+  (void)upper;
+  for (x = 0; x < num_pixels; ++x) out[x] = VP8LAddPixels(in[x], ARGB_BLACK);
+}
 static void PredictorAdd1_C(const uint32_t* in, const uint32_t* upper,
                            int num_pixels, uint32_t* out) {
  int i;
  uint32_t left = out[-1];
+  (void)upper;
  for (i = 0; i < num_pixels; ++i) {
    out[i] = left = VP8LAddPixels(in[i], left);
  }
-  (void)upper;
 }
 GENERATE_PREDICTOR_ADD(Predictor2_C, PredictorAdd2_C)
 GENERATE_PREDICTOR_ADD(Predictor3_C, PredictorAdd3_C)
@@ -270,14 +275,14 @@ void VP8LTransformColorInverse_C(const VP8LMultipliers* const m,
  int i;
  for (i = 0; i < num_pixels; ++i) {
    const uint32_t argb = src[i];
-    const uint32_t green = argb >> 8;
+    const int8_t green = (int8_t)(argb >> 8);
    const uint32_t red = argb >> 16;
    int new_red = red & 0xff;
    int new_blue = argb & 0xff;
    new_red += ColorTransformDelta(m->green_to_red_, green);
    new_red &= 0xff;
    new_blue += ColorTransformDelta(m->green_to_blue_, green);
-    new_blue += ColorTransformDelta(m->red_to_blue_, new_red);
+    new_blue += ColorTransformDelta(m->red_to_blue_, (int8_t)new_red);
    new_blue &= 0xff;
    dst[i] = (argb & 0xff00ff00u) | (new_red << 16) | (new_blue);
  }
@@ -177,6 +177,7 @@ uint32_t VP8LSubPixels(uint32_t a, uint32_t b) {
 static void PREDICTOR_ADD(const uint32_t* in, const uint32_t* upper, \
                          int num_pixels, uint32_t* out) {           \
  int x;                                                             \
+  assert(upper != NULL);                                             \
  for (x = 0; x < num_pixels; ++x) {                                 \
    const uint32_t pred = (PREDICTOR)(out[x - 1], upper + x);        \
    out[x] = VP8LAddPixels(in[x], pred);                             \
@@ -189,6 +190,7 @@ static void PREDICTOR_ADD(const uint32_t* in, const uint32_t* upper, \
 static void PREDICTOR_SUB(const uint32_t* in, const uint32_t* upper, \
                          int num_pixels, uint32_t* out) {           \
  int x;                                                             \
+  assert(upper != NULL);                                             \
  for (x = 0; x < num_pixels; ++x) {                                 \
    const uint32_t pred = (PREDICTOR)(in[x - 1], upper + x);         \
    out[x] = VP8LSubPixels(in[x], pred);                             \
@@ -515,13 +515,17 @@ static WEBP_INLINE int ColorTransformDelta(int8_t color_pred, int8_t color) {
  return ((int)color_pred * color) >> 5;
 }

+static WEBP_INLINE int8_t U32ToS8(uint32_t v) {
+  return (int8_t)(v & 0xff);
+}
+
 void VP8LTransformColor_C(const VP8LMultipliers* const m, uint32_t* data,
                          int num_pixels) {
  int i;
  for (i = 0; i < num_pixels; ++i) {
    const uint32_t argb = data[i];
-    const uint32_t green = argb >> 8;
-    const uint32_t red = argb >> 16;
+    const int8_t green = U32ToS8(argb >>  8);
+    const int8_t red   = U32ToS8(argb >> 16);
    int new_red = red & 0xff;
    int new_blue = argb & 0xff;
    new_red -= ColorTransformDelta(m->green_to_red_, green);
@@ -535,7 +539,7 @@ void VP8LTransformColor_C(const VP8LMultipliers* const m, uint32_t* data,

 static WEBP_INLINE uint8_t TransformColorRed(uint8_t green_to_red,
                                             uint32_t argb) {
-  const uint32_t green = argb >> 8;
+  const int8_t green = U32ToS8(argb >> 8);
  int new_red = argb >> 16;
  new_red -= ColorTransformDelta(green_to_red, green);
  return (new_red & 0xff);
@@ -544,9 +548,9 @@ static WEBP_INLINE uint8_t TransformColorRed(uint8_t green_to_red,
 static WEBP_INLINE uint8_t TransformColorBlue(uint8_t green_to_blue,
                                              uint8_t red_to_blue,
                                              uint32_t argb) {
-  const uint32_t green = argb >> 8;
-  const uint32_t red = argb >> 16;
-  uint8_t new_blue = argb;
+  const int8_t green = U32ToS8(argb >>  8);
+  const int8_t red   = U32ToS8(argb >> 16);
+  uint8_t new_blue = argb & 0xff;
  new_blue -= ColorTransformDelta(green_to_blue, green);
  new_blue -= ColorTransformDelta(red_to_blue, red);
  return (new_blue & 0xff);
@@ -558,7 +562,7 @@ void VP8LCollectColorRedTransforms_C(const uint32_t* argb, int stride,
  while (tile_height-- > 0) {
    int x;
    for (x = 0; x < tile_width; ++x) {
-      ++histo[TransformColorRed(green_to_red, argb[x])];
+      ++histo[TransformColorRed((uint8_t)green_to_red, argb[x])];
    }
    argb += stride;
  }
@@ -571,7 +575,8 @@ void VP8LCollectColorBlueTransforms_C(const uint32_t* argb, int stride,
  while (tile_height-- > 0) {
    int x;
    for (x = 0; x < tile_width; ++x) {
-      ++histo[TransformColorBlue(green_to_blue, red_to_blue, argb[x])];
+      ++histo[TransformColorBlue((uint8_t)green_to_blue, (uint8_t)red_to_blue,
+                                 argb[x])];
    }
    argb += stride;
  }
@@ -363,7 +363,7 @@ static void BundleColorMap_SSE2(const uint8_t* const row, int width, int xbits,
  assert(xbits <= 3);
  switch (xbits) {
    case 0: {
-      const __m128i ff = _mm_set1_epi16(0xff00);
+      const __m128i ff = _mm_set1_epi16((short)0xff00);
      const __m128i zero = _mm_setzero_si128();
      // Store 0xff000000 | (row[x] << 8).
      for (x = 0; x + 16 <= width; x += 16, dst += 16) {
@@ -382,7 +382,7 @@ static void BundleColorMap_SSE2(const uint8_t* const row, int width, int xbits,
      break;
    }
    case 1: {
-      const __m128i ff = _mm_set1_epi16(0xff00);
+      const __m128i ff = _mm_set1_epi16((short)0xff00);
      const __m128i mul = _mm_set1_epi16(0x110);
      for (x = 0; x + 16 <= width; x += 16, dst += 8) {
        // 0a0b | (where a/b are 4 bits).
@@ -455,8 +455,9 @@ static void PredictorSub0_SSE2(const uint32_t* in, const uint32_t* upper,
    _mm_storeu_si128((__m128i*)&out[i], res);
  }
  if (i != num_pixels) {
-    VP8LPredictorsSub_C[0](in + i, upper + i, num_pixels - i, out + i);
+    VP8LPredictorsSub_C[0](in + i, NULL, num_pixels - i, out + i);
  }
+  (void)upper;
 }

 #define GENERATE_PREDICTOR_1(X, IN)                                           \
@@ -51,9 +51,9 @@ static void CollectColorBlueTransforms_SSE41(const uint32_t* argb, int stride,
                                             int histo[]) {
  const __m128i mults_r = _mm_set1_epi16(CST_5b(red_to_blue));
  const __m128i mults_g = _mm_set1_epi16(CST_5b(green_to_blue));
-  const __m128i mask_g = _mm_set1_epi16(0xff00);   // green mask
-  const __m128i mask_gb = _mm_set1_epi32(0xffff);  // green/blue mask
-  const __m128i mask_b = _mm_set1_epi16(0x00ff);   // blue mask
+  const __m128i mask_g = _mm_set1_epi16((short)0xff00);   // green mask
+  const __m128i mask_gb = _mm_set1_epi32(0xffff);         // green/blue mask
+  const __m128i mask_b = _mm_set1_epi16(0x00ff);          // blue mask
  const __m128i shuffler_lo = _mm_setr_epi8(-1, 2, -1, 6, -1, 10, -1, 14, -1,
                                            -1, -1, -1, -1, -1, -1, -1);
  const __m128i shuffler_hi = _mm_setr_epi8(-1, -1, -1, -1, -1, -1, -1, -1, -1,
@@ -191,8 +191,9 @@ static void PredictorAdd0_SSE2(const uint32_t* in, const uint32_t* upper,
    _mm_storeu_si128((__m128i*)&out[i], res);
  }
  if (i != num_pixels) {
-    VP8LPredictorsAdd_C[0](in + i, upper + i, num_pixels - i, out + i);
+    VP8LPredictorsAdd_C[0](in + i, NULL, num_pixels - i, out + i);
  }
+  (void)upper;
 }

 // Predictor1: left.
@@ -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));                                \
@@ -10,6 +10,8 @@
 #ifndef WEBP_DSP_QUANT_H_
 #define WEBP_DSP_QUANT_H_

+#include <string.h>
+
 #include "src/dsp/dsp.h"
 #include "src/webp/types.h"

@@ -67,4 +69,17 @@ static WEBP_INLINE int IsFlat(const int16_t* levels, int num_blocks,
 #endif  // defined(WEBP_USE_NEON) && !defined(WEBP_ANDROID_NEON) &&
        // !defined(WEBP_HAVE_NEON_RTCD)

+static WEBP_INLINE int IsFlatSource16(const uint8_t* src) {
+  const uint32_t v = src[0] * 0x01010101u;
+  int i;
+  for (i = 0; i < 16; ++i) {
+    if (memcmp(src + 0, &v, 4) || memcmp(src +  4, &v, 4) ||
+        memcmp(src + 8, &v, 4) || memcmp(src + 12, &v, 4)) {
+      return 0;
+    }
+    src += BPS;
+  }
+  return 1;
+}
+
 #endif  // WEBP_DSP_QUANT_H_
@@ -109,8 +109,7 @@ void WebPRescalerExportRowExpand_C(WebPRescaler* const wrk) {
    for (x_out = 0; x_out < x_out_max; ++x_out) {
      const uint32_t J = frow[x_out];
      const int v = (int)MULT_FIX(J, wrk->fy_scale);
-      assert(v >= 0 && v <= 255);
-      dst[x_out] = v;
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
    }
  } else {
    const uint32_t B = WEBP_RESCALER_FRAC(-wrk->y_accum, wrk->y_sub);
@@ -120,8 +119,7 @@ void WebPRescalerExportRowExpand_C(WebPRescaler* const wrk) {
                       + (uint64_t)B * irow[x_out];
      const uint32_t J = (uint32_t)((I + ROUNDER) >> WEBP_RESCALER_RFIX);
      const int v = (int)MULT_FIX(J, wrk->fy_scale);
-      assert(v >= 0 && v <= 255);
-      dst[x_out] = v;
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
    }
  }
 }
@@ -138,17 +136,15 @@ void WebPRescalerExportRowShrink_C(WebPRescaler* const wrk) {
  assert(!wrk->y_expand);
  if (yscale) {
    for (x_out = 0; x_out < x_out_max; ++x_out) {
-      const uint32_t frac = (uint32_t)MULT_FIX(frow[x_out], yscale);
-      const int v = (int)MULT_FIX_FLOOR(irow[x_out] - frac, wrk->fxy_scale);
-      assert(v >= 0 && v <= 255);
-      dst[x_out] = v;
+      const uint32_t frac = (uint32_t)MULT_FIX_FLOOR(frow[x_out], yscale);
+      const int v = (int)MULT_FIX(irow[x_out] - frac, wrk->fxy_scale);
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
      irow[x_out] = frac;   // new fractional start
    }
  } else {
    for (x_out = 0; x_out < x_out_max; ++x_out) {
      const int v = (int)MULT_FIX(irow[x_out], wrk->fxy_scale);
-      assert(v >= 0 && v <= 255);
-      dst[x_out] = v;
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
      irow[x_out] = 0;
    }
  }
@@ -107,10 +107,9 @@ static void ExportRowShrink_MIPSdspR2(WebPRescaler* const wrk) {
      );
    }
    for (i = 0; i < (x_out_max & 0x3); ++i) {
-      const uint32_t frac = (uint32_t)MULT_FIX(*frow++, yscale);
-      const int v = (int)MULT_FIX_FLOOR(*irow - frac, wrk->fxy_scale);
-      assert(v >= 0 && v <= 255);
-      *dst++ = v;
+      const uint32_t frac = (uint32_t)MULT_FIX_FLOOR(*frow++, yscale);
+      const int v = (int)MULT_FIX(*irow - frac, wrk->fxy_scale);
+      *dst++ = (v > 255) ? 255u : (uint8_t)v;
      *irow++ = frac;   // new fractional start
    }
  } else {
@@ -157,8 +156,7 @@ static void ExportRowShrink_MIPSdspR2(WebPRescaler* const wrk) {
    }
    for (i = 0; i < (x_out_max & 0x3); ++i) {
      const int v = (int)MULT_FIX_FLOOR(*irow, wrk->fxy_scale);
-      assert(v >= 0 && v <= 255);
-      *dst++ = v;
+      *dst++ = (v > 255) ? 255u : (uint8_t)v;
      *irow++ = 0;
    }
  }
@@ -219,8 +217,7 @@ static void ExportRowExpand_MIPSdspR2(WebPRescaler* const wrk) {
    for (i = 0; i < (x_out_max & 0x3); ++i) {
      const uint32_t J = *frow++;
      const int v = (int)MULT_FIX(J, wrk->fy_scale);
-      assert(v >= 0 && v <= 255);
-      *dst++ = v;
+      *dst++ = (v > 255) ? 255u : (uint8_t)v;
    }
  } else {
    const uint32_t B = WEBP_RESCALER_FRAC(-wrk->y_accum, wrk->y_sub);
@@ -291,8 +288,7 @@ static void ExportRowExpand_MIPSdspR2(WebPRescaler* const wrk) {
                       + (uint64_t)B * *irow++;
      const uint32_t J = (uint32_t)((I + ROUNDER) >> WEBP_RESCALER_RFIX);
      const int v = (int)MULT_FIX(J, wrk->fy_scale);
-      assert(v >= 0 && v <= 255);
-      *dst++ = v;
+      *dst++ = (v > 255) ? 255u : (uint8_t)v;
    }
  }
 }
@@ -166,8 +166,7 @@ static WEBP_INLINE void ExportRowExpand_0(const uint32_t* frow, uint8_t* dst,
    for (x_out = 0; x_out < length; ++x_out) {
      const uint32_t J = frow[x_out];
      const int v = (int)MULT_FIX(J, wrk->fy_scale);
-      assert(v >= 0 && v <= 255);
-      dst[x_out] = v;
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
    }
  }
 }
@@ -241,8 +240,7 @@ static WEBP_INLINE void ExportRowExpand_1(const uint32_t* frow, uint32_t* irow,
                       + (uint64_t)B * irow[x_out];
      const uint32_t J = (uint32_t)((I + ROUNDER) >> WEBP_RESCALER_RFIX);
      const int v = (int)MULT_FIX(J, wrk->fy_scale);
-      assert(v >= 0 && v <= 255);
-      dst[x_out] = v;
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
    }
  }
 }
@@ -342,10 +340,9 @@ static WEBP_INLINE void ExportRowShrink_0(const uint32_t* frow, uint32_t* irow,
      length -= 4;
    }
    for (x_out = 0; x_out < length; ++x_out) {
-      const uint32_t frac = (uint32_t)MULT_FIX(frow[x_out], yscale);
-      const int v = (int)MULT_FIX_FLOOR(irow[x_out] - frac, wrk->fxy_scale);
-      assert(v >= 0 && v <= 255);
-      dst[x_out] = v;
+      const uint32_t frac = (uint32_t)MULT_FIX_FLOOR(frow[x_out], yscale);
+      const int v = (int)MULT_FIX(irow[x_out] - frac, wrk->fxy_scale);
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
      irow[x_out] = frac;
    }
  }
@@ -406,8 +403,7 @@ static WEBP_INLINE void ExportRowShrink_1(uint32_t* irow, uint8_t* dst,
    }
    for (x_out = 0; x_out < length; ++x_out) {
      const int v = (int)MULT_FIX(irow[x_out], wrk->fxy_scale);
-      assert(v >= 0 && v <= 255);
-      dst[x_out] = v;
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
      irow[x_out] = 0;
    }
  }
@@ -81,14 +81,13 @@ static void RescalerExportRowExpand_NEON(WebPRescaler* const wrk) {
      const uint32x4_t B1 = MULT_FIX(A1, fy_scale_half);
      const uint16x4_t C0 = vmovn_u32(B0);
      const uint16x4_t C1 = vmovn_u32(B1);
-      const uint8x8_t D = vmovn_u16(vcombine_u16(C0, C1));
+      const uint8x8_t D = vqmovn_u16(vcombine_u16(C0, C1));
      vst1_u8(dst + x_out, D);
    }
    for (; x_out < x_out_max; ++x_out) {
      const uint32_t J = frow[x_out];
      const int v = (int)MULT_FIX_C(J, fy_scale);
-      assert(v >= 0 && v <= 255);
-      dst[x_out] = v;
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
    }
  } else {
    const uint32_t B = WEBP_RESCALER_FRAC(-wrk->y_accum, wrk->y_sub);
@@ -102,7 +101,7 @@ static void RescalerExportRowExpand_NEON(WebPRescaler* const wrk) {
      const uint32x4_t D1 = MULT_FIX(C1, fy_scale_half);
      const uint16x4_t E0 = vmovn_u32(D0);
      const uint16x4_t E1 = vmovn_u32(D1);
-      const uint8x8_t F = vmovn_u16(vcombine_u16(E0, E1));
+      const uint8x8_t F = vqmovn_u16(vcombine_u16(E0, E1));
      vst1_u8(dst + x_out, F);
    }
    for (; x_out < x_out_max; ++x_out) {
@@ -110,8 +109,7 @@ static void RescalerExportRowExpand_NEON(WebPRescaler* const wrk) {
                       + (uint64_t)B * irow[x_out];
      const uint32_t J = (uint32_t)((I + ROUNDER) >> WEBP_RESCALER_RFIX);
      const int v = (int)MULT_FIX_C(J, fy_scale);
-      assert(v >= 0 && v <= 255);
-      dst[x_out] = v;
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
    }
  }
 }
@@ -135,23 +133,22 @@ static void RescalerExportRowShrink_NEON(WebPRescaler* const wrk) {
    for (x_out = 0; x_out < max_span; x_out += 8) {
      LOAD_32x8(frow + x_out, in0, in1);
      LOAD_32x8(irow + x_out, in2, in3);
-      const uint32x4_t A0 = MULT_FIX(in0, yscale_half);
-      const uint32x4_t A1 = MULT_FIX(in1, yscale_half);
+      const uint32x4_t A0 = MULT_FIX_FLOOR(in0, yscale_half);
+      const uint32x4_t A1 = MULT_FIX_FLOOR(in1, yscale_half);
      const uint32x4_t B0 = vqsubq_u32(in2, A0);
      const uint32x4_t B1 = vqsubq_u32(in3, A1);
-      const uint32x4_t C0 = MULT_FIX_FLOOR(B0, fxy_scale_half);
-      const uint32x4_t C1 = MULT_FIX_FLOOR(B1, fxy_scale_half);
+      const uint32x4_t C0 = MULT_FIX(B0, fxy_scale_half);
+      const uint32x4_t C1 = MULT_FIX(B1, fxy_scale_half);
      const uint16x4_t D0 = vmovn_u32(C0);
      const uint16x4_t D1 = vmovn_u32(C1);
-      const uint8x8_t E = vmovn_u16(vcombine_u16(D0, D1));
+      const uint8x8_t E = vqmovn_u16(vcombine_u16(D0, D1));
      vst1_u8(dst + x_out, E);
      STORE_32x8(A0, A1, irow + x_out);
    }
    for (; x_out < x_out_max; ++x_out) {
-      const uint32_t frac = (uint32_t)MULT_FIX_C(frow[x_out], yscale);
-      const int v = (int)MULT_FIX_FLOOR_C(irow[x_out] - frac, fxy_scale);
-      assert(v >= 0 && v <= 255);
-      dst[x_out] = v;
+      const uint32_t frac = (uint32_t)MULT_FIX_FLOOR_C(frow[x_out], yscale);
+      const int v = (int)MULT_FIX_C(irow[x_out] - frac, fxy_scale);
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
      irow[x_out] = frac;   // new fractional start
    }
  } else {
@@ -161,14 +158,13 @@ static void RescalerExportRowShrink_NEON(WebPRescaler* const wrk) {
      const uint32x4_t A1 = MULT_FIX(in1, fxy_scale_half);
      const uint16x4_t B0 = vmovn_u32(A0);
      const uint16x4_t B1 = vmovn_u32(A1);
-      const uint8x8_t C = vmovn_u16(vcombine_u16(B0, B1));
+      const uint8x8_t C = vqmovn_u16(vcombine_u16(B0, B1));
      vst1_u8(dst + x_out, C);
      STORE_32x8(zero, zero, irow + x_out);
    }
    for (; x_out < x_out_max; ++x_out) {
      const int v = (int)MULT_FIX_C(irow[x_out], fxy_scale);
-      assert(v >= 0 && v <= 255);
-      dst[x_out] = v;
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
      irow[x_out] = 0;
    }
  }
@@ -225,35 +225,6 @@ static WEBP_INLINE void ProcessRow_SSE2(const __m128i* const A0,
  _mm_storel_epi64((__m128i*)dst, G);
 }

-static WEBP_INLINE void ProcessRow_Floor_SSE2(const __m128i* const A0,
-                                              const __m128i* const A1,
-                                              const __m128i* const A2,
-                                              const __m128i* const A3,
-                                              const __m128i* const mult,
-                                              uint8_t* const dst) {
-  const __m128i mask = _mm_set_epi32(0xffffffffu, 0, 0xffffffffu, 0);
-  const __m128i B0 = _mm_mul_epu32(*A0, *mult);
-  const __m128i B1 = _mm_mul_epu32(*A1, *mult);
-  const __m128i B2 = _mm_mul_epu32(*A2, *mult);
-  const __m128i B3 = _mm_mul_epu32(*A3, *mult);
-  const __m128i D0 = _mm_srli_epi64(B0, WEBP_RESCALER_RFIX);
-  const __m128i D1 = _mm_srli_epi64(B1, WEBP_RESCALER_RFIX);
-#if (WEBP_RESCALER_RFIX < 32)
-  const __m128i D2 =
-      _mm_and_si128(_mm_slli_epi64(B2, 32 - WEBP_RESCALER_RFIX), mask);
-  const __m128i D3 =
-      _mm_and_si128(_mm_slli_epi64(B3, 32 - WEBP_RESCALER_RFIX), mask);
-#else
-  const __m128i D2 = _mm_and_si128(B2, mask);
-  const __m128i D3 = _mm_and_si128(B3, mask);
-#endif
-  const __m128i E0 = _mm_or_si128(D0, D2);
-  const __m128i E1 = _mm_or_si128(D1, D3);
-  const __m128i F = _mm_packs_epi32(E0, E1);
-  const __m128i G = _mm_packus_epi16(F, F);
-  _mm_storel_epi64((__m128i*)dst, G);
-}
-
 static void RescalerExportRowExpand_SSE2(WebPRescaler* const wrk) {
  int x_out;
  uint8_t* const dst = wrk->dst;
@@ -274,8 +245,7 @@ static void RescalerExportRowExpand_SSE2(WebPRescaler* const wrk) {
    for (; x_out < x_out_max; ++x_out) {
      const uint32_t J = frow[x_out];
      const int v = (int)MULT_FIX(J, wrk->fy_scale);
-      assert(v >= 0 && v <= 255);
-      dst[x_out] = v;
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
    }
  } else {
    const uint32_t B = WEBP_RESCALER_FRAC(-wrk->y_accum, wrk->y_sub);
@@ -308,8 +278,7 @@ static void RescalerExportRowExpand_SSE2(WebPRescaler* const wrk) {
                       + (uint64_t)B * irow[x_out];
      const uint32_t J = (uint32_t)((I + ROUNDER) >> WEBP_RESCALER_RFIX);
      const int v = (int)MULT_FIX(J, wrk->fy_scale);
-      assert(v >= 0 && v <= 255);
-      dst[x_out] = v;
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
    }
  }
 }
@@ -328,20 +297,15 @@ static void RescalerExportRowShrink_SSE2(WebPRescaler* const wrk) {
    const int scale_xy = wrk->fxy_scale;
    const __m128i mult_xy = _mm_set_epi32(0, scale_xy, 0, scale_xy);
    const __m128i mult_y = _mm_set_epi32(0, yscale, 0, yscale);
-    const __m128i rounder = _mm_set_epi32(0, ROUNDER, 0, ROUNDER);
    for (x_out = 0; x_out + 8 <= x_out_max; x_out += 8) {
      __m128i A0, A1, A2, A3, B0, B1, B2, B3;
      LoadDispatchAndMult_SSE2(irow + x_out, NULL, &A0, &A1, &A2, &A3);
      LoadDispatchAndMult_SSE2(frow + x_out, &mult_y, &B0, &B1, &B2, &B3);
      {
-        const __m128i C0 = _mm_add_epi64(B0, rounder);
-        const __m128i C1 = _mm_add_epi64(B1, rounder);
-        const __m128i C2 = _mm_add_epi64(B2, rounder);
-        const __m128i C3 = _mm_add_epi64(B3, rounder);
-        const __m128i D0 = _mm_srli_epi64(C0, WEBP_RESCALER_RFIX);   // = frac
-        const __m128i D1 = _mm_srli_epi64(C1, WEBP_RESCALER_RFIX);
-        const __m128i D2 = _mm_srli_epi64(C2, WEBP_RESCALER_RFIX);
-        const __m128i D3 = _mm_srli_epi64(C3, WEBP_RESCALER_RFIX);
+        const __m128i D0 = _mm_srli_epi64(B0, WEBP_RESCALER_RFIX);   // = frac
+        const __m128i D1 = _mm_srli_epi64(B1, WEBP_RESCALER_RFIX);
+        const __m128i D2 = _mm_srli_epi64(B2, WEBP_RESCALER_RFIX);
+        const __m128i D3 = _mm_srli_epi64(B3, WEBP_RESCALER_RFIX);
        const __m128i E0 = _mm_sub_epi64(A0, D0);   // irow[x] - frac
        const __m128i E1 = _mm_sub_epi64(A1, D1);
        const __m128i E2 = _mm_sub_epi64(A2, D2);
@@ -352,14 +316,13 @@ static void RescalerExportRowShrink_SSE2(WebPRescaler* const wrk) {
        const __m128i G1 = _mm_or_si128(D1, F3);
        _mm_storeu_si128((__m128i*)(irow + x_out + 0), G0);
        _mm_storeu_si128((__m128i*)(irow + x_out + 4), G1);
-        ProcessRow_Floor_SSE2(&E0, &E1, &E2, &E3, &mult_xy, dst + x_out);
+        ProcessRow_SSE2(&E0, &E1, &E2, &E3, &mult_xy, dst + x_out);
      }
    }
    for (; x_out < x_out_max; ++x_out) {
-      const uint32_t frac = (int)MULT_FIX(frow[x_out], yscale);
-      const int v = (int)MULT_FIX_FLOOR(irow[x_out] - frac, wrk->fxy_scale);
-      assert(v >= 0 && v <= 255);
-      dst[x_out] = v;
+      const uint32_t frac = (int)MULT_FIX_FLOOR(frow[x_out], yscale);
+      const int v = (int)MULT_FIX(irow[x_out] - frac, wrk->fxy_scale);
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
      irow[x_out] = frac;   // new fractional start
    }
  } else {
@@ -375,8 +338,7 @@ static void RescalerExportRowShrink_SSE2(WebPRescaler* const wrk) {
    }
    for (; x_out < x_out_max; ++x_out) {
      const int v = (int)MULT_FIX(irow[x_out], scale);
-      assert(v >= 0 && v <= 255);
-      dst[x_out] = v;
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
      irow[x_out] = 0;
    }
  }
@@ -576,9 +576,9 @@ static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bot_y,        \
  const uint32_t l_uv = ((cur_u[0]) | ((cur_v[0]) << 16));               \
  const uint32_t uv0 = (3 * tl_uv + l_uv + 0x00020002u) >> 2;            \
  const uint8_t* ptop_y = &top_y[1];                                     \
-  uint8_t *ptop_dst = top_dst + XSTEP;                                   \
+  uint8_t* ptop_dst = top_dst + XSTEP;                                   \
  const uint8_t* pbot_y = &bot_y[1];                                     \
-  uint8_t *pbot_dst = bot_dst + XSTEP;                                   \
+  uint8_t* pbot_dst = bot_dst + XSTEP;                                   \
                                                                         \
  FUNC(top_y[0], uv0 & 0xff, (uv0 >> 16), top_dst);                      \
  if (bot_y != NULL) {                                                   \
@@ -58,8 +58,8 @@
 } while (0)

 // Turn the macro into a function for reducing code-size when non-critical
-static void Upsample16Pixels_NEON(const uint8_t *r1, const uint8_t *r2,
-                                  uint8_t *out) {
+static void Upsample16Pixels_NEON(const uint8_t* r1, const uint8_t* r2,
+                                  uint8_t* out) {
  UPSAMPLE_16PIXELS(r1, r2, out);
 }

@@ -190,14 +190,14 @@ static const int16_t kCoeffs1[4] = { 19077, 26149, 6419, 13320 };
 }

 #define NEON_UPSAMPLE_FUNC(FUNC_NAME, FMT, XSTEP)                       \
-static void FUNC_NAME(const uint8_t *top_y, const uint8_t *bottom_y,    \
-                      const uint8_t *top_u, const uint8_t *top_v,       \
-                      const uint8_t *cur_u, const uint8_t *cur_v,       \
-                      uint8_t *top_dst, uint8_t *bottom_dst, int len) { \
+static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bottom_y,    \
+                      const uint8_t* top_u, const uint8_t* top_v,       \
+                      const uint8_t* cur_u, const uint8_t* cur_v,       \
+                      uint8_t* top_dst, uint8_t* bottom_dst, int len) { \
  int block;                                                            \
  /* 16 byte aligned array to cache reconstructed u and v */            \
  uint8_t uv_buf[2 * 32 + 15];                                          \
-  uint8_t *const r_uv = (uint8_t*)((uintptr_t)(uv_buf + 15) & ~15);     \
+  uint8_t* const r_uv = (uint8_t*)((uintptr_t)(uv_buf + 15) & ~15);     \
  const int uv_len = (len + 1) >> 1;                                    \
  /* 9 pixels must be read-able for each block */                       \
  const int num_blocks = (uv_len - 1) >> 3;                             \
@@ -191,13 +191,14 @@ void VP8LHashChainClear(VP8LHashChain* const p) {

 // -----------------------------------------------------------------------------

-#define HASH_MULTIPLIER_HI (0xc6a4a793ULL)
-#define HASH_MULTIPLIER_LO (0x5bd1e996ULL)
+static const uint32_t kHashMultiplierHi = 0xc6a4a793u;
+static const uint32_t kHashMultiplierLo = 0x5bd1e996u;

-static WEBP_INLINE uint32_t GetPixPairHash64(const uint32_t* const argb) {
+static WEBP_UBSAN_IGNORE_UNSIGNED_OVERFLOW WEBP_INLINE
+uint32_t GetPixPairHash64(const uint32_t* const argb) {
  uint32_t key;
-  key  = (argb[1] * HASH_MULTIPLIER_HI) & 0xffffffffu;
-  key += (argb[0] * HASH_MULTIPLIER_LO) & 0xffffffffu;
+  key  = argb[1] * kHashMultiplierHi;
+  key += argb[0] * kHashMultiplierLo;
  key = key >> (32 - HASH_BITS);
  return key;
 }
@@ -641,7 +641,7 @@ static void HistogramAnalyzeEntropyBin(VP8LHistogramSet* const image_histo,
 // Merges some histograms with same bin_id together if it's advantageous.
 // Sets the remaining histograms to NULL.
 static void HistogramCombineEntropyBin(VP8LHistogramSet* const image_histo,
-                                       int *num_used,
+                                       int* num_used,
                                       const uint16_t* const clusters,
                                       uint16_t* const cluster_mappings,
                                       VP8LHistogram* cur_combo,
@@ -929,9 +929,8 @@ static int HistogramCombineStochastic(VP8LHistogramSet* const image_histo,
  }

  mappings = (int*) WebPSafeMalloc(*num_used, sizeof(*mappings));
-  if (mappings == NULL || !HistoQueueInit(&histo_queue, kHistoQueueSize)) {
-    goto End;
-  }
+  if (mappings == NULL) return 0;
+  if (!HistoQueueInit(&histo_queue, kHistoQueueSize)) goto End;
  // Fill the initial mapping.
  for (j = 0, iter = 0; iter < image_histo->size; ++iter) {
    if (histograms[iter] == NULL) continue;
@@ -29,11 +29,15 @@
 #define USE_INVERSE_ALPHA_TABLE

 #ifdef WORDS_BIGENDIAN
-#define ALPHA_OFFSET 0   // uint32_t 0xff000000 is 0xff,00,00,00 in memory
+// uint32_t 0xff000000 is 0xff,00,00,00 in memory
+#define CHANNEL_OFFSET(i) (i)
 #else
-#define ALPHA_OFFSET 3   // uint32_t 0xff000000 is 0x00,00,00,ff in memory
+// uint32_t 0xff000000 is 0x00,00,00,ff in memory
+#define CHANNEL_OFFSET(i) (3-(i))
 #endif

+#define ALPHA_OFFSET CHANNEL_OFFSET(0)
+
 //------------------------------------------------------------------------------
 // Detection of non-trivial transparency

@@ -997,10 +1001,10 @@ static int PictureARGBToYUVA(WebPPicture* picture, WebPEncCSP colorspace,
    return WebPEncodingSetError(picture, VP8_ENC_ERROR_INVALID_CONFIGURATION);
  } else {
    const uint8_t* const argb = (const uint8_t*)picture->argb;
-    const uint8_t* const a = argb + (0 ^ ALPHA_OFFSET);
-    const uint8_t* const r = argb + (1 ^ ALPHA_OFFSET);
-    const uint8_t* const g = argb + (2 ^ ALPHA_OFFSET);
-    const uint8_t* const b = argb + (3 ^ ALPHA_OFFSET);
+    const uint8_t* const a = argb + CHANNEL_OFFSET(0);
+    const uint8_t* const r = argb + CHANNEL_OFFSET(1);
+    const uint8_t* const g = argb + CHANNEL_OFFSET(2);
+    const uint8_t* const b = argb + CHANNEL_OFFSET(3);

    picture->colorspace = WEBP_YUV420;
    return ImportYUVAFromRGBA(r, g, b, a, 4, 4 * picture->argb_stride,
@@ -1050,7 +1054,7 @@ int WebPPictureYUVAToARGB(WebPPicture* picture) {
    const int height = picture->height;
    const int argb_stride = 4 * picture->argb_stride;
    uint8_t* dst = (uint8_t*)picture->argb;
-    const uint8_t *cur_u = picture->u, *cur_v = picture->v, *cur_y = picture->y;
+    const uint8_t* cur_u = picture->u, *cur_v = picture->v, *cur_y = picture->y;
    WebPUpsampleLinePairFunc upsample =
        WebPGetLinePairConverter(ALPHA_OFFSET > 0);

@@ -202,7 +202,7 @@ static uint32_t NearLossless(uint32_t value, uint32_t predict,
  }
  if ((value >> 24) == 0 || (value >> 24) == 0xff) {
    // Preserve transparency of fully transparent or fully opaque pixels.
-    a = NearLosslessDiff(value >> 24, predict >> 24);
+    a = NearLosslessDiff((value >> 24) & 0xff, (predict >> 24) & 0xff);
  } else {
    a = NearLosslessComponent(value >> 24, predict >> 24, 0xff, quantization);
  }
@@ -215,12 +215,12 @@ static uint32_t NearLossless(uint32_t value, uint32_t predict,
    // The amount by which green has been adjusted during quantization. It is
    // subtracted from red and blue for compensation, to avoid accumulating two
    // quantization errors in them.
-    green_diff = NearLosslessDiff(new_green, value >> 8);
+    green_diff = NearLosslessDiff(new_green, (value >> 8) & 0xff);
  }
-  r = NearLosslessComponent(NearLosslessDiff(value >> 16, green_diff),
+  r = NearLosslessComponent(NearLosslessDiff((value >> 16) & 0xff, green_diff),
                            (predict >> 16) & 0xff, 0xff - new_green,
                            quantization);
-  b = NearLosslessComponent(NearLosslessDiff(value, green_diff),
+  b = NearLosslessComponent(NearLosslessDiff(value & 0xff, green_diff),
                            predict & 0xff, 0xff - new_green, quantization);
  return ((uint32_t)a << 24) | ((uint32_t)r << 16) | ((uint32_t)g << 8) | b;
 }
@@ -587,7 +587,7 @@ static void GetBestGreenToRed(
      }
    }
  }
-  best_tx->green_to_red_ = green_to_red_best;
+  best_tx->green_to_red_ = (green_to_red_best & 0xff);
 }

 static float GetPredictionCostCrossColorBlue(
@@ -666,8 +666,8 @@ static void GetBestGreenRedToBlue(
      break;  // out of iter-loop.
    }
  }
-  best_tx->green_to_blue_ = green_to_blue_best;
-  best_tx->red_to_blue_ = red_to_blue_best;
+  best_tx->green_to_blue_ = green_to_blue_best & 0xff;
+  best_tx->red_to_blue_ = red_to_blue_best & 0xff;
 }
 #undef kGreenRedToBlueMaxIters
 #undef kGreenRedToBlueNumAxis
@@ -33,7 +33,7 @@

 // number of non-zero coeffs below which we consider the block very flat
 // (and apply a penalty to complex predictions)
-#define FLATNESS_LIMIT_I16 10      // I16 mode
+#define FLATNESS_LIMIT_I16 0       // I16 mode (special case)
 #define FLATNESS_LIMIT_I4  3       // I4 mode
 #define FLATNESS_LIMIT_UV  2       // UV mode
 #define FLATNESS_PENALTY   140     // roughly ~1bit per block
@@ -988,6 +988,7 @@ static void PickBestIntra16(VP8EncIterator* const it, VP8ModeScore* rd) {
  VP8ModeScore* rd_cur = &rd_tmp;
  VP8ModeScore* rd_best = rd;
  int mode;
+  int is_flat = IsFlatSource16(it->yuv_in_ + Y_OFF_ENC);

  rd->mode_i16 = -1;
  for (mode = 0; mode < NUM_PRED_MODES; ++mode) {
@@ -1003,10 +1004,14 @@ static void PickBestIntra16(VP8EncIterator* const it, VP8ModeScore* rd) {
        tlambda ? MULT_8B(tlambda, VP8TDisto16x16(src, tmp_dst, kWeightY)) : 0;
    rd_cur->H = VP8FixedCostsI16[mode];
    rd_cur->R = VP8GetCostLuma16(it, rd_cur);
-    if (mode > 0 &&
-        IsFlat(rd_cur->y_ac_levels[0], kNumBlocks, FLATNESS_LIMIT_I16)) {
-      // penalty to avoid flat area to be mispredicted by complex mode
-      rd_cur->R += FLATNESS_PENALTY * kNumBlocks;
+    if (is_flat) {
+      // refine the first impression (which was in pixel space)
+      is_flat = IsFlat(rd_cur->y_ac_levels[0], kNumBlocks, FLATNESS_LIMIT_I16);
+      if (is_flat) {
+        // Block is very flat. We put emphasis on the distortion being very low!
+        rd_cur->D *= 2;
+        rd_cur->SD *= 2;
+      }
    }

    // Since we always examine Intra16 first, we can overwrite *rd directly.
@@ -1087,7 +1092,8 @@ static int PickBestIntra4(VP8EncIterator* const it, VP8ModeScore* const rd) {
                  : 0;
      rd_tmp.H = mode_costs[mode];

-      // Add flatness penalty
+      // Add flatness penalty, to avoid flat area to be mispredicted
+      // by a complex mode.
      if (mode > 0 && IsFlat(tmp_levels, kNumBlocks, FLATNESS_LIMIT_I4)) {
        rd_tmp.R = FLATNESS_PENALTY * kNumBlocks;
      } else {
@@ -1242,11 +1248,19 @@ static void RefineUsingDistortion(VP8EncIterator* const it,
      if (mode > 0 && VP8FixedCostsI16[mode] > bit_limit) {
        continue;
      }
+
      if (score < best_score) {
        best_mode = mode;
        best_score = score;
      }
    }
+    if (it->x_ == 0 || it->y_ == 0) {
+      // avoid starting a checkerboard resonance from the border. See bug #432.
+      if (IsFlatSource16(src)) {
+        best_mode = (it->x_ == 0) ? 0 : 2;
+        try_both_modes = 0;  // stick to i16
+      }
+    }
    VP8SetIntra16Mode(it, best_mode);
    // we'll reconstruct later, if i16 mode actually gets selected
  }
@@ -31,8 +31,8 @@ extern "C" {

 // version numbers
 #define ENC_MAJ_VERSION 1
-#define ENC_MIN_VERSION 0
-#define ENC_REV_VERSION 2
+#define ENC_MIN_VERSION 1
+#define ENC_REV_VERSION 0

 enum { MAX_LF_LEVELS = 64,       // Maximum loop filter level
       MAX_VARIABLE_LEVEL = 67,  // last (inclusive) level with variable cost
@@ -249,7 +249,7 @@ typedef struct {
  int           percent0_;         // saved initial progress percent

  DError        left_derr_;        // left error diffusion (u/v)
-  DError       *top_derr_;         // top diffusion error - NULL if disabled
+  DError*       top_derr_;         // top diffusion error - NULL if disabled

  uint8_t* y_left_;    // left luma samples (addressable from index -1 to 15).
  uint8_t* u_left_;    // left u samples (addressable from index -1 to 7)
@@ -28,8 +28,8 @@ extern "C" {
 // Defines and constants.

 #define MUX_MAJ_VERSION 1
-#define MUX_MIN_VERSION 0
-#define MUX_REV_VERSION 2
+#define MUX_MIN_VERSION 1
+#define MUX_REV_VERSION 0

 // Chunk object.
 typedef struct WebPChunk WebPChunk;
@@ -100,7 +100,7 @@ static int MuxImageParse(const WebPChunk* const chunk, int copy_data,
                         WebPMuxImage* const wpi) {
  const uint8_t* bytes = chunk->data_.bytes;
  size_t size = chunk->data_.size;
-  const uint8_t* const last = bytes + size;
+  const uint8_t* const last = (bytes == NULL) ? NULL : bytes + size;
  WebPChunk subchunk;
  size_t subchunk_size;
  WebPChunk** unknown_chunk_list = &wpi->unknown_;
@@ -104,7 +104,8 @@ void VP8LoadNewBytes(VP8BitReader* const br) {
 }

 // Read a bit with proba 'prob'. Speed-critical function!
-static WEBP_INLINE int VP8GetBit(VP8BitReader* const br, int prob) {
+static WEBP_INLINE int VP8GetBit(VP8BitReader* const br,
+                                 int prob, const char label[]) {
  // Don't move this declaration! It makes a big speed difference to store
  // 'range' *before* calling VP8LoadNewBytes(), even if this function doesn't
  // alter br->range_ value.
@@ -129,13 +130,14 @@ static WEBP_INLINE int VP8GetBit(VP8BitReader* const br, int prob) {
      br->bits_ -= shift;
    }
    br->range_ = range - 1;
+    BT_TRACK(br);
    return bit;
  }
 }

 // simplified version of VP8GetBit() for prob=0x80 (note shift is always 1 here)
 static WEBP_UBSAN_IGNORE_UNSIGNED_OVERFLOW WEBP_INLINE
-int VP8GetSigned(VP8BitReader* const br, int v) {
+int VP8GetSigned(VP8BitReader* const br, int v, const char label[]) {
  if (br->bits_ < 0) {
    VP8LoadNewBytes(br);
  }
@@ -148,11 +150,13 @@ int VP8GetSigned(VP8BitReader* const br, int v) {
    br->range_ += mask;
    br->range_ |= 1;
    br->value_ -= (bit_t)((split + 1) & mask) << pos;
+    BT_TRACK(br);
    return (v ^ mask) - mask;
  }
 }

-static WEBP_INLINE int VP8GetBitAlt(VP8BitReader* const br, int prob) {
+static WEBP_INLINE int VP8GetBitAlt(VP8BitReader* const br,
+                                    int prob, const char label[]) {
  // Don't move this declaration! It makes a big speed difference to store
  // 'range' *before* calling VP8LoadNewBytes(), even if this function doesn't
  // alter br->range_ value.
@@ -179,6 +183,7 @@ static WEBP_INLINE int VP8GetBitAlt(VP8BitReader* const br, int prob) {
      br->bits_ -= shift;
    }
    br->range_ = range;
+    BT_TRACK(br);
    return bit;
  }
 }
@@ -102,17 +102,18 @@ void VP8LoadFinalBytes(VP8BitReader* const br) {
 //------------------------------------------------------------------------------
 // Higher-level calls

-uint32_t VP8GetValue(VP8BitReader* const br, int bits) {
+uint32_t VP8GetValue(VP8BitReader* const br, int bits, const char label[]) {
  uint32_t v = 0;
  while (bits-- > 0) {
-    v |= VP8GetBit(br, 0x80) << bits;
+    v |= VP8GetBit(br, 0x80, label) << bits;
  }
  return v;
 }

-int32_t VP8GetSignedValue(VP8BitReader* const br, int bits) {
-  const int value = VP8GetValue(br, bits);
-  return VP8Get(br) ? -value : value;
+int32_t VP8GetSignedValue(VP8BitReader* const br, int bits,
+                          const char label[]) {
+  const int value = VP8GetValue(br, bits, label);
+  return VP8Get(br, label) ? -value : value;
 }

 //------------------------------------------------------------------------------
@@ -220,3 +221,78 @@ uint32_t VP8LReadBits(VP8LBitReader* const br, int n_bits) {
 }

 //------------------------------------------------------------------------------
+// Bit-tracing tool
+
+#if (BITTRACE > 0)
+
+#include <stdlib.h>   // for atexit()
+#include <stdio.h>
+#include <string.h>
+
+#define MAX_NUM_LABELS 32
+static struct {
+  const char* label;
+  int size;
+  int count;
+} kLabels[MAX_NUM_LABELS];
+
+static int last_label = 0;
+static int last_pos = 0;
+static const uint8_t* buf_start = NULL;
+static int init_done = 0;
+
+static void PrintBitTraces(void) {
+  int i;
+  int scale = 1;
+  int total = 0;
+  const char* units = "bits";
+#if (BITTRACE == 2)
+  scale = 8;
+  units = "bytes";
+#endif
+  for (i = 0; i < last_label; ++i) total += kLabels[i].size;
+  if (total < 1) total = 1;   // avoid rounding errors
+  printf("=== Bit traces ===\n");
+  for (i = 0; i < last_label; ++i) {
+    const int skip = 16 - (int)strlen(kLabels[i].label);
+    const int value = (kLabels[i].size + scale - 1) / scale;
+    assert(skip > 0);
+    printf("%s \%*s: %6d %s   \t[%5.2f%%] [count: %7d]\n",
+           kLabels[i].label, skip, "", value, units,
+           100.f * kLabels[i].size / total,
+           kLabels[i].count);
+  }
+  total = (total + scale - 1) / scale;
+  printf("Total: %d %s\n", total, units);
+}
+
+void BitTrace(const struct VP8BitReader* const br, const char label[]) {
+  int i, pos;
+  if (!init_done) {
+    memset(kLabels, 0, sizeof(kLabels));
+    atexit(PrintBitTraces);
+    buf_start = br->buf_;
+    init_done = 1;
+  }
+  pos = (int)(br->buf_ - buf_start) * 8 - br->bits_;
+  // if there's a too large jump, we've changed partition -> reset counter
+  if (abs(pos - last_pos) > 32) {
+    buf_start = br->buf_;
+    pos = 0;
+    last_pos = 0;
+  }
+  if (br->range_ >= 0x7f) pos += kVP8Log2Range[br->range_ - 0x7f];
+  for (i = 0; i < last_label; ++i) {
+    if (!strcmp(label, kLabels[i].label)) break;
+  }
+  if (i == MAX_NUM_LABELS) abort();   // overflow!
+  kLabels[i].label = label;
+  kLabels[i].size += pos - last_pos;
+  kLabels[i].count += 1;
+  if (i == last_label) ++last_label;
+  last_pos = pos;
+}
+
+#endif  // BITTRACE > 0
+
+//------------------------------------------------------------------------------
@@ -21,6 +21,27 @@
 #endif
 #include "src/webp/types.h"

+// Warning! This macro triggers quite some MACRO wizardry around func signature!
+#if !defined(BITTRACE)
+#define BITTRACE 0    // 0 = off, 1 = print bits, 2 = print bytes
+#endif
+
+#if (BITTRACE > 0)
+struct VP8BitReader;
+extern void BitTrace(const struct VP8BitReader* const br, const char label[]);
+#define BT_TRACK(br) BitTrace(br, label)
+#define VP8Get(BR, L) VP8GetValue(BR, 1, L)
+#else
+#define BT_TRACK(br)
+// We'll REMOVE the 'const char label[]' from all signatures and calls (!!):
+#define VP8GetValue(BR, N, L) VP8GetValue(BR, N)
+#define VP8Get(BR, L) VP8GetValue(BR, 1, L)
+#define VP8GetSignedValue(BR, N, L) VP8GetSignedValue(BR, N)
+#define VP8GetBit(BR, P, L) VP8GetBit(BR, P)
+#define VP8GetBitAlt(BR, P, L) VP8GetBitAlt(BR, P)
+#define VP8GetSigned(BR, V, L) VP8GetSigned(BR, V)
+#endif
+
 #ifdef __cplusplus
 extern "C" {
 #endif
@@ -92,17 +113,15 @@ void VP8BitReaderSetBuffer(VP8BitReader* const br,
 void VP8RemapBitReader(VP8BitReader* const br, ptrdiff_t offset);

 // return the next value made of 'num_bits' bits
-uint32_t VP8GetValue(VP8BitReader* const br, int num_bits);
-static WEBP_INLINE uint32_t VP8Get(VP8BitReader* const br) {
-  return VP8GetValue(br, 1);
-}
+uint32_t VP8GetValue(VP8BitReader* const br, int num_bits, const char label[]);

 // return the next value with sign-extension.
-int32_t VP8GetSignedValue(VP8BitReader* const br, int num_bits);
+int32_t VP8GetSignedValue(VP8BitReader* const br, int num_bits,
+                          const char label[]);

 // bit_reader_inl.h will implement the following methods:
-//   static WEBP_INLINE int VP8GetBit(VP8BitReader* const br, int prob)
-//   static WEBP_INLINE int VP8GetSigned(VP8BitReader* const br, int v)
+//   static WEBP_INLINE int VP8GetBit(VP8BitReader* const br, int prob, ...)
+//   static WEBP_INLINE int VP8GetSigned(VP8BitReader* const br, int v, ...)
 // and should be included by the .c files that actually need them.
 // This is to avoid recompiling the whole library whenever this file is touched,
 // and also allowing platform-specific ad-hoc hacks.
@@ -70,7 +70,7 @@ static void Flush(VP8BitWriter* const bw) {
      const int value = (bits & 0x100) ? 0x00 : 0xff;
      for (; bw->run_ > 0; --bw->run_) bw->buf_[pos++] = value;
    }
-    bw->buf_[pos++] = bits;
+    bw->buf_[pos++] = bits & 0xff;
    bw->pos_ = pos;
  } else {
    bw->run_++;   // delay writing of bytes 0xff, pending eventual carry.
@@ -17,6 +17,7 @@

 #include <assert.h>

+#include "src/dsp/dsp.h"
 #include "src/webp/types.h"

 #ifdef __cplusplus
@@ -25,15 +26,16 @@ extern "C" {

 // Main color cache struct.
 typedef struct {
-  uint32_t *colors_;  // color entries
+  uint32_t* colors_;  // color entries
  int hash_shift_;    // Hash shift: 32 - hash_bits_.
  int hash_bits_;
 } VP8LColorCache;

-static const uint64_t kHashMul = 0x1e35a7bdull;
+static const uint32_t kHashMul = 0x1e35a7bdu;

-static WEBP_INLINE int VP8LHashPix(uint32_t argb, int shift) {
-  return (int)(((argb * kHashMul) & 0xffffffffu) >> shift);
+static WEBP_UBSAN_IGNORE_UNSIGNED_OVERFLOW WEBP_INLINE
+int VP8LHashPix(uint32_t argb, int shift) {
+  return (int)((argb * kHashMul) >> shift);
 }

 static WEBP_INLINE uint32_t VP8LColorCacheLookup(
@@ -91,7 +91,8 @@ static int BuildHuffmanTable(HuffmanCode* const root_table, int root_bits,

  assert(code_lengths_size != 0);
  assert(code_lengths != NULL);
-  assert(root_table != NULL);
+  assert((root_table != NULL && sorted != NULL) ||
+         (root_table == NULL && sorted == NULL));
  assert(root_bits > 0);

  // Build histogram of code lengths.
@@ -120,16 +121,22 @@ static int BuildHuffmanTable(HuffmanCode* const root_table, int root_bits,
  for (symbol = 0; symbol < code_lengths_size; ++symbol) {
    const int symbol_code_length = code_lengths[symbol];
    if (code_lengths[symbol] > 0) {
-      sorted[offset[symbol_code_length]++] = symbol;
+      if (sorted != NULL) {
+        sorted[offset[symbol_code_length]++] = symbol;
+      } else {
+        offset[symbol_code_length]++;
+      }
    }
  }

  // Special case code with only one value.
  if (offset[MAX_ALLOWED_CODE_LENGTH] == 1) {
-    HuffmanCode code;
-    code.bits = 0;
-    code.value = (uint16_t)sorted[0];
-    ReplicateValue(table, 1, total_size, code);
+    if (sorted != NULL) {
+      HuffmanCode code;
+      code.bits = 0;
+      code.value = (uint16_t)sorted[0];
+      ReplicateValue(table, 1, total_size, code);
+    }
    return total_size;
  }

@@ -151,6 +158,7 @@ static int BuildHuffmanTable(HuffmanCode* const root_table, int root_bits,
      if (num_open < 0) {
        return 0;
      }
+      if (root_table == NULL) continue;
      for (; count[len] > 0; --count[len]) {
        HuffmanCode code;
        code.bits = (uint8_t)len;
@@ -169,6 +177,7 @@ static int BuildHuffmanTable(HuffmanCode* const root_table, int root_bits,
      if (num_open < 0) {
        return 0;
      }
+      if (root_table == NULL) continue;
      for (; count[len] > 0; --count[len]) {
        HuffmanCode code;
        if ((key & mask) != low) {
@@ -206,7 +215,10 @@ int VP8LBuildHuffmanTable(HuffmanCode* const root_table, int root_bits,
                          const int code_lengths[], int code_lengths_size) {
  int total_size;
  assert(code_lengths_size <= MAX_CODE_LENGTHS_SIZE);
-  if (code_lengths_size <= SORTED_SIZE_CUTOFF) {
+  if (root_table == NULL) {
+    total_size = BuildHuffmanTable(NULL, root_bits,
+                                   code_lengths, code_lengths_size, NULL);
+  } else if (code_lengths_size <= SORTED_SIZE_CUTOFF) {
    // use local stack-allocated array.
    uint16_t sorted[SORTED_SIZE_CUTOFF];
    total_size = BuildHuffmanTable(root_table, root_bits,
@@ -78,6 +78,8 @@ void VP8LHtreeGroupsFree(HTreeGroup* const htree_groups);
 // the huffman table.
 // Returns built table size or 0 in case of error (invalid tree or
 // memory error).
+// If root_table is NULL, it returns 0 if a lookup cannot be built, something
+// > 0 otherwise (but not the table size).
 int VP8LBuildHuffmanTable(HuffmanCode* const root_table, int root_bits,
                          const int code_lengths[], int code_lengths_size);

@@ -84,14 +84,14 @@ int WebPRescalerGetScaledDimensions(int src_width, int src_height,
    int height = *scaled_height;

    // if width is unspecified, scale original proportionally to height ratio.
-    if (width == 0) {
+    if (width == 0 && src_height > 0) {
      width =
-          (int)(((uint64_t)src_width * height + src_height / 2) / src_height);
+          (int)(((uint64_t)src_width * height + src_height - 1) / src_height);
    }
    // if height is unspecified, scale original proportionally to width ratio.
-    if (height == 0) {
+    if (height == 0 && src_width > 0) {
      height =
-          (int)(((uint64_t)src_height * width + src_width / 2) / src_width);
+          (int)(((uint64_t)src_height * width + src_width - 1) / src_width);
    }
    // Check if the overall dimensions still make sense.
    if (width <= 0 || height <= 0) {
@@ -73,7 +73,7 @@ typedef struct {
 #endif

 static int pthread_create(pthread_t* const thread, const void* attr,
-                          unsigned int (__stdcall *start)(void*), void* arg) {
+                          unsigned int (__stdcall* start)(void*), void* arg) {
  (void)attr;
 #ifdef USE_CREATE_THREAD
  *thread = CreateThread(NULL,   /* lpThreadAttributes */
@@ -217,8 +217,12 @@ static THREADFN ThreadLoop(void* ptr) {
      done = 1;
    }
    // signal to the main thread that we're done (for Sync())
-    pthread_cond_signal(&impl->condition_);
+    // Note the associated mutex does not need to be held when signaling the
+    // condition. Unlocking the mutex first may improve performance in some
+    // implementations, avoiding the case where the waiting thread can't
+    // reacquire the mutex when woken.
    pthread_mutex_unlock(&impl->mutex_);
+    pthread_cond_signal(&impl->condition_);
  }
  return THREAD_RETURN(NULL);    // Thread is finished
 }
@@ -240,7 +244,13 @@ static void ChangeState(WebPWorker* const worker, WebPWorkerStatus new_status) {
    // assign new status and release the working thread if needed
    if (new_status != OK) {
      worker->status_ = new_status;
+      // Note the associated mutex does not need to be held when signaling the
+      // condition. Unlocking the mutex first may improve performance in some
+      // implementations, avoiding the case where the waiting thread can't
+      // reacquire the mutex when woken.
+      pthread_mutex_unlock(&impl->mutex_);
      pthread_cond_signal(&impl->condition_);
+      return;
    }
  }
  pthread_mutex_unlock(&impl->mutex_);
@@ -216,9 +216,14 @@ void WebPSafeFree(void* const ptr) {
  free(ptr);
 }

-// Public API function.
+// Public API functions.
+
+void* WebPMalloc(size_t size) {
+  return WebPSafeMalloc(1, size);
+}
+
 void WebPFree(void* ptr) {
-  free(ptr);
+  WebPSafeFree(ptr);
 }

 //------------------------------------------------------------------------------
@@ -92,14 +92,14 @@ static WEBP_INLINE uint32_t GetLE32(const uint8_t* const data) {
 // Store 16, 24 or 32 bits in little-endian order.
 static WEBP_INLINE void PutLE16(uint8_t* const data, int val) {
  assert(val < (1 << 16));
-  data[0] = (val >> 0);
-  data[1] = (val >> 8);
+  data[0] = (val >> 0) & 0xff;
+  data[1] = (val >> 8) & 0xff;
 }

 static WEBP_INLINE void PutLE24(uint8_t* const data, int val) {
  assert(val < (1 << 24));
  PutLE16(data, val & 0xffff);
-  data[2] = (val >> 16);
+  data[2] = (val >> 16) & 0xff;
 }

 static WEBP_INLINE void PutLE32(uint8_t* const data, uint32_t val) {
@@ -20,7 +20,7 @@
 extern "C" {
 #endif

-#define WEBP_DECODER_ABI_VERSION 0x0208    // MAJOR(8b) + MINOR(8b)
+#define WEBP_DECODER_ABI_VERSION 0x0209    // MAJOR(8b) + MINOR(8b)

 // Note: forward declaring enumerations is not allowed in (strict) C and C++,
 // the types are left here for reference.
@@ -91,9 +91,6 @@ WEBP_EXTERN uint8_t* WebPDecodeYUV(const uint8_t* data, size_t data_size,
                                   uint8_t** u, uint8_t** v,
                                   int* stride, int* uv_stride);

-// Releases memory returned by the WebPDecode*() functions above.
-WEBP_EXTERN void WebPFree(void* ptr);
-
 // These five functions are variants of the above ones, that decode the image
 // directly into a pre-allocated buffer 'output_buffer'. The maximum storage
 // available in this buffer is indicated by 'output_buffer_size'. If this
@@ -20,7 +20,7 @@
 extern "C" {
 #endif

-#define WEBP_ENCODER_ABI_VERSION 0x020e    // MAJOR(8b) + MINOR(8b)
+#define WEBP_ENCODER_ABI_VERSION 0x020f    // MAJOR(8b) + MINOR(8b)

 // Note: forward declaring enumerations is not allowed in (strict) C and C++,
 // the types are left here for reference.
@@ -62,6 +62,10 @@ WEBP_EXTERN size_t WebPEncodeBGRA(const uint8_t* bgra,
 // These functions are the equivalent of the above, but compressing in a
 // lossless manner. Files are usually larger than lossy format, but will
 // not suffer any compression loss.
+// Note these functions, like the lossy versions, use the library's default
+// settings. For lossless this means 'exact' is disabled. RGB values in
+// transparent areas will be modified to improve compression. To avoid this,
+// use WebPEncode() and set WebPConfig::exact to 1.
 WEBP_EXTERN size_t WebPEncodeLosslessRGB(const uint8_t* rgb,
                                         int width, int height, int stride,
                                         uint8_t** output);
@@ -75,9 +79,6 @@ WEBP_EXTERN size_t WebPEncodeLosslessBGRA(const uint8_t* bgra,
                                          int width, int height, int stride,
                                          uint8_t** output);

-// Releases memory returned by the WebPEncode*() functions above.
-WEBP_EXTERN void WebPFree(void* ptr);
-
 //------------------------------------------------------------------------------
 // Coding parameters

@@ -302,7 +303,7 @@ struct WebPPicture {
  // YUV input (mostly used for input to lossy compression)
  WebPEncCSP colorspace;     // colorspace: should be YUV420 for now (=Y'CbCr).
  int width, height;         // dimensions (less or equal to WEBP_MAX_DIMENSION)
-  uint8_t *y, *u, *v;        // pointers to luma/chroma planes.
+  uint8_t* y, *u, *v;        // pointers to luma/chroma planes.
  int y_stride, uv_stride;   // luma/chroma strides.
  uint8_t* a;                // pointer to the alpha plane
  int a_stride;              // stride of the alpha plane
@@ -346,7 +347,7 @@ struct WebPPicture {
  uint32_t pad3[3];       // padding for later use

  // Unused for now
-  uint8_t *pad4, *pad5;
+  uint8_t* pad4, *pad5;
  uint32_t pad6[8];       // padding for later use

  // PRIVATE FIELDS
@@ -57,7 +57,7 @@ extern "C" {
  WebPMuxGetChunk(mux, "ICCP", &icc_profile);
  // ... (Consume icc_data).
  WebPMuxDelete(mux);
-  free(data);
+  WebPFree(data);
 */

 // Note: forward declaring enumerations is not allowed in (strict) C and C++,
@@ -245,7 +245,7 @@ WEBP_EXTERN WebPMuxError WebPMuxPushFrame(
    WebPMux* mux, const WebPMuxFrameInfo* frame, int copy_data);

 // Gets the nth frame from the mux object.
-// The content of 'frame->bitstream' is allocated using malloc(), and NOT
+// The content of 'frame->bitstream' is allocated using WebPMalloc(), and NOT
 // owned by the 'mux' object. It MUST be deallocated by the caller by calling
 // WebPDataClear().
 // nth=0 has a special meaning - last position.
@@ -376,10 +376,10 @@ WEBP_EXTERN WebPMuxError WebPMuxNumChunks(const WebPMux* mux,
 // Assembles all chunks in WebP RIFF format and returns in 'assembled_data'.
 // This function also validates the mux object.
 // Note: The content of 'assembled_data' will be ignored and overwritten.
-// Also, the content of 'assembled_data' is allocated using malloc(), and NOT
-// owned by the 'mux' object. It MUST be deallocated by the caller by calling
-// WebPDataClear(). It's always safe to call WebPDataClear() upon return,
-// even in case of error.
+// Also, the content of 'assembled_data' is allocated using WebPMalloc(), and
+// NOT owned by the 'mux' object. It MUST be deallocated by the caller by
+// calling WebPDataClear(). It's always safe to call WebPDataClear() upon
+// return, even in case of error.
 // Parameters:
 //   mux - (in/out) object whose chunks are to be assembled
 //   assembled_data - (out) assembled WebP data
@@ -14,7 +14,6 @@
 #ifndef WEBP_WEBP_MUX_TYPES_H_
 #define WEBP_WEBP_MUX_TYPES_H_

-#include <stdlib.h>  // free()
 #include <string.h>  // memset()
 #include "./types.h"

@@ -56,6 +55,7 @@ typedef enum WebPMuxAnimBlend {

 // Data type used to describe 'raw' data, e.g., chunk data
 // (ICC profile, metadata) and WebP compressed image data.
+// 'bytes' memory must be allocated using WebPMalloc() and such.
 struct WebPData {
  const uint8_t* bytes;
  size_t size;
@@ -68,11 +68,11 @@ static WEBP_INLINE void WebPDataInit(WebPData* webp_data) {
  }
 }

-// Clears the contents of the 'webp_data' object by calling free(). Does not
-// deallocate the object itself.
+// Clears the contents of the 'webp_data' object by calling WebPFree().
+// Does not deallocate the object itself.
 static WEBP_INLINE void WebPDataClear(WebPData* webp_data) {
  if (webp_data != NULL) {
-    free((void*)webp_data->bytes);
+    WebPFree((void*)webp_data->bytes);
    WebPDataInit(webp_data);
  }
 }
@@ -83,7 +83,7 @@ static WEBP_INLINE int WebPDataCopy(const WebPData* src, WebPData* dst) {
  if (src == NULL || dst == NULL) return 0;
  WebPDataInit(dst);
  if (src->bytes != NULL && src->size != 0) {
-    dst->bytes = (uint8_t*)malloc(src->size);
+    dst->bytes = (uint8_t*)WebPMalloc(src->size);
    if (dst->bytes == NULL) return 0;
    memcpy((void*)dst->bytes, src->bytes, src->size);
    dst->size = src->size;
@@ -7,7 +7,7 @@
 // be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
-//  Common types
+//  Common types + memory wrappers
 //
 // Author: Skal (pascal.massimino@gmail.com)

@@ -49,4 +49,20 @@ typedef long long int int64_t;
 // Macro to check ABI compatibility (same major revision number)
 #define WEBP_ABI_IS_INCOMPATIBLE(a, b) (((a) >> 8) != ((b) >> 8))

+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// Allocates 'size' bytes of memory. Returns NULL upon error. Memory
+// must be deallocated by calling WebPFree(). This function is made available
+// by the core 'libwebp' library.
+WEBP_EXTERN void* WebPMalloc(size_t size);
+
+// Releases memory returned by the WebPDecode*() functions (from decode.h).
+WEBP_EXTERN void WebPFree(void* ptr);
+
+#ifdef __cplusplus
+}    // extern "C"
+#endif
+
 #endif  // WEBP_WEBP_TYPES_H_
@@ -63,7 +63,7 @@ Attribute::~Attribute () {}

 namespace {

-struct NameCompare: std::binary_function <const char *, const char *, bool>
+struct NameCompare
 {
    bool
    operator () (const char *x, const char *y) const
@@ -45,7 +45,7 @@


 // GCC and Visual Studio SSE2 compiler flags
-#if defined __SSE2__ || (_MSC_VER >= 1300 && !_M_CEE_PURE)
+#if defined __SSE2__ || (defined(_M_X64) || _M_IX86_FP == 2)
    #define IMF_HAVE_SSE2 1
 #endif

@@ -60,6 +60,7 @@
 #include "ImathNamespace.h"

 #include <iostream>
+#include <algorithm>

 IMATH_INTERNAL_NAMESPACE_HEADER_ENTER

--- a/Show More
+++ b/Show More