Attachment 357696 Details for Bug 192864 – Patch

[patch] Patch

bug-192864-20181219110903.patch (text/plain), 891.42 KB, created by youenn fablet on 2018-12-19 11:09:06 PST

(hide)

Description:

Filename:

MIME Type:

Creator: youenn fablet

Created: 2018-12-19 11:09:06 PST

Size: 891.42 KB

patch

obsolete

>Subversion Revision: 239282
>diff --git a/Source/ThirdParty/libwebrtc/ChangeLog b/Source/ThirdParty/libwebrtc/ChangeLog
>index cc9deee9c607b733af24f0872d058bac84420cd0..308e447e9f82b26114ba8b2c488a9e39f6245011 100644
>--- a/Source/ThirdParty/libwebrtc/ChangeLog
>+++ b/Source/ThirdParty/libwebrtc/ChangeLog
>@@ -1,3 +1,12 @@
>+2018-12-19  Youenn Fablet  <youenn@apple.com>
>+
>+        Refresh libyuv to M72
>+        https://bugs.webkit.org/show_bug.cgi?id=192864
>+
>+        Reviewed by NOBODY (OOPS!).
>+
>+        * Source/third_party/libyuv: Resynced.
>+
> 2018-12-19  Youenn Fablet  <youenn@apple.com>
> 
>         Refresh usrsctplib to M72
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/.gn b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/.gn
>index 26a4a62d1994e5abd6f534a9b6b20b71bacb07e9..fc66637f362a70b5c2321c86888fa824fd5bac0c 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/.gn
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/.gn
>@@ -29,7 +29,7 @@ exec_script_whitelist = build_dotfile_settings.exec_script_whitelist +
>                         [ "//build_overrides/build.gni" ]
> 
> default_args = {
>-  mac_sdk_min = "10.11"
>+  mac_sdk_min = "10.12"
> 
>   # LibYUV does not want to switch to C++14 yet.
>   use_cxx11 = true
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/.vpython b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/.vpython
>index cf74427542e4b6a1dfb2243d4ad58b175acf9964..defaf5df722ace398fe39febdf7086f5f3ad79cd 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/.vpython
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/.vpython
>@@ -30,3 +30,16 @@ wheel: <
>   name: "infra/python/wheels/psutil/${platform}_${py_python}_${py_abi}"
>   version: "version:5.2.2"
> >
>+
>+# Used by:
>+#   third_party/catapult
>+wheel: <
>+  name: "infra/python/wheels/pypiwin32/${vpython_platform}"
>+  version: "version:219"
>+  match_tag: <
>+    platform: "win32"
>+  >
>+  match_tag: <
>+    platform: "win_amd64"
>+  >
>+>
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/Android.bp b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/Android.bp
>index 7d95a7865abe182356dcb849efaa52c019988f6f..691686aad330fad7acd62f83a19f1ee7f2f363b0 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/Android.bp
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/Android.bp
>@@ -11,6 +11,7 @@ cc_library {
>         "source/compare_gcc.cc",
>         "source/compare_neon.cc",
>         "source/compare_neon64.cc",
>+        "source/compare_mmi.cc",
>         "source/compare_msa.cc",
>         "source/convert.cc",
>         "source/convert_argb.cc",
>@@ -25,12 +26,14 @@ cc_library {
>         "source/rotate_argb.cc",
>         "source/rotate_common.cc",
>         "source/rotate_gcc.cc",
>+        "source/rotate_mmi.cc",
>         "source/rotate_msa.cc",
>         "source/rotate_neon.cc",
>         "source/rotate_neon64.cc",
>         "source/row_any.cc",
>         "source/row_common.cc",
>         "source/row_gcc.cc",
>+        "source/row_mmi.cc",
>         "source/row_msa.cc",
>         "source/row_neon.cc",
>         "source/row_neon64.cc",
>@@ -39,11 +42,11 @@ cc_library {
>         "source/scale_argb.cc",
>         "source/scale_common.cc",
>         "source/scale_gcc.cc",
>+        "source/scale_mmi.cc",
>         "source/scale_msa.cc",
>         "source/scale_neon.cc",
>         "source/scale_neon64.cc",
>         "source/video_common.cc",
>-
>         "source/convert_jpeg.cc",
>         "source/mjpeg_decoder.cc",
>         "source/mjpeg_validate.cc",
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/Android.mk b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/Android.mk
>index dbc6cad37ab3ca5b09ea342639935ef48f1dae50..0975d64a3860695947680e5e63db45efe6dbc0c0 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/Android.mk
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/Android.mk
>@@ -9,6 +9,7 @@ LOCAL_SRC_FILES := \
>     source/compare.cc           \
>     source/compare_common.cc    \
>     source/compare_gcc.cc       \
>+    source/compare_mmi.cc       \
>     source/compare_msa.cc       \
>     source/compare_neon.cc      \
>     source/compare_neon64.cc    \
>@@ -25,12 +26,14 @@ LOCAL_SRC_FILES := \
>     source/rotate_argb.cc       \
>     source/rotate_common.cc     \
>     source/rotate_gcc.cc        \
>+    source/rotate_mmi.cc        \
>     source/rotate_msa.cc        \
>     source/rotate_neon.cc       \
>     source/rotate_neon64.cc     \
>     source/row_any.cc           \
>     source/row_common.cc        \
>     source/row_gcc.cc           \
>+    source/row_mmi.cc           \
>     source/row_msa.cc           \
>     source/row_neon.cc          \
>     source/row_neon64.cc        \
>@@ -39,6 +42,7 @@ LOCAL_SRC_FILES := \
>     source/scale_argb.cc        \
>     source/scale_common.cc      \
>     source/scale_gcc.cc         \
>+    source/scale_mmi.cc         \
>     source/scale_msa.cc         \
>     source/scale_neon.cc        \
>     source/scale_neon64.cc      \
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/BUILD.gn b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/BUILD.gn
>index 03ce499e6d9db8d2decdbadd94f877b403a24d6d..8904fd6c69913235520ae97fac7cd10d9217dc07 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/BUILD.gn
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/BUILD.gn
>@@ -69,6 +69,10 @@ group("libyuv") {
>     deps += [ ":libyuv_msa" ]
>   }
> 
>+  if (libyuv_use_mmi) {
>+    deps += [ ":libyuv_mmi" ]
>+  }
>+
>   if (!is_ios) {
>     # Make sure that clients of libyuv link with libjpeg. This can't go in
>     # libyuv_internal because in Windows x64 builds that will generate a clang
>@@ -229,6 +233,24 @@ if (libyuv_use_msa) {
>   }
> }
> 
>+if (libyuv_use_mmi) {
>+  static_library("libyuv_mmi") {
>+    sources = [
>+      # MMI Source Files
>+      "source/compare_mmi.cc",
>+      "source/rotate_mmi.cc",
>+      "source/row_mmi.cc",
>+      "source/scale_mmi.cc",
>+    ]
>+
>+    deps = [
>+      ":libyuv_internal",
>+    ]
>+
>+    public_configs = [ ":libyuv_config" ]
>+  }
>+}
>+
> if (libyuv_include_tests) {
>   config("libyuv_unittest_warnings_config") {
>     if (!is_win) {
>@@ -329,7 +351,6 @@ if (libyuv_include_tests) {
>     ]
>     deps = [
>       ":libyuv",
>-      "//build/config:exe_and_shlib_deps",  # for asan on llvm libc++
>     ]
>     if (is_linux) {
>       cflags = [ "-fexceptions" ]
>@@ -343,7 +364,6 @@ if (libyuv_include_tests) {
>     ]
>     deps = [
>       ":libyuv",
>-      "//build/config:exe_and_shlib_deps",  # for new[] on llvm libc++
>     ]
>     if (is_linux) {
>       cflags = [ "-fexceptions" ]
>@@ -359,7 +379,6 @@ if (libyuv_include_tests) {
>     ]
>     deps = [
>       ":libyuv",
>-      "//build/config:exe_and_shlib_deps",  # for new[] on llvm libc++
>     ]
> 
>     if (!is_ios && !libyuv_disable_jpeg) {
>@@ -374,7 +393,6 @@ if (libyuv_include_tests) {
>     ]
>     deps = [
>       ":libyuv",
>-      "//build/config:exe_and_shlib_deps",  # for asan on llvm libc++
>     ]
>   }
> }
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/CMakeLists.txt b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/CMakeLists.txt
>index 6420371de98314b82ec39f541c37c8a82abcec48..ed4948f066e23c927cf02a87c63c8b1b99a0931c 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/CMakeLists.txt
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/CMakeLists.txt
>@@ -59,7 +59,11 @@ if(TEST)
>   endif()
> 
>   add_executable(libyuv_unittest ${ly_unittest_sources})
>-  target_link_libraries(libyuv_unittest ${ly_lib_name} ${GTEST_LIBRARY} pthread)
>+  target_link_libraries(libyuv_unittest ${ly_lib_name} ${GTEST_LIBRARY})
>+  find_library(PTHREAD_LIBRARY pthread)
>+  if(NOT PTHREAD_LIBRARY STREQUAL "PTHREAD_LIBRARY-NOTFOUND")
>+    target_link_libraries(libyuv_unittest pthread)
>+  endif()
>   if (JPEG_FOUND)
>     target_link_libraries(libyuv_unittest ${JPEG_LIBRARY})
>   endif()
>@@ -68,14 +72,18 @@ if(TEST)
>     target_link_libraries(libyuv_unittest glibc-compat)
>   endif()
> 
>-  target_link_libraries(libyuv_unittest gflags)
>+  find_library(GFLAGS_LIBRARY gflags)
>+  if(NOT GFLAGS_LIBRARY STREQUAL "GFLAGS_LIBRARY-NOTFOUND")
>+    target_link_libraries(libyuv_unittest gflags)
>+    add_definitions(-DLIBYUV_USE_GFLAGS)
>+  endif()
> endif()
> 
> 
> # install the conversion tool, .so, .a, and all the header files
> INSTALL ( PROGRAMS ${CMAKE_BINARY_DIR}/yuvconvert			DESTINATION bin )
> INSTALL ( TARGETS ${ly_lib_static}						DESTINATION lib )
>-INSTALL ( TARGETS ${ly_lib_shared} LIBRARY				DESTINATION lib )
>+INSTALL ( TARGETS ${ly_lib_shared} LIBRARY				DESTINATION lib RUNTIME DESTINATION bin )
> INSTALL ( DIRECTORY ${PROJECT_SOURCE_DIR}/include/		DESTINATION include )
> 
> # create the .deb and .rpm packages using cpack
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/DEPS b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/DEPS
>index 266327470b7c0fa2c3631a3d0b28439e18c7d253..ac20e06276c867c6003962cf68c544bb04f16112 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/DEPS
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/DEPS
>@@ -1,7 +1,7 @@
> vars = {
>   'chromium_git': 'https://chromium.googlesource.com',
>-  'chromium_revision': '80f0c5570649c35a869429b2ab8c381a0a7246cb',
>-  'swarming_revision': '88229872dd17e71658fe96763feaa77915d8cbd6',
>+  'chromium_revision': '35b72bf255d6519506b7e732f9c74205d2ab452d',
>+  'swarming_revision': '486c9b53c4d54dd4b95bb6ce0e31160e600dfc11',
>   # Three lines of non-changing comments so that
>   # the commit queue can handle CLs rolling lss
>   # and whatever else without interference from each other.
>@@ -9,34 +9,36 @@ vars = {
>   # Three lines of non-changing comments so that
>   # the commit queue can handle CLs rolling catapult
>   # and whatever else without interference from each other.
>-  'catapult_revision': 'e7298f36f7912f2caa122086cfbe71734d04b73f',
>+  'catapult_revision': '0d25dda9b148bcd2dad9e1080b1dc57eaf9d2c2a',
> }
> 
> deps = {
>   'src/build':
>-    Var('chromium_git') + '/chromium/src/build' + '@' + '39738e75b27f39d4c0030a0b11d5d2ddd34715f7',
>+    Var('chromium_git') + '/chromium/src/build' + '@' + 'f79db013c75bff172913707cd762eba847838fea',
>   'src/buildtools':
>-    Var('chromium_git') + '/chromium/buildtools.git' + '@' + 'a09e064635a49f08e585e3b173d5fbc3dd3f485e',
>+    Var('chromium_git') + '/chromium/buildtools.git' + '@' + '9a90d9aaadeb5e04327ed05775f45132e4b3523f',
>   'src/testing':
>-    Var('chromium_git') + '/chromium/src/testing' + '@' + '5f7e36cad6434fd3d65674af96653a23ecc9f694',
>+    Var('chromium_git') + '/chromium/src/testing' + '@' + 'd2fde4ae5b8d0a5021e6f79d2f4a62e83ba348bc',
>   'src/third_party':
>-    Var('chromium_git') + '/chromium/src/third_party' + '@' + '2c0ced3ddbf84ce1c0759b277d9538da42f23650',
>+    Var('chromium_git') + '/chromium/src/third_party' + '@' + 'f931bb4f2bdcb327d066052df1914cab4bd68c50',
>   'src/third_party/catapult':
>     Var('chromium_git') + '/catapult.git' + '@' + Var('catapult_revision'),
>   'src/third_party/colorama/src':
>     Var('chromium_git') + '/external/colorama.git' + '@' + '799604a1041e9b3bc5d2789ecbd7e8db2e18e6b8',
>+  'src/third_party/freetype/src':
>+    Var('chromium_git') + '/chromium/src/third_party/freetype2.git' + '@' + '578bcf103a12fb742cdb314565819011d1ac12a7',
>   'src/third_party/googletest/src':
>-    Var('chromium_git') + '/external/github.com/google/googletest.git' + '@' + '0062e4869f07a3ef235703ddf63af604b712446c',
>+    Var('chromium_git') + '/external/github.com/google/googletest.git' + '@' + 'd5266326752f0a1dadbd310932d8f4fd8c3c5e7d',
>+  'src/third_party/harfbuzz-ng/src':
>+    Var('chromium_git') + '/external/github.com/harfbuzz/harfbuzz.git' + '@' + '2b76767bf572364d3d647cdd139f2044a7ad06b2',
>   'src/third_party/libjpeg_turbo':
>     Var('chromium_git') + '/chromium/deps/libjpeg_turbo.git' + '@' + 'a1750dbc79a8792dde3d3f7d7d8ac28ba01ac9dd',
>   'src/third_party/yasm/source/patched-yasm':
>-    Var('chromium_git') + '/chromium/deps/yasm/patched-yasm.git' + '@' + 'b98114e18d8b9b84586b10d24353ab8616d4c5fc',
>+    Var('chromium_git') + '/chromium/deps/yasm/patched-yasm.git' + '@' + '720b70524a4424b15fc57e82263568c8ba0496ad',
>   'src/tools':
>-    Var('chromium_git') + '/chromium/src/tools' + '@' + '6202b67fc46a9984097caf237e12e3b8f7a9f7da',
>-  'src/tools/gyp':
>-    Var('chromium_git') + '/external/gyp.git' + '@' + 'd61a9397e668fa9843c4aa7da9e79460fe590bfb',
>-   'src/tools/swarming_client':
>-     Var('chromium_git') + '/infra/luci/client-py.git' + '@' +  Var('swarming_revision'),
>+    Var('chromium_git') + '/chromium/src/tools' + '@' + 'f2c6ed916b94176158763400de308c2afd56b259',
>+  'src/tools/swarming_client':
>+    Var('chromium_git') + '/infra/luci/client-py.git' + '@' +  Var('swarming_revision'),
> 
>   # libyuv-only dependencies (not present in Chromium).
>   'src/third_party/gflags':
>@@ -52,40 +54,136 @@ deps = {
>   },
> 
>   # Android deps:
>-  'src/third_party/auto/src': {
>-    'url': Var('chromium_git') + '/external/github.com/google/auto.git' + '@' + '8a81a858ae7b78a1aef71ac3905fade0bbd64e82',
>-    'condition': 'checkout_android',
>+  'src/third_party/accessibility_test_framework': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/accessibility-test-framework',
>+              'version': 'version:2.1-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>   },
>   'src/third_party/auto/src': {
>     'url': Var('chromium_git') + '/external/github.com/google/auto.git' + '@' + '8a81a858ae7b78a1aef71ac3905fade0bbd64e82',
>     'condition': 'checkout_android',
>   },
>   'src/base': {
>-    'url': Var('chromium_git') + '/chromium/src/base' + '@' + 'ac3d2b81181b085a9952cb83dba748420eefe691',
>+    'url': Var('chromium_git') + '/chromium/src/base' + '@' + '6c0497f398c5f6e6af0c66fbf4d77e875eb3f2b1',
>     'condition': 'checkout_android',
>   },
>+  'src/third_party/bazel': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/bazel',
>+              'version': 'version:0.10.0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+  'src/third_party/bouncycastle': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/bouncycastle',
>+              'version': 'version:1.46-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>   'src/third_party/android_ndk': {
>-    'url': Var('chromium_git') + '/android_ndk.git' + '@' + 'e951c37287c7d8cd915bf8d4149fd4a06d808b55',
>+    'url': Var('chromium_git') + '/android_ndk.git' + '@' + '5cd86312e794bdf542a3685c6f10cbb96072990b',
>     'condition': 'checkout_android',
>   },
>+  'src/third_party/android_support_test_runner': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_support_test_runner',
>+              'version': 'version:0.5-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>   'src/third_party/android_tools': {
>-    'url': Var('chromium_git') + '/android_tools.git' + '@' + '9a70d48fcdd68cd0e7e968f342bd767ee6323bd1',
>+    'url': Var('chromium_git') + '/android_tools.git' + '@' + '130499e25286f4d56acafa252fee09f3cc595c49',
>     'condition': 'checkout_android',
>   },
>+  'src/third_party/byte_buddy': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/byte_buddy',
>+              'version': 'version:1.4.17-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>   'src/third_party/ced/src': {
>     'url': Var('chromium_git') + '/external/github.com/google/compact_enc_det.git' + '@' + '94c367a1fe3a13207f4b22604fcfd1d9f9ddf6d9',
>     'condition': 'checkout_android',
>   },
>   'src/third_party/errorprone/lib': {
>-      'url': Var('chromium_git') + '/chromium/third_party/errorprone.git' + '@' + 'ecc57c2b00627667874744b9ad8efe10734d97a8',
>+      'url': Var('chromium_git') + '/chromium/third_party/errorprone.git' + '@' + '980d49e839aa4984015efed34b0134d4b2c9b6d7',
>       'condition': 'checkout_android',
>   },
>   'src/third_party/findbugs': {
>     'url': Var('chromium_git') + '/chromium/deps/findbugs.git' + '@' + '4275d9ac8610db6b1bc9a5e887f97e41b33fac67',
>     'condition': 'checkout_android',
>   },
>+  'src/third_party/gson': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/gson',
>+              'version': 'version:2.8.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+  'src/third_party/guava': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/guava',
>+              'version': 'version:23.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+  'src/third_party/hamcrest': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/hamcrest',
>+              'version': 'version:1.3-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>   'src/third_party/icu': {
>-    'url': Var('chromium_git') + '/chromium/deps/icu.git' + '@' + 'd888fd2a1be890f4d35e43f68d6d79f42519a357',
>+    'url': Var('chromium_git') + '/chromium/deps/icu.git' + '@' + '297a4dd02b9d36c92ab9b4f121e433c9c3bc14f8',
>+  },
>+  'src/third_party/icu4j': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/icu4j',
>+              'version': 'version:53.1-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+  'src/third_party/intellij': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/intellij',
>+              'version': 'version:12.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>   },
>   'src/third_party/jsr-305/src': {
>     'url': Var('chromium_git') + '/external/jsr-305.git' + '@' + '642c508235471f7220af6d5df2d3210e3bfc0919',
>@@ -96,25 +194,85 @@ deps = {
>     'condition': 'checkout_android',
>   },
>   'src/third_party/mockito/src': {
>-    'url': Var('chromium_git') + '/external/mockito/mockito.git' + '@' + 'de83ad4598ad4cf5ea53c69a8a8053780b04b850',
>+    'url': Var('chromium_git') + '/external/mockito/mockito.git' + '@' + '04a2a289a4222f80ad20717c25144981210d2eac',
>     'condition': 'checkout_android',
>   },
>+  'src/third_party/objenesis': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/objenesis',
>+              'version': 'version:2.4-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+  'src/third_party/ow2_asm': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/ow2_asm',
>+              'version': 'version:5.0.1-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+  'src/third_party/r8': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/r8',
>+              'version': 'version:1.0.30',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>   'src/third_party/requests/src': {
>     'url': Var('chromium_git') + '/external/github.com/kennethreitz/requests.git' + '@' + 'f172b30356d821d180fa4ecfa3e71c7274a32de4',
>     'condition': 'checkout_android',
>   },
>+  'src/third_party/robolectric': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/robolectric',
>+              'version': 'version:3.5.1',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>   'src/third_party/robolectric/robolectric': {
>     'url': Var('chromium_git') + '/external/robolectric.git' + '@' + '7e067f1112e1502caa742f7be72d37b5678d3403',
>     'condition': 'checkout_android',
>   },
>+  'src/third_party/sqlite4java': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/sqlite4java',
>+              'version': 'version:0.282-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>   'src/third_party/ub-uiautomator/lib': {
>     'url': Var('chromium_git') + '/chromium/third_party/ub-uiautomator.git' + '@' + '00270549ce3161ae72ceb24712618ea28b4f9434',
>     'condition': 'checkout_android',
>   },
>+  'src/third_party/xstream': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/xstream',
>+              'version': 'version:1.4.8-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
> 
>   # iOS deps:
>   'src/ios': {
>-    'url': Var('chromium_git') + '/chromium/src/ios' + '@' + '299ef76e844a74a1f2f4ce7f06d101861fb49aba',
>+    'url': Var('chromium_git') + '/chromium/src/ios' + '@' + '8e45eb00dffadde7e0669a881991e237b0b7a8eb',
>     'condition': 'checkout_ios'
>   },
> 
>@@ -124,6 +282,472 @@ deps = {
>     'url': Var('chromium_git') + '/chromium/deps/yasm/binaries.git' + '@' + '52f9b3f4b0aa06da24ef8b123058bb61ee468881',
>     'condition': 'checkout_win',
>   },
>+
>+  # === ANDROID_DEPS Generated Code Start ===
>+  # Generated by //tools/android/roll/android_deps/fetch_all.sh
>+  'src/third_party/android_deps/libs/android_arch_core_common': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/android_arch_core_common',
>+              'version': 'version:1.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/android_arch_lifecycle_common': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/android_arch_lifecycle_common',
>+              'version': 'version:1.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/android_arch_lifecycle_runtime': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/android_arch_lifecycle_runtime',
>+              'version': 'version:1.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_android_support_animated_vector_drawable': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_android_support_animated_vector_drawable',
>+              'version': 'version:27.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_android_support_appcompat_v7': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_android_support_appcompat_v7',
>+              'version': 'version:27.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_android_support_cardview_v7': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_android_support_cardview_v7',
>+              'version': 'version:27.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_android_support_design': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_android_support_design',
>+              'version': 'version:27.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_android_support_gridlayout_v7': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_android_support_gridlayout_v7',
>+              'version': 'version:27.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_android_support_leanback_v17': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_android_support_leanback_v17',
>+              'version': 'version:27.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_android_support_mediarouter_v7': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_android_support_mediarouter_v7',
>+              'version': 'version:27.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_android_support_multidex': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_android_support_multidex',
>+              'version': 'version:1.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_android_support_palette_v7': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_android_support_palette_v7',
>+              'version': 'version:27.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_android_support_preference_leanback_v17': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_android_support_preference_leanback_v17',
>+              'version': 'version:27.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_android_support_preference_v14': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_android_support_preference_v14',
>+              'version': 'version:27.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_android_support_preference_v7': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_android_support_preference_v7',
>+              'version': 'version:27.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_android_support_recyclerview_v7': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_android_support_recyclerview_v7',
>+              'version': 'version:27.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_android_support_support_annotations': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_android_support_support_annotations',
>+              'version': 'version:27.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_android_support_support_compat': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_android_support_support_compat',
>+              'version': 'version:27.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_android_support_support_core_ui': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_android_support_support_core_ui',
>+              'version': 'version:27.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_android_support_support_core_utils': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_android_support_support_core_utils',
>+              'version': 'version:27.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_android_support_support_fragment': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_android_support_support_fragment',
>+              'version': 'version:27.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_android_support_support_media_compat': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_android_support_support_media_compat',
>+              'version': 'version:27.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_android_support_support_v13': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_android_support_support_v13',
>+              'version': 'version:27.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_android_support_support_v4': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_android_support_support_v4',
>+              'version': 'version:27.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_android_support_support_vector_drawable': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_android_support_support_vector_drawable',
>+              'version': 'version:27.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_android_support_transition': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_android_support_transition',
>+              'version': 'version:27.0.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_google_android_gms_play_services_auth': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_google_android_gms_play_services_auth',
>+              'version': 'version:12.0.1-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_google_android_gms_play_services_auth_api_phone': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_google_android_gms_play_services_auth_api_phone',
>+              'version': 'version:12.0.1-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_google_android_gms_play_services_auth_base': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_google_android_gms_play_services_auth_base',
>+              'version': 'version:12.0.1-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_google_android_gms_play_services_base': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_google_android_gms_play_services_base',
>+              'version': 'version:12.0.1-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_google_android_gms_play_services_basement': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_google_android_gms_play_services_basement',
>+              'version': 'version:12.0.1-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_google_android_gms_play_services_cast': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_google_android_gms_play_services_cast',
>+              'version': 'version:12.0.1-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_google_android_gms_play_services_cast_framework': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_google_android_gms_play_services_cast_framework',
>+              'version': 'version:12.0.1-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_google_android_gms_play_services_fido': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_google_android_gms_play_services_fido',
>+              'version': 'version:12.0.1-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_google_android_gms_play_services_gcm': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_google_android_gms_play_services_gcm',
>+              'version': 'version:12.0.1-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_google_android_gms_play_services_iid': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_google_android_gms_play_services_iid',
>+              'version': 'version:12.0.1-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_google_android_gms_play_services_instantapps': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_google_android_gms_play_services_instantapps',
>+              'version': 'version:12.0.1-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_google_android_gms_play_services_location': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_google_android_gms_play_services_location',
>+              'version': 'version:12.0.1-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_google_android_gms_play_services_tasks': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_google_android_gms_play_services_tasks',
>+              'version': 'version:12.0.1-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_google_android_gms_play_services_vision': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_google_android_gms_play_services_vision',
>+              'version': 'version:12.0.1-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_google_android_gms_play_services_vision_common': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_google_android_gms_play_services_vision_common',
>+              'version': 'version:12.0.1-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  'src/third_party/android_deps/libs/com_google_android_play_core': {
>+      'packages': [
>+          {
>+              'package': 'chromium/third_party/android_deps/libs/com_google_android_play_core',
>+              'version': 'version:1.3.0-cr0',
>+          },
>+      ],
>+      'condition': 'checkout_android',
>+      'dep_type': 'cipd',
>+  },
>+
>+  # === ANDROID_DEPS Generated Code End ===
> }
> 
> # Define rules for which include paths are allowed in our source.
>@@ -197,6 +821,12 @@ hooks = [
>     'pattern': '.',
>     'action': ['python', 'src/build/vs_toolchain.py', 'update'],
>   },
>+  {
>+    # Update the Mac toolchain if necessary.
>+    'name': 'mac_toolchain',
>+    'pattern': '.',
>+    'action': ['python', 'src/build/mac_toolchain.py'],
>+  },
>   # Pull binutils for linux, enabled debug fission for faster linking /
>   # debugging when used with clang on Ubuntu Precise.
>   # https://code.google.com/p/chromium/issues/detail?id=352046
>@@ -213,7 +843,7 @@ hooks = [
>     # Note: On Win, this should run after win_toolchain, as it may use it.
>     'name': 'clang',
>     'pattern': '.',
>-    'action': ['python', 'src/tools/clang/scripts/update.py', '--if-needed'],
>+    'action': ['python', 'src/tools/clang/scripts/update.py'],
>   },
>   {
>     # Update LASTCHANGE.
>@@ -325,12 +955,15 @@ hooks = [
>     ],
>   },
>   {
>-    'name': 'Android CIPD Ensure',
>+    # We used to use src as a CIPD root. We moved it to a different directory
>+    # in crrev.com/c/930178 but left the clobber here to ensure that that CL
>+    # could be reverted safely. This can be safely removed once crbug.com/794764
>+    # is resolved.
>+    'name': 'Android Clobber Deprecated CIPD Root',
>     'pattern': '.',
>     'condition': 'checkout_android',
>-    'action': ['src/build/cipd/cipd_wrapper.py',
>-               '--chromium-root', 'src',
>-               '--ensure-file', 'src/build/cipd/android/android.ensure',
>+    'action': ['src/build/cipd/clobber_cipd_root.py',
>+               '--root', 'src',
>     ],
>   },
>   # Android dependencies. Many are downloaded using Google Storage these days.
>@@ -340,10 +973,11 @@ hooks = [
>     # This downloads SDK extras and puts them in the
>     # third_party/android_tools/sdk/extras directory.
>     'name': 'sdkextras',
>+    'condition': 'checkout_android',
>     'pattern': '.',
>     # When adding a new sdk extras package to download, add the package
>     # directory and zip file to .gitignore in third_party/android_tools.
>-    'action': ['python',
>+    'action': ['vpython',
>                'src/build/android/play_services/update.py',
>                'download'
>     ],
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/LICENSE_THIRD_PARTY b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/LICENSE_THIRD_PARTY
>deleted file mode 100644
>index a71591e77104194f464733625331235ed838958e..0000000000000000000000000000000000000000
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/LICENSE_THIRD_PARTY
>+++ /dev/null
>@@ -1,8 +0,0 @@
>-This source tree contains third party source code which is governed by third
>-party licenses. This file contains references to files which are under other
>-licenses than the one provided in the LICENSE file in the root of the source
>-tree.
>-
>-Files governed by third party licenses:
>-source/x86inc.asm
>-
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/README.chromium b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/README.chromium
>index c25373e23ce4214f22882c6d9904bea39681820c..4a239b3cbf0eb68479b746d14c495f857690387f 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/README.chromium
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/README.chromium
>@@ -1,6 +1,6 @@
> Name: libyuv
> URL: http://code.google.com/p/libyuv/
>-Version: 1703
>+Version: 1722
> License: BSD
> License File: LICENSE
> 
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/README.md b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/README.md
>index 7b6619220b8dcd183948083221de49df6fb4b2a7..db70b7f08d3ead3bfcdaa206faf392e5f56d1418 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/README.md
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/README.md
>@@ -10,9 +10,9 @@
> 
> ### Development
> 
>-See [Getting started] [1] for instructions on how to get started developing.
>+See [Getting started][1] for instructions on how to get started developing.
> 
>-You can also browse the [docs directory] [2] for more documentation.
>+You can also browse the [docs directory][2] for more documentation.
> 
>-[1]: https://chromium.googlesource.com/libyuv/libyuv/+/master/docs/getting_started.md
>-[2]: https://chromium.googlesource.com/libyuv/libyuv/+/master/docs/
>+[1]: ./docs/getting_started.md
>+[2]: ./docs/
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/all.gyp b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/all.gyp
>deleted file mode 100644
>index 88a74842716a81cdfea319b034b825ac73847604..0000000000000000000000000000000000000000
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/all.gyp
>+++ /dev/null
>@@ -1,21 +0,0 @@
>-# Copyright 2013 The LibYuv Project Authors. All rights reserved.
>-#
>-# Use of this source code is governed by a BSD-style license
>-# that can be found in the LICENSE file in the root of the source
>-# tree. An additional intellectual property rights grant can be found
>-# in the file PATENTS. All contributing project authors may
>-# be found in the AUTHORS file in the root of the source tree.
>-
>-# all.gyp and All target are for benefit of android gyp build.
>-{
>-  'targets': [
>-    {
>-      'target_name': 'All',
>-      'type': 'none',
>-      'dependencies': [
>-        'libyuv.gyp:*',
>-        'libyuv_test.gyp:*',
>-      ],
>-    },
>-  ],
>-}
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/build_overrides/build.gni b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/build_overrides/build.gni
>index 9bb73ab023acf7d062388e1e17d3771137329895..6d8319b965e318b563d1e5ca3c95f4f62a630d29 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/build_overrides/build.gni
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/build_overrides/build.gni
>@@ -33,6 +33,14 @@ ubsan_vptr_blacklist_path =
> # so we just ignore that assert. See https://crbug.com/648948 for more info.
> ignore_elf32_limitations = true
> 
>-# Use system Xcode installation instead of the Chromium bundled Mac toolchain,
>-# since it contains only SDK 10.11, not 10.12 which WebRTC needs.
>-use_system_xcode = true
>+# Use bundled hermetic Xcode installation maintained by Chromium,
>+# except for local iOS builds where it is unsupported.
>+if (host_os == "mac") {
>+  _result = exec_script("//build/mac/should_use_hermetic_xcode.py",
>+                        [ target_os ],
>+                        "value")
>+  assert(_result != 2,
>+         "Do not allow building targets with the default" +
>+             "hermetic toolchain if the minimum OS version is not met.")
>+  use_system_xcode = _result == 0
>+}
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/docs/environment_variables.md b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/docs/environment_variables.md
>index c28d83e7dc14905e9506e4f7eee8ac35957ffb16..cd8159ad5a87776cf9d5eebf45be6ec1ad4e638d 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/docs/environment_variables.md
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/docs/environment_variables.md
>@@ -6,7 +6,10 @@ For test purposes, environment variables can be set to control libyuv behavior.
> 
> By default the cpu is detected and the most advanced form of SIMD is used.  But you can disable instruction sets selectively, or completely, falling back on C code.  Set the variable to 1 to disable the specified instruction set.
> 
>+## All CPUs
>     LIBYUV_DISABLE_ASM
>+
>+## Intel CPUs
>     LIBYUV_DISABLE_X86
>     LIBYUV_DISABLE_SSE2
>     LIBYUV_DISABLE_SSSE3
>@@ -14,12 +17,25 @@ By default the cpu is detected and the most advanced form of SIMD is used.  But
>     LIBYUV_DISABLE_SSE42
>     LIBYUV_DISABLE_AVX
>     LIBYUV_DISABLE_AVX2
>-    LIBYUV_DISABLE_AVX512BW
>     LIBYUV_DISABLE_ERMS
>     LIBYUV_DISABLE_FMA3
>-    LIBYUV_DISABLE_MSA
>+    LIBYUV_DISABLE_F16C
>+    LIBYUV_DISABLE_AVX512BW
>+    LIBYUV_DISABLE_AVX512VL
>+    LIBYUV_DISABLE_AVX512VBMI
>+    LIBYUV_DISABLE_AVX512VBMI2
>+    LIBYUV_DISABLE_AVX512VBITALG
>+    LIBYUV_DISABLE_AVX512VPOPCNTDQ
>+    LIBYUV_DISABLE_GFNI
>+
>+## ARM CPUs
>+
>     LIBYUV_DISABLE_NEON
> 
>+## MIPS CPUs
>+    LIBYUV_DISABLE_MSA
>+    LIBYUV_DISABLE_MMI
>+
> # Test Width/Height/Repeat
> 
> The unittests default to a small image (128x72) to run fast.  This can be set by environment variable to test a specific resolutions.
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/docs/formats.md b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/docs/formats.md
>index f78f57bb4c41a3e150842a0c4fb5c043e33656f0..97e8ce05f482ba97c0617fa74c7fa737d72e7878 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/docs/formats.md
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/docs/formats.md
>@@ -66,7 +66,7 @@ The following is extracted from video_common.h as a complete list of formats sup
>       // 1 Primary Compressed YUV format.
>       FOURCC_MJPG = FOURCC('M', 'J', 'P', 'G'),
> 
>-      // 7 Auxiliary YUV variations: 3 with U and V planes are swapped, 1 Alias.
>+      // 8 Auxiliary YUV variations: 3 with U and V planes are swapped, 1 Alias.
>       FOURCC_YV12 = FOURCC('Y', 'V', '1', '2'),
>       FOURCC_YV16 = FOURCC('Y', 'V', '1', '6'),
>       FOURCC_YV24 = FOURCC('Y', 'V', '2', '4'),
>@@ -74,6 +74,7 @@ The following is extracted from video_common.h as a complete list of formats sup
>       FOURCC_J420 = FOURCC('J', '4', '2', '0'),
>       FOURCC_J400 = FOURCC('J', '4', '0', '0'),  // unofficial fourcc
>       FOURCC_H420 = FOURCC('H', '4', '2', '0'),  // unofficial fourcc
>+      FOURCC_H422 = FOURCC('H', '4', '2', '2'),  // unofficial fourcc
> 
>       // 14 Auxiliary aliases.  CanonicalFourCC() maps these to canonical fourcc.
>       FOURCC_IYUV = FOURCC('I', 'Y', 'U', 'V'),  // Alias for I420.
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/docs/getting_started.md b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/docs/getting_started.md
>index 09297b66a5cc784de596690fb4bfe36ad4232b87..4426b606e07561c606c997618c125c75dc1131a4 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/docs/getting_started.md
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/docs/getting_started.md
>@@ -27,7 +27,7 @@ Then you'll get a .gclient file like:
>       },
>     ];
> 
>-For iOS add `;target_os=['ios'];` to your OSX .gclient and run `GYP_DEFINES="OS=ios" gclient sync.`
>+For iOS add `;target_os=['ios'];` to your OSX .gclient and run `gclient sync.`
> 
> Browse the Git reprository: https://chromium.googlesource.com/libyuv/libyuv/+/master
> 
>@@ -48,11 +48,8 @@ For Android add `;target_os=['android'];` to your Linux .gclient
> 
> Then run:
> 
>-    export GYP_DEFINES="OS=android"
>     gclient sync
> 
>-The sync will generate native build files for your environment using gyp (Windows: Visual Studio, OSX: XCode, Linux: make). This generation can also be forced manually: `gclient runhooks`
>-
> To get just the source (not buildable):
> 
>     git clone https://chromium.googlesource.com/libyuv/libyuv
>@@ -98,11 +95,15 @@ arm64
> 
> ios simulator
> 
>-    gn gen out/Release "--args=is_debug=false target_os=\"ios\" ios_enable_code_signing=false target_cpu=\"x86\""
>-    gn gen out/Debug "--args=is_debug=true target_os=\"ios\" ios_enable_code_signing=false target_cpu=\"x86\""
>+    gn gen out/Release "--args=is_debug=false target_os=\"ios\" ios_enable_code_signing=false use_xcode_clang=true target_cpu=\"x86\""
>+    gn gen out/Debug "--args=is_debug=true target_os=\"ios\" ios_enable_code_signing=false use_xcode_clang=true target_cpu=\"x86\""
>     ninja -v -C out/Debug libyuv_unittest
>     ninja -v -C out/Release libyuv_unittest
> 
>+ios disassembly
>+
>+    otool -tV ./out/Release/obj/libyuv_neon/row_neon64.o >row_neon64.txt
>+
> ### Android
> https://code.google.com/p/chromium/wiki/AndroidBuildInstructions
> 
>@@ -131,8 +132,8 @@ ia32
> 
> mips
> 
>-    gn gen out/Release "--args=is_debug=false target_os=\"android\" target_cpu=\"mips64el\" mips_arch_variant=\"r6\" mips_use_msa=true is_component_build=true is_clang=true"
>-    gn gen out/Debug "--args=is_debug=true target_os=\"android\" target_cpu=\"mips64el\" mips_arch_variant=\"r6\" mips_use_msa=true is_component_build=true is_clang=true"
>+    gn gen out/Release "--args=is_debug=false target_os=\"android\" target_cpu=\"mips64el\" mips_arch_variant=\"r6\" mips_use_msa=true is_component_build=true"
>+    gn gen out/Debug "--args=is_debug=true target_os=\"android\" target_cpu=\"mips64el\" mips_arch_variant=\"r6\" mips_use_msa=true is_component_build=true"
>     ninja -v -C out/Debug libyuv_unittest
>     ninja -v -C out/Release libyuv_unittest
> 
>@@ -144,17 +145,19 @@ arm disassembly:
> 
>     third_party/android_ndk/toolchains/aarch64-linux-android-4.9/prebuilt/linux-x86_64/bin/aarch64-linux-android-objdump -d ./out/Release/obj/libyuv_neon/row_neon64.o >row_neon64.txt
> 
>+    Caveat: Disassembly may require optimize_max be disabled in BUILD.gn
>+
> Running tests:
> 
>-    build/android/test_runner.py gtest -s libyuv_unittest -t 7200 --verbose --release --gtest_filter=*
>+    out/Release/bin/run_libyuv_unittest -vv --gtest_filter=*
> 
> Running test as benchmark:
> 
>-    build/android/test_runner.py gtest -s libyuv_unittest -t 7200 --verbose --release --gtest_filter=* -a "--libyuv_width=1280 --libyuv_height=720 --libyuv_repeat=999 --libyuv_flags=-1  --libyuv_cpu_info=-1"
>+    out/Release/bin/run_libyuv_unittest -vv --gtest_filter=* --libyuv_width=1280 --libyuv_height=720 --libyuv_repeat=999 --libyuv_flags=-1  --libyuv_cpu_info=-1
> 
> Running test with C code:
> 
>-    build/android/test_runner.py gtest -s libyuv_unittest -t 7200 --verbose --release --gtest_filter=* -a "--libyuv_width=1280 --libyuv_height=720 --libyuv_repeat=999 --libyuv_flags=1 --libyuv_cpu_info=1"
>+    out/Release/bin/run_libyuv_unittest -vv --gtest_filter=* --libyuv_width=1280 --libyuv_height=720 --libyuv_repeat=999 --libyuv_flags=1 --libyuv_cpu_info=1
> 
> ### Build targets
> 
>@@ -172,6 +175,15 @@ Running test with C code:
>     ninja -v -C out/Debug libyuv_unittest
>     ninja -v -C out/Release libyuv_unittest
> 
>+### MIPS Linux
>+
>+mips
>+
>+   gn gen out/Release "--args=is_debug=false target_os=\"linux\" target_cpu=\"mips64el\" mips_arch_variant=\"loongson3\" mips_use_mmi=true is_component_build=false use_sysroot=false use_gold=false"
>+   gn gen out/Debug "--args=is_debug=true target_os=\"linux\" target_cpu=\"mips64el\" mips_arch_variant=\"loongson3\" mips_use_mmi=true is_component_build=false use_sysroot=false use_gold=false"
>+   ninja -v -C out/Debug libyuv_unittest
>+   ninja -v -C out/Release libyuv_unittest
>+
> ## Building the Library with make
> 
> ### Linux
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/gyp_libyuv b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/gyp_libyuv
>deleted file mode 100755
>index 445b924f16b0a814f626eccebb1a2b02e04922ff..0000000000000000000000000000000000000000
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/gyp_libyuv
>+++ /dev/null
>@@ -1,101 +0,0 @@
>-#!/usr/bin/env python
>-#
>-# Copyright 2014 The LibYuv Project Authors. All rights reserved.
>-#
>-# Use of this source code is governed by a BSD-style license
>-# that can be found in the LICENSE file in the root of the source
>-# tree. An additional intellectual property rights grant can be found
>-# in the file PATENTS. All contributing project authors may
>-# be found in the AUTHORS file in the root of the source tree.
>-
>-# This script is used to run GYP for libyuv. It contains selected parts of the
>-# main function from the src/build/gyp_chromium file.
>-
>-import glob
>-import os
>-import shlex
>-import sys
>-
>-checkout_root = os.path.dirname(os.path.realpath(__file__))
>-
>-sys.path.insert(0, os.path.join(checkout_root, 'build'))
>-import gyp_chromium
>-import gyp_helper
>-import vs_toolchain
>-
>-sys.path.insert(0, os.path.join(checkout_root, 'tools', 'gyp', 'pylib'))
>-import gyp
>-
>-def GetSupplementalFiles():
>-  """Returns a list of the supplemental files that are included in all GYP
>-  sources."""
>-  # Can't use the one in gyp_chromium since the directory location of the root
>-  # is different.
>-  return glob.glob(os.path.join(checkout_root, '*', 'supplement.gypi'))
>-
>-
>-if __name__ == '__main__':
>-  args = sys.argv[1:]
>-
>-  if int(os.environ.get('GYP_CHROMIUM_NO_ACTION', 0)):
>-    print 'Skipping gyp_libyuv due to GYP_CHROMIUM_NO_ACTION env var.'
>-    sys.exit(0)
>-
>-  # This could give false positives since it doesn't actually do real option
>-  # parsing.  Oh well.
>-  gyp_file_specified = False
>-  for arg in args:
>-    if arg.endswith('.gyp'):
>-      gyp_file_specified = True
>-      break
>-
>-  # If we didn't get a file, assume 'all.gyp' in the root of the checkout.
>-  if not gyp_file_specified:
>-    # Because of a bug in gyp, simply adding the abspath to all.gyp doesn't
>-    # work, but chdir'ing and adding the relative path does. Spooky :/
>-    os.chdir(checkout_root)
>-    args.append('all.gyp')
>-
>-  # There shouldn't be a circular dependency relationship between .gyp files,
>-  args.append('--no-circular-check')
>-
>-  # Default to ninja unless GYP_GENERATORS is set.
>-  if not os.environ.get('GYP_GENERATORS'):
>-    os.environ['GYP_GENERATORS'] = 'ninja'
>-
>-  vs2013_runtime_dll_dirs = None
>-  if int(os.environ.get('DEPOT_TOOLS_WIN_TOOLCHAIN', '1')):
>-    vs2013_runtime_dll_dirs = vs_toolchain.SetEnvironmentAndGetRuntimeDllDirs()
>-
>-  # Enforce gyp syntax checking. This adds about 20% execution time.
>-  args.append('--check')
>-
>-  supplemental_includes = gyp_chromium.GetSupplementalFiles()
>-  gyp_vars_dict = gyp_chromium.GetGypVars(supplemental_includes)
>-
>-  # Automatically turn on crosscompile support for platforms that need it.
>-  if all(('ninja' in os.environ.get('GYP_GENERATORS', ''),
>-          gyp_vars_dict.get('OS') in ['android', 'ios'],
>-          'GYP_CROSSCOMPILE' not in os.environ)):
>-    os.environ['GYP_CROSSCOMPILE'] = '1'
>-
>-  args.extend(['-I' + i for i in
>-               gyp_chromium.additional_include_files(supplemental_includes,
>-                                                     args)])
>-
>-  # Set the gyp depth variable to the root of the checkout.
>-  args.append('--depth=' + os.path.relpath(checkout_root))
>-
>-  print 'Updating projects from gyp files...'
>-  sys.stdout.flush()
>-
>-  # Off we go...
>-  gyp_rc = gyp.main(args)
>-
>-  if vs2013_runtime_dll_dirs:
>-    x64_runtime, x86_runtime = vs2013_runtime_dll_dirs
>-    vs_toolchain.CopyVsRuntimeDlls(
>-        os.path.join(checkout_root, gyp_chromium.GetOutputDirectory()),
>-        (x86_runtime, x64_runtime))
>-
>-  sys.exit(gyp_rc)
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/gyp_libyuv.py b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/gyp_libyuv.py
>deleted file mode 100644
>index bb32ec39df3953cb98c25f1d80b3fd4a27f6acf5..0000000000000000000000000000000000000000
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/gyp_libyuv.py
>+++ /dev/null
>@@ -1,28 +0,0 @@
>-#!/usr/bin/env python
>-#
>-# Copyright 2014 The LibYuv Project Authors. All rights reserved.
>-#
>-# Use of this source code is governed by a BSD-style license
>-# that can be found in the LICENSE file in the root of the source
>-# tree. An additional intellectual property rights grant can be found
>-# in the file PATENTS. All contributing project authors may
>-# be found in the AUTHORS file in the root of the source tree.
>-
>-
>-# This script is a modified copy of the src/build/gyp_chromium.py file.
>-# It is needed for parallel processing.
>-
>-# This file is (possibly, depending on python version) imported by
>-# gyp_libyuv when GYP_PARALLEL=1 and it creates sub-processes
>-# through the multiprocessing library.
>-
>-# Importing in Python 2.6 (fixed in 2.7) on Windows doesn't search for
>-# imports that don't end in .py (and aren't directories with an
>-# __init__.py). This wrapper makes "import gyp_libyuv" work with
>-# those old versions and makes it possible to execute gyp_libyuv.py
>-# directly on Windows where the extension is useful.
>-
>-import os
>-
>-path = os.path.abspath(os.path.split(__file__)[0])
>-execfile(os.path.join(path, 'gyp_libyuv'))
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/basic_types.h b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/basic_types.h
>index 01d9dfc773628adf90c4c223dd0926b7ce7ff2f0..1bea67f2f2547081684704fc61a0acfdcd5e5de2 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/basic_types.h
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/basic_types.h
>@@ -29,6 +29,8 @@ typedef signed char int8_t;
> #else
> #include <stdint.h>  // for uintptr_t and C99 types
> #endif               // defined(_MSC_VER) && (_MSC_VER < 1600)
>+// Types are deprecated.  Enable this macro for legacy types.
>+#ifdef LIBYUV_LEGACY_TYPES
> typedef uint64_t uint64;
> typedef int64_t int64;
> typedef uint32_t uint32;
>@@ -37,6 +39,7 @@ typedef uint16_t uint16;
> typedef int16_t int16;
> typedef uint8_t uint8;
> typedef int8_t int8;
>+#endif  // LIBYUV_LEGACY_TYPES
> #endif  // INT_TYPES_DEFINED
> 
> #if !defined(LIBYUV_API)
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/compare_row.h b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/compare_row.h
>index 72ee740600af1c643a2bf1473f735ce829a144f3..e95b9d93eb2c64a69712abe83a32ff4f9997351c 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/compare_row.h
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/compare_row.h
>@@ -45,12 +45,6 @@ extern "C" {
> #endif  // clang >= 3.4
> #endif  // __clang__
> 
>-// The following are available for Visual C:
>-#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && \
>-    (defined(VISUALC_HAS_AVX2) || defined(CLANG_HAS_AVX2))
>-#define HAS_HASHDJB2_AVX2
>-#endif
>-
> // The following are available for Visual C and GCC:
> #if !defined(LIBYUV_DISABLE_X86) && \
>     (defined(__x86_64__) || defined(__i386__) || defined(_M_IX86))
>@@ -60,7 +54,7 @@ extern "C" {
> #endif
> 
> // The following are available for Visual C and clangcl 32 bit:
>-#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && \
>+#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && defined(_MSC_VER) && \
>     (defined(VISUALC_HAS_AVX2) || defined(CLANG_HAS_AVX2))
> #define HAS_HASHDJB2_AVX2
> #define HAS_SUMSQUAREERROR_AVX2
>@@ -90,6 +84,11 @@ extern "C" {
> #define HAS_SUMSQUAREERROR_MSA
> #endif
> 
>+#if !defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)
>+#define HAS_HAMMINGDISTANCE_MMI
>+#define HAS_SUMSQUAREERROR_MMI
>+#endif
>+
> uint32_t HammingDistance_C(const uint8_t* src_a,
>                            const uint8_t* src_b,
>                            int count);
>@@ -108,7 +107,9 @@ uint32_t HammingDistance_NEON(const uint8_t* src_a,
> uint32_t HammingDistance_MSA(const uint8_t* src_a,
>                              const uint8_t* src_b,
>                              int count);
>-
>+uint32_t HammingDistance_MMI(const uint8_t* src_a,
>+                             const uint8_t* src_b,
>+                             int count);
> uint32_t SumSquareError_C(const uint8_t* src_a,
>                           const uint8_t* src_b,
>                           int count);
>@@ -124,6 +125,9 @@ uint32_t SumSquareError_NEON(const uint8_t* src_a,
> uint32_t SumSquareError_MSA(const uint8_t* src_a,
>                             const uint8_t* src_b,
>                             int count);
>+uint32_t SumSquareError_MMI(const uint8_t* src_a,
>+                            const uint8_t* src_b,
>+                            int count);
> 
> uint32_t HashDjb2_C(const uint8_t* src, int count, uint32_t seed);
> uint32_t HashDjb2_SSE41(const uint8_t* src, int count, uint32_t seed);
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/convert.h b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/convert.h
>index d12ef24f799792a5695bf51e67e1e972892b0125..d8b47a838487e06cb96d17e398b7379a9ec3bd7e 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/convert.h
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/convert.h
>@@ -42,6 +42,21 @@ int I444ToI420(const uint8_t* src_y,
>                int width,
>                int height);
> 
>+// Convert I444 to NV21.
>+LIBYUV_API
>+int I444ToNV21(const uint8_t* src_y,
>+               int src_stride_y,
>+               const uint8_t* src_u,
>+               int src_stride_u,
>+               const uint8_t* src_v,
>+               int src_stride_v,
>+               uint8_t* dst_y,
>+               int dst_stride_y,
>+               uint8_t* dst_vu,
>+               int dst_stride_vu,
>+               int width,
>+               int height);
>+
> // Convert I422 to I420.
> LIBYUV_API
> int I422ToI420(const uint8_t* src_y,
>@@ -59,6 +74,21 @@ int I422ToI420(const uint8_t* src_y,
>                int width,
>                int height);
> 
>+// Convert I422 to NV21.
>+LIBYUV_API
>+int I422ToNV21(const uint8_t* src_y,
>+               int src_stride_y,
>+               const uint8_t* src_u,
>+               int src_stride_u,
>+               const uint8_t* src_v,
>+               int src_stride_v,
>+               uint8_t* dst_y,
>+               int dst_stride_y,
>+               uint8_t* dst_vu,
>+               int dst_stride_vu,
>+               int width,
>+               int height);
>+
> // Copy I420 to I420.
> #define I420ToI420 I420Copy
> LIBYUV_API
>@@ -127,6 +157,17 @@ int I400ToI420(const uint8_t* src_y,
>                int width,
>                int height);
> 
>+// Convert I400 (grey) to NV21.
>+LIBYUV_API
>+int I400ToNV21(const uint8_t* src_y,
>+               int src_stride_y,
>+               uint8_t* dst_y,
>+               int dst_stride_y,
>+               uint8_t* dst_vu,
>+               int dst_stride_vu,
>+               int width,
>+               int height);
>+
> #define J400ToJ420 I400ToI420
> 
> // Convert NV12 to I420.
>@@ -350,6 +391,19 @@ int MJPGToI420(const uint8_t* sample,
>                int dst_width,
>                int dst_height);
> 
>+// JPEG to NV21
>+LIBYUV_API
>+int MJPGToNV21(const uint8_t* sample,
>+               size_t sample_size,
>+               uint8_t* dst_y,
>+               int dst_stride_y,
>+               uint8_t* dst_vu,
>+               int dst_stride_vu,
>+               int src_width,
>+               int src_height,
>+               int dst_width,
>+               int dst_height);
>+
> // Query size of MJPG in pixels.
> LIBYUV_API
> int MJPGSize(const uint8_t* sample,
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/convert_argb.h b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/convert_argb.h
>index ab772b6c32334c8eac8600c281c9aa3f082a75b9..891c57de6808658a4ccce2847f8be11af98b0000 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/convert_argb.h
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/convert_argb.h
>@@ -298,6 +298,28 @@ int NV21ToRGB24(const uint8_t* src_y,
>                 int width,
>                 int height);
> 
>+// Convert NV12 to RAW.
>+LIBYUV_API
>+int NV12ToRAW(const uint8_t* src_y,
>+              int src_stride_y,
>+              const uint8_t* src_uv,
>+              int src_stride_uv,
>+              uint8_t* dst_raw,
>+              int dst_stride_raw,
>+              int width,
>+              int height);
>+
>+// Convert NV21 to RAW.
>+LIBYUV_API
>+int NV21ToRAW(const uint8_t* src_y,
>+              int src_stride_y,
>+              const uint8_t* src_vu,
>+              int src_stride_vu,
>+              uint8_t* dst_raw,
>+              int dst_stride_raw,
>+              int width,
>+              int height);
>+
> // Convert M420 to ARGB.
> LIBYUV_API
> int M420ToARGB(const uint8_t* src_m420,
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/convert_from.h b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/convert_from.h
>index 5cd8a4bfc04cc24603e70edf87bb5b3aa2dee556..861418d07b3cf967e5e89004353582b0f0f1a942 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/convert_from.h
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/convert_from.h
>@@ -239,6 +239,30 @@ int I420ToRGB565(const uint8_t* src_y,
>                  int width,
>                  int height);
> 
>+LIBYUV_API
>+int J420ToRGB565(const uint8_t* src_y,
>+                 int src_stride_y,
>+                 const uint8_t* src_u,
>+                 int src_stride_u,
>+                 const uint8_t* src_v,
>+                 int src_stride_v,
>+                 uint8_t* dst_rgb565,
>+                 int dst_stride_rgb565,
>+                 int width,
>+                 int height);
>+
>+LIBYUV_API
>+int H420ToRGB565(const uint8_t* src_y,
>+                 int src_stride_y,
>+                 const uint8_t* src_u,
>+                 int src_stride_u,
>+                 const uint8_t* src_v,
>+                 int src_stride_v,
>+                 uint8_t* dst_rgb565,
>+                 int dst_stride_rgb565,
>+                 int width,
>+                 int height);
>+
> LIBYUV_API
> int I422ToRGB565(const uint8_t* src_y,
>                  int src_stride_y,
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/cpu_id.h b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/cpu_id.h
>index 91480c68b0133c90a0389b3d527566b595d386a8..b01cd25c574f1ed5646154c87148644b72c62c26 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/cpu_id.h
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/cpu_id.h
>@@ -48,6 +48,7 @@ static const int kCpuHasAVX512VPOPCNTDQ = 0x100000;
> // These flags are only valid on MIPS processors.
> static const int kCpuHasMIPS = 0x200000;
> static const int kCpuHasMSA = 0x400000;
>+static const int kCpuHasMMI = 0x800000;
> 
> // Optional init function. TestCpuFlag does an auto-init.
> // Returns cpu_info flags.
>@@ -80,6 +81,31 @@ int ArmCpuCaps(const char* cpuinfo_name);
> LIBYUV_API
> int MaskCpuFlags(int enable_flags);
> 
>+// Sets the CPU flags to |cpu_flags|, bypassing the detection code. |cpu_flags|
>+// should be a valid combination of the kCpuHas constants above and include
>+// kCpuInitialized. Use this method when running in a sandboxed process where
>+// the detection code might fail (as it might access /proc/cpuinfo). In such
>+// cases the cpu_info can be obtained from a non sandboxed process by calling
>+// InitCpuFlags() and passed to the sandboxed process (via command line
>+// parameters, IPC...) which can then call this method to initialize the CPU
>+// flags.
>+// Notes:
>+// - when specifying 0 for |cpu_flags|, the auto initialization is enabled
>+//   again.
>+// - enabling CPU features that are not supported by the CPU will result in
>+//   undefined behavior.
>+// TODO(fbarchard): consider writing a helper function that translates from
>+// other library CPU info to libyuv CPU info and add a .md doc that explains
>+// CPU detection.
>+static __inline void SetCpuFlags(int cpu_flags) {
>+  LIBYUV_API extern int cpu_info_;
>+#ifdef __ATOMIC_RELAXED
>+  __atomic_store_n(&cpu_info_, cpu_flags, __ATOMIC_RELAXED);
>+#else
>+  cpu_info_ = cpu_flags;
>+#endif
>+}
>+
> // Low level cpuid for X86. Returns zeros on other CPUs.
> // eax is the info type that you want.
> // ecx is typically the cpu number, and should normally be zero.
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/macros_msa.h b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/macros_msa.h
>index 921eb0714d66faf26cbcda7ec88762fff410d52f..29997ce11fd12ded65c8d6f8603f52ba4cd041dd 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/macros_msa.h
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/macros_msa.h
>@@ -16,30 +16,30 @@
> #include <stdint.h>
> 
> #if (__mips_isa_rev >= 6)
>-#define LW(psrc)                                        \
>-  ({                                                    \
>-    uint8_t* psrc_lw_m = (uint8_t*)(psrc); /* NOLINT */ \
>-    uint32_t val_m;                                     \
>-    asm volatile("lw  %[val_m],  %[psrc_lw_m]  \n"      \
>-                 : [val_m] "=r"(val_m)                  \
>-                 : [psrc_lw_m] "m"(*psrc_lw_m));        \
>-    val_m;                                              \
>+#define LW(psrc)                                       \
>+  ({                                                   \
>+    const uint8_t* psrc_lw_m = (const uint8_t*)(psrc); \
>+    uint32_t val_m;                                    \
>+    asm volatile("lw  %[val_m],  %[psrc_lw_m]  \n"     \
>+                 : [val_m] "=r"(val_m)                 \
>+                 : [psrc_lw_m] "m"(*psrc_lw_m));       \
>+    val_m;                                             \
>   })
> 
> #if (__mips == 64)
>-#define LD(psrc)                                        \
>-  ({                                                    \
>-    uint8_t* psrc_ld_m = (uint8_t*)(psrc); /* NOLINT */ \
>-    uint64_t val_m = 0;                                 \
>-    asm volatile("ld  %[val_m],  %[psrc_ld_m]  \n"      \
>-                 : [val_m] "=r"(val_m)                  \
>-                 : [psrc_ld_m] "m"(*psrc_ld_m));        \
>-    val_m;                                              \
>+#define LD(psrc)                                       \
>+  ({                                                   \
>+    const uint8_t* psrc_ld_m = (const uint8_t*)(psrc); \
>+    uint64_t val_m = 0;                                \
>+    asm volatile("ld  %[val_m],  %[psrc_ld_m]  \n"     \
>+                 : [val_m] "=r"(val_m)                 \
>+                 : [psrc_ld_m] "m"(*psrc_ld_m));       \
>+    val_m;                                             \
>   })
> #else  // !(__mips == 64)
> #define LD(psrc)                                                         \
>   ({                                                                     \
>-    uint8_t* psrc_ld_m = (uint8_t*)(psrc); /* NOLINT */                  \
>+    const uint8_t* psrc_ld_m = (const uint8_t*)(psrc);                   \
>     uint32_t val0_m, val1_m;                                             \
>     uint64_t val_m = 0;                                                  \
>     val0_m = LW(psrc_ld_m);                                              \
>@@ -81,30 +81,30 @@
>   })
> #endif  // !(__mips == 64)
> #else   // !(__mips_isa_rev >= 6)
>-#define LW(psrc)                                        \
>-  ({                                                    \
>-    uint8_t* psrc_lw_m = (uint8_t*)(psrc); /* NOLINT */ \
>-    uint32_t val_m;                                     \
>-    asm volatile("ulw  %[val_m],  %[psrc_lw_m]  \n"     \
>-                 : [val_m] "=r"(val_m)                  \
>-                 : [psrc_lw_m] "m"(*psrc_lw_m));        \
>-    val_m;                                              \
>+#define LW(psrc)                                       \
>+  ({                                                   \
>+    const uint8_t* psrc_lw_m = (const uint8_t*)(psrc); \
>+    uint32_t val_m;                                    \
>+    asm volatile("ulw  %[val_m],  %[psrc_lw_m]  \n"    \
>+                 : [val_m] "=r"(val_m)                 \
>+                 : [psrc_lw_m] "m"(*psrc_lw_m));       \
>+    val_m;                                             \
>   })
> 
> #if (__mips == 64)
>-#define LD(psrc)                                        \
>-  ({                                                    \
>-    uint8_t* psrc_ld_m = (uint8_t*)(psrc); /* NOLINT */ \
>-    uint64_t val_m = 0;                                 \
>-    asm volatile("uld  %[val_m],  %[psrc_ld_m]  \n"     \
>-                 : [val_m] "=r"(val_m)                  \
>-                 : [psrc_ld_m] "m"(*psrc_ld_m));        \
>-    val_m;                                              \
>+#define LD(psrc)                                       \
>+  ({                                                   \
>+    const uint8_t* psrc_ld_m = (const uint8_t*)(psrc); \
>+    uint64_t val_m = 0;                                \
>+    asm volatile("uld  %[val_m],  %[psrc_ld_m]  \n"    \
>+                 : [val_m] "=r"(val_m)                 \
>+                 : [psrc_ld_m] "m"(*psrc_ld_m));       \
>+    val_m;                                             \
>   })
> #else  // !(__mips == 64)
> #define LD(psrc)                                                         \
>   ({                                                                     \
>-    uint8_t* psrc_ld_m = (uint8_t*)(psrc); /* NOLINT */                  \
>+    const uint8_t* psrc_ld_m = (const uint8_t*)(psrc);                   \
>     uint32_t val0_m, val1_m;                                             \
>     uint64_t val_m = 0;                                                  \
>     val0_m = LW(psrc_ld_m);                                              \
>@@ -138,7 +138,7 @@
> 
> // TODO(fbarchard): Consider removing __VAR_ARGS versions.
> #define LD_B(RTYPE, psrc) *((RTYPE*)(psrc)) /* NOLINT */
>-#define LD_UB(...) LD_B(v16u8, __VA_ARGS__)
>+#define LD_UB(...) LD_B(const v16u8, __VA_ARGS__)
> 
> #define ST_B(RTYPE, in, pdst) *((RTYPE*)(pdst)) = (in) /* NOLINT */
> #define ST_UB(...) ST_B(v16u8, __VA_ARGS__)
>@@ -158,14 +158,14 @@
>     out0 = LD_B(RTYPE, (psrc));                \
>     out1 = LD_B(RTYPE, (psrc) + stride);       \
>   }
>-#define LD_UB2(...) LD_B2(v16u8, __VA_ARGS__)
>+#define LD_UB2(...) LD_B2(const 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 LD_UB4(...) LD_B4(const v16u8, __VA_ARGS__)
> 
> /* Description : Store two vectors with stride each having 16 'byte' sized
>                  elements
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/mjpeg_decoder.h b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/mjpeg_decoder.h
>index 6c12633387fad8234ddb8f0691564aaed029d7ee..275f8d4c18532003248bf4c51d0bce36b737c9a8 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/mjpeg_decoder.h
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/mjpeg_decoder.h
>@@ -168,8 +168,8 @@ class LIBYUV_API MJpegDecoder {
>   int GetComponentScanlinePadding(int component);
> 
>   // A buffer holding the input data for a frame.
>-  Buffer buf_{};
>-  BufferVector buf_vec_{};
>+  Buffer buf_;
>+  BufferVector buf_vec_;
> 
>   jpeg_decompress_struct* decompress_struct_;
>   jpeg_source_mgr* source_mgr_;
>@@ -181,12 +181,12 @@ class LIBYUV_API MJpegDecoder {
> 
>   // Temporaries used to point to scanline outputs.
>   int num_outbufs_;  // Outermost size of all arrays below.
>-  uint8_t*** scanlines_{};
>-  int* scanlines_sizes_{};
>+  uint8_t*** scanlines_;
>+  int* scanlines_sizes_;
>   // Temporary buffer used for decoding when we can't decode directly to the
>   // output buffers. Large enough for just one iMCU row.
>-  uint8_t** databuf_{};
>-  int* databuf_strides_{};
>+  uint8_t** databuf_;
>+  int* databuf_strides_;
> };
> 
> }  // namespace libyuv
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/rotate_row.h b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/rotate_row.h
>index 5edc0fcf13a1b632fb4a2ca127567e5de031cacb..022293eef2c9580c2ca46d94633db8e84b58f9ba 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/rotate_row.h
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/rotate_row.h
>@@ -60,6 +60,11 @@ extern "C" {
> #define HAS_TRANSPOSEUVWX16_MSA
> #endif
> 
>+#if !defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)
>+#define HAS_TRANSPOSEWX8_MMI
>+#define HAS_TRANSPOSEUVWX8_MMI
>+#endif
>+
> void TransposeWxH_C(const uint8_t* src,
>                     int src_stride,
>                     uint8_t* dst,
>@@ -87,6 +92,11 @@ void TransposeWx8_SSSE3(const uint8_t* src,
>                         uint8_t* dst,
>                         int dst_stride,
>                         int width);
>+void TransposeWx8_MMI(const uint8_t* src,
>+                      int src_stride,
>+                      uint8_t* dst,
>+                      int dst_stride,
>+                      int width);
> void TransposeWx8_Fast_SSSE3(const uint8_t* src,
>                              int src_stride,
>                              uint8_t* dst,
>@@ -108,6 +118,11 @@ void TransposeWx8_Any_SSSE3(const uint8_t* src,
>                             uint8_t* dst,
>                             int dst_stride,
>                             int width);
>+void TransposeWx8_Any_MMI(const uint8_t* src,
>+                          int src_stride,
>+                          uint8_t* dst,
>+                          int dst_stride,
>+                          int width);
> void TransposeWx8_Fast_Any_SSSE3(const uint8_t* src,
>                                  int src_stride,
>                                  uint8_t* dst,
>@@ -156,6 +171,13 @@ void TransposeUVWx8_NEON(const uint8_t* src,
>                          uint8_t* dst_b,
>                          int dst_stride_b,
>                          int width);
>+void TransposeUVWx8_MMI(const uint8_t* src,
>+                        int src_stride,
>+                        uint8_t* dst_a,
>+                        int dst_stride_a,
>+                        uint8_t* dst_b,
>+                        int dst_stride_b,
>+                        int width);
> void TransposeUVWx16_MSA(const uint8_t* src,
>                          int src_stride,
>                          uint8_t* dst_a,
>@@ -178,6 +200,13 @@ void TransposeUVWx8_Any_NEON(const uint8_t* src,
>                              uint8_t* dst_b,
>                              int dst_stride_b,
>                              int width);
>+void TransposeUVWx8_Any_MMI(const uint8_t* src,
>+                            int src_stride,
>+                            uint8_t* dst_a,
>+                            int dst_stride_a,
>+                            uint8_t* dst_b,
>+                            int dst_stride_b,
>+                            int width);
> void TransposeUVWx16_Any_MSA(const uint8_t* src,
>                              int src_stride,
>                              uint8_t* dst_a,
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/row.h b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/row.h
>index b5a42d0e62cfcf5c43fe753e78c92e669442cec0..cc948878f668979c9836eea680830a92bb126ee6 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/row.h
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/row.h
>@@ -55,6 +55,15 @@ extern "C" {
> #endif  // clang >= 3.4
> #endif  // __clang__
> 
>+// clang >= 6.0.0 required for AVX512.
>+#if defined(__clang__) && (defined(__x86_64__) || defined(__i386__))
>+// clang in xcode follows a different versioning scheme.
>+// TODO(fbarchard): fix xcode 9 ios b/789.
>+#if (__clang_major__ >= 7) && !defined(__APPLE__)
>+#define CLANG_HAS_AVX512 1
>+#endif  // clang >= 7
>+#endif  // __clang__
>+
> // Visual C 2012 required for AVX2.
> #if defined(_M_IX86) && !defined(__clang__) && defined(_MSC_VER) && \
>     _MSC_VER >= 1700
>@@ -105,8 +114,10 @@ extern "C" {
> #define HAS_MIRRORROW_SSSE3
> #define HAS_MIRRORUVROW_SSSE3
> #define HAS_NV12TOARGBROW_SSSE3
>+#define HAS_NV12TORGB24ROW_SSSE3
> #define HAS_NV12TORGB565ROW_SSSE3
> #define HAS_NV21TOARGBROW_SSSE3
>+#define HAS_NV21TORGB24ROW_SSSE3
> #define HAS_RAWTOARGBROW_SSSE3
> #define HAS_RAWTORGB24ROW_SSSE3
> #define HAS_RAWTOYROW_SSSE3
>@@ -200,8 +211,10 @@ extern "C" {
> #define HAS_MERGEUVROW_AVX2
> #define HAS_MIRRORROW_AVX2
> #define HAS_NV12TOARGBROW_AVX2
>+#define HAS_NV12TORGB24ROW_AVX2
> #define HAS_NV12TORGB565ROW_AVX2
> #define HAS_NV21TOARGBROW_AVX2
>+#define HAS_NV21TORGB24ROW_AVX2
> #define HAS_SPLITUVROW_AVX2
> #define HAS_UYVYTOARGBROW_AVX2
> #define HAS_UYVYTOUV422ROW_AVX2
>@@ -230,7 +243,7 @@ extern "C" {
> 
> // The following are available for AVX2 Visual C and clangcl 32 bit:
> // TODO(fbarchard): Port to gcc.
>-#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && \
>+#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && defined(_MSC_VER) && \
>     (defined(VISUALC_HAS_AVX2) || defined(CLANG_HAS_AVX2))
> #define HAS_ARGB1555TOARGBROW_AVX2
> #define HAS_ARGB4444TOARGBROW_AVX2
>@@ -271,10 +284,13 @@ extern "C" {
>     (defined(CLANG_HAS_AVX2) || defined(GCC_HAS_AVX2))
> #define HAS_ABGRTOAR30ROW_AVX2
> #define HAS_ARGBTOAR30ROW_AVX2
>+// Fix AVX2 b:118386049 segfault
>+//#define HAS_ARGBTORAWROW_AVX2
>+//#define HAS_ARGBTORGB24ROW_AVX2
> #define HAS_CONVERT16TO8ROW_AVX2
> #define HAS_CONVERT8TO16ROW_AVX2
>-#define HAS_I210TOARGBROW_AVX2
> #define HAS_I210TOAR30ROW_AVX2
>+#define HAS_I210TOARGBROW_AVX2
> #define HAS_I422TOAR30ROW_AVX2
> #define HAS_I422TOUYVYROW_AVX2
> #define HAS_I422TOYUY2ROW_AVX2
>@@ -282,6 +298,16 @@ extern "C" {
> #define HAS_MULTIPLYROW_16_AVX2
> #endif
> 
>+// The following are available for AVX512 clang x86 platforms:
>+// TODO(fbarchard): Port to GCC and Visual C
>+// TODO(fbarchard): re-enable HAS_ARGBTORGB24ROW_AVX512VBMI. Issue libyuv:789
>+#if !defined(LIBYUV_DISABLE_X86) &&                                       \
>+    (defined(__x86_64__) || (defined(__i386__) && !defined(_MSC_VER))) && \
>+    (defined(CLANG_HAS_AVX512))
>+// Fix AVX2 b:118386049 segfault
>+//#define HAS_ARGBTORGB24ROW_AVX512VBMI
>+#endif
>+
> // The following are available on Neon platforms:
> #if !defined(LIBYUV_DISABLE_NEON) && \
>     (defined(__aarch64__) || defined(__ARM_NEON__) || defined(LIBYUV_NEON))
>@@ -459,6 +485,81 @@ extern "C" {
> #define HAS_YUY2TOYROW_MSA
> #endif
> 
>+#if !defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)
>+#define HAS_ABGRTOUVROW_MMI
>+#define HAS_ABGRTOYROW_MMI
>+#define HAS_ARGB1555TOARGBROW_MMI
>+#define HAS_ARGB1555TOUVROW_MMI
>+#define HAS_ARGB1555TOYROW_MMI
>+#define HAS_ARGB4444TOARGBROW_MMI
>+#define HAS_ARGB4444TOUVROW_MMI
>+#define HAS_ARGB4444TOYROW_MMI
>+#define HAS_ARGBADDROW_MMI
>+#define HAS_ARGBATTENUATEROW_MMI
>+#define HAS_ARGBBLENDROW_MMI
>+#define HAS_ARGBCOLORMATRIXROW_MMI
>+#define HAS_ARGBCOPYALPHAROW_MMI
>+#define HAS_ARGBCOPYYTOALPHAROW_MMI
>+#define HAS_ARGBEXTRACTALPHAROW_MMI
>+#define HAS_ARGBGRAYROW_MMI
>+#define HAS_ARGBMIRRORROW_MMI
>+#define HAS_ARGBMULTIPLYROW_MMI
>+#define HAS_ARGBSEPIAROW_MMI
>+#define HAS_ARGBSHADEROW_MMI
>+#define HAS_ARGBSHUFFLEROW_MMI
>+#define HAS_ARGBSUBTRACTROW_MMI
>+#define HAS_ARGBTOARGB1555ROW_MMI
>+#define HAS_ARGBTOARGB4444ROW_MMI
>+#define HAS_ARGBTORAWROW_MMI
>+#define HAS_ARGBTORGB24ROW_MMI
>+#define HAS_ARGBTORGB565DITHERROW_MMI
>+#define HAS_ARGBTORGB565ROW_MMI
>+#define HAS_ARGBTOUV444ROW_MMI
>+#define HAS_ARGBTOUVJROW_MMI
>+#define HAS_ARGBTOUVROW_MMI
>+#define HAS_ARGBTOYJROW_MMI
>+#define HAS_ARGBTOYROW_MMI
>+#define HAS_BGRATOUVROW_MMI
>+#define HAS_BGRATOYROW_MMI
>+#define HAS_BLENDPLANEROW_MMI
>+#define HAS_COMPUTECUMULATIVESUMROW_MMI
>+#define HAS_CUMULATIVESUMTOAVERAGEROW_MMI
>+#define HAS_HALFFLOATROW_MMI
>+#define HAS_I400TOARGBROW_MMI
>+#define HAS_I422TOUYVYROW_MMI
>+#define HAS_I422TOYUY2ROW_MMI
>+#define HAS_INTERPOLATEROW_MMI
>+#define HAS_J400TOARGBROW_MMI
>+#define HAS_MERGERGBROW_MMI
>+#define HAS_MERGEUVROW_MMI
>+#define HAS_MIRRORROW_MMI
>+#define HAS_MIRRORUVROW_MMI
>+#define HAS_RAWTOARGBROW_MMI
>+#define HAS_RAWTORGB24ROW_MMI
>+#define HAS_RAWTOUVROW_MMI
>+#define HAS_RAWTOYROW_MMI
>+#define HAS_RGB24TOARGBROW_MMI
>+#define HAS_RGB24TOUVROW_MMI
>+#define HAS_RGB24TOYROW_MMI
>+#define HAS_RGB565TOARGBROW_MMI
>+#define HAS_RGB565TOUVROW_MMI
>+#define HAS_RGB565TOYROW_MMI
>+#define HAS_RGBATOUVROW_MMI
>+#define HAS_RGBATOYROW_MMI
>+#define HAS_SOBELROW_MMI
>+#define HAS_SOBELTOPLANEROW_MMI
>+#define HAS_SOBELXROW_MMI
>+#define HAS_SOBELXYROW_MMI
>+#define HAS_SOBELYROW_MMI
>+#define HAS_SPLITRGBROW_MMI
>+#define HAS_SPLITUVROW_MMI
>+#define HAS_UYVYTOUVROW_MMI
>+#define HAS_UYVYTOYROW_MMI
>+#define HAS_YUY2TOUV422ROW_MMI
>+#define HAS_YUY2TOUVROW_MMI
>+#define HAS_YUY2TOYROW_MMI
>+#endif
>+
> #if defined(_MSC_VER) && !defined(__CLR_VER) && !defined(__clang__)
> #if defined(VISUALC_HAS_AVX2)
> #define SIMD_ALIGNED(var) __declspec(align(32)) var
>@@ -813,6 +914,8 @@ void ARGBToYRow_NEON(const uint8_t* src_argb, uint8_t* dst_y, int width);
> void ARGBToYJRow_NEON(const uint8_t* src_argb, uint8_t* dst_y, int width);
> void ARGBToYRow_MSA(const uint8_t* src_argb0, uint8_t* dst_y, int width);
> void ARGBToYJRow_MSA(const uint8_t* src_argb0, uint8_t* dst_y, int width);
>+void ARGBToYRow_MMI(const uint8_t* src_argb0, uint8_t* dst_y, int width);
>+void ARGBToYJRow_MMI(const uint8_t* src_argb0, uint8_t* dst_y, int width);
> void ARGBToUV444Row_NEON(const uint8_t* src_argb,
>                          uint8_t* dst_u,
>                          uint8_t* dst_v,
>@@ -831,6 +934,15 @@ void ARGBToUVRow_MSA(const uint8_t* src_argb0,
>                      uint8_t* dst_u,
>                      uint8_t* dst_v,
>                      int width);
>+void ARGBToUV444Row_MMI(const uint8_t* src_argb,
>+                        uint8_t* dst_u,
>+                        uint8_t* dst_v,
>+                        int width);
>+void ARGBToUVRow_MMI(const uint8_t* src_argb0,
>+                     int src_stride_argb,
>+                     uint8_t* dst_u,
>+                     uint8_t* dst_v,
>+                     int width);
> void ARGBToUVJRow_NEON(const uint8_t* src_argb,
>                        int src_stride_argb,
>                        uint8_t* dst_u,
>@@ -916,6 +1028,51 @@ void ARGB1555ToUVRow_MSA(const uint8_t* src_argb1555,
>                          uint8_t* dst_u,
>                          uint8_t* dst_v,
>                          int width);
>+void ARGBToUVJRow_MMI(const uint8_t* src_rgb0,
>+                      int src_stride_rgb,
>+                      uint8_t* dst_u,
>+                      uint8_t* dst_v,
>+                      int width);
>+void BGRAToUVRow_MMI(const uint8_t* src_rgb0,
>+                     int src_stride_rgb,
>+                     uint8_t* dst_u,
>+                     uint8_t* dst_v,
>+                     int width);
>+void ABGRToUVRow_MMI(const uint8_t* src_rgb0,
>+                     int src_stride_rgb,
>+                     uint8_t* dst_u,
>+                     uint8_t* dst_v,
>+                     int width);
>+void RGBAToUVRow_MMI(const uint8_t* src_rgb0,
>+                     int src_stride_rgb,
>+                     uint8_t* dst_u,
>+                     uint8_t* dst_v,
>+                     int width);
>+void RGB24ToUVRow_MMI(const uint8_t* src_rgb0,
>+                      int src_stride_rgb,
>+                      uint8_t* dst_u,
>+                      uint8_t* dst_v,
>+                      int width);
>+void RAWToUVRow_MMI(const uint8_t* src_rgb0,
>+                    int src_stride_rgb,
>+                    uint8_t* dst_u,
>+                    uint8_t* dst_v,
>+                    int width);
>+void RGB565ToUVRow_MMI(const uint8_t* src_rgb565,
>+                       int src_stride_rgb565,
>+                       uint8_t* dst_u,
>+                       uint8_t* dst_v,
>+                       int width);
>+void ARGB1555ToUVRow_MMI(const uint8_t* src_argb1555,
>+                         int src_stride_argb1555,
>+                         uint8_t* dst_u,
>+                         uint8_t* dst_v,
>+                         int width);
>+void ARGB4444ToUVRow_MMI(const uint8_t* src_argb4444,
>+                         int src_stride_argb4444,
>+                         uint8_t* dst_u,
>+                         uint8_t* dst_v,
>+                         int width);
> void BGRAToYRow_NEON(const uint8_t* src_bgra, uint8_t* dst_y, int width);
> void ABGRToYRow_NEON(const uint8_t* src_abgr, uint8_t* dst_y, int width);
> void RGBAToYRow_NEON(const uint8_t* src_rgba, uint8_t* dst_y, int width);
>@@ -935,6 +1092,15 @@ void RGB24ToYRow_MSA(const uint8_t* src_argb0, uint8_t* dst_y, int width);
> void RAWToYRow_MSA(const uint8_t* src_argb0, uint8_t* dst_y, int width);
> void RGB565ToYRow_MSA(const uint8_t* src_rgb565, uint8_t* dst_y, int width);
> void ARGB1555ToYRow_MSA(const uint8_t* src_argb1555, uint8_t* dst_y, int width);
>+void BGRAToYRow_MMI(const uint8_t* src_argb0, uint8_t* dst_y, int width);
>+void ABGRToYRow_MMI(const uint8_t* src_argb0, uint8_t* dst_y, int width);
>+void RGBAToYRow_MMI(const uint8_t* src_argb0, uint8_t* dst_y, int width);
>+void RGB24ToYRow_MMI(const uint8_t* src_argb0, uint8_t* dst_y, int width);
>+void RAWToYRow_MMI(const uint8_t* src_argb0, uint8_t* dst_y, int width);
>+void RGB565ToYRow_MMI(const uint8_t* src_rgb565, uint8_t* dst_y, int width);
>+void ARGB1555ToYRow_MMI(const uint8_t* src_argb1555, uint8_t* dst_y, int width);
>+void ARGB4444ToYRow_MMI(const uint8_t* src_argb4444, uint8_t* dst_y, int width);
>+
> void ARGBToYRow_C(const uint8_t* src_argb0, uint8_t* dst_y, int width);
> void ARGBToYJRow_C(const uint8_t* src_argb0, uint8_t* dst_y, int width);
> void BGRAToYRow_C(const uint8_t* src_argb0, uint8_t* dst_y, int width);
>@@ -977,6 +1143,20 @@ void RGB565ToYRow_Any_MSA(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
> void ARGB1555ToYRow_Any_MSA(const uint8_t* src_ptr,
>                             uint8_t* dst_ptr,
>                             int width);
>+void BGRAToYRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
>+void ABGRToYRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
>+void RGBAToYRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
>+void ARGBToYJRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
>+void ARGBToYRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
>+void RGB24ToYRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
>+void RAWToYRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
>+void RGB565ToYRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
>+void ARGB1555ToYRow_Any_MMI(const uint8_t* src_ptr,
>+                            uint8_t* dst_ptr,
>+                            int width);
>+void ARGB4444ToYRow_Any_MMI(const uint8_t* src_ptr,
>+                            uint8_t* dst_ptr,
>+                            int width);
> 
> void ARGBToUVRow_AVX2(const uint8_t* src_argb0,
>                       int src_stride_argb,
>@@ -1066,6 +1246,15 @@ void ARGBToUVRow_Any_MSA(const uint8_t* src_ptr,
>                          uint8_t* dst_u,
>                          uint8_t* dst_v,
>                          int width);
>+void ARGBToUV444Row_Any_MMI(const uint8_t* src_ptr,
>+                            uint8_t* dst_u,
>+                            uint8_t* dst_v,
>+                            int width);
>+void ARGBToUVRow_Any_MMI(const uint8_t* src_ptr,
>+                         int src_stride_ptr,
>+                         uint8_t* dst_u,
>+                         uint8_t* dst_v,
>+                         int width);
> void ARGBToUVJRow_Any_NEON(const uint8_t* src_ptr,
>                            int src_stride_ptr,
>                            uint8_t* dst_u,
>@@ -1151,6 +1340,51 @@ void ARGB1555ToUVRow_Any_MSA(const uint8_t* src_ptr,
>                              uint8_t* dst_u,
>                              uint8_t* dst_v,
>                              int width);
>+void ARGBToUVJRow_Any_MMI(const uint8_t* src_ptr,
>+                          int src_stride_ptr,
>+                          uint8_t* dst_u,
>+                          uint8_t* dst_v,
>+                          int width);
>+void BGRAToUVRow_Any_MMI(const uint8_t* src_ptr,
>+                         int src_stride_ptr,
>+                         uint8_t* dst_u,
>+                         uint8_t* dst_v,
>+                         int width);
>+void ABGRToUVRow_Any_MMI(const uint8_t* src_ptr,
>+                         int src_stride_ptr,
>+                         uint8_t* dst_u,
>+                         uint8_t* dst_v,
>+                         int width);
>+void RGBAToUVRow_Any_MMI(const uint8_t* src_ptr,
>+                         int src_stride_ptr,
>+                         uint8_t* dst_u,
>+                         uint8_t* dst_v,
>+                         int width);
>+void RGB24ToUVRow_Any_MMI(const uint8_t* src_ptr,
>+                          int src_stride_ptr,
>+                          uint8_t* dst_u,
>+                          uint8_t* dst_v,
>+                          int width);
>+void RAWToUVRow_Any_MMI(const uint8_t* src_ptr,
>+                        int src_stride_ptr,
>+                        uint8_t* dst_u,
>+                        uint8_t* dst_v,
>+                        int width);
>+void RGB565ToUVRow_Any_MMI(const uint8_t* src_ptr,
>+                           int src_stride_ptr,
>+                           uint8_t* dst_u,
>+                           uint8_t* dst_v,
>+                           int width);
>+void ARGB1555ToUVRow_Any_MMI(const uint8_t* src_ptr,
>+                             int src_stride_ptr,
>+                             uint8_t* dst_u,
>+                             uint8_t* dst_v,
>+                             int width);
>+void ARGB4444ToUVRow_Any_MMI(const uint8_t* src_ptr,
>+                             int src_stride_ptr,
>+                             uint8_t* dst_u,
>+                             uint8_t* dst_v,
>+                             int width);
> void ARGBToUVRow_C(const uint8_t* src_rgb0,
>                    int src_stride_rgb,
>                    uint8_t* dst_u,
>@@ -1230,12 +1464,14 @@ void MirrorRow_AVX2(const uint8_t* src, uint8_t* dst, int width);
> void MirrorRow_SSSE3(const uint8_t* src, uint8_t* dst, int width);
> void MirrorRow_NEON(const uint8_t* src, uint8_t* dst, int width);
> void MirrorRow_MSA(const uint8_t* src, uint8_t* dst, int width);
>+void MirrorRow_MMI(const uint8_t* src, uint8_t* dst, int width);
> void MirrorRow_C(const uint8_t* src, uint8_t* dst, int width);
> void MirrorRow_Any_AVX2(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
> void MirrorRow_Any_SSSE3(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
> void MirrorRow_Any_SSE2(const uint8_t* src, uint8_t* dst, int width);
> void MirrorRow_Any_NEON(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
> void MirrorRow_Any_MSA(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
>+void MirrorRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
> 
> void MirrorUVRow_SSSE3(const uint8_t* src,
>                        uint8_t* dst_u,
>@@ -1249,6 +1485,10 @@ void MirrorUVRow_MSA(const uint8_t* src_uv,
>                      uint8_t* dst_u,
>                      uint8_t* dst_v,
>                      int width);
>+void MirrorUVRow_MMI(const uint8_t* src_uv,
>+                     uint8_t* dst_u,
>+                     uint8_t* dst_v,
>+                     int width);
> void MirrorUVRow_C(const uint8_t* src_uv,
>                    uint8_t* dst_u,
>                    uint8_t* dst_v,
>@@ -1258,6 +1498,7 @@ void ARGBMirrorRow_AVX2(const uint8_t* src, uint8_t* dst, int width);
> void ARGBMirrorRow_SSE2(const uint8_t* src, uint8_t* dst, int width);
> void ARGBMirrorRow_NEON(const uint8_t* src, uint8_t* dst, int width);
> void ARGBMirrorRow_MSA(const uint8_t* src, uint8_t* dst, int width);
>+void ARGBMirrorRow_MMI(const uint8_t* src, uint8_t* dst, int width);
> void ARGBMirrorRow_C(const uint8_t* src, uint8_t* dst, int width);
> void ARGBMirrorRow_Any_AVX2(const uint8_t* src_ptr,
>                             uint8_t* dst_ptr,
>@@ -1269,6 +1510,7 @@ void ARGBMirrorRow_Any_NEON(const uint8_t* src_ptr,
>                             uint8_t* dst_ptr,
>                             int width);
> void ARGBMirrorRow_Any_MSA(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
>+void ARGBMirrorRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
> 
> void SplitUVRow_C(const uint8_t* src_uv,
>                   uint8_t* dst_u,
>@@ -1290,6 +1532,10 @@ void SplitUVRow_MSA(const uint8_t* src_uv,
>                     uint8_t* dst_u,
>                     uint8_t* dst_v,
>                     int width);
>+void SplitUVRow_MMI(const uint8_t* src_uv,
>+                    uint8_t* dst_u,
>+                    uint8_t* dst_v,
>+                    int width);
> void SplitUVRow_Any_SSE2(const uint8_t* src_ptr,
>                          uint8_t* dst_u,
>                          uint8_t* dst_v,
>@@ -1306,6 +1552,10 @@ void SplitUVRow_Any_MSA(const uint8_t* src_ptr,
>                         uint8_t* dst_u,
>                         uint8_t* dst_v,
>                         int width);
>+void SplitUVRow_Any_MMI(const uint8_t* src_ptr,
>+                        uint8_t* dst_u,
>+                        uint8_t* dst_v,
>+                        int width);
> 
> void MergeUVRow_C(const uint8_t* src_u,
>                   const uint8_t* src_v,
>@@ -1327,6 +1577,10 @@ void MergeUVRow_MSA(const uint8_t* src_u,
>                     const uint8_t* src_v,
>                     uint8_t* dst_uv,
>                     int width);
>+void MergeUVRow_MMI(const uint8_t* src_u,
>+                    const uint8_t* src_v,
>+                    uint8_t* dst_uv,
>+                    int width);
> void MergeUVRow_Any_SSE2(const uint8_t* y_buf,
>                          const uint8_t* uv_buf,
>                          uint8_t* dst_ptr,
>@@ -1343,6 +1597,10 @@ void MergeUVRow_Any_MSA(const uint8_t* y_buf,
>                         const uint8_t* uv_buf,
>                         uint8_t* dst_ptr,
>                         int width);
>+void MergeUVRow_Any_MMI(const uint8_t* y_buf,
>+                        const uint8_t* uv_buf,
>+                        uint8_t* dst_ptr,
>+                        int width);
> 
> void SplitRGBRow_C(const uint8_t* src_rgb,
>                    uint8_t* dst_r,
>@@ -1359,6 +1617,11 @@ void SplitRGBRow_NEON(const uint8_t* src_rgb,
>                       uint8_t* dst_g,
>                       uint8_t* dst_b,
>                       int width);
>+void SplitRGBRow_MMI(const uint8_t* src_rgb,
>+                     uint8_t* dst_r,
>+                     uint8_t* dst_g,
>+                     uint8_t* dst_b,
>+                     int width);
> void SplitRGBRow_Any_SSSE3(const uint8_t* src_ptr,
>                            uint8_t* dst_r,
>                            uint8_t* dst_g,
>@@ -1369,6 +1632,11 @@ void SplitRGBRow_Any_NEON(const uint8_t* src_ptr,
>                           uint8_t* dst_g,
>                           uint8_t* dst_b,
>                           int width);
>+void SplitRGBRow_Any_MMI(const uint8_t* src_ptr,
>+                         uint8_t* dst_r,
>+                         uint8_t* dst_g,
>+                         uint8_t* dst_b,
>+                         int width);
> 
> void MergeRGBRow_C(const uint8_t* src_r,
>                    const uint8_t* src_g,
>@@ -1385,6 +1653,11 @@ void MergeRGBRow_NEON(const uint8_t* src_r,
>                       const uint8_t* src_b,
>                       uint8_t* dst_rgb,
>                       int width);
>+void MergeRGBRow_MMI(const uint8_t* src_r,
>+                     const uint8_t* src_g,
>+                     const uint8_t* src_b,
>+                     uint8_t* dst_rgb,
>+                     int width);
> void MergeRGBRow_Any_SSSE3(const uint8_t* y_buf,
>                            const uint8_t* u_buf,
>                            const uint8_t* v_buf,
>@@ -1395,6 +1668,11 @@ void MergeRGBRow_Any_NEON(const uint8_t* src_r,
>                           const uint8_t* src_b,
>                           uint8_t* dst_rgb,
>                           int width);
>+void MergeRGBRow_Any_MMI(const uint8_t* src_r,
>+                         const uint8_t* src_g,
>+                         const uint8_t* src_b,
>+                         uint8_t* dst_rgb,
>+                         int width);
> 
> void MergeUVRow_16_C(const uint16_t* src_u,
>                      const uint16_t* src_v,
>@@ -1473,12 +1751,16 @@ void CopyRow_16_C(const uint16_t* src, uint16_t* dst, int count);
> void ARGBCopyAlphaRow_C(const uint8_t* src, uint8_t* dst, int width);
> void ARGBCopyAlphaRow_SSE2(const uint8_t* src, uint8_t* dst, int width);
> void ARGBCopyAlphaRow_AVX2(const uint8_t* src, uint8_t* dst, int width);
>+void ARGBCopyAlphaRow_MMI(const uint8_t* src, uint8_t* dst, int width);
> void ARGBCopyAlphaRow_Any_SSE2(const uint8_t* src_ptr,
>                                uint8_t* dst_ptr,
>                                int width);
> void ARGBCopyAlphaRow_Any_AVX2(const uint8_t* src_ptr,
>                                uint8_t* dst_ptr,
>                                int width);
>+void ARGBCopyAlphaRow_Any_MMI(const uint8_t* src_ptr,
>+                              uint8_t* dst_ptr,
>+                              int width);
> 
> void ARGBExtractAlphaRow_C(const uint8_t* src_argb, uint8_t* dst_a, int width);
> void ARGBExtractAlphaRow_SSE2(const uint8_t* src_argb,
>@@ -1493,6 +1775,9 @@ void ARGBExtractAlphaRow_NEON(const uint8_t* src_argb,
> void ARGBExtractAlphaRow_MSA(const uint8_t* src_argb,
>                              uint8_t* dst_a,
>                              int width);
>+void ARGBExtractAlphaRow_MMI(const uint8_t* src_argb,
>+                             uint8_t* dst_a,
>+                             int width);
> void ARGBExtractAlphaRow_Any_SSE2(const uint8_t* src_ptr,
>                                   uint8_t* dst_ptr,
>                                   int width);
>@@ -1505,16 +1790,23 @@ void ARGBExtractAlphaRow_Any_NEON(const uint8_t* src_ptr,
> void ARGBExtractAlphaRow_Any_MSA(const uint8_t* src_ptr,
>                                  uint8_t* dst_ptr,
>                                  int width);
>+void ARGBExtractAlphaRow_Any_MMI(const uint8_t* src_ptr,
>+                                 uint8_t* dst_ptr,
>+                                 int width);
> 
> void ARGBCopyYToAlphaRow_C(const uint8_t* src, uint8_t* dst, int width);
> void ARGBCopyYToAlphaRow_SSE2(const uint8_t* src, uint8_t* dst, int width);
> void ARGBCopyYToAlphaRow_AVX2(const uint8_t* src, uint8_t* dst, int width);
>+void ARGBCopyYToAlphaRow_MMI(const uint8_t* src, uint8_t* dst, int width);
> void ARGBCopyYToAlphaRow_Any_SSE2(const uint8_t* src_ptr,
>                                   uint8_t* dst_ptr,
>                                   int width);
> void ARGBCopyYToAlphaRow_Any_AVX2(const uint8_t* src_ptr,
>                                   uint8_t* dst_ptr,
>                                   int width);
>+void ARGBCopyYToAlphaRow_Any_MMI(const uint8_t* src_ptr,
>+                                 uint8_t* dst_ptr,
>+                                 int width);
> 
> void SetRow_C(uint8_t* dst, uint8_t v8, int width);
> void SetRow_MSA(uint8_t* dst, uint8_t v8, int width);
>@@ -1552,6 +1844,10 @@ void ARGBShuffleRow_MSA(const uint8_t* src_argb,
>                         uint8_t* dst_argb,
>                         const uint8_t* shuffler,
>                         int width);
>+void ARGBShuffleRow_MMI(const uint8_t* src_argb,
>+                        uint8_t* dst_argb,
>+                        const uint8_t* shuffler,
>+                        int width);
> void ARGBShuffleRow_Any_SSSE3(const uint8_t* src_ptr,
>                               uint8_t* dst_ptr,
>                               const uint8_t* param,
>@@ -1568,6 +1864,10 @@ void ARGBShuffleRow_Any_MSA(const uint8_t* src_ptr,
>                             uint8_t* dst_ptr,
>                             const uint8_t* param,
>                             int width);
>+void ARGBShuffleRow_Any_MMI(const uint8_t* src_ptr,
>+                            uint8_t* dst_ptr,
>+                            const uint8_t* param,
>+                            int width);
> 
> void RGB24ToARGBRow_SSSE3(const uint8_t* src_rgb24,
>                           uint8_t* dst_argb,
>@@ -1591,28 +1891,40 @@ void RGB24ToARGBRow_NEON(const uint8_t* src_rgb24,
>                          uint8_t* dst_argb,
>                          int width);
> void RGB24ToARGBRow_MSA(const uint8_t* src_rgb24, uint8_t* dst_argb, int width);
>+void RGB24ToARGBRow_MMI(const uint8_t* src_rgb24, uint8_t* dst_argb, int width);
> void RAWToARGBRow_NEON(const uint8_t* src_raw, uint8_t* dst_argb, int width);
> void RAWToARGBRow_MSA(const uint8_t* src_raw, uint8_t* dst_argb, int width);
>+void RAWToARGBRow_MMI(const uint8_t* src_raw, uint8_t* dst_argb, int width);
> void RAWToRGB24Row_NEON(const uint8_t* src_raw, uint8_t* dst_rgb24, int width);
> void RAWToRGB24Row_MSA(const uint8_t* src_raw, uint8_t* dst_rgb24, int width);
>+void RAWToRGB24Row_MMI(const uint8_t* src_raw, uint8_t* dst_rgb24, int width);
> void RGB565ToARGBRow_NEON(const uint8_t* src_rgb565,
>                           uint8_t* dst_argb,
>                           int width);
> void RGB565ToARGBRow_MSA(const uint8_t* src_rgb565,
>                          uint8_t* dst_argb,
>                          int width);
>+void RGB565ToARGBRow_MMI(const uint8_t* src_rgb565,
>+                         uint8_t* dst_argb,
>+                         int width);
> void ARGB1555ToARGBRow_NEON(const uint8_t* src_argb1555,
>                             uint8_t* dst_argb,
>                             int width);
> void ARGB1555ToARGBRow_MSA(const uint8_t* src_argb1555,
>                            uint8_t* dst_argb,
>                            int width);
>+void ARGB1555ToARGBRow_MMI(const uint8_t* src_argb1555,
>+                           uint8_t* dst_argb,
>+                           int width);
> void ARGB4444ToARGBRow_NEON(const uint8_t* src_argb4444,
>                             uint8_t* dst_argb,
>                             int width);
> void ARGB4444ToARGBRow_MSA(const uint8_t* src_argb4444,
>                            uint8_t* dst_argb,
>                            int width);
>+void ARGB4444ToARGBRow_MMI(const uint8_t* src_argb4444,
>+                           uint8_t* dst_argb,
>+                           int width);
> void RGB24ToARGBRow_C(const uint8_t* src_rgb24, uint8_t* dst_argb, int width);
> void RAWToARGBRow_C(const uint8_t* src_raw, uint8_t* dst_argb, int width);
> void RAWToRGB24Row_C(const uint8_t* src_raw, uint8_t* dst_rgb24, int width);
>@@ -1663,24 +1975,35 @@ void RGB24ToARGBRow_Any_NEON(const uint8_t* src_ptr,
> void RGB24ToARGBRow_Any_MSA(const uint8_t* src_ptr,
>                             uint8_t* dst_ptr,
>                             int width);
>+void RGB24ToARGBRow_Any_MMI(const uint8_t* src_ptr,
>+                            uint8_t* dst_ptr,
>+                            int width);
> void RAWToARGBRow_Any_NEON(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
> void RAWToARGBRow_Any_MSA(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
>+void RAWToARGBRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
> void RAWToRGB24Row_Any_NEON(const uint8_t* src_ptr,
>                             uint8_t* dst_ptr,
>                             int width);
> void RAWToRGB24Row_Any_MSA(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
>+void RAWToRGB24Row_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
> void RGB565ToARGBRow_Any_NEON(const uint8_t* src_ptr,
>                               uint8_t* dst_ptr,
>                               int width);
> void RGB565ToARGBRow_Any_MSA(const uint8_t* src_ptr,
>                              uint8_t* dst_ptr,
>                              int width);
>+void RGB565ToARGBRow_Any_MMI(const uint8_t* src_ptr,
>+                             uint8_t* dst_ptr,
>+                             int width);
> void ARGB1555ToARGBRow_Any_NEON(const uint8_t* src_ptr,
>                                 uint8_t* dst_ptr,
>                                 int width);
> void ARGB1555ToARGBRow_Any_MSA(const uint8_t* src_ptr,
>                                uint8_t* dst_ptr,
>                                int width);
>+void ARGB1555ToARGBRow_Any_MMI(const uint8_t* src_ptr,
>+                               uint8_t* dst_ptr,
>+                               int width);
> void ARGB4444ToARGBRow_Any_NEON(const uint8_t* src_ptr,
>                                 uint8_t* dst_ptr,
>                                 int width);
>@@ -1688,6 +2011,9 @@ void ARGB4444ToARGBRow_Any_NEON(const uint8_t* src_ptr,
> void ARGB4444ToARGBRow_Any_MSA(const uint8_t* src_ptr,
>                                uint8_t* dst_ptr,
>                                int width);
>+void ARGB4444ToARGBRow_Any_MMI(const uint8_t* src_ptr,
>+                               uint8_t* dst_ptr,
>+                               int width);
> 
> void ARGBToRGB24Row_SSSE3(const uint8_t* src, uint8_t* dst, int width);
> void ARGBToRAWRow_SSSE3(const uint8_t* src, uint8_t* dst, int width);
>@@ -1697,6 +2023,11 @@ void ARGBToARGB4444Row_SSE2(const uint8_t* src, uint8_t* dst, int width);
> void ABGRToAR30Row_SSSE3(const uint8_t* src, uint8_t* dst, int width);
> void ARGBToAR30Row_SSSE3(const uint8_t* src, uint8_t* dst, int width);
> 
>+void ARGBToRAWRow_AVX2(const uint8_t* src, uint8_t* dst, int width);
>+void ARGBToRGB24Row_AVX2(const uint8_t* src, uint8_t* dst, int width);
>+
>+void ARGBToRGB24Row_AVX512VBMI(const uint8_t* src, uint8_t* dst, int width);
>+
> void ARGBToRGB565DitherRow_C(const uint8_t* src_argb,
>                              uint8_t* dst_rgb,
>                              const uint32_t dither4,
>@@ -1751,6 +2082,20 @@ void ARGBToRGB565DitherRow_MSA(const uint8_t* src_argb,
>                                const uint32_t dither4,
>                                int width);
> 
>+void ARGBToRGB24Row_MMI(const uint8_t* src_argb, uint8_t* dst_rgb, int width);
>+void ARGBToRAWRow_MMI(const uint8_t* src_argb, uint8_t* dst_rgb, int width);
>+void ARGBToRGB565Row_MMI(const uint8_t* src_argb, uint8_t* dst_rgb, int width);
>+void ARGBToARGB1555Row_MMI(const uint8_t* src_argb,
>+                           uint8_t* dst_rgb,
>+                           int width);
>+void ARGBToARGB4444Row_MMI(const uint8_t* src_argb,
>+                           uint8_t* dst_rgb,
>+                           int width);
>+void ARGBToRGB565DitherRow_MMI(const uint8_t* src_argb,
>+                               uint8_t* dst_rgb,
>+                               const uint32_t dither4,
>+                               int width);
>+
> void ARGBToRGBARow_C(const uint8_t* src_argb, uint8_t* dst_rgb, int width);
> void ARGBToRGB24Row_C(const uint8_t* src_argb, uint8_t* dst_rgb, int width);
> void ARGBToRAWRow_C(const uint8_t* src_argb, uint8_t* dst_rgb, int width);
>@@ -1764,6 +2109,7 @@ void J400ToARGBRow_SSE2(const uint8_t* src_y, uint8_t* dst_argb, int width);
> void J400ToARGBRow_AVX2(const uint8_t* src_y, uint8_t* dst_argb, int width);
> void J400ToARGBRow_NEON(const uint8_t* src_y, uint8_t* dst_argb, int width);
> void J400ToARGBRow_MSA(const uint8_t* src_y, uint8_t* dst_argb, int width);
>+void J400ToARGBRow_MMI(const uint8_t* src_y, uint8_t* dst_argb, int width);
> void J400ToARGBRow_C(const uint8_t* src_y, uint8_t* dst_argb, int width);
> void J400ToARGBRow_Any_SSE2(const uint8_t* src_ptr,
>                             uint8_t* dst_ptr,
>@@ -1775,6 +2121,7 @@ void J400ToARGBRow_Any_NEON(const uint8_t* src_ptr,
>                             uint8_t* dst_ptr,
>                             int width);
> void J400ToARGBRow_Any_MSA(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
>+void J400ToARGBRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
> 
> void I444ToARGBRow_C(const uint8_t* src_y,
>                      const uint8_t* src_u,
>@@ -1979,11 +2326,31 @@ void NV12ToARGBRow_AVX2(const uint8_t* y_buf,
>                         uint8_t* dst_argb,
>                         const struct YuvConstants* yuvconstants,
>                         int width);
>+void NV12ToRGB24Row_SSSE3(const uint8_t* src_y,
>+                          const uint8_t* src_uv,
>+                          uint8_t* dst_rgb24,
>+                          const struct YuvConstants* yuvconstants,
>+                          int width);
>+void NV21ToRGB24Row_SSSE3(const uint8_t* src_y,
>+                          const uint8_t* src_vu,
>+                          uint8_t* dst_rgb24,
>+                          const struct YuvConstants* yuvconstants,
>+                          int width);
> void NV12ToRGB565Row_SSSE3(const uint8_t* src_y,
>                            const uint8_t* src_uv,
>                            uint8_t* dst_rgb565,
>                            const struct YuvConstants* yuvconstants,
>                            int width);
>+void NV12ToRGB24Row_AVX2(const uint8_t* src_y,
>+                         const uint8_t* src_uv,
>+                         uint8_t* dst_rgb24,
>+                         const struct YuvConstants* yuvconstants,
>+                         int width);
>+void NV21ToRGB24Row_AVX2(const uint8_t* src_y,
>+                         const uint8_t* src_vu,
>+                         uint8_t* dst_rgb24,
>+                         const struct YuvConstants* yuvconstants,
>+                         int width);
> void NV12ToRGB565Row_AVX2(const uint8_t* src_y,
>                           const uint8_t* src_uv,
>                           uint8_t* dst_rgb565,
>@@ -2169,6 +2536,26 @@ void NV21ToARGBRow_Any_AVX2(const uint8_t* y_buf,
>                             uint8_t* dst_ptr,
>                             const struct YuvConstants* yuvconstants,
>                             int width);
>+void NV12ToRGB24Row_Any_SSSE3(const uint8_t* y_buf,
>+                              const uint8_t* uv_buf,
>+                              uint8_t* dst_ptr,
>+                              const struct YuvConstants* yuvconstants,
>+                              int width);
>+void NV21ToRGB24Row_Any_SSSE3(const uint8_t* y_buf,
>+                              const uint8_t* uv_buf,
>+                              uint8_t* dst_ptr,
>+                              const struct YuvConstants* yuvconstants,
>+                              int width);
>+void NV12ToRGB24Row_Any_AVX2(const uint8_t* y_buf,
>+                             const uint8_t* uv_buf,
>+                             uint8_t* dst_ptr,
>+                             const struct YuvConstants* yuvconstants,
>+                             int width);
>+void NV21ToRGB24Row_Any_AVX2(const uint8_t* y_buf,
>+                             const uint8_t* uv_buf,
>+                             uint8_t* dst_ptr,
>+                             const struct YuvConstants* yuvconstants,
>+                             int width);
> void NV12ToRGB565Row_Any_SSSE3(const uint8_t* y_buf,
>                                const uint8_t* uv_buf,
>                                uint8_t* dst_ptr,
>@@ -2255,6 +2642,7 @@ void I400ToARGBRow_SSE2(const uint8_t* y_buf, uint8_t* dst_argb, int width);
> void I400ToARGBRow_AVX2(const uint8_t* y_buf, uint8_t* dst_argb, int width);
> void I400ToARGBRow_NEON(const uint8_t* src_y, uint8_t* dst_argb, int width);
> void I400ToARGBRow_MSA(const uint8_t* src_y, uint8_t* dst_argb, int width);
>+void I400ToARGBRow_MMI(const uint8_t* src_y, uint8_t* dst_argb, int width);
> void I400ToARGBRow_Any_SSE2(const uint8_t* src_ptr,
>                             uint8_t* dst_ptr,
>                             int width);
>@@ -2265,6 +2653,7 @@ void I400ToARGBRow_Any_NEON(const uint8_t* src_ptr,
>                             uint8_t* dst_ptr,
>                             int width);
> void I400ToARGBRow_Any_MSA(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
>+void I400ToARGBRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
> 
> // ARGB preattenuated alpha blend.
> void ARGBBlendRow_SSSE3(const uint8_t* src_argb0,
>@@ -2279,6 +2668,10 @@ void ARGBBlendRow_MSA(const uint8_t* src_argb0,
>                       const uint8_t* src_argb1,
>                       uint8_t* dst_argb,
>                       int width);
>+void ARGBBlendRow_MMI(const uint8_t* src_argb0,
>+                      const uint8_t* src_argb1,
>+                      uint8_t* dst_argb,
>+                      int width);
> void ARGBBlendRow_C(const uint8_t* src_argb0,
>                     const uint8_t* src_argb1,
>                     uint8_t* dst_argb,
>@@ -2305,6 +2698,16 @@ void BlendPlaneRow_Any_AVX2(const uint8_t* y_buf,
>                             const uint8_t* v_buf,
>                             uint8_t* dst_ptr,
>                             int width);
>+void BlendPlaneRow_MMI(const uint8_t* src0,
>+                       const uint8_t* src1,
>+                       const uint8_t* alpha,
>+                       uint8_t* dst,
>+                       int width);
>+void BlendPlaneRow_Any_MMI(const uint8_t* y_buf,
>+                           const uint8_t* u_buf,
>+                           const uint8_t* v_buf,
>+                           uint8_t* dst_ptr,
>+                           int width);
> void BlendPlaneRow_C(const uint8_t* src0,
>                      const uint8_t* src1,
>                      const uint8_t* alpha,
>@@ -2349,6 +2752,14 @@ void ARGBMultiplyRow_Any_MSA(const uint8_t* y_buf,
>                              const uint8_t* uv_buf,
>                              uint8_t* dst_ptr,
>                              int width);
>+void ARGBMultiplyRow_MMI(const uint8_t* src_argb0,
>+                         const uint8_t* src_argb1,
>+                         uint8_t* dst_argb,
>+                         int width);
>+void ARGBMultiplyRow_Any_MMI(const uint8_t* y_buf,
>+                             const uint8_t* uv_buf,
>+                             uint8_t* dst_ptr,
>+                             int width);
> 
> // ARGB add images.
> void ARGBAddRow_C(const uint8_t* src_argb0,
>@@ -2387,6 +2798,14 @@ void ARGBAddRow_Any_MSA(const uint8_t* y_buf,
>                         const uint8_t* uv_buf,
>                         uint8_t* dst_ptr,
>                         int width);
>+void ARGBAddRow_MMI(const uint8_t* src_argb0,
>+                    const uint8_t* src_argb1,
>+                    uint8_t* dst_argb,
>+                    int width);
>+void ARGBAddRow_Any_MMI(const uint8_t* y_buf,
>+                        const uint8_t* uv_buf,
>+                        uint8_t* dst_ptr,
>+                        int width);
> 
> // ARGB subtract images. Same API as Blend, but these require
> // pointer and width alignment for SSE2.
>@@ -2426,6 +2845,14 @@ void ARGBSubtractRow_Any_MSA(const uint8_t* y_buf,
>                              const uint8_t* uv_buf,
>                              uint8_t* dst_ptr,
>                              int width);
>+void ARGBSubtractRow_MMI(const uint8_t* src_argb0,
>+                         const uint8_t* src_argb1,
>+                         uint8_t* dst_argb,
>+                         int width);
>+void ARGBSubtractRow_Any_MMI(const uint8_t* y_buf,
>+                             const uint8_t* uv_buf,
>+                             uint8_t* dst_ptr,
>+                             int width);
> 
> void ARGBToRGB24Row_Any_SSSE3(const uint8_t* src_ptr,
>                               uint8_t* dst_ptr,
>@@ -2448,7 +2875,13 @@ void ABGRToAR30Row_Any_SSSE3(const uint8_t* src_ptr,
> void ARGBToAR30Row_Any_SSSE3(const uint8_t* src_ptr,
>                              uint8_t* dst_ptr,
>                              int width);
>-
>+void ARGBToRAWRow_Any_AVX2(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
>+void ARGBToRGB24Row_Any_AVX2(const uint8_t* src_ptr,
>+                             uint8_t* dst_ptr,
>+                             int width);
>+void ARGBToRGB24Row_Any_AVX512VBMI(const uint8_t* src_ptr,
>+                                   uint8_t* dst_ptr,
>+                                   int width);
> void ARGBToRGB565DitherRow_Any_SSE2(const uint8_t* src_ptr,
>                                     uint8_t* dst_ptr,
>                                     const uint32_t param,
>@@ -2509,6 +2942,24 @@ void ARGBToRGB565DitherRow_Any_MSA(const uint8_t* src_ptr,
>                                    const uint32_t param,
>                                    int width);
> 
>+void ARGBToRGB24Row_Any_MMI(const uint8_t* src_ptr,
>+                            uint8_t* dst_ptr,
>+                            int width);
>+void ARGBToRAWRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
>+void ARGBToRGB565Row_Any_MMI(const uint8_t* src_ptr,
>+                             uint8_t* dst_ptr,
>+                             int width);
>+void ARGBToARGB1555Row_Any_MMI(const uint8_t* src_ptr,
>+                               uint8_t* dst_ptr,
>+                               int width);
>+void ARGBToARGB4444Row_Any_MMI(const uint8_t* src_ptr,
>+                               uint8_t* dst_ptr,
>+                               int width);
>+void ARGBToRGB565DitherRow_Any_MMI(const uint8_t* src_ptr,
>+                                   uint8_t* dst_ptr,
>+                                   const uint32_t param,
>+                                   int width);
>+
> void I444ToARGBRow_Any_NEON(const uint8_t* y_buf,
>                             const uint8_t* u_buf,
>                             const uint8_t* v_buf,
>@@ -2695,15 +3146,25 @@ void YUY2ToUV422Row_NEON(const uint8_t* src_yuy2,
>                          uint8_t* dst_v,
>                          int width);
> void YUY2ToYRow_MSA(const uint8_t* src_yuy2, uint8_t* dst_y, int width);
>+void YUY2ToYRow_MMI(const uint8_t* src_yuy2, uint8_t* dst_y, int width);
> void YUY2ToUVRow_MSA(const uint8_t* src_yuy2,
>                      int src_stride_yuy2,
>                      uint8_t* dst_u,
>                      uint8_t* dst_v,
>                      int width);
>+void YUY2ToUVRow_MMI(const uint8_t* src_yuy2,
>+                     int src_stride_yuy2,
>+                     uint8_t* dst_u,
>+                     uint8_t* dst_v,
>+                     int width);
> void YUY2ToUV422Row_MSA(const uint8_t* src_yuy2,
>                         uint8_t* dst_u,
>                         uint8_t* dst_v,
>                         int width);
>+void YUY2ToUV422Row_MMI(const uint8_t* src_yuy2,
>+                        uint8_t* dst_u,
>+                        uint8_t* dst_v,
>+                        int width);
> void YUY2ToYRow_C(const uint8_t* src_yuy2, uint8_t* dst_y, int width);
> void YUY2ToUVRow_C(const uint8_t* src_yuy2,
>                    int src_stride_yuy2,
>@@ -2745,15 +3206,25 @@ void YUY2ToUV422Row_Any_NEON(const uint8_t* src_ptr,
>                              uint8_t* dst_v,
>                              int width);
> void YUY2ToYRow_Any_MSA(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
>+void YUY2ToYRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
> void YUY2ToUVRow_Any_MSA(const uint8_t* src_ptr,
>                          int src_stride_ptr,
>                          uint8_t* dst_u,
>                          uint8_t* dst_v,
>                          int width);
>+void YUY2ToUVRow_Any_MMI(const uint8_t* src_ptr,
>+                         int src_stride_ptr,
>+                         uint8_t* dst_u,
>+                         uint8_t* dst_v,
>+                         int width);
> void YUY2ToUV422Row_Any_MSA(const uint8_t* src_ptr,
>                             uint8_t* dst_u,
>                             uint8_t* dst_v,
>                             int width);
>+void YUY2ToUV422Row_Any_MMI(const uint8_t* src_ptr,
>+                            uint8_t* dst_u,
>+                            uint8_t* dst_v,
>+                            int width);
> void UYVYToYRow_AVX2(const uint8_t* src_uyvy, uint8_t* dst_y, int width);
> void UYVYToUVRow_AVX2(const uint8_t* src_uyvy,
>                       int stride_uyvy,
>@@ -2795,15 +3266,25 @@ void UYVYToUV422Row_NEON(const uint8_t* src_uyvy,
>                          uint8_t* dst_v,
>                          int width);
> void UYVYToYRow_MSA(const uint8_t* src_uyvy, uint8_t* dst_y, int width);
>+void UYVYToYRow_MMI(const uint8_t* src_uyvy, uint8_t* dst_y, int width);
> void UYVYToUVRow_MSA(const uint8_t* src_uyvy,
>                      int src_stride_uyvy,
>                      uint8_t* dst_u,
>                      uint8_t* dst_v,
>                      int width);
>+void UYVYToUVRow_MMI(const uint8_t* src_uyvy,
>+                     int src_stride_uyvy,
>+                     uint8_t* dst_u,
>+                     uint8_t* dst_v,
>+                     int width);
> void UYVYToUV422Row_MSA(const uint8_t* src_uyvy,
>                         uint8_t* dst_u,
>                         uint8_t* dst_v,
>                         int width);
>+void UYVYToUV422Row_MMI(const uint8_t* src_uyvy,
>+                        uint8_t* dst_u,
>+                        uint8_t* dst_v,
>+                        int width);
> 
> void UYVYToYRow_C(const uint8_t* src_uyvy, uint8_t* dst_y, int width);
> void UYVYToUVRow_C(const uint8_t* src_uyvy,
>@@ -2846,15 +3327,25 @@ void UYVYToUV422Row_Any_NEON(const uint8_t* src_ptr,
>                              uint8_t* dst_v,
>                              int width);
> void UYVYToYRow_Any_MSA(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
>+void UYVYToYRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
> void UYVYToUVRow_Any_MSA(const uint8_t* src_ptr,
>                          int src_stride_ptr,
>                          uint8_t* dst_u,
>                          uint8_t* dst_v,
>                          int width);
>+void UYVYToUVRow_Any_MMI(const uint8_t* src_ptr,
>+                         int src_stride_ptr,
>+                         uint8_t* dst_u,
>+                         uint8_t* dst_v,
>+                         int width);
> void UYVYToUV422Row_Any_MSA(const uint8_t* src_ptr,
>                             uint8_t* dst_u,
>                             uint8_t* dst_v,
>                             int width);
>+void UYVYToUV422Row_Any_MMI(const uint8_t* src_ptr,
>+                            uint8_t* dst_u,
>+                            uint8_t* dst_v,
>+                            int width);
> 
> void I422ToYUY2Row_C(const uint8_t* src_y,
>                      const uint8_t* src_u,
>@@ -2931,21 +3422,41 @@ void I422ToYUY2Row_MSA(const uint8_t* src_y,
>                        const uint8_t* src_v,
>                        uint8_t* dst_yuy2,
>                        int width);
>+void I422ToYUY2Row_MMI(const uint8_t* src_y,
>+                       const uint8_t* src_u,
>+                       const uint8_t* src_v,
>+                       uint8_t* dst_yuy2,
>+                       int width);
> void I422ToUYVYRow_MSA(const uint8_t* src_y,
>                        const uint8_t* src_u,
>                        const uint8_t* src_v,
>                        uint8_t* dst_uyvy,
>                        int width);
>+void I422ToUYVYRow_MMI(const uint8_t* src_y,
>+                       const uint8_t* src_u,
>+                       const uint8_t* src_v,
>+                       uint8_t* dst_uyvy,
>+                       int width);
> void I422ToYUY2Row_Any_MSA(const uint8_t* y_buf,
>                            const uint8_t* u_buf,
>                            const uint8_t* v_buf,
>                            uint8_t* dst_ptr,
>                            int width);
>+void I422ToYUY2Row_Any_MMI(const uint8_t* y_buf,
>+                           const uint8_t* u_buf,
>+                           const uint8_t* v_buf,
>+                           uint8_t* dst_ptr,
>+                           int width);
> void I422ToUYVYRow_Any_MSA(const uint8_t* y_buf,
>                            const uint8_t* u_buf,
>                            const uint8_t* v_buf,
>                            uint8_t* dst_ptr,
>                            int width);
>+void I422ToUYVYRow_Any_MMI(const uint8_t* y_buf,
>+                           const uint8_t* u_buf,
>+                           const uint8_t* v_buf,
>+                           uint8_t* dst_ptr,
>+                           int width);
> 
> // Effects related row functions.
> void ARGBAttenuateRow_C(const uint8_t* src_argb, uint8_t* dst_argb, int width);
>@@ -2961,6 +3472,9 @@ void ARGBAttenuateRow_NEON(const uint8_t* src_argb,
> void ARGBAttenuateRow_MSA(const uint8_t* src_argb,
>                           uint8_t* dst_argb,
>                           int width);
>+void ARGBAttenuateRow_MMI(const uint8_t* src_argb,
>+                          uint8_t* dst_argb,
>+                          int width);
> void ARGBAttenuateRow_Any_SSSE3(const uint8_t* src_ptr,
>                                 uint8_t* dst_ptr,
>                                 int width);
>@@ -2973,6 +3487,9 @@ void ARGBAttenuateRow_Any_NEON(const uint8_t* src_ptr,
> void ARGBAttenuateRow_Any_MSA(const uint8_t* src_ptr,
>                               uint8_t* dst_ptr,
>                               int width);
>+void ARGBAttenuateRow_Any_MMI(const uint8_t* src_ptr,
>+                              uint8_t* dst_ptr,
>+                              int width);
> 
> // Inverse table for unattenuate, shared by C and SSE2.
> extern const uint32_t fixed_invtbl8[256];
>@@ -2996,11 +3513,13 @@ void ARGBGrayRow_C(const uint8_t* src_argb, uint8_t* dst_argb, int width);
> void ARGBGrayRow_SSSE3(const uint8_t* src_argb, uint8_t* dst_argb, int width);
> void ARGBGrayRow_NEON(const uint8_t* src_argb, uint8_t* dst_argb, int width);
> void ARGBGrayRow_MSA(const uint8_t* src_argb, uint8_t* dst_argb, int width);
>+void ARGBGrayRow_MMI(const uint8_t* src_argb, uint8_t* dst_argb, int width);
> 
> void ARGBSepiaRow_C(uint8_t* dst_argb, int width);
> void ARGBSepiaRow_SSSE3(uint8_t* dst_argb, int width);
> void ARGBSepiaRow_NEON(uint8_t* dst_argb, int width);
> void ARGBSepiaRow_MSA(uint8_t* dst_argb, int width);
>+void ARGBSepiaRow_MMI(uint8_t* dst_argb, int width);
> 
> void ARGBColorMatrixRow_C(const uint8_t* src_argb,
>                           uint8_t* dst_argb,
>@@ -3018,6 +3537,10 @@ void ARGBColorMatrixRow_MSA(const uint8_t* src_argb,
>                             uint8_t* dst_argb,
>                             const int8_t* matrix_argb,
>                             int width);
>+void ARGBColorMatrixRow_MMI(const uint8_t* src_argb,
>+                            uint8_t* dst_argb,
>+                            const int8_t* matrix_argb,
>+                            int width);
> 
> void ARGBColorTableRow_C(uint8_t* dst_argb,
>                          const uint8_t* table_argb,
>@@ -3070,6 +3593,10 @@ void ARGBShadeRow_MSA(const uint8_t* src_argb,
>                       uint8_t* dst_argb,
>                       int width,
>                       uint32_t value);
>+void ARGBShadeRow_MMI(const uint8_t* src_argb,
>+                      uint8_t* dst_argb,
>+                      int width,
>+                      uint32_t value);
> 
> // Used for blur.
> void CumulativeSumToAverageRow_SSE2(const int32_t* topleft,
>@@ -3083,6 +3610,11 @@ void ComputeCumulativeSumRow_SSE2(const uint8_t* row,
>                                   const int32_t* previous_cumsum,
>                                   int width);
> 
>+void ComputeCumulativeSumRow_MMI(const uint8_t* row,
>+                                 int32_t* cumsum,
>+                                 const int32_t* previous_cumsum,
>+                                 int width);
>+
> void CumulativeSumToAverageRow_C(const int32_t* tl,
>                                  const int32_t* bl,
>                                  int w,
>@@ -3133,6 +3665,11 @@ void InterpolateRow_MSA(uint8_t* dst_ptr,
>                         ptrdiff_t src_stride,
>                         int width,
>                         int source_y_fraction);
>+void InterpolateRow_MMI(uint8_t* dst_ptr,
>+                        const uint8_t* src_ptr,
>+                        ptrdiff_t src_stride,
>+                        int width,
>+                        int source_y_fraction);
> void InterpolateRow_Any_NEON(uint8_t* dst_ptr,
>                              const uint8_t* src_ptr,
>                              ptrdiff_t src_stride_ptr,
>@@ -3153,6 +3690,11 @@ void InterpolateRow_Any_MSA(uint8_t* dst_ptr,
>                             ptrdiff_t src_stride_ptr,
>                             int width,
>                             int source_y_fraction);
>+void InterpolateRow_Any_MMI(uint8_t* dst_ptr,
>+                            const uint8_t* src_ptr,
>+                            ptrdiff_t src_stride_ptr,
>+                            int width,
>+                            int source_y_fraction);
> 
> void InterpolateRow_16_C(uint16_t* dst_ptr,
>                          const uint16_t* src_ptr,
>@@ -3181,6 +3723,11 @@ void SobelXRow_MSA(const uint8_t* src_y0,
>                    const uint8_t* src_y2,
>                    uint8_t* dst_sobelx,
>                    int width);
>+void SobelXRow_MMI(const uint8_t* src_y0,
>+                   const uint8_t* src_y1,
>+                   const uint8_t* src_y2,
>+                   uint8_t* dst_sobelx,
>+                   int width);
> void SobelYRow_C(const uint8_t* src_y0,
>                  const uint8_t* src_y1,
>                  uint8_t* dst_sobely,
>@@ -3197,6 +3744,10 @@ void SobelYRow_MSA(const uint8_t* src_y0,
>                    const uint8_t* src_y1,
>                    uint8_t* dst_sobely,
>                    int width);
>+void SobelYRow_MMI(const uint8_t* src_y0,
>+                   const uint8_t* src_y1,
>+                   uint8_t* dst_sobely,
>+                   int width);
> void SobelRow_C(const uint8_t* src_sobelx,
>                 const uint8_t* src_sobely,
>                 uint8_t* dst_argb,
>@@ -3213,6 +3764,10 @@ void SobelRow_MSA(const uint8_t* src_sobelx,
>                   const uint8_t* src_sobely,
>                   uint8_t* dst_argb,
>                   int width);
>+void SobelRow_MMI(const uint8_t* src_sobelx,
>+                  const uint8_t* src_sobely,
>+                  uint8_t* dst_argb,
>+                  int width);
> void SobelToPlaneRow_C(const uint8_t* src_sobelx,
>                        const uint8_t* src_sobely,
>                        uint8_t* dst_y,
>@@ -3229,6 +3784,10 @@ void SobelToPlaneRow_MSA(const uint8_t* src_sobelx,
>                          const uint8_t* src_sobely,
>                          uint8_t* dst_y,
>                          int width);
>+void SobelToPlaneRow_MMI(const uint8_t* src_sobelx,
>+                         const uint8_t* src_sobely,
>+                         uint8_t* dst_y,
>+                         int width);
> void SobelXYRow_C(const uint8_t* src_sobelx,
>                   const uint8_t* src_sobely,
>                   uint8_t* dst_argb,
>@@ -3245,6 +3804,10 @@ void SobelXYRow_MSA(const uint8_t* src_sobelx,
>                     const uint8_t* src_sobely,
>                     uint8_t* dst_argb,
>                     int width);
>+void SobelXYRow_MMI(const uint8_t* src_sobelx,
>+                    const uint8_t* src_sobely,
>+                    uint8_t* dst_argb,
>+                    int width);
> void SobelRow_Any_SSE2(const uint8_t* y_buf,
>                        const uint8_t* uv_buf,
>                        uint8_t* dst_ptr,
>@@ -3257,6 +3820,10 @@ void SobelRow_Any_MSA(const uint8_t* y_buf,
>                       const uint8_t* uv_buf,
>                       uint8_t* dst_ptr,
>                       int width);
>+void SobelRow_Any_MMI(const uint8_t* y_buf,
>+                      const uint8_t* uv_buf,
>+                      uint8_t* dst_ptr,
>+                      int width);
> void SobelToPlaneRow_Any_SSE2(const uint8_t* y_buf,
>                               const uint8_t* uv_buf,
>                               uint8_t* dst_ptr,
>@@ -3269,6 +3836,10 @@ void SobelToPlaneRow_Any_MSA(const uint8_t* y_buf,
>                              const uint8_t* uv_buf,
>                              uint8_t* dst_ptr,
>                              int width);
>+void SobelToPlaneRow_Any_MMI(const uint8_t* y_buf,
>+                             const uint8_t* uv_buf,
>+                             uint8_t* dst_ptr,
>+                             int width);
> void SobelXYRow_Any_SSE2(const uint8_t* y_buf,
>                          const uint8_t* uv_buf,
>                          uint8_t* dst_ptr,
>@@ -3281,6 +3852,10 @@ void SobelXYRow_Any_MSA(const uint8_t* y_buf,
>                         const uint8_t* uv_buf,
>                         uint8_t* dst_ptr,
>                         int width);
>+void SobelXYRow_Any_MMI(const uint8_t* y_buf,
>+                        const uint8_t* uv_buf,
>+                        uint8_t* dst_ptr,
>+                        int width);
> 
> void ARGBPolynomialRow_C(const uint8_t* src_argb,
>                          uint8_t* dst_argb,
>@@ -3358,9 +3933,9 @@ void ByteToFloatRow_NEON(const uint8_t* src,
>                          float* dst,
>                          float scale,
>                          int width);
>-void ByteToFloatRow_Any_NEON(const uint8_t* src,
>-                             float* dst,
>-                             float scale,
>+void ByteToFloatRow_Any_NEON(const uint8_t* src_ptr,
>+                             float* dst_ptr,
>+                             float param,
>                              int width);
> 
> void ARGBLumaColorTableRow_C(const uint8_t* src_argb,
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/scale_row.h b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/scale_row.h
>index 7194ba09f842ba7ffb5b295b06977ddf91237e31..6e207a9c64e119d43a39e77043f994b96ff5ee33 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/scale_row.h
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/scale_row.h
>@@ -58,6 +58,7 @@ extern "C" {
>     (defined(_M_IX86) || defined(__x86_64__) || defined(__i386__))
> #define HAS_FIXEDDIV1_X86
> #define HAS_FIXEDDIV_X86
>+#define HAS_SCALEADDROW_SSE2
> #define HAS_SCALEARGBCOLS_SSE2
> #define HAS_SCALEARGBCOLSUP2_SSE2
> #define HAS_SCALEARGBFILTERCOLS_SSSE3
>@@ -69,7 +70,6 @@ extern "C" {
> #define HAS_SCALEROWDOWN34_SSSE3
> #define HAS_SCALEROWDOWN38_SSSE3
> #define HAS_SCALEROWDOWN4_SSSE3
>-#define HAS_SCALEADDROW_SSE2
> #endif
> 
> // The following are available on all x86 platforms, but
>@@ -86,7 +86,9 @@ extern "C" {
> // The following are available on Neon platforms:
> #if !defined(LIBYUV_DISABLE_NEON) && \
>     (defined(__ARM_NEON__) || defined(LIBYUV_NEON) || defined(__aarch64__))
>+#define HAS_SCALEADDROW_NEON
> #define HAS_SCALEARGBCOLS_NEON
>+#define HAS_SCALEARGBFILTERCOLS_NEON
> #define HAS_SCALEARGBROWDOWN2_NEON
> #define HAS_SCALEARGBROWDOWNEVEN_NEON
> #define HAS_SCALEFILTERCOLS_NEON
>@@ -94,7 +96,6 @@ extern "C" {
> #define HAS_SCALEROWDOWN34_NEON
> #define HAS_SCALEROWDOWN38_NEON
> #define HAS_SCALEROWDOWN4_NEON
>-#define HAS_SCALEARGBFILTERCOLS_NEON
> #endif
> 
> #if !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)
>@@ -110,6 +111,23 @@ extern "C" {
> #define HAS_SCALEROWDOWN4_MSA
> #endif
> 
>+#if !defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)
>+#define HAS_FIXEDDIV1_MIPS
>+#define HAS_FIXEDDIV_MIPS
>+#define HAS_SCALEADDROW_16_MMI
>+#define HAS_SCALEADDROW_MMI
>+#define HAS_SCALEARGBCOLS_MMI
>+#define HAS_SCALEARGBCOLSUP2_MMI
>+#define HAS_SCALEARGBROWDOWN2_MMI
>+#define HAS_SCALEARGBROWDOWNEVEN_MMI
>+#define HAS_SCALECOLS_16_MMI
>+#define HAS_SCALECOLS_MMI
>+#define HAS_SCALEROWDOWN2_16_MMI
>+#define HAS_SCALEROWDOWN2_MMI
>+#define HAS_SCALEROWDOWN4_16_MMI
>+#define HAS_SCALEROWDOWN4_MMI
>+#endif
>+
> // Scale ARGB vertically with bilinear interpolation.
> void ScalePlaneVertical(int src_height,
>                         int dst_width,
>@@ -147,12 +165,17 @@ enum FilterMode ScaleFilterReduce(int src_width,
> // Divide num by div and return as 16.16 fixed point result.
> int FixedDiv_C(int num, int div);
> int FixedDiv_X86(int num, int div);
>+int FixedDiv_MIPS(int num, int div);
> // Divide num - 1 by div - 1 and return as 16.16 fixed point result.
> int FixedDiv1_C(int num, int div);
> int FixedDiv1_X86(int num, int div);
>+int FixedDiv1_MIPS(int num, int div);
> #ifdef HAS_FIXEDDIV_X86
> #define FixedDiv FixedDiv_X86
> #define FixedDiv1 FixedDiv1_X86
>+#elif defined HAS_FIXEDDIV_MIPS
>+#define FixedDiv FixedDiv_MIPS
>+#define FixedDiv1 FixedDiv1_MIPS
> #else
> #define FixedDiv FixedDiv_C
> #define FixedDiv1 FixedDiv1_C
>@@ -569,6 +592,16 @@ void ScaleARGBCols_Any_MSA(uint8_t* dst_ptr,
>                            int dst_width,
>                            int x,
>                            int dx);
>+void ScaleARGBCols_MMI(uint8_t* dst_argb,
>+                       const uint8_t* src_argb,
>+                       int dst_width,
>+                       int x,
>+                       int dx);
>+void ScaleARGBCols_Any_MMI(uint8_t* dst_ptr,
>+                           const uint8_t* src_ptr,
>+                           int dst_width,
>+                           int x,
>+                           int dx);
> 
> // ARGB Row functions
> void ScaleARGBRowDown2_SSE2(const uint8_t* src_argb,
>@@ -607,6 +640,18 @@ void ScaleARGBRowDown2Box_MSA(const uint8_t* src_argb,
>                               ptrdiff_t src_stride,
>                               uint8_t* dst_argb,
>                               int dst_width);
>+void ScaleARGBRowDown2_MMI(const uint8_t* src_argb,
>+                           ptrdiff_t src_stride,
>+                           uint8_t* dst_argb,
>+                           int dst_width);
>+void ScaleARGBRowDown2Linear_MMI(const uint8_t* src_argb,
>+                                 ptrdiff_t src_stride,
>+                                 uint8_t* dst_argb,
>+                                 int dst_width);
>+void ScaleARGBRowDown2Box_MMI(const uint8_t* src_argb,
>+                              ptrdiff_t src_stride,
>+                              uint8_t* dst_argb,
>+                              int dst_width);
> void ScaleARGBRowDown2_Any_SSE2(const uint8_t* src_ptr,
>                                 ptrdiff_t src_stride,
>                                 uint8_t* dst_ptr,
>@@ -643,7 +688,18 @@ void ScaleARGBRowDown2Box_Any_MSA(const uint8_t* src_ptr,
>                                   ptrdiff_t src_stride,
>                                   uint8_t* dst_ptr,
>                                   int dst_width);
>-
>+void ScaleARGBRowDown2_Any_MMI(const uint8_t* src_ptr,
>+                               ptrdiff_t src_stride,
>+                               uint8_t* dst_ptr,
>+                               int dst_width);
>+void ScaleARGBRowDown2Linear_Any_MMI(const uint8_t* src_ptr,
>+                                     ptrdiff_t src_stride,
>+                                     uint8_t* dst_ptr,
>+                                     int dst_width);
>+void ScaleARGBRowDown2Box_Any_MMI(const uint8_t* src_ptr,
>+                                  ptrdiff_t src_stride,
>+                                  uint8_t* dst_ptr,
>+                                  int dst_width);
> void ScaleARGBRowDownEven_SSE2(const uint8_t* src_argb,
>                                ptrdiff_t src_stride,
>                                int src_stepx,
>@@ -674,6 +730,16 @@ void ScaleARGBRowDownEvenBox_MSA(const uint8_t* src_argb,
>                                  int src_stepx,
>                                  uint8_t* dst_argb,
>                                  int dst_width);
>+void ScaleARGBRowDownEven_MMI(const uint8_t* src_argb,
>+                              ptrdiff_t src_stride,
>+                              int32_t src_stepx,
>+                              uint8_t* dst_argb,
>+                              int dst_width);
>+void ScaleARGBRowDownEvenBox_MMI(const uint8_t* src_argb,
>+                                 ptrdiff_t src_stride,
>+                                 int src_stepx,
>+                                 uint8_t* dst_argb,
>+                                 int dst_width);
> void ScaleARGBRowDownEven_Any_SSE2(const uint8_t* src_ptr,
>                                    ptrdiff_t src_stride,
>                                    int src_stepx,
>@@ -704,6 +770,16 @@ void ScaleARGBRowDownEvenBox_Any_MSA(const uint8_t* src_ptr,
>                                      int src_stepx,
>                                      uint8_t* dst_ptr,
>                                      int dst_width);
>+void ScaleARGBRowDownEven_Any_MMI(const uint8_t* src_ptr,
>+                                  ptrdiff_t src_stride,
>+                                  int32_t src_stepx,
>+                                  uint8_t* dst_ptr,
>+                                  int dst_width);
>+void ScaleARGBRowDownEvenBox_Any_MMI(const uint8_t* src_ptr,
>+                                     ptrdiff_t src_stride,
>+                                     int src_stepx,
>+                                     uint8_t* dst_ptr,
>+                                     int dst_width);
> 
> // ScaleRowDown2Box also used by planar functions
> // NEON downscalers with interpolation.
>@@ -936,6 +1012,93 @@ void ScaleRowDown34_1_Box_Any_MSA(const uint8_t* src_ptr,
>                                   uint8_t* dst_ptr,
>                                   int dst_width);
> 
>+void ScaleRowDown2_MMI(const uint8_t* src_ptr,
>+                       ptrdiff_t src_stride,
>+                       uint8_t* dst,
>+                       int dst_width);
>+void ScaleRowDown2_16_MMI(const uint16_t* src_ptr,
>+                          ptrdiff_t src_stride,
>+                          uint16_t* dst,
>+                          int dst_width);
>+void ScaleRowDown2Linear_MMI(const uint8_t* src_ptr,
>+                             ptrdiff_t src_stride,
>+                             uint8_t* dst,
>+                             int dst_width);
>+void ScaleRowDown2Linear_16_MMI(const uint16_t* src_ptr,
>+                                ptrdiff_t src_stride,
>+                                uint16_t* dst,
>+                                int dst_width);
>+void ScaleRowDown2Box_MMI(const uint8_t* src_ptr,
>+                          ptrdiff_t src_stride,
>+                          uint8_t* dst,
>+                          int dst_width);
>+void ScaleRowDown2Box_16_MMI(const uint16_t* src_ptr,
>+                             ptrdiff_t src_stride,
>+                             uint16_t* dst,
>+                             int dst_width);
>+void ScaleRowDown2Box_Odd_MMI(const uint8_t* src_ptr,
>+                              ptrdiff_t src_stride,
>+                              uint8_t* dst,
>+                              int dst_width);
>+void ScaleRowDown4_MMI(const uint8_t* src_ptr,
>+                       ptrdiff_t src_stride,
>+                       uint8_t* dst,
>+                       int dst_width);
>+void ScaleRowDown4_16_MMI(const uint16_t* src_ptr,
>+                          ptrdiff_t src_stride,
>+                          uint16_t* dst,
>+                          int dst_width);
>+void ScaleRowDown4Box_MMI(const uint8_t* src_ptr,
>+                          ptrdiff_t src_stride,
>+                          uint8_t* dst,
>+                          int dst_width);
>+void ScaleRowDown4Box_16_MMI(const uint16_t* src_ptr,
>+                             ptrdiff_t src_stride,
>+                             uint16_t* dst,
>+                             int dst_width);
>+void ScaleAddRow_MMI(const uint8_t* src_ptr, uint16_t* dst_ptr, int src_width);
>+void ScaleAddRow_16_MMI(const uint16_t* src_ptr,
>+                        uint32_t* dst_ptr,
>+                        int src_width);
>+void ScaleColsUp2_MMI(uint8_t* dst_ptr,
>+                      const uint8_t* src_ptr,
>+                      int dst_width,
>+                      int x,
>+                      int dx);
>+void ScaleColsUp2_16_MMI(uint16_t* dst_ptr,
>+                         const uint16_t* src_ptr,
>+                         int dst_width,
>+                         int x,
>+                         int dx);
>+void ScaleARGBColsUp2_MMI(uint8_t* dst_argb,
>+                          const uint8_t* src_argb,
>+                          int dst_width,
>+                          int x,
>+                          int dx);
>+
>+void ScaleRowDown2_Any_MMI(const uint8_t* src_ptr,
>+                           ptrdiff_t src_stride,
>+                           uint8_t* dst_ptr,
>+                           int dst_width);
>+void ScaleRowDown2Linear_Any_MMI(const uint8_t* src_ptr,
>+                                 ptrdiff_t src_stride,
>+                                 uint8_t* dst_ptr,
>+                                 int dst_width);
>+void ScaleRowDown2Box_Any_MMI(const uint8_t* src_ptr,
>+                              ptrdiff_t src_stride,
>+                              uint8_t* dst_ptr,
>+                              int dst_width);
>+void ScaleRowDown4_Any_MMI(const uint8_t* src_ptr,
>+                           ptrdiff_t src_stride,
>+                           uint8_t* dst_ptr,
>+                           int dst_width);
>+void ScaleRowDown4Box_Any_MMI(const uint8_t* src_ptr,
>+                              ptrdiff_t src_stride,
>+                              uint8_t* dst_ptr,
>+                              int dst_width);
>+void ScaleAddRow_Any_MMI(const uint8_t* src_ptr,
>+                         uint16_t* dst_ptr,
>+                         int src_width);
> #ifdef __cplusplus
> }  // extern "C"
> }  // namespace libyuv
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/version.h b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/version.h
>index 21522cf351c8c53610ec2e8430ddc0a42cd58772..1a38ba7d6dd41d9e9c30650587f764aa4878b3c4 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/version.h
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/version.h
>@@ -11,6 +11,6 @@
> #ifndef INCLUDE_LIBYUV_VERSION_H_
> #define INCLUDE_LIBYUV_VERSION_H_
> 
>-#define LIBYUV_VERSION 1703
>+#define LIBYUV_VERSION 1722
> 
> #endif  // INCLUDE_LIBYUV_VERSION_H_
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/video_common.h b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/video_common.h
>index bcef378b5a4e5b2c3af9f364e153c7753671dd31..ffcbdbf1b0c1455079931e4796e111a497275a94 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/video_common.h
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/include/libyuv/video_common.h
>@@ -30,7 +30,8 @@ extern "C" {
> #ifdef __cplusplus
> #define FOURCC(a, b, c, d)                                        \
>   ((static_cast<uint32_t>(a)) | (static_cast<uint32_t>(b) << 8) | \
>-   (static_cast<uint32_t>(c) << 16) | (static_cast<uint32_t>(d) << 24))
>+   (static_cast<uint32_t>(c) << 16) | /* NOLINT */                \
>+   (static_cast<uint32_t>(d) << 24))  /* NOLINT */
> #else
> #define FOURCC(a, b, c, d)                                     \
>   (((uint32_t)(a)) | ((uint32_t)(b) << 8) |       /* NOLINT */ \
>@@ -79,7 +80,7 @@ enum FourCC {
>   // 1 Primary Compressed YUV format.
>   FOURCC_MJPG = FOURCC('M', 'J', 'P', 'G'),
> 
>-  // 7 Auxiliary YUV variations: 3 with U and V planes are swapped, 1 Alias.
>+  // 8 Auxiliary YUV variations: 3 with U and V planes are swapped, 1 Alias.
>   FOURCC_YV12 = FOURCC('Y', 'V', '1', '2'),
>   FOURCC_YV16 = FOURCC('Y', 'V', '1', '6'),
>   FOURCC_YV24 = FOURCC('Y', 'V', '2', '4'),
>@@ -87,6 +88,7 @@ enum FourCC {
>   FOURCC_J420 = FOURCC('J', '4', '2', '0'),
>   FOURCC_J400 = FOURCC('J', '4', '0', '0'),  // unofficial fourcc
>   FOURCC_H420 = FOURCC('H', '4', '2', '0'),  // unofficial fourcc
>+  FOURCC_H422 = FOURCC('H', '4', '2', '2'),  // unofficial fourcc
> 
>   // 14 Auxiliary aliases.  CanonicalFourCC() maps these to canonical fourcc.
>   FOURCC_IYUV = FOURCC('I', 'Y', 'U', 'V'),  // Alias for I420.
>@@ -155,6 +157,7 @@ enum FourCCBpp {
>   FOURCC_BPP_J420 = 12,
>   FOURCC_BPP_J400 = 8,
>   FOURCC_BPP_H420 = 12,
>+  FOURCC_BPP_H422 = 16,
>   FOURCC_BPP_H010 = 24,
>   FOURCC_BPP_MJPG = 0,  // 0 means unknown.
>   FOURCC_BPP_H264 = 0,
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/infra/config/PRESUBMIT.py b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/infra/config/PRESUBMIT.py
>new file mode 100644
>index 0000000000000000000000000000000000000000..89eaa5192c2b3333031b3580dfd0eab59e96328f
>--- /dev/null
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/infra/config/PRESUBMIT.py
>@@ -0,0 +1,15 @@
>+# Copyright 2018 The LibYuv Project Authors. All rights reserved.
>+#
>+# Use of this source code is governed by a BSD-style license
>+# that can be found in the LICENSE file in the root of the source
>+# tree. An additional intellectual property rights grant can be found
>+# in the file PATENTS. All contributing project authors may
>+# be found in the AUTHORS file in the root of the source tree.
>+
>+
>+def CheckChangeOnUpload(input_api, output_api):
>+  return input_api.canned_checks.CheckChangedLUCIConfigs(input_api, output_api)
>+
>+
>+def CheckChangeOnCommit(input_api, output_api):
>+  return input_api.canned_checks.CheckChangedLUCIConfigs(input_api, output_api)
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/infra/config/cq.cfg b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/infra/config/cq.cfg
>index 604de7814d9420614a3b372d377282dd14113412..7bcc05955a38d9bceaf8bce5fb37658dac0e211e 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/infra/config/cq.cfg
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/infra/config/cq.cfg
>@@ -2,7 +2,6 @@
> # at http://luci-config.appspot.com/schemas/projects/refs:cq.cfg.
> 
> version: 1
>-cq_name: "libyuv"
> cq_status_url: "https://chromium-cq-status.appspot.com"
> git_repo_url: "https://chromium.googlesource.com/libyuv/libyuv.git"
> 
>@@ -16,7 +15,7 @@ verifiers {
> 
>   try_job {
>     buckets {
>-      name: "master.tryserver.libyuv"
>+      name: "luci.libyuv.try"
>       builders { name: "win" }
>       builders { name: "win_rel" }
>       builders { name: "win_x64_rel" }
>@@ -47,10 +46,6 @@ verifiers {
>       builders { name: "android_arm64" }
>       builders { name: "android_x86" }
>       builders { name: "android_x64" }
>-      builders {
>-        name: "android_mips"
>-        experiment_percentage: 100
>-      }
>     }
>   }
> }
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/libyuv.gni b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/libyuv.gni
>index 89e4d38232767e84c5fa76498a1edf21acfa6087..8df40ba2d77b3b5cf2b6fd27bd69200e93db5390 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/libyuv.gni
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/libyuv.gni
>@@ -13,8 +13,11 @@ import("//build/config/mips.gni")
> declare_args() {
>   libyuv_include_tests = !build_with_chromium
>   libyuv_disable_jpeg = false
>-  libyuv_use_neon = (current_cpu == "arm64" ||
>-      (current_cpu == "arm" && (arm_use_neon || arm_optionally_use_neon)))
>-  libyuv_use_msa = (current_cpu == "mips64el" || current_cpu == "mipsel") &&
>-    mips_use_msa
>+  libyuv_use_neon =
>+      current_cpu == "arm64" ||
>+      (current_cpu == "arm" && (arm_use_neon || arm_optionally_use_neon))
>+  libyuv_use_msa =
>+      (current_cpu == "mips64el" || current_cpu == "mipsel") && mips_use_msa
>+  libyuv_use_mmi =
>+      (current_cpu == "mips64el" || current_cpu == "mipsel") && mips_use_mmi
> }
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/libyuv.gyp b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/libyuv.gyp
>deleted file mode 100644
>index e853ba3197a94bcad91011ac6384ce84daa1db10..0000000000000000000000000000000000000000
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/libyuv.gyp
>+++ /dev/null
>@@ -1,161 +0,0 @@
>-# Copyright 2011 The LibYuv Project Authors. All rights reserved.
>-#
>-# Use of this source code is governed by a BSD-style license
>-# that can be found in the LICENSE file in the root of the source
>-# tree. An additional intellectual property rights grant can be found
>-# in the file PATENTS. All contributing project authors may
>-# be found in the AUTHORS file in the root of the source tree.
>-
>-{
>-  'includes': [
>-    'libyuv.gypi',
>-  ],
>-  # Make sure that if we are being compiled to an xcodeproj, nothing tries to
>-  # include a .pch.
>-  'xcode_settings': {
>-    'GCC_PREFIX_HEADER': '',
>-    'GCC_PRECOMPILE_PREFIX_HEADER': 'NO',
>-  },
>-  'variables': {
>-    'use_system_libjpeg%': 0,
>-    # Can be enabled if your jpeg has GYP support.
>-    'libyuv_disable_jpeg%': 1,
>-    # 'chromium_code' treats libyuv as internal and increases warning level.
>-    'chromium_code': 1,
>-    # clang compiler default variable usable by other apps that include libyuv.
>-    'clang%': 0,
>-    # Link-Time Optimizations.
>-    'use_lto%': 0,
>-    'mips_msa%': 0,  # Default to msa off.
>-    'build_neon': 0,
>-    'build_msa': 0,
>-    'conditions': [
>-       ['(target_arch == "armv7" or target_arch == "armv7s" or \
>-       (target_arch == "arm" and arm_version >= 7) or target_arch == "arm64")\
>-       and (arm_neon == 1 or arm_neon_optional == 1)', {
>-         'build_neon': 1,
>-       }],
>-       ['(target_arch == "mipsel" or target_arch == "mips64el")\
>-       and (mips_msa == 1)',
>-       {
>-         'build_msa': 1,
>-       }],
>-    ],
>-  },
>-
>-  'targets': [
>-    {
>-      'target_name': 'libyuv',
>-      # Change type to 'shared_library' to build .so or .dll files.
>-      'type': 'static_library',
>-      'variables': {
>-        'optimize': 'max',  # enable O2 and ltcg.
>-      },
>-      # Allows libyuv.a redistributable library without external dependencies.
>-      'standalone_static_library': 1,
>-      'conditions': [
>-       # Disable -Wunused-parameter
>-        ['clang == 1', {
>-          'cflags': [
>-            '-Wno-unused-parameter',
>-         ],
>-        }],
>-        ['build_neon != 0', {
>-          'defines': [
>-            'LIBYUV_NEON',
>-          ],
>-          'cflags!': [
>-            '-mfpu=vfp',
>-            '-mfpu=vfpv3',
>-            '-mfpu=vfpv3-d16',
>-            # '-mthumb',  # arm32 not thumb
>-          ],
>-          'conditions': [
>-            # Disable LTO in libyuv_neon target due to gcc 4.9 compiler bug.
>-            ['clang == 0 and use_lto == 1', {
>-              'cflags!': [
>-                '-flto',
>-                '-ffat-lto-objects',
>-              ],
>-            }],
>-            # arm64 does not need -mfpu=neon option as neon is not optional
>-            ['target_arch != "arm64"', {
>-              'cflags': [
>-                '-mfpu=neon',
>-                # '-marm',  # arm32 not thumb
>-              ],
>-            }],
>-          ],
>-        }],
>-        ['build_msa != 0', {
>-          'defines': [
>-            'LIBYUV_MSA',
>-          ],
>-        }],
>-        ['OS != "ios" and libyuv_disable_jpeg != 1', {
>-          'defines': [
>-            'HAVE_JPEG'
>-          ],
>-          'conditions': [
>-            # Caveat system jpeg support may not support motion jpeg
>-            [ 'use_system_libjpeg == 1', {
>-              'dependencies': [
>-                 '<(DEPTH)/third_party/libjpeg/libjpeg.gyp:libjpeg',
>-              ],
>-            }, {
>-              'dependencies': [
>-                 '<(DEPTH)/third_party/libjpeg_turbo/libjpeg.gyp:libjpeg',
>-              ],
>-            }],
>-            [ 'use_system_libjpeg == 1', {
>-              'link_settings': {
>-                'libraries': [
>-                  '-ljpeg',
>-                ],
>-              }
>-            }],
>-          ],
>-        }],
>-      ], #conditions
>-      'defines': [
>-        # Enable the following 3 macros to turn off assembly for specified CPU.
>-        # 'LIBYUV_DISABLE_X86',
>-        # 'LIBYUV_DISABLE_NEON',
>-        # Enable the following macro to build libyuv as a shared library (dll).
>-        # 'LIBYUV_USING_SHARED_LIBRARY',
>-        # TODO(fbarchard): Make these into gyp defines.
>-      ],
>-      'include_dirs': [
>-        'include',
>-        '.',
>-      ],
>-      'direct_dependent_settings': {
>-        'include_dirs': [
>-          'include',
>-          '.',
>-        ],
>-        'conditions': [
>-          ['OS == "android" and target_arch == "arm64"', {
>-            'ldflags': [
>-              '-Wl,--dynamic-linker,/system/bin/linker64',
>-            ],
>-          }],
>-          ['OS == "android" and target_arch != "arm64"', {
>-            'ldflags': [
>-              '-Wl,--dynamic-linker,/system/bin/linker',
>-            ],
>-          }],
>-        ], #conditions
>-      },
>-      'sources': [
>-        '<@(libyuv_sources)',
>-      ],
>-    },
>-  ], # targets.
>-}
>-
>-# Local Variables:
>-# tab-width:2
>-# indent-tabs-mode:nil
>-# End:
>-# vim: set expandtab tabstop=2 shiftwidth=2:
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/libyuv.gypi b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/libyuv.gypi
>deleted file mode 100644
>index 9467adfc515d7e45e5fda706a32b565ddfe492ed..0000000000000000000000000000000000000000
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/libyuv.gypi
>+++ /dev/null
>@@ -1,81 +0,0 @@
>-# Copyright 2014 The LibYuv Project Authors. All rights reserved.
>-#
>-# Use of this source code is governed by a BSD-style license
>-# that can be found in the LICENSE file in the root of the source
>-# tree. An additional intellectual property rights grant can be found
>-# in the file PATENTS. All contributing project authors may
>-# be found in the AUTHORS file in the root of the source tree.
>-
>-{
>-  'variables': {
>-    'libyuv_sources': [
>-      # includes.
>-      'include/libyuv.h',
>-      'include/libyuv/basic_types.h',
>-      'include/libyuv/compare.h',
>-      'include/libyuv/convert.h',
>-      'include/libyuv/convert_argb.h',
>-      'include/libyuv/convert_from.h',
>-      'include/libyuv/convert_from_argb.h',
>-      'include/libyuv/cpu_id.h',
>-      'include/libyuv/macros_msa.h',
>-      'include/libyuv/mjpeg_decoder.h',
>-      'include/libyuv/planar_functions.h',
>-      'include/libyuv/rotate.h',
>-      'include/libyuv/rotate_argb.h',
>-      'include/libyuv/rotate_row.h',
>-      'include/libyuv/row.h',
>-      'include/libyuv/scale.h',
>-      'include/libyuv/scale_argb.h',
>-      'include/libyuv/scale_row.h',
>-      'include/libyuv/version.h',
>-      'include/libyuv/video_common.h',
>-
>-      # sources.
>-      'source/compare.cc',
>-      'source/compare_common.cc',
>-      'source/compare_gcc.cc',
>-      'source/compare_msa.cc',
>-      'source/compare_neon.cc',
>-      'source/compare_neon64.cc',
>-      'source/compare_win.cc',
>-      'source/convert.cc',
>-      'source/convert_argb.cc',
>-      'source/convert_from.cc',
>-      'source/convert_from_argb.cc',
>-      'source/convert_jpeg.cc',
>-      'source/convert_to_argb.cc',
>-      'source/convert_to_i420.cc',
>-      'source/cpu_id.cc',
>-      'source/mjpeg_decoder.cc',
>-      'source/mjpeg_validate.cc',
>-      'source/planar_functions.cc',
>-      'source/rotate.cc',
>-      'source/rotate_any.cc',
>-      'source/rotate_argb.cc',
>-      'source/rotate_common.cc',
>-      'source/rotate_gcc.cc',
>-      'source/rotate_msa.cc',
>-      'source/rotate_neon.cc',
>-      'source/rotate_neon64.cc',
>-      'source/rotate_win.cc',
>-      'source/row_any.cc',
>-      'source/row_common.cc',
>-      'source/row_gcc.cc',
>-      'source/row_msa.cc',
>-      'source/row_neon.cc',
>-      'source/row_neon64.cc',
>-      'source/row_win.cc',
>-      'source/scale.cc',
>-      'source/scale_any.cc',
>-      'source/scale_argb.cc',
>-      'source/scale_common.cc',
>-      'source/scale_gcc.cc',
>-      'source/scale_msa.cc',
>-      'source/scale_neon.cc',
>-      'source/scale_neon64.cc',
>-      'source/scale_win.cc',
>-      'source/video_common.cc',
>-    ],
>-  }
>-}
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/libyuv_nacl.gyp b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/libyuv_nacl.gyp
>deleted file mode 100644
>index b8fe57ee3e2116267fa5a6db47f5864662ae4430..0000000000000000000000000000000000000000
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/libyuv_nacl.gyp
>+++ /dev/null
>@@ -1,37 +0,0 @@
>-# Copyright 2014 The LibYuv Project Authors. All rights reserved.
>-#
>-# Use of this source code is governed by a BSD-style license
>-# that can be found in the LICENSE file in the root of the source
>-# tree. An additional intellectual property rights grant can be found
>-# in the file PATENTS. All contributing project authors may
>-# be found in the AUTHORS file in the root of the source tree.
>-
>-{
>-  'includes': [
>-    'libyuv.gypi',
>-    '../../native_client/build/untrusted.gypi',
>-  ],
>-  'targets': [
>-    {
>-      'target_name': 'libyuv_nacl',
>-      'type': 'none',
>-      'variables': {
>-        'nlib_target': 'libyuv_nacl.a',
>-        'build_glibc': 0,
>-        'build_newlib': 0,
>-        'build_pnacl_newlib': 1,
>-      },
>-      'include_dirs': [
>-        'include',
>-      ],
>-      'direct_dependent_settings': {
>-        'include_dirs': [
>-          'include',
>-        ],
>-      },
>-      'sources': [
>-        '<@(libyuv_sources)',
>-      ],
>-    },  # target libyuv_nacl
>-  ]
>-}
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/libyuv_test.gyp b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/libyuv_test.gyp
>deleted file mode 100644
>index 5fe154c610a70ef9acdf0f40076d807d86856094..0000000000000000000000000000000000000000
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/libyuv_test.gyp
>+++ /dev/null
>@@ -1,203 +0,0 @@
>-# Copyright 2011 The LibYuv Project Authors. All rights reserved.
>-#
>-# Use of this source code is governed by a BSD-style license
>-# that can be found in the LICENSE file in the root of the source
>-# tree. An additional intellectual property rights grant can be found
>-# in the file PATENTS. All contributing project authors may
>-# be found in the AUTHORS file in the root of the source tree.
>-
>-{
>-  'variables': {
>-    # Can be enabled if your jpeg has GYP support.
>-    'libyuv_disable_jpeg%': 1,
>-    'mips_msa%': 0,  # Default to msa off.
>-  },
>-  'targets': [
>-    {
>-      'target_name': 'libyuv_unittest',
>-      'type': '<(gtest_target_type)',
>-      'dependencies': [
>-        'libyuv.gyp:libyuv',
>-        'testing/gtest.gyp:gtest',
>-        'third_party/gflags/gflags.gyp:gflags',
>-      ],
>-      'direct_dependent_settings': {
>-        'defines': [
>-          'GTEST_RELATIVE_PATH',
>-        ],
>-      },
>-      'export_dependent_settings': [
>-        '<(DEPTH)/testing/gtest.gyp:gtest',
>-      ],
>-      'sources': [
>-        # headers
>-        'unit_test/unit_test.h',
>-
>-        # sources
>-        'unit_test/basictypes_test.cc',
>-        'unit_test/compare_test.cc',
>-        'unit_test/color_test.cc',
>-        'unit_test/convert_test.cc',
>-        'unit_test/cpu_test.cc',
>-        'unit_test/cpu_thread_test.cc',
>-        'unit_test/math_test.cc',
>-        'unit_test/planar_test.cc',
>-        'unit_test/rotate_argb_test.cc',
>-        'unit_test/rotate_test.cc',
>-        'unit_test/scale_argb_test.cc',
>-        'unit_test/scale_test.cc',
>-        'unit_test/unit_test.cc',
>-        'unit_test/video_common_test.cc',
>-      ],
>-      'conditions': [
>-        ['OS=="linux"', {
>-          'cflags': [
>-            '-fexceptions',
>-          ],
>-        }],
>-        [ 'OS == "ios"', {
>-          'xcode_settings': {
>-            'DEBUGGING_SYMBOLS': 'YES',
>-            'DEBUG_INFORMATION_FORMAT' : 'dwarf-with-dsym',
>-            # Work around compile issue with isosim.mm, see
>-            # https://code.google.com/p/libyuv/issues/detail?id=548 for details.
>-            'WARNING_CFLAGS': [
>-              '-Wno-sometimes-uninitialized',
>-            ],
>-          },
>-          'cflags': [
>-            '-Wno-sometimes-uninitialized',
>-          ],
>-        }],
>-        [ 'OS != "ios" and libyuv_disable_jpeg != 1', {
>-          'defines': [
>-            'HAVE_JPEG',
>-          ],
>-        }],
>-        ['OS=="android"', {
>-          'dependencies': [
>-            '<(DEPTH)/testing/android/native_test.gyp:native_test_native_code',
>-          ],
>-        }],
>-        # TODO(YangZhang): These lines can be removed when high accuracy
>-        # YUV to RGB to Neon is ported.
>-        [ '(target_arch == "armv7" or target_arch == "armv7s" \
>-          or (target_arch == "arm" and arm_version >= 7) \
>-          or target_arch == "arm64") \
>-          and (arm_neon == 1 or arm_neon_optional == 1)', {
>-          'defines': [
>-            'LIBYUV_NEON'
>-          ],
>-        }],
>-        [ '(target_arch == "mipsel" or target_arch == "mips64el") \
>-          and (mips_msa == 1)', {
>-          'defines': [
>-            'LIBYUV_MSA'
>-          ],
>-        }],
>-      ], # conditions
>-      'defines': [
>-        # Enable the following 3 macros to turn off assembly for specified CPU.
>-        # 'LIBYUV_DISABLE_X86',
>-        # 'LIBYUV_DISABLE_NEON',
>-        # Enable the following macro to build libyuv as a shared library (dll).
>-        # 'LIBYUV_USING_SHARED_LIBRARY',
>-      ],
>-    },
>-    {
>-      'target_name': 'compare',
>-      'type': 'executable',
>-      'dependencies': [
>-        'libyuv.gyp:libyuv',
>-      ],
>-      'sources': [
>-        # sources
>-        'util/compare.cc',
>-      ],
>-      'conditions': [
>-        ['OS=="linux"', {
>-          'cflags': [
>-            '-fexceptions',
>-          ],
>-        }],
>-      ], # conditions
>-    },
>-    {
>-      'target_name': 'yuvconvert',
>-      'type': 'executable',
>-      'dependencies': [
>-        'libyuv.gyp:libyuv',
>-      ],
>-      'sources': [
>-        # sources
>-        'util/yuvconvert.cc',
>-      ],
>-      'conditions': [
>-        ['OS=="linux"', {
>-          'cflags': [
>-            '-fexceptions',
>-          ],
>-        }],
>-      ], # conditions
>-    },
>-    # TODO(fbarchard): Enable SSE2 and OpenMP for better performance.
>-    {
>-      'target_name': 'psnr',
>-      'type': 'executable',
>-      'sources': [
>-        # sources
>-        'util/psnr_main.cc',
>-        'util/psnr.cc',
>-        'util/ssim.cc',
>-      ],
>-      'dependencies': [
>-        'libyuv.gyp:libyuv',
>-      ],
>-      'conditions': [
>-        [ 'OS != "ios" and libyuv_disable_jpeg != 1', {
>-          'defines': [
>-            'HAVE_JPEG',
>-          ],
>-        }],
>-      ], # conditions
>-    },
>-
>-    {
>-      'target_name': 'cpuid',
>-      'type': 'executable',
>-      'sources': [
>-        # sources
>-        'util/cpuid.c',
>-      ],
>-      'dependencies': [
>-        'libyuv.gyp:libyuv',
>-      ],
>-    },
>-  ], # targets
>-  'conditions': [
>-    ['OS=="android"', {
>-      'targets': [
>-        {
>-          'target_name': 'yuv_unittest_apk',
>-          'type': 'none',
>-          'variables': {
>-            'test_suite_name': 'yuv_unittest',
>-            'input_shlib_path': '<(SHARED_LIB_DIR)/(SHARED_LIB_PREFIX)libyuv_unittest<(SHARED_LIB_SUFFIX)',
>-          },
>-          'includes': [
>-            'build/apk_test.gypi',
>-          ],
>-          'dependencies': [
>-            'libyuv_unittest',
>-          ],
>-        },
>-      ],
>-    }],
>-  ],
>-}
>-
>-# Local Variables:
>-# tab-width:2
>-# indent-tabs-mode:nil
>-# End:
>-# vim: set expandtab tabstop=2 shiftwidth=2:
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/linux.mk b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/linux.mk
>index b84c89f912e066416f2a2477ab720a6a477cb66f..e9a26a79b209564b3f4e6e80ac14f1dd83768ae1 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/linux.mk
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/linux.mk
>@@ -13,6 +13,8 @@ LOCAL_OBJ_FILES := \
> 	source/compare.o           \
> 	source/compare_common.o    \
> 	source/compare_gcc.o       \
>+	source/compare_mmi.o       \
>+	source/compare_msa.o       \
> 	source/compare_neon64.o    \
> 	source/compare_neon.o      \
> 	source/compare_win.o       \
>@@ -32,12 +34,16 @@ LOCAL_OBJ_FILES := \
> 	source/rotate.o            \
> 	source/rotate_common.o     \
> 	source/rotate_gcc.o        \
>+	source/rotate_mmi.o        \
>+	source/rotate_msa.o        \
> 	source/rotate_neon64.o     \
> 	source/rotate_neon.o       \
> 	source/rotate_win.o        \
> 	source/row_any.o           \
> 	source/row_common.o        \
> 	source/row_gcc.o           \
>+	source/row_mmi.o           \
>+	source/row_msa.o           \
> 	source/row_neon64.o        \
> 	source/row_neon.o          \
> 	source/row_win.o           \
>@@ -46,6 +52,8 @@ LOCAL_OBJ_FILES := \
> 	source/scale.o             \
> 	source/scale_common.o      \
> 	source/scale_gcc.o         \
>+	source/scale_mmi.o         \
>+	source/scale_msa.o         \
> 	source/scale_neon64.o      \
> 	source/scale_neon.o        \
> 	source/scale_win.o         \
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/public.mk b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/public.mk
>index 090d8cb659fa8198ec0ec95cd083c42d7afa2db5..1342307a56e30142ee3128f819d9eabfab9cb8cc 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/public.mk
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/public.mk
>@@ -1,13 +1,13 @@
>-# This file contains all the common make variables which are useful for
>-# anyone depending on this library.
>-# Note that dependencies on NDK are not directly listed since NDK auto adds
>-# them.
>-
>-LIBYUV_INCLUDES := $(LIBYUV_PATH)/include
>-
>-LIBYUV_C_FLAGS :=
>-
>-LIBYUV_CPP_FLAGS :=
>-
>-LIBYUV_LDLIBS :=
>-LIBYUV_DEP_MODULES :=
>+# This file contains all the common make variables which are useful for
>+# anyone depending on this library.
>+# Note that dependencies on NDK are not directly listed since NDK auto adds
>+# them.
>+
>+LIBYUV_INCLUDES := $(LIBYUV_PATH)/include
>+
>+LIBYUV_C_FLAGS :=
>+
>+LIBYUV_CPP_FLAGS :=
>+
>+LIBYUV_LDLIBS :=
>+LIBYUV_DEP_MODULES :=
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/compare.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/compare.cc
>index 50e3abd0556e3862a7da070e4555daa2ed4bf45c..5aa3a4db86da6421f5ecbb1142d25fb5b0352b87 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/compare.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/compare.cc
>@@ -154,6 +154,12 @@ uint64_t ComputeHammingDistance(const uint8_t* src_a,
>     HammingDistance = HammingDistance_MSA;
>   }
> #endif
>+#if defined(HAS_HAMMINGDISTANCE_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    HammingDistance = HammingDistance_MMI;
>+  }
>+#endif
>+
> #ifdef _OPENMP
> #pragma omp parallel for reduction(+ : diff)
> #endif
>@@ -210,6 +216,11 @@ uint64_t ComputeSumSquareError(const uint8_t* src_a,
>     SumSquareError = SumSquareError_MSA;
>   }
> #endif
>+#if defined(HAS_SUMSQUAREERROR_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    SumSquareError = SumSquareError_MMI;
>+  }
>+#endif
> #ifdef _OPENMP
> #pragma omp parallel for reduction(+ : sse)
> #endif
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/compare_common.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/compare_common.cc
>index 633466addb53e48b4ae839187c6b59cdd23c8eb6..d4b170ad9862b5efe05b7a9fcedca5b305fb6b97 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/compare_common.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/compare_common.cc
>@@ -55,7 +55,7 @@ uint32_t HammingDistance_C(const uint8_t* src_a,
> 
>   int i;
>   for (i = 0; i < count - 3; i += 4) {
>-    uint32_t x = *((uint32_t*)src_a) ^ *((uint32_t*)src_b);  // NOLINT
>+    uint32_t x = *((const uint32_t*)src_a) ^ *((const uint32_t*)src_b);
>     uint32_t u = x - ((x >> 1) & 0x55555555);
>     u = ((u >> 2) & 0x33333333) + (u & 0x33333333);
>     diff += ((((u + (u >> 4)) & 0x0f0f0f0f) * 0x01010101) >> 24);
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/compare_mmi.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/compare_mmi.cc
>new file mode 100644
>index 0000000000000000000000000000000000000000..7640d9468cb8986c120cc96ad3b3c83bd3ce3678
>--- /dev/null
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/compare_mmi.cc
>@@ -0,0 +1,123 @@
>+/*
>+ *  Copyright 2012 The LibYuv Project Authors. All rights reserved.
>+ *
>+ *  Use of this source code is governed by a BSD-style license
>+ *  that can be found in the LICENSE file in the root of the source
>+ *  tree. An additional intellectual property rights grant can be found
>+ *  in the file PATENTS. All contributing project authors may
>+ *  be found in the AUTHORS file in the root of the source tree.
>+ */
>+
>+#include "libyuv/basic_types.h"
>+
>+#include "libyuv/compare_row.h"
>+
>+#ifdef __cplusplus
>+namespace libyuv {
>+extern "C" {
>+#endif
>+
>+// This module is for Mips MMI.
>+#if !defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)
>+
>+// Hakmem method for hamming distance.
>+uint32_t HammingDistance_MMI(const uint8_t* src_a,
>+                             const uint8_t* src_b,
>+                             int count) {
>+  uint32_t diff = 0u;
>+
>+  uint64_t temp = 0, temp1 = 0, ta = 0, tb = 0;
>+  uint64_t c1 = 0x5555555555555555;
>+  uint64_t c2 = 0x3333333333333333;
>+  uint64_t c3 = 0x0f0f0f0f0f0f0f0f;
>+  uint32_t c4 = 0x01010101;
>+  uint64_t s1 = 1, s2 = 2, s3 = 4;
>+  __asm__ volatile(
>+      "1:	\n\t"
>+      "ldc1   %[ta],    0(%[src_a])          \n\t"
>+      "ldc1   %[tb],    0(%[src_b])          \n\t"
>+      "xor    %[temp],  %[ta],      %[tb]    \n\t"
>+      "psrlw  %[temp1], %[temp],    %[s1]    \n\t"  // temp1=x>>1
>+      "and    %[temp1], %[temp1],   %[c1]    \n\t"  // temp1&=c1
>+      "psubw  %[temp1], %[temp],    %[temp1] \n\t"  // x-temp1
>+      "and    %[temp],  %[temp1],   %[c2]    \n\t"  // t = (u&c2)
>+      "psrlw  %[temp1], %[temp1],   %[s2]    \n\t"  // u>>2
>+      "and    %[temp1], %[temp1],   %[c2]    \n\t"  // u>>2 & c2
>+      "paddw  %[temp1], %[temp1],   %[temp]  \n\t"  // t1 = t1+t
>+      "psrlw  %[temp],  %[temp1],   %[s3]    \n\t"  // u>>4
>+      "paddw  %[temp1], %[temp1],   %[temp]  \n\t"  // u+(u>>4)
>+      "and    %[temp1], %[temp1],   %[c3]    \n\t"  //&c3
>+      "dmfc1  $t0,      %[temp1]             \n\t"
>+      "dsrl32 $t0,      $t0,        0        \n\t "
>+      "mul    $t0,      $t0,        %[c4]    \n\t"
>+      "dsrl   $t0,      $t0,        24       \n\t"
>+      "dadd   %[diff],  %[diff],    $t0      \n\t"
>+      "dmfc1  $t0,      %[temp1]             \n\t"
>+      "mul    $t0,      $t0,        %[c4]    \n\t"
>+      "dsrl   $t0,      $t0,        24       \n\t"
>+      "dadd   %[diff],  %[diff],    $t0      \n\t"
>+      "daddiu %[src_a], %[src_a],   8        \n\t"
>+      "daddiu %[src_b], %[src_b],   8        \n\t"
>+      "addiu  %[count], %[count],  -8        \n\t"
>+      "bgtz   %[count], 1b \n\t"
>+      "nop                            \n\t"
>+      : [diff] "+r"(diff), [src_a] "+r"(src_a), [src_b] "+r"(src_b),
>+        [count] "+r"(count), [ta] "+f"(ta), [tb] "+f"(tb), [temp] "+f"(temp),
>+        [temp1] "+f"(temp1)
>+      : [c1] "f"(c1), [c2] "f"(c2), [c3] "f"(c3), [c4] "r"(c4), [s1] "f"(s1),
>+        [s2] "f"(s2), [s3] "f"(s3)
>+      : "memory");
>+  return diff;
>+}
>+
>+uint32_t SumSquareError_MMI(const uint8_t* src_a,
>+                            const uint8_t* src_b,
>+                            int count) {
>+  uint32_t sse = 0u;
>+  uint32_t sse_hi = 0u, sse_lo = 0u;
>+
>+  uint64_t src1, src2;
>+  uint64_t diff, diff_hi, diff_lo;
>+  uint64_t sse_sum, sse_tmp;
>+
>+  const uint64_t mask = 0x0ULL;
>+
>+  __asm__ volatile(
>+      "xor        %[sse_sum],      %[sse_sum],        %[sse_sum]    \n\t"
>+
>+      "1:                                                           \n\t"
>+      "ldc1       %[src1],         0x00(%[src_a])                   \n\t"
>+      "ldc1       %[src2],         0x00(%[src_b])                   \n\t"
>+      "pasubub    %[diff],         %[src1],           %[src2]       \n\t"
>+      "punpcklbh  %[diff_lo],      %[diff],           %[mask]       \n\t"
>+      "punpckhbh  %[diff_hi],      %[diff],           %[mask]       \n\t"
>+      "pmaddhw    %[sse_tmp],      %[diff_lo],        %[diff_lo]    \n\t"
>+      "paddw      %[sse_sum],      %[sse_sum],        %[sse_tmp]    \n\t"
>+      "pmaddhw    %[sse_tmp],      %[diff_hi],        %[diff_hi]    \n\t"
>+      "paddw      %[sse_sum],      %[sse_sum],        %[sse_tmp]    \n\t"
>+
>+      "daddiu     %[src_a],        %[src_a],          0x08          \n\t"
>+      "daddiu     %[src_b],        %[src_b],          0x08          \n\t"
>+      "daddiu     %[count],        %[count],         -0x08          \n\t"
>+      "bnez       %[count],        1b                               \n\t"
>+
>+      "mfc1       %[sse_lo],       %[sse_sum]                       \n\t"
>+      "mfhc1      %[sse_hi],       %[sse_sum]                       \n\t"
>+      "daddu      %[sse],          %[sse_hi],         %[sse_lo]     \n\t"
>+      : [sse] "+&r"(sse), [diff] "=&f"(diff), [src1] "=&f"(src1),
>+        [src2] "=&f"(src2), [diff_lo] "=&f"(diff_lo), [diff_hi] "=&f"(diff_hi),
>+        [sse_sum] "=&f"(sse_sum), [sse_tmp] "=&f"(sse_tmp),
>+        [sse_hi] "+&r"(sse_hi), [sse_lo] "+&r"(sse_lo)
>+      : [src_a] "r"(src_a), [src_b] "r"(src_b), [count] "r"(count),
>+        [mask] "f"(mask)
>+      : "memory");
>+
>+  return sse;
>+}
>+
>+#endif  // !defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)
>+
>+#ifdef __cplusplus
>+}  // extern "C"
>+}  // namespace libyuv
>+#endif
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/compare_msa.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/compare_msa.cc
>index e944235f02060baaae70f29da4c427d549dd810c..0b807d37bee353735e70746731c86b57dbf0e739 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/compare_msa.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/compare_msa.cc
>@@ -1,97 +1,97 @@
>-/*
>- *  Copyright 2017 The LibYuv Project Authors. All rights reserved.
>- *
>- *  Use of this source code is governed by a BSD-style license
>- *  that can be found in the LICENSE file in the root of the source
>- *  tree. An additional intellectual property rights grant can be found
>- *  in the file PATENTS. All contributing project authors may
>- *  be found in the AUTHORS file in the root of the source tree.
>- */
>-
>-#include "libyuv/basic_types.h"
>-
>-#include "libyuv/compare_row.h"
>-#include "libyuv/row.h"
>-
>-// This module is for GCC MSA
>-#if !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)
>-#include "libyuv/macros_msa.h"
>-
>-#ifdef __cplusplus
>-namespace libyuv {
>-extern "C" {
>-#endif
>-
>-uint32_t HammingDistance_MSA(const uint8_t* src_a,
>-                             const uint8_t* src_b,
>-                             int count) {
>-  uint32_t diff = 0u;
>-  int i;
>-  v16u8 src0, src1, src2, src3;
>-  v2i64 vec0 = {0}, vec1 = {0};
>-
>-  for (i = 0; i < count; i += 32) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_a, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_a, 16);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)src_b, 0);
>-    src3 = (v16u8)__msa_ld_b((v16i8*)src_b, 16);
>-    src0 ^= src2;
>-    src1 ^= src3;
>-    vec0 += __msa_pcnt_d((v2i64)src0);
>-    vec1 += __msa_pcnt_d((v2i64)src1);
>-    src_a += 32;
>-    src_b += 32;
>-  }
>-
>-  vec0 += vec1;
>-  diff = (uint32_t)__msa_copy_u_w((v4i32)vec0, 0);
>-  diff += (uint32_t)__msa_copy_u_w((v4i32)vec0, 2);
>-  return diff;
>-}
>-
>-uint32_t SumSquareError_MSA(const uint8_t* src_a,
>-                            const uint8_t* src_b,
>-                            int count) {
>-  uint32_t sse = 0u;
>-  int i;
>-  v16u8 src0, src1, src2, src3;
>-  v8i16 vec0, vec1, vec2, vec3;
>-  v4i32 reg0 = {0}, reg1 = {0}, reg2 = {0}, reg3 = {0};
>-  v2i64 tmp0;
>-
>-  for (i = 0; i < count; i += 32) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_a, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_a, 16);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)src_b, 0);
>-    src3 = (v16u8)__msa_ld_b((v16i8*)src_b, 16);
>-    vec0 = (v8i16)__msa_ilvr_b((v16i8)src2, (v16i8)src0);
>-    vec1 = (v8i16)__msa_ilvl_b((v16i8)src2, (v16i8)src0);
>-    vec2 = (v8i16)__msa_ilvr_b((v16i8)src3, (v16i8)src1);
>-    vec3 = (v8i16)__msa_ilvl_b((v16i8)src3, (v16i8)src1);
>-    vec0 = __msa_hsub_u_h((v16u8)vec0, (v16u8)vec0);
>-    vec1 = __msa_hsub_u_h((v16u8)vec1, (v16u8)vec1);
>-    vec2 = __msa_hsub_u_h((v16u8)vec2, (v16u8)vec2);
>-    vec3 = __msa_hsub_u_h((v16u8)vec3, (v16u8)vec3);
>-    reg0 = __msa_dpadd_s_w(reg0, vec0, vec0);
>-    reg1 = __msa_dpadd_s_w(reg1, vec1, vec1);
>-    reg2 = __msa_dpadd_s_w(reg2, vec2, vec2);
>-    reg3 = __msa_dpadd_s_w(reg3, vec3, vec3);
>-    src_a += 32;
>-    src_b += 32;
>-  }
>-
>-  reg0 += reg1;
>-  reg2 += reg3;
>-  reg0 += reg2;
>-  tmp0 = __msa_hadd_s_d(reg0, reg0);
>-  sse = (uint32_t)__msa_copy_u_w((v4i32)tmp0, 0);
>-  sse += (uint32_t)__msa_copy_u_w((v4i32)tmp0, 2);
>-  return sse;
>-}
>-
>-#ifdef __cplusplus
>-}  // extern "C"
>-}  // namespace libyuv
>-#endif
>-
>-#endif  // !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)
>+/*
>+ *  Copyright 2017 The LibYuv Project Authors. All rights reserved.
>+ *
>+ *  Use of this source code is governed by a BSD-style license
>+ *  that can be found in the LICENSE file in the root of the source
>+ *  tree. An additional intellectual property rights grant can be found
>+ *  in the file PATENTS. All contributing project authors may
>+ *  be found in the AUTHORS file in the root of the source tree.
>+ */
>+
>+#include "libyuv/basic_types.h"
>+
>+#include "libyuv/compare_row.h"
>+#include "libyuv/row.h"
>+
>+// This module is for GCC MSA
>+#if !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)
>+#include "libyuv/macros_msa.h"
>+
>+#ifdef __cplusplus
>+namespace libyuv {
>+extern "C" {
>+#endif
>+
>+uint32_t HammingDistance_MSA(const uint8_t* src_a,
>+                             const uint8_t* src_b,
>+                             int count) {
>+  uint32_t diff = 0u;
>+  int i;
>+  v16u8 src0, src1, src2, src3;
>+  v2i64 vec0 = {0}, vec1 = {0};
>+
>+  for (i = 0; i < count; i += 32) {
>+    src0 = (v16u8)__msa_ld_b((v16i8*)src_a, 0);
>+    src1 = (v16u8)__msa_ld_b((v16i8*)src_a, 16);
>+    src2 = (v16u8)__msa_ld_b((v16i8*)src_b, 0);
>+    src3 = (v16u8)__msa_ld_b((v16i8*)src_b, 16);
>+    src0 ^= src2;
>+    src1 ^= src3;
>+    vec0 += __msa_pcnt_d((v2i64)src0);
>+    vec1 += __msa_pcnt_d((v2i64)src1);
>+    src_a += 32;
>+    src_b += 32;
>+  }
>+
>+  vec0 += vec1;
>+  diff = (uint32_t)__msa_copy_u_w((v4i32)vec0, 0);
>+  diff += (uint32_t)__msa_copy_u_w((v4i32)vec0, 2);
>+  return diff;
>+}
>+
>+uint32_t SumSquareError_MSA(const uint8_t* src_a,
>+                            const uint8_t* src_b,
>+                            int count) {
>+  uint32_t sse = 0u;
>+  int i;
>+  v16u8 src0, src1, src2, src3;
>+  v8i16 vec0, vec1, vec2, vec3;
>+  v4i32 reg0 = {0}, reg1 = {0}, reg2 = {0}, reg3 = {0};
>+  v2i64 tmp0;
>+
>+  for (i = 0; i < count; i += 32) {
>+    src0 = (v16u8)__msa_ld_b((v16i8*)src_a, 0);
>+    src1 = (v16u8)__msa_ld_b((v16i8*)src_a, 16);
>+    src2 = (v16u8)__msa_ld_b((v16i8*)src_b, 0);
>+    src3 = (v16u8)__msa_ld_b((v16i8*)src_b, 16);
>+    vec0 = (v8i16)__msa_ilvr_b((v16i8)src2, (v16i8)src0);
>+    vec1 = (v8i16)__msa_ilvl_b((v16i8)src2, (v16i8)src0);
>+    vec2 = (v8i16)__msa_ilvr_b((v16i8)src3, (v16i8)src1);
>+    vec3 = (v8i16)__msa_ilvl_b((v16i8)src3, (v16i8)src1);
>+    vec0 = __msa_hsub_u_h((v16u8)vec0, (v16u8)vec0);
>+    vec1 = __msa_hsub_u_h((v16u8)vec1, (v16u8)vec1);
>+    vec2 = __msa_hsub_u_h((v16u8)vec2, (v16u8)vec2);
>+    vec3 = __msa_hsub_u_h((v16u8)vec3, (v16u8)vec3);
>+    reg0 = __msa_dpadd_s_w(reg0, vec0, vec0);
>+    reg1 = __msa_dpadd_s_w(reg1, vec1, vec1);
>+    reg2 = __msa_dpadd_s_w(reg2, vec2, vec2);
>+    reg3 = __msa_dpadd_s_w(reg3, vec3, vec3);
>+    src_a += 32;
>+    src_b += 32;
>+  }
>+
>+  reg0 += reg1;
>+  reg2 += reg3;
>+  reg0 += reg2;
>+  tmp0 = __msa_hadd_s_d(reg0, reg0);
>+  sse = (uint32_t)__msa_copy_u_w((v4i32)tmp0, 0);
>+  sse += (uint32_t)__msa_copy_u_w((v4i32)tmp0, 2);
>+  return sse;
>+}
>+
>+#ifdef __cplusplus
>+}  // extern "C"
>+}  // namespace libyuv
>+#endif
>+
>+#endif  // !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert.cc
>index 375cc732c1d710ebd8ccbc44ac1d4ff4a8a81e2f..4b8d0dc57f7d4106e39f5a0114036a10ae2bad0f 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert.cc
>@@ -215,6 +215,195 @@ int I422ToI420(const uint8_t* src_y,
>                     dst_v, dst_stride_v, width, height, src_uv_width, height);
> }
> 
>+// TODO(fbarchard): Implement row conversion.
>+LIBYUV_API
>+int I422ToNV21(const uint8_t* src_y,
>+               int src_stride_y,
>+               const uint8_t* src_u,
>+               int src_stride_u,
>+               const uint8_t* src_v,
>+               int src_stride_v,
>+               uint8_t* dst_y,
>+               int dst_stride_y,
>+               uint8_t* dst_vu,
>+               int dst_stride_vu,
>+               int width,
>+               int height) {
>+  int halfwidth = (width + 1) >> 1;
>+  int halfheight = (height + 1) >> 1;
>+  // Negative height means invert the image.
>+  if (height < 0) {
>+    height = -height;
>+    halfheight = (height + 1) >> 1;
>+    src_y = src_y + (height - 1) * src_stride_y;
>+    src_u = src_u + (height - 1) * src_stride_u;
>+    src_v = src_v + (height - 1) * src_stride_v;
>+    src_stride_y = -src_stride_y;
>+    src_stride_u = -src_stride_u;
>+    src_stride_v = -src_stride_v;
>+  }
>+
>+  // Allocate u and v buffers
>+  align_buffer_64(plane_u, halfwidth * halfheight * 2);
>+  uint8_t* plane_v = plane_u + halfwidth * halfheight;
>+
>+  I422ToI420(src_y, src_stride_y, src_u, src_stride_u, src_v, src_stride_v,
>+             dst_y, dst_stride_y, plane_u, halfwidth, plane_v, halfwidth, width,
>+             height);
>+  MergeUVPlane(plane_v, halfwidth, plane_u, halfwidth, dst_vu, dst_stride_vu,
>+               halfwidth, halfheight);
>+  free_aligned_buffer_64(plane_u);
>+  return 0;
>+}
>+
>+#ifdef I422TONV21_ROW_VERSION
>+// Unittest fails for this version.
>+// 422 chroma is 1/2 width, 1x height
>+// 420 chroma is 1/2 width, 1/2 height
>+// Swap src_u and src_v to implement I422ToNV12
>+LIBYUV_API
>+int I422ToNV21(const uint8_t* src_y,
>+               int src_stride_y,
>+               const uint8_t* src_u,
>+               int src_stride_u,
>+               const uint8_t* src_v,
>+               int src_stride_v,
>+               uint8_t* dst_y,
>+               int dst_stride_y,
>+               uint8_t* dst_vu,
>+               int dst_stride_vu,
>+               int width,
>+               int height) {
>+  int y;
>+  void (*MergeUVRow)(const uint8_t* src_u, const uint8_t* src_v,
>+                     uint8_t* dst_uv, int width) = MergeUVRow_C;
>+  void (*InterpolateRow)(uint8_t * dst_ptr, const uint8_t* src_ptr,
>+                         ptrdiff_t src_stride, int dst_width,
>+                         int source_y_fraction) = InterpolateRow_C;
>+  int halfwidth = (width + 1) >> 1;
>+  int halfheight = (height + 1) >> 1;
>+  if (!src_u || !src_v || !dst_vu || width <= 0 || height == 0) {
>+    return -1;
>+  }
>+  // Negative height means invert the image.
>+  if (height < 0) {
>+    height = -height;
>+    halfheight = (height + 1) >> 1;
>+    src_y = src_y + (height - 1) * src_stride_y;
>+    src_u = src_u + (halfheight - 1) * src_stride_u;
>+    src_v = src_v + (halfheight - 1) * src_stride_v;
>+    src_stride_y = -src_stride_y;
>+    src_stride_u = -src_stride_u;
>+    src_stride_v = -src_stride_v;
>+  }
>+#if defined(HAS_MERGEUVROW_SSE2)
>+  if (TestCpuFlag(kCpuHasSSE2)) {
>+    MergeUVRow = MergeUVRow_Any_SSE2;
>+    if (IS_ALIGNED(halfwidth, 16)) {
>+      MergeUVRow = MergeUVRow_SSE2;
>+    }
>+  }
>+#endif
>+#if defined(HAS_MERGEUVROW_AVX2)
>+  if (TestCpuFlag(kCpuHasAVX2)) {
>+    MergeUVRow = MergeUVRow_Any_AVX2;
>+    if (IS_ALIGNED(halfwidth, 32)) {
>+      MergeUVRow = MergeUVRow_AVX2;
>+    }
>+  }
>+#endif
>+#if defined(HAS_MERGEUVROW_NEON)
>+  if (TestCpuFlag(kCpuHasNEON)) {
>+    MergeUVRow = MergeUVRow_Any_NEON;
>+    if (IS_ALIGNED(halfwidth, 16)) {
>+      MergeUVRow = MergeUVRow_NEON;
>+    }
>+  }
>+#endif
>+#if defined(HAS_MERGEUVROW_MSA)
>+  if (TestCpuFlag(kCpuHasMSA)) {
>+    MergeUVRow = MergeUVRow_Any_MSA;
>+    if (IS_ALIGNED(halfwidth, 16)) {
>+      MergeUVRow = MergeUVRow_MSA;
>+    }
>+  }
>+#endif
>+#if defined(HAS_MERGEUVROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    MergeUVRow = MergeUVRow_Any_MMI;
>+    if (IS_ALIGNED(halfwidth, 8)) {
>+      MergeUVRow = MergeUVRow_MMI;
>+    }
>+  }
>+#endif
>+#if defined(HAS_INTERPOLATEROW_SSSE3)
>+  if (TestCpuFlag(kCpuHasSSSE3)) {
>+    InterpolateRow = InterpolateRow_Any_SSSE3;
>+    if (IS_ALIGNED(width, 16)) {
>+      InterpolateRow = InterpolateRow_SSSE3;
>+    }
>+  }
>+#endif
>+#if defined(HAS_INTERPOLATEROW_AVX2)
>+  if (TestCpuFlag(kCpuHasAVX2)) {
>+    InterpolateRow = InterpolateRow_Any_AVX2;
>+    if (IS_ALIGNED(width, 32)) {
>+      InterpolateRow = InterpolateRow_AVX2;
>+    }
>+  }
>+#endif
>+#if defined(HAS_INTERPOLATEROW_NEON)
>+  if (TestCpuFlag(kCpuHasNEON)) {
>+    InterpolateRow = InterpolateRow_Any_NEON;
>+    if (IS_ALIGNED(width, 16)) {
>+      InterpolateRow = InterpolateRow_NEON;
>+    }
>+  }
>+#endif
>+#if defined(HAS_INTERPOLATEROW_MSA)
>+  if (TestCpuFlag(kCpuHasMSA)) {
>+    InterpolateRow = InterpolateRow_Any_MSA;
>+    if (IS_ALIGNED(width, 32)) {
>+      InterpolateRow = InterpolateRow_MSA;
>+    }
>+  }
>+#endif
>+#if defined(HAS_INTERPOLATEROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    InterpolateRow = InterpolateRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      InterpolateRow = InterpolateRow_MMI;
>+    }
>+  }
>+#endif
>+
>+  if (dst_y) {
>+    CopyPlane(src_y, src_stride_y, dst_y, dst_stride_y, halfwidth, height);
>+  }
>+  {
>+    // Allocate 2 rows of vu.
>+    int awidth = halfwidth * 2;
>+    align_buffer_64(row_vu_0, awidth * 2);
>+    uint8_t* row_vu_1 = row_vu_0 + awidth;
>+
>+    for (y = 0; y < height - 1; y += 2) {
>+      MergeUVRow(src_v, src_u, row_vu_0, halfwidth);
>+      MergeUVRow(src_v + src_stride_v, src_u + src_stride_u, row_vu_1,
>+                 halfwidth);
>+      InterpolateRow(dst_vu, row_vu_0, awidth, awidth, 128);
>+      src_u += src_stride_u * 2;
>+      src_v += src_stride_v * 2;
>+      dst_vu += dst_stride_vu;
>+    }
>+    if (height & 1) {
>+      MergeUVRow(src_v, src_u, dst_vu, halfwidth);
>+    }
>+    free_aligned_buffer_64(row_vu_0);
>+  }
>+  return 0;
>+}
>+#endif  // I422TONV21_ROW_VERSION
>+
> // 444 chroma is 1x width, 1x height
> // 420 chroma is 1/2 width, 1/2 height
> LIBYUV_API
>@@ -237,6 +426,46 @@ int I444ToI420(const uint8_t* src_y,
>                     dst_v, dst_stride_v, width, height, width, height);
> }
> 
>+// TODO(fbarchard): Implement row conversion.
>+LIBYUV_API
>+int I444ToNV21(const uint8_t* src_y,
>+               int src_stride_y,
>+               const uint8_t* src_u,
>+               int src_stride_u,
>+               const uint8_t* src_v,
>+               int src_stride_v,
>+               uint8_t* dst_y,
>+               int dst_stride_y,
>+               uint8_t* dst_vu,
>+               int dst_stride_vu,
>+               int width,
>+               int height) {
>+  int halfwidth = (width + 1) >> 1;
>+  int halfheight = (height + 1) >> 1;
>+  // Negative height means invert the image.
>+  if (height < 0) {
>+    height = -height;
>+    halfheight = (height + 1) >> 1;
>+    src_y = src_y + (height - 1) * src_stride_y;
>+    src_u = src_u + (height - 1) * src_stride_u;
>+    src_v = src_v + (height - 1) * src_stride_v;
>+    src_stride_y = -src_stride_y;
>+    src_stride_u = -src_stride_u;
>+    src_stride_v = -src_stride_v;
>+  }
>+  // Allocate u and v buffers
>+  align_buffer_64(plane_u, halfwidth * halfheight * 2);
>+  uint8_t* plane_v = plane_u + halfwidth * halfheight;
>+
>+  I444ToI420(src_y, src_stride_y, src_u, src_stride_u, src_v, src_stride_v,
>+             dst_y, dst_stride_y, plane_u, halfwidth, plane_v, halfwidth, width,
>+             height);
>+  MergeUVPlane(plane_v, halfwidth, plane_u, halfwidth, dst_vu, dst_stride_vu,
>+               halfwidth, halfheight);
>+  free_aligned_buffer_64(plane_u);
>+  return 0;
>+}
>+
> // I400 is greyscale typically used in MJPG
> LIBYUV_API
> int I400ToI420(const uint8_t* src_y,
>@@ -269,6 +498,35 @@ int I400ToI420(const uint8_t* src_y,
>   return 0;
> }
> 
>+// I400 is greyscale typically used in MJPG
>+LIBYUV_API
>+int I400ToNV21(const uint8_t* src_y,
>+               int src_stride_y,
>+               uint8_t* dst_y,
>+               int dst_stride_y,
>+               uint8_t* dst_vu,
>+               int dst_stride_vu,
>+               int width,
>+               int height) {
>+  int halfwidth = (width + 1) >> 1;
>+  int halfheight = (height + 1) >> 1;
>+  if (!dst_vu || width <= 0 || height == 0) {
>+    return -1;
>+  }
>+  // Negative height means invert the image.
>+  if (height < 0) {
>+    height = -height;
>+    halfheight = (height + 1) >> 1;
>+    src_y = src_y + (height - 1) * src_stride_y;
>+    src_stride_y = -src_stride_y;
>+  }
>+  if (dst_y) {
>+    CopyPlane(src_y, src_stride_y, dst_y, dst_stride_y, width, height);
>+  }
>+  SetPlane(dst_vu, dst_stride_vu, halfwidth * 2, halfheight, 128);
>+  return 0;
>+}
>+
> static void CopyPlane2(const uint8_t* src,
>                        int src_stride_0,
>                        int src_stride_1,
>@@ -502,6 +760,18 @@ int YUY2ToI420(const uint8_t* src_yuy2,
>     }
>   }
> #endif
>+#if defined(HAS_YUY2TOYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    YUY2ToYRow = YUY2ToYRow_Any_MMI;
>+    YUY2ToUVRow = YUY2ToUVRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      YUY2ToYRow = YUY2ToYRow_MMI;
>+      if (IS_ALIGNED(width, 16)) {
>+        YUY2ToUVRow = YUY2ToUVRow_MMI;
>+      }
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height - 1; y += 2) {
>     YUY2ToUVRow(src_yuy2, src_stride_yuy2, dst_u, dst_v, width);
>@@ -583,6 +853,16 @@ int UYVYToI420(const uint8_t* src_uyvy,
>     }
>   }
> #endif
>+#if defined(HAS_UYVYTOYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    UYVYToYRow = UYVYToYRow_Any_MMI;
>+    UYVYToUVRow = UYVYToUVRow_Any_MMI;
>+    if (IS_ALIGNED(width, 16)) {
>+      UYVYToYRow = UYVYToYRow_MMI;
>+      UYVYToUVRow = UYVYToUVRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height - 1; y += 2) {
>     UYVYToUVRow(src_uyvy, src_stride_uyvy, dst_u, dst_v, width);
>@@ -679,6 +959,22 @@ int ARGBToI420(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBTOYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToYRow = ARGBToYRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      ARGBToYRow = ARGBToYRow_MMI;
>+    }
>+  }
>+#endif
>+#if defined(HAS_ARGBTOUVROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToUVRow = ARGBToUVRow_Any_MMI;
>+    if (IS_ALIGNED(width, 16)) {
>+      ARGBToUVRow = ARGBToUVRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height - 1; y += 2) {
>     ARGBToUVRow(src_argb, src_stride_argb, dst_u, dst_v, width);
>@@ -765,6 +1061,22 @@ int BGRAToI420(const uint8_t* src_bgra,
>     }
>   }
> #endif
>+#if defined(HAS_BGRATOYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    BGRAToYRow = BGRAToYRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      BGRAToYRow = BGRAToYRow_MMI;
>+    }
>+  }
>+#endif
>+#if defined(HAS_BGRATOUVROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    BGRAToUVRow = BGRAToUVRow_Any_MMI;
>+    if (IS_ALIGNED(width, 16)) {
>+      BGRAToUVRow = BGRAToUVRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height - 1; y += 2) {
>     BGRAToUVRow(src_bgra, src_stride_bgra, dst_u, dst_v, width);
>@@ -851,6 +1163,22 @@ int ABGRToI420(const uint8_t* src_abgr,
>     }
>   }
> #endif
>+#if defined(HAS_ABGRTOYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ABGRToYRow = ABGRToYRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      ABGRToYRow = ABGRToYRow_MMI;
>+    }
>+  }
>+#endif
>+#if defined(HAS_ABGRTOUVROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ABGRToUVRow = ABGRToUVRow_Any_MMI;
>+    if (IS_ALIGNED(width, 16)) {
>+      ABGRToUVRow = ABGRToUVRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height - 1; y += 2) {
>     ABGRToUVRow(src_abgr, src_stride_abgr, dst_u, dst_v, width);
>@@ -937,6 +1265,22 @@ int RGBAToI420(const uint8_t* src_rgba,
>     }
>   }
> #endif
>+#if defined(HAS_RGBATOYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    RGBAToYRow = RGBAToYRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      RGBAToYRow = RGBAToYRow_MMI;
>+    }
>+  }
>+#endif
>+#if defined(HAS_RGBATOUVROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    RGBAToUVRow = RGBAToUVRow_Any_MMI;
>+    if (IS_ALIGNED(width, 16)) {
>+      RGBAToUVRow = RGBAToUVRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height - 1; y += 2) {
>     RGBAToUVRow(src_rgba, src_stride_rgba, dst_u, dst_v, width);
>@@ -967,7 +1311,8 @@ int RGB24ToI420(const uint8_t* src_rgb24,
>                 int width,
>                 int height) {
>   int y;
>-#if (defined(HAS_RGB24TOYROW_NEON) || defined(HAS_RGB24TOYROW_MSA))
>+#if (defined(HAS_RGB24TOYROW_NEON) || defined(HAS_RGB24TOYROW_MSA) || \
>+     defined(HAS_RGB24TOYROW_MMI))
>   void (*RGB24ToUVRow)(const uint8_t* src_rgb24, int src_stride_rgb24,
>                        uint8_t* dst_u, uint8_t* dst_v, int width) =
>       RGB24ToUVRow_C;
>@@ -1013,6 +1358,17 @@ int RGB24ToI420(const uint8_t* src_rgb24,
>       RGB24ToUVRow = RGB24ToUVRow_MSA;
>     }
>   }
>+#elif defined(HAS_RGB24TOYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    RGB24ToUVRow = RGB24ToUVRow_Any_MMI;
>+    RGB24ToYRow = RGB24ToYRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      RGB24ToYRow = RGB24ToYRow_MMI;
>+      if (IS_ALIGNED(width, 16)) {
>+        RGB24ToUVRow = RGB24ToUVRow_MMI;
>+      }
>+    }
>+  }
> // Other platforms do intermediate conversion from RGB24 to ARGB.
> #else
> #if defined(HAS_RGB24TOARGBROW_SSSE3)
>@@ -1046,14 +1402,16 @@ int RGB24ToI420(const uint8_t* src_rgb24,
> #endif
> 
>   {
>-#if !(defined(HAS_RGB24TOYROW_NEON) || defined(HAS_RGB24TOYROW_MSA))
>+#if !(defined(HAS_RGB24TOYROW_NEON) || defined(HAS_RGB24TOYROW_MSA) || \
>+      defined(HAS_RGB24TOYROW_MMI))
>     // Allocate 2 rows of ARGB.
>     const int kRowSize = (width * 4 + 31) & ~31;
>     align_buffer_64(row, kRowSize * 2);
> #endif
> 
>     for (y = 0; y < height - 1; y += 2) {
>-#if (defined(HAS_RGB24TOYROW_NEON) || defined(HAS_RGB24TOYROW_MSA))
>+#if (defined(HAS_RGB24TOYROW_NEON) || defined(HAS_RGB24TOYROW_MSA) || \
>+     defined(HAS_RGB24TOYROW_MMI))
>       RGB24ToUVRow(src_rgb24, src_stride_rgb24, dst_u, dst_v, width);
>       RGB24ToYRow(src_rgb24, dst_y, width);
>       RGB24ToYRow(src_rgb24 + src_stride_rgb24, dst_y + dst_stride_y, width);
>@@ -1070,7 +1428,8 @@ int RGB24ToI420(const uint8_t* src_rgb24,
>       dst_v += dst_stride_v;
>     }
>     if (height & 1) {
>-#if (defined(HAS_RGB24TOYROW_NEON) || defined(HAS_RGB24TOYROW_MSA))
>+#if (defined(HAS_RGB24TOYROW_NEON) || defined(HAS_RGB24TOYROW_MSA) || \
>+     defined(HAS_RGB24TOYROW_MMI))
>       RGB24ToUVRow(src_rgb24, 0, dst_u, dst_v, width);
>       RGB24ToYRow(src_rgb24, dst_y, width);
> #else
>@@ -1079,7 +1438,8 @@ int RGB24ToI420(const uint8_t* src_rgb24,
>       ARGBToYRow(row, dst_y, width);
> #endif
>     }
>-#if !(defined(HAS_RGB24TOYROW_NEON) || defined(HAS_RGB24TOYROW_MSA))
>+#if !(defined(HAS_RGB24TOYROW_NEON) || defined(HAS_RGB24TOYROW_MSA) || \
>+      defined(HAS_RGB24TOYROW_MMI))
>     free_aligned_buffer_64(row);
> #endif
>   }
>@@ -1099,7 +1459,8 @@ int RAWToI420(const uint8_t* src_raw,
>               int width,
>               int height) {
>   int y;
>-#if (defined(HAS_RAWTOYROW_NEON) || defined(HAS_RAWTOYROW_MSA))
>+#if (defined(HAS_RAWTOYROW_NEON) || defined(HAS_RAWTOYROW_MSA) || \
>+     defined(HAS_RAWTOYROW_MMI))
>   void (*RAWToUVRow)(const uint8_t* src_raw, int src_stride_raw, uint8_t* dst_u,
>                      uint8_t* dst_v, int width) = RAWToUVRow_C;
>   void (*RAWToYRow)(const uint8_t* src_raw, uint8_t* dst_y, int width) =
>@@ -1144,6 +1505,17 @@ int RAWToI420(const uint8_t* src_raw,
>       RAWToUVRow = RAWToUVRow_MSA;
>     }
>   }
>+#elif defined(HAS_RAWTOYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    RAWToUVRow = RAWToUVRow_Any_MMI;
>+    RAWToYRow = RAWToYRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      RAWToYRow = RAWToYRow_MMI;
>+      if (IS_ALIGNED(width, 16)) {
>+        RAWToUVRow = RAWToUVRow_MMI;
>+      }
>+    }
>+  }
> // Other platforms do intermediate conversion from RAW to ARGB.
> #else
> #if defined(HAS_RAWTOARGBROW_SSSE3)
>@@ -1177,14 +1549,16 @@ int RAWToI420(const uint8_t* src_raw,
> #endif
> 
>   {
>-#if !(defined(HAS_RAWTOYROW_NEON) || defined(HAS_RAWTOYROW_MSA))
>+#if !(defined(HAS_RAWTOYROW_NEON) || defined(HAS_RAWTOYROW_MSA) || \
>+      defined(HAS_RAWTOYROW_MMI))
>     // Allocate 2 rows of ARGB.
>     const int kRowSize = (width * 4 + 31) & ~31;
>     align_buffer_64(row, kRowSize * 2);
> #endif
> 
>     for (y = 0; y < height - 1; y += 2) {
>-#if (defined(HAS_RAWTOYROW_NEON) || defined(HAS_RAWTOYROW_MSA))
>+#if (defined(HAS_RAWTOYROW_NEON) || defined(HAS_RAWTOYROW_MSA) || \
>+     defined(HAS_RAWTOYROW_MMI))
>       RAWToUVRow(src_raw, src_stride_raw, dst_u, dst_v, width);
>       RAWToYRow(src_raw, dst_y, width);
>       RAWToYRow(src_raw + src_stride_raw, dst_y + dst_stride_y, width);
>@@ -1201,7 +1575,8 @@ int RAWToI420(const uint8_t* src_raw,
>       dst_v += dst_stride_v;
>     }
>     if (height & 1) {
>-#if (defined(HAS_RAWTOYROW_NEON) || defined(HAS_RAWTOYROW_MSA))
>+#if (defined(HAS_RAWTOYROW_NEON) || defined(HAS_RAWTOYROW_MSA) || \
>+     defined(HAS_RAWTOYROW_MMI))
>       RAWToUVRow(src_raw, 0, dst_u, dst_v, width);
>       RAWToYRow(src_raw, dst_y, width);
> #else
>@@ -1210,7 +1585,8 @@ int RAWToI420(const uint8_t* src_raw,
>       ARGBToYRow(row, dst_y, width);
> #endif
>     }
>-#if !(defined(HAS_RAWTOYROW_NEON) || defined(HAS_RAWTOYROW_MSA))
>+#if !(defined(HAS_RAWTOYROW_NEON) || defined(HAS_RAWTOYROW_MSA) || \
>+      defined(HAS_RAWTOYROW_MMI))
>     free_aligned_buffer_64(row);
> #endif
>   }
>@@ -1230,7 +1606,8 @@ int RGB565ToI420(const uint8_t* src_rgb565,
>                  int width,
>                  int height) {
>   int y;
>-#if (defined(HAS_RGB565TOYROW_NEON) || defined(HAS_RGB565TOYROW_MSA))
>+#if (defined(HAS_RGB565TOYROW_NEON) || defined(HAS_RGB565TOYROW_MSA) || \
>+     defined(HAS_RGB565TOYROW_MMI))
>   void (*RGB565ToUVRow)(const uint8_t* src_rgb565, int src_stride_rgb565,
>                         uint8_t* dst_u, uint8_t* dst_v, int width) =
>       RGB565ToUVRow_C;
>@@ -1276,6 +1653,17 @@ int RGB565ToI420(const uint8_t* src_rgb565,
>       RGB565ToUVRow = RGB565ToUVRow_MSA;
>     }
>   }
>+#elif defined(HAS_RGB565TOYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    RGB565ToUVRow = RGB565ToUVRow_Any_MMI;
>+    RGB565ToYRow = RGB565ToYRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      RGB565ToYRow = RGB565ToYRow_MMI;
>+      if (IS_ALIGNED(width, 16)) {
>+        RGB565ToUVRow = RGB565ToUVRow_MMI;
>+      }
>+    }
>+  }
> // Other platforms do intermediate conversion from RGB565 to ARGB.
> #else
> #if defined(HAS_RGB565TOARGBROW_SSE2)
>@@ -1316,13 +1704,15 @@ int RGB565ToI420(const uint8_t* src_rgb565,
> #endif
> #endif
>   {
>-#if !(defined(HAS_RGB565TOYROW_NEON) || defined(HAS_RGB565TOYROW_MSA))
>+#if !(defined(HAS_RGB565TOYROW_NEON) || defined(HAS_RGB565TOYROW_MSA) || \
>+      defined(HAS_RGB565TOYROW_MMI))
>     // Allocate 2 rows of ARGB.
>     const int kRowSize = (width * 4 + 31) & ~31;
>     align_buffer_64(row, kRowSize * 2);
> #endif
>     for (y = 0; y < height - 1; y += 2) {
>-#if (defined(HAS_RGB565TOYROW_NEON) || defined(HAS_RGB565TOYROW_MSA))
>+#if (defined(HAS_RGB565TOYROW_NEON) || defined(HAS_RGB565TOYROW_MSA) || \
>+     defined(HAS_RGB565TOYROW_MMI))
>       RGB565ToUVRow(src_rgb565, src_stride_rgb565, dst_u, dst_v, width);
>       RGB565ToYRow(src_rgb565, dst_y, width);
>       RGB565ToYRow(src_rgb565 + src_stride_rgb565, dst_y + dst_stride_y, width);
>@@ -1339,7 +1729,8 @@ int RGB565ToI420(const uint8_t* src_rgb565,
>       dst_v += dst_stride_v;
>     }
>     if (height & 1) {
>-#if (defined(HAS_RGB565TOYROW_NEON) || defined(HAS_RGB565TOYROW_MSA))
>+#if (defined(HAS_RGB565TOYROW_NEON) || defined(HAS_RGB565TOYROW_MSA) || \
>+     defined(HAS_RGB565TOYROW_MMI))
>       RGB565ToUVRow(src_rgb565, 0, dst_u, dst_v, width);
>       RGB565ToYRow(src_rgb565, dst_y, width);
> #else
>@@ -1348,7 +1739,8 @@ int RGB565ToI420(const uint8_t* src_rgb565,
>       ARGBToYRow(row, dst_y, width);
> #endif
>     }
>-#if !(defined(HAS_RGB565TOYROW_NEON) || defined(HAS_RGB565TOYROW_MSA))
>+#if !(defined(HAS_RGB565TOYROW_NEON) || defined(HAS_RGB565TOYROW_MSA) || \
>+      defined(HAS_RGB565TOYROW_MMI))
>     free_aligned_buffer_64(row);
> #endif
>   }
>@@ -1368,7 +1760,8 @@ int ARGB1555ToI420(const uint8_t* src_argb1555,
>                    int width,
>                    int height) {
>   int y;
>-#if (defined(HAS_ARGB1555TOYROW_NEON) || defined(HAS_ARGB1555TOYROW_MSA))
>+#if (defined(HAS_ARGB1555TOYROW_NEON) || defined(HAS_ARGB1555TOYROW_MSA) || \
>+     defined(HAS_ARGB1555TOYROW_MMI))
>   void (*ARGB1555ToUVRow)(const uint8_t* src_argb1555, int src_stride_argb1555,
>                           uint8_t* dst_u, uint8_t* dst_v, int width) =
>       ARGB1555ToUVRow_C;
>@@ -1415,6 +1808,17 @@ int ARGB1555ToI420(const uint8_t* src_argb1555,
>       ARGB1555ToUVRow = ARGB1555ToUVRow_MSA;
>     }
>   }
>+#elif defined(HAS_ARGB1555TOYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGB1555ToUVRow = ARGB1555ToUVRow_Any_MMI;
>+    ARGB1555ToYRow = ARGB1555ToYRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      ARGB1555ToYRow = ARGB1555ToYRow_MMI;
>+      if (IS_ALIGNED(width, 16)) {
>+        ARGB1555ToUVRow = ARGB1555ToUVRow_MMI;
>+      }
>+    }
>+  }
> // Other platforms do intermediate conversion from ARGB1555 to ARGB.
> #else
> #if defined(HAS_ARGB1555TOARGBROW_SSE2)
>@@ -1455,14 +1859,16 @@ int ARGB1555ToI420(const uint8_t* src_argb1555,
> #endif
> #endif
>   {
>-#if !(defined(HAS_ARGB1555TOYROW_NEON) || defined(HAS_ARGB1555TOYROW_MSA))
>+#if !(defined(HAS_ARGB1555TOYROW_NEON) || defined(HAS_ARGB1555TOYROW_MSA) || \
>+      defined(HAS_ARGB1555TOYROW_MMI))
>     // Allocate 2 rows of ARGB.
>     const int kRowSize = (width * 4 + 31) & ~31;
>     align_buffer_64(row, kRowSize * 2);
> #endif
> 
>     for (y = 0; y < height - 1; y += 2) {
>-#if (defined(HAS_ARGB1555TOYROW_NEON) || defined(HAS_ARGB1555TOYROW_MSA))
>+#if (defined(HAS_ARGB1555TOYROW_NEON) || defined(HAS_ARGB1555TOYROW_MSA) || \
>+     defined(HAS_ARGB1555TOYROW_MMI))
>       ARGB1555ToUVRow(src_argb1555, src_stride_argb1555, dst_u, dst_v, width);
>       ARGB1555ToYRow(src_argb1555, dst_y, width);
>       ARGB1555ToYRow(src_argb1555 + src_stride_argb1555, dst_y + dst_stride_y,
>@@ -1481,7 +1887,8 @@ int ARGB1555ToI420(const uint8_t* src_argb1555,
>       dst_v += dst_stride_v;
>     }
>     if (height & 1) {
>-#if (defined(HAS_ARGB1555TOYROW_NEON) || defined(HAS_ARGB1555TOYROW_MSA))
>+#if (defined(HAS_ARGB1555TOYROW_NEON) || defined(HAS_ARGB1555TOYROW_MSA) || \
>+     defined(HAS_ARGB1555TOYROW_MMI))
>       ARGB1555ToUVRow(src_argb1555, 0, dst_u, dst_v, width);
>       ARGB1555ToYRow(src_argb1555, dst_y, width);
> #else
>@@ -1490,7 +1897,8 @@ int ARGB1555ToI420(const uint8_t* src_argb1555,
>       ARGBToYRow(row, dst_y, width);
> #endif
>     }
>-#if !(defined(HAS_ARGB1555TOYROW_NEON) || defined(HAS_ARGB1555TOYROW_MSA))
>+#if !(defined(HAS_ARGB1555TOYROW_NEON) || defined(HAS_ARGB1555TOYROW_MSA) || \
>+      defined(HAS_ARGB1555TOYROW_MMI))
>     free_aligned_buffer_64(row);
> #endif
>   }
>@@ -1510,7 +1918,7 @@ int ARGB4444ToI420(const uint8_t* src_argb4444,
>                    int width,
>                    int height) {
>   int y;
>-#if defined(HAS_ARGB4444TOYROW_NEON)
>+#if (defined(HAS_ARGB4444TOYROW_NEON) || defined(HAS_ARGB4444TOYROW_MMI))
>   void (*ARGB4444ToUVRow)(const uint8_t* src_argb4444, int src_stride_argb4444,
>                           uint8_t* dst_u, uint8_t* dst_v, int width) =
>       ARGB4444ToUVRow_C;
>@@ -1548,6 +1956,17 @@ int ARGB4444ToI420(const uint8_t* src_argb4444,
>       }
>     }
>   }
>+#elif defined(HAS_ARGB4444TOYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGB4444ToUVRow = ARGB4444ToUVRow_Any_MMI;
>+    ARGB4444ToYRow = ARGB4444ToYRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      ARGB4444ToYRow = ARGB4444ToYRow_MMI;
>+      if (IS_ALIGNED(width, 16)) {
>+        ARGB4444ToUVRow = ARGB4444ToUVRow_MMI;
>+      }
>+    }
>+  }
> // Other platforms do intermediate conversion from ARGB4444 to ARGB.
> #else
> #if defined(HAS_ARGB4444TOARGBROW_SSE2)
>@@ -1606,17 +2025,29 @@ int ARGB4444ToI420(const uint8_t* src_argb4444,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBTOYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToUVRow = ARGBToUVRow_Any_MMI;
>+    ARGBToYRow = ARGBToYRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      ARGBToYRow = ARGBToYRow_MMI;
>+      if (IS_ALIGNED(width, 16)) {
>+        ARGBToUVRow = ARGBToUVRow_MMI;
>+      }
>+    }
>+  }
>+#endif
> #endif
> 
>   {
>-#if !defined(HAS_ARGB4444TOYROW_NEON)
>+#if !(defined(HAS_ARGB4444TOYROW_NEON) || defined(HAS_ARGB4444TOYROW_MMI))
>     // Allocate 2 rows of ARGB.
>     const int kRowSize = (width * 4 + 31) & ~31;
>     align_buffer_64(row, kRowSize * 2);
> #endif
> 
>     for (y = 0; y < height - 1; y += 2) {
>-#if defined(HAS_ARGB4444TOYROW_NEON)
>+#if (defined(HAS_ARGB4444TOYROW_NEON) || defined(HAS_ARGB4444TOYROW_MMI))
>       ARGB4444ToUVRow(src_argb4444, src_stride_argb4444, dst_u, dst_v, width);
>       ARGB4444ToYRow(src_argb4444, dst_y, width);
>       ARGB4444ToYRow(src_argb4444 + src_stride_argb4444, dst_y + dst_stride_y,
>@@ -1635,7 +2066,7 @@ int ARGB4444ToI420(const uint8_t* src_argb4444,
>       dst_v += dst_stride_v;
>     }
>     if (height & 1) {
>-#if defined(HAS_ARGB4444TOYROW_NEON)
>+#if (defined(HAS_ARGB4444TOYROW_NEON) || defined(HAS_ARGB4444TOYROW_MMI))
>       ARGB4444ToUVRow(src_argb4444, 0, dst_u, dst_v, width);
>       ARGB4444ToYRow(src_argb4444, dst_y, width);
> #else
>@@ -1644,7 +2075,7 @@ int ARGB4444ToI420(const uint8_t* src_argb4444,
>       ARGBToYRow(row, dst_y, width);
> #endif
>     }
>-#if !defined(HAS_ARGB4444TOYROW_NEON)
>+#if !(defined(HAS_ARGB4444TOYROW_NEON) || defined(HAS_ARGB4444TOYROW_MMI))
>     free_aligned_buffer_64(row);
> #endif
>   }
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert_argb.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert_argb.cc
>index 4c317ae30c68d61d00fad58c4b3e8a62d8c33b04..b376a0f38767652afe736c324ffb46a5e68f5034 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert_argb.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert_argb.cc
>@@ -885,6 +885,14 @@ static int I420AlphaToARGBMatrix(const uint8_t* src_y,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBATTENUATEROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBAttenuateRow = ARGBAttenuateRow_Any_MMI;
>+    if (IS_ALIGNED(width, 2)) {
>+      ARGBAttenuateRow = ARGBAttenuateRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     I422AlphaToARGBRow(src_y, src_u, src_v, src_a, dst_argb, yuvconstants,
>@@ -1004,6 +1012,14 @@ int I400ToARGB(const uint8_t* src_y,
>     }
>   }
> #endif
>+#if defined(HAS_I400TOARGBROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    I400ToARGBRow = I400ToARGBRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      I400ToARGBRow = I400ToARGBRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     I400ToARGBRow(src_y, dst_argb, width);
>@@ -1070,6 +1086,14 @@ int J400ToARGB(const uint8_t* src_y,
>       J400ToARGBRow = J400ToARGBRow_MSA;
>     }
>   }
>+#endif
>+#if defined(HAS_J400TOARGBROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    J400ToARGBRow = J400ToARGBRow_Any_MMI;
>+    if (IS_ALIGNED(width, 4)) {
>+      J400ToARGBRow = J400ToARGBRow_MMI;
>+    }
>+  }
> #endif
>   for (y = 0; y < height; ++y) {
>     J400ToARGBRow(src_y, dst_argb, width);
>@@ -1201,6 +1225,14 @@ int RGB24ToARGB(const uint8_t* src_rgb24,
>     }
>   }
> #endif
>+#if defined(HAS_RGB24TOARGBROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    RGB24ToARGBRow = RGB24ToARGBRow_Any_MMI;
>+    if (IS_ALIGNED(width, 4)) {
>+      RGB24ToARGBRow = RGB24ToARGBRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     RGB24ToARGBRow(src_rgb24, dst_argb, width);
>@@ -1260,6 +1292,14 @@ int RAWToARGB(const uint8_t* src_raw,
>     }
>   }
> #endif
>+#if defined(HAS_RAWTOARGBROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    RAWToARGBRow = RAWToARGBRow_Any_MMI;
>+    if (IS_ALIGNED(width, 4)) {
>+      RAWToARGBRow = RAWToARGBRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     RAWToARGBRow(src_raw, dst_argb, width);
>@@ -1327,6 +1367,14 @@ int RGB565ToARGB(const uint8_t* src_rgb565,
>     }
>   }
> #endif
>+#if defined(HAS_RGB565TOARGBROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    RGB565ToARGBRow = RGB565ToARGBRow_Any_MMI;
>+    if (IS_ALIGNED(width, 4)) {
>+      RGB565ToARGBRow = RGB565ToARGBRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     RGB565ToARGBRow(src_rgb565, dst_argb, width);
>@@ -1394,6 +1442,14 @@ int ARGB1555ToARGB(const uint8_t* src_argb1555,
>     }
>   }
> #endif
>+#if defined(HAS_ARGB1555TOARGBROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGB1555ToARGBRow = ARGB1555ToARGBRow_Any_MMI;
>+    if (IS_ALIGNED(width, 4)) {
>+      ARGB1555ToARGBRow = ARGB1555ToARGBRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     ARGB1555ToARGBRow(src_argb1555, dst_argb, width);
>@@ -1461,6 +1517,14 @@ int ARGB4444ToARGB(const uint8_t* src_argb4444,
>     }
>   }
> #endif
>+#if defined(HAS_ARGB4444TOARGBROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGB4444ToARGBRow = ARGB4444ToARGBRow_Any_MMI;
>+    if (IS_ALIGNED(width, 4)) {
>+      ARGB4444ToARGBRow = ARGB4444ToARGBRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     ARGB4444ToARGBRow(src_argb4444, dst_argb, width);
>@@ -1789,6 +1853,22 @@ static int NV12ToRGB24Matrix(const uint8_t* src_y,
>     }
>   }
> #endif
>+#if defined(HAS_NV12TORGB24ROW_SSSE3)
>+  if (TestCpuFlag(kCpuHasSSSE3)) {
>+    NV12ToRGB24Row = NV12ToRGB24Row_Any_SSSE3;
>+    if (IS_ALIGNED(width, 16)) {
>+      NV12ToRGB24Row = NV12ToRGB24Row_SSSE3;
>+    }
>+  }
>+#endif
>+#if defined(HAS_NV12TORGB24ROW_AVX2)
>+  if (TestCpuFlag(kCpuHasAVX2)) {
>+    NV12ToRGB24Row = NV12ToRGB24Row_Any_AVX2;
>+    if (IS_ALIGNED(width, 32)) {
>+      NV12ToRGB24Row = NV12ToRGB24Row_AVX2;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     NV12ToRGB24Row(src_y, src_uv, dst_rgb24, yuvconstants, width);
>@@ -1832,6 +1912,22 @@ static int NV21ToRGB24Matrix(const uint8_t* src_y,
>     }
>   }
> #endif
>+#if defined(HAS_NV21TORGB24ROW_SSSE3)
>+  if (TestCpuFlag(kCpuHasSSSE3)) {
>+    NV21ToRGB24Row = NV21ToRGB24Row_Any_SSSE3;
>+    if (IS_ALIGNED(width, 16)) {
>+      NV21ToRGB24Row = NV21ToRGB24Row_SSSE3;
>+    }
>+  }
>+#endif
>+#if defined(HAS_NV21TORGB24ROW_AVX2)
>+  if (TestCpuFlag(kCpuHasAVX2)) {
>+    NV21ToRGB24Row = NV21ToRGB24Row_Any_AVX2;
>+    if (IS_ALIGNED(width, 32)) {
>+      NV21ToRGB24Row = NV21ToRGB24Row_AVX2;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     NV21ToRGB24Row(src_y, src_vu, dst_rgb24, yuvconstants, width);
>@@ -1844,8 +1940,7 @@ static int NV21ToRGB24Matrix(const uint8_t* src_y,
>   return 0;
> }
> 
>-// TODO(fbarchard): \(fbarchard): NV12ToRAW can be implemented by mirrored
>-// matrix. Convert NV12 to RGB24.
>+// Convert NV12 to RGB24.
> LIBYUV_API
> int NV12ToRGB24(const uint8_t* src_y,
>                 int src_stride_y,
>@@ -1875,6 +1970,34 @@ int NV21ToRGB24(const uint8_t* src_y,
>                            width, height);
> }
> 
>+// Convert NV12 to RAW.
>+LIBYUV_API
>+int NV12ToRAW(const uint8_t* src_y,
>+              int src_stride_y,
>+              const uint8_t* src_uv,
>+              int src_stride_uv,
>+              uint8_t* dst_raw,
>+              int dst_stride_raw,
>+              int width,
>+              int height) {
>+  return NV21ToRGB24Matrix(src_y, src_stride_y, src_uv, src_stride_uv, dst_raw,
>+                           dst_stride_raw, &kYvuI601Constants, width, height);
>+}
>+
>+// Convert NV21 to RAW.
>+LIBYUV_API
>+int NV21ToRAW(const uint8_t* src_y,
>+              int src_stride_y,
>+              const uint8_t* src_vu,
>+              int src_stride_vu,
>+              uint8_t* dst_raw,
>+              int dst_stride_raw,
>+              int width,
>+              int height) {
>+  return NV12ToRGB24Matrix(src_y, src_stride_y, src_vu, src_stride_vu, dst_raw,
>+                           dst_stride_raw, &kYvuI601Constants, width, height);
>+}
>+
> // Convert M420 to ARGB.
> LIBYUV_API
> int M420ToARGB(const uint8_t* src_m420,
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert_from.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert_from.cc
>index b5587ced625702f13d0bb3b353fa1ecedf61ffe6..706067bb29ba501016f685ee2e71bb48b5ae25dd 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert_from.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert_from.cc
>@@ -302,6 +302,14 @@ int I420ToYUY2(const uint8_t* src_y,
>     }
>   }
> #endif
>+#if defined(HAS_I422TOYUY2ROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    I422ToYUY2Row = I422ToYUY2Row_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      I422ToYUY2Row = I422ToYUY2Row_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height - 1; y += 2) {
>     I422ToYUY2Row(src_y, src_u, src_v, dst_yuy2, width);
>@@ -381,6 +389,14 @@ int I422ToUYVY(const uint8_t* src_y,
>     }
>   }
> #endif
>+#if defined(HAS_I422TOUYVYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    I422ToUYVYRow = I422ToUYVYRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      I422ToUYVYRow = I422ToUYVYRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     I422ToUYVYRow(src_y, src_u, src_v, dst_uyvy, width);
>@@ -448,6 +464,14 @@ int I420ToUYVY(const uint8_t* src_y,
>     }
>   }
> #endif
>+#if defined(HAS_I422TOUYVYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    I422ToUYVYRow = I422ToUYVYRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      I422ToUYVYRow = I422ToUYVYRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height - 1; y += 2) {
>     I422ToUYVYRow(src_y, src_u, src_v, dst_uyvy, width);
>@@ -906,18 +930,19 @@ int I420ToARGB4444(const uint8_t* src_y,
>   return 0;
> }
> 
>-// Convert I420 to RGB565.
>+// Convert I420 to RGB565 with specified color matrix.
> LIBYUV_API
>-int I420ToRGB565(const uint8_t* src_y,
>-                 int src_stride_y,
>-                 const uint8_t* src_u,
>-                 int src_stride_u,
>-                 const uint8_t* src_v,
>-                 int src_stride_v,
>-                 uint8_t* dst_rgb565,
>-                 int dst_stride_rgb565,
>-                 int width,
>-                 int height) {
>+int I420ToRGB565Matrix(const uint8_t* src_y,
>+                       int src_stride_y,
>+                       const uint8_t* src_u,
>+                       int src_stride_u,
>+                       const uint8_t* src_v,
>+                       int src_stride_v,
>+                       uint8_t* dst_rgb565,
>+                       int dst_stride_rgb565,
>+                       const struct YuvConstants* yuvconstants,
>+                       int width,
>+                       int height) {
>   int y;
>   void (*I422ToRGB565Row)(const uint8_t* y_buf, const uint8_t* u_buf,
>                           const uint8_t* v_buf, uint8_t* rgb_buf,
>@@ -966,7 +991,7 @@ int I420ToRGB565(const uint8_t* src_y,
> #endif
> 
>   for (y = 0; y < height; ++y) {
>-    I422ToRGB565Row(src_y, src_u, src_v, dst_rgb565, &kYuvI601Constants, width);
>+    I422ToRGB565Row(src_y, src_u, src_v, dst_rgb565, yuvconstants, width);
>     dst_rgb565 += dst_stride_rgb565;
>     src_y += src_stride_y;
>     if (y & 1) {
>@@ -977,6 +1002,57 @@ int I420ToRGB565(const uint8_t* src_y,
>   return 0;
> }
> 
>+// Convert I420 to RGB565.
>+LIBYUV_API
>+int I420ToRGB565(const uint8_t* src_y,
>+                 int src_stride_y,
>+                 const uint8_t* src_u,
>+                 int src_stride_u,
>+                 const uint8_t* src_v,
>+                 int src_stride_v,
>+                 uint8_t* dst_rgb565,
>+                 int dst_stride_rgb565,
>+                 int width,
>+                 int height) {
>+  return I420ToRGB565Matrix(src_y, src_stride_y, src_u, src_stride_u, src_v,
>+                            src_stride_v, dst_rgb565, dst_stride_rgb565,
>+                            &kYuvI601Constants, width, height);
>+}
>+
>+// Convert J420 to RGB565.
>+LIBYUV_API
>+int J420ToRGB565(const uint8_t* src_y,
>+                 int src_stride_y,
>+                 const uint8_t* src_u,
>+                 int src_stride_u,
>+                 const uint8_t* src_v,
>+                 int src_stride_v,
>+                 uint8_t* dst_rgb565,
>+                 int dst_stride_rgb565,
>+                 int width,
>+                 int height) {
>+  return I420ToRGB565Matrix(src_y, src_stride_y, src_u, src_stride_u, src_v,
>+                            src_stride_v, dst_rgb565, dst_stride_rgb565,
>+                            &kYuvJPEGConstants, width, height);
>+}
>+
>+// Convert H420 to RGB565.
>+LIBYUV_API
>+int H420ToRGB565(const uint8_t* src_y,
>+                 int src_stride_y,
>+                 const uint8_t* src_u,
>+                 int src_stride_u,
>+                 const uint8_t* src_v,
>+                 int src_stride_v,
>+                 uint8_t* dst_rgb565,
>+                 int dst_stride_rgb565,
>+                 int width,
>+                 int height) {
>+  return I420ToRGB565Matrix(src_y, src_stride_y, src_u, src_stride_u, src_v,
>+                            src_stride_v, dst_rgb565, dst_stride_rgb565,
>+                            &kYuvH709Constants, width, height);
>+}
>+
> // Convert I422 to RGB565.
> LIBYUV_API
> int I422ToRGB565(const uint8_t* src_y,
>@@ -1154,8 +1230,8 @@ int I420ToRGB565Dither(const uint8_t* src_y,
>     for (y = 0; y < height; ++y) {
>       I422ToARGBRow(src_y, src_u, src_v, row_argb, &kYuvI601Constants, width);
>       ARGBToRGB565DitherRow(row_argb, dst_rgb565,
>-                            *(uint32_t*)(dither4x4 + ((y & 3) << 2)),  // NOLINT
>-                            width);                                    // NOLINT
>+                            *(const uint32_t*)(dither4x4 + ((y & 3) << 2)),
>+                            width);
>       dst_rgb565 += dst_stride_rgb565;
>       src_y += src_stride_y;
>       if (y & 1) {
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert_from_argb.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert_from_argb.cc
>index 16b838458f02d07767e23468d94b612d5dab2849..1b070c10f756fcba861c5ab1b6a292073994c740 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert_from_argb.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert_from_argb.cc
>@@ -76,6 +76,14 @@ int ARGBToI444(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBTOUV444ROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToUV444Row = ARGBToUV444Row_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      ARGBToUV444Row = ARGBToUV444Row_MMI;
>+    }
>+  }
>+#endif
> #if defined(HAS_ARGBTOYROW_SSSE3)
>   if (TestCpuFlag(kCpuHasSSSE3)) {
>     ARGBToYRow = ARGBToYRow_Any_SSSE3;
>@@ -108,6 +116,14 @@ int ARGBToI444(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBTOYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToYRow = ARGBToYRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      ARGBToYRow = ARGBToYRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     ARGBToUV444Row(src_argb, dst_u, dst_v, width);
>@@ -208,6 +224,23 @@ int ARGBToI422(const uint8_t* src_argb,
>   }
> #endif
> 
>+#if defined(HAS_ARGBTOYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToYRow = ARGBToYRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      ARGBToYRow = ARGBToYRow_MMI;
>+    }
>+  }
>+#endif
>+#if defined(HAS_ARGBTOUVROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToUVRow = ARGBToUVRow_Any_MMI;
>+    if (IS_ALIGNED(width, 16)) {
>+      ARGBToUVRow = ARGBToUVRow_MMI;
>+    }
>+  }
>+#endif
>+
>   for (y = 0; y < height; ++y) {
>     ARGBToUVRow(src_argb, 0, dst_u, dst_v, width);
>     ARGBToYRow(src_argb, dst_y, width);
>@@ -298,6 +331,22 @@ int ARGBToNV12(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBTOYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToYRow = ARGBToYRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      ARGBToYRow = ARGBToYRow_MMI;
>+    }
>+  }
>+#endif
>+#if defined(HAS_ARGBTOUVROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToUVRow = ARGBToUVRow_Any_MMI;
>+    if (IS_ALIGNED(width, 16)) {
>+      ARGBToUVRow = ARGBToUVRow_MMI;
>+    }
>+  }
>+#endif
> #if defined(HAS_MERGEUVROW_SSE2)
>   if (TestCpuFlag(kCpuHasSSE2)) {
>     MergeUVRow_ = MergeUVRow_Any_SSE2;
>@@ -329,6 +378,14 @@ int ARGBToNV12(const uint8_t* src_argb,
>       MergeUVRow_ = MergeUVRow_MSA;
>     }
>   }
>+#endif
>+#if defined(HAS_MERGEUVROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    MergeUVRow_ = MergeUVRow_Any_MMI;
>+    if (IS_ALIGNED(halfwidth, 8)) {
>+      MergeUVRow_ = MergeUVRow_MMI;
>+    }
>+  }
> #endif
>   {
>     // Allocate a rows of uv.
>@@ -434,6 +491,23 @@ int ARGBToNV21(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBTOYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToYRow = ARGBToYRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      ARGBToYRow = ARGBToYRow_MMI;
>+    }
>+  }
>+#endif
>+#if defined(HAS_ARGBTOUVROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToUVRow = ARGBToUVRow_Any_MMI;
>+    if (IS_ALIGNED(width, 16)) {
>+      ARGBToUVRow = ARGBToUVRow_MMI;
>+    }
>+  }
>+#endif
>+
> #if defined(HAS_MERGEUVROW_SSE2)
>   if (TestCpuFlag(kCpuHasSSE2)) {
>     MergeUVRow_ = MergeUVRow_Any_SSE2;
>@@ -465,6 +539,14 @@ int ARGBToNV21(const uint8_t* src_argb,
>       MergeUVRow_ = MergeUVRow_MSA;
>     }
>   }
>+#endif
>+#if defined(HAS_MERGEUVROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    MergeUVRow_ = MergeUVRow_Any_MMI;
>+    if (IS_ALIGNED(halfwidth, 8)) {
>+      MergeUVRow_ = MergeUVRow_MMI;
>+    }
>+  }
> #endif
>   {
>     // Allocate a rows of uv.
>@@ -575,6 +657,22 @@ int ARGBToYUY2(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBTOYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToYRow = ARGBToYRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      ARGBToYRow = ARGBToYRow_MMI;
>+    }
>+  }
>+#endif
>+#if defined(HAS_ARGBTOUVROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToUVRow = ARGBToUVRow_Any_MMI;
>+    if (IS_ALIGNED(width, 16)) {
>+      ARGBToUVRow = ARGBToUVRow_MMI;
>+    }
>+  }
>+#endif
> #if defined(HAS_I422TOYUY2ROW_SSE2)
>   if (TestCpuFlag(kCpuHasSSE2)) {
>     I422ToYUY2Row = I422ToYUY2Row_Any_SSE2;
>@@ -607,6 +705,14 @@ int ARGBToYUY2(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_I422TOYUY2ROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    I422ToYUY2Row = I422ToYUY2Row_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      I422ToYUY2Row = I422ToYUY2Row_MMI;
>+    }
>+  }
>+#endif
> 
>   {
>     // Allocate a rows of yuv.
>@@ -712,6 +818,22 @@ int ARGBToUYVY(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBTOYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToYRow = ARGBToYRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      ARGBToYRow = ARGBToYRow_MMI;
>+    }
>+  }
>+#endif
>+#if defined(HAS_ARGBTOUVROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToUVRow = ARGBToUVRow_Any_MMI;
>+    if (IS_ALIGNED(width, 16)) {
>+      ARGBToUVRow = ARGBToUVRow_MMI;
>+    }
>+  }
>+#endif
> #if defined(HAS_I422TOUYVYROW_SSE2)
>   if (TestCpuFlag(kCpuHasSSE2)) {
>     I422ToUYVYRow = I422ToUYVYRow_Any_SSE2;
>@@ -744,6 +866,14 @@ int ARGBToUYVY(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_I422TOUYVYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    I422ToUYVYRow = I422ToUYVYRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      I422ToUYVYRow = I422ToUYVYRow_MMI;
>+    }
>+  }
>+#endif
> 
>   {
>     // Allocate a rows of yuv.
>@@ -821,6 +951,14 @@ int ARGBToI400(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBTOYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToYRow = ARGBToYRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      ARGBToYRow = ARGBToYRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     ARGBToYRow(src_argb, dst_y, width);
>@@ -879,6 +1017,22 @@ int ARGBToRGB24(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBTORGB24ROW_AVX2)
>+  if (TestCpuFlag(kCpuHasAVX2)) {
>+    ARGBToRGB24Row = ARGBToRGB24Row_Any_AVX2;
>+    if (IS_ALIGNED(width, 32)) {
>+      ARGBToRGB24Row = ARGBToRGB24Row_AVX2;
>+    }
>+  }
>+#endif
>+#if defined(HAS_ARGBTORGB24ROW_AVX512VBMI)
>+  if (TestCpuFlag(kCpuHasAVX512VBMI)) {
>+    ARGBToRGB24Row = ARGBToRGB24Row_Any_AVX512VBMI;
>+    if (IS_ALIGNED(width, 32)) {
>+      ARGBToRGB24Row = ARGBToRGB24Row_AVX512VBMI;
>+    }
>+  }
>+#endif
> #if defined(HAS_ARGBTORGB24ROW_NEON)
>   if (TestCpuFlag(kCpuHasNEON)) {
>     ARGBToRGB24Row = ARGBToRGB24Row_Any_NEON;
>@@ -895,6 +1049,14 @@ int ARGBToRGB24(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBTORGB24ROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToRGB24Row = ARGBToRGB24Row_Any_MMI;
>+    if (IS_ALIGNED(width, 4)) {
>+      ARGBToRGB24Row = ARGBToRGB24Row_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     ARGBToRGB24Row(src_argb, dst_rgb24, width);
>@@ -937,6 +1099,14 @@ int ARGBToRAW(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBTORAWROW_AVX2)
>+  if (TestCpuFlag(kCpuHasAVX2)) {
>+    ARGBToRAWRow = ARGBToRAWRow_Any_AVX2;
>+    if (IS_ALIGNED(width, 32)) {
>+      ARGBToRAWRow = ARGBToRAWRow_AVX2;
>+    }
>+  }
>+#endif
> #if defined(HAS_ARGBTORAWROW_NEON)
>   if (TestCpuFlag(kCpuHasNEON)) {
>     ARGBToRAWRow = ARGBToRAWRow_Any_NEON;
>@@ -953,6 +1123,14 @@ int ARGBToRAW(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBTORAWROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToRAWRow = ARGBToRAWRow_Any_MMI;
>+    if (IS_ALIGNED(width, 4)) {
>+      ARGBToRAWRow = ARGBToRAWRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     ARGBToRAWRow(src_argb, dst_raw, width);
>@@ -1023,11 +1201,19 @@ int ARGBToRGB565Dither(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBTORGB565DITHERROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToRGB565DitherRow = ARGBToRGB565DitherRow_Any_MMI;
>+    if (IS_ALIGNED(width, 4)) {
>+      ARGBToRGB565DitherRow = ARGBToRGB565DitherRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     ARGBToRGB565DitherRow(src_argb, dst_rgb565,
>-                          *(uint32_t*)(dither4x4 + ((y & 3) << 2)),  // NOLINT
>-                          width); /* NOLINT */
>+                          *(const uint32_t*)(dither4x4 + ((y & 3) << 2)),
>+                          width);
>     src_argb += src_stride_argb;
>     dst_rgb565 += dst_stride_rgb565;
>   }
>@@ -1092,6 +1278,14 @@ int ARGBToRGB565(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBTORGB565ROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToRGB565Row = ARGBToRGB565Row_Any_MMI;
>+    if (IS_ALIGNED(width, 4)) {
>+      ARGBToRGB565Row = ARGBToRGB565Row_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     ARGBToRGB565Row(src_argb, dst_rgb565, width);
>@@ -1158,6 +1352,14 @@ int ARGBToARGB1555(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBTOARGB1555ROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToARGB1555Row = ARGBToARGB1555Row_Any_MMI;
>+    if (IS_ALIGNED(width, 4)) {
>+      ARGBToARGB1555Row = ARGBToARGB1555Row_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     ARGBToARGB1555Row(src_argb, dst_argb1555, width);
>@@ -1224,6 +1426,14 @@ int ARGBToARGB4444(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBTOARGB4444ROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToARGB4444Row = ARGBToARGB4444Row_Any_MMI;
>+    if (IS_ALIGNED(width, 4)) {
>+      ARGBToARGB4444Row = ARGBToARGB4444Row_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     ARGBToARGB4444Row(src_argb, dst_argb4444, width);
>@@ -1400,6 +1610,14 @@ int ARGBToJ420(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBTOYJROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToYJRow = ARGBToYJRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      ARGBToYJRow = ARGBToYJRow_MMI;
>+    }
>+  }
>+#endif
> #if defined(HAS_ARGBTOUVJROW_MSA)
>   if (TestCpuFlag(kCpuHasMSA)) {
>     ARGBToUVJRow = ARGBToUVJRow_Any_MSA;
>@@ -1408,6 +1626,14 @@ int ARGBToJ420(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBTOUVJROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToUVJRow = ARGBToUVJRow_Any_MMI;
>+    if (IS_ALIGNED(width, 16)) {
>+      ARGBToUVJRow = ARGBToUVJRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height - 1; y += 2) {
>     ARGBToUVJRow(src_argb, src_stride_argb, dst_u, dst_v, width);
>@@ -1501,6 +1727,14 @@ int ARGBToJ422(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBTOYJROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToYJRow = ARGBToYJRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      ARGBToYJRow = ARGBToYJRow_MMI;
>+    }
>+  }
>+#endif
> #if defined(HAS_ARGBTOUVJROW_MSA)
>   if (TestCpuFlag(kCpuHasMSA)) {
>     ARGBToUVJRow = ARGBToUVJRow_Any_MSA;
>@@ -1509,6 +1743,14 @@ int ARGBToJ422(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBTOUVJROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToUVJRow = ARGBToUVJRow_Any_MMI;
>+    if (IS_ALIGNED(width, 16)) {
>+      ARGBToUVJRow = ARGBToUVJRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     ARGBToUVJRow(src_argb, 0, dst_u, dst_v, width);
>@@ -1578,6 +1820,14 @@ int ARGBToJ400(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBTOYJROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToYJRow = ARGBToYJRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      ARGBToYJRow = ARGBToYJRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     ARGBToYJRow(src_argb, dst_yj, width);
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert_jpeg.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert_jpeg.cc
>index ae3cc18cd247c2fe976ef7d7ebe48b8da67a6727..56a95c5770787accf1e76f66b0e9e3e5bd385133 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert_jpeg.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert_jpeg.cc
>@@ -190,7 +190,144 @@ int MJPGToI420(const uint8_t* sample,
>   return ret ? 0 : 1;
> }
> 
>-#ifdef HAVE_JPEG
>+struct NV21Buffers {
>+  uint8_t* y;
>+  int y_stride;
>+  uint8_t* vu;
>+  int vu_stride;
>+  int w;
>+  int h;
>+};
>+
>+static void JpegI420ToNV21(void* opaque,
>+                           const uint8_t* const* data,
>+                           const int* strides,
>+                           int rows) {
>+  NV21Buffers* dest = (NV21Buffers*)(opaque);
>+  I420ToNV21(data[0], strides[0], data[1], strides[1], data[2], strides[2],
>+             dest->y, dest->y_stride, dest->vu, dest->vu_stride, dest->w, rows);
>+  dest->y += rows * dest->y_stride;
>+  dest->vu += ((rows + 1) >> 1) * dest->vu_stride;
>+  dest->h -= rows;
>+}
>+
>+static void JpegI422ToNV21(void* opaque,
>+                           const uint8_t* const* data,
>+                           const int* strides,
>+                           int rows) {
>+  NV21Buffers* dest = (NV21Buffers*)(opaque);
>+  I422ToNV21(data[0], strides[0], data[1], strides[1], data[2], strides[2],
>+             dest->y, dest->y_stride, dest->vu, dest->vu_stride, dest->w, rows);
>+  dest->y += rows * dest->y_stride;
>+  dest->vu += ((rows + 1) >> 1) * dest->vu_stride;
>+  dest->h -= rows;
>+}
>+
>+static void JpegI444ToNV21(void* opaque,
>+                           const uint8_t* const* data,
>+                           const int* strides,
>+                           int rows) {
>+  NV21Buffers* dest = (NV21Buffers*)(opaque);
>+  I444ToNV21(data[0], strides[0], data[1], strides[1], data[2], strides[2],
>+             dest->y, dest->y_stride, dest->vu, dest->vu_stride, dest->w, rows);
>+  dest->y += rows * dest->y_stride;
>+  dest->vu += ((rows + 1) >> 1) * dest->vu_stride;
>+  dest->h -= rows;
>+}
>+
>+static void JpegI400ToNV21(void* opaque,
>+                           const uint8_t* const* data,
>+                           const int* strides,
>+                           int rows) {
>+  NV21Buffers* dest = (NV21Buffers*)(opaque);
>+  I400ToNV21(data[0], strides[0], dest->y, dest->y_stride, dest->vu,
>+             dest->vu_stride, dest->w, rows);
>+  dest->y += rows * dest->y_stride;
>+  dest->vu += ((rows + 1) >> 1) * dest->vu_stride;
>+  dest->h -= rows;
>+}
>+
>+// MJPG (Motion JPeg) to NV21
>+LIBYUV_API
>+int MJPGToNV21(const uint8_t* sample,
>+               size_t sample_size,
>+               uint8_t* dst_y,
>+               int dst_stride_y,
>+               uint8_t* dst_vu,
>+               int dst_stride_vu,
>+               int src_width,
>+               int src_height,
>+               int dst_width,
>+               int dst_height) {
>+  if (sample_size == kUnknownDataSize) {
>+    // ERROR: MJPEG frame size unknown
>+    return -1;
>+  }
>+
>+  // TODO(fbarchard): Port MJpeg to C.
>+  MJpegDecoder mjpeg_decoder;
>+  LIBYUV_BOOL ret = mjpeg_decoder.LoadFrame(sample, sample_size);
>+  if (ret && (mjpeg_decoder.GetWidth() != src_width ||
>+              mjpeg_decoder.GetHeight() != src_height)) {
>+    // ERROR: MJPEG frame has unexpected dimensions
>+    mjpeg_decoder.UnloadFrame();
>+    return 1;  // runtime failure
>+  }
>+  if (ret) {
>+    NV21Buffers bufs = {dst_y,         dst_stride_y, dst_vu,
>+                        dst_stride_vu, dst_width,    dst_height};
>+    // YUV420
>+    if (mjpeg_decoder.GetColorSpace() == MJpegDecoder::kColorSpaceYCbCr &&
>+        mjpeg_decoder.GetNumComponents() == 3 &&
>+        mjpeg_decoder.GetVertSampFactor(0) == 2 &&
>+        mjpeg_decoder.GetHorizSampFactor(0) == 2 &&
>+        mjpeg_decoder.GetVertSampFactor(1) == 1 &&
>+        mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
>+        mjpeg_decoder.GetVertSampFactor(2) == 1 &&
>+        mjpeg_decoder.GetHorizSampFactor(2) == 1) {
>+      ret = mjpeg_decoder.DecodeToCallback(&JpegI420ToNV21, &bufs, dst_width,
>+                                           dst_height);
>+      // YUV422
>+    } else if (mjpeg_decoder.GetColorSpace() ==
>+                   MJpegDecoder::kColorSpaceYCbCr &&
>+               mjpeg_decoder.GetNumComponents() == 3 &&
>+               mjpeg_decoder.GetVertSampFactor(0) == 1 &&
>+               mjpeg_decoder.GetHorizSampFactor(0) == 2 &&
>+               mjpeg_decoder.GetVertSampFactor(1) == 1 &&
>+               mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
>+               mjpeg_decoder.GetVertSampFactor(2) == 1 &&
>+               mjpeg_decoder.GetHorizSampFactor(2) == 1) {
>+      ret = mjpeg_decoder.DecodeToCallback(&JpegI422ToNV21, &bufs, dst_width,
>+                                           dst_height);
>+      // YUV444
>+    } else if (mjpeg_decoder.GetColorSpace() ==
>+                   MJpegDecoder::kColorSpaceYCbCr &&
>+               mjpeg_decoder.GetNumComponents() == 3 &&
>+               mjpeg_decoder.GetVertSampFactor(0) == 1 &&
>+               mjpeg_decoder.GetHorizSampFactor(0) == 1 &&
>+               mjpeg_decoder.GetVertSampFactor(1) == 1 &&
>+               mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
>+               mjpeg_decoder.GetVertSampFactor(2) == 1 &&
>+               mjpeg_decoder.GetHorizSampFactor(2) == 1) {
>+      ret = mjpeg_decoder.DecodeToCallback(&JpegI444ToNV21, &bufs, dst_width,
>+                                           dst_height);
>+      // YUV400
>+    } else if (mjpeg_decoder.GetColorSpace() ==
>+                   MJpegDecoder::kColorSpaceGrayscale &&
>+               mjpeg_decoder.GetNumComponents() == 1 &&
>+               mjpeg_decoder.GetVertSampFactor(0) == 1 &&
>+               mjpeg_decoder.GetHorizSampFactor(0) == 1) {
>+      ret = mjpeg_decoder.DecodeToCallback(&JpegI400ToNV21, &bufs, dst_width,
>+                                           dst_height);
>+    } else {
>+      // Unknown colorspace.
>+      mjpeg_decoder.UnloadFrame();
>+      return 1;
>+    }
>+  }
>+  return ret ? 0 : 1;
>+}
>+
> struct ARGBBuffers {
>   uint8_t* argb;
>   int argb_stride;
>@@ -322,9 +459,8 @@ int MJPGToARGB(const uint8_t* sample,
>   }
>   return ret ? 0 : 1;
> }
>-#endif
> 
>-#endif
>+#endif  // HAVE_JPEG
> 
> #ifdef __cplusplus
> }  // extern "C"
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert_to_argb.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert_to_argb.cc
>index 677e5d56fcc98481af504b10a2708435a6cebf9d..bde1aa8891bc87a20182c0f31dde1ee4c50505fe 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert_to_argb.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert_to_argb.cc
>@@ -28,6 +28,14 @@ extern "C" {
> // src_height is used to compute location of planes, and indicate inversion
> // sample_size is measured in bytes and is the size of the frame.
> //   With MJPEG it is the compressed size of the frame.
>+
>+// TODO(fbarchard): Add the following:
>+// H010ToARGB
>+// I010ToARGB
>+// J400ToARGB
>+// J422ToARGB
>+// J444ToARGB
>+
> LIBYUV_API
> int ConvertToARGB(const uint8_t* sample,
>                   size_t sample_size,
>@@ -123,6 +131,16 @@ int ConvertToARGB(const uint8_t* sample,
>       r = RGBAToARGB(src, src_width * 4, dst_argb, dst_stride_argb, crop_width,
>                      inv_crop_height);
>       break;
>+    case FOURCC_AR30:
>+      src = sample + (src_width * crop_y + crop_x) * 4;
>+      r = AR30ToARGB(src, src_width * 4, dst_argb, dst_stride_argb, crop_width,
>+                     inv_crop_height);
>+      break;
>+    case FOURCC_AB30:
>+      src = sample + (src_width * crop_y + crop_x) * 4;
>+      r = AB30ToARGB(src, src_width * 4, dst_argb, dst_stride_argb, crop_width,
>+                     inv_crop_height);
>+      break;
>     case FOURCC_RGBP:
>       src = sample + (src_width * crop_y + crop_x) * 2;
>       r = RGB565ToARGB(src, src_width * 2, dst_argb, dst_stride_argb,
>@@ -147,13 +165,15 @@ int ConvertToARGB(const uint8_t* sample,
>     // Biplanar formats
>     case FOURCC_NV12:
>       src = sample + (src_width * crop_y + crop_x);
>-      src_uv = sample + aligned_src_width * (src_height + crop_y / 2) + crop_x;
>+      src_uv =
>+          sample + aligned_src_width * (abs_src_height + crop_y / 2) + crop_x;
>       r = NV12ToARGB(src, src_width, src_uv, aligned_src_width, dst_argb,
>                      dst_stride_argb, crop_width, inv_crop_height);
>       break;
>     case FOURCC_NV21:
>       src = sample + (src_width * crop_y + crop_x);
>-      src_uv = sample + aligned_src_width * (src_height + crop_y / 2) + crop_x;
>+      src_uv =
>+          sample + aligned_src_width * (abs_src_height + crop_y / 2) + crop_x;
>       // Call NV12 but with u and v parameters swapped.
>       r = NV21ToARGB(src, src_width, src_uv, aligned_src_width, dst_argb,
>                      dst_stride_argb, crop_width, inv_crop_height);
>@@ -163,6 +183,7 @@ int ConvertToARGB(const uint8_t* sample,
>       r = M420ToARGB(src, src_width, dst_argb, dst_stride_argb, crop_width,
>                      inv_crop_height);
>       break;
>+
>     // Triplanar formats
>     case FOURCC_I420:
>     case FOURCC_YV12: {
>@@ -187,16 +208,27 @@ int ConvertToARGB(const uint8_t* sample,
>       break;
>     }
> 
>-    case FOURCC_J420: {
>+    case FOURCC_H420: {
>+      int halfwidth = (src_width + 1) / 2;
>+      int halfheight = (abs_src_height + 1) / 2;
>       const uint8_t* src_y = sample + (src_width * crop_y + crop_x);
>-      const uint8_t* src_u;
>-      const uint8_t* src_v;
>+      const uint8_t* src_u = sample + src_width * abs_src_height +
>+                             (halfwidth * crop_y + crop_x) / 2;
>+      const uint8_t* src_v = sample + src_width * abs_src_height +
>+                             halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
>+      r = H420ToARGB(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
>+                     dst_argb, dst_stride_argb, crop_width, inv_crop_height);
>+      break;
>+    }
>+
>+    case FOURCC_J420: {
>       int halfwidth = (src_width + 1) / 2;
>       int halfheight = (abs_src_height + 1) / 2;
>-      src_u = sample + src_width * abs_src_height +
>-              (halfwidth * crop_y + crop_x) / 2;
>-      src_v = sample + src_width * abs_src_height +
>-              halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
>+      const uint8_t* src_y = sample + (src_width * crop_y + crop_x);
>+      const uint8_t* src_u = sample + src_width * abs_src_height +
>+                             (halfwidth * crop_y + crop_x) / 2;
>+      const uint8_t* src_v = sample + src_width * abs_src_height +
>+                             halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
>       r = J420ToARGB(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
>                      dst_argb, dst_stride_argb, crop_width, inv_crop_height);
>       break;
>@@ -204,10 +236,10 @@ int ConvertToARGB(const uint8_t* sample,
> 
>     case FOURCC_I422:
>     case FOURCC_YV16: {
>+      int halfwidth = (src_width + 1) / 2;
>       const uint8_t* src_y = sample + src_width * crop_y + crop_x;
>       const uint8_t* src_u;
>       const uint8_t* src_v;
>-      int halfwidth = (src_width + 1) / 2;
>       if (format == FOURCC_YV16) {
>         src_v = sample + src_width * abs_src_height + halfwidth * crop_y +
>                 crop_x / 2;
>@@ -223,6 +255,19 @@ int ConvertToARGB(const uint8_t* sample,
>                      dst_argb, dst_stride_argb, crop_width, inv_crop_height);
>       break;
>     }
>+
>+    case FOURCC_H422: {
>+      int halfwidth = (src_width + 1) / 2;
>+      const uint8_t* src_y = sample + src_width * crop_y + crop_x;
>+      const uint8_t* src_u =
>+          sample + src_width * abs_src_height + halfwidth * crop_y + crop_x / 2;
>+      const uint8_t* src_v = sample + src_width * abs_src_height +
>+                             halfwidth * (abs_src_height + crop_y) + crop_x / 2;
>+      r = H422ToARGB(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
>+                     dst_argb, dst_stride_argb, crop_width, inv_crop_height);
>+      break;
>+    }
>+
>     case FOURCC_I444:
>     case FOURCC_YV24: {
>       const uint8_t* src_y = sample + src_width * crop_y + crop_x;
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert_to_i420.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert_to_i420.cc
>index 1bed9d6440db92a696b2167c88873963e4c29f05..584be0ac33aded539ebd4ea26fa43b002522e2dd 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert_to_i420.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/convert_to_i420.cc
>@@ -155,6 +155,7 @@ int ConvertToI420(const uint8_t* sample,
>                      dst_stride_u, dst_v, dst_stride_v, crop_width,
>                      inv_crop_height);
>       break;
>+    // TODO(fbarchard): Add AR30 and AB30
>     case FOURCC_I400:
>       src = sample + src_width * crop_y + crop_x;
>       r = I400ToI420(src, src_width, dst_y, dst_stride_y, dst_u, dst_stride_u,
>@@ -163,7 +164,7 @@ int ConvertToI420(const uint8_t* sample,
>     // Biplanar formats
>     case FOURCC_NV12:
>       src = sample + (src_width * crop_y + crop_x);
>-      src_uv = sample + (src_width * src_height) +
>+      src_uv = sample + (src_width * abs_src_height) +
>                ((crop_y / 2) * aligned_src_width) + ((crop_x / 2) * 2);
>       r = NV12ToI420Rotate(src, src_width, src_uv, aligned_src_width, dst_y,
>                            dst_stride_y, dst_u, dst_stride_u, dst_v,
>@@ -171,7 +172,7 @@ int ConvertToI420(const uint8_t* sample,
>       break;
>     case FOURCC_NV21:
>       src = sample + (src_width * crop_y + crop_x);
>-      src_uv = sample + (src_width * src_height) +
>+      src_uv = sample + (src_width * abs_src_height) +
>                ((crop_y / 2) * aligned_src_width) + ((crop_x / 2) * 2);
>       // Call NV12 but with dst_u and dst_v parameters swapped.
>       r = NV12ToI420Rotate(src, src_width, src_uv, aligned_src_width, dst_y,
>@@ -192,15 +193,15 @@ int ConvertToI420(const uint8_t* sample,
>       int halfwidth = (src_width + 1) / 2;
>       int halfheight = (abs_src_height + 1) / 2;
>       if (format == FOURCC_YV12) {
>-        src_v = sample + src_width * abs_src_height +
>-                (halfwidth * crop_y + crop_x) / 2;
>+        src_v = sample + src_width * abs_src_height + halfwidth * (crop_y / 2) +
>+                (crop_x / 2);
>         src_u = sample + src_width * abs_src_height +
>-                halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
>+                halfwidth * (halfheight + (crop_y / 2)) + (crop_x / 2);
>       } else {
>-        src_u = sample + src_width * abs_src_height +
>-                (halfwidth * crop_y + crop_x) / 2;
>+        src_u = sample + src_width * abs_src_height + halfwidth * (crop_y / 2) +
>+                (crop_x / 2);
>         src_v = sample + src_width * abs_src_height +
>-                halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
>+                halfwidth * (halfheight + (crop_y / 2)) + (crop_x / 2);
>       }
>       r = I420Rotate(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
>                      dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v,
>@@ -215,14 +216,14 @@ int ConvertToI420(const uint8_t* sample,
>       int halfwidth = (src_width + 1) / 2;
>       if (format == FOURCC_YV16) {
>         src_v = sample + src_width * abs_src_height + halfwidth * crop_y +
>-                crop_x / 2;
>+                (crop_x / 2);
>         src_u = sample + src_width * abs_src_height +
>-                halfwidth * (abs_src_height + crop_y) + crop_x / 2;
>+                halfwidth * (abs_src_height + crop_y) + (crop_x / 2);
>       } else {
>         src_u = sample + src_width * abs_src_height + halfwidth * crop_y +
>-                crop_x / 2;
>+                (crop_x / 2);
>         src_v = sample + src_width * abs_src_height +
>-                halfwidth * (abs_src_height + crop_y) + crop_x / 2;
>+                halfwidth * (abs_src_height + crop_y) + (crop_x / 2);
>       }
>       r = I422ToI420(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
>                      dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v,
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/cpu_id.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/cpu_id.cc
>index 446aad12078b4e4a963ff58335efb8018c32c713..48e2b61526d1325136d738425b1f34993d7352d9 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/cpu_id.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/cpu_id.cc
>@@ -19,10 +19,6 @@
> #include <immintrin.h>  // For _xgetbv()
> #endif
> 
>-#if !defined(__native_client__)
>-#include <stdlib.h>  // For getenv()
>-#endif
>-
> // For ArmCpuCaps() but unittested on all platforms
> #include <stdio.h>
> #include <string.h>
>@@ -177,6 +173,9 @@ LIBYUV_API SAFEBUFFERS int MipsCpuCaps(const char* cpuinfo_name,
>     if (strcmp(ase, " msa") == 0) {
>       return kCpuHasMSA;
>     }
>+    if (strcmp(ase, " mmi") == 0) {
>+      return kCpuHasMMI;
>+    }
>     return 0;
>   }
>   while (fgets(cpuinfo_line, sizeof(cpuinfo_line) - 1, f)) {
>@@ -189,31 +188,21 @@ LIBYUV_API SAFEBUFFERS int MipsCpuCaps(const char* cpuinfo_name,
>         }
>         return 0;
>       }
>+    } else if (memcmp(cpuinfo_line, "cpu model", 9) == 0) {
>+      char* p = strstr(cpuinfo_line, "Loongson-3");
>+      if (p) {
>+        fclose(f);
>+        if (strcmp(ase, " mmi") == 0) {
>+          return kCpuHasMMI;
>+        }
>+        return 0;
>+      }
>     }
>   }
>   fclose(f);
>   return 0;
> }
> 
>-// Test environment variable for disabling CPU features. Any non-zero value
>-// to disable. Zero ignored to make it easy to set the variable on/off.
>-#if !defined(__native_client__) && !defined(_M_ARM)
>-
>-static LIBYUV_BOOL TestEnv(const char* name) {
>-  const char* var = getenv(name);
>-  if (var) {
>-    if (var[0] != '0') {
>-      return LIBYUV_TRUE;
>-    }
>-  }
>-  return LIBYUV_FALSE;
>-}
>-#else  // nacl does not support getenv().
>-static LIBYUV_BOOL TestEnv(const char*) {
>-  return LIBYUV_FALSE;
>-}
>-#endif
>-
> static SAFEBUFFERS int GetCpuFlags(void) {
>   int cpu_info = 0;
> #if !defined(__pnacl__) && !defined(__CLR_VER) &&                   \
>@@ -251,52 +240,14 @@ static SAFEBUFFERS int GetCpuFlags(void) {
>       cpu_info |= (cpu_info7[2] & 0x00000100) ? kCpuHasGFNI : 0;
>     }
>   }
>-
>-  // TODO(fbarchard): Consider moving these to gtest
>-  // Environment variable overrides for testing.
>-  if (TestEnv("LIBYUV_DISABLE_X86")) {
>-    cpu_info &= ~kCpuHasX86;
>-  }
>-  if (TestEnv("LIBYUV_DISABLE_SSE2")) {
>-    cpu_info &= ~kCpuHasSSE2;
>-  }
>-  if (TestEnv("LIBYUV_DISABLE_SSSE3")) {
>-    cpu_info &= ~kCpuHasSSSE3;
>-  }
>-  if (TestEnv("LIBYUV_DISABLE_SSE41")) {
>-    cpu_info &= ~kCpuHasSSE41;
>-  }
>-  if (TestEnv("LIBYUV_DISABLE_SSE42")) {
>-    cpu_info &= ~kCpuHasSSE42;
>-  }
>-  if (TestEnv("LIBYUV_DISABLE_AVX")) {
>-    cpu_info &= ~kCpuHasAVX;
>-  }
>-  if (TestEnv("LIBYUV_DISABLE_AVX2")) {
>-    cpu_info &= ~kCpuHasAVX2;
>-  }
>-  if (TestEnv("LIBYUV_DISABLE_ERMS")) {
>-    cpu_info &= ~kCpuHasERMS;
>-  }
>-  if (TestEnv("LIBYUV_DISABLE_FMA3")) {
>-    cpu_info &= ~kCpuHasFMA3;
>-  }
>-  if (TestEnv("LIBYUV_DISABLE_F16C")) {
>-    cpu_info &= ~kCpuHasF16C;
>-  }
>-  if (TestEnv("LIBYUV_DISABLE_AVX512BW")) {
>-    cpu_info &= ~kCpuHasAVX512BW;
>-  }
>-
> #endif
> #if defined(__mips__) && defined(__linux__)
> #if defined(__mips_msa)
>   cpu_info = MipsCpuCaps("/proc/cpuinfo", " msa");
>+#elif defined(_MIPS_ARCH_LOONGSON3A)
>+  cpu_info = MipsCpuCaps("/proc/cpuinfo", " mmi");
> #endif
>   cpu_info |= kCpuHasMIPS;
>-  if (getenv("LIBYUV_DISABLE_MSA")) {
>-    cpu_info &= ~kCpuHasMSA;
>-  }
> #endif
> #if defined(__arm__) || defined(__aarch64__)
> // gcc -mfpu=neon defines __ARM_NEON__
>@@ -315,13 +266,7 @@ static SAFEBUFFERS int GetCpuFlags(void) {
>   cpu_info = ArmCpuCaps("/proc/cpuinfo");
> #endif
>   cpu_info |= kCpuHasARM;
>-  if (TestEnv("LIBYUV_DISABLE_NEON")) {
>-    cpu_info &= ~kCpuHasNEON;
>-  }
> #endif  // __arm__
>-  if (TestEnv("LIBYUV_DISABLE_ASM")) {
>-    cpu_info = 0;
>-  }
>   cpu_info |= kCpuInitialized;
>   return cpu_info;
> }
>@@ -330,11 +275,7 @@ static SAFEBUFFERS int GetCpuFlags(void) {
> LIBYUV_API
> int MaskCpuFlags(int enable_flags) {
>   int cpu_info = GetCpuFlags() & enable_flags;
>-#ifdef __ATOMIC_RELAXED
>-  __atomic_store_n(&cpu_info_, cpu_info, __ATOMIC_RELAXED);
>-#else
>-  cpu_info_ = cpu_info;
>-#endif
>+  SetCpuFlags(cpu_info);
>   return cpu_info;
> }
> 
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/planar_functions.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/planar_functions.cc
>index 5eae3f763a7290aa83b0031aa6725bbb367a6f18..b49bf0a0b4b6e34cf8464db6e4ce1da3318a43ec 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/planar_functions.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/planar_functions.cc
>@@ -410,6 +410,14 @@ void SplitUVPlane(const uint8_t* src_uv,
>     }
>   }
> #endif
>+#if defined(HAS_SPLITUVROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    SplitUVRow = SplitUVRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      SplitUVRow = SplitUVRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     // Copy a row of UV.
>@@ -478,6 +486,14 @@ void MergeUVPlane(const uint8_t* src_u,
>     }
>   }
> #endif
>+#if defined(HAS_MERGEUVROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    MergeUVRow = MergeUVRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      MergeUVRow = MergeUVRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     // Merge a row of U and V into a row of UV.
>@@ -537,6 +553,14 @@ void SplitRGBPlane(const uint8_t* src_rgb,
>     }
>   }
> #endif
>+#if defined(HAS_SPLITRGBROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    SplitRGBRow = SplitRGBRow_Any_MMI;
>+    if (IS_ALIGNED(width, 4)) {
>+      SplitRGBRow = SplitRGBRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     // Copy a row of RGB.
>@@ -593,6 +617,14 @@ void MergeRGBPlane(const uint8_t* src_r,
>     }
>   }
> #endif
>+#if defined(HAS_MERGERGBROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    MergeRGBRow = MergeRGBRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      MergeRGBRow = MergeRGBRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     // Merge a row of U and V into a row of RGB.
>@@ -651,6 +683,14 @@ void MirrorPlane(const uint8_t* src_y,
>     }
>   }
> #endif
>+#if defined(HAS_MIRRORROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    MirrorRow = MirrorRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      MirrorRow = MirrorRow_MMI;
>+    }
>+  }
>+#endif
> 
>   // Mirror plane
>   for (y = 0; y < height; ++y) {
>@@ -734,6 +774,16 @@ int YUY2ToI422(const uint8_t* src_yuy2,
>     }
>   }
> #endif
>+#if defined(HAS_YUY2TOYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    YUY2ToYRow = YUY2ToYRow_Any_MMI;
>+    YUY2ToUV422Row = YUY2ToUV422Row_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      YUY2ToYRow = YUY2ToYRow_MMI;
>+      YUY2ToUV422Row = YUY2ToUV422Row_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     YUY2ToUV422Row(src_yuy2, dst_u, dst_v, width);
>@@ -820,6 +870,16 @@ int UYVYToI422(const uint8_t* src_uyvy,
>     }
>   }
> #endif
>+#if defined(HAS_UYVYTOYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    UYVYToYRow = UYVYToYRow_Any_MMI;
>+    UYVYToUV422Row = UYVYToUV422Row_Any_MMI;
>+    if (IS_ALIGNED(width, 16)) {
>+      UYVYToYRow = UYVYToYRow_MMI;
>+      UYVYToUV422Row = UYVYToUV422Row_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     UYVYToUV422Row(src_uyvy, dst_u, dst_v, width);
>@@ -890,6 +950,14 @@ int YUY2ToY(const uint8_t* src_yuy2,
>     }
>   }
> #endif
>+#if defined(HAS_YUY2TOYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    YUY2ToYRow = YUY2ToYRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      YUY2ToYRow = YUY2ToYRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     YUY2ToYRow(src_yuy2, dst_y, width);
>@@ -1015,6 +1083,14 @@ int ARGBMirror(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBMIRRORROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBMirrorRow = ARGBMirrorRow_Any_MMI;
>+    if (IS_ALIGNED(width, 2)) {
>+      ARGBMirrorRow = ARGBMirrorRow_MMI;
>+    }
>+  }
>+#endif
> 
>   // Mirror plane
>   for (y = 0; y < height; ++y) {
>@@ -1047,6 +1123,11 @@ ARGBBlendRow GetARGBBlend() {
>   if (TestCpuFlag(kCpuHasMSA)) {
>     ARGBBlendRow = ARGBBlendRow_MSA;
>   }
>+#endif
>+#if defined(HAS_ARGBBLENDROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBBlendRow = ARGBBlendRow_MMI;
>+  }
> #endif
>   return ARGBBlendRow;
> }
>@@ -1140,6 +1221,14 @@ int BlendPlane(const uint8_t* src_y0,
>     }
>   }
> #endif
>+#if defined(HAS_BLENDPLANEROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    BlendPlaneRow = BlendPlaneRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      BlendPlaneRow = BlendPlaneRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     BlendPlaneRow(src_y0, src_y1, alpha, dst_y, width);
>@@ -1215,6 +1304,14 @@ int I420Blend(const uint8_t* src_y0,
>       BlendPlaneRow = BlendPlaneRow_AVX2;
>     }
>   }
>+#endif
>+#if defined(HAS_BLENDPLANEROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    BlendPlaneRow = BlendPlaneRow_Any_MMI;
>+    if (IS_ALIGNED(halfwidth, 8)) {
>+      BlendPlaneRow = BlendPlaneRow_MMI;
>+    }
>+  }
> #endif
>   if (!IS_ALIGNED(width, 2)) {
>     ScaleRowDown2 = ScaleRowDown2Box_Odd_C;
>@@ -1252,6 +1349,17 @@ int I420Blend(const uint8_t* src_y0,
>     }
>   }
> #endif
>+#if defined(HAS_SCALEROWDOWN2_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ScaleRowDown2 = ScaleRowDown2Box_Odd_MMI;
>+    if (IS_ALIGNED(width, 2)) {
>+      ScaleRowDown2 = ScaleRowDown2Box_Any_MMI;
>+      if (IS_ALIGNED(halfwidth, 8)) {
>+        ScaleRowDown2 = ScaleRowDown2Box_MMI;
>+      }
>+    }
>+  }
>+#endif
> 
>   // Row buffer for intermediate alpha pixels.
>   align_buffer_64(halfalpha, halfwidth);
>@@ -1337,6 +1445,14 @@ int ARGBMultiply(const uint8_t* src_argb0,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBMULTIPLYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBMultiplyRow = ARGBMultiplyRow_Any_MMI;
>+    if (IS_ALIGNED(width, 2)) {
>+      ARGBMultiplyRow = ARGBMultiplyRow_MMI;
>+    }
>+  }
>+#endif
> 
>   // Multiply plane
>   for (y = 0; y < height; ++y) {
>@@ -1414,6 +1530,14 @@ int ARGBAdd(const uint8_t* src_argb0,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBADDROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBAddRow = ARGBAddRow_Any_MMI;
>+    if (IS_ALIGNED(width, 2)) {
>+      ARGBAddRow = ARGBAddRow_MMI;
>+    }
>+  }
>+#endif
> 
>   // Add plane
>   for (y = 0; y < height; ++y) {
>@@ -1486,6 +1610,14 @@ int ARGBSubtract(const uint8_t* src_argb0,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBSUBTRACTROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBSubtractRow = ARGBSubtractRow_Any_MMI;
>+    if (IS_ALIGNED(width, 2)) {
>+      ARGBSubtractRow = ARGBSubtractRow_MMI;
>+    }
>+  }
>+#endif
> 
>   // Subtract plane
>   for (y = 0; y < height; ++y) {
>@@ -1718,6 +1850,14 @@ int RAWToRGB24(const uint8_t* src_raw,
>     }
>   }
> #endif
>+#if defined(HAS_RAWTORGB24ROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    RAWToRGB24Row = RAWToRGB24Row_Any_MMI;
>+    if (IS_ALIGNED(width, 4)) {
>+      RAWToRGB24Row = RAWToRGB24Row_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     RAWToRGB24Row(src_raw, dst_rgb24, width);
>@@ -1939,6 +2079,14 @@ int ARGBAttenuate(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBATTENUATEROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBAttenuateRow = ARGBAttenuateRow_Any_MMI;
>+    if (IS_ALIGNED(width, 2)) {
>+      ARGBAttenuateRow = ARGBAttenuateRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     ARGBAttenuateRow(src_argb, dst_argb, width);
>@@ -2039,6 +2187,11 @@ int ARGBGrayTo(const uint8_t* src_argb,
>     ARGBGrayRow = ARGBGrayRow_MSA;
>   }
> #endif
>+#if defined(HAS_ARGBGRAYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI) && IS_ALIGNED(width, 2)) {
>+    ARGBGrayRow = ARGBGrayRow_MMI;
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     ARGBGrayRow(src_argb, dst_argb, width);
>@@ -2084,6 +2237,11 @@ int ARGBGray(uint8_t* dst_argb,
>     ARGBGrayRow = ARGBGrayRow_MSA;
>   }
> #endif
>+#if defined(HAS_ARGBGRAYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI) && IS_ALIGNED(width, 2)) {
>+    ARGBGrayRow = ARGBGrayRow_MMI;
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     ARGBGrayRow(dst, dst, width);
>@@ -2127,6 +2285,11 @@ int ARGBSepia(uint8_t* dst_argb,
>     ARGBSepiaRow = ARGBSepiaRow_MSA;
>   }
> #endif
>+#if defined(HAS_ARGBSEPIAROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI) && IS_ALIGNED(width, 2)) {
>+    ARGBSepiaRow = ARGBSepiaRow_MMI;
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     ARGBSepiaRow(dst, width);
>@@ -2177,6 +2340,11 @@ int ARGBColorMatrix(const uint8_t* src_argb,
>   if (TestCpuFlag(kCpuHasMSA) && IS_ALIGNED(width, 8)) {
>     ARGBColorMatrixRow = ARGBColorMatrixRow_MSA;
>   }
>+#endif
>+#if defined(HAS_ARGBCOLORMATRIXROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI) && IS_ALIGNED(width, 2)) {
>+    ARGBColorMatrixRow = ARGBColorMatrixRow_MMI;
>+  }
> #endif
>   for (y = 0; y < height; ++y) {
>     ARGBColorMatrixRow(src_argb, dst_argb, matrix_argb, width);
>@@ -2372,6 +2540,12 @@ int ARGBComputeCumulativeSum(const uint8_t* src_argb,
>     ComputeCumulativeSumRow = ComputeCumulativeSumRow_SSE2;
>   }
> #endif
>+#if defined(HAS_CUMULATIVESUMTOAVERAGEROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ComputeCumulativeSumRow = ComputeCumulativeSumRow_MMI;
>+  }
>+#endif
>+
>   memset(dst_cumsum, 0, width * sizeof(dst_cumsum[0]) * 4);  // 4 int per pixel.
>   for (y = 0; y < height; ++y) {
>     ComputeCumulativeSumRow(src_argb, dst_cumsum, previous_cumsum, width);
>@@ -2429,6 +2603,11 @@ int ARGBBlur(const uint8_t* src_argb,
>     ComputeCumulativeSumRow = ComputeCumulativeSumRow_SSE2;
>     CumulativeSumToAverageRow = CumulativeSumToAverageRow_SSE2;
>   }
>+#endif
>+#if defined(HAS_CUMULATIVESUMTOAVERAGEROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ComputeCumulativeSumRow = ComputeCumulativeSumRow_MMI;
>+  }
> #endif
>   // Compute enough CumulativeSum for first row to be blurred. After this
>   // one row of CumulativeSum is updated at a time.
>@@ -2536,6 +2715,11 @@ int ARGBShade(const uint8_t* src_argb,
>     ARGBShadeRow = ARGBShadeRow_MSA;
>   }
> #endif
>+#if defined(HAS_ARGBSHADEROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI) && IS_ALIGNED(width, 2)) {
>+    ARGBShadeRow = ARGBShadeRow_MMI;
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     ARGBShadeRow(src_argb, dst_argb, width, value);
>@@ -2607,6 +2791,14 @@ int InterpolatePlane(const uint8_t* src0,
>     }
>   }
> #endif
>+#if defined(HAS_INTERPOLATEROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    InterpolateRow = InterpolateRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      InterpolateRow = InterpolateRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     InterpolateRow(dst, src0, src1 - src0, width, interpolation);
>@@ -2730,6 +2922,14 @@ int ARGBShuffle(const uint8_t* src_bgra,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBSHUFFLEROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBShuffleRow = ARGBShuffleRow_Any_MMI;
>+    if (IS_ALIGNED(width, 2)) {
>+      ARGBShuffleRow = ARGBShuffleRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     ARGBShuffleRow(src_bgra, dst_argb, shuffler, width);
>@@ -2801,6 +3001,14 @@ static int ARGBSobelize(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBTOYJROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBToYJRow = ARGBToYJRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      ARGBToYJRow = ARGBToYJRow_MMI;
>+    }
>+  }
>+#endif
> 
> #if defined(HAS_SOBELYROW_SSE2)
>   if (TestCpuFlag(kCpuHasSSE2)) {
>@@ -2817,6 +3025,11 @@ static int ARGBSobelize(const uint8_t* src_argb,
>     SobelYRow = SobelYRow_MSA;
>   }
> #endif
>+#if defined(HAS_SOBELYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    SobelYRow = SobelYRow_MMI;
>+  }
>+#endif
> #if defined(HAS_SOBELXROW_SSE2)
>   if (TestCpuFlag(kCpuHasSSE2)) {
>     SobelXRow = SobelXRow_SSE2;
>@@ -2831,6 +3044,11 @@ static int ARGBSobelize(const uint8_t* src_argb,
>   if (TestCpuFlag(kCpuHasMSA)) {
>     SobelXRow = SobelXRow_MSA;
>   }
>+#endif
>+#if defined(HAS_SOBELXROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    SobelXRow = SobelXRow_MMI;
>+  }
> #endif
>   {
>     // 3 rows with edges before/after.
>@@ -2913,6 +3131,14 @@ int ARGBSobel(const uint8_t* src_argb,
>       SobelRow = SobelRow_MSA;
>     }
>   }
>+#endif
>+#if defined(HAS_SOBELROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    SobelRow = SobelRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      SobelRow = SobelRow_MMI;
>+    }
>+  }
> #endif
>   return ARGBSobelize(src_argb, src_stride_argb, dst_argb, dst_stride_argb,
>                       width, height, SobelRow);
>@@ -2951,6 +3177,14 @@ int ARGBSobelToPlane(const uint8_t* src_argb,
>       SobelToPlaneRow = SobelToPlaneRow_MSA;
>     }
>   }
>+#endif
>+#if defined(HAS_SOBELTOPLANEROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    SobelToPlaneRow = SobelToPlaneRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      SobelToPlaneRow = SobelToPlaneRow_MMI;
>+    }
>+  }
> #endif
>   return ARGBSobelize(src_argb, src_stride_argb, dst_y, dst_stride_y, width,
>                       height, SobelToPlaneRow);
>@@ -2990,6 +3224,14 @@ int ARGBSobelXY(const uint8_t* src_argb,
>       SobelXYRow = SobelXYRow_MSA;
>     }
>   }
>+#endif
>+#if defined(HAS_SOBELXYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    SobelXYRow = SobelXYRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      SobelXYRow = SobelXYRow_MMI;
>+    }
>+  }
> #endif
>   return ARGBSobelize(src_argb, src_stride_argb, dst_argb, dst_stride_argb,
>                       width, height, SobelXYRow);
>@@ -3228,6 +3470,14 @@ int ARGBCopyAlpha(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBCOPYALPHAROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBCopyAlphaRow = ARGBCopyAlphaRow_Any_MMI;
>+    if (IS_ALIGNED(width, 2)) {
>+      ARGBCopyAlphaRow = ARGBCopyAlphaRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     ARGBCopyAlphaRow(src_argb, dst_argb, width);
>@@ -3286,6 +3536,12 @@ int ARGBExtractAlpha(const uint8_t* src_argb,
>                                                 : ARGBExtractAlphaRow_Any_MSA;
>   }
> #endif
>+#if defined(HAS_ARGBEXTRACTALPHAROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBExtractAlphaRow = IS_ALIGNED(width, 8) ? ARGBExtractAlphaRow_MMI
>+                                               : ARGBExtractAlphaRow_Any_MMI;
>+  }
>+#endif
> 
>   for (int y = 0; y < height; ++y) {
>     ARGBExtractAlphaRow(src_argb, dst_a, width);
>@@ -3337,6 +3593,14 @@ int ARGBCopyYToAlpha(const uint8_t* src_y,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBCOPYYTOALPHAROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBCopyYToAlphaRow = ARGBCopyYToAlphaRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      ARGBCopyYToAlphaRow = ARGBCopyYToAlphaRow_MMI;
>+    }
>+  }
>+#endif
> 
>   for (y = 0; y < height; ++y) {
>     ARGBCopyYToAlphaRow(src_y, dst_argb, width);
>@@ -3406,6 +3670,14 @@ int YUY2ToNV12(const uint8_t* src_yuy2,
>     }
>   }
> #endif
>+#if defined(HAS_SPLITUVROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    SplitUVRow = SplitUVRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      SplitUVRow = SplitUVRow_MMI;
>+    }
>+  }
>+#endif
> #if defined(HAS_INTERPOLATEROW_SSSE3)
>   if (TestCpuFlag(kCpuHasSSSE3)) {
>     InterpolateRow = InterpolateRow_Any_SSSE3;
>@@ -3438,6 +3710,14 @@ int YUY2ToNV12(const uint8_t* src_yuy2,
>     }
>   }
> #endif
>+#if defined(HAS_INTERPOLATEROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    InterpolateRow = InterpolateRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      InterpolateRow = InterpolateRow_MMI;
>+    }
>+  }
>+#endif
> 
>   {
>     int awidth = halfwidth * 2;
>@@ -3522,6 +3802,14 @@ int UYVYToNV12(const uint8_t* src_uyvy,
>     }
>   }
> #endif
>+#if defined(HAS_SPLITUVROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    SplitUVRow = SplitUVRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      SplitUVRow = SplitUVRow_MMI;
>+    }
>+  }
>+#endif
> #if defined(HAS_INTERPOLATEROW_SSSE3)
>   if (TestCpuFlag(kCpuHasSSSE3)) {
>     InterpolateRow = InterpolateRow_Any_SSSE3;
>@@ -3554,6 +3842,14 @@ int UYVYToNV12(const uint8_t* src_uyvy,
>     }
>   }
> #endif
>+#if defined(HAS_INTERPOLATEROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    InterpolateRow = InterpolateRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      InterpolateRow = InterpolateRow_MMI;
>+    }
>+  }
>+#endif
> 
>   {
>     int awidth = halfwidth * 2;
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/rotate.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/rotate.cc
>index f2bed85b7552f250d419ec2889c1920ba540298a..f28a06d38a9756674edfcab0f68cd405e33bd4c3 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/rotate.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/rotate.cc
>@@ -49,6 +49,11 @@ void TransposePlane(const uint8_t* src,
>     }
>   }
> #endif
>+#if defined(HAS_TRANSPOSEWX8_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    TransposeWx8 = TransposeWx8_MMI;
>+  }
>+#endif
> #if defined(HAS_TRANSPOSEWX8_FAST_SSSE3)
>   if (TestCpuFlag(kCpuHasSSSE3)) {
>     TransposeWx8 = TransposeWx8_Fast_Any_SSSE3;
>@@ -166,6 +171,14 @@ void RotatePlane180(const uint8_t* src,
>     }
>   }
> #endif
>+#if defined(HAS_MIRRORROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    MirrorRow = MirrorRow_Any_MMI;
>+    if (IS_ALIGNED(width, 8)) {
>+      MirrorRow = MirrorRow_MMI;
>+    }
>+  }
>+#endif
> #if defined(HAS_COPYROW_SSE2)
>   if (TestCpuFlag(kCpuHasSSE2)) {
>     CopyRow = IS_ALIGNED(width, 32) ? CopyRow_SSE2 : CopyRow_Any_SSE2;
>@@ -186,6 +199,11 @@ void RotatePlane180(const uint8_t* src,
>     CopyRow = IS_ALIGNED(width, 32) ? CopyRow_NEON : CopyRow_Any_NEON;
>   }
> #endif
>+#if defined(HAS_COPYROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    CopyRow = IS_ALIGNED(width, 8) ? CopyRow_MMI : CopyRow_Any_MMI;
>+  }
>+#endif
> 
>   // Odd height will harmlessly mirror the middle row twice.
>   for (y = 0; y < half_height; ++y) {
>@@ -232,6 +250,14 @@ void TransposeUV(const uint8_t* src,
>     }
>   }
> #endif
>+#if defined(HAS_TRANSPOSEUVWX8_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    TransposeUVWx8 = TransposeUVWx8_Any_MMI;
>+    if (IS_ALIGNED(width, 4)) {
>+      TransposeUVWx8 = TransposeUVWx8_MMI;
>+    }
>+  }
>+#endif
> #if defined(HAS_TRANSPOSEUVWX16_MSA)
>   if (TestCpuFlag(kCpuHasMSA)) {
>     TransposeUVWx16 = TransposeUVWx16_Any_MSA;
>@@ -331,6 +357,11 @@ void RotateUV180(const uint8_t* src,
>     MirrorUVRow = MirrorUVRow_MSA;
>   }
> #endif
>+#if defined(HAS_MIRRORUVROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI) && IS_ALIGNED(width, 8)) {
>+    MirrorUVRow = MirrorUVRow_MMI;
>+  }
>+#endif
> 
>   dst_a += dst_stride_a * (height - 1);
>   dst_b += dst_stride_b * (height - 1);
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/rotate_any.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/rotate_any.cc
>index c2752e6222c02df6824aa8511fc3d0c5832de3a7..b3baf084d0c8281fa99801620a4724ece95235d6 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/rotate_any.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/rotate_any.cc
>@@ -35,6 +35,9 @@ TANY(TransposeWx8_Any_NEON, TransposeWx8_NEON, 7)
> #ifdef HAS_TRANSPOSEWX8_SSSE3
> TANY(TransposeWx8_Any_SSSE3, TransposeWx8_SSSE3, 7)
> #endif
>+#ifdef HAS_TRANSPOSEWX8_MMI
>+TANY(TransposeWx8_Any_MMI, TransposeWx8_MMI, 7)
>+#endif
> #ifdef HAS_TRANSPOSEWX8_FAST_SSSE3
> TANY(TransposeWx8_Fast_Any_SSSE3, TransposeWx8_Fast_SSSE3, 15)
> #endif
>@@ -62,6 +65,9 @@ TUVANY(TransposeUVWx8_Any_NEON, TransposeUVWx8_NEON, 7)
> #ifdef HAS_TRANSPOSEUVWX8_SSE2
> TUVANY(TransposeUVWx8_Any_SSE2, TransposeUVWx8_SSE2, 7)
> #endif
>+#ifdef HAS_TRANSPOSEUVWX8_MMI
>+TUVANY(TransposeUVWx8_Any_MMI, TransposeUVWx8_MMI, 7)
>+#endif
> #ifdef HAS_TRANSPOSEUVWX16_MSA
> TUVANY(TransposeUVWx16_Any_MSA, TransposeUVWx16_MSA, 7)
> #endif
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/rotate_argb.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/rotate_argb.cc
>index 5a6e05376f17690040900a2c49afb13344828f50..a93fd55f9214fa058c4bf53f9ae863a30a13596b 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/rotate_argb.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/rotate_argb.cc
>@@ -56,6 +56,14 @@ static void ARGBTranspose(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_SCALEARGBROWDOWNEVEN_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ScaleARGBRowDownEven = ScaleARGBRowDownEven_Any_MMI;
>+    if (IS_ALIGNED(height, 4)) {  // Width of dest.
>+      ScaleARGBRowDownEven = ScaleARGBRowDownEven_MMI;
>+    }
>+  }
>+#endif
> 
>   for (i = 0; i < width; ++i) {  // column of source to row of dest.
>     ScaleARGBRowDownEven(src_argb, 0, src_pixel_step, dst_argb, height);
>@@ -142,6 +150,14 @@ void ARGBRotate180(const uint8_t* src_argb,
>     }
>   }
> #endif
>+#if defined(HAS_ARGBMIRRORROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ARGBMirrorRow = ARGBMirrorRow_Any_MMI;
>+    if (IS_ALIGNED(width, 2)) {
>+      ARGBMirrorRow = ARGBMirrorRow_MMI;
>+    }
>+  }
>+#endif
> #if defined(HAS_COPYROW_SSE2)
>   if (TestCpuFlag(kCpuHasSSE2)) {
>     CopyRow = IS_ALIGNED(width * 4, 32) ? CopyRow_SSE2 : CopyRow_Any_SSE2;
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/rotate_mmi.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/rotate_mmi.cc
>new file mode 100644
>index 0000000000000000000000000000000000000000..f8de60834d9cab7a33e0d2ed5205a2ab00972e19
>--- /dev/null
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/rotate_mmi.cc
>@@ -0,0 +1,291 @@
>+/*
>+ *  Copyright 2011 The LibYuv Project Authors. All rights reserved.
>+ *
>+ *  Use of this source code is governed by a BSD-style license
>+ *  that can be found in the LICENSE file in the root of the source
>+ *  tree. An additional intellectual property rights grant can be found
>+ *  in the file PATENTS. All contributing project authors may
>+ *  be found in the AUTHORS file in the root of the source tree.
>+ */
>+
>+#include "libyuv/rotate_row.h"
>+#include "libyuv/row.h"
>+
>+#ifdef __cplusplus
>+namespace libyuv {
>+extern "C" {
>+#endif
>+
>+// This module is for Mips MMI.
>+#if !defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)
>+
>+void TransposeWx8_MMI(const uint8_t* src,
>+                      int src_stride,
>+                      uint8_t* dst,
>+                      int dst_stride,
>+                      int width) {
>+  uint64_t tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6;
>+  uint64_t tmp7, tmp8, tmp9, tmp10, tmp11, tmp12, tmp13;
>+  uint8_t* src_tmp = nullptr;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "ldc1       %[tmp12],        0x00(%[src])                     \n\t"
>+      "dadd       %[src_tmp],      %[src],         %[src_stride]    \n\t"
>+      "ldc1       %[tmp13],        0x00(%[src_tmp])                 \n\t"
>+
>+      /* tmp0 = (00 10 01 11 02 12 03 13) */
>+      "punpcklbh  %[tmp0],         %[tmp12],       %[tmp13]         \n\t"
>+      /* tmp1 = (04 14 05 15 06 16 07 17) */
>+      "punpckhbh  %[tmp1],         %[tmp12],       %[tmp13]         \n\t"
>+
>+      "dadd       %[src_tmp],      %[src_tmp],     %[src_stride]    \n\t"
>+      "ldc1       %[tmp12],        0x00(%[src_tmp])                 \n\t"
>+      "dadd       %[src_tmp],      %[src_tmp],     %[src_stride]    \n\t"
>+      "ldc1       %[tmp13],        0x00(%[src_tmp])                 \n\t"
>+
>+      /* tmp2 = (20 30 21 31 22 32 23 33) */
>+      "punpcklbh  %[tmp2],         %[tmp12],       %[tmp13]         \n\t"
>+      /* tmp3 = (24 34 25 35 26 36 27 37) */
>+      "punpckhbh  %[tmp3],         %[tmp12],       %[tmp13]         \n\t"
>+
>+      /* tmp4 = (00 10 20 30 01 11 21 31) */
>+      "punpcklhw  %[tmp4],         %[tmp0],        %[tmp2]          \n\t"
>+      /* tmp5 = (02 12 22 32 03 13 23 33) */
>+      "punpckhhw  %[tmp5],         %[tmp0],        %[tmp2]          \n\t"
>+      /* tmp6 = (04 14 24 34 05 15 25 35) */
>+      "punpcklhw  %[tmp6],         %[tmp1],        %[tmp3]          \n\t"
>+      /* tmp7 = (06 16 26 36 07 17 27 37) */
>+      "punpckhhw  %[tmp7],         %[tmp1],        %[tmp3]          \n\t"
>+
>+      "dadd       %[src_tmp],      %[src_tmp],     %[src_stride]    \n\t"
>+      "ldc1       %[tmp12],        0x00(%[src_tmp])                 \n\t"
>+      "dadd       %[src_tmp],      %[src_tmp],     %[src_stride]    \n\t"
>+      "ldc1       %[tmp13],        0x00(%[src_tmp])                 \n\t"
>+
>+      /* tmp0 = (40 50 41 51 42 52 43 53) */
>+      "punpcklbh  %[tmp0],         %[tmp12],       %[tmp13]         \n\t"
>+      /* tmp1 = (44 54 45 55 46 56 47 57) */
>+      "punpckhbh  %[tmp1],         %[tmp12],       %[tmp13]         \n\t"
>+
>+      "dadd       %[src_tmp],      %[src_tmp],     %[src_stride]    \n\t"
>+      "ldc1       %[tmp12],        0x00(%[src_tmp])                 \n\t"
>+      "dadd       %[src_tmp],      %[src_tmp],     %[src_stride]    \n\t"
>+      "ldc1       %[tmp13],        0x00(%[src_tmp])                 \n\t"
>+
>+      /* tmp2 = (60 70 61 71 62 72 63 73) */
>+      "punpcklbh  %[tmp2],         %[tmp12],       %[tmp13]         \n\t"
>+      /* tmp3 = (64 74 65 75 66 76 67 77) */
>+      "punpckhbh  %[tmp3],         %[tmp12],       %[tmp13]         \n\t"
>+
>+      /* tmp8 = (40 50 60 70 41 51 61 71) */
>+      "punpcklhw  %[tmp8],         %[tmp0],        %[tmp2]          \n\t"
>+      /* tmp9 = (42 52 62 72 43 53 63 73) */
>+      "punpckhhw  %[tmp9],         %[tmp0],        %[tmp2]          \n\t"
>+      /* tmp10 = (44 54 64 74 45 55 65 75) */
>+      "punpcklhw  %[tmp10],        %[tmp1],        %[tmp3]          \n\t"
>+      /* tmp11 = (46 56 66 76 47 57 67 77) */
>+      "punpckhhw  %[tmp11],        %[tmp1],        %[tmp3]          \n\t"
>+
>+      /* tmp0 = (00 10 20 30 40 50 60 70) */
>+      "punpcklwd  %[tmp0],         %[tmp4],        %[tmp8]          \n\t"
>+      /* tmp1 = (01 11 21 31 41 51 61 71) */
>+      "punpckhwd  %[tmp1],         %[tmp4],        %[tmp8]          \n\t"
>+      "gssdlc1    %[tmp0],         0x07(%[dst])                     \n\t"
>+      "gssdrc1    %[tmp0],         0x00(%[dst])                     \n\t"
>+      "dadd       %[dst],          %[dst],         %[dst_stride]    \n\t"
>+      "gssdlc1    %[tmp1],         0x07(%[dst])                     \n\t"
>+      "gssdrc1    %[tmp1],         0x00(%[dst])                     \n\t"
>+
>+      /* tmp0 = (02 12 22 32 42 52 62 72) */
>+      "punpcklwd  %[tmp0],         %[tmp5],        %[tmp9]          \n\t"
>+      /* tmp1 = (03 13 23 33 43 53 63 73) */
>+      "punpckhwd  %[tmp1],         %[tmp5],        %[tmp9]          \n\t"
>+      "dadd       %[dst],          %[dst],         %[dst_stride]    \n\t"
>+      "gssdlc1    %[tmp0],         0x07(%[dst])                     \n\t"
>+      "gssdrc1    %[tmp0],         0x00(%[dst])                     \n\t"
>+      "dadd       %[dst],          %[dst],         %[dst_stride]    \n\t"
>+      "gssdlc1    %[tmp1],         0x07(%[dst])                     \n\t"
>+      "gssdrc1    %[tmp1],         0x00(%[dst])                     \n\t"
>+
>+      /* tmp0 = (04 14 24 34 44 54 64 74) */
>+      "punpcklwd  %[tmp0],         %[tmp6],        %[tmp10]         \n\t"
>+      /* tmp1 = (05 15 25 35 45 55 65 75) */
>+      "punpckhwd  %[tmp1],         %[tmp6],        %[tmp10]         \n\t"
>+      "dadd       %[dst],          %[dst],         %[dst_stride]    \n\t"
>+      "gssdlc1    %[tmp0],         0x07(%[dst])                     \n\t"
>+      "gssdrc1    %[tmp0],         0x00(%[dst])                     \n\t"
>+      "dadd       %[dst],          %[dst],         %[dst_stride]    \n\t"
>+      "gssdlc1    %[tmp1],         0x07(%[dst])                     \n\t"
>+      "gssdrc1    %[tmp1],         0x00(%[dst])                     \n\t"
>+
>+      /* tmp0 = (06 16 26 36 46 56 66 76) */
>+      "punpcklwd  %[tmp0],         %[tmp7],        %[tmp11]         \n\t"
>+      /* tmp1 = (07 17 27 37 47 57 67 77) */
>+      "punpckhwd  %[tmp1],         %[tmp7],        %[tmp11]         \n\t"
>+      "dadd       %[dst],          %[dst],         %[dst_stride]    \n\t"
>+      "gssdlc1    %[tmp0],         0x07(%[dst])                     \n\t"
>+      "gssdrc1    %[tmp0],         0x00(%[dst])                     \n\t"
>+      "dadd       %[dst],          %[dst],         %[dst_stride]    \n\t"
>+      "gssdlc1    %[tmp1],         0x07(%[dst])                     \n\t"
>+      "gssdrc1    %[tmp1],         0x00(%[dst])                     \n\t"
>+
>+      "dadd       %[dst],          %[dst],         %[dst_stride]    \n\t"
>+      "daddi      %[src],          %[src],          0x08            \n\t"
>+      "daddi      %[width],        %[width],       -0x08            \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+
>+      : [tmp0] "=&f"(tmp0), [tmp1] "=&f"(tmp1), [tmp2] "=&f"(tmp2),
>+        [tmp3] "=&f"(tmp3), [tmp4] "=&f"(tmp4), [tmp5] "=&f"(tmp5),
>+        [tmp6] "=&f"(tmp6), [tmp7] "=&f"(tmp7), [tmp8] "=&f"(tmp8),
>+        [tmp9] "=&f"(tmp9), [tmp10] "=&f"(tmp10), [tmp11] "=&f"(tmp11),
>+        [tmp12] "=&f"(tmp12), [tmp13] "=&f"(tmp13), [dst] "+&r"(dst),
>+        [src_tmp] "+&r"(src_tmp)
>+      : [src] "r"(src), [width] "r"(width), [src_stride] "r"(src_stride),
>+        [dst_stride] "r"(dst_stride)
>+      : "memory");
>+}
>+
>+void TransposeUVWx8_MMI(const uint8_t* src,
>+                        int src_stride,
>+                        uint8_t* dst_a,
>+                        int dst_stride_a,
>+                        uint8_t* dst_b,
>+                        int dst_stride_b,
>+                        int width) {
>+  uint64_t tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6;
>+  uint64_t tmp7, tmp8, tmp9, tmp10, tmp11, tmp12, tmp13;
>+  uint8_t* src_tmp = nullptr;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      /* tmp12 = (u00 v00 u01 v01 u02 v02 u03 v03) */
>+      "ldc1       %[tmp12],        0x00(%[src])                     \n\t"
>+      "dadd       %[src_tmp],      %[src],         %[src_stride]    \n\t"
>+      /* tmp13 = (u10 v10 u11 v11 u12 v12 u13 v13) */
>+      "ldc1       %[tmp13],        0x00(%[src_tmp])                  \n\t"
>+
>+      /* tmp0 = (u00 u10 v00 v10 u01 u11 v01 v11) */
>+      "punpcklbh  %[tmp0],         %[tmp12],       %[tmp13]         \n\t"
>+      /* tmp1 = (u02 u12 v02 v12 u03 u13 v03 v13) */
>+      "punpckhbh  %[tmp1],         %[tmp12],       %[tmp13]         \n\t"
>+
>+      "dadd       %[src_tmp],      %[src_tmp],     %[src_stride]    \n\t"
>+      /* tmp12 = (u20 v20 u21 v21 u22 v22 u23 v23) */
>+      "ldc1       %[tmp12],        0x00(%[src_tmp])                 \n\t"
>+      "dadd       %[src_tmp],      %[src_tmp],     %[src_stride]    \n\t"
>+      /* tmp13 = (u30 v30 u31 v31 u32 v32 u33 v33) */
>+      "ldc1       %[tmp13],        0x00(%[src_tmp])                 \n\t"
>+
>+      /* tmp2 = (u20 u30 v20 v30 u21 u31 v21 v31) */
>+      "punpcklbh  %[tmp2],         %[tmp12],       %[tmp13]         \n\t"
>+      /* tmp3 = (u22 u32 v22 v32 u23 u33 v23 v33) */
>+      "punpckhbh  %[tmp3],         %[tmp12],       %[tmp13]         \n\t"
>+
>+      /* tmp4 = (u00 u10 u20 u30 v00 v10 v20 v30) */
>+      "punpcklhw  %[tmp4],         %[tmp0],        %[tmp2]          \n\t"
>+      /* tmp5 = (u01 u11 u21 u31 v01 v11 v21 v31) */
>+      "punpckhhw  %[tmp5],         %[tmp0],        %[tmp2]          \n\t"
>+      /* tmp6 = (u02 u12 u22 u32 v02 v12 v22 v32) */
>+      "punpcklhw  %[tmp6],         %[tmp1],        %[tmp3]          \n\t"
>+      /* tmp7 = (u03 u13 u23 u33 v03 v13 v23 v33) */
>+      "punpckhhw  %[tmp7],         %[tmp1],        %[tmp3]          \n\t"
>+
>+      "dadd       %[src_tmp],     %[src_tmp],      %[src_stride]    \n\t"
>+      /* tmp12 = (u40 v40 u41 v41 u42 v42 u43 v43) */
>+      "ldc1       %[tmp12],        0x00(%[src_tmp])                 \n\t"
>+      /* tmp13 = (u50 v50 u51 v51 u52 v52 u53 v53) */
>+      "dadd       %[src_tmp],      %[src_tmp],     %[src_stride]    \n\t"
>+      "ldc1       %[tmp13],        0x00(%[src_tmp])                 \n\t"
>+
>+      /* tmp0 = (u40 u50 v40 v50 u41 u51 v41 v51) */
>+      "punpcklbh  %[tmp0],         %[tmp12],       %[tmp13]         \n\t"
>+      /* tmp1 = (u42 u52 v42 v52 u43 u53 v43 v53) */
>+      "punpckhbh  %[tmp1],         %[tmp12],       %[tmp13]         \n\t"
>+
>+      "dadd       %[src_tmp],      %[src_tmp],     %[src_stride]    \n\t"
>+      /* tmp12 = (u60 v60 u61 v61 u62 v62 u63 v63) */
>+      "ldc1       %[tmp12],        0x00(%[src_tmp])                 \n\t"
>+      /* tmp13 = (u70 v70 u71 v71 u72 v72 u73 v73) */
>+      "dadd       %[src_tmp],      %[src_tmp],     %[src_stride]    \n\t"
>+      "ldc1       %[tmp13],        0x00(%[src_tmp])                 \n\t"
>+
>+      /* tmp2 = (u60 u70 v60 v70 u61 u71 v61 v71) */
>+      "punpcklbh  %[tmp2],         %[tmp12],       %[tmp13]         \n\t"
>+      /* tmp3 = (u62 u72 v62 v72 u63 u73 v63 v73) */
>+      "punpckhbh  %[tmp3],         %[tmp12],       %[tmp13]         \n\t"
>+
>+      /* tmp8 = (u40 u50 u60 u70 v40 v50 v60 v70) */
>+      "punpcklhw  %[tmp8],         %[tmp0],        %[tmp2]          \n\t"
>+      /* tmp9 = (u41 u51 u61 u71 v41 v51 v61 v71) */
>+      "punpckhhw  %[tmp9],         %[tmp0],        %[tmp2]          \n\t"
>+      /* tmp10 = (u42 u52 u62 u72 v42 v52 v62 v72) */
>+      "punpcklhw  %[tmp10],        %[tmp1],        %[tmp3]          \n\t"
>+      /* tmp11 = (u43 u53 u63 u73 v43 v53 v63 v73) */
>+      "punpckhhw  %[tmp11],        %[tmp1],        %[tmp3]          \n\t"
>+
>+      /* tmp0 = (u00 u10 u20 u30 u40 u50 u60 u70) */
>+      "punpcklwd  %[tmp0],         %[tmp4],        %[tmp8]          \n\t"
>+      /* tmp1 = (v00 v10 v20 v30 v40 v50 v60 v70) */
>+      "punpckhwd  %[tmp1],         %[tmp4],        %[tmp8]          \n\t"
>+      "gssdlc1    %[tmp0],         0x07(%[dst_a])                   \n\t"
>+      "gssdrc1    %[tmp0],         0x00(%[dst_a])                   \n\t"
>+      "gssdlc1    %[tmp1],         0x07(%[dst_b])                   \n\t"
>+      "gssdrc1    %[tmp1],         0x00(%[dst_b])                   \n\t"
>+
>+      /* tmp0 = (u01 u11 u21 u31 u41 u51 u61 u71) */
>+      "punpcklwd  %[tmp0],         %[tmp5],        %[tmp9]          \n\t"
>+      /* tmp1 = (v01 v11 v21 v31 v41 v51 v61 v71) */
>+      "punpckhwd  %[tmp1],         %[tmp5],        %[tmp9]          \n\t"
>+      "dadd       %[dst_a],        %[dst_a],       %[dst_stride_a]  \n\t"
>+      "gssdlc1    %[tmp0],         0x07(%[dst_a])                   \n\t"
>+      "gssdrc1    %[tmp0],         0x00(%[dst_a])                   \n\t"
>+      "dadd       %[dst_b],        %[dst_b],       %[dst_stride_b]  \n\t"
>+      "gssdlc1    %[tmp1],         0x07(%[dst_b])                   \n\t"
>+      "gssdrc1    %[tmp1],         0x00(%[dst_b])                   \n\t"
>+
>+      /* tmp0 = (u02 u12 u22 u32 u42 u52 u62 u72) */
>+      "punpcklwd  %[tmp0],         %[tmp6],        %[tmp10]         \n\t"
>+      /* tmp1 = (v02 v12 v22 v32 v42 v52 v62 v72) */
>+      "punpckhwd  %[tmp1],         %[tmp6],        %[tmp10]         \n\t"
>+      "dadd       %[dst_a],        %[dst_a],       %[dst_stride_a]  \n\t"
>+      "gssdlc1    %[tmp0],         0x07(%[dst_a])                   \n\t"
>+      "gssdrc1    %[tmp0],         0x00(%[dst_a])                   \n\t"
>+      "dadd       %[dst_b],        %[dst_b],       %[dst_stride_b]  \n\t"
>+      "gssdlc1    %[tmp1],         0x07(%[dst_b])                   \n\t"
>+      "gssdrc1    %[tmp1],         0x00(%[dst_b])                   \n\t"
>+
>+      /* tmp0 = (u03 u13 u23 u33 u43 u53 u63 u73) */
>+      "punpcklwd  %[tmp0],         %[tmp7],        %[tmp11]         \n\t"
>+      /* tmp1 = (v03 v13 v23 v33 v43 v53 v63 v73) */
>+      "punpckhwd  %[tmp1],         %[tmp7],        %[tmp11]         \n\t"
>+      "dadd       %[dst_a],        %[dst_a],       %[dst_stride_a]  \n\t"
>+      "gssdlc1    %[tmp0],         0x07(%[dst_a])                   \n\t"
>+      "gssdrc1    %[tmp0],         0x00(%[dst_a])                   \n\t"
>+      "dadd       %[dst_b],        %[dst_b],       %[dst_stride_b]  \n\t"
>+      "gssdlc1    %[tmp1],         0x07(%[dst_b])                   \n\t"
>+      "gssdrc1    %[tmp1],         0x00(%[dst_b])                   \n\t"
>+
>+      "dadd       %[dst_a],        %[dst_a],       %[dst_stride_a]  \n\t"
>+      "dadd       %[dst_b],        %[dst_b],       %[dst_stride_b]  \n\t"
>+      "daddiu     %[src],          %[src],          0x08            \n\t"
>+      "daddi      %[width],        %[width],       -0x04            \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+
>+      : [tmp0] "=&f"(tmp0), [tmp1] "=&f"(tmp1), [tmp2] "=&f"(tmp2),
>+        [tmp3] "=&f"(tmp3), [tmp4] "=&f"(tmp4), [tmp5] "=&f"(tmp5),
>+        [tmp6] "=&f"(tmp6), [tmp7] "=&f"(tmp7), [tmp8] "=&f"(tmp8),
>+        [tmp9] "=&f"(tmp9), [tmp10] "=&f"(tmp10), [tmp11] "=&f"(tmp11),
>+        [tmp12] "=&f"(tmp12), [tmp13] "=&f"(tmp13), [dst_a] "+&r"(dst_a),
>+        [dst_b] "+&r"(dst_b), [src_tmp] "+&r"(src_tmp)
>+      : [src] "r"(src), [width] "r"(width), [dst_stride_a] "r"(dst_stride_a),
>+        [dst_stride_b] "r"(dst_stride_b), [src_stride] "r"(src_stride)
>+      : "memory");
>+}
>+
>+#endif  // !defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)
>+
>+#ifdef __cplusplus
>+}  // extern "C"
>+}  // namespace libyuv
>+#endif
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/row_any.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/row_any.cc
>index cc5914dd29d0d0106177e583598b87053f6dc721..031a8f6490ef15e080d877855ab1aac1d7010697 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/row_any.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/row_any.cc
>@@ -92,6 +92,9 @@ ANY31(MergeRGBRow_Any_SSSE3, MergeRGBRow_SSSE3, 0, 0, 3, 15)
> #ifdef HAS_MERGERGBROW_NEON
> ANY31(MergeRGBRow_Any_NEON, MergeRGBRow_NEON, 0, 0, 3, 15)
> #endif
>+#ifdef HAS_MERGERGBROW_MMI
>+ANY31(MergeRGBRow_Any_MMI, MergeRGBRow_MMI, 0, 0, 3, 7)
>+#endif
> #ifdef HAS_I422TOYUY2ROW_SSE2
> ANY31(I422ToYUY2Row_Any_SSE2, I422ToYUY2Row_SSE2, 1, 1, 4, 15)
> ANY31(I422ToUYVYRow_Any_SSE2, I422ToUYVYRow_SSE2, 1, 1, 4, 15)
>@@ -106,44 +109,54 @@ ANY31(I422ToYUY2Row_Any_NEON, I422ToYUY2Row_NEON, 1, 1, 4, 15)
> #ifdef HAS_I422TOYUY2ROW_MSA
> ANY31(I422ToYUY2Row_Any_MSA, I422ToYUY2Row_MSA, 1, 1, 4, 31)
> #endif
>+#ifdef HAS_I422TOYUY2ROW_MMI
>+ANY31(I422ToYUY2Row_Any_MMI, I422ToYUY2Row_MMI, 1, 1, 4, 7)
>+#endif
> #ifdef HAS_I422TOUYVYROW_NEON
> ANY31(I422ToUYVYRow_Any_NEON, I422ToUYVYRow_NEON, 1, 1, 4, 15)
> #endif
> #ifdef HAS_I422TOUYVYROW_MSA
> ANY31(I422ToUYVYRow_Any_MSA, I422ToUYVYRow_MSA, 1, 1, 4, 31)
> #endif
>+#ifdef HAS_I422TOUYVYROW_MMI
>+ANY31(I422ToUYVYRow_Any_MMI, I422ToUYVYRow_MMI, 1, 1, 4, 7)
>+#endif
> #ifdef HAS_BLENDPLANEROW_AVX2
> ANY31(BlendPlaneRow_Any_AVX2, BlendPlaneRow_AVX2, 0, 0, 1, 31)
> #endif
> #ifdef HAS_BLENDPLANEROW_SSSE3
> ANY31(BlendPlaneRow_Any_SSSE3, BlendPlaneRow_SSSE3, 0, 0, 1, 7)
> #endif
>+#ifdef HAS_BLENDPLANEROW_MMI
>+ANY31(BlendPlaneRow_Any_MMI, BlendPlaneRow_MMI, 0, 0, 1, 7)
>+#endif
> #undef ANY31
> 
> // Note that odd width replication includes 444 due to implementation
> // on arm that subsamples 444 to 422 internally.
> // Any 3 planes to 1 with yuvconstants
>-#define ANY31C(NAMEANY, ANY_SIMD, UVSHIFT, DUVSHIFT, BPP, MASK)                \
>-  void NAMEANY(const uint8_t* y_buf, const uint8_t* u_buf,                     \
>-               const uint8_t* v_buf, uint8_t* dst_ptr,                         \
>-               const struct YuvConstants* yuvconstants, int width) {           \
>-    SIMD_ALIGNED(uint8_t temp[64 * 4]);                                        \
>-    memset(temp, 0, 64 * 3); /* for YUY2 and msan */                           \
>-    int r = width & MASK;                                                      \
>-    int n = width & ~MASK;                                                     \
>-    if (n > 0) {                                                               \
>-      ANY_SIMD(y_buf, u_buf, v_buf, dst_ptr, yuvconstants, n);                 \
>-    }                                                                          \
>-    memcpy(temp, y_buf + n, r);                                                \
>-    memcpy(temp + 64, u_buf + (n >> UVSHIFT), SS(r, UVSHIFT));                 \
>-    memcpy(temp + 128, v_buf + (n >> UVSHIFT), SS(r, UVSHIFT));                \
>-    if (width & 1) {                                                           \
>-      temp[64 + SS(r, UVSHIFT)] = temp[64 + SS(r, UVSHIFT) - 1];               \
>-      temp[128 + SS(r, UVSHIFT)] = temp[128 + SS(r, UVSHIFT) - 1];             \
>-    }                                                                          \
>-    ANY_SIMD(temp, temp + 64, temp + 128, temp + 192, yuvconstants, MASK + 1); \
>-    memcpy(dst_ptr + (n >> DUVSHIFT) * BPP, temp + 192,                        \
>-           SS(r, DUVSHIFT) * BPP);                                             \
>+#define ANY31C(NAMEANY, ANY_SIMD, UVSHIFT, DUVSHIFT, BPP, MASK)      \
>+  void NAMEANY(const uint8_t* y_buf, const uint8_t* u_buf,           \
>+               const uint8_t* v_buf, uint8_t* dst_ptr,               \
>+               const struct YuvConstants* yuvconstants, int width) { \
>+    SIMD_ALIGNED(uint8_t temp[128 * 4]);                             \
>+    memset(temp, 0, 128 * 3); /* for YUY2 and msan */                \
>+    int r = width & MASK;                                            \
>+    int n = width & ~MASK;                                           \
>+    if (n > 0) {                                                     \
>+      ANY_SIMD(y_buf, u_buf, v_buf, dst_ptr, yuvconstants, n);       \
>+    }                                                                \
>+    memcpy(temp, y_buf + n, r);                                      \
>+    memcpy(temp + 128, u_buf + (n >> UVSHIFT), SS(r, UVSHIFT));      \
>+    memcpy(temp + 256, v_buf + (n >> UVSHIFT), SS(r, UVSHIFT));      \
>+    if (width & 1) {                                                 \
>+      temp[128 + SS(r, UVSHIFT)] = temp[128 + SS(r, UVSHIFT) - 1];   \
>+      temp[256 + SS(r, UVSHIFT)] = temp[256 + SS(r, UVSHIFT) - 1];   \
>+    }                                                                \
>+    ANY_SIMD(temp, temp + 128, temp + 256, temp + 384, yuvconstants, \
>+             MASK + 1);                                              \
>+    memcpy(dst_ptr + (n >> DUVSHIFT) * BPP, temp + 384,              \
>+           SS(r, DUVSHIFT) * BPP);                                   \
>   }
> 
> #ifdef HAS_I422TOARGBROW_SSSE3
>@@ -161,10 +174,10 @@ ANY31C(I422ToRGBARow_Any_SSSE3, I422ToRGBARow_SSSE3, 1, 0, 4, 7)
> ANY31C(I422ToARGB4444Row_Any_SSSE3, I422ToARGB4444Row_SSSE3, 1, 0, 2, 7)
> ANY31C(I422ToARGB1555Row_Any_SSSE3, I422ToARGB1555Row_SSSE3, 1, 0, 2, 7)
> ANY31C(I422ToRGB565Row_Any_SSSE3, I422ToRGB565Row_SSSE3, 1, 0, 2, 7)
>-ANY31C(I422ToRGB24Row_Any_SSSE3, I422ToRGB24Row_SSSE3, 1, 0, 3, 7)
>+ANY31C(I422ToRGB24Row_Any_SSSE3, I422ToRGB24Row_SSSE3, 1, 0, 3, 15)
> #endif  // HAS_I444TOARGBROW_SSSE3
> #ifdef HAS_I422TORGB24ROW_AVX2
>-ANY31C(I422ToRGB24Row_Any_AVX2, I422ToRGB24Row_AVX2, 1, 0, 3, 15)
>+ANY31C(I422ToRGB24Row_Any_AVX2, I422ToRGB24Row_AVX2, 1, 0, 3, 31)
> #endif
> #ifdef HAS_I422TOARGBROW_AVX2
> ANY31C(I422ToARGBRow_Any_AVX2, I422ToARGBRow_AVX2, 1, 0, 4, 15)
>@@ -270,6 +283,9 @@ ANY21(MergeUVRow_Any_NEON, MergeUVRow_NEON, 0, 1, 1, 2, 15)
> #ifdef HAS_MERGEUVROW_MSA
> ANY21(MergeUVRow_Any_MSA, MergeUVRow_MSA, 0, 1, 1, 2, 15)
> #endif
>+#ifdef HAS_MERGEUVROW_MMI
>+ANY21(MergeUVRow_Any_MMI, MergeUVRow_MMI, 0, 1, 1, 2, 7)
>+#endif
> 
> // Math functions.
> #ifdef HAS_ARGBMULTIPLYROW_SSE2
>@@ -302,12 +318,21 @@ ANY21(ARGBSubtractRow_Any_NEON, ARGBSubtractRow_NEON, 0, 4, 4, 4, 7)
> #ifdef HAS_ARGBMULTIPLYROW_MSA
> ANY21(ARGBMultiplyRow_Any_MSA, ARGBMultiplyRow_MSA, 0, 4, 4, 4, 3)
> #endif
>+#ifdef HAS_ARGBMULTIPLYROW_MMI
>+ANY21(ARGBMultiplyRow_Any_MMI, ARGBMultiplyRow_MMI, 0, 4, 4, 4, 1)
>+#endif
> #ifdef HAS_ARGBADDROW_MSA
> ANY21(ARGBAddRow_Any_MSA, ARGBAddRow_MSA, 0, 4, 4, 4, 7)
> #endif
>+#ifdef HAS_ARGBADDROW_MMI
>+ANY21(ARGBAddRow_Any_MMI, ARGBAddRow_MMI, 0, 4, 4, 4, 1)
>+#endif
> #ifdef HAS_ARGBSUBTRACTROW_MSA
> ANY21(ARGBSubtractRow_Any_MSA, ARGBSubtractRow_MSA, 0, 4, 4, 4, 7)
> #endif
>+#ifdef HAS_ARGBSUBTRACTROW_MMI
>+ANY21(ARGBSubtractRow_Any_MMI, ARGBSubtractRow_MMI, 0, 4, 4, 4, 1)
>+#endif
> #ifdef HAS_SOBELROW_SSE2
> ANY21(SobelRow_Any_SSE2, SobelRow_SSE2, 0, 1, 1, 4, 15)
> #endif
>@@ -317,6 +342,9 @@ ANY21(SobelRow_Any_NEON, SobelRow_NEON, 0, 1, 1, 4, 7)
> #ifdef HAS_SOBELROW_MSA
> ANY21(SobelRow_Any_MSA, SobelRow_MSA, 0, 1, 1, 4, 15)
> #endif
>+#ifdef HAS_SOBELROW_MMI
>+ANY21(SobelRow_Any_MMI, SobelRow_MMI, 0, 1, 1, 4, 7)
>+#endif
> #ifdef HAS_SOBELTOPLANEROW_SSE2
> ANY21(SobelToPlaneRow_Any_SSE2, SobelToPlaneRow_SSE2, 0, 1, 1, 1, 15)
> #endif
>@@ -326,6 +354,9 @@ ANY21(SobelToPlaneRow_Any_NEON, SobelToPlaneRow_NEON, 0, 1, 1, 1, 15)
> #ifdef HAS_SOBELTOPLANEROW_MSA
> ANY21(SobelToPlaneRow_Any_MSA, SobelToPlaneRow_MSA, 0, 1, 1, 1, 31)
> #endif
>+#ifdef HAS_SOBELTOPLANEROW_MMI
>+ANY21(SobelToPlaneRow_Any_MMI, SobelToPlaneRow_MMI, 0, 1, 1, 1, 7)
>+#endif
> #ifdef HAS_SOBELXYROW_SSE2
> ANY21(SobelXYRow_Any_SSE2, SobelXYRow_SSE2, 0, 1, 1, 4, 15)
> #endif
>@@ -335,24 +366,27 @@ ANY21(SobelXYRow_Any_NEON, SobelXYRow_NEON, 0, 1, 1, 4, 7)
> #ifdef HAS_SOBELXYROW_MSA
> ANY21(SobelXYRow_Any_MSA, SobelXYRow_MSA, 0, 1, 1, 4, 15)
> #endif
>+#ifdef HAS_SOBELXYROW_MMI
>+ANY21(SobelXYRow_Any_MMI, SobelXYRow_MMI, 0, 1, 1, 4, 7)
>+#endif
> #undef ANY21
> 
> // Any 2 planes to 1 with yuvconstants
> #define ANY21C(NAMEANY, ANY_SIMD, UVSHIFT, SBPP, SBPP2, BPP, MASK)            \
>   void NAMEANY(const uint8_t* y_buf, const uint8_t* uv_buf, uint8_t* dst_ptr, \
>                const struct YuvConstants* yuvconstants, int width) {          \
>-    SIMD_ALIGNED(uint8_t temp[64 * 3]);                                       \
>-    memset(temp, 0, 64 * 2); /* for msan */                                   \
>+    SIMD_ALIGNED(uint8_t temp[128 * 3]);                                      \
>+    memset(temp, 0, 128 * 2); /* for msan */                                  \
>     int r = width & MASK;                                                     \
>     int n = width & ~MASK;                                                    \
>     if (n > 0) {                                                              \
>       ANY_SIMD(y_buf, uv_buf, dst_ptr, yuvconstants, n);                      \
>     }                                                                         \
>     memcpy(temp, y_buf + n * SBPP, r * SBPP);                                 \
>-    memcpy(temp + 64, uv_buf + (n >> UVSHIFT) * SBPP2,                        \
>+    memcpy(temp + 128, uv_buf + (n >> UVSHIFT) * SBPP2,                       \
>            SS(r, UVSHIFT) * SBPP2);                                           \
>-    ANY_SIMD(temp, temp + 64, temp + 128, yuvconstants, MASK + 1);            \
>-    memcpy(dst_ptr + n * BPP, temp + 128, r * BPP);                           \
>+    ANY_SIMD(temp, temp + 128, temp + 256, yuvconstants, MASK + 1);           \
>+    memcpy(dst_ptr + n * BPP, temp + 256, r * BPP);                           \
>   }
> 
> // Biplanar to RGB.
>@@ -386,6 +420,18 @@ ANY21C(NV12ToRGB24Row_Any_NEON, NV12ToRGB24Row_NEON, 1, 1, 2, 3, 7)
> #ifdef HAS_NV21TORGB24ROW_NEON
> ANY21C(NV21ToRGB24Row_Any_NEON, NV21ToRGB24Row_NEON, 1, 1, 2, 3, 7)
> #endif
>+#ifdef HAS_NV12TORGB24ROW_SSSE3
>+ANY21C(NV12ToRGB24Row_Any_SSSE3, NV12ToRGB24Row_SSSE3, 1, 1, 2, 3, 15)
>+#endif
>+#ifdef HAS_NV21TORGB24ROW_SSSE3
>+ANY21C(NV21ToRGB24Row_Any_SSSE3, NV21ToRGB24Row_SSSE3, 1, 1, 2, 3, 15)
>+#endif
>+#ifdef HAS_NV12TORGB24ROW_AVX2
>+ANY21C(NV12ToRGB24Row_Any_AVX2, NV12ToRGB24Row_AVX2, 1, 1, 2, 3, 31)
>+#endif
>+#ifdef HAS_NV21TORGB24ROW_AVX2
>+ANY21C(NV21ToRGB24Row_Any_AVX2, NV21ToRGB24Row_AVX2, 1, 1, 2, 3, 31)
>+#endif
> #ifdef HAS_NV12TORGB565ROW_SSSE3
> ANY21C(NV12ToRGB565Row_Any_SSSE3, NV12ToRGB565Row_SSSE3, 1, 1, 2, 2, 7)
> #endif
>@@ -431,6 +477,15 @@ ANY11(ARGBToRGB565Row_Any_SSE2, ARGBToRGB565Row_SSE2, 0, 4, 2, 3)
> ANY11(ARGBToARGB1555Row_Any_SSE2, ARGBToARGB1555Row_SSE2, 0, 4, 2, 3)
> ANY11(ARGBToARGB4444Row_Any_SSE2, ARGBToARGB4444Row_SSE2, 0, 4, 2, 3)
> #endif
>+#if defined(HAS_ARGBTORGB24ROW_AVX2)
>+ANY11(ARGBToRGB24Row_Any_AVX2, ARGBToRGB24Row_AVX2, 0, 4, 3, 31)
>+#endif
>+#if defined(HAS_ARGBTORGB24ROW_AVX512VBMI)
>+ANY11(ARGBToRGB24Row_Any_AVX512VBMI, ARGBToRGB24Row_AVX512VBMI, 0, 4, 3, 31)
>+#endif
>+#if defined(HAS_ARGBTORAWROW_AVX2)
>+ANY11(ARGBToRAWRow_Any_AVX2, ARGBToRAWRow_AVX2, 0, 4, 3, 31)
>+#endif
> #if defined(HAS_ARGBTORGB565ROW_AVX2)
> ANY11(ARGBToRGB565Row_Any_AVX2, ARGBToRGB565Row_AVX2, 0, 4, 2, 7)
> #endif
>@@ -499,12 +554,24 @@ ANY11(ARGBToARGB4444Row_Any_MSA, ARGBToARGB4444Row_MSA, 0, 4, 2, 7)
> ANY11(J400ToARGBRow_Any_MSA, J400ToARGBRow_MSA, 0, 1, 4, 15)
> ANY11(I400ToARGBRow_Any_MSA, I400ToARGBRow_MSA, 0, 1, 4, 15)
> #endif
>+#if defined(HAS_ARGBTORGB24ROW_MMI)
>+ANY11(ARGBToRGB24Row_Any_MMI, ARGBToRGB24Row_MMI, 0, 4, 3, 3)
>+ANY11(ARGBToRAWRow_Any_MMI, ARGBToRAWRow_MMI, 0, 4, 3, 3)
>+ANY11(ARGBToRGB565Row_Any_MMI, ARGBToRGB565Row_MMI, 0, 4, 2, 3)
>+ANY11(ARGBToARGB1555Row_Any_MMI, ARGBToARGB1555Row_MMI, 0, 4, 2, 3)
>+ANY11(ARGBToARGB4444Row_Any_MMI, ARGBToARGB4444Row_MMI, 0, 4, 2, 3)
>+ANY11(J400ToARGBRow_Any_MMI, J400ToARGBRow_MMI, 0, 1, 4, 3)
>+ANY11(I400ToARGBRow_Any_MMI, I400ToARGBRow_MMI, 0, 1, 4, 7)
>+#endif
> #if defined(HAS_RAWTORGB24ROW_NEON)
> ANY11(RAWToRGB24Row_Any_NEON, RAWToRGB24Row_NEON, 0, 3, 3, 7)
> #endif
> #if defined(HAS_RAWTORGB24ROW_MSA)
> ANY11(RAWToRGB24Row_Any_MSA, RAWToRGB24Row_MSA, 0, 3, 3, 15)
> #endif
>+#if defined(HAS_RAWTORGB24ROW_MMI)
>+ANY11(RAWToRGB24Row_Any_MMI, RAWToRGB24Row_MMI, 0, 3, 3, 3)
>+#endif
> #ifdef HAS_ARGBTOYROW_AVX2
> ANY11(ARGBToYRow_Any_AVX2, ARGBToYRow_AVX2, 0, 4, 1, 31)
> #endif
>@@ -536,57 +603,87 @@ ANY11(ARGBToYRow_Any_NEON, ARGBToYRow_NEON, 0, 4, 1, 7)
> #ifdef HAS_ARGBTOYROW_MSA
> ANY11(ARGBToYRow_Any_MSA, ARGBToYRow_MSA, 0, 4, 1, 15)
> #endif
>+#ifdef HAS_ARGBTOYROW_MMI
>+ANY11(ARGBToYRow_Any_MMI, ARGBToYRow_MMI, 0, 4, 1, 7)
>+#endif
> #ifdef HAS_ARGBTOYJROW_NEON
> ANY11(ARGBToYJRow_Any_NEON, ARGBToYJRow_NEON, 0, 4, 1, 7)
> #endif
> #ifdef HAS_ARGBTOYJROW_MSA
> ANY11(ARGBToYJRow_Any_MSA, ARGBToYJRow_MSA, 0, 4, 1, 15)
> #endif
>+#ifdef HAS_ARGBTOYJROW_MMI
>+ANY11(ARGBToYJRow_Any_MMI, ARGBToYJRow_MMI, 0, 4, 1, 7)
>+#endif
> #ifdef HAS_BGRATOYROW_NEON
> ANY11(BGRAToYRow_Any_NEON, BGRAToYRow_NEON, 0, 4, 1, 7)
> #endif
> #ifdef HAS_BGRATOYROW_MSA
> ANY11(BGRAToYRow_Any_MSA, BGRAToYRow_MSA, 0, 4, 1, 15)
> #endif
>+#ifdef HAS_BGRATOYROW_MMI
>+ANY11(BGRAToYRow_Any_MMI, BGRAToYRow_MMI, 0, 4, 1, 7)
>+#endif
> #ifdef HAS_ABGRTOYROW_NEON
> ANY11(ABGRToYRow_Any_NEON, ABGRToYRow_NEON, 0, 4, 1, 7)
> #endif
> #ifdef HAS_ABGRTOYROW_MSA
> ANY11(ABGRToYRow_Any_MSA, ABGRToYRow_MSA, 0, 4, 1, 7)
> #endif
>+#ifdef HAS_ABGRTOYROW_MMI
>+ANY11(ABGRToYRow_Any_MMI, ABGRToYRow_MMI, 0, 4, 1, 7)
>+#endif
> #ifdef HAS_RGBATOYROW_NEON
> ANY11(RGBAToYRow_Any_NEON, RGBAToYRow_NEON, 0, 4, 1, 7)
> #endif
> #ifdef HAS_RGBATOYROW_MSA
> ANY11(RGBAToYRow_Any_MSA, RGBAToYRow_MSA, 0, 4, 1, 15)
> #endif
>+#ifdef HAS_RGBATOYROW_MMI
>+ANY11(RGBAToYRow_Any_MMI, RGBAToYRow_MMI, 0, 4, 1, 7)
>+#endif
> #ifdef HAS_RGB24TOYROW_NEON
> ANY11(RGB24ToYRow_Any_NEON, RGB24ToYRow_NEON, 0, 3, 1, 7)
> #endif
> #ifdef HAS_RGB24TOYROW_MSA
> ANY11(RGB24ToYRow_Any_MSA, RGB24ToYRow_MSA, 0, 3, 1, 15)
> #endif
>+#ifdef HAS_RGB24TOYROW_MMI
>+ANY11(RGB24ToYRow_Any_MMI, RGB24ToYRow_MMI, 0, 3, 1, 7)
>+#endif
> #ifdef HAS_RAWTOYROW_NEON
> ANY11(RAWToYRow_Any_NEON, RAWToYRow_NEON, 0, 3, 1, 7)
> #endif
> #ifdef HAS_RAWTOYROW_MSA
> ANY11(RAWToYRow_Any_MSA, RAWToYRow_MSA, 0, 3, 1, 15)
> #endif
>+#ifdef HAS_RAWTOYROW_MMI
>+ANY11(RAWToYRow_Any_MMI, RAWToYRow_MMI, 0, 3, 1, 7)
>+#endif
> #ifdef HAS_RGB565TOYROW_NEON
> ANY11(RGB565ToYRow_Any_NEON, RGB565ToYRow_NEON, 0, 2, 1, 7)
> #endif
> #ifdef HAS_RGB565TOYROW_MSA
> ANY11(RGB565ToYRow_Any_MSA, RGB565ToYRow_MSA, 0, 2, 1, 15)
> #endif
>+#ifdef HAS_RGB565TOYROW_MMI
>+ANY11(RGB565ToYRow_Any_MMI, RGB565ToYRow_MMI, 0, 2, 1, 7)
>+#endif
> #ifdef HAS_ARGB1555TOYROW_NEON
> ANY11(ARGB1555ToYRow_Any_NEON, ARGB1555ToYRow_NEON, 0, 2, 1, 7)
> #endif
> #ifdef HAS_ARGB1555TOYROW_MSA
> ANY11(ARGB1555ToYRow_Any_MSA, ARGB1555ToYRow_MSA, 0, 2, 1, 15)
> #endif
>+#ifdef HAS_ARGB1555TOYROW_MMI
>+ANY11(ARGB1555ToYRow_Any_MMI, ARGB1555ToYRow_MMI, 0, 2, 1, 7)
>+#endif
> #ifdef HAS_ARGB4444TOYROW_NEON
> ANY11(ARGB4444ToYRow_Any_NEON, ARGB4444ToYRow_NEON, 0, 2, 1, 7)
> #endif
>+#ifdef HAS_ARGB4444TOYROW_MMI
>+ANY11(ARGB4444ToYRow_Any_MMI, ARGB4444ToYRow_MMI, 0, 2, 1, 7)
>+#endif
> #ifdef HAS_YUY2TOYROW_NEON
> ANY11(YUY2ToYRow_Any_NEON, YUY2ToYRow_NEON, 1, 4, 1, 15)
> #endif
>@@ -596,39 +693,60 @@ ANY11(UYVYToYRow_Any_NEON, UYVYToYRow_NEON, 1, 4, 1, 15)
> #ifdef HAS_YUY2TOYROW_MSA
> ANY11(YUY2ToYRow_Any_MSA, YUY2ToYRow_MSA, 1, 4, 1, 31)
> #endif
>+#ifdef HAS_YUY2TOYROW_MMI
>+ANY11(YUY2ToYRow_Any_MMI, YUY2ToYRow_MMI, 1, 4, 1, 7)
>+#endif
> #ifdef HAS_UYVYTOYROW_MSA
> ANY11(UYVYToYRow_Any_MSA, UYVYToYRow_MSA, 1, 4, 1, 31)
> #endif
>+#ifdef HAS_UYVYTOYROW_MMI
>+ANY11(UYVYToYRow_Any_MMI, UYVYToYRow_MMI, 1, 4, 1, 15)
>+#endif
> #ifdef HAS_RGB24TOARGBROW_NEON
> ANY11(RGB24ToARGBRow_Any_NEON, RGB24ToARGBRow_NEON, 0, 3, 4, 7)
> #endif
> #ifdef HAS_RGB24TOARGBROW_MSA
> ANY11(RGB24ToARGBRow_Any_MSA, RGB24ToARGBRow_MSA, 0, 3, 4, 15)
> #endif
>+#ifdef HAS_RGB24TOARGBROW_MMI
>+ANY11(RGB24ToARGBRow_Any_MMI, RGB24ToARGBRow_MMI, 0, 3, 4, 3)
>+#endif
> #ifdef HAS_RAWTOARGBROW_NEON
> ANY11(RAWToARGBRow_Any_NEON, RAWToARGBRow_NEON, 0, 3, 4, 7)
> #endif
> #ifdef HAS_RAWTOARGBROW_MSA
> ANY11(RAWToARGBRow_Any_MSA, RAWToARGBRow_MSA, 0, 3, 4, 15)
> #endif
>+#ifdef HAS_RAWTOARGBROW_MMI
>+ANY11(RAWToARGBRow_Any_MMI, RAWToARGBRow_MMI, 0, 3, 4, 3)
>+#endif
> #ifdef HAS_RGB565TOARGBROW_NEON
> ANY11(RGB565ToARGBRow_Any_NEON, RGB565ToARGBRow_NEON, 0, 2, 4, 7)
> #endif
> #ifdef HAS_RGB565TOARGBROW_MSA
> ANY11(RGB565ToARGBRow_Any_MSA, RGB565ToARGBRow_MSA, 0, 2, 4, 15)
> #endif
>+#ifdef HAS_RGB565TOARGBROW_MMI
>+ANY11(RGB565ToARGBRow_Any_MMI, RGB565ToARGBRow_MMI, 0, 2, 4, 3)
>+#endif
> #ifdef HAS_ARGB1555TOARGBROW_NEON
> ANY11(ARGB1555ToARGBRow_Any_NEON, ARGB1555ToARGBRow_NEON, 0, 2, 4, 7)
> #endif
> #ifdef HAS_ARGB1555TOARGBROW_MSA
> ANY11(ARGB1555ToARGBRow_Any_MSA, ARGB1555ToARGBRow_MSA, 0, 2, 4, 15)
> #endif
>+#ifdef HAS_ARGB1555TOARGBROW_MMI
>+ANY11(ARGB1555ToARGBRow_Any_MMI, ARGB1555ToARGBRow_MMI, 0, 2, 4, 3)
>+#endif
> #ifdef HAS_ARGB4444TOARGBROW_NEON
> ANY11(ARGB4444ToARGBRow_Any_NEON, ARGB4444ToARGBRow_NEON, 0, 2, 4, 7)
> #endif
> #ifdef HAS_ARGB4444TOARGBROW_MSA
> ANY11(ARGB4444ToARGBRow_Any_MSA, ARGB4444ToARGBRow_MSA, 0, 2, 4, 15)
> #endif
>+#ifdef HAS_ARGB4444TOARGBROW_MMI
>+ANY11(ARGB4444ToARGBRow_Any_MMI, ARGB4444ToARGBRow_MMI, 0, 2, 4, 3)
>+#endif
> #ifdef HAS_ARGBATTENUATEROW_SSSE3
> ANY11(ARGBAttenuateRow_Any_SSSE3, ARGBAttenuateRow_SSSE3, 0, 4, 4, 3)
> #endif
>@@ -647,11 +765,14 @@ ANY11(ARGBAttenuateRow_Any_NEON, ARGBAttenuateRow_NEON, 0, 4, 4, 7)
> #ifdef HAS_ARGBATTENUATEROW_MSA
> ANY11(ARGBAttenuateRow_Any_MSA, ARGBAttenuateRow_MSA, 0, 4, 4, 7)
> #endif
>+#ifdef HAS_ARGBATTENUATEROW_MMI
>+ANY11(ARGBAttenuateRow_Any_MMI, ARGBAttenuateRow_MMI, 0, 4, 4, 1)
>+#endif
> #ifdef HAS_ARGBEXTRACTALPHAROW_SSE2
> ANY11(ARGBExtractAlphaRow_Any_SSE2, ARGBExtractAlphaRow_SSE2, 0, 4, 1, 7)
> #endif
> #ifdef HAS_ARGBEXTRACTALPHAROW_AVX2
>-ANY11(ARGBExtractAlphaRow_Any_AVX2, ARGBExtractAlphaRow_AVX2, 0, 4, 1, 32)
>+ANY11(ARGBExtractAlphaRow_Any_AVX2, ARGBExtractAlphaRow_AVX2, 0, 4, 1, 31)
> #endif
> #ifdef HAS_ARGBEXTRACTALPHAROW_NEON
> ANY11(ARGBExtractAlphaRow_Any_NEON, ARGBExtractAlphaRow_NEON, 0, 4, 1, 15)
>@@ -659,6 +780,9 @@ ANY11(ARGBExtractAlphaRow_Any_NEON, ARGBExtractAlphaRow_NEON, 0, 4, 1, 15)
> #ifdef HAS_ARGBEXTRACTALPHAROW_MSA
> ANY11(ARGBExtractAlphaRow_Any_MSA, ARGBExtractAlphaRow_MSA, 0, 4, 1, 15)
> #endif
>+#ifdef HAS_ARGBEXTRACTALPHAROW_MMI
>+ANY11(ARGBExtractAlphaRow_Any_MMI, ARGBExtractAlphaRow_MMI, 0, 4, 1, 7)
>+#endif
> #undef ANY11
> 
> // Any 1 to 1 blended.  Destination is read, modify, write.
>@@ -683,12 +807,18 @@ ANY11B(ARGBCopyAlphaRow_Any_AVX2, ARGBCopyAlphaRow_AVX2, 0, 4, 4, 15)
> #ifdef HAS_ARGBCOPYALPHAROW_SSE2
> ANY11B(ARGBCopyAlphaRow_Any_SSE2, ARGBCopyAlphaRow_SSE2, 0, 4, 4, 7)
> #endif
>+#ifdef HAS_ARGBCOPYALPHAROW_MMI
>+ANY11B(ARGBCopyAlphaRow_Any_MMI, ARGBCopyAlphaRow_MMI, 0, 4, 4, 1)
>+#endif
> #ifdef HAS_ARGBCOPYYTOALPHAROW_AVX2
> ANY11B(ARGBCopyYToAlphaRow_Any_AVX2, ARGBCopyYToAlphaRow_AVX2, 0, 1, 4, 15)
> #endif
> #ifdef HAS_ARGBCOPYYTOALPHAROW_SSE2
> ANY11B(ARGBCopyYToAlphaRow_Any_SSE2, ARGBCopyYToAlphaRow_SSE2, 0, 1, 4, 7)
> #endif
>+#ifdef HAS_ARGBCOPYYTOALPHAROW_MMI
>+ANY11B(ARGBCopyYToAlphaRow_Any_MMI, ARGBCopyYToAlphaRow_MMI, 0, 1, 4, 7)
>+#endif
> #undef ANY11B
> 
> // Any 1 to 1 with parameter.
>@@ -738,6 +868,14 @@ ANY11P(ARGBToRGB565DitherRow_Any_MSA,
>        2,
>        7)
> #endif
>+#if defined(HAS_ARGBTORGB565DITHERROW_MMI)
>+ANY11P(ARGBToRGB565DitherRow_Any_MMI,
>+       ARGBToRGB565DitherRow_MMI,
>+       const uint32_t,
>+       4,
>+       2,
>+       3)
>+#endif
> #ifdef HAS_ARGBSHUFFLEROW_SSSE3
> ANY11P(ARGBShuffleRow_Any_SSSE3, ARGBShuffleRow_SSSE3, const uint8_t*, 4, 4, 7)
> #endif
>@@ -750,6 +888,10 @@ ANY11P(ARGBShuffleRow_Any_NEON, ARGBShuffleRow_NEON, const uint8_t*, 4, 4, 3)
> #ifdef HAS_ARGBSHUFFLEROW_MSA
> ANY11P(ARGBShuffleRow_Any_MSA, ARGBShuffleRow_MSA, const uint8_t*, 4, 4, 7)
> #endif
>+#ifdef HAS_ARGBSHUFFLEROW_MMI
>+ANY11P(ARGBShuffleRow_Any_MMI, ARGBShuffleRow_MMI, const uint8_t*, 4, 4, 1)
>+#endif
>+#undef ANY11P
> #undef ANY11P
> 
> // Any 1 to 1 with parameter and shorts.  BPP measures in shorts.
>@@ -918,6 +1060,9 @@ ANY11T(InterpolateRow_Any_NEON, InterpolateRow_NEON, 1, 1, 15)
> #ifdef HAS_INTERPOLATEROW_MSA
> ANY11T(InterpolateRow_Any_MSA, InterpolateRow_MSA, 1, 1, 31)
> #endif
>+#ifdef HAS_INTERPOLATEROW_MMI
>+ANY11T(InterpolateRow_Any_MMI, InterpolateRow_MMI, 1, 1, 7)
>+#endif
> #undef ANY11T
> 
> // Any 1 to 1 mirror.
>@@ -947,6 +1092,9 @@ ANY11M(MirrorRow_Any_NEON, MirrorRow_NEON, 1, 15)
> #ifdef HAS_MIRRORROW_MSA
> ANY11M(MirrorRow_Any_MSA, MirrorRow_MSA, 1, 63)
> #endif
>+#ifdef HAS_MIRRORROW_MMI
>+ANY11M(MirrorRow_Any_MMI, MirrorRow_MMI, 1, 7)
>+#endif
> #ifdef HAS_ARGBMIRRORROW_AVX2
> ANY11M(ARGBMirrorRow_Any_AVX2, ARGBMirrorRow_AVX2, 4, 7)
> #endif
>@@ -959,6 +1107,9 @@ ANY11M(ARGBMirrorRow_Any_NEON, ARGBMirrorRow_NEON, 4, 3)
> #ifdef HAS_ARGBMIRRORROW_MSA
> ANY11M(ARGBMirrorRow_Any_MSA, ARGBMirrorRow_MSA, 4, 15)
> #endif
>+#ifdef HAS_ARGBMIRRORROW_MMI
>+ANY11M(ARGBMirrorRow_Any_MMI, ARGBMirrorRow_MMI, 4, 1)
>+#endif
> #undef ANY11M
> 
> // Any 1 plane. (memset)
>@@ -1017,6 +1168,9 @@ ANY12(SplitUVRow_Any_NEON, SplitUVRow_NEON, 0, 2, 0, 15)
> #ifdef HAS_SPLITUVROW_MSA
> ANY12(SplitUVRow_Any_MSA, SplitUVRow_MSA, 0, 2, 0, 31)
> #endif
>+#ifdef HAS_SPLITUVROW_MMI
>+ANY12(SplitUVRow_Any_MMI, SplitUVRow_MMI, 0, 2, 0, 7)
>+#endif
> #ifdef HAS_ARGBTOUV444ROW_SSSE3
> ANY12(ARGBToUV444Row_Any_SSSE3, ARGBToUV444Row_SSSE3, 0, 4, 0, 15)
> #endif
>@@ -1038,6 +1192,11 @@ ANY12(ARGBToUV444Row_Any_MSA, ARGBToUV444Row_MSA, 0, 4, 0, 15)
> ANY12(YUY2ToUV422Row_Any_MSA, YUY2ToUV422Row_MSA, 1, 4, 1, 31)
> ANY12(UYVYToUV422Row_Any_MSA, UYVYToUV422Row_MSA, 1, 4, 1, 31)
> #endif
>+#ifdef HAS_YUY2TOUV422ROW_MMI
>+ANY12(ARGBToUV444Row_Any_MMI, ARGBToUV444Row_MMI, 0, 4, 0, 7)
>+ANY12(UYVYToUV422Row_Any_MMI, UYVYToUV422Row_MMI, 1, 4, 1, 15)
>+ANY12(YUY2ToUV422Row_Any_MMI, YUY2ToUV422Row_MMI, 1, 4, 1, 15)
>+#endif
> #undef ANY12
> 
> // Any 1 to 3.  Outputs RGB planes.
>@@ -1064,6 +1223,9 @@ ANY13(SplitRGBRow_Any_SSSE3, SplitRGBRow_SSSE3, 3, 15)
> #ifdef HAS_SPLITRGBROW_NEON
> ANY13(SplitRGBRow_Any_NEON, SplitRGBRow_NEON, 3, 15)
> #endif
>+#ifdef HAS_SPLITRGBROW_MMI
>+ANY13(SplitRGBRow_Any_MMI, SplitRGBRow_MMI, 3, 3)
>+#endif
> 
> // Any 1 to 2 with source stride (2 rows of source).  Outputs UV planes.
> // 128 byte row allows for 32 avx ARGB pixels.
>@@ -1118,57 +1280,87 @@ ANY12S(ARGBToUVRow_Any_NEON, ARGBToUVRow_NEON, 0, 4, 15)
> #ifdef HAS_ARGBTOUVROW_MSA
> ANY12S(ARGBToUVRow_Any_MSA, ARGBToUVRow_MSA, 0, 4, 31)
> #endif
>+#ifdef HAS_ARGBTOUVROW_MMI
>+ANY12S(ARGBToUVRow_Any_MMI, ARGBToUVRow_MMI, 0, 4, 15)
>+#endif
> #ifdef HAS_ARGBTOUVJROW_NEON
> ANY12S(ARGBToUVJRow_Any_NEON, ARGBToUVJRow_NEON, 0, 4, 15)
> #endif
> #ifdef HAS_ARGBTOUVJROW_MSA
> ANY12S(ARGBToUVJRow_Any_MSA, ARGBToUVJRow_MSA, 0, 4, 31)
> #endif
>+#ifdef HAS_ARGBTOUVJROW_MMI
>+ANY12S(ARGBToUVJRow_Any_MMI, ARGBToUVJRow_MMI, 0, 4, 15)
>+#endif
> #ifdef HAS_BGRATOUVROW_NEON
> ANY12S(BGRAToUVRow_Any_NEON, BGRAToUVRow_NEON, 0, 4, 15)
> #endif
> #ifdef HAS_BGRATOUVROW_MSA
> ANY12S(BGRAToUVRow_Any_MSA, BGRAToUVRow_MSA, 0, 4, 31)
> #endif
>+#ifdef HAS_BGRATOUVROW_MMI
>+ANY12S(BGRAToUVRow_Any_MMI, BGRAToUVRow_MMI, 0, 4, 15)
>+#endif
> #ifdef HAS_ABGRTOUVROW_NEON
> ANY12S(ABGRToUVRow_Any_NEON, ABGRToUVRow_NEON, 0, 4, 15)
> #endif
> #ifdef HAS_ABGRTOUVROW_MSA
> ANY12S(ABGRToUVRow_Any_MSA, ABGRToUVRow_MSA, 0, 4, 31)
> #endif
>+#ifdef HAS_ABGRTOUVROW_MMI
>+ANY12S(ABGRToUVRow_Any_MMI, ABGRToUVRow_MMI, 0, 4, 15)
>+#endif
> #ifdef HAS_RGBATOUVROW_NEON
> ANY12S(RGBAToUVRow_Any_NEON, RGBAToUVRow_NEON, 0, 4, 15)
> #endif
> #ifdef HAS_RGBATOUVROW_MSA
> ANY12S(RGBAToUVRow_Any_MSA, RGBAToUVRow_MSA, 0, 4, 31)
> #endif
>+#ifdef HAS_RGBATOUVROW_MMI
>+ANY12S(RGBAToUVRow_Any_MMI, RGBAToUVRow_MMI, 0, 4, 15)
>+#endif
> #ifdef HAS_RGB24TOUVROW_NEON
> ANY12S(RGB24ToUVRow_Any_NEON, RGB24ToUVRow_NEON, 0, 3, 15)
> #endif
> #ifdef HAS_RGB24TOUVROW_MSA
> ANY12S(RGB24ToUVRow_Any_MSA, RGB24ToUVRow_MSA, 0, 3, 15)
> #endif
>+#ifdef HAS_RGB24TOUVROW_MMI
>+ANY12S(RGB24ToUVRow_Any_MMI, RGB24ToUVRow_MMI, 0, 3, 15)
>+#endif
> #ifdef HAS_RAWTOUVROW_NEON
> ANY12S(RAWToUVRow_Any_NEON, RAWToUVRow_NEON, 0, 3, 15)
> #endif
> #ifdef HAS_RAWTOUVROW_MSA
> ANY12S(RAWToUVRow_Any_MSA, RAWToUVRow_MSA, 0, 3, 15)
> #endif
>+#ifdef HAS_RAWTOUVROW_MMI
>+ANY12S(RAWToUVRow_Any_MMI, RAWToUVRow_MMI, 0, 3, 15)
>+#endif
> #ifdef HAS_RGB565TOUVROW_NEON
> ANY12S(RGB565ToUVRow_Any_NEON, RGB565ToUVRow_NEON, 0, 2, 15)
> #endif
> #ifdef HAS_RGB565TOUVROW_MSA
> ANY12S(RGB565ToUVRow_Any_MSA, RGB565ToUVRow_MSA, 0, 2, 15)
> #endif
>+#ifdef HAS_RGB565TOUVROW_MMI
>+ANY12S(RGB565ToUVRow_Any_MMI, RGB565ToUVRow_MMI, 0, 2, 15)
>+#endif
> #ifdef HAS_ARGB1555TOUVROW_NEON
> ANY12S(ARGB1555ToUVRow_Any_NEON, ARGB1555ToUVRow_NEON, 0, 2, 15)
> #endif
> #ifdef HAS_ARGB1555TOUVROW_MSA
> ANY12S(ARGB1555ToUVRow_Any_MSA, ARGB1555ToUVRow_MSA, 0, 2, 15)
> #endif
>+#ifdef HAS_ARGB1555TOUVROW_MMI
>+ANY12S(ARGB1555ToUVRow_Any_MMI, ARGB1555ToUVRow_MMI, 0, 2, 15)
>+#endif
> #ifdef HAS_ARGB4444TOUVROW_NEON
> ANY12S(ARGB4444ToUVRow_Any_NEON, ARGB4444ToUVRow_NEON, 0, 2, 15)
> #endif
>+#ifdef HAS_ARGB4444TOUVROW_MMI
>+ANY12S(ARGB4444ToUVRow_Any_MMI, ARGB4444ToUVRow_MMI, 0, 2, 15)
>+#endif
> #ifdef HAS_YUY2TOUVROW_NEON
> ANY12S(YUY2ToUVRow_Any_NEON, YUY2ToUVRow_NEON, 1, 4, 15)
> #endif
>@@ -1178,9 +1370,15 @@ ANY12S(UYVYToUVRow_Any_NEON, UYVYToUVRow_NEON, 1, 4, 15)
> #ifdef HAS_YUY2TOUVROW_MSA
> ANY12S(YUY2ToUVRow_Any_MSA, YUY2ToUVRow_MSA, 1, 4, 31)
> #endif
>+#ifdef HAS_YUY2TOUVROW_MMI
>+ANY12S(YUY2ToUVRow_Any_MMI, YUY2ToUVRow_MMI, 1, 4, 15)
>+#endif
> #ifdef HAS_UYVYTOUVROW_MSA
> ANY12S(UYVYToUVRow_Any_MSA, UYVYToUVRow_MSA, 1, 4, 31)
> #endif
>+#ifdef HAS_UYVYTOUVROW_MMI
>+ANY12S(UYVYToUVRow_Any_MMI, UYVYToUVRow_MMI, 1, 4, 15)
>+#endif
> #undef ANY12S
> 
> #ifdef __cplusplus
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/row_common.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/row_common.cc
>index da97821f796440dd12beb6e05f4ff626842d2bd6..2bbc5adbf145f36231b203bf7f1cbea4468b9965 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/row_common.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/row_common.cc
>@@ -181,7 +181,7 @@ void ARGB4444ToARGBRow_C(const uint8_t* src_argb4444,
> void AR30ToARGBRow_C(const uint8_t* src_ar30, uint8_t* dst_argb, int width) {
>   int x;
>   for (x = 0; x < width; ++x) {
>-    uint32_t ar30 = *(uint32_t*)src_ar30;
>+    uint32_t ar30 = *(const uint32_t*)src_ar30;
>     uint32_t b = (ar30 >> 2) & 0xff;
>     uint32_t g = (ar30 >> 12) & 0xff;
>     uint32_t r = (ar30 >> 22) & 0xff;
>@@ -195,7 +195,7 @@ void AR30ToARGBRow_C(const uint8_t* src_ar30, uint8_t* dst_argb, int width) {
> void AR30ToABGRRow_C(const uint8_t* src_ar30, uint8_t* dst_abgr, int width) {
>   int x;
>   for (x = 0; x < width; ++x) {
>-    uint32_t ar30 = *(uint32_t*)src_ar30;
>+    uint32_t ar30 = *(const uint32_t*)src_ar30;
>     uint32_t b = (ar30 >> 2) & 0xff;
>     uint32_t g = (ar30 >> 12) & 0xff;
>     uint32_t r = (ar30 >> 22) & 0xff;
>@@ -209,7 +209,7 @@ void AR30ToABGRRow_C(const uint8_t* src_ar30, uint8_t* dst_abgr, int width) {
> void AR30ToAB30Row_C(const uint8_t* src_ar30, uint8_t* dst_ab30, int width) {
>   int x;
>   for (x = 0; x < width; ++x) {
>-    uint32_t ar30 = *(uint32_t*)src_ar30;
>+    uint32_t ar30 = *(const uint32_t*)src_ar30;
>     uint32_t b = ar30 & 0x3ff;
>     uint32_t ga = ar30 & 0xc00ffc00;
>     uint32_t r = (ar30 >> 20) & 0x3ff;
>@@ -2762,6 +2762,13 @@ void ARGBPolynomialRow_C(const uint8_t* src_argb,
> // simply extract the low bits of the exponent and the high
> // bits of the mantissa from our float and we're done.
> 
>+// Work around GCC 7 punning warning -Wstrict-aliasing
>+#if defined(__GNUC__)
>+typedef uint32_t __attribute__((__may_alias__)) uint32_alias_t;
>+#else
>+typedef uint32_t uint32_alias_t;
>+#endif
>+
> void HalfFloatRow_C(const uint16_t* src,
>                     uint16_t* dst,
>                     float scale,
>@@ -2770,7 +2777,7 @@ void HalfFloatRow_C(const uint16_t* src,
>   float mult = 1.9259299444e-34f * scale;
>   for (i = 0; i < width; ++i) {
>     float value = src[i] * mult;
>-    dst[i] = (uint16_t)((*(uint32_t*)&value) >> 13);
>+    dst[i] = (uint16_t)((*(const uint32_alias_t*)&value) >> 13);
>   }
> }
> 
>@@ -2953,6 +2960,94 @@ void NV12ToRGB565Row_SSSE3(const uint8_t* src_y,
> }
> #endif
> 
>+#if defined(HAS_NV12TORGB24ROW_SSSE3)
>+void NV12ToRGB24Row_SSSE3(const uint8_t* src_y,
>+                          const uint8_t* src_uv,
>+                          uint8_t* dst_rgb24,
>+                          const struct YuvConstants* yuvconstants,
>+                          int width) {
>+  // Row buffer for intermediate ARGB pixels.
>+  SIMD_ALIGNED(uint8_t row[MAXTWIDTH * 4]);
>+  while (width > 0) {
>+    int twidth = width > MAXTWIDTH ? MAXTWIDTH : width;
>+    NV12ToARGBRow_SSSE3(src_y, src_uv, row, yuvconstants, twidth);
>+    ARGBToRGB24Row_SSSE3(row, dst_rgb24, twidth);
>+    src_y += twidth;
>+    src_uv += twidth;
>+    dst_rgb24 += twidth * 3;
>+    width -= twidth;
>+  }
>+}
>+#endif
>+
>+#if defined(HAS_NV21TORGB24ROW_SSSE3)
>+void NV21ToRGB24Row_SSSE3(const uint8_t* src_y,
>+                          const uint8_t* src_vu,
>+                          uint8_t* dst_rgb24,
>+                          const struct YuvConstants* yuvconstants,
>+                          int width) {
>+  // Row buffer for intermediate ARGB pixels.
>+  SIMD_ALIGNED(uint8_t row[MAXTWIDTH * 4]);
>+  while (width > 0) {
>+    int twidth = width > MAXTWIDTH ? MAXTWIDTH : width;
>+    NV21ToARGBRow_SSSE3(src_y, src_vu, row, yuvconstants, twidth);
>+    ARGBToRGB24Row_SSSE3(row, dst_rgb24, twidth);
>+    src_y += twidth;
>+    src_vu += twidth;
>+    dst_rgb24 += twidth * 3;
>+    width -= twidth;
>+  }
>+}
>+#endif
>+
>+#if defined(HAS_NV12TORGB24ROW_AVX2)
>+void NV12ToRGB24Row_AVX2(const uint8_t* src_y,
>+                         const uint8_t* src_uv,
>+                         uint8_t* dst_rgb24,
>+                         const struct YuvConstants* yuvconstants,
>+                         int width) {
>+  // Row buffer for intermediate ARGB pixels.
>+  SIMD_ALIGNED(uint8_t row[MAXTWIDTH * 4]);
>+  while (width > 0) {
>+    int twidth = width > MAXTWIDTH ? MAXTWIDTH : width;
>+    NV12ToARGBRow_AVX2(src_y, src_uv, row, yuvconstants, twidth);
>+#if defined(HAS_ARGBTORGB24ROW_AVX2)
>+    ARGBToRGB24Row_AVX2(row, dst_rgb24, twidth);
>+#else
>+    ARGBToRGB24Row_SSSE3(row, dst_rgb24, twidth);
>+#endif
>+    src_y += twidth;
>+    src_uv += twidth;
>+    dst_rgb24 += twidth * 3;
>+    width -= twidth;
>+  }
>+}
>+#endif
>+
>+#if defined(HAS_NV21TORGB24ROW_AVX2)
>+void NV21ToRGB24Row_AVX2(const uint8_t* src_y,
>+                         const uint8_t* src_vu,
>+                         uint8_t* dst_rgb24,
>+                         const struct YuvConstants* yuvconstants,
>+                         int width) {
>+  // Row buffer for intermediate ARGB pixels.
>+  SIMD_ALIGNED(uint8_t row[MAXTWIDTH * 4]);
>+  while (width > 0) {
>+    int twidth = width > MAXTWIDTH ? MAXTWIDTH : width;
>+    NV21ToARGBRow_AVX2(src_y, src_vu, row, yuvconstants, twidth);
>+#if defined(HAS_ARGBTORGB24ROW_AVX2)
>+    ARGBToRGB24Row_AVX2(row, dst_rgb24, twidth);
>+#else
>+    ARGBToRGB24Row_SSSE3(row, dst_rgb24, twidth);
>+#endif
>+    src_y += twidth;
>+    src_vu += twidth;
>+    dst_rgb24 += twidth * 3;
>+    width -= twidth;
>+  }
>+}
>+#endif
>+
> #if defined(HAS_I422TORGB565ROW_AVX2)
> void I422ToRGB565Row_AVX2(const uint8_t* src_y,
>                           const uint8_t* src_u,
>@@ -3042,8 +3137,11 @@ void I422ToRGB24Row_AVX2(const uint8_t* src_y,
>   while (width > 0) {
>     int twidth = width > MAXTWIDTH ? MAXTWIDTH : width;
>     I422ToARGBRow_AVX2(src_y, src_u, src_v, row, yuvconstants, twidth);
>-    // TODO(fbarchard): ARGBToRGB24Row_AVX2
>+#if defined(HAS_ARGBTORGB24ROW_AVX2)
>+    ARGBToRGB24Row_AVX2(row, dst_rgb24, twidth);
>+#else
>     ARGBToRGB24Row_SSSE3(row, dst_rgb24, twidth);
>+#endif
>     src_y += twidth;
>     src_u += twidth / 2;
>     src_v += twidth / 2;
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/row_gcc.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/row_gcc.cc
>index 95845c2592f7c84307c37b00bf18274ea83b8cc9..8d3cb81cec2e25e2addc7f6e52f351bb616a4a6b 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/row_gcc.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/row_gcc.cc
>@@ -505,6 +505,149 @@ void ARGBToRAWRow_SSSE3(const uint8_t* src, uint8_t* dst, int width) {
>       : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6");
> }
> 
>+#ifdef HAS_ARGBTORGB24ROW_AVX2
>+// vpermd for 12+12 to 24
>+static const lvec32 kPermdRGB24_AVX = {0, 1, 2, 4, 5, 6, 3, 7};
>+
>+void ARGBToRGB24Row_AVX2(const uint8_t* src, uint8_t* dst, int width) {
>+  asm volatile(
>+      "vbroadcastf128 %3,%%ymm6                  \n"
>+      "vmovdqa    %4,%%ymm7                      \n"
>+
>+      LABELALIGN
>+      "1:                                        \n"
>+      "vmovdqu    (%0),%%ymm0                    \n"
>+      "vmovdqu    0x20(%0),%%ymm1                \n"
>+      "vmovdqu    0x40(%0),%%ymm2                \n"
>+      "vmovdqu    0x60(%0),%%ymm3                \n"
>+      "lea        0x80(%0),%0                    \n"
>+      "vpshufb    %%ymm6,%%ymm0,%%ymm0           \n"  // xxx0yyy0
>+      "vpshufb    %%ymm6,%%ymm1,%%ymm1           \n"
>+      "vpshufb    %%ymm6,%%ymm2,%%ymm2           \n"
>+      "vpshufb    %%ymm6,%%ymm3,%%ymm3           \n"
>+      "vpermd     %%ymm0,%%ymm7,%%ymm0           \n"  // pack to 24 bytes
>+      "vpermd     %%ymm1,%%ymm7,%%ymm1           \n"
>+      "vpermd     %%ymm2,%%ymm7,%%ymm2           \n"
>+      "vpermd     %%ymm3,%%ymm7,%%ymm3           \n"
>+      "vpermq     $0x3f,%%ymm1,%%ymm4            \n"  // combine 24 + 8
>+      "vpor       %%ymm4,%%ymm0,%%ymm0           \n"
>+      "vmovdqu    %%ymm0,(%1)                    \n"
>+      "vpermq     $0xf9,%%ymm1,%%ymm1            \n"  // combine 16 + 16
>+      "vpermq     $0x4f,%%ymm2,%%ymm4            \n"
>+      "vpor       %%ymm4,%%ymm1,%%ymm1           \n"
>+      "vmovdqu    %%ymm1,0x20(%1)                \n"
>+      "vpermq     $0xfe,%%ymm2,%%ymm2            \n"  // combine 8 + 24
>+      "vpermq     $0x93,%%ymm3,%%ymm3            \n"
>+      "vpor       %%ymm3,%%ymm2,%%ymm2           \n"
>+      "vmovdqu    %%ymm2,0x40(%1)                \n"
>+      "lea        0x60(%1),%1                    \n"
>+      "sub        $0x20,%2                       \n"
>+      "jg         1b                             \n"
>+      "vzeroupper                                \n"
>+      : "+r"(src),                     // %0
>+        "+r"(dst),                     // %1
>+        "+r"(width)                    // %2
>+      : "m"(kShuffleMaskARGBToRGB24),  // %3
>+        "m"(kPermdRGB24_AVX)           // %4
>+      : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
>+        "xmm7");
>+}
>+#endif
>+
>+#ifdef HAS_ARGBTORGB24ROW_AVX512VBMI
>+// Shuffle table for converting ARGBToRGB24
>+static const ulvec8 kPermARGBToRGB24_0 = {
>+    0u,  1u,  2u,  4u,  5u,  6u,  8u,  9u,  10u, 12u, 13u,
>+    14u, 16u, 17u, 18u, 20u, 21u, 22u, 24u, 25u, 26u, 28u,
>+    29u, 30u, 32u, 33u, 34u, 36u, 37u, 38u, 40u, 41u};
>+static const ulvec8 kPermARGBToRGB24_1 = {
>+    10u, 12u, 13u, 14u, 16u, 17u, 18u, 20u, 21u, 22u, 24u,
>+    25u, 26u, 28u, 29u, 30u, 32u, 33u, 34u, 36u, 37u, 38u,
>+    40u, 41u, 42u, 44u, 45u, 46u, 48u, 49u, 50u, 52u};
>+static const ulvec8 kPermARGBToRGB24_2 = {
>+    21u, 22u, 24u, 25u, 26u, 28u, 29u, 30u, 32u, 33u, 34u,
>+    36u, 37u, 38u, 40u, 41u, 42u, 44u, 45u, 46u, 48u, 49u,
>+    50u, 52u, 53u, 54u, 56u, 57u, 58u, 60u, 61u, 62u};
>+
>+void ARGBToRGB24Row_AVX512VBMI(const uint8_t* src, uint8_t* dst, int width) {
>+  asm volatile(
>+      "vmovdqa    %3,%%ymm5                      \n"
>+      "vmovdqa    %4,%%ymm6                      \n"
>+      "vmovdqa    %5,%%ymm7                      \n"
>+
>+      LABELALIGN
>+      "1:                                        \n"
>+      "vmovdqu    (%0),%%ymm0                    \n"
>+      "vmovdqu    0x20(%0),%%ymm1                \n"
>+      "vmovdqu    0x40(%0),%%ymm2                \n"
>+      "vmovdqu    0x60(%0),%%ymm3                \n"
>+      "lea        0x80(%0),%0                    \n"
>+      "vpermt2b   %%ymm1,%%ymm5,%%ymm0           \n"
>+      "vpermt2b   %%ymm2,%%ymm6,%%ymm1           \n"
>+      "vpermt2b   %%ymm3,%%ymm7,%%ymm2           \n"
>+      "vmovdqu    %%ymm0,(%1)                    \n"
>+      "vmovdqu    %%ymm1,0x20(%1)                \n"
>+      "vmovdqu    %%ymm2,0x40(%1)                \n"
>+      "lea        0x60(%1),%1                    \n"
>+      "sub        $0x20,%2                       \n"
>+      "jg         1b                             \n"
>+      "vzeroupper                                \n"
>+      : "+r"(src),                // %0
>+        "+r"(dst),                // %1
>+        "+r"(width)               // %2
>+      : "m"(kPermARGBToRGB24_0),  // %3
>+        "m"(kPermARGBToRGB24_1),  // %4
>+        "m"(kPermARGBToRGB24_2)   // %5
>+      : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm5", "xmm6", "xmm7");
>+}
>+#endif
>+
>+#ifdef HAS_ARGBTORAWROW_AVX2
>+void ARGBToRAWRow_AVX2(const uint8_t* src, uint8_t* dst, int width) {
>+  asm volatile(
>+      "vbroadcastf128 %3,%%ymm6                  \n"
>+      "vmovdqa    %4,%%ymm7                      \n"
>+
>+      LABELALIGN
>+      "1:                                        \n"
>+      "vmovdqu    (%0),%%ymm0                    \n"
>+      "vmovdqu    0x20(%0),%%ymm1                \n"
>+      "vmovdqu    0x40(%0),%%ymm2                \n"
>+      "vmovdqu    0x60(%0),%%ymm3                \n"
>+      "lea        0x80(%0),%0                    \n"
>+      "vpshufb    %%ymm6,%%ymm0,%%ymm0           \n"  // xxx0yyy0
>+      "vpshufb    %%ymm6,%%ymm1,%%ymm1           \n"
>+      "vpshufb    %%ymm6,%%ymm2,%%ymm2           \n"
>+      "vpshufb    %%ymm6,%%ymm3,%%ymm3           \n"
>+      "vpermd     %%ymm0,%%ymm7,%%ymm0           \n"  // pack to 24 bytes
>+      "vpermd     %%ymm1,%%ymm7,%%ymm1           \n"
>+      "vpermd     %%ymm2,%%ymm7,%%ymm2           \n"
>+      "vpermd     %%ymm3,%%ymm7,%%ymm3           \n"
>+      "vpermq     $0x3f,%%ymm1,%%ymm4            \n"  // combine 24 + 8
>+      "vpor       %%ymm4,%%ymm0,%%ymm0           \n"
>+      "vmovdqu    %%ymm0,(%1)                    \n"
>+      "vpermq     $0xf9,%%ymm1,%%ymm1            \n"  // combine 16 + 16
>+      "vpermq     $0x4f,%%ymm2,%%ymm4            \n"
>+      "vpor       %%ymm4,%%ymm1,%%ymm1           \n"
>+      "vmovdqu    %%ymm1,0x20(%1)                \n"
>+      "vpermq     $0xfe,%%ymm2,%%ymm2            \n"  // combine 8 + 24
>+      "vpermq     $0x93,%%ymm3,%%ymm3            \n"
>+      "vpor       %%ymm3,%%ymm2,%%ymm2           \n"
>+      "vmovdqu    %%ymm2,0x40(%1)                \n"
>+      "lea        0x60(%1),%1                    \n"
>+      "sub        $0x20,%2                       \n"
>+      "jg         1b                             \n"
>+      "vzeroupper                                \n"
>+      : "+r"(src),                   // %0
>+        "+r"(dst),                   // %1
>+        "+r"(width)                  // %2
>+      : "m"(kShuffleMaskARGBToRAW),  // %3
>+        "m"(kPermdRGB24_AVX)         // %4
>+      : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
>+        "xmm7");
>+}
>+#endif
>+
> void ARGBToRGB565Row_SSE2(const uint8_t* src, uint8_t* dst, int width) {
>   asm volatile(
>       "pcmpeqb   %%xmm3,%%xmm3                   \n"
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/row_mmi.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/row_mmi.cc
>new file mode 100644
>index 0000000000000000000000000000000000000000..d8726d09374d90457ea9de67f28dbc9c9db262f4
>--- /dev/null
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/row_mmi.cc
>@@ -0,0 +1,6042 @@
>+/*
>+ *  Copyright 2011 The LibYuv Project Authors. All rights reserved.
>+ *
>+ *  Use of this source code is governed by a BSD-style license
>+ *  that can be found in the LICENSE file in the root of the source
>+ *  tree. An additional intellectual property rights grant can be found
>+ *  in the file PATENTS. All contributing project authors may
>+ *  be found in the AUTHORS file in the root of the source tree.
>+ */
>+#include "libyuv/row.h"
>+
>+#include <string.h>  // For memcpy and memset.
>+
>+#include "libyuv/basic_types.h"
>+
>+#ifdef __cplusplus
>+namespace libyuv {
>+extern "C" {
>+#endif
>+
>+// This module is for Mips MMI.
>+#if !defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)
>+
>+void RGB24ToARGBRow_MMI(const uint8_t* src_rgb24,
>+                        uint8_t* dst_argb,
>+                        int width) {
>+  uint64_t src0, src1, dest;
>+  const uint64_t mask = 0xff000000ULL;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gslwlc1    %[src0],         0x03(%[src_ptr])                 \n\t"
>+      "gslwrc1    %[src0],         0x00(%[src_ptr])                 \n\t"
>+      "gslwlc1    %[src1],         0x06(%[src_ptr])                 \n\t"
>+      "gslwrc1    %[src1],         0x03(%[src_ptr])                 \n\t"
>+
>+      "or         %[src0],         %[src0],           %[mask]       \n\t"
>+      "or         %[src1],         %[src1],           %[mask]       \n\t"
>+      "punpcklwd  %[dest],         %[src0],           %[src1]       \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "gslwlc1    %[src0],         0x09(%[src_ptr])                 \n\t"
>+      "gslwrc1    %[src0],         0x06(%[src_ptr])                 \n\t"
>+      "gslwlc1    %[src1],         0x0c(%[src_ptr])                 \n\t"
>+      "gslwrc1    %[src1],         0x09(%[src_ptr])                 \n\t"
>+
>+      "or         %[src0],         %[src0],           %[mask]       \n\t"
>+      "or         %[src1],         %[src1],           %[mask]       \n\t"
>+      "punpcklwd  %[dest],         %[src0],           %[src1]       \n\t"
>+      "gssdlc1    %[dest],         0x0f(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x08(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x0c          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x10          \n\t"
>+      "daddi      %[width],        %[width],         -0x04          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src0] "=&f"(src0), [src1] "=&f"(src1), [dest] "=&f"(dest)
>+      : [src_ptr] "r"(src_rgb24), [dst_ptr] "r"(dst_argb), [width] "r"(width),
>+        [mask] "f"(mask)
>+      : "memory");
>+}
>+
>+void RAWToARGBRow_MMI(const uint8_t* src_raw, uint8_t* dst_argb, int width) {
>+  uint64_t src0, src1, dest;
>+  const uint64_t mask0 = 0x0;
>+  const uint64_t mask1 = 0xff000000ULL;
>+  const uint64_t mask2 = 0xc6;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gslwlc1    %[src0],         0x03(%[src_ptr])                 \n\t"
>+      "gslwrc1    %[src0],         0x00(%[src_ptr])                 \n\t"
>+      "gslwlc1    %[src1],         0x06(%[src_ptr])                 \n\t"
>+      "gslwrc1    %[src1],         0x03(%[src_ptr])                 \n\t"
>+
>+      "or         %[src0],         %[src0],           %[mask1]      \n\t"
>+      "punpcklbh  %[src0],         %[src0],           %[mask0]      \n\t"
>+      "pshufh     %[src0],         %[src0],           %[mask2]      \n\t"
>+      "or         %[src1],         %[src1],           %[mask1]      \n\t"
>+      "punpcklbh  %[src1],         %[src1],           %[mask0]      \n\t"
>+      "pshufh     %[src1],         %[src1],           %[mask2]      \n\t"
>+      "packushb   %[dest],         %[src0],           %[src1]       \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "gslwlc1    %[src0],         0x09(%[src_ptr])                 \n\t"
>+      "gslwrc1    %[src0],         0x06(%[src_ptr])                 \n\t"
>+      "gslwlc1    %[src1],         0x0c(%[src_ptr])                 \n\t"
>+      "gslwrc1    %[src1],         0x09(%[src_ptr])                 \n\t"
>+
>+      "or         %[src0],         %[src0],           %[mask1]      \n\t"
>+      "punpcklbh  %[src0],         %[src0],           %[mask0]      \n\t"
>+      "pshufh     %[src0],         %[src0],           %[mask2]      \n\t"
>+      "or         %[src1],         %[src1],           %[mask1]      \n\t"
>+      "punpcklbh  %[src1],         %[src1],           %[mask0]      \n\t"
>+      "pshufh     %[src1],         %[src1],           %[mask2]      \n\t"
>+      "packushb   %[dest],         %[src0],           %[src1]       \n\t"
>+      "gssdlc1    %[dest],         0x0f(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x08(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x0c          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x10          \n\t"
>+      "daddi      %[width],        %[width],         -0x04          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src0] "=&f"(src0), [src1] "=&f"(src1), [dest] "=&f"(dest)
>+      : [src_ptr] "r"(src_raw), [dst_ptr] "r"(dst_argb), [mask0] "f"(mask0),
>+        [mask1] "f"(mask1), [mask2] "f"(mask2), [width] "r"(width)
>+      : "memory");
>+}
>+
>+void RAWToRGB24Row_MMI(const uint8_t* src_raw, uint8_t* dst_rgb24, int width) {
>+  uint64_t src0, src1;
>+  uint64_t ftmp[4];
>+  uint64_t mask0 = 0xc6;
>+  uint64_t mask1 = 0x6c;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src_raw])                 \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src_raw])                 \n\t"
>+      "gslwrc1    %[src1],         0x08(%[src_raw])                 \n\t"
>+      "gslwlc1    %[src1],         0x0b(%[src_raw])                 \n\t"
>+
>+      "punpcklbh  %[ftmp0],        %[src0],           %[zero]       \n\t"
>+      "pshufh     %[ftmp0],        %[ftmp0],          %[mask0]      \n\t"
>+      "punpckhbh  %[ftmp1],        %[src0],           %[zero]       \n\t"
>+      "punpcklbh  %[src1],         %[src1],           %[zero]       \n\t"
>+      "pextrh     %[ftmp2],        %[ftmp0],          %[three]      \n\t"
>+      "pextrh     %[ftmp3],        %[ftmp1],          %[one]        \n\t"
>+      "pinsrh_3   %[ftmp0],        %[ftmp0],          %[ftmp3]      \n\t"
>+      "pextrh     %[ftmp3],        %[ftmp1],          %[two]        \n\t"
>+      "pinsrh_1   %[ftmp1],        %[ftmp1],          %[ftmp2]      \n\t"
>+      "pshufh     %[src1],         %[src1],           %[mask1]      \n\t"
>+      "pextrh     %[ftmp2],        %[src1],           %[zero]       \n\t"
>+      "pinsrh_2   %[ftmp1],        %[ftmp1],          %[ftmp2]      \n\t"
>+      "pinsrh_0   %[src1],         %[src1],           %[ftmp3]      \n\t"
>+      "packushb   %[ftmp0],        %[ftmp0],          %[ftmp1]      \n\t"
>+      "packushb   %[src1],         %[src1],           %[zero]       \n\t"
>+
>+      "gssdrc1    %[ftmp0],        0x00(%[dst_rgb24])               \n\t"
>+      "gssdlc1    %[ftmp0],        0x07(%[dst_rgb24])               \n\t"
>+      "gsswrc1    %[src1],         0x08(%[dst_rgb24])               \n\t"
>+      "gsswlc1    %[src1],         0x0b(%[dst_rgb24])               \n\t"
>+
>+      "daddiu     %[src_raw],      %[src_raw],        0x0c          \n\t"
>+      "daddiu     %[dst_rgb24],    %[dst_rgb24],      0x0c          \n\t"
>+      "daddiu     %[width],        %[width],         -0x04          \n\t"
>+      "bgtz       %[width],        1b                               \n\t"
>+      : [src0] "=&f"(src0), [src1] "=&f"(src1), [ftmp0] "=&f"(ftmp[0]),
>+        [ftmp1] "=&f"(ftmp[1]), [ftmp2] "=&f"(ftmp[2]), [ftmp3] "=&f"(ftmp[3])
>+      : [src_raw] "r"(src_raw), [dst_rgb24] "r"(dst_rgb24), [width] "r"(width),
>+        [mask0] "f"(mask0), [mask1] "f"(mask1), [zero] "f"(0x00),
>+        [one] "f"(0x01), [two] "f"(0x02), [three] "f"(0x03)
>+      : "memory");
>+}
>+
>+void RGB565ToARGBRow_MMI(const uint8_t* src_rgb565,
>+                         uint8_t* dst_argb,
>+                         int width) {
>+  uint64_t ftmp[5];
>+  uint64_t c0 = 0x001f001f001f001f;
>+  uint64_t c1 = 0x00ff00ff00ff00ff;
>+  uint64_t c2 = 0x0007000700070007;
>+  __asm__ volatile(
>+      "1:                                                      \n\t"
>+      "gsldrc1   %[src0],       0x00(%[src_rgb565])            \n\t"
>+      "gsldlc1   %[src0],       0x07(%[src_rgb565])            \n\t"
>+      "psrlh     %[src1],       %[src0],             %[eight]  \n\t"
>+      "and       %[b],          %[src0],             %[c0]     \n\t"
>+      "and       %[src0],       %[src0],             %[c1]     \n\t"
>+      "psrlh     %[src0],       %[src0],             %[five]   \n\t"
>+      "and       %[g],          %[src1],             %[c2]     \n\t"
>+      "psllh     %[g],          %[g],                %[three]  \n\t"
>+      "or        %[g],          %[src0],             %[g]      \n\t"
>+      "psrlh     %[r],          %[src1],             %[three]  \n\t"
>+      "psllh     %[src0],       %[b],                %[three]  \n\t"
>+      "psrlh     %[src1],       %[b],                %[two]    \n\t"
>+      "or        %[b],          %[src0],             %[src1]   \n\t"
>+      "psllh     %[src0],       %[g],                %[two]    \n\t"
>+      "psrlh     %[src1],       %[g],                %[four]   \n\t"
>+      "or        %[g],          %[src0],             %[src1]   \n\t"
>+      "psllh     %[src0],       %[r],                %[three]  \n\t"
>+      "psrlh     %[src1],       %[r],                %[two]    \n\t"
>+      "or        %[r],          %[src0],             %[src1]   \n\t"
>+      "packushb  %[b],          %[b],                %[r]      \n\t"
>+      "packushb  %[g],          %[g],                %[c1]     \n\t"
>+      "punpcklbh %[src0],       %[b],                %[g]      \n\t"
>+      "punpckhbh %[src1],       %[b],                %[g]      \n\t"
>+      "punpcklhw %[r],          %[src0],             %[src1]   \n\t"
>+      "gssdrc1   %[r],          0x00(%[dst_argb])              \n\t"
>+      "gssdlc1   %[r],          0x07(%[dst_argb])              \n\t"
>+      "punpckhhw %[r],          %[src0],             %[src1]   \n\t"
>+      "gssdrc1   %[r],          0x08(%[dst_argb])              \n\t"
>+      "gssdlc1   %[r],          0x0f(%[dst_argb])              \n\t"
>+      "daddiu    %[src_rgb565], %[src_rgb565],       0x08      \n\t"
>+      "daddiu    %[dst_argb],   %[dst_argb],         0x10      \n\t"
>+      "daddiu    %[width],      %[width],           -0x04      \n\t"
>+      "bgtz      %[width],     1b                              \n\t"
>+      : [src0] "=&f"(ftmp[0]), [src1] "=&f"(ftmp[1]), [b] "=&f"(ftmp[2]),
>+        [g] "=&f"(ftmp[3]), [r] "=&f"(ftmp[4])
>+      : [src_rgb565] "r"(src_rgb565), [dst_argb] "r"(dst_argb),
>+        [width] "r"(width), [c0] "f"(c0), [c1] "f"(c1), [c2] "f"(c2),
>+        [eight] "f"(0x08), [five] "f"(0x05), [three] "f"(0x03), [two] "f"(0x02),
>+        [four] "f"(0x04)
>+      : "memory");
>+}
>+
>+void ARGB1555ToARGBRow_MMI(const uint8_t* src_argb1555,
>+                           uint8_t* dst_argb,
>+                           int width) {
>+  uint64_t ftmp[6];
>+  uint64_t c0 = 0x001f001f001f001f;
>+  uint64_t c1 = 0x00ff00ff00ff00ff;
>+  uint64_t c2 = 0x0003000300030003;
>+  uint64_t c3 = 0x007c007c007c007c;
>+  uint64_t c4 = 0x0001000100010001;
>+  __asm__ volatile(
>+      "1:                                                         \n\t"
>+      "gsldrc1   %[src0],         0x00(%[src_argb1555])           \n\t"
>+      "gsldlc1   %[src0],         0x07(%[src_argb1555])           \n\t"
>+      "psrlh     %[src1],         %[src0],              %[eight]  \n\t"
>+      "and       %[b],            %[src0],              %[c0]     \n\t"
>+      "and       %[src0],         %[src0],              %[c1]     \n\t"
>+      "psrlh     %[src0],         %[src0],              %[five]   \n\t"
>+      "and       %[g],            %[src1],              %[c2]     \n\t"
>+      "psllh     %[g],            %[g],                 %[three]  \n\t"
>+      "or        %[g],            %[src0],              %[g]      \n\t"
>+      "and       %[r],            %[src1],              %[c3]     \n\t"
>+      "psrlh     %[r],            %[r],                 %[two]    \n\t"
>+      "psrlh     %[a],            %[src1],              %[seven]  \n\t"
>+      "psllh     %[src0],         %[b],                 %[three]  \n\t"
>+      "psrlh     %[src1],         %[b],                 %[two]    \n\t"
>+      "or        %[b],            %[src0],              %[src1]   \n\t"
>+      "psllh     %[src0],         %[g],                 %[three]  \n\t"
>+      "psrlh     %[src1],         %[g],                 %[two]    \n\t"
>+      "or        %[g],            %[src0],              %[src1]   \n\t"
>+      "psllh     %[src0],         %[r],                 %[three]  \n\t"
>+      "psrlh     %[src1],         %[r],                 %[two]    \n\t"
>+      "or        %[r],            %[src0],              %[src1]   \n\t"
>+      "xor       %[a],            %[a],                 %[c1]     \n\t"
>+      "paddb     %[a],            %[a],                 %[c4]     \n\t"
>+      "packushb  %[b],            %[b],                 %[r]      \n\t"
>+      "packushb  %[g],            %[g],                 %[a]      \n\t"
>+      "punpcklbh %[src0],         %[b],                 %[g]      \n\t"
>+      "punpckhbh %[src1],         %[b],                 %[g]      \n\t"
>+      "punpcklhw %[r],            %[src0],              %[src1]   \n\t"
>+      "gssdrc1   %[r],            0x00(%[dst_argb])               \n\t"
>+      "gssdlc1   %[r],            0x07(%[dst_argb])               \n\t"
>+      "punpckhhw %[r],            %[src0],              %[src1]   \n\t"
>+      "gssdrc1   %[r],            0x08(%[dst_argb])               \n\t"
>+      "gssdlc1   %[r],            0x0f(%[dst_argb])               \n\t"
>+      "daddiu    %[src_argb1555], %[src_argb1555],      0x08      \n\t"
>+      "daddiu    %[dst_argb],     %[dst_argb],          0x10      \n\t"
>+      "daddiu    %[width],        %[width],            -0x04      \n\t"
>+      "bgtz      %[width],        1b                              \n\t"
>+      : [src0] "=&f"(ftmp[0]), [src1] "=&f"(ftmp[1]), [b] "=&f"(ftmp[2]),
>+        [g] "=&f"(ftmp[3]), [r] "=&f"(ftmp[4]), [a] "=&f"(ftmp[5])
>+      : [src_argb1555] "r"(src_argb1555), [dst_argb] "r"(dst_argb),
>+        [width] "r"(width), [c0] "f"(c0), [c1] "f"(c1), [c2] "f"(c2),
>+        [c3] "f"(c3), [c4] "f"(c4), [eight] "f"(0x08), [five] "f"(0x05),
>+        [three] "f"(0x03), [two] "f"(0x02), [seven] "f"(0x07)
>+      : "memory");
>+}
>+
>+void ARGB4444ToARGBRow_MMI(const uint8_t* src_argb4444,
>+                           uint8_t* dst_argb,
>+                           int width) {
>+  uint64_t ftmp[6];
>+  uint64_t c0 = 0x000f000f000f000f;
>+  uint64_t c1 = 0x00ff00ff00ff00ff;
>+  __asm__ volatile(
>+      "1:                                                          \n\t"
>+      "gsldrc1   %[src0],         0x00(%[src_argb4444])            \n\t"
>+      "gsldlc1   %[src0],         0x07(%[src_argb4444])            \n\t"
>+      "psrlh     %[src1],         %[src0],              %[eight]   \n\t"
>+      "and       %[b],            %[src0],              %[c0]      \n\t"
>+      "and       %[src0],         %[src0],              %[c1]      \n\t"
>+      "psrlh     %[g],            %[src0],              %[four]    \n\t"
>+      "and       %[r],            %[src1],              %[c0]      \n\t"
>+      "psrlh     %[a],            %[src1],              %[four]    \n\t"
>+      "psllh     %[src0],         %[b],                 %[four]    \n\t"
>+      "or        %[b],            %[src0],              %[b]       \n\t"
>+      "psllh     %[src0],         %[g],                 %[four]    \n\t"
>+      "or        %[g],            %[src0],              %[g]       \n\t"
>+      "psllh     %[src0],         %[r],                 %[four]    \n\t"
>+      "or        %[r],            %[src0],              %[r]       \n\t"
>+      "psllh     %[src0],         %[a],                 %[four]    \n\t"
>+      "or        %[a],            %[src0],              %[a]       \n\t"
>+      "packushb  %[b],            %[b],                 %[r]       \n\t"
>+      "packushb  %[g],            %[g],                 %[a]       \n\t"
>+      "punpcklbh %[src0],         %[b],                 %[g]       \n\t"
>+      "punpckhbh %[src1],         %[b],                 %[g]       \n\t"
>+      "punpcklhw %[r],            %[src0],              %[src1]    \n\t"
>+      "gssdrc1   %[r],            0x00(%[dst_argb])                \n\t"
>+      "gssdlc1   %[r],            0x07(%[dst_argb])                \n\t"
>+      "punpckhhw %[r],            %[src0],              %[src1]    \n\t"
>+      "gssdrc1   %[r],            0x08(%[dst_argb])                \n\t"
>+      "gssdlc1   %[r],            0x0f(%[dst_argb])                \n\t"
>+      "daddiu    %[src_argb4444], %[src_argb4444],      0x08       \n\t"
>+      "daddiu    %[dst_argb],     %[dst_argb],          0x10       \n\t"
>+      "daddiu    %[width],        %[width],            -0x04       \n\t"
>+      "bgtz      %[width],        1b                               \n\t"
>+      : [src0] "=&f"(ftmp[0]), [src1] "=&f"(ftmp[1]), [b] "=&f"(ftmp[2]),
>+        [g] "=&f"(ftmp[3]), [r] "=&f"(ftmp[4]), [a] "=&f"(ftmp[5])
>+      : [src_argb4444] "r"(src_argb4444), [dst_argb] "r"(dst_argb),
>+        [width] "r"(width), [c0] "f"(c0), [c1] "f"(c1), [eight] "f"(0x08),
>+        [four] "f"(0x04)
>+      : "memory");
>+}
>+
>+void ARGBToRGB24Row_MMI(const uint8_t* src_argb, uint8_t* dst_rgb, int width) {
>+  uint64_t src;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gslwlc1    %[src],          0x03(%[src_ptr])                 \n\t"
>+      "gslwrc1    %[src],          0x00(%[src_ptr])                 \n\t"
>+      "gsswlc1    %[src],          0x03(%[dst_ptr])                 \n\t"
>+      "gsswrc1    %[src],          0x00(%[dst_ptr])                 \n\t"
>+
>+      "gslwlc1    %[src],          0x07(%[src_ptr])                 \n\t"
>+      "gslwrc1    %[src],          0x04(%[src_ptr])                 \n\t"
>+      "gsswlc1    %[src],          0x06(%[dst_ptr])                 \n\t"
>+      "gsswrc1    %[src],          0x03(%[dst_ptr])                 \n\t"
>+
>+      "gslwlc1    %[src],          0x0b(%[src_ptr])                 \n\t"
>+      "gslwrc1    %[src],          0x08(%[src_ptr])                 \n\t"
>+      "gsswlc1    %[src],          0x09(%[dst_ptr])                 \n\t"
>+      "gsswrc1    %[src],          0x06(%[dst_ptr])                 \n\t"
>+
>+      "gslwlc1    %[src],          0x0f(%[src_ptr])                 \n\t"
>+      "gslwrc1    %[src],          0x0c(%[src_ptr])                 \n\t"
>+      "gsswlc1    %[src],          0x0c(%[dst_ptr])                 \n\t"
>+      "gsswrc1    %[src],          0x09(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x10          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x0c          \n\t"
>+      "daddi      %[width],        %[width],         -0x04          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src] "=&f"(src)
>+      : [src_ptr] "r"(src_argb), [dst_ptr] "r"(dst_rgb), [width] "r"(width)
>+      : "memory");
>+}
>+
>+void ARGBToRAWRow_MMI(const uint8_t* src_argb, uint8_t* dst_rgb, int width) {
>+  uint64_t src0, src1;
>+  uint64_t ftmp[3];
>+  uint64_t mask0 = 0xc6;
>+  uint64_t mask1 = 0x18;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src_argb])                \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src_argb])                \n\t"
>+      "gsldrc1    %[src1],         0x08(%[src_argb])                \n\t"
>+      "gsldlc1    %[src1],         0x0f(%[src_argb])                \n\t"
>+
>+      "punpcklbh  %[ftmp0],        %[src0],           %[zero]       \n\t"
>+      "pshufh     %[ftmp0],        %[ftmp0],          %[mask0]      \n\t"
>+      "punpckhbh  %[ftmp1],        %[src0],           %[zero]       \n\t"
>+      "punpcklbh  %[ftmp2],        %[src1],           %[zero]       \n\t"
>+      "punpckhbh  %[src1],         %[src1],           %[zero]       \n\t"
>+
>+      "pextrh     %[src0],         %[ftmp1],          %[two]        \n\t"
>+      "pinsrh_3   %[ftmp0],        %[ftmp0],          %[src0]       \n\t"
>+      "pshufh     %[ftmp1],        %[ftmp1],          %[one]        \n\t"
>+
>+      "pextrh     %[src0],         %[ftmp2],          %[two]        \n\t"
>+      "pinsrh_2   %[ftmp1],        %[ftmp1],          %[src0]       \n\t"
>+      "pextrh     %[src0],         %[ftmp2],          %[one]        \n\t"
>+      "pinsrh_3   %[ftmp1],        %[ftmp1],          %[src0]       \n\t"
>+      "pextrh     %[src0],         %[ftmp2],          %[zero]       \n\t"
>+      "pshufh     %[src1],         %[src1],           %[mask1]      \n\t"
>+      "pinsrh_0   %[src1],         %[src1],           %[src0]       \n\t"
>+      "packushb   %[ftmp0],        %[ftmp0],          %[ftmp1]      \n\t"
>+      "packushb   %[src1],         %[src1],           %[zero]       \n\t"
>+
>+      "gssdrc1    %[ftmp0],        0x00(%[dst_rgb])                 \n\t"
>+      "gssdlc1    %[ftmp0],        0x07(%[dst_rgb])                 \n\t"
>+      "gsswrc1    %[src1],         0x08(%[dst_rgb])                 \n\t"
>+      "gsswlc1    %[src1],         0x0b(%[dst_rgb])                 \n\t"
>+
>+      "daddiu     %[src_argb],     %[src_argb],       0x10          \n\t"
>+      "daddiu     %[dst_rgb],      %[dst_rgb],        0x0c          \n\t"
>+      "daddiu     %[width],        %[width],         -0x04          \n\t"
>+      "bgtz       %[width],        1b                               \n\t"
>+      : [src0] "=&f"(src0), [src1] "=&f"(src1), [ftmp0] "=&f"(ftmp[0]),
>+        [ftmp1] "=&f"(ftmp[1]), [ftmp2] "=&f"(ftmp[2])
>+      : [src_argb] "r"(src_argb), [dst_rgb] "r"(dst_rgb), [width] "r"(width),
>+        [mask0] "f"(mask0), [mask1] "f"(mask1), [zero] "f"(0x00),
>+        [one] "f"(0x01), [two] "f"(0x02)
>+      : "memory");
>+}
>+
>+void ARGBToRGB565Row_MMI(const uint8_t* src_argb, uint8_t* dst_rgb, int width) {
>+  uint64_t src0, src1;
>+  uint64_t ftmp[3];
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src_argb])                \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src_argb])                \n\t"
>+      "gsldrc1    %[src1],         0x08(%[src_argb])                \n\t"
>+      "gsldlc1    %[src1],         0x0f(%[src_argb])                \n\t"
>+
>+      "punpcklbh  %[b],            %[src0],           %[src1]       \n\t"
>+      "punpckhbh  %[g],            %[src0],           %[src1]       \n\t"
>+      "punpcklbh  %[src0],         %[b],              %[g]          \n\t"
>+      "punpckhbh  %[src1],         %[b],              %[g]          \n\t"
>+      "punpcklbh  %[b],            %[src0],           %[zero]       \n\t"
>+      "punpckhbh  %[g],            %[src0],           %[zero]       \n\t"
>+      "punpcklbh  %[r],            %[src1],           %[zero]       \n\t"
>+
>+      "psrlh      %[b],            %[b],              %[three]      \n\t"
>+      "psrlh      %[g],            %[g],              %[two]        \n\t"
>+      "psrlh      %[r],            %[r],              %[three]      \n\t"
>+
>+      "psllh      %[g],            %[g],              %[five]       \n\t"
>+      "psllh      %[r],            %[r],              %[eleven]     \n\t"
>+      "or         %[b],            %[b],              %[g]          \n\t"
>+      "or         %[b],            %[b],              %[r]          \n\t"
>+
>+      "gssdrc1    %[b],            0x00(%[dst_rgb])                 \n\t"
>+      "gssdlc1    %[b],            0x07(%[dst_rgb])                 \n\t"
>+
>+      "daddiu     %[src_argb],     %[src_argb],       0x10          \n\t"
>+      "daddiu     %[dst_rgb],      %[dst_rgb],        0x08          \n\t"
>+      "daddiu     %[width],        %[width],         -0x04          \n\t"
>+      "bgtz       %[width],        1b                               \n\t"
>+      : [src0] "=&f"(src0), [src1] "=&f"(src1), [b] "=&f"(ftmp[0]),
>+        [g] "=&f"(ftmp[1]), [r] "=&f"(ftmp[2])
>+      : [src_argb] "r"(src_argb), [dst_rgb] "r"(dst_rgb), [width] "r"(width),
>+        [zero] "f"(0x00), [two] "f"(0x02), [three] "f"(0x03), [five] "f"(0x05),
>+        [eleven] "f"(0x0b)
>+      : "memory");
>+}
>+
>+// dither4 is a row of 4 values from 4x4 dither matrix.
>+// The 4x4 matrix contains values to increase RGB.  When converting to
>+// fewer bits (565) this provides an ordered dither.
>+// The order in the 4x4 matrix in first byte is upper left.
>+// The 4 values are passed as an int, then referenced as an array, so
>+// endian will not affect order of the original matrix.  But the dither4
>+// will containing the first pixel in the lower byte for little endian
>+// or the upper byte for big endian.
>+void ARGBToRGB565DitherRow_MMI(const uint8_t* src_argb,
>+                               uint8_t* dst_rgb,
>+                               const uint32_t dither4,
>+                               int width) {
>+  uint64_t src0, src1;
>+  uint64_t ftmp[3];
>+  uint64_t c0 = 0x00ff00ff00ff00ff;
>+
>+  __asm__ volatile(
>+      "punpcklbh  %[dither],       %[dither],         %[zero]       \n\t"
>+      "1:                                                           \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src_argb])                \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src_argb])                \n\t"
>+      "gsldrc1    %[src1],         0x08(%[src_argb])                \n\t"
>+      "gsldlc1    %[src1],         0x0f(%[src_argb])                \n\t"
>+
>+      "punpcklbh  %[b],            %[src0],           %[src1]       \n\t"
>+      "punpckhbh  %[g],            %[src0],           %[src1]       \n\t"
>+      "punpcklbh  %[src0],         %[b],              %[g]          \n\t"
>+      "punpckhbh  %[src1],         %[b],              %[g]          \n\t"
>+      "punpcklbh  %[b],            %[src0],           %[zero]       \n\t"
>+      "punpckhbh  %[g],            %[src0],           %[zero]       \n\t"
>+      "punpcklbh  %[r],            %[src1],           %[zero]       \n\t"
>+
>+      "paddh      %[b],            %[b],              %[dither]     \n\t"
>+      "paddh      %[g],            %[g],              %[dither]     \n\t"
>+      "paddh      %[r],            %[r],              %[dither]     \n\t"
>+      "pcmpgth    %[src0],         %[b],              %[c0]         \n\t"
>+      "or         %[src0],         %[src0],           %[b]          \n\t"
>+      "and        %[b],            %[src0],           %[c0]         \n\t"
>+      "pcmpgth    %[src0],         %[g],              %[c0]         \n\t"
>+      "or         %[src0],         %[src0],           %[g]          \n\t"
>+      "and        %[g],            %[src0],           %[c0]         \n\t"
>+      "pcmpgth    %[src0],         %[r],              %[c0]         \n\t"
>+      "or         %[src0],         %[src0],           %[r]          \n\t"
>+      "and        %[r],            %[src0],           %[c0]         \n\t"
>+
>+      "psrlh      %[b],            %[b],              %[three]      \n\t"
>+      "psrlh      %[g],            %[g],              %[two]        \n\t"
>+      "psrlh      %[r],            %[r],              %[three]      \n\t"
>+
>+      "psllh      %[g],            %[g],              %[five]       \n\t"
>+      "psllh      %[r],            %[r],              %[eleven]     \n\t"
>+      "or         %[b],            %[b],              %[g]          \n\t"
>+      "or         %[b],            %[b],              %[r]          \n\t"
>+
>+      "gssdrc1    %[b],            0x00(%[dst_rgb])                 \n\t"
>+      "gssdlc1    %[b],            0x07(%[dst_rgb])                 \n\t"
>+
>+      "daddiu     %[src_argb],     %[src_argb],       0x10          \n\t"
>+      "daddiu     %[dst_rgb],      %[dst_rgb],        0x08          \n\t"
>+      "daddiu     %[width],        %[width],         -0x04          \n\t"
>+      "bgtz       %[width],        1b                               \n\t"
>+      : [src0] "=&f"(src0), [src1] "=&f"(src1), [b] "=&f"(ftmp[0]),
>+        [g] "=&f"(ftmp[1]), [r] "=&f"(ftmp[2])
>+      : [src_argb] "r"(src_argb), [dst_rgb] "r"(dst_rgb), [width] "r"(width),
>+        [dither] "f"(dither4), [c0] "f"(c0), [zero] "f"(0x00), [two] "f"(0x02),
>+        [three] "f"(0x03), [five] "f"(0x05), [eleven] "f"(0x0b)
>+      : "memory");
>+}
>+
>+void ARGBToARGB1555Row_MMI(const uint8_t* src_argb,
>+                           uint8_t* dst_rgb,
>+                           int width) {
>+  uint64_t src0, src1;
>+  uint64_t ftmp[4];
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src_argb])                \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src_argb])                \n\t"
>+      "gsldrc1    %[src1],         0x08(%[src_argb])                \n\t"
>+      "gsldlc1    %[src1],         0x0f(%[src_argb])                \n\t"
>+
>+      "punpcklbh  %[b],            %[src0],           %[src1]       \n\t"
>+      "punpckhbh  %[g],            %[src0],           %[src1]       \n\t"
>+      "punpcklbh  %[src0],         %[b],              %[g]          \n\t"
>+      "punpckhbh  %[src1],         %[b],              %[g]          \n\t"
>+      "punpcklbh  %[b],            %[src0],           %[zero]       \n\t"
>+      "punpckhbh  %[g],            %[src0],           %[zero]       \n\t"
>+      "punpcklbh  %[r],            %[src1],           %[zero]       \n\t"
>+      "punpckhbh  %[a],            %[src1],           %[zero]       \n\t"
>+
>+      "psrlh      %[b],            %[b],              %[three]      \n\t"
>+      "psrlh      %[g],            %[g],              %[three]      \n\t"
>+      "psrlh      %[r],            %[r],              %[three]      \n\t"
>+      "psrlh      %[a],            %[a],              %[seven]      \n\t"
>+
>+      "psllh      %[g],            %[g],              %[five]       \n\t"
>+      "psllh      %[r],            %[r],              %[ten]        \n\t"
>+      "psllh      %[a],            %[a],              %[fifteen]    \n\t"
>+      "or         %[b],            %[b],              %[g]          \n\t"
>+      "or         %[b],            %[b],              %[r]          \n\t"
>+      "or         %[b],            %[b],              %[a]          \n\t"
>+
>+      "gssdrc1    %[b],            0x00(%[dst_rgb])                 \n\t"
>+      "gssdlc1    %[b],            0x07(%[dst_rgb])                 \n\t"
>+
>+      "daddiu     %[src_argb],     %[src_argb],       0x10          \n\t"
>+      "daddiu     %[dst_rgb],      %[dst_rgb],        0x08          \n\t"
>+      "daddiu     %[width],        %[width],         -0x04          \n\t"
>+      "bgtz       %[width],        1b                               \n\t"
>+      : [src0] "=&f"(src0), [src1] "=&f"(src1), [b] "=&f"(ftmp[0]),
>+        [g] "=&f"(ftmp[1]), [r] "=&f"(ftmp[2]), [a] "=&f"(ftmp[3])
>+      : [src_argb] "r"(src_argb), [dst_rgb] "r"(dst_rgb), [width] "r"(width),
>+        [zero] "f"(0x00), [three] "f"(0x03), [five] "f"(0x05),
>+        [seven] "f"(0x07), [ten] "f"(0x0a), [fifteen] "f"(0x0f)
>+      : "memory");
>+}
>+
>+void ARGBToARGB4444Row_MMI(const uint8_t* src_argb,
>+                           uint8_t* dst_rgb,
>+                           int width) {
>+  uint64_t src0, src1;
>+  uint64_t ftmp[4];
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src_argb])                \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src_argb])                \n\t"
>+      "gsldrc1    %[src1],         0x08(%[src_argb])                \n\t"
>+      "gsldlc1    %[src1],         0x0f(%[src_argb])                \n\t"
>+
>+      "punpcklbh  %[b],            %[src0],           %[src1]       \n\t"
>+      "punpckhbh  %[g],            %[src0],           %[src1]       \n\t"
>+      "punpcklbh  %[src0],         %[b],              %[g]          \n\t"
>+      "punpckhbh  %[src1],         %[b],              %[g]          \n\t"
>+      "punpcklbh  %[b],            %[src0],           %[zero]       \n\t"
>+      "punpckhbh  %[g],            %[src0],           %[zero]       \n\t"
>+      "punpcklbh  %[r],            %[src1],           %[zero]       \n\t"
>+      "punpckhbh  %[a],            %[src1],           %[zero]       \n\t"
>+
>+      "psrlh      %[b],            %[b],              %[four]       \n\t"
>+      "psrlh      %[g],            %[g],              %[four]       \n\t"
>+      "psrlh      %[r],            %[r],              %[four]       \n\t"
>+      "psrlh      %[a],            %[a],              %[four]       \n\t"
>+
>+      "psllh      %[g],            %[g],              %[four]       \n\t"
>+      "psllh      %[r],            %[r],              %[eight]      \n\t"
>+      "psllh      %[a],            %[a],              %[twelve]     \n\t"
>+      "or         %[b],            %[b],              %[g]          \n\t"
>+      "or         %[b],            %[b],              %[r]          \n\t"
>+      "or         %[b],            %[b],              %[a]          \n\t"
>+
>+      "gssdrc1    %[b],            0x00(%[dst_rgb])                 \n\t"
>+      "gssdlc1    %[b],            0x07(%[dst_rgb])                 \n\t"
>+
>+      "daddiu     %[src_argb],     %[src_argb],       0x10          \n\t"
>+      "daddiu     %[dst_rgb],      %[dst_rgb],        0x08          \n\t"
>+      "daddiu     %[width],        %[width],         -0x04          \n\t"
>+      "bgtz       %[width],        1b                               \n\t"
>+      : [src0] "=&f"(src0), [src1] "=&f"(src1), [b] "=&f"(ftmp[0]),
>+        [g] "=&f"(ftmp[1]), [r] "=&f"(ftmp[2]), [a] "=&f"(ftmp[3])
>+      : [src_argb] "r"(src_argb), [dst_rgb] "r"(dst_rgb), [width] "r"(width),
>+        [zero] "f"(0x00), [four] "f"(0x04), [eight] "f"(0x08),
>+        [twelve] "f"(0x0c)
>+      : "memory");
>+}
>+
>+void ARGBToYRow_MMI(const uint8_t* src_argb0, uint8_t* dst_y, int width) {
>+  uint64_t src, src_hi, src_lo;
>+  uint64_t dest0, dest1, dest2, dest3;
>+  const uint64_t value = 0x1080;
>+  const uint64_t mask = 0x0001004200810019;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[src],          0x07(%[src_argb0])               \n\t"
>+      "gsldrc1    %[src],          0x00(%[src_argb0])               \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask]       \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask]       \n\t"
>+      "punpcklwd  %[src],          %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[dest0],        %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest0],        %[dest0],          %[src]        \n\t"
>+      "psrlw      %[dest0],        %[dest0],          %[eight]      \n\t"
>+
>+      "gsldlc1    %[src],          0x0f(%[src_argb0])               \n\t"
>+      "gsldrc1    %[src],          0x08(%[src_argb0])               \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask]       \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask]       \n\t"
>+      "punpcklwd  %[src],          %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[dest1],        %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest1],        %[dest1],          %[src]        \n\t"
>+      "psrlw      %[dest1],        %[dest1],          %[eight]      \n\t"
>+
>+      "gsldlc1    %[src],          0x17(%[src_argb0])               \n\t"
>+      "gsldrc1    %[src],          0x10(%[src_argb0])               \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask]       \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask]       \n\t"
>+      "punpcklwd  %[src],          %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[dest2],        %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest2],        %[dest2],          %[src]        \n\t"
>+      "psrlw      %[dest2],        %[dest2],          %[eight]      \n\t"
>+
>+      "gsldlc1    %[src],          0x1f(%[src_argb0])               \n\t"
>+      "gsldrc1    %[src],          0x18(%[src_argb0])               \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask]       \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask]       \n\t"
>+      "punpcklwd  %[src],          %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[dest3],        %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest3],        %[dest3],          %[src]        \n\t"
>+      "psrlw      %[dest3],        %[dest3],          %[eight]      \n\t"
>+
>+      "packsswh   %[src_lo],       %[dest0],          %[dest1]      \n\t"
>+      "packsswh   %[src_hi],       %[dest2],          %[dest3]      \n\t"
>+      "packushb   %[dest0],        %[src_lo],         %[src_hi]     \n\t"
>+      "gssdlc1    %[dest0],        0x07(%[dst_y])                   \n\t"
>+      "gssdrc1    %[dest0],        0x00(%[dst_y])                   \n\t"
>+
>+      "daddiu     %[src_argb0],    %[src_argb0],      0x20          \n\t"
>+      "daddiu     %[dst_y],        %[dst_y],          0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x08          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src] "=&f"(src), [src_hi] "=&f"(src_hi), [src_lo] "=&f"(src_lo),
>+        [dest0] "=&f"(dest0), [dest1] "=&f"(dest1), [dest2] "=&f"(dest2),
>+        [dest3] "=&f"(dest3)
>+      : [src_argb0] "r"(src_argb0), [dst_y] "r"(dst_y), [width] "r"(width),
>+        [mask] "f"(mask), [value] "f"(value), [eight] "f"(0x08),
>+        [zero] "f"(0x00)
>+      : "memory");
>+}
>+
>+void ARGBToUVRow_MMI(const uint8_t* src_rgb0,
>+                     int src_stride_rgb,
>+                     uint8_t* dst_u,
>+                     uint8_t* dst_v,
>+                     int width) {
>+  uint64_t src_rgb1;
>+  uint64_t ftmp[12];
>+  const uint64_t value = 0x4040;
>+  const uint64_t mask_u = 0x0026004a00700002;
>+  const uint64_t mask_v = 0x00020070005e0012;
>+
>+  __asm__ volatile(
>+      "1:                                                               \n\t"
>+      "daddu      %[src_rgb1],     %[src_rgb0],       %[src_stride_rgb] \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x00(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x07(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "dsll       %[dest0_u],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[dest0_u],      %[dest0_u],        %[value]          \n\t"
>+      "pinsrh_3   %[dest0_v],      %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[dest0_u],      %[dest0_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest0_v],      %[dest0_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x08(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x0f(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x08(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x0f(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "dsll       %[src_lo],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[src_lo],       %[src_lo],         %[value]          \n\t"
>+      "pinsrh_3   %[src_hi],       %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest0_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest0_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest0_u],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest0_u],      %[dest0_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest0_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest0_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest0_v],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest0_v],      %[dest0_v],        %[eight]          \n\t"
>+
>+      "gsldrc1    %[src0],         0x10(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x17(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x10(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x17(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "dsll       %[dest1_u],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[dest1_u],      %[dest1_u],        %[value]          \n\t"
>+      "pinsrh_3   %[dest1_v],      %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[dest1_u],      %[dest1_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest1_v],      %[dest1_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x18(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x1f(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x18(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x1f(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "dsll       %[src_lo],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[src_lo],       %[src_lo],         %[value]          \n\t"
>+      "pinsrh_3   %[src_hi],       %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest1_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest1_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest1_u],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest1_u],      %[dest1_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest1_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest1_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest1_v],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest1_v],      %[dest1_v],        %[eight]          \n\t"
>+
>+      "gsldrc1    %[src0],         0x20(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x27(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x20(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x27(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "dsll       %[dest2_u],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[dest2_u],      %[dest2_u],        %[value]          \n\t"
>+      "pinsrh_3   %[dest2_v],      %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[dest2_u],      %[dest2_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest2_v],      %[dest2_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x28(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x2f(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x28(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x2f(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "dsll       %[src_lo],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[src_lo],       %[src_lo],         %[value]          \n\t"
>+      "pinsrh_3   %[src_hi],       %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest2_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest2_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest2_u],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest2_u],      %[dest2_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest2_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest2_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest2_v],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest2_v],      %[dest2_v],        %[eight]          \n\t"
>+
>+      "gsldrc1    %[src0],         0x30(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x37(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x30(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x37(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "dsll       %[dest3_u],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[dest3_u],      %[dest3_u],        %[value]          \n\t"
>+      "pinsrh_3   %[dest3_v],      %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[dest3_u],      %[dest3_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest3_v],      %[dest3_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x38(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x3f(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x38(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x3f(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "dsll       %[src_lo],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[src_lo],       %[src_lo],         %[value]          \n\t"
>+      "pinsrh_3   %[src_hi],       %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest3_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest3_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest3_u],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest3_u],      %[dest3_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest3_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest3_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest3_v],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest3_v],      %[dest3_v],        %[eight]          \n\t"
>+
>+      "packsswh   %[src0],         %[dest0_u],        %[dest1_u]        \n\t"
>+      "packsswh   %[src1],         %[dest2_u],        %[dest3_u]        \n\t"
>+      "packushb   %[dest0_u],      %[src0],           %[src1]           \n\t"
>+      "gssdlc1    %[dest0_u],      0x07(%[dst_u])                       \n\t"
>+      "gssdrc1    %[dest0_u],      0x00(%[dst_u])                       \n\t"
>+
>+      "packsswh   %[src0],         %[dest0_v],        %[dest1_v]        \n\t"
>+      "packsswh   %[src1],         %[dest2_v],        %[dest3_v]        \n\t"
>+      "packushb   %[dest0_v],      %[src0],           %[src1]           \n\t"
>+      "gssdlc1    %[dest0_v],      0x07(%[dst_v])                       \n\t"
>+      "gssdrc1    %[dest0_v],      0x00(%[dst_v])                       \n\t"
>+
>+      "daddiu     %[src_rgb0],     %[src_rgb0],       0x40              \n\t"
>+      "daddiu     %[dst_u],        %[dst_u],          0x08              \n\t"
>+      "daddiu     %[dst_v],        %[dst_v],          0x08              \n\t"
>+      "daddi      %[width],        %[width],         -0x10              \n\t"
>+      "bgtz       %[width],        1b                                   \n\t"
>+      : [src_rgb1] "=&r"(src_rgb1), [src0] "=&f"(ftmp[0]),
>+        [src1] "=&f"(ftmp[1]), [src_lo] "=&f"(ftmp[2]), [src_hi] "=&f"(ftmp[3]),
>+        [dest0_u] "=&f"(ftmp[4]), [dest0_v] "=&f"(ftmp[5]),
>+        [dest1_u] "=&f"(ftmp[6]), [dest1_v] "=&f"(ftmp[7]),
>+        [dest2_u] "=&f"(ftmp[8]), [dest2_v] "=&f"(ftmp[9]),
>+        [dest3_u] "=&f"(ftmp[10]), [dest3_v] "=&f"(ftmp[11])
>+      : [src_rgb0] "r"(src_rgb0), [src_stride_rgb] "r"(src_stride_rgb),
>+        [dst_u] "r"(dst_u), [dst_v] "r"(dst_v), [width] "r"(width),
>+        [mask_u] "f"(mask_u), [mask_v] "f"(mask_v), [value] "f"(value),
>+        [zero] "f"(0x00), [eight] "f"(0x08), [two] "f"(0x02),
>+        [sixteen] "f"(0x10)
>+      : "memory");
>+}
>+
>+void BGRAToYRow_MMI(const uint8_t* src_argb0, uint8_t* dst_y, int width) {
>+  uint64_t src, src_hi, src_lo;
>+  uint64_t dest0, dest1, dest2, dest3;
>+  const uint64_t value = 0x1080;
>+  const uint64_t mask = 0x0019008100420001;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[src],          0x07(%[src_argb0])               \n\t"
>+      "gsldrc1    %[src],          0x00(%[src_argb0])               \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[zero]       \n\t"
>+      "pinsrh_0   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask]       \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[zero]       \n\t"
>+      "pinsrh_0   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask]       \n\t"
>+      "punpcklwd  %[src],          %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[dest0],        %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest0],        %[dest0],          %[src]        \n\t"
>+      "psrlw      %[dest0],        %[dest0],          %[eight]      \n\t"
>+
>+      "gsldlc1    %[src],          0x0f(%[src_argb0])               \n\t"
>+      "gsldrc1    %[src],          0x08(%[src_argb0])               \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[zero]       \n\t"
>+      "pinsrh_0   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask]       \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[zero]       \n\t"
>+      "pinsrh_0   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask]       \n\t"
>+      "punpcklwd  %[src],          %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[dest1],        %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest1],        %[dest1],          %[src]        \n\t"
>+      "psrlw      %[dest1],        %[dest1],          %[eight]      \n\t"
>+
>+      "gsldlc1    %[src],          0x17(%[src_argb0])               \n\t"
>+      "gsldrc1    %[src],          0x10(%[src_argb0])               \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[zero]       \n\t"
>+      "pinsrh_0   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask]       \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[zero]       \n\t"
>+      "pinsrh_0   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask]       \n\t"
>+      "punpcklwd  %[src],          %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[dest2],        %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest2],        %[dest2],          %[src]        \n\t"
>+      "psrlw      %[dest2],        %[dest2],          %[eight]      \n\t"
>+
>+      "gsldlc1    %[src],          0x1f(%[src_argb0])               \n\t"
>+      "gsldrc1    %[src],          0x18(%[src_argb0])               \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[zero]       \n\t"
>+      "pinsrh_0   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask]       \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[zero]       \n\t"
>+      "pinsrh_0   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask]       \n\t"
>+      "punpcklwd  %[src],          %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[dest3],        %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest3],        %[dest3],          %[src]        \n\t"
>+      "psrlw      %[dest3],        %[dest3],          %[eight]      \n\t"
>+
>+      "packsswh   %[src_lo],       %[dest0],          %[dest1]      \n\t"
>+      "packsswh   %[src_hi],       %[dest2],          %[dest3]      \n\t"
>+      "packushb   %[dest0],        %[src_lo],         %[src_hi]     \n\t"
>+      "gssdlc1    %[dest0],        0x07(%[dst_y])                   \n\t"
>+      "gssdrc1    %[dest0],        0x00(%[dst_y])                   \n\t"
>+
>+      "daddiu     %[src_argb0],    %[src_argb0],      0x20          \n\t"
>+      "daddiu     %[dst_y],        %[dst_y],          0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x08          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src] "=&f"(src), [src_hi] "=&f"(src_hi), [src_lo] "=&f"(src_lo),
>+        [dest0] "=&f"(dest0), [dest1] "=&f"(dest1), [dest2] "=&f"(dest2),
>+        [dest3] "=&f"(dest3)
>+      : [src_argb0] "r"(src_argb0), [dst_y] "r"(dst_y), [width] "r"(width),
>+        [mask] "f"(mask), [value] "f"(value), [eight] "f"(0x08),
>+        [zero] "f"(0x00)
>+      : "memory");
>+}
>+
>+void BGRAToUVRow_MMI(const uint8_t* src_rgb0,
>+                     int src_stride_rgb,
>+                     uint8_t* dst_u,
>+                     uint8_t* dst_v,
>+                     int width) {
>+  uint64_t src_rgb1;
>+  uint64_t ftmp[12];
>+  const uint64_t value = 0x4040;
>+  const uint64_t mask_u = 0x00020070004a0026;
>+  const uint64_t mask_v = 0x0012005e00700002;
>+
>+  __asm__ volatile(
>+      "1:                                                               \n\t"
>+      "daddu      %[src_rgb1],     %[src_rgb0],       %[src_stride_rgb] \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x00(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x07(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "dsrl       %[dest0_u],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_3   %[dest0_u],      %[dest0_u],        %[value]          \n\t"
>+      "pinsrh_0   %[dest0_v],      %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[dest0_u],      %[dest0_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest0_v],      %[dest0_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x08(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x0f(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x08(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x0f(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "dsrl       %[src_lo],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[value]          \n\t"
>+      "pinsrh_0   %[src_hi],       %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest0_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest0_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest0_u],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest0_u],      %[dest0_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest0_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest0_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest0_v],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest0_v],      %[dest0_v],        %[eight]          \n\t"
>+
>+      "gsldrc1    %[src0],         0x10(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x17(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x10(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x17(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "dsrl       %[dest1_u],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_3   %[dest1_u],      %[dest1_u],        %[value]          \n\t"
>+      "pinsrh_0   %[dest1_v],      %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[dest1_u],      %[dest1_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest1_v],      %[dest1_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x18(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x1f(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x18(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x1f(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "dsrl       %[src_lo],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[value]          \n\t"
>+      "pinsrh_0   %[src_hi],       %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest1_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest1_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest1_u],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest1_u],      %[dest1_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest1_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest1_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest1_v],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest1_v],      %[dest1_v],        %[eight]          \n\t"
>+
>+      "gsldrc1    %[src0],         0x20(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x27(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x20(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x27(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "dsrl       %[dest2_u],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_3   %[dest2_u],      %[dest2_u],        %[value]          \n\t"
>+      "pinsrh_0   %[dest2_v],      %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[dest2_u],      %[dest2_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest2_v],      %[dest2_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x28(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x2f(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x28(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x2f(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]        \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "dsrl       %[src_lo],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[value]          \n\t"
>+      "pinsrh_0   %[src_hi],       %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest2_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest2_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest2_u],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest2_u],      %[dest2_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest2_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest2_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest2_v],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest2_v],      %[dest2_v],        %[eight]          \n\t"
>+
>+      "gsldrc1    %[src0],         0x30(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x37(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x30(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x37(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "dsrl       %[dest3_u],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_3   %[dest3_u],      %[dest3_u],        %[value]          \n\t"
>+      "pinsrh_0   %[dest3_v],      %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[dest3_u],      %[dest3_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest3_v],      %[dest3_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x38(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x3f(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x38(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x3f(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "dsrl       %[src_lo],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[value]          \n\t"
>+      "pinsrh_0   %[src_hi],       %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest3_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest3_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest3_u],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest3_u],      %[dest3_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest3_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest3_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest3_v],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest3_v],      %[dest3_v],        %[eight]          \n\t"
>+
>+      "packsswh   %[src0],         %[dest0_u],        %[dest1_u]        \n\t"
>+      "packsswh   %[src1],         %[dest2_u],        %[dest3_u]        \n\t"
>+      "packushb   %[dest0_u],      %[src0],           %[src1]           \n\t"
>+      "gssdlc1    %[dest0_u],      0x07(%[dst_u])                       \n\t"
>+      "gssdrc1    %[dest0_u],      0x00(%[dst_u])                       \n\t"
>+
>+      "packsswh   %[src0],         %[dest0_v],        %[dest1_v]        \n\t"
>+      "packsswh   %[src1],         %[dest2_v],        %[dest3_v]        \n\t"
>+      "packushb   %[dest0_v],      %[src0],           %[src1]           \n\t"
>+      "gssdlc1    %[dest0_v],      0x07(%[dst_v])                       \n\t"
>+      "gssdrc1    %[dest0_v],      0x00(%[dst_v])                       \n\t"
>+
>+      "daddiu     %[src_rgb0],     %[src_rgb0],       0x40              \n\t"
>+      "daddiu     %[dst_u],        %[dst_u],          0x08              \n\t"
>+      "daddiu     %[dst_v],        %[dst_v],          0x08              \n\t"
>+      "daddi      %[width],        %[width],         -0x10              \n\t"
>+      "bgtz       %[width],        1b                                   \n\t"
>+      : [src_rgb1] "=&r"(src_rgb1), [src0] "=&f"(ftmp[0]),
>+        [src1] "=&f"(ftmp[1]), [src_lo] "=&f"(ftmp[2]), [src_hi] "=&f"(ftmp[3]),
>+        [dest0_u] "=&f"(ftmp[4]), [dest0_v] "=&f"(ftmp[5]),
>+        [dest1_u] "=&f"(ftmp[6]), [dest1_v] "=&f"(ftmp[7]),
>+        [dest2_u] "=&f"(ftmp[8]), [dest2_v] "=&f"(ftmp[9]),
>+        [dest3_u] "=&f"(ftmp[10]), [dest3_v] "=&f"(ftmp[11])
>+      : [src_rgb0] "r"(src_rgb0), [src_stride_rgb] "r"(src_stride_rgb),
>+        [dst_u] "r"(dst_u), [dst_v] "r"(dst_v), [width] "r"(width),
>+        [mask_u] "f"(mask_u), [mask_v] "f"(mask_v), [value] "f"(value),
>+        [zero] "f"(0x00), [eight] "f"(0x08), [two] "f"(0x02),
>+        [sixteen] "f"(0x10)
>+      : "memory");
>+}
>+
>+void ABGRToYRow_MMI(const uint8_t* src_argb0, uint8_t* dst_y, int width) {
>+  uint64_t src, src_hi, src_lo;
>+  uint64_t dest0, dest1, dest2, dest3;
>+  const uint64_t value = 0x1080;
>+  const uint64_t mask = 0x0001001900810042;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[src],          0x07(%[src_argb0])               \n\t"
>+      "gsldrc1    %[src],          0x00(%[src_argb0])               \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask]       \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask]       \n\t"
>+      "punpcklwd  %[src],          %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[dest0],        %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest0],        %[dest0],          %[src]        \n\t"
>+      "psrlw      %[dest0],        %[dest0],          %[eight]      \n\t"
>+
>+      "gsldlc1    %[src],          0x0f(%[src_argb0])               \n\t"
>+      "gsldrc1    %[src],          0x08(%[src_argb0])               \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask]       \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask]       \n\t"
>+      "punpcklwd  %[src],          %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[dest1],        %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest1],        %[dest1],          %[src]        \n\t"
>+      "psrlw      %[dest1],        %[dest1],          %[eight]      \n\t"
>+
>+      "gsldlc1    %[src],          0x17(%[src_argb0])               \n\t"
>+      "gsldrc1    %[src],          0x10(%[src_argb0])               \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask]       \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask]       \n\t"
>+      "punpcklwd  %[src],          %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[dest2],        %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest2],        %[dest2],          %[src]        \n\t"
>+      "psrlw      %[dest2],        %[dest2],          %[eight]      \n\t"
>+
>+      "gsldlc1    %[src],          0x1f(%[src_argb0])               \n\t"
>+      "gsldrc1    %[src],          0x18(%[src_argb0])               \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask]       \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask]       \n\t"
>+      "punpcklwd  %[src],          %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[dest3],        %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest3],        %[dest3],          %[src]        \n\t"
>+      "psrlw      %[dest3],        %[dest3],          %[eight]      \n\t"
>+
>+      "packsswh   %[src_lo],       %[dest0],          %[dest1]      \n\t"
>+      "packsswh   %[src_hi],       %[dest2],          %[dest3]      \n\t"
>+      "packushb   %[dest0],        %[src_lo],         %[src_hi]     \n\t"
>+      "gssdlc1    %[dest0],        0x07(%[dst_y])                   \n\t"
>+      "gssdrc1    %[dest0],        0x00(%[dst_y])                   \n\t"
>+
>+      "daddiu     %[src_argb0],    %[src_argb0],      0x20          \n\t"
>+      "daddiu     %[dst_y],        %[dst_y],          0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x08          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src] "=&f"(src), [src_hi] "=&f"(src_hi), [src_lo] "=&f"(src_lo),
>+        [dest0] "=&f"(dest0), [dest1] "=&f"(dest1), [dest2] "=&f"(dest2),
>+        [dest3] "=&f"(dest3)
>+      : [src_argb0] "r"(src_argb0), [dst_y] "r"(dst_y), [width] "r"(width),
>+        [mask] "f"(mask), [value] "f"(value), [eight] "f"(0x08),
>+        [zero] "f"(0x00)
>+      : "memory");
>+}
>+
>+void ABGRToUVRow_MMI(const uint8_t* src_rgb0,
>+                     int src_stride_rgb,
>+                     uint8_t* dst_u,
>+                     uint8_t* dst_v,
>+                     int width) {
>+  uint64_t src_rgb1;
>+  uint64_t ftmp[12];
>+  const uint64_t value = 0x4040;
>+  const uint64_t mask_u = 0x00020070004a0026;
>+  const uint64_t mask_v = 0x0012005e00700002;
>+
>+  __asm__ volatile(
>+      "1:                                                               \n\t"
>+      "daddu      %[src_rgb1],     %[src_rgb0],       %[src_stride_rgb] \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x00(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x07(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "pinsrh_3   %[dest0_u],      %[src0],           %[value]          \n\t"
>+      "dsll       %[dest0_v],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[dest0_v],      %[dest0_v],        %[value]          \n\t"
>+      "pmaddhw    %[dest0_u],      %[dest0_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest0_v],      %[dest0_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x08(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x0f(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x08(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x0f(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "pinsrh_3   %[src_lo],       %[src0],           %[value]          \n\t"
>+      "dsll       %[src_hi],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[src_hi],       %[src_hi],         %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest0_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest0_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest0_u],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest0_u],      %[dest0_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest0_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest0_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest0_v],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest0_v],      %[dest0_v],        %[eight]          \n\t"
>+
>+      "gsldrc1    %[src0],         0x10(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x17(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x10(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x17(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "pinsrh_3   %[dest1_u],      %[src0],           %[value]          \n\t"
>+      "dsll       %[dest1_v],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[dest1_v],      %[dest1_v],        %[value]          \n\t"
>+      "pmaddhw    %[dest1_u],      %[dest1_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest1_v],      %[dest1_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x18(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x1f(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x18(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x1f(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "pinsrh_3   %[src_lo],       %[src0],           %[value]          \n\t"
>+      "dsll       %[src_hi],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[src_hi],       %[src_hi],         %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest1_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest1_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest1_u],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest1_u],      %[dest1_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest1_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest1_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest1_v],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest1_v],      %[dest1_v],        %[eight]          \n\t"
>+
>+      "gsldrc1    %[src0],         0x20(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x27(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x20(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x27(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "pinsrh_3   %[dest2_u],      %[src0],           %[value]          \n\t"
>+      "dsll       %[dest2_v],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[dest2_v],      %[dest2_v],        %[value]          \n\t"
>+      "pmaddhw    %[dest2_u],      %[dest2_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest2_v],      %[dest2_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x28(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x2f(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x28(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x2f(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "pinsrh_3   %[src_lo],       %[src0],           %[value]          \n\t"
>+      "dsll       %[src_hi],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[src_hi],       %[src_hi],         %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest2_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest2_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest2_u],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest2_u],      %[dest2_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest2_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest2_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest2_v],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest2_v],      %[dest2_v],        %[eight]          \n\t"
>+
>+      "gsldrc1    %[src0],         0x30(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x37(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x30(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x37(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "pinsrh_3   %[dest3_u],      %[src0],           %[value]          \n\t"
>+      "dsll       %[dest3_v],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[dest3_v],      %[dest3_v],        %[value]          \n\t"
>+      "pmaddhw    %[dest3_u],      %[dest3_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest3_v],      %[dest3_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x38(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x3f(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x38(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x3f(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "pinsrh_3   %[src_lo],       %[src0],           %[value]          \n\t"
>+      "dsll       %[src_hi],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[src_hi],       %[src_hi],         %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest3_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest3_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest3_u],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest3_u],      %[dest3_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest3_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest3_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest3_v],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest3_v],      %[dest3_v],        %[eight]          \n\t"
>+
>+      "packsswh   %[src0],         %[dest0_u],        %[dest1_u]        \n\t"
>+      "packsswh   %[src1],         %[dest2_u],        %[dest3_u]        \n\t"
>+      "packushb   %[dest0_u],      %[src0],           %[src1]           \n\t"
>+      "gssdlc1    %[dest0_u],      0x07(%[dst_u])                       \n\t"
>+      "gssdrc1    %[dest0_u],      0x00(%[dst_u])                       \n\t"
>+
>+      "packsswh   %[src0],         %[dest0_v],        %[dest1_v]        \n\t"
>+      "packsswh   %[src1],         %[dest2_v],        %[dest3_v]        \n\t"
>+      "packushb   %[dest0_v],      %[src0],           %[src1]           \n\t"
>+      "gssdlc1    %[dest0_v],      0x07(%[dst_v])                       \n\t"
>+      "gssdrc1    %[dest0_v],      0x00(%[dst_v])                       \n\t"
>+
>+      "daddiu     %[src_rgb0],     %[src_rgb0],       0x40              \n\t"
>+      "daddiu     %[dst_u],        %[dst_u],          0x08              \n\t"
>+      "daddiu     %[dst_v],        %[dst_v],          0x08              \n\t"
>+      "daddi      %[width],        %[width],         -0x10              \n\t"
>+      "bgtz       %[width],        1b                                   \n\t"
>+      : [src_rgb1] "=&r"(src_rgb1), [src0] "=&f"(ftmp[0]),
>+        [src1] "=&f"(ftmp[1]), [src_lo] "=&f"(ftmp[2]), [src_hi] "=&f"(ftmp[3]),
>+        [dest0_u] "=&f"(ftmp[4]), [dest0_v] "=&f"(ftmp[5]),
>+        [dest1_u] "=&f"(ftmp[6]), [dest1_v] "=&f"(ftmp[7]),
>+        [dest2_u] "=&f"(ftmp[8]), [dest2_v] "=&f"(ftmp[9]),
>+        [dest3_u] "=&f"(ftmp[10]), [dest3_v] "=&f"(ftmp[11])
>+      : [src_rgb0] "r"(src_rgb0), [src_stride_rgb] "r"(src_stride_rgb),
>+        [dst_u] "r"(dst_u), [dst_v] "r"(dst_v), [width] "r"(width),
>+        [mask_u] "f"(mask_u), [mask_v] "f"(mask_v), [value] "f"(value),
>+        [zero] "f"(0x00), [eight] "f"(0x08), [two] "f"(0x02),
>+        [sixteen] "f"(0x10)
>+      : "memory");
>+}
>+
>+void RGBAToYRow_MMI(const uint8_t* src_argb0, uint8_t* dst_y, int width) {
>+  uint64_t src, src_hi, src_lo;
>+  uint64_t dest0, dest1, dest2, dest3;
>+  const uint64_t value = 0x1080;
>+  const uint64_t mask = 0x0042008100190001;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[src],          0x07(%[src_argb0])               \n\t"
>+      "gsldrc1    %[src],          0x00(%[src_argb0])               \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[zero]       \n\t"
>+      "pinsrh_0   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask]       \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[zero]       \n\t"
>+      "pinsrh_0   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask]       \n\t"
>+      "punpcklwd  %[src],          %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[dest0],        %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest0],        %[dest0],          %[src]        \n\t"
>+      "psrlw      %[dest0],        %[dest0],          %[eight]      \n\t"
>+
>+      "gsldlc1    %[src],          0x0f(%[src_argb0])               \n\t"
>+      "gsldrc1    %[src],          0x08(%[src_argb0])               \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[zero]       \n\t"
>+      "pinsrh_0   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask]       \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[zero]       \n\t"
>+      "pinsrh_0   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask]       \n\t"
>+      "punpcklwd  %[src],          %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[dest1],        %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest1],        %[dest1],          %[src]        \n\t"
>+      "psrlw      %[dest1],        %[dest1],          %[eight]      \n\t"
>+
>+      "gsldlc1    %[src],          0x17(%[src_argb0])               \n\t"
>+      "gsldrc1    %[src],          0x10(%[src_argb0])               \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[zero]       \n\t"
>+      "pinsrh_0   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask]       \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[zero]       \n\t"
>+      "pinsrh_0   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask]       \n\t"
>+      "punpcklwd  %[src],          %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[dest2],        %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest2],        %[dest2],          %[src]        \n\t"
>+      "psrlw      %[dest2],        %[dest2],          %[eight]      \n\t"
>+
>+      "gsldlc1    %[src],          0x1f(%[src_argb0])               \n\t"
>+      "gsldrc1    %[src],          0x18(%[src_argb0])               \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[zero]       \n\t"
>+      "pinsrh_0   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask]       \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[zero]       \n\t"
>+      "pinsrh_0   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask]       \n\t"
>+      "punpcklwd  %[src],          %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[dest3],        %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest3],        %[dest3],          %[src]        \n\t"
>+      "psrlw      %[dest3],        %[dest3],          %[eight]      \n\t"
>+
>+      "packsswh   %[src_lo],       %[dest0],          %[dest1]      \n\t"
>+      "packsswh   %[src_hi],       %[dest2],          %[dest3]      \n\t"
>+      "packushb   %[dest0],        %[src_lo],         %[src_hi]     \n\t"
>+      "gssdlc1    %[dest0],        0x07(%[dst_y])                   \n\t"
>+      "gssdrc1    %[dest0],        0x00(%[dst_y])                   \n\t"
>+
>+      "daddiu     %[src_argb0],    %[src_argb0],      0x20          \n\t"
>+      "daddiu     %[dst_y],        %[dst_y],          0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x08          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src] "=&f"(src), [src_hi] "=&f"(src_hi), [src_lo] "=&f"(src_lo),
>+        [dest0] "=&f"(dest0), [dest1] "=&f"(dest1), [dest2] "=&f"(dest2),
>+        [dest3] "=&f"(dest3)
>+      : [src_argb0] "r"(src_argb0), [dst_y] "r"(dst_y), [width] "r"(width),
>+        [mask] "f"(mask), [value] "f"(value), [eight] "f"(0x08),
>+        [zero] "f"(0x00)
>+      : "memory");
>+}
>+
>+void RGBAToUVRow_MMI(const uint8_t* src_rgb0,
>+                     int src_stride_rgb,
>+                     uint8_t* dst_u,
>+                     uint8_t* dst_v,
>+                     int width) {
>+  uint64_t src_rgb1;
>+  uint64_t ftmp[12];
>+  const uint64_t value = 0x4040;
>+  const uint64_t mask_u = 0x0026004a00700002;
>+  const uint64_t mask_v = 0x00020070005e0012;
>+
>+  __asm__ volatile(
>+      "1:                                                               \n\t"
>+      "daddu      %[src_rgb1],     %[src_rgb0],       %[src_stride_rgb] \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x00(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x07(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "pinsrh_0   %[dest0_u],      %[src0],           %[value]          \n\t"
>+      "dsrl       %[dest0_v],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_3   %[dest0_v],      %[dest0_v],        %[value]          \n\t"
>+      "pmaddhw    %[dest0_u],      %[dest0_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest0_v],      %[dest0_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x08(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x0f(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x08(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x0f(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "pinsrh_0   %[src_lo],       %[src0],           %[value]          \n\t"
>+      "dsrl       %[src_hi],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest0_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest0_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest0_u],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest0_u],      %[dest0_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest0_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest0_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest0_v],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest0_v],      %[dest0_v],        %[eight]          \n\t"
>+
>+      "gsldrc1    %[src0],         0x10(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x17(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x10(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x17(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "pinsrh_0   %[dest1_u],      %[src0],           %[value]          \n\t"
>+      "dsrl       %[dest1_v],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_3   %[dest1_v],      %[dest1_v],        %[value]          \n\t"
>+      "pmaddhw    %[dest1_u],      %[dest1_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest1_v],      %[dest1_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x18(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x1f(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x18(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x1f(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "pinsrh_0   %[src_lo],       %[src0],           %[value]          \n\t"
>+      "dsrl       %[src_hi],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest1_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest1_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest1_u],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest1_u],      %[dest1_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest1_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest1_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest1_v],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest1_v],      %[dest1_v],        %[eight]          \n\t"
>+
>+      "gsldrc1    %[src0],         0x20(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x27(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x20(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x27(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "pinsrh_0   %[dest2_u],      %[src0],           %[value]          \n\t"
>+      "dsrl       %[dest2_v],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_3   %[dest2_v],      %[dest2_v],        %[value]          \n\t"
>+      "pmaddhw    %[dest2_u],      %[dest2_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest2_v],      %[dest2_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x28(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x2f(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x28(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x2f(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "pinsrh_0   %[src_lo],       %[src0],           %[value]          \n\t"
>+      "dsrl       %[src_hi],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest2_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest2_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest2_u],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest2_u],      %[dest2_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest2_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest2_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest2_v],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest2_v],      %[dest2_v],        %[eight]          \n\t"
>+
>+      "gsldrc1    %[src0],         0x30(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x37(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x30(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x37(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "pinsrh_0   %[dest3_u],      %[src0],           %[value]          \n\t"
>+      "dsrl       %[dest3_v],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_3   %[dest3_v],      %[dest3_v],        %[value]          \n\t"
>+      "pmaddhw    %[dest3_u],      %[dest3_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest3_v],      %[dest3_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x38(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x3f(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x38(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x3f(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "pinsrh_0   %[src_lo],       %[src0],           %[value]          \n\t"
>+      "dsrl       %[src_hi],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest3_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest3_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest3_u],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest3_u],      %[dest3_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest3_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest3_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest3_v],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest3_v],      %[dest3_v],        %[eight]          \n\t"
>+
>+      "packsswh   %[src0],         %[dest0_u],        %[dest1_u]        \n\t"
>+      "packsswh   %[src1],         %[dest2_u],        %[dest3_u]        \n\t"
>+      "packushb   %[dest0_u],      %[src0],           %[src1]           \n\t"
>+      "gssdlc1    %[dest0_u],      0x07(%[dst_u])                       \n\t"
>+      "gssdrc1    %[dest0_u],      0x00(%[dst_u])                       \n\t"
>+
>+      "packsswh   %[src0],         %[dest0_v],        %[dest1_v]        \n\t"
>+      "packsswh   %[src1],         %[dest2_v],        %[dest3_v]        \n\t"
>+      "packushb   %[dest0_v],      %[src0],           %[src1]           \n\t"
>+      "gssdlc1    %[dest0_v],      0x07(%[dst_v])                       \n\t"
>+      "gssdrc1    %[dest0_v],      0x00(%[dst_v])                       \n\t"
>+
>+      "daddiu     %[src_rgb0],     %[src_rgb0],       0x40              \n\t"
>+      "daddiu     %[dst_u],        %[dst_u],          0x08              \n\t"
>+      "daddiu     %[dst_v],        %[dst_v],          0x08              \n\t"
>+      "daddi      %[width],        %[width],         -0x10              \n\t"
>+      "bgtz       %[width],        1b                                   \n\t"
>+      : [src_rgb1] "=&r"(src_rgb1), [src0] "=&f"(ftmp[0]),
>+        [src1] "=&f"(ftmp[1]), [src_lo] "=&f"(ftmp[2]), [src_hi] "=&f"(ftmp[3]),
>+        [dest0_u] "=&f"(ftmp[4]), [dest0_v] "=&f"(ftmp[5]),
>+        [dest1_u] "=&f"(ftmp[6]), [dest1_v] "=&f"(ftmp[7]),
>+        [dest2_u] "=&f"(ftmp[8]), [dest2_v] "=&f"(ftmp[9]),
>+        [dest3_u] "=&f"(ftmp[10]), [dest3_v] "=&f"(ftmp[11])
>+      : [src_rgb0] "r"(src_rgb0), [src_stride_rgb] "r"(src_stride_rgb),
>+        [dst_u] "r"(dst_u), [dst_v] "r"(dst_v), [width] "r"(width),
>+        [mask_u] "f"(mask_u), [mask_v] "f"(mask_v), [value] "f"(value),
>+        [zero] "f"(0x00), [eight] "f"(0x08), [two] "f"(0x02),
>+        [sixteen] "f"(0x10)
>+      : "memory");
>+}
>+
>+void RGB24ToYRow_MMI(const uint8_t* src_argb0, uint8_t* dst_y, int width) {
>+  uint64_t src, src_hi, src_lo;
>+  uint64_t dest0, dest1, dest2, dest3;
>+  const uint64_t value = 0x1080;
>+  const uint64_t mask = 0x0001004200810019;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[src],          0x07(%[src_argb0])               \n\t"
>+      "gsldrc1    %[src],          0x00(%[src_argb0])               \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask]       \n\t"
>+      "dsll       %[src],          %[src],            %[eight]      \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask]       \n\t"
>+      "punpcklwd  %[src],          %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[dest0],        %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest0],        %[dest0],          %[src]        \n\t"
>+      "psrlw      %[dest0],        %[dest0],          %[eight]      \n\t"
>+
>+      "gsldlc1    %[src],          0x0d(%[src_argb0])               \n\t"
>+      "gsldrc1    %[src],          0x06(%[src_argb0])               \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask]       \n\t"
>+      "dsll       %[src],          %[src],            %[eight]      \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask]       \n\t"
>+      "punpcklwd  %[src],          %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[dest1],        %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest1],        %[dest1],          %[src]        \n\t"
>+      "psrlw      %[dest1],        %[dest1],          %[eight]      \n\t"
>+
>+      "gsldlc1    %[src],          0x13(%[src_argb0])               \n\t"
>+      "gsldrc1    %[src],          0x0c(%[src_argb0])               \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask]       \n\t"
>+      "dsll       %[src],          %[src],            %[eight]      \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask]       \n\t"
>+      "punpcklwd  %[src],          %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[dest2],        %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest2],        %[dest2],          %[src]        \n\t"
>+      "psrlw      %[dest2],        %[dest2],          %[eight]      \n\t"
>+
>+      "gsldlc1    %[src],          0x19(%[src_argb0])               \n\t"
>+      "gsldrc1    %[src],          0x12(%[src_argb0])               \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask]       \n\t"
>+      "dsll       %[src],          %[src],            %[eight]      \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask]       \n\t"
>+      "punpcklwd  %[src],          %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[dest3],        %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest3],        %[dest3],          %[src]        \n\t"
>+      "psrlw      %[dest3],        %[dest3],          %[eight]      \n\t"
>+
>+      "packsswh   %[src_lo],       %[dest0],          %[dest1]      \n\t"
>+      "packsswh   %[src_hi],       %[dest2],          %[dest3]      \n\t"
>+      "packushb   %[dest0],        %[src_lo],         %[src_hi]     \n\t"
>+      "gssdlc1    %[dest0],        0x07(%[dst_y])                   \n\t"
>+      "gssdrc1    %[dest0],        0x00(%[dst_y])                   \n\t"
>+
>+      "daddiu     %[src_argb0],    %[src_argb0],      0x18          \n\t"
>+      "daddiu     %[dst_y],        %[dst_y],          0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x08          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src] "=&f"(src), [src_hi] "=&f"(src_hi), [src_lo] "=&f"(src_lo),
>+        [dest0] "=&f"(dest0), [dest1] "=&f"(dest1), [dest2] "=&f"(dest2),
>+        [dest3] "=&f"(dest3)
>+      : [src_argb0] "r"(src_argb0), [dst_y] "r"(dst_y), [width] "r"(width),
>+        [mask] "f"(mask), [value] "f"(value), [eight] "f"(0x08),
>+        [zero] "f"(0x00)
>+      : "memory");
>+}
>+
>+void RGB24ToUVRow_MMI(const uint8_t* src_rgb0,
>+                      int src_stride_rgb,
>+                      uint8_t* dst_u,
>+                      uint8_t* dst_v,
>+                      int width) {
>+  uint64_t src_rgb1;
>+  uint64_t ftmp[12];
>+  const uint64_t value = 0x4040;
>+  const uint64_t mask_u = 0x0026004a00700002;
>+  const uint64_t mask_v = 0x00020070005e0012;
>+
>+  __asm__ volatile(
>+      "1:                                                               \n\t"
>+      "daddu      %[src_rgb1],     %[src_rgb0],       %[src_stride_rgb] \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x00(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x07(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "dsll       %[src0],         %[src0],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "dsll       %[src1],         %[src1],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "dsll       %[dest0_u],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[dest0_u],      %[dest0_u],        %[value]          \n\t"
>+      "pinsrh_3   %[dest0_v],      %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[dest0_u],      %[dest0_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest0_v],      %[dest0_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x06(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x0d(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x06(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x0d(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "dsll       %[src0],         %[src0],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "dsll       %[src1],         %[src1],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "dsll       %[src_lo],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[src_lo],       %[src_lo],         %[value]          \n\t"
>+      "pinsrh_3   %[src_hi],       %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest0_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest0_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest0_u],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest0_u],      %[dest0_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest0_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest0_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest0_v],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest0_v],      %[dest0_v],        %[eight]          \n\t"
>+
>+      "gsldrc1    %[src0],         0x0c(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x13(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x0c(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x13(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "dsll       %[src0],         %[src0],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "dsll       %[src1],         %[src1],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "dsll       %[dest1_u],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[dest1_u],      %[dest1_u],        %[value]          \n\t"
>+      "pinsrh_3   %[dest1_v],      %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[dest1_u],      %[dest1_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest1_v],      %[dest1_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x12(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x19(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x12(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x19(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "dsll       %[src0],         %[src0],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "dsll       %[src1],         %[src1],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "dsll       %[src_lo],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[src_lo],       %[src_lo],         %[value]          \n\t"
>+      "pinsrh_3   %[src_hi],       %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest1_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest1_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest1_u],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest1_u],      %[dest1_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest1_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest1_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest1_v],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest1_v],      %[dest1_v],        %[eight]          \n\t"
>+
>+      "gsldrc1    %[src0],         0x18(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x1f(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x18(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x1f(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "dsll       %[src0],         %[src0],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "dsll       %[src1],         %[src1],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "dsll       %[dest2_u],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[dest2_u],      %[dest2_u],        %[value]          \n\t"
>+      "pinsrh_3   %[dest2_v],      %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[dest2_u],      %[dest2_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest2_v],      %[dest2_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x1e(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x25(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x1e(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x25(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "dsll       %[src0],         %[src0],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "dsll       %[src1],         %[src1],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "dsll       %[src_lo],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[src_lo],       %[src_lo],         %[value]          \n\t"
>+      "pinsrh_3   %[src_hi],       %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest2_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest2_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest2_u],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest2_u],      %[dest2_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest2_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest2_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest2_v],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest2_v],      %[dest2_v],        %[eight]          \n\t"
>+
>+      "gsldrc1    %[src0],         0x24(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x2b(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x24(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x2b(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "dsll       %[src0],         %[src0],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "dsll       %[src1],         %[src1],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "dsll       %[dest3_u],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[dest3_u],      %[dest3_u],        %[value]          \n\t"
>+      "pinsrh_3   %[dest3_v],      %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[dest3_u],      %[dest3_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest3_v],      %[dest3_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x2a(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x31(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x2a(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x31(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "dsll       %[src0],         %[src0],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "dsll       %[src1],         %[src1],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "dsll       %[src_lo],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[src_lo],       %[src_lo],         %[value]          \n\t"
>+      "pinsrh_3   %[src_hi],       %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest3_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest3_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest3_u],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest3_u],      %[dest3_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest3_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest3_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest3_v],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest3_v],      %[dest3_v],        %[eight]          \n\t"
>+
>+      "packsswh   %[src0],         %[dest0_u],        %[dest1_u]        \n\t"
>+      "packsswh   %[src1],         %[dest2_u],        %[dest3_u]        \n\t"
>+      "packushb   %[dest0_u],      %[src0],           %[src1]           \n\t"
>+      "gssdlc1    %[dest0_u],      0x07(%[dst_u])                       \n\t"
>+      "gssdrc1    %[dest0_u],      0x00(%[dst_u])                       \n\t"
>+
>+      "packsswh   %[src0],         %[dest0_v],        %[dest1_v]        \n\t"
>+      "packsswh   %[src1],         %[dest2_v],        %[dest3_v]        \n\t"
>+      "packushb   %[dest0_v],      %[src0],           %[src1]           \n\t"
>+      "gssdlc1    %[dest0_v],      0x07(%[dst_v])                       \n\t"
>+      "gssdrc1    %[dest0_v],      0x00(%[dst_v])                       \n\t"
>+
>+      "daddiu     %[src_rgb0],     %[src_rgb0],       0x30              \n\t"
>+      "daddiu     %[dst_u],        %[dst_u],          0x08              \n\t"
>+      "daddiu     %[dst_v],        %[dst_v],          0x08              \n\t"
>+      "daddi      %[width],        %[width],         -0x10              \n\t"
>+      "bgtz       %[width],        1b                                   \n\t"
>+      : [src_rgb1] "=&r"(src_rgb1), [src0] "=&f"(ftmp[0]),
>+        [src1] "=&f"(ftmp[1]), [src_lo] "=&f"(ftmp[2]), [src_hi] "=&f"(ftmp[3]),
>+        [dest0_u] "=&f"(ftmp[4]), [dest0_v] "=&f"(ftmp[5]),
>+        [dest1_u] "=&f"(ftmp[6]), [dest1_v] "=&f"(ftmp[7]),
>+        [dest2_u] "=&f"(ftmp[8]), [dest2_v] "=&f"(ftmp[9]),
>+        [dest3_u] "=&f"(ftmp[10]), [dest3_v] "=&f"(ftmp[11])
>+      : [src_rgb0] "r"(src_rgb0), [src_stride_rgb] "r"(src_stride_rgb),
>+        [dst_u] "r"(dst_u), [dst_v] "r"(dst_v), [width] "r"(width),
>+        [mask_u] "f"(mask_u), [mask_v] "f"(mask_v), [value] "f"(value),
>+        [zero] "f"(0x00), [eight] "f"(0x08), [two] "f"(0x02),
>+        [sixteen] "f"(0x10)
>+      : "memory");
>+}
>+
>+void RAWToYRow_MMI(const uint8_t* src_argb0, uint8_t* dst_y, int width) {
>+  uint64_t src, src_hi, src_lo;
>+  uint64_t dest0, dest1, dest2, dest3;
>+  const uint64_t value = 0x1080;
>+  const uint64_t mask = 0x0001001900810042;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[src],          0x07(%[src_argb0])               \n\t"
>+      "gsldrc1    %[src],          0x00(%[src_argb0])               \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask]       \n\t"
>+      "dsll       %[src],          %[src],            %[eight]      \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask]       \n\t"
>+      "punpcklwd  %[src],          %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[dest0],        %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest0],        %[dest0],          %[src]        \n\t"
>+      "psrlw      %[dest0],        %[dest0],          %[eight]      \n\t"
>+
>+      "gsldlc1    %[src],          0x0d(%[src_argb0])               \n\t"
>+      "gsldrc1    %[src],          0x06(%[src_argb0])               \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask]       \n\t"
>+      "dsll       %[src],          %[src],            %[eight]      \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask]       \n\t"
>+      "punpcklwd  %[src],          %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[dest1],        %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest1],        %[dest1],          %[src]        \n\t"
>+      "psrlw      %[dest1],        %[dest1],          %[eight]      \n\t"
>+
>+      "gsldlc1    %[src],          0x13(%[src_argb0])               \n\t"
>+      "gsldrc1    %[src],          0x0c(%[src_argb0])               \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask]       \n\t"
>+      "dsll       %[src],          %[src],            %[eight]      \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask]       \n\t"
>+      "punpcklwd  %[src],          %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[dest2],        %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest2],        %[dest2],          %[src]        \n\t"
>+      "psrlw      %[dest2],        %[dest2],          %[eight]      \n\t"
>+
>+      "gsldlc1    %[src],          0x19(%[src_argb0])               \n\t"
>+      "gsldrc1    %[src],          0x12(%[src_argb0])               \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask]       \n\t"
>+      "dsll       %[src],          %[src],            %[eight]      \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[zero]       \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask]       \n\t"
>+      "punpcklwd  %[src],          %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[dest3],        %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest3],        %[dest3],          %[src]        \n\t"
>+      "psrlw      %[dest3],        %[dest3],          %[eight]      \n\t"
>+
>+      "packsswh   %[src_lo],       %[dest0],          %[dest1]      \n\t"
>+      "packsswh   %[src_hi],       %[dest2],          %[dest3]      \n\t"
>+      "packushb   %[dest0],        %[src_lo],         %[src_hi]     \n\t"
>+      "gssdlc1    %[dest0],        0x07(%[dst_y])                   \n\t"
>+      "gssdrc1    %[dest0],        0x00(%[dst_y])                   \n\t"
>+
>+      "daddiu     %[src_argb0],    %[src_argb0],      0x18          \n\t"
>+      "daddiu     %[dst_y],        %[dst_y],          0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x08          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src] "=&f"(src), [src_hi] "=&f"(src_hi), [src_lo] "=&f"(src_lo),
>+        [dest0] "=&f"(dest0), [dest1] "=&f"(dest1), [dest2] "=&f"(dest2),
>+        [dest3] "=&f"(dest3)
>+      : [src_argb0] "r"(src_argb0), [dst_y] "r"(dst_y), [width] "r"(width),
>+        [mask] "f"(mask), [value] "f"(value), [eight] "f"(0x08),
>+        [zero] "f"(0x00)
>+      : "memory");
>+}
>+
>+void RAWToUVRow_MMI(const uint8_t* src_rgb0,
>+                    int src_stride_rgb,
>+                    uint8_t* dst_u,
>+                    uint8_t* dst_v,
>+                    int width) {
>+  uint64_t src_rgb1;
>+  uint64_t ftmp[12];
>+  const uint64_t value = 0x4040;
>+  const uint64_t mask_u = 0x00020070004a0026;
>+  const uint64_t mask_v = 0x0012005e00700002;
>+
>+  __asm__ volatile(
>+      "1:                                                               \n\t"
>+      "daddu      %[src_rgb1],     %[src_rgb0],       %[src_stride_rgb] \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x00(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x07(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "dsll       %[src0],         %[src0],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "dsll       %[src1],         %[src1],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "pinsrh_3   %[dest0_u],      %[src0],           %[value]          \n\t"
>+      "dsll       %[dest0_v],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[dest0_v],      %[dest0_v],        %[value]          \n\t"
>+      "pmaddhw    %[dest0_u],      %[dest0_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest0_v],      %[dest0_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x06(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x0d(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x06(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x0d(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "dsll       %[src0],         %[src0],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "dsll       %[src1],         %[src1],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "pinsrh_3   %[src_lo],       %[src0],           %[value]          \n\t"
>+      "dsll       %[src_hi],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[src_hi],       %[src_hi],         %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest0_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest0_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest0_u],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest0_u],      %[dest0_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest0_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest0_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest0_v],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest0_v],      %[dest0_v],        %[eight]          \n\t"
>+
>+      "gsldrc1    %[src0],         0x0c(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x13(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x0c(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x13(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "dsll       %[src0],         %[src0],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "dsll       %[src1],         %[src1],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "pinsrh_3   %[dest1_u],      %[src0],           %[value]          \n\t"
>+      "dsll       %[dest1_v],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[dest1_v],      %[dest1_v],        %[value]          \n\t"
>+      "pmaddhw    %[dest1_u],      %[dest1_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest1_v],      %[dest1_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x12(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x19(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x12(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x19(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "dsll       %[src0],         %[src0],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "dsll       %[src1],         %[src1],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "pinsrh_3   %[src_lo],       %[src0],           %[value]          \n\t"
>+      "dsll       %[src_hi],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[src_hi],       %[src_hi],         %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest1_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest1_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest1_u],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest1_u],      %[dest1_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest1_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest1_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest1_v],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest1_v],      %[dest1_v],        %[eight]          \n\t"
>+
>+      "gsldrc1    %[src0],         0x18(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x1f(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x18(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x1f(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "dsll       %[src0],         %[src0],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "dsll       %[src1],         %[src1],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "pinsrh_3   %[dest2_u],      %[src0],           %[value]          \n\t"
>+      "dsll       %[dest2_v],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[dest2_v],      %[dest2_v],        %[value]          \n\t"
>+      "pmaddhw    %[dest2_u],      %[dest2_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest2_v],      %[dest2_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x1e(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x25(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x1e(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x25(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "dsll       %[src0],         %[src0],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "dsll       %[src1],         %[src1],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "pinsrh_3   %[src_lo],       %[src0],           %[value]          \n\t"
>+      "dsll       %[src_hi],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[src_hi],       %[src_hi],         %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest2_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest2_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest2_u],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest2_u],      %[dest2_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest2_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest2_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest2_v],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest2_v],      %[dest2_v],        %[eight]          \n\t"
>+
>+      "gsldrc1    %[src0],         0x24(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x2b(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x24(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x2b(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "dsll       %[src0],         %[src0],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "dsll       %[src1],         %[src1],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "pinsrh_3   %[dest3_u],      %[src0],           %[value]          \n\t"
>+      "dsll       %[dest3_v],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[dest3_v],      %[dest3_v],        %[value]          \n\t"
>+      "pmaddhw    %[dest3_u],      %[dest3_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest3_v],      %[dest3_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x2a(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x31(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x2a(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x31(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "dsll       %[src0],         %[src0],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src_lo],         %[src_hi]         \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_lo]         \n\t"
>+      "dsll       %[src1],         %[src1],           %[eight]          \n\t"
>+      "punpckhbh  %[src_hi],       %[src1],           %[zero]           \n\t"
>+      "paddh      %[src0],         %[src0],           %[src_hi]         \n\t"
>+      "psrlh      %[src0],         %[src0],           %[two]            \n\t"
>+      "pinsrh_3   %[src_lo],       %[src0],           %[value]          \n\t"
>+      "dsll       %[src_hi],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[src_hi],       %[src_hi],         %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest3_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest3_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest3_u],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest3_u],      %[dest3_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest3_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest3_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest3_v],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest3_v],      %[dest3_v],        %[eight]          \n\t"
>+
>+      "packsswh   %[src0],         %[dest0_u],        %[dest1_u]        \n\t"
>+      "packsswh   %[src1],         %[dest2_u],        %[dest3_u]        \n\t"
>+      "packushb   %[dest0_u],      %[src0],           %[src1]           \n\t"
>+      "gssdlc1    %[dest0_u],      0x07(%[dst_u])                       \n\t"
>+      "gssdrc1    %[dest0_u],      0x00(%[dst_u])                       \n\t"
>+
>+      "packsswh   %[src0],         %[dest0_v],        %[dest1_v]        \n\t"
>+      "packsswh   %[src1],         %[dest2_v],        %[dest3_v]        \n\t"
>+      "packushb   %[dest0_v],      %[src0],           %[src1]           \n\t"
>+      "gssdlc1    %[dest0_v],      0x07(%[dst_v])                       \n\t"
>+      "gssdrc1    %[dest0_v],      0x00(%[dst_v])                       \n\t"
>+
>+      "daddiu     %[src_rgb0],     %[src_rgb0],       0x30              \n\t"
>+      "daddiu     %[dst_u],        %[dst_u],          0x08              \n\t"
>+      "daddiu     %[dst_v],        %[dst_v],          0x08              \n\t"
>+      "daddi      %[width],        %[width],         -0x10              \n\t"
>+      "bgtz       %[width],        1b                                   \n\t"
>+      : [src_rgb1] "=&r"(src_rgb1), [src0] "=&f"(ftmp[0]),
>+        [src1] "=&f"(ftmp[1]), [src_lo] "=&f"(ftmp[2]), [src_hi] "=&f"(ftmp[3]),
>+        [dest0_u] "=&f"(ftmp[4]), [dest0_v] "=&f"(ftmp[5]),
>+        [dest1_u] "=&f"(ftmp[6]), [dest1_v] "=&f"(ftmp[7]),
>+        [dest2_u] "=&f"(ftmp[8]), [dest2_v] "=&f"(ftmp[9]),
>+        [dest3_u] "=&f"(ftmp[10]), [dest3_v] "=&f"(ftmp[11])
>+      : [src_rgb0] "r"(src_rgb0), [src_stride_rgb] "r"(src_stride_rgb),
>+        [dst_u] "r"(dst_u), [dst_v] "r"(dst_v), [width] "r"(width),
>+        [mask_u] "f"(mask_u), [mask_v] "f"(mask_v), [value] "f"(value),
>+        [zero] "f"(0x00), [eight] "f"(0x08), [two] "f"(0x02),
>+        [sixteen] "f"(0x10)
>+      : "memory");
>+}
>+
>+void ARGBToYJRow_MMI(const uint8_t* src_argb0, uint8_t* dst_y, int width) {
>+  uint64_t src, src_hi, src_lo;
>+  uint64_t dest, dest0, dest1, dest2, dest3;
>+  uint64_t tmp0, tmp1;
>+  const uint64_t shift = 0x07;
>+  const uint64_t value = 0x0040;
>+  const uint64_t mask0 = 0x0;
>+  const uint64_t mask1 = 0x00010026004B000FULL;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[src],          0x07(%[src_ptr])                 \n\t"
>+      "gsldrc1    %[src],          0x00(%[src_ptr])                 \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[mask0]      \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask1]      \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[mask0]      \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask1]      \n\t"
>+      "punpcklwd  %[tmp0],         %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[tmp1],         %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest0],        %[tmp0],           %[tmp1]       \n\t"
>+      "psrlw      %[dest0],        %[dest0],          %[shift]      \n\t"
>+
>+      "gsldlc1    %[src],          0x0f(%[src_ptr])                 \n\t"
>+      "gsldrc1    %[src],          0x08(%[src_ptr])                 \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[mask0]      \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask1]      \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[mask0]      \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask1]      \n\t"
>+      "punpcklwd  %[tmp0],         %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[tmp1],         %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest1],        %[tmp0],           %[tmp1]       \n\t"
>+      "psrlw      %[dest1],        %[dest1],          %[shift]      \n\t"
>+
>+      "gsldlc1    %[src],          0x17(%[src_ptr])                 \n\t"
>+      "gsldrc1    %[src],          0x10(%[src_ptr])                 \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[mask0]      \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask1]      \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[mask0]      \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask1]      \n\t"
>+      "punpcklwd  %[tmp0],         %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[tmp1],         %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest2],        %[tmp0],           %[tmp1]       \n\t"
>+      "psrlw      %[dest2],        %[dest2],          %[shift]      \n\t"
>+
>+      "gsldlc1    %[src],          0x1f(%[src_ptr])                 \n\t"
>+      "gsldrc1    %[src],          0x18(%[src_ptr])                 \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[mask0]      \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[value]      \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask1]      \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[mask0]      \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]      \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask1]      \n\t"
>+      "punpcklwd  %[tmp0],         %[src_lo],         %[src_hi]     \n\t"
>+      "punpckhwd  %[tmp1],         %[src_lo],         %[src_hi]     \n\t"
>+      "paddw      %[dest3],        %[tmp0],           %[tmp1]       \n\t"
>+      "psrlw      %[dest3],        %[dest3],          %[shift]      \n\t"
>+
>+      "packsswh   %[tmp0],         %[dest0],          %[dest1]      \n\t"
>+      "packsswh   %[tmp1],         %[dest2],          %[dest3]      \n\t"
>+      "packushb   %[dest],         %[tmp0],           %[tmp1]       \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x20          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x08          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src] "=&f"(src), [dest] "=&f"(dest), [src_hi] "=&f"(src_hi),
>+        [src_lo] "=&f"(src_lo), [dest0] "=&f"(dest0), [dest1] "=&f"(dest1),
>+        [dest2] "=&f"(dest2), [dest3] "=&f"(dest3), [tmp0] "=&f"(tmp0),
>+        [tmp1] "=&f"(tmp1)
>+      : [src_ptr] "r"(src_argb0), [dst_ptr] "r"(dst_y), [mask0] "f"(mask0),
>+        [mask1] "f"(mask1), [shift] "f"(shift), [value] "f"(value),
>+        [width] "r"(width)
>+      : "memory");
>+}
>+
>+void ARGBToUVJRow_MMI(const uint8_t* src_rgb0,
>+                      int src_stride_rgb,
>+                      uint8_t* dst_u,
>+                      uint8_t* dst_v,
>+                      int width) {
>+  uint64_t src_rgb1;
>+  uint64_t ftmp[12];
>+  const uint64_t value = 0x4040;
>+  const uint64_t mask_u = 0x002b0054007f0002;
>+  const uint64_t mask_v = 0x0002007f006b0014;
>+
>+  __asm__ volatile(
>+      "1:                                                               \n\t"
>+      "daddu      %[src_rgb1],     %[src_rgb0],       %[src_stride_rgb] \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x00(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x07(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "punpcklbh  %[src0],         %[src1],           %[zero]           \n\t"
>+      "punpckhbh  %[src1],         %[src1],           %[zero]           \n\t"
>+      "pavgh      %[src0],         %[src_lo],         %[src0]           \n\t"
>+      "pavgh      %[src1],         %[src_hi],         %[src1]           \n\t"
>+      "pavgh      %[src0],         %[src0],           %[src1]           \n\t"
>+      "dsll       %[dest0_u],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[dest0_u],      %[dest0_u],        %[value]          \n\t"
>+      "pinsrh_3   %[dest0_v],      %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[dest0_u],      %[dest0_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest0_v],      %[dest0_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x08(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x0f(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x08(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x0f(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "punpcklbh  %[src0],         %[src1],           %[zero]           \n\t"
>+      "punpckhbh  %[src1],         %[src1],           %[zero]           \n\t"
>+      "pavgh      %[src0],         %[src_lo],         %[src0]           \n\t"
>+      "pavgh      %[src1],         %[src_hi],         %[src1]           \n\t"
>+      "pavgh      %[src0],         %[src0],           %[src1]           \n\t"
>+      "dsll       %[src_lo],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[src_lo],       %[src_lo],         %[value]          \n\t"
>+      "pinsrh_3   %[src_hi],       %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest0_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest0_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest0_u],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest0_u],      %[dest0_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest0_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest0_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest0_v],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest0_v],      %[dest0_v],        %[eight]          \n\t"
>+
>+      "gsldrc1    %[src0],         0x10(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x17(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x10(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x17(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "punpcklbh  %[src0],         %[src1],           %[zero]           \n\t"
>+      "punpckhbh  %[src1],         %[src1],           %[zero]           \n\t"
>+      "pavgh      %[src0],         %[src_lo],         %[src0]           \n\t"
>+      "pavgh      %[src1],         %[src_hi],         %[src1]           \n\t"
>+      "pavgh      %[src0],         %[src0],           %[src1]           \n\t"
>+      "dsll       %[dest1_u],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[dest1_u],      %[dest1_u],        %[value]          \n\t"
>+      "pinsrh_3   %[dest1_v],      %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[dest1_u],      %[dest1_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest1_v],      %[dest1_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x18(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x1f(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x18(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x1f(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "punpcklbh  %[src0],         %[src1],           %[zero]           \n\t"
>+      "punpckhbh  %[src1],         %[src1],           %[zero]           \n\t"
>+      "pavgh      %[src0],         %[src_lo],         %[src0]           \n\t"
>+      "pavgh      %[src1],         %[src_hi],         %[src1]           \n\t"
>+      "pavgh      %[src0],         %[src0],           %[src1]           \n\t"
>+      "dsll       %[src_lo],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[src_lo],       %[src_lo],         %[value]          \n\t"
>+      "pinsrh_3   %[src_hi],       %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest1_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest1_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest1_u],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest1_u],      %[dest1_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest1_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest1_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest1_v],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest1_v],      %[dest1_v],        %[eight]          \n\t"
>+
>+      "gsldrc1    %[src0],         0x20(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x27(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x20(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x27(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "punpcklbh  %[src0],         %[src1],           %[zero]           \n\t"
>+      "punpckhbh  %[src1],         %[src1],           %[zero]           \n\t"
>+      "pavgh      %[src0],         %[src_lo],         %[src0]           \n\t"
>+      "pavgh      %[src1],         %[src_hi],         %[src1]           \n\t"
>+      "pavgh      %[src0],         %[src0],           %[src1]           \n\t"
>+      "dsll       %[dest2_u],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[dest2_u],      %[dest2_u],        %[value]          \n\t"
>+      "pinsrh_3   %[dest2_v],      %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[dest2_u],      %[dest2_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest2_v],      %[dest2_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x28(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x2f(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x28(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x2f(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "punpcklbh  %[src0],         %[src1],           %[zero]           \n\t"
>+      "punpckhbh  %[src1],         %[src1],           %[zero]           \n\t"
>+      "pavgh      %[src0],         %[src_lo],         %[src0]           \n\t"
>+      "pavgh      %[src1],         %[src_hi],         %[src1]           \n\t"
>+      "pavgh      %[src0],         %[src0],           %[src1]           \n\t"
>+      "dsll       %[src_lo],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[src_lo],       %[src_lo],         %[value]          \n\t"
>+      "pinsrh_3   %[src_hi],       %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest2_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest2_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest2_u],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest2_u],      %[dest2_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest2_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest2_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest2_v],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest2_v],      %[dest2_v],        %[eight]          \n\t"
>+
>+      "gsldrc1    %[src0],         0x30(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x37(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x30(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x37(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "punpcklbh  %[src0],         %[src1],           %[zero]           \n\t"
>+      "punpckhbh  %[src1],         %[src1],           %[zero]           \n\t"
>+      "pavgh      %[src0],         %[src_lo],         %[src0]           \n\t"
>+      "pavgh      %[src1],         %[src_hi],         %[src1]           \n\t"
>+      "pavgh      %[src0],         %[src0],           %[src1]           \n\t"
>+      "dsll       %[dest3_u],      %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[dest3_u],      %[dest3_u],        %[value]          \n\t"
>+      "pinsrh_3   %[dest3_v],      %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[dest3_u],      %[dest3_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest3_v],      %[dest3_v],        %[mask_v]         \n\t"
>+
>+      "gsldrc1    %[src0],         0x38(%[src_rgb0])                    \n\t"
>+      "gsldlc1    %[src0],         0x3f(%[src_rgb0])                    \n\t"
>+      "gsldrc1    %[src1],         0x38(%[src_rgb1])                    \n\t"
>+      "gsldlc1    %[src1],         0x3f(%[src_rgb1])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "punpcklbh  %[src0],         %[src1],           %[zero]           \n\t"
>+      "punpckhbh  %[src1],         %[src1],           %[zero]           \n\t"
>+      "pavgh      %[src0],         %[src_lo],         %[src0]           \n\t"
>+      "pavgh      %[src1],         %[src_hi],         %[src1]           \n\t"
>+      "pavgh      %[src0],         %[src0],           %[src1]           \n\t"
>+      "dsll       %[src_lo],       %[src0],           %[sixteen]        \n\t"
>+      "pinsrh_0   %[src_lo],       %[src_lo],         %[value]          \n\t"
>+      "pinsrh_3   %[src_hi],       %[src0],           %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest3_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest3_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest3_u],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest3_u],      %[dest3_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest3_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest3_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest3_v],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest3_v],      %[dest3_v],        %[eight]          \n\t"
>+
>+      "packsswh   %[src0],         %[dest0_u],        %[dest1_u]        \n\t"
>+      "packsswh   %[src1],         %[dest2_u],        %[dest3_u]        \n\t"
>+      "packushb   %[dest0_u],      %[src0],           %[src1]           \n\t"
>+      "gssdlc1    %[dest0_u],      0x07(%[dst_u])                       \n\t"
>+      "gssdrc1    %[dest0_u],      0x00(%[dst_u])                       \n\t"
>+
>+      "packsswh   %[src0],         %[dest0_v],        %[dest1_v]        \n\t"
>+      "packsswh   %[src1],         %[dest2_v],        %[dest3_v]        \n\t"
>+      "packushb   %[dest0_v],      %[src0],           %[src1]           \n\t"
>+      "gssdlc1    %[dest0_v],      0x07(%[dst_v])                       \n\t"
>+      "gssdrc1    %[dest0_v],      0x00(%[dst_v])                       \n\t"
>+
>+      "daddiu     %[src_rgb0],     %[src_rgb0],       0x40              \n\t"
>+      "daddiu     %[dst_u],        %[dst_u],          0x08              \n\t"
>+      "daddiu     %[dst_v],        %[dst_v],          0x08              \n\t"
>+      "daddi      %[width],        %[width],         -0x10              \n\t"
>+      "bgtz       %[width],        1b                                   \n\t"
>+      : [src_rgb1] "=&r"(src_rgb1), [src0] "=&f"(ftmp[0]),
>+        [src1] "=&f"(ftmp[1]), [src_lo] "=&f"(ftmp[2]), [src_hi] "=&f"(ftmp[3]),
>+        [dest0_u] "=&f"(ftmp[4]), [dest0_v] "=&f"(ftmp[5]),
>+        [dest1_u] "=&f"(ftmp[6]), [dest1_v] "=&f"(ftmp[7]),
>+        [dest2_u] "=&f"(ftmp[8]), [dest2_v] "=&f"(ftmp[9]),
>+        [dest3_u] "=&f"(ftmp[10]), [dest3_v] "=&f"(ftmp[11])
>+      : [src_rgb0] "r"(src_rgb0), [src_stride_rgb] "r"(src_stride_rgb),
>+        [dst_u] "r"(dst_u), [dst_v] "r"(dst_v), [width] "r"(width),
>+        [mask_u] "f"(mask_u), [mask_v] "f"(mask_v), [value] "f"(value),
>+        [zero] "f"(0x00), [eight] "f"(0x08), [two] "f"(0x02),
>+        [sixteen] "f"(0x10)
>+      : "memory");
>+}
>+
>+void RGB565ToYRow_MMI(const uint8_t* src_rgb565, uint8_t* dst_y, int width) {
>+  uint64_t ftmp[11];
>+  const uint64_t value = 0x1080108010801080;
>+  const uint64_t mask = 0x0001004200810019;
>+  uint64_t c0 = 0x001f001f001f001f;
>+  uint64_t c1 = 0x00ff00ff00ff00ff;
>+  uint64_t c2 = 0x0007000700070007;
>+  __asm__ volatile(
>+      "1:                                                            \n\t"
>+      "gsldrc1    %[src0],        0x00(%[src_rgb565])                \n\t"
>+      "gsldlc1    %[src0],        0x07(%[src_rgb565])                \n\t"
>+      "psrlh      %[src1],        %[src0],             %[eight]      \n\t"
>+      "and        %[b],           %[src0],             %[c0]         \n\t"
>+      "and        %[src0],        %[src0],             %[c1]         \n\t"
>+      "psrlh      %[src0],        %[src0],             %[five]       \n\t"
>+      "and        %[g],           %[src1],             %[c2]         \n\t"
>+      "psllh      %[g],           %[g],                %[three]      \n\t"
>+      "or         %[g],           %[src0],             %[g]          \n\t"
>+      "psrlh      %[r],           %[src1],             %[three]      \n\t"
>+      "psllh      %[src0],        %[b],                %[three]      \n\t"
>+      "psrlh      %[src1],        %[b],                %[two]        \n\t"
>+      "or         %[b],           %[src0],             %[src1]       \n\t"
>+      "psllh      %[src0],        %[g],                %[two]        \n\t"
>+      "psrlh      %[src1],        %[g],                %[four]       \n\t"
>+      "or         %[g],           %[src0],             %[src1]       \n\t"
>+      "psllh      %[src0],        %[r],                %[three]      \n\t"
>+      "psrlh      %[src1],        %[r],                %[two]        \n\t"
>+      "or         %[r],           %[src0],             %[src1]       \n\t"
>+      "punpcklhw  %[src0],        %[b],                %[r]          \n\t"
>+      "punpcklhw  %[src1],        %[g],                %[value]      \n\t"
>+      "punpcklhw  %[src_lo],      %[src0],             %[src1]       \n\t"
>+      "punpckhhw  %[src_hi],      %[src0],             %[src1]       \n\t"
>+      "pmaddhw    %[src_lo],      %[src_lo],           %[mask]       \n\t"
>+      "pmaddhw    %[src_hi],      %[src_hi],           %[mask]       \n\t"
>+      "punpcklwd  %[src0],        %[src_lo],           %[src_hi]     \n\t"
>+      "punpckhwd  %[src1],        %[src_lo],           %[src_hi]     \n\t"
>+      "paddw      %[dest0],       %[src0],             %[src1]       \n\t"
>+      "psrlw      %[dest0],       %[dest0],            %[eight]      \n\t"
>+
>+      "punpckhhw  %[src0],        %[b],                %[r]          \n\t"
>+      "punpckhhw  %[src1],        %[g],                %[value]      \n\t"
>+      "punpcklhw  %[src_lo],      %[src0],             %[src1]       \n\t"
>+      "punpckhhw  %[src_hi],      %[src0],             %[src1]       \n\t"
>+      "pmaddhw    %[src_lo],      %[src_lo],           %[mask]       \n\t"
>+      "pmaddhw    %[src_hi],      %[src_hi],           %[mask]       \n\t"
>+      "punpcklwd  %[src0],        %[src_lo],           %[src_hi]     \n\t"
>+      "punpckhwd  %[src1],        %[src_lo],           %[src_hi]     \n\t"
>+      "paddw      %[dest1],       %[src0],             %[src1]       \n\t"
>+      "psrlw      %[dest1],       %[dest1],            %[eight]      \n\t"
>+
>+      "gsldrc1    %[src0],        0x08(%[src_rgb565])                \n\t"
>+      "gsldlc1    %[src0],        0x0f(%[src_rgb565])                \n\t"
>+      "psrlh      %[src1],        %[src0],             %[eight]      \n\t"
>+      "and        %[b],           %[src0],             %[c0]         \n\t"
>+      "and        %[src0],        %[src0],             %[c1]         \n\t"
>+      "psrlh      %[src0],        %[src0],             %[five]       \n\t"
>+      "and        %[g],           %[src1],             %[c2]         \n\t"
>+      "psllh      %[g],           %[g],                %[three]      \n\t"
>+      "or         %[g],           %[src0],             %[g]          \n\t"
>+      "psrlh      %[r],           %[src1],             %[three]      \n\t"
>+      "psllh      %[src0],        %[b],                %[three]      \n\t"
>+      "psrlh      %[src1],        %[b],                %[two]        \n\t"
>+      "or         %[b],           %[src0],             %[src1]       \n\t"
>+      "psllh      %[src0],        %[g],                %[two]        \n\t"
>+      "psrlh      %[src1],        %[g],                %[four]       \n\t"
>+      "or         %[g],           %[src0],             %[src1]       \n\t"
>+      "psllh      %[src0],        %[r],                %[three]      \n\t"
>+      "psrlh      %[src1],        %[r],                %[two]        \n\t"
>+      "or         %[r],           %[src0],             %[src1]       \n\t"
>+      "punpcklhw  %[src0],        %[b],                %[r]          \n\t"
>+      "punpcklhw  %[src1],        %[g],                %[value]      \n\t"
>+      "punpcklhw  %[src_lo],      %[src0],             %[src1]       \n\t"
>+      "punpckhhw  %[src_hi],      %[src0],             %[src1]       \n\t"
>+      "pmaddhw    %[src_lo],      %[src_lo],           %[mask]       \n\t"
>+      "pmaddhw    %[src_hi],      %[src_hi],           %[mask]       \n\t"
>+      "punpcklwd  %[src0],        %[src_lo],           %[src_hi]     \n\t"
>+      "punpckhwd  %[src1],        %[src_lo],           %[src_hi]     \n\t"
>+      "paddw      %[dest2],       %[src0],             %[src1]       \n\t"
>+      "psrlw      %[dest2],       %[dest2],            %[eight]      \n\t"
>+
>+      "punpckhhw  %[src0],        %[b],                %[r]          \n\t"
>+      "punpckhhw  %[src1],        %[g],                %[value]      \n\t"
>+      "punpcklhw  %[src_lo],      %[src0],             %[src1]       \n\t"
>+      "punpckhhw  %[src_hi],      %[src0],             %[src1]       \n\t"
>+      "pmaddhw    %[src_lo],      %[src_lo],           %[mask]       \n\t"
>+      "pmaddhw    %[src_hi],      %[src_hi],           %[mask]       \n\t"
>+      "punpcklwd  %[src0],        %[src_lo],           %[src_hi]     \n\t"
>+      "punpckhwd  %[src1],        %[src_lo],           %[src_hi]     \n\t"
>+      "paddw      %[dest3],       %[src0],             %[src1]       \n\t"
>+      "psrlw      %[dest3],       %[dest3],            %[eight]      \n\t"
>+
>+      "packsswh   %[src_lo],      %[dest0],            %[dest1]      \n\t"
>+      "packsswh   %[src_hi],      %[dest2],            %[dest3]      \n\t"
>+      "packushb   %[dest0],       %[src_lo],           %[src_hi]     \n\t"
>+      "gssdlc1    %[dest0],       0x07(%[dst_y])                     \n\t"
>+      "gssdrc1    %[dest0],       0x00(%[dst_y])                     \n\t"
>+
>+      "daddiu    %[src_rgb565],   %[src_rgb565],       0x10          \n\t"
>+      "daddiu    %[dst_y],        %[dst_y],            0x08          \n\t"
>+      "daddiu    %[width],        %[width],           -0x08          \n\t"
>+      "bgtz      %[width],        1b                                 \n\t"
>+      : [src0] "=&f"(ftmp[0]), [src1] "=&f"(ftmp[1]), [src_lo] "=&f"(ftmp[2]),
>+        [src_hi] "=&f"(ftmp[3]), [b] "=&f"(ftmp[4]), [g] "=&f"(ftmp[5]),
>+        [r] "=&f"(ftmp[6]), [dest0] "=&f"(ftmp[7]), [dest1] "=&f"(ftmp[8]),
>+        [dest2] "=&f"(ftmp[9]), [dest3] "=&f"(ftmp[10])
>+      : [src_rgb565] "r"(src_rgb565), [dst_y] "r"(dst_y), [value] "f"(value),
>+        [width] "r"(width), [c0] "f"(c0), [c1] "f"(c1), [c2] "f"(c2),
>+        [mask] "f"(mask), [eight] "f"(0x08), [five] "f"(0x05),
>+        [three] "f"(0x03), [two] "f"(0x02), [four] "f"(0x04)
>+      : "memory");
>+}
>+
>+void ARGB1555ToYRow_MMI(const uint8_t* src_argb1555,
>+                        uint8_t* dst_y,
>+                        int width) {
>+  uint64_t ftmp[11];
>+  const uint64_t value = 0x1080108010801080;
>+  const uint64_t mask = 0x0001004200810019;
>+  uint64_t c0 = 0x001f001f001f001f;
>+  uint64_t c1 = 0x00ff00ff00ff00ff;
>+  uint64_t c2 = 0x0003000300030003;
>+  uint64_t c3 = 0x007c007c007c007c;
>+  __asm__ volatile(
>+      "1:                                                            \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src_argb1555])             \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src_argb1555])             \n\t"
>+      "psrlh      %[src1],         %[src0],              %[eight]    \n\t"
>+      "and        %[b],            %[src0],              %[c0]       \n\t"
>+      "and        %[src0],         %[src0],              %[c1]       \n\t"
>+      "psrlh      %[src0],         %[src0],              %[five]     \n\t"
>+      "and        %[g],            %[src1],              %[c2]       \n\t"
>+      "psllh      %[g],            %[g],                 %[three]    \n\t"
>+      "or         %[g],            %[src0],              %[g]        \n\t"
>+      "and        %[r],            %[src1],              %[c3]       \n\t"
>+      "psrlh      %[r],            %[r],                 %[two]      \n\t"
>+      "psllh      %[src0],         %[b],                 %[three]    \n\t"
>+      "psrlh      %[src1],         %[b],                 %[two]      \n\t"
>+      "or         %[b],            %[src0],              %[src1]     \n\t"
>+      "psllh      %[src0],         %[g],                 %[three]    \n\t"
>+      "psrlh      %[src1],         %[g],                 %[two]      \n\t"
>+      "or         %[g],            %[src0],              %[src1]     \n\t"
>+      "psllh      %[src0],         %[r],                 %[three]    \n\t"
>+      "psrlh      %[src1],         %[r],                 %[two]      \n\t"
>+      "or         %[r],            %[src0],              %[src1]     \n\t"
>+      "punpcklhw  %[src0],         %[b],                 %[r]        \n\t"
>+      "punpcklhw  %[src1],         %[g],                 %[value]    \n\t"
>+      "punpcklhw  %[src_lo],       %[src0],              %[src1]     \n\t"
>+      "punpckhhw  %[src_hi],       %[src0],              %[src1]     \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],            %[mask]     \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],            %[mask]     \n\t"
>+      "punpcklwd  %[src0],         %[src_lo],            %[src_hi]   \n\t"
>+      "punpckhwd  %[src1],         %[src_lo],            %[src_hi]   \n\t"
>+      "paddw      %[dest0],        %[src0],              %[src1]     \n\t"
>+      "psrlw      %[dest0],        %[dest0],             %[eight]    \n\t"
>+
>+      "punpckhhw  %[src0],         %[b],                 %[r]        \n\t"
>+      "punpckhhw  %[src1],         %[g],                 %[value]    \n\t"
>+      "punpcklhw  %[src_lo],       %[src0],              %[src1]     \n\t"
>+      "punpckhhw  %[src_hi],       %[src0],              %[src1]     \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],            %[mask]     \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],            %[mask]     \n\t"
>+      "punpcklwd  %[src0],         %[src_lo],            %[src_hi]   \n\t"
>+      "punpckhwd  %[src1],         %[src_lo],            %[src_hi]   \n\t"
>+      "paddw      %[dest1],        %[src0],              %[src1]     \n\t"
>+      "psrlw      %[dest1],        %[dest1],             %[eight]    \n\t"
>+
>+      "gsldrc1    %[src0],         0x08(%[src_argb1555])             \n\t"
>+      "gsldlc1    %[src0],         0x0f(%[src_argb1555])             \n\t"
>+      "psrlh      %[src1],         %[src0],              %[eight]    \n\t"
>+      "and        %[b],            %[src0],              %[c0]       \n\t"
>+      "and        %[src0],         %[src0],              %[c1]       \n\t"
>+      "psrlh      %[src0],         %[src0],              %[five]     \n\t"
>+      "and        %[g],            %[src1],              %[c2]       \n\t"
>+      "psllh      %[g],            %[g],                 %[three]    \n\t"
>+      "or         %[g],            %[src0],              %[g]        \n\t"
>+      "and        %[r],            %[src1],              %[c3]       \n\t"
>+      "psrlh      %[r],            %[r],                 %[two]      \n\t"
>+      "psllh      %[src0],         %[b],                 %[three]    \n\t"
>+      "psrlh      %[src1],         %[b],                 %[two]      \n\t"
>+      "or         %[b],            %[src0],              %[src1]     \n\t"
>+      "psllh      %[src0],         %[g],                 %[three]    \n\t"
>+      "psrlh      %[src1],         %[g],                 %[two]      \n\t"
>+      "or         %[g],            %[src0],              %[src1]     \n\t"
>+      "psllh      %[src0],         %[r],                 %[three]    \n\t"
>+      "psrlh      %[src1],         %[r],                 %[two]      \n\t"
>+      "or         %[r],            %[src0],              %[src1]     \n\t"
>+      "punpcklhw  %[src0],         %[b],                 %[r]        \n\t"
>+      "punpcklhw  %[src1],         %[g],                 %[value]    \n\t"
>+      "punpcklhw  %[src_lo],       %[src0],              %[src1]     \n\t"
>+      "punpckhhw  %[src_hi],       %[src0],              %[src1]     \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],            %[mask]     \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],            %[mask]     \n\t"
>+      "punpcklwd  %[src0],         %[src_lo],            %[src_hi]   \n\t"
>+      "punpckhwd  %[src1],         %[src_lo],            %[src_hi]   \n\t"
>+      "paddw      %[dest2],        %[src0],              %[src1]     \n\t"
>+      "psrlw      %[dest2],        %[dest2],             %[eight]    \n\t"
>+
>+      "punpckhhw  %[src0],         %[b],                 %[r]        \n\t"
>+      "punpckhhw  %[src1],         %[g],                 %[value]    \n\t"
>+      "punpcklhw  %[src_lo],       %[src0],              %[src1]     \n\t"
>+      "punpckhhw  %[src_hi],       %[src0],              %[src1]     \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],            %[mask]     \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],            %[mask]     \n\t"
>+      "punpcklwd  %[src0],         %[src_lo],            %[src_hi]   \n\t"
>+      "punpckhwd  %[src1],         %[src_lo],            %[src_hi]   \n\t"
>+      "paddw      %[dest3],        %[src0],              %[src1]     \n\t"
>+      "psrlw      %[dest3],        %[dest3],             %[eight]    \n\t"
>+
>+      "packsswh   %[src_lo],       %[dest0],             %[dest1]    \n\t"
>+      "packsswh   %[src_hi],       %[dest2],             %[dest3]    \n\t"
>+      "packushb   %[dest0],        %[src_lo],            %[src_hi]   \n\t"
>+      "gssdlc1    %[dest0],        0x07(%[dst_y])                    \n\t"
>+      "gssdrc1    %[dest0],        0x00(%[dst_y])                    \n\t"
>+
>+      "daddiu     %[src_argb1555], %[src_argb1555],      0x10        \n\t"
>+      "daddiu     %[dst_y],        %[dst_y],             0x08        \n\t"
>+      "daddiu     %[width],        %[width],            -0x08        \n\t"
>+      "bgtz       %[width],        1b                                \n\t"
>+      : [src0] "=&f"(ftmp[0]), [src1] "=&f"(ftmp[1]), [src_lo] "=&f"(ftmp[2]),
>+        [src_hi] "=&f"(ftmp[3]), [b] "=&f"(ftmp[4]), [g] "=&f"(ftmp[5]),
>+        [r] "=&f"(ftmp[6]), [dest0] "=&f"(ftmp[7]), [dest1] "=&f"(ftmp[8]),
>+        [dest2] "=&f"(ftmp[9]), [dest3] "=&f"(ftmp[10])
>+      : [src_argb1555] "r"(src_argb1555), [dst_y] "r"(dst_y),
>+        [width] "r"(width), [value] "f"(value), [mask] "f"(mask), [c0] "f"(c0),
>+        [c1] "f"(c1), [c2] "f"(c2), [c3] "f"(c3), [eight] "f"(0x08),
>+        [five] "f"(0x05), [three] "f"(0x03), [two] "f"(0x02), [seven] "f"(0x07)
>+      : "memory");
>+}
>+
>+void ARGB4444ToYRow_MMI(const uint8_t* src_argb4444,
>+                        uint8_t* dst_y,
>+                        int width) {
>+  uint64_t ftmp[11];
>+  uint64_t value = 0x1080108010801080;
>+  uint64_t mask = 0x0001004200810019;
>+  uint64_t c0 = 0x000f000f000f000f;
>+  uint64_t c1 = 0x00ff00ff00ff00ff;
>+  __asm__ volatile(
>+      "1:                                                            \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src_argb4444])             \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src_argb4444])             \n\t"
>+      "psrlh      %[src1],         %[src0],              %[eight]    \n\t"
>+      "and        %[b],            %[src0],              %[c0]       \n\t"
>+      "and        %[src0],         %[src0],              %[c1]       \n\t"
>+      "psrlh      %[g],            %[src0],              %[four]     \n\t"
>+      "and        %[r],            %[src1],              %[c0]       \n\t"
>+      "psllh      %[src0],         %[b],                 %[four]     \n\t"
>+      "or         %[b],            %[src0],              %[b]        \n\t"
>+      "psllh      %[src0],         %[g],                 %[four]     \n\t"
>+      "or         %[g],            %[src0],              %[g]        \n\t"
>+      "psllh      %[src0],         %[r],                 %[four]     \n\t"
>+      "or         %[r],            %[src0],              %[r]        \n\t"
>+      "punpcklhw  %[src0],         %[b],                 %[r]        \n\t"
>+      "punpcklhw  %[src1],         %[g],                 %[value]    \n\t"
>+      "punpcklhw  %[src_lo],       %[src0],              %[src1]     \n\t"
>+      "punpckhhw  %[src_hi],       %[src0],              %[src1]     \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],            %[mask]     \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],            %[mask]     \n\t"
>+      "punpcklwd  %[src0],         %[src_lo],            %[src_hi]   \n\t"
>+      "punpckhwd  %[src1],         %[src_lo],            %[src_hi]   \n\t"
>+      "paddw      %[dest0],        %[src0],              %[src1]     \n\t"
>+      "psrlw      %[dest0],        %[dest0],             %[eight]    \n\t"
>+
>+      "punpckhhw  %[src0],         %[b],                 %[r]        \n\t"
>+      "punpckhhw  %[src1],         %[g],                 %[value]    \n\t"
>+      "punpcklhw  %[src_lo],       %[src0],              %[src1]     \n\t"
>+      "punpckhhw  %[src_hi],       %[src0],              %[src1]     \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],            %[mask]     \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],            %[mask]     \n\t"
>+      "punpcklwd  %[src0],         %[src_lo],            %[src_hi]   \n\t"
>+      "punpckhwd  %[src1],         %[src_lo],            %[src_hi]   \n\t"
>+      "paddw      %[dest1],        %[src0],              %[src1]     \n\t"
>+      "psrlw      %[dest1],        %[dest1],             %[eight]    \n\t"
>+
>+      "gsldrc1    %[src0],         0x08(%[src_argb4444])             \n\t"
>+      "gsldlc1    %[src0],         0x0f(%[src_argb4444])             \n\t"
>+      "psrlh      %[src1],         %[src0],              %[eight]    \n\t"
>+      "and        %[b],            %[src0],              %[c0]       \n\t"
>+      "and        %[src0],         %[src0],              %[c1]       \n\t"
>+      "psrlh      %[g],            %[src0],              %[four]     \n\t"
>+      "and        %[r],            %[src1],              %[c0]       \n\t"
>+      "psllh      %[src0],         %[b],                 %[four]     \n\t"
>+      "or         %[b],            %[src0],              %[b]        \n\t"
>+      "psllh      %[src0],         %[g],                 %[four]     \n\t"
>+      "or         %[g],            %[src0],              %[g]        \n\t"
>+      "psllh      %[src0],         %[r],                 %[four]     \n\t"
>+      "or         %[r],            %[src0],              %[r]        \n\t"
>+      "punpcklhw  %[src0],         %[b],                 %[r]        \n\t"
>+      "punpcklhw  %[src1],         %[g],                 %[value]    \n\t"
>+      "punpcklhw  %[src_lo],       %[src0],              %[src1]     \n\t"
>+      "punpckhhw  %[src_hi],       %[src0],              %[src1]     \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],            %[mask]     \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],            %[mask]     \n\t"
>+      "punpcklwd  %[src0],         %[src_lo],            %[src_hi]   \n\t"
>+      "punpckhwd  %[src1],         %[src_lo],            %[src_hi]   \n\t"
>+      "paddw      %[dest2],        %[src0],              %[src1]     \n\t"
>+      "psrlw      %[dest2],        %[dest2],             %[eight]    \n\t"
>+
>+      "punpckhhw  %[src0],         %[b],                 %[r]        \n\t"
>+      "punpckhhw  %[src1],         %[g],                 %[value]    \n\t"
>+      "punpcklhw  %[src_lo],       %[src0],              %[src1]     \n\t"
>+      "punpckhhw  %[src_hi],       %[src0],              %[src1]     \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],            %[mask]     \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],            %[mask]     \n\t"
>+      "punpcklwd  %[src0],         %[src_lo],            %[src_hi]   \n\t"
>+      "punpckhwd  %[src1],         %[src_lo],            %[src_hi]   \n\t"
>+      "paddw      %[dest3],        %[src0],              %[src1]     \n\t"
>+      "psrlw      %[dest3],        %[dest3],             %[eight]    \n\t"
>+
>+      "packsswh   %[src_lo],       %[dest0],             %[dest1]    \n\t"
>+      "packsswh   %[src_hi],       %[dest2],             %[dest3]    \n\t"
>+      "packushb   %[dest0],        %[src_lo],            %[src_hi]   \n\t"
>+      "gssdlc1    %[dest0],        0x07(%[dst_y])                    \n\t"
>+      "gssdrc1    %[dest0],        0x00(%[dst_y])                    \n\t"
>+
>+      "daddiu     %[src_argb4444], %[src_argb4444],      0x10        \n\t"
>+      "daddiu     %[dst_y],        %[dst_y],             0x08        \n\t"
>+      "daddiu     %[width],        %[width],            -0x08        \n\t"
>+      "bgtz       %[width],        1b                                \n\t"
>+      : [src0] "=&f"(ftmp[0]), [src1] "=&f"(ftmp[1]), [src_lo] "=&f"(ftmp[2]),
>+        [src_hi] "=&f"(ftmp[3]), [b] "=&f"(ftmp[4]), [g] "=&f"(ftmp[5]),
>+        [r] "=&f"(ftmp[6]), [dest0] "=&f"(ftmp[7]), [dest1] "=&f"(ftmp[8]),
>+        [dest2] "=&f"(ftmp[9]), [dest3] "=&f"(ftmp[10])
>+      : [src_argb4444] "r"(src_argb4444), [dst_y] "r"(dst_y),
>+        [width] "r"(width), [value] "f"(value), [mask] "f"(mask), [c0] "f"(c0),
>+        [c1] "f"(c1), [eight] "f"(0x08), [four] "f"(0x04)
>+      : "memory");
>+}
>+
>+void RGB565ToUVRow_MMI(const uint8_t* src_rgb565,
>+                       int src_stride_rgb565,
>+                       uint8_t* dst_u,
>+                       uint8_t* dst_v,
>+                       int width) {
>+  uint64_t ftmp[13];
>+  uint64_t value = 0x2020202020202020;
>+  uint64_t mask_u = 0x0026004a00700002;
>+  uint64_t mask_v = 0x00020070005e0012;
>+  uint64_t mask = 0x93;
>+  uint64_t c0 = 0x001f001f001f001f;
>+  uint64_t c1 = 0x00ff00ff00ff00ff;
>+  uint64_t c2 = 0x0007000700070007;
>+  __asm__ volatile(
>+      "daddu      %[next_rgb565], %[src_rgb565],       %[next_rgb565]   \n\t"
>+      "1:                                                               \n\t"
>+      "gsldrc1    %[src0],        0x00(%[src_rgb565])                   \n\t"
>+      "gsldlc1    %[src0],        0x07(%[src_rgb565])                   \n\t"
>+      "gsldrc1    %[src1],        0x00(%[next_rgb565])                  \n\t"
>+      "gsldlc1    %[src1],        0x07(%[next_rgb565])                  \n\t"
>+      "psrlh      %[dest0_u],     %[src0],             %[eight]         \n\t"
>+      "and        %[b0],          %[src0],             %[c0]            \n\t"
>+      "and        %[src0],        %[src0],             %[c1]            \n\t"
>+      "psrlh      %[src0],        %[src0],             %[five]          \n\t"
>+      "and        %[g0],          %[dest0_u],          %[c2]            \n\t"
>+      "psllh      %[g0],          %[g0],               %[three]         \n\t"
>+      "or         %[g0],          %[src0],             %[g0]            \n\t"
>+      "psrlh      %[r0],          %[dest0_u],          %[three]         \n\t"
>+      "psrlh      %[src0],        %[src1],             %[eight]         \n\t"
>+      "and        %[dest0_u],     %[src1],             %[c0]            \n\t"
>+      "and        %[src1],        %[src1],             %[c1]            \n\t"
>+      "psrlh      %[src1],        %[src1],             %[five]          \n\t"
>+      "and        %[dest0_v],     %[src0],             %[c2]            \n\t"
>+      "psllh      %[dest0_v],     %[dest0_v],          %[three]         \n\t"
>+      "or         %[dest0_v],     %[src1],             %[dest0_v]       \n\t"
>+      "psrlh      %[src0],        %[src0],             %[three]         \n\t"
>+      "paddh      %[b0],          %[b0],               %[dest0_u]       \n\t"
>+      "paddh      %[g0],          %[g0],               %[dest0_v]       \n\t"
>+      "paddh      %[r0],          %[r0],               %[src0]          \n\t"
>+      "punpcklhw  %[src0],        %[b0],               %[r0]            \n\t"
>+      "punpckhhw  %[src1],        %[b0],               %[r0]            \n\t"
>+      "punpcklwd  %[dest0_u],     %[src0],             %[src1]          \n\t"
>+      "punpckhwd  %[dest0_v],     %[src0],             %[src1]          \n\t"
>+      "paddh      %[src0],        %[dest0_u],          %[dest0_v]       \n\t"
>+      "psrlh      %[b0],          %[src0],             %[six]           \n\t"
>+      "psllh      %[r0],          %[src0],             %[one]           \n\t"
>+      "or         %[b0],          %[b0],               %[r0]            \n\t"
>+      "punpcklhw  %[src0],        %[g0],               %[value]         \n\t"
>+      "punpckhhw  %[src1],        %[g0],               %[value]         \n\t"
>+      "punpcklwd  %[dest0_u],     %[src0],             %[src1]          \n\t"
>+      "punpckhwd  %[dest0_v],     %[src0],             %[src1]          \n\t"
>+      "paddh      %[g0],          %[dest0_u],          %[dest0_v]       \n\t"
>+      "punpcklhw  %[src0],        %[b0],               %[g0]            \n\t"
>+      "punpckhhw  %[src1],        %[b0],               %[g0]            \n\t"
>+
>+      "pmaddhw    %[dest0_v],     %[src0],             %[mask_v]        \n\t"
>+      "pshufh     %[dest0_u],     %[src0],             %[mask]          \n\t"
>+      "pmaddhw    %[dest0_u],     %[dest0_u],          %[mask_u]        \n\t"
>+      "pmaddhw    %[g0],          %[src1],             %[mask_v]        \n\t"
>+      "pshufh     %[b0],          %[src1],             %[mask]          \n\t"
>+      "pmaddhw    %[b0],          %[b0],               %[mask_u]        \n\t"
>+
>+      "punpcklwd  %[src0],        %[dest0_u],          %[b0]            \n\t"
>+      "punpckhwd  %[src1],        %[dest0_u],          %[b0]            \n\t"
>+      "psubw      %[dest0_u],     %[src0],             %[src1]          \n\t"
>+      "psraw      %[dest0_u],     %[dest0_u],          %[eight]         \n\t"
>+      "punpcklwd  %[src0],        %[dest0_v],          %[g0]            \n\t"
>+      "punpckhwd  %[src1],        %[dest0_v],          %[g0]            \n\t"
>+      "psubw      %[dest0_v],     %[src1],             %[src0]          \n\t"
>+      "psraw      %[dest0_v],     %[dest0_v],          %[eight]         \n\t"
>+
>+      "gsldrc1    %[src0],        0x08(%[src_rgb565])                   \n\t"
>+      "gsldlc1    %[src0],        0x0f(%[src_rgb565])                   \n\t"
>+      "gsldrc1    %[src1],        0x08(%[next_rgb565])                  \n\t"
>+      "gsldlc1    %[src1],        0x0f(%[next_rgb565])                  \n\t"
>+      "psrlh      %[dest1_u],     %[src0],             %[eight]         \n\t"
>+      "and        %[b0],          %[src0],             %[c0]            \n\t"
>+      "and        %[src0],        %[src0],             %[c1]            \n\t"
>+      "psrlh      %[src0],        %[src0],             %[five]          \n\t"
>+      "and        %[g0],          %[dest1_u],          %[c2]            \n\t"
>+      "psllh      %[g0],          %[g0],               %[three]         \n\t"
>+      "or         %[g0],          %[src0],             %[g0]            \n\t"
>+      "psrlh      %[r0],          %[dest1_u],          %[three]         \n\t"
>+      "psrlh      %[src0],        %[src1],             %[eight]         \n\t"
>+      "and        %[dest1_u],     %[src1],             %[c0]            \n\t"
>+      "and        %[src1],        %[src1],             %[c1]            \n\t"
>+      "psrlh      %[src1],        %[src1],             %[five]          \n\t"
>+      "and        %[dest1_v],     %[src0],             %[c2]            \n\t"
>+      "psllh      %[dest1_v],     %[dest1_v],          %[three]         \n\t"
>+      "or         %[dest1_v],     %[src1],             %[dest1_v]       \n\t"
>+      "psrlh      %[src0],        %[src0],             %[three]         \n\t"
>+      "paddh      %[b0],          %[b0],               %[dest1_u]       \n\t"
>+      "paddh      %[g0],          %[g0],               %[dest1_v]       \n\t"
>+      "paddh      %[r0],          %[r0],               %[src0]          \n\t"
>+      "punpcklhw  %[src0],        %[b0],               %[r0]            \n\t"
>+      "punpckhhw  %[src1],        %[b0],               %[r0]            \n\t"
>+      "punpcklwd  %[dest1_u],     %[src0],             %[src1]          \n\t"
>+      "punpckhwd  %[dest1_v],     %[src0],             %[src1]          \n\t"
>+      "paddh      %[src0],        %[dest1_u],          %[dest1_v]       \n\t"
>+      "psrlh      %[b0],          %[src0],             %[six]           \n\t"
>+      "psllh      %[r0],          %[src0],             %[one]           \n\t"
>+      "or         %[b0],          %[b0],               %[r0]            \n\t"
>+      "punpcklhw  %[src0],        %[g0],               %[value]         \n\t"
>+      "punpckhhw  %[src1],        %[g0],               %[value]         \n\t"
>+      "punpcklwd  %[dest1_u],     %[src0],             %[src1]          \n\t"
>+      "punpckhwd  %[dest1_v],     %[src0],             %[src1]          \n\t"
>+      "paddh      %[g0],          %[dest1_u],          %[dest1_v]       \n\t"
>+      "punpcklhw  %[src0],        %[b0],               %[g0]            \n\t"
>+      "punpckhhw  %[src1],        %[b0],               %[g0]            \n\t"
>+
>+      "pmaddhw    %[dest1_v],     %[src0],             %[mask_v]        \n\t"
>+      "pshufh     %[dest1_u],     %[src0],             %[mask]          \n\t"
>+      "pmaddhw    %[dest1_u],     %[dest1_u],          %[mask_u]        \n\t"
>+      "pmaddhw    %[g0],          %[src1],             %[mask_v]        \n\t"
>+      "pshufh     %[b0],          %[src1],             %[mask]          \n\t"
>+      "pmaddhw    %[b0],          %[b0],               %[mask_u]        \n\t"
>+
>+      "punpcklwd  %[src0],        %[dest1_u],          %[b0]            \n\t"
>+      "punpckhwd  %[src1],        %[dest1_u],          %[b0]            \n\t"
>+      "psubw      %[dest1_u],     %[src0],             %[src1]          \n\t"
>+      "psraw      %[dest1_u],     %[dest1_u],          %[eight]         \n\t"
>+      "punpcklwd  %[src0],        %[dest1_v],          %[g0]            \n\t"
>+      "punpckhwd  %[src1],        %[dest1_v],          %[g0]            \n\t"
>+      "psubw      %[dest1_v],     %[src1],             %[src0]          \n\t"
>+      "psraw      %[dest1_v],     %[dest1_v],          %[eight]         \n\t"
>+
>+      "gsldrc1    %[src0],        0x10(%[src_rgb565])                   \n\t"
>+      "gsldlc1    %[src0],        0x17(%[src_rgb565])                   \n\t"
>+      "gsldrc1    %[src1],        0x10(%[next_rgb565])                  \n\t"
>+      "gsldlc1    %[src1],        0x17(%[next_rgb565])                  \n\t"
>+      "psrlh      %[dest2_u],     %[src0],             %[eight]         \n\t"
>+      "and        %[b0],          %[src0],             %[c0]            \n\t"
>+      "and        %[src0],        %[src0],             %[c1]            \n\t"
>+      "psrlh      %[src0],        %[src0],             %[five]          \n\t"
>+      "and        %[g0],          %[dest2_u],          %[c2]            \n\t"
>+      "psllh      %[g0],          %[g0],               %[three]         \n\t"
>+      "or         %[g0],          %[src0],             %[g0]            \n\t"
>+      "psrlh      %[r0],          %[dest2_u],          %[three]         \n\t"
>+      "psrlh      %[src0],        %[src1],             %[eight]         \n\t"
>+      "and        %[dest2_u],     %[src1],             %[c0]            \n\t"
>+      "and        %[src1],        %[src1],             %[c1]            \n\t"
>+      "psrlh      %[src1],        %[src1],             %[five]          \n\t"
>+      "and        %[dest2_v],     %[src0],             %[c2]            \n\t"
>+      "psllh      %[dest2_v],     %[dest2_v],          %[three]         \n\t"
>+      "or         %[dest2_v],     %[src1],             %[dest2_v]       \n\t"
>+      "psrlh      %[src0],        %[src0],             %[three]         \n\t"
>+      "paddh      %[b0],          %[b0],               %[dest2_u]       \n\t"
>+      "paddh      %[g0],          %[g0],               %[dest2_v]       \n\t"
>+      "paddh      %[r0],          %[r0],               %[src0]          \n\t"
>+      "punpcklhw  %[src0],        %[b0],               %[r0]            \n\t"
>+      "punpckhhw  %[src1],        %[b0],               %[r0]            \n\t"
>+      "punpcklwd  %[dest2_u],     %[src0],             %[src1]          \n\t"
>+      "punpckhwd  %[dest2_v],     %[src0],             %[src1]          \n\t"
>+      "paddh      %[src0],        %[dest2_u],          %[dest2_v]       \n\t"
>+      "psrlh      %[b0],          %[src0],             %[six]           \n\t"
>+      "psllh      %[r0],          %[src0],             %[one]           \n\t"
>+      "or         %[b0],          %[b0],               %[r0]            \n\t"
>+      "punpcklhw  %[src0],        %[g0],               %[value]         \n\t"
>+      "punpckhhw  %[src1],        %[g0],               %[value]         \n\t"
>+      "punpcklwd  %[dest2_u],     %[src0],             %[src1]          \n\t"
>+      "punpckhwd  %[dest2_v],     %[src0],             %[src1]          \n\t"
>+      "paddh      %[g0],          %[dest2_u],          %[dest2_v]       \n\t"
>+      "punpcklhw  %[src0],        %[b0],               %[g0]            \n\t"
>+      "punpckhhw  %[src1],        %[b0],               %[g0]            \n\t"
>+
>+      "pmaddhw    %[dest2_v],     %[src0],             %[mask_v]        \n\t"
>+      "pshufh     %[dest2_u],     %[src0],             %[mask]          \n\t"
>+      "pmaddhw    %[dest2_u],     %[dest2_u],          %[mask_u]        \n\t"
>+      "pmaddhw    %[g0],          %[src1],             %[mask_v]        \n\t"
>+      "pshufh     %[b0],          %[src1],             %[mask]          \n\t"
>+      "pmaddhw    %[b0],          %[b0],               %[mask_u]        \n\t"
>+
>+      "punpcklwd  %[src0],        %[dest2_u],          %[b0]            \n\t"
>+      "punpckhwd  %[src1],        %[dest2_u],          %[b0]            \n\t"
>+      "psubw      %[dest2_u],     %[src0],             %[src1]          \n\t"
>+      "psraw      %[dest2_u],     %[dest2_u],          %[eight]         \n\t"
>+      "punpcklwd  %[src0],        %[dest2_v],          %[g0]            \n\t"
>+      "punpckhwd  %[src1],        %[dest2_v],          %[g0]            \n\t"
>+      "psubw      %[dest2_v],     %[src1],             %[src0]          \n\t"
>+      "psraw      %[dest2_v],     %[dest2_v],          %[eight]         \n\t"
>+
>+      "gsldrc1    %[src0],        0x18(%[src_rgb565])                   \n\t"
>+      "gsldlc1    %[src0],        0x1f(%[src_rgb565])                   \n\t"
>+      "gsldrc1    %[src1],        0x18(%[next_rgb565])                  \n\t"
>+      "gsldlc1    %[src1],        0x1f(%[next_rgb565])                  \n\t"
>+      "psrlh      %[dest3_u],     %[src0],             %[eight]         \n\t"
>+      "and        %[b0],          %[src0],             %[c0]            \n\t"
>+      "and        %[src0],        %[src0],             %[c1]            \n\t"
>+      "psrlh      %[src0],        %[src0],             %[five]          \n\t"
>+      "and        %[g0],          %[dest3_u],          %[c2]            \n\t"
>+      "psllh      %[g0],          %[g0],               %[three]         \n\t"
>+      "or         %[g0],          %[src0],             %[g0]            \n\t"
>+      "psrlh      %[r0],          %[dest3_u],          %[three]         \n\t"
>+      "psrlh      %[src0],        %[src1],             %[eight]         \n\t"
>+      "and        %[dest3_u],     %[src1],             %[c0]            \n\t"
>+      "and        %[src1],        %[src1],             %[c1]            \n\t"
>+      "psrlh      %[src1],        %[src1],             %[five]          \n\t"
>+      "and        %[dest3_v],     %[src0],             %[c2]            \n\t"
>+      "psllh      %[dest3_v],     %[dest3_v],          %[three]         \n\t"
>+      "or         %[dest3_v],     %[src1],             %[dest3_v]       \n\t"
>+      "psrlh      %[src0],        %[src0],             %[three]         \n\t"
>+      "paddh      %[b0],          %[b0],               %[dest3_u]       \n\t"
>+      "paddh      %[g0],          %[g0],               %[dest3_v]       \n\t"
>+      "paddh      %[r0],          %[r0],               %[src0]          \n\t"
>+      "punpcklhw  %[src0],        %[b0],               %[r0]            \n\t"
>+      "punpckhhw  %[src1],        %[b0],               %[r0]            \n\t"
>+      "punpcklwd  %[dest3_u],     %[src0],             %[src1]          \n\t"
>+      "punpckhwd  %[dest3_v],     %[src0],             %[src1]          \n\t"
>+      "paddh      %[src0],        %[dest3_u],          %[dest3_v]       \n\t"
>+      "psrlh      %[b0],          %[src0],             %[six]           \n\t"
>+      "psllh      %[r0],          %[src0],             %[one]           \n\t"
>+      "or         %[b0],          %[b0],               %[r0]            \n\t"
>+      "punpcklhw  %[src0],        %[g0],               %[value]         \n\t"
>+      "punpckhhw  %[src1],        %[g0],               %[value]         \n\t"
>+      "punpcklwd  %[dest3_u],     %[src0],             %[src1]          \n\t"
>+      "punpckhwd  %[dest3_v],     %[src0],             %[src1]          \n\t"
>+      "paddh      %[g0],          %[dest3_u],          %[dest3_v]       \n\t"
>+      "punpcklhw  %[src0],        %[b0],               %[g0]            \n\t"
>+      "punpckhhw  %[src1],        %[b0],               %[g0]            \n\t"
>+
>+      "pmaddhw    %[dest3_v],     %[src0],             %[mask_v]        \n\t"
>+      "pshufh     %[dest3_u],     %[src0],             %[mask]          \n\t"
>+      "pmaddhw    %[dest3_u],     %[dest3_u],          %[mask_u]        \n\t"
>+      "pmaddhw    %[g0],          %[src1],             %[mask_v]        \n\t"
>+      "pshufh     %[b0],          %[src1],             %[mask]          \n\t"
>+      "pmaddhw    %[b0],          %[b0],               %[mask_u]        \n\t"
>+
>+      "punpcklwd  %[src0],        %[dest3_u],          %[b0]            \n\t"
>+      "punpckhwd  %[src1],        %[dest3_u],          %[b0]            \n\t"
>+      "psubw      %[dest3_u],     %[src0],             %[src1]          \n\t"
>+      "psraw      %[dest3_u],     %[dest3_u],          %[eight]         \n\t"
>+      "punpcklwd  %[src0],        %[dest3_v],          %[g0]            \n\t"
>+      "punpckhwd  %[src1],        %[dest3_v],          %[g0]            \n\t"
>+      "psubw      %[dest3_v],     %[src1],             %[src0]          \n\t"
>+      "psraw      %[dest3_v],     %[dest3_v],          %[eight]         \n\t"
>+
>+      "packsswh   %[src0],        %[dest0_u],          %[dest1_u]       \n\t"
>+      "packsswh   %[src1],        %[dest2_u],          %[dest3_u]       \n\t"
>+      "packushb   %[dest0_u],     %[src0],             %[src1]          \n\t"
>+      "gssdlc1    %[dest0_u],     0x07(%[dst_u])                        \n\t"
>+      "gssdrc1    %[dest0_u],     0x00(%[dst_u])                        \n\t"
>+      "packsswh   %[src0],        %[dest0_v],          %[dest1_v]       \n\t"
>+      "packsswh   %[src1],        %[dest2_v],          %[dest3_v]       \n\t"
>+      "packushb   %[dest0_v],     %[src0],             %[src1]          \n\t"
>+      "gssdlc1    %[dest0_v],     0x07(%[dst_v])                        \n\t"
>+      "gssdrc1    %[dest0_v],     0x00(%[dst_v])                        \n\t"
>+
>+      "daddiu    %[src_rgb565],   %[src_rgb565],       0x20             \n\t"
>+      "daddiu    %[next_rgb565],  %[next_rgb565],      0x20             \n\t"
>+      "daddiu    %[dst_u],        %[dst_u],            0x08             \n\t"
>+      "daddiu    %[dst_v],        %[dst_v],            0x08             \n\t"
>+      "daddiu    %[width],        %[width],           -0x10             \n\t"
>+      "bgtz      %[width],        1b                                    \n\t"
>+      : [src0] "=&f"(ftmp[0]), [src1] "=&f"(ftmp[1]), [b0] "=&f"(ftmp[2]),
>+        [g0] "=&f"(ftmp[3]), [r0] "=&f"(ftmp[4]), [dest0_u] "=&f"(ftmp[5]),
>+        [dest1_u] "=&f"(ftmp[6]), [dest2_u] "=&f"(ftmp[7]),
>+        [dest3_u] "=&f"(ftmp[8]), [dest0_v] "=&f"(ftmp[9]),
>+        [dest1_v] "=&f"(ftmp[10]), [dest2_v] "=&f"(ftmp[11]),
>+        [dest3_v] "=&f"(ftmp[12])
>+      : [src_rgb565] "r"(src_rgb565), [next_rgb565] "r"(src_stride_rgb565),
>+        [dst_u] "r"(dst_u), [dst_v] "r"(dst_v), [width] "r"(width),
>+        [value] "f"(value), [c0] "f"(c0), [c1] "f"(c1), [c2] "f"(c2),
>+        [mask] "f"(mask), [mask_u] "f"(mask_u), [mask_v] "f"(mask_v),
>+        [eight] "f"(0x08), [six] "f"(0x06), [five] "f"(0x05), [three] "f"(0x03),
>+        [one] "f"(0x01)
>+      : "memory");
>+}
>+
>+void ARGB1555ToUVRow_MMI(const uint8_t* src_argb1555,
>+                         int src_stride_argb1555,
>+                         uint8_t* dst_u,
>+                         uint8_t* dst_v,
>+                         int width) {
>+  uint64_t ftmp[11];
>+  uint64_t value = 0x2020202020202020;
>+  uint64_t mask_u = 0x0026004a00700002;
>+  uint64_t mask_v = 0x00020070005e0012;
>+  uint64_t mask = 0x93;
>+  uint64_t c0 = 0x001f001f001f001f;
>+  uint64_t c1 = 0x00ff00ff00ff00ff;
>+  uint64_t c2 = 0x0003000300030003;
>+  uint64_t c3 = 0x007c007c007c007c;
>+  __asm__ volatile(
>+      "daddu      %[next_argb1555], %[src_argb1555],      %[next_argb1555] \n\t"
>+      "1:                                                                  \n\t"
>+      "gsldrc1    %[src0],          0x00(%[src_argb1555])                  \n\t"
>+      "gsldlc1    %[src0],          0x07(%[src_argb1555])                  \n\t"
>+      "gsldrc1    %[src1],          0x00(%[next_argb1555])                 \n\t"
>+      "gsldlc1    %[src1],          0x07(%[next_argb1555])                 \n\t"
>+      "psrlh      %[dest0_u],       %[src0],               %[eight]        \n\t"
>+      "and        %[b0],            %[src0],               %[c0]           \n\t"
>+      "and        %[src0],          %[src0],               %[c1]           \n\t"
>+      "psrlh      %[src0],          %[src0],               %[five]         \n\t"
>+      "and        %[g0],            %[dest0_u],            %[c2]           \n\t"
>+      "psllh      %[g0],            %[g0],                 %[three]        \n\t"
>+      "or         %[g0],            %[src0],               %[g0]           \n\t"
>+      "and        %[r0],            %[dest0_u],            %[c3]           \n\t"
>+      "psrlh      %[r0],            %[r0],                 %[two]          \n\t"
>+      "psrlh      %[src0],          %[src1],               %[eight]        \n\t"
>+      "and        %[dest0_u],       %[src1],               %[c0]           \n\t"
>+      "and        %[src1],          %[src1],               %[c1]           \n\t"
>+      "psrlh      %[src1],          %[src1],               %[five]         \n\t"
>+      "and        %[dest0_v],       %[src0],               %[c2]           \n\t"
>+      "psllh      %[dest0_v],       %[dest0_v],            %[three]        \n\t"
>+      "or         %[dest0_v],       %[src1],               %[dest0_v]      \n\t"
>+      "and        %[src0],          %[src0],               %[c3]           \n\t"
>+      "psrlh      %[src0],          %[src0],               %[two]          \n\t"
>+      "paddh      %[b0],            %[b0],                 %[dest0_u]      \n\t"
>+      "paddh      %[g0],            %[g0],                 %[dest0_v]      \n\t"
>+      "paddh      %[r0],            %[r0],                 %[src0]         \n\t"
>+      "punpcklhw  %[src0],          %[b0],                 %[r0]           \n\t"
>+      "punpckhhw  %[src1],          %[b0],                 %[r0]           \n\t"
>+      "punpcklwd  %[dest0_u],       %[src0],               %[src1]         \n\t"
>+      "punpckhwd  %[dest0_v],       %[src0],               %[src1]         \n\t"
>+      "paddh      %[src0],          %[dest0_u],            %[dest0_v]      \n\t"
>+      "psrlh      %[b0],            %[src0],               %[six]          \n\t"
>+      "psllh      %[r0],            %[src0],               %[one]          \n\t"
>+      "or         %[b0],            %[b0],                 %[r0]           \n\t"
>+      "psrlh      %[r0],            %[g0],                 %[six]          \n\t"
>+      "psllh      %[g0],            %[g0],                 %[one]          \n\t"
>+      "or         %[g0],            %[g0],                 %[r0]           \n\t"
>+      "punpcklhw  %[src0],          %[g0],                 %[value]        \n\t"
>+      "punpckhhw  %[src1],          %[g0],                 %[value]        \n\t"
>+      "punpcklwd  %[dest0_u],       %[src0],               %[src1]         \n\t"
>+      "punpckhwd  %[dest0_v],       %[src0],               %[src1]         \n\t"
>+      "paddh      %[g0],            %[dest0_u],            %[dest0_v]      \n\t"
>+      "punpcklhw  %[src0],          %[b0],                 %[g0]           \n\t"
>+      "punpckhhw  %[src1],          %[b0],                 %[g0]           \n\t"
>+
>+      "pmaddhw    %[dest0_v],       %[src0],               %[mask_v]       \n\t"
>+      "pshufh     %[dest0_u],       %[src0],               %[mask]         \n\t"
>+      "pmaddhw    %[dest0_u],       %[dest0_u],            %[mask_u]       \n\t"
>+      "pmaddhw    %[g0],            %[src1],               %[mask_v]       \n\t"
>+      "pshufh     %[b0],            %[src1],               %[mask]         \n\t"
>+      "pmaddhw    %[b0],            %[b0],                 %[mask_u]       \n\t"
>+
>+      "punpcklwd  %[src0],          %[dest0_u],            %[b0]           \n\t"
>+      "punpckhwd  %[src1],          %[dest0_u],            %[b0]           \n\t"
>+      "psubw      %[dest0_u],       %[src0],               %[src1]         \n\t"
>+      "psraw      %[dest0_u],       %[dest0_u],            %[eight]        \n\t"
>+      "punpcklwd  %[src0],          %[dest0_v],            %[g0]           \n\t"
>+      "punpckhwd  %[src1],          %[dest0_v],            %[g0]           \n\t"
>+      "psubw      %[dest0_v],       %[src1],               %[src0]         \n\t"
>+      "psraw      %[dest0_v],       %[dest0_v],            %[eight]        \n\t"
>+
>+      "gsldrc1    %[src0],          0x08(%[src_argb1555])                  \n\t"
>+      "gsldlc1    %[src0],          0x0f(%[src_argb1555])                  \n\t"
>+      "gsldrc1    %[src1],          0x08(%[next_argb1555])                 \n\t"
>+      "gsldlc1    %[src1],          0x0f(%[next_argb1555])                 \n\t"
>+      "psrlh      %[dest1_u],       %[src0],               %[eight]        \n\t"
>+      "and        %[b0],            %[src0],               %[c0]           \n\t"
>+      "and        %[src0],          %[src0],               %[c1]           \n\t"
>+      "psrlh      %[src0],          %[src0],               %[five]         \n\t"
>+      "and        %[g0],            %[dest1_u],            %[c2]           \n\t"
>+      "psllh      %[g0],            %[g0],                 %[three]        \n\t"
>+      "or         %[g0],            %[src0],               %[g0]           \n\t"
>+      "and        %[r0],            %[dest1_u],            %[c3]           \n\t"
>+      "psrlh      %[r0],            %[r0],                 %[two]          \n\t"
>+      "psrlh      %[src0],          %[src1],               %[eight]        \n\t"
>+      "and        %[dest1_u],       %[src1],               %[c0]           \n\t"
>+      "and        %[src1],          %[src1],               %[c1]           \n\t"
>+      "psrlh      %[src1],          %[src1],               %[five]         \n\t"
>+      "and        %[dest1_v],       %[src0],               %[c2]           \n\t"
>+      "psllh      %[dest1_v],       %[dest1_v],            %[three]        \n\t"
>+      "or         %[dest1_v],       %[src1],               %[dest1_v]      \n\t"
>+      "and        %[src0],          %[src0],               %[c3]           \n\t"
>+      "psrlh      %[src0],          %[src0],               %[two]          \n\t"
>+      "paddh      %[b0],            %[b0],                 %[dest1_u]      \n\t"
>+      "paddh      %[g0],            %[g0],                 %[dest1_v]      \n\t"
>+      "paddh      %[r0],            %[r0],                 %[src0]         \n\t"
>+      "punpcklhw  %[src0],          %[b0],                 %[r0]           \n\t"
>+      "punpckhhw  %[src1],          %[b0],                 %[r0]           \n\t"
>+      "punpcklwd  %[dest1_u],       %[src0],               %[src1]         \n\t"
>+      "punpckhwd  %[dest1_v],       %[src0],               %[src1]         \n\t"
>+      "paddh      %[src0],          %[dest1_u],            %[dest1_v]      \n\t"
>+      "psrlh      %[b0],            %[src0],               %[six]          \n\t"
>+      "psllh      %[r0],            %[src0],               %[one]          \n\t"
>+      "or         %[b0],            %[b0],                 %[r0]           \n\t"
>+      "psrlh      %[r0],            %[g0],                 %[six]          \n\t"
>+      "psllh      %[g0],            %[g0],                 %[one]          \n\t"
>+      "or         %[g0],            %[g0],                 %[r0]           \n\t"
>+      "punpcklhw  %[src0],          %[g0],                 %[value]        \n\t"
>+      "punpckhhw  %[src1],          %[g0],                 %[value]        \n\t"
>+      "punpcklwd  %[dest1_u],       %[src0],               %[src1]         \n\t"
>+      "punpckhwd  %[dest1_v],       %[src0],               %[src1]         \n\t"
>+      "paddh      %[g0],            %[dest1_u],            %[dest1_v]      \n\t"
>+      "punpcklhw  %[src0],          %[b0],                 %[g0]           \n\t"
>+      "punpckhhw  %[src1],          %[b0],                 %[g0]           \n\t"
>+
>+      "pmaddhw    %[dest1_v],       %[src0],               %[mask_v]       \n\t"
>+      "pshufh     %[dest1_u],       %[src0],               %[mask]         \n\t"
>+      "pmaddhw    %[dest1_u],       %[dest1_u],            %[mask_u]       \n\t"
>+      "pmaddhw    %[g0],            %[src1],               %[mask_v]       \n\t"
>+      "pshufh     %[b0],            %[src1],               %[mask]         \n\t"
>+      "pmaddhw    %[b0],            %[b0],                 %[mask_u]       \n\t"
>+
>+      "punpcklwd  %[src0],          %[dest1_u],            %[b0]           \n\t"
>+      "punpckhwd  %[src1],          %[dest1_u],            %[b0]           \n\t"
>+      "psubw      %[dest1_u],       %[src0],               %[src1]         \n\t"
>+      "psraw      %[dest1_u],       %[dest1_u],            %[eight]        \n\t"
>+      "punpcklwd  %[src0],          %[dest1_v],            %[g0]           \n\t"
>+      "punpckhwd  %[src1],          %[dest1_v],            %[g0]           \n\t"
>+      "psubw      %[dest1_v],       %[src1],               %[src0]         \n\t"
>+      "psraw      %[dest1_v],       %[dest1_v],            %[eight]        \n\t"
>+
>+      "packsswh   %[dest0_u],       %[dest0_u],            %[dest1_u]      \n\t"
>+      "packsswh   %[dest1_u],       %[dest0_v],            %[dest1_v]      \n\t"
>+
>+      "gsldrc1    %[src0],          0x10(%[src_argb1555])                  \n\t"
>+      "gsldlc1    %[src0],          0x17(%[src_argb1555])                  \n\t"
>+      "gsldrc1    %[src1],          0x10(%[next_argb1555])                 \n\t"
>+      "gsldlc1    %[src1],          0x17(%[next_argb1555])                 \n\t"
>+      "psrlh      %[dest2_u],       %[src0],               %[eight]        \n\t"
>+      "and        %[b0],            %[src0],               %[c0]           \n\t"
>+      "and        %[src0],          %[src0],               %[c1]           \n\t"
>+      "psrlh      %[src0],          %[src0],               %[five]         \n\t"
>+      "and        %[g0],            %[dest2_u],            %[c2]           \n\t"
>+      "psllh      %[g0],            %[g0],                 %[three]        \n\t"
>+      "or         %[g0],            %[src0],               %[g0]           \n\t"
>+      "and        %[r0],            %[dest2_u],            %[c3]           \n\t"
>+      "psrlh      %[r0],            %[r0],                 %[two]          \n\t"
>+      "psrlh      %[src0],          %[src1],               %[eight]        \n\t"
>+      "and        %[dest2_u],       %[src1],               %[c0]           \n\t"
>+      "and        %[src1],          %[src1],               %[c1]           \n\t"
>+      "psrlh      %[src1],          %[src1],               %[five]         \n\t"
>+      "and        %[dest0_v],       %[src0],               %[c2]           \n\t"
>+      "psllh      %[dest0_v],       %[dest0_v],            %[three]        \n\t"
>+      "or         %[dest0_v],       %[src1],               %[dest0_v]      \n\t"
>+      "and        %[src0],          %[src0],               %[c3]           \n\t"
>+      "psrlh      %[src0],          %[src0],               %[two]          \n\t"
>+      "paddh      %[b0],            %[b0],                 %[dest2_u]      \n\t"
>+      "paddh      %[g0],            %[g0],                 %[dest0_v]      \n\t"
>+      "paddh      %[r0],            %[r0],                 %[src0]         \n\t"
>+      "punpcklhw  %[src0],          %[b0],                 %[r0]           \n\t"
>+      "punpckhhw  %[src1],          %[b0],                 %[r0]           \n\t"
>+      "punpcklwd  %[dest2_u],       %[src0],               %[src1]         \n\t"
>+      "punpckhwd  %[dest0_v],       %[src0],               %[src1]         \n\t"
>+      "paddh      %[src0],          %[dest2_u],            %[dest0_v]      \n\t"
>+      "psrlh      %[b0],            %[src0],               %[six]          \n\t"
>+      "psllh      %[r0],            %[src0],               %[one]          \n\t"
>+      "or         %[b0],            %[b0],                 %[r0]           \n\t"
>+      "psrlh      %[r0],            %[g0],                 %[six]          \n\t"
>+      "psllh      %[g0],            %[g0],                 %[one]          \n\t"
>+      "or         %[g0],            %[g0],                 %[r0]           \n\t"
>+      "punpcklhw  %[src0],          %[g0],                 %[value]        \n\t"
>+      "punpckhhw  %[src1],          %[g0],                 %[value]        \n\t"
>+      "punpcklwd  %[dest2_u],       %[src0],               %[src1]         \n\t"
>+      "punpckhwd  %[dest0_v],       %[src0],               %[src1]         \n\t"
>+      "paddh      %[g0],            %[dest2_u],            %[dest0_v]      \n\t"
>+      "punpcklhw  %[src0],          %[b0],                 %[g0]           \n\t"
>+      "punpckhhw  %[src1],          %[b0],                 %[g0]           \n\t"
>+
>+      "pmaddhw    %[dest0_v],       %[src0],               %[mask_v]       \n\t"
>+      "pshufh     %[dest2_u],       %[src0],               %[mask]         \n\t"
>+      "pmaddhw    %[dest2_u],       %[dest2_u],            %[mask_u]       \n\t"
>+      "pmaddhw    %[g0],            %[src1],               %[mask_v]       \n\t"
>+      "pshufh     %[b0],            %[src1],               %[mask]         \n\t"
>+      "pmaddhw    %[b0],            %[b0],                 %[mask_u]       \n\t"
>+
>+      "punpcklwd  %[src0],          %[dest2_u],            %[b0]           \n\t"
>+      "punpckhwd  %[src1],          %[dest2_u],            %[b0]           \n\t"
>+      "psubw      %[dest2_u],       %[src0],               %[src1]         \n\t"
>+      "psraw      %[dest2_u],       %[dest2_u],            %[eight]        \n\t"
>+      "punpcklwd  %[src0],          %[dest0_v],            %[g0]           \n\t"
>+      "punpckhwd  %[src1],          %[dest0_v],            %[g0]           \n\t"
>+      "psubw      %[dest0_v],       %[src1],               %[src0]         \n\t"
>+      "psraw      %[dest0_v],       %[dest0_v],            %[eight]        \n\t"
>+
>+      "gsldrc1    %[src0],          0x18(%[src_argb1555])                  \n\t"
>+      "gsldlc1    %[src0],          0x1f(%[src_argb1555])                  \n\t"
>+      "gsldrc1    %[src1],          0x18(%[next_argb1555])                 \n\t"
>+      "gsldlc1    %[src1],          0x1f(%[next_argb1555])                 \n\t"
>+      "psrlh      %[dest3_u],       %[src0],               %[eight]        \n\t"
>+      "and        %[b0],            %[src0],               %[c0]           \n\t"
>+      "and        %[src0],          %[src0],               %[c1]           \n\t"
>+      "psrlh      %[src0],          %[src0],               %[five]         \n\t"
>+      "and        %[g0],            %[dest3_u],            %[c2]           \n\t"
>+      "psllh      %[g0],            %[g0],                 %[three]        \n\t"
>+      "or         %[g0],            %[src0],               %[g0]           \n\t"
>+      "and        %[r0],            %[dest3_u],            %[c3]           \n\t"
>+      "psrlh      %[r0],            %[r0],                 %[two]          \n\t"
>+      "psrlh      %[src0],          %[src1],               %[eight]        \n\t"
>+      "and        %[dest3_u],       %[src1],               %[c0]           \n\t"
>+      "and        %[src1],          %[src1],               %[c1]           \n\t"
>+      "psrlh      %[src1],          %[src1],               %[five]         \n\t"
>+      "and        %[dest1_v],       %[src0],               %[c2]           \n\t"
>+      "psllh      %[dest1_v],       %[dest1_v],            %[three]        \n\t"
>+      "or         %[dest1_v],       %[src1],               %[dest1_v]      \n\t"
>+      "and        %[src0],          %[src0],               %[c3]           \n\t"
>+      "psrlh      %[src0],          %[src0],               %[two]          \n\t"
>+      "paddh      %[b0],            %[b0],                 %[dest3_u]      \n\t"
>+      "paddh      %[g0],            %[g0],                 %[dest1_v]      \n\t"
>+      "paddh      %[r0],            %[r0],                 %[src0]         \n\t"
>+      "punpcklhw  %[src0],          %[b0],                 %[r0]           \n\t"
>+      "punpckhhw  %[src1],          %[b0],                 %[r0]           \n\t"
>+      "punpcklwd  %[dest3_u],       %[src0],               %[src1]         \n\t"
>+      "punpckhwd  %[dest1_v],       %[src0],               %[src1]         \n\t"
>+      "paddh      %[src0],          %[dest3_u],            %[dest1_v]      \n\t"
>+      "psrlh      %[b0],            %[src0],               %[six]          \n\t"
>+      "psllh      %[r0],            %[src0],               %[one]          \n\t"
>+      "or         %[b0],            %[b0],                 %[r0]           \n\t"
>+      "psrlh      %[r0],            %[g0],                 %[six]          \n\t"
>+      "psllh      %[g0],            %[g0],                 %[one]          \n\t"
>+      "or         %[g0],            %[g0],                 %[r0]           \n\t"
>+      "punpcklhw  %[src0],          %[g0],                 %[value]        \n\t"
>+      "punpckhhw  %[src1],          %[g0],                 %[value]        \n\t"
>+      "punpcklwd  %[dest3_u],       %[src0],               %[src1]         \n\t"
>+      "punpckhwd  %[dest1_v],       %[src0],               %[src1]         \n\t"
>+      "paddh      %[g0],            %[dest3_u],            %[dest1_v]      \n\t"
>+      "punpcklhw  %[src0],          %[b0],                 %[g0]           \n\t"
>+      "punpckhhw  %[src1],          %[b0],                 %[g0]           \n\t"
>+
>+      "pmaddhw    %[dest1_v],       %[src0],               %[mask_v]       \n\t"
>+      "pshufh     %[dest3_u],       %[src0],               %[mask]         \n\t"
>+      "pmaddhw    %[dest3_u],       %[dest3_u],            %[mask_u]       \n\t"
>+      "pmaddhw    %[g0],            %[src1],               %[mask_v]       \n\t"
>+      "pshufh     %[b0],            %[src1],               %[mask]         \n\t"
>+      "pmaddhw    %[b0],            %[b0],                 %[mask_u]       \n\t"
>+
>+      "punpcklwd  %[src0],          %[dest3_u],            %[b0]           \n\t"
>+      "punpckhwd  %[src1],          %[dest3_u],            %[b0]           \n\t"
>+      "psubw      %[dest3_u],       %[src0],               %[src1]         \n\t"
>+      "psraw      %[dest3_u],       %[dest3_u],            %[eight]        \n\t"
>+      "punpcklwd  %[src0],          %[dest1_v],            %[g0]           \n\t"
>+      "punpckhwd  %[src1],          %[dest1_v],            %[g0]           \n\t"
>+      "psubw      %[dest1_v],       %[src1],               %[src0]         \n\t"
>+      "psraw      %[dest1_v],       %[dest1_v],            %[eight]        \n\t"
>+
>+      "packsswh   %[src1],          %[dest2_u],            %[dest3_u]      \n\t"
>+      "packushb   %[dest0_u],       %[dest0_u],            %[src1]         \n\t"
>+      "gssdlc1    %[dest0_u],       0x07(%[dst_u])                         \n\t"
>+      "gssdrc1    %[dest0_u],       0x00(%[dst_u])                         \n\t"
>+      "packsswh   %[src1],          %[dest0_v],            %[dest1_v]      \n\t"
>+      "packushb   %[dest0_v],       %[dest1_u],            %[src1]         \n\t"
>+      "gssdlc1    %[dest0_v],       0x07(%[dst_v])                         \n\t"
>+      "gssdrc1    %[dest0_v],       0x00(%[dst_v])                         \n\t"
>+
>+      "daddiu    %[src_argb1555],   %[src_argb1555],       0x20            \n\t"
>+      "daddiu    %[next_argb1555],  %[next_argb1555],      0x20            \n\t"
>+      "daddiu    %[dst_u],          %[dst_u],              0x08            \n\t"
>+      "daddiu    %[dst_v],          %[dst_v],              0x08            \n\t"
>+      "daddiu    %[width],          %[width],             -0x10            \n\t"
>+      "bgtz      %[width],          1b                                     \n\t"
>+      : [src0] "=&f"(ftmp[0]), [src1] "=&f"(ftmp[1]), [b0] "=&f"(ftmp[2]),
>+        [g0] "=&f"(ftmp[3]), [r0] "=&f"(ftmp[4]), [dest0_u] "=&f"(ftmp[5]),
>+        [dest1_u] "=&f"(ftmp[6]), [dest2_u] "=&f"(ftmp[7]),
>+        [dest3_u] "=&f"(ftmp[8]), [dest0_v] "=&f"(ftmp[9]),
>+        [dest1_v] "=&f"(ftmp[10])
>+      : [src_argb1555] "r"(src_argb1555),
>+        [next_argb1555] "r"(src_stride_argb1555), [dst_u] "r"(dst_u),
>+        [dst_v] "r"(dst_v), [width] "r"(width), [value] "f"(value),
>+        [c0] "f"(c0), [c1] "f"(c1), [c2] "f"(c2), [c3] "f"(c3),
>+        [mask] "f"(mask), [mask_u] "f"(mask_u), [mask_v] "f"(mask_v),
>+        [eight] "f"(0x08), [six] "f"(0x06), [five] "f"(0x05), [three] "f"(0x03),
>+        [two] "f"(0x02), [one] "f"(0x01)
>+      : "memory");
>+}
>+
>+void ARGB4444ToUVRow_MMI(const uint8_t* src_argb4444,
>+                         int src_stride_argb4444,
>+                         uint8_t* dst_u,
>+                         uint8_t* dst_v,
>+                         int width) {
>+  uint64_t ftmp[13];
>+  uint64_t value = 0x2020202020202020;
>+  uint64_t mask_u = 0x0026004a00700002;
>+  uint64_t mask_v = 0x00020070005e0012;
>+  uint64_t mask = 0x93;
>+  uint64_t c0 = 0x000f000f000f000f;
>+  uint64_t c1 = 0x00ff00ff00ff00ff;
>+  __asm__ volatile(
>+      "daddu      %[next_argb4444], %[src_argb4444],      %[next_argb4444] \n\t"
>+      "1:                                                                  \n\t"
>+      "gsldrc1    %[src0],          0x00(%[src_argb4444])                  \n\t"
>+      "gsldlc1    %[src0],          0x07(%[src_argb4444])                  \n\t"
>+      "gsldrc1    %[src1],          0x00(%[next_argb4444])                 \n\t"
>+      "gsldlc1    %[src1],          0x07(%[next_argb4444])                 \n\t"
>+      "psrlh      %[dest0_u],       %[src0],               %[eight]        \n\t"
>+      "and        %[b0],            %[src0],               %[c0]           \n\t"
>+      "and        %[src0],          %[src0],               %[c1]           \n\t"
>+      "psrlh      %[g0],            %[src0],               %[four]         \n\t"
>+      "and        %[r0],            %[dest0_u],            %[c0]           \n\t"
>+      "psrlh      %[src0],          %[src1],               %[eight]        \n\t"
>+      "and        %[dest0_u],       %[src1],               %[c0]           \n\t"
>+      "and        %[src1],          %[src1],               %[c1]           \n\t"
>+      "psrlh      %[dest0_v],       %[src1],               %[four]         \n\t"
>+      "and        %[src0],          %[src0],               %[c0]           \n\t"
>+      "paddh      %[b0],            %[b0],                 %[dest0_u]      \n\t"
>+      "paddh      %[g0],            %[g0],                 %[dest0_v]      \n\t"
>+      "paddh      %[r0],            %[r0],                 %[src0]         \n\t"
>+      "punpcklhw  %[src0],          %[b0],                 %[r0]           \n\t"
>+      "punpckhhw  %[src1],          %[b0],                 %[r0]           \n\t"
>+      "punpcklwd  %[dest0_u],       %[src0],               %[src1]         \n\t"
>+      "punpckhwd  %[dest0_v],       %[src0],               %[src1]         \n\t"
>+      "paddh      %[src0],          %[dest0_u],            %[dest0_v]      \n\t"
>+      "psrlh      %[b0],            %[src0],               %[four]         \n\t"
>+      "psllh      %[r0],            %[src0],               %[two]          \n\t"
>+      "or         %[b0],            %[b0],                 %[r0]           \n\t"
>+      "psrlh      %[r0],            %[g0],                 %[four]         \n\t"
>+      "psllh      %[g0],            %[g0],                 %[two]          \n\t"
>+      "or         %[g0],            %[g0],                 %[r0]           \n\t"
>+      "punpcklhw  %[src0],          %[g0],                 %[value]        \n\t"
>+      "punpckhhw  %[src1],          %[g0],                 %[value]        \n\t"
>+      "punpcklwd  %[dest0_u],       %[src0],               %[src1]         \n\t"
>+      "punpckhwd  %[dest0_v],       %[src0],               %[src1]         \n\t"
>+      "paddh      %[g0],            %[dest0_u],            %[dest0_v]      \n\t"
>+      "punpcklhw  %[src0],          %[b0],                 %[g0]           \n\t"
>+      "punpckhhw  %[src1],          %[b0],                 %[g0]           \n\t"
>+
>+      "pmaddhw    %[dest0_v],       %[src0],               %[mask_v]       \n\t"
>+      "pshufh     %[dest0_u],       %[src0],               %[mask]         \n\t"
>+      "pmaddhw    %[dest0_u],       %[dest0_u],            %[mask_u]       \n\t"
>+      "pmaddhw    %[g0],            %[src1],               %[mask_v]       \n\t"
>+      "pshufh     %[b0],            %[src1],               %[mask]         \n\t"
>+      "pmaddhw    %[b0],            %[b0],                 %[mask_u]       \n\t"
>+
>+      "punpcklwd  %[src0],          %[dest0_u],            %[b0]           \n\t"
>+      "punpckhwd  %[src1],          %[dest0_u],            %[b0]           \n\t"
>+      "psubw      %[dest0_u],       %[src0],               %[src1]         \n\t"
>+      "psraw      %[dest0_u],       %[dest0_u],            %[eight]        \n\t"
>+      "punpcklwd  %[src0],          %[dest0_v],            %[g0]           \n\t"
>+      "punpckhwd  %[src1],          %[dest0_v],            %[g0]           \n\t"
>+      "psubw      %[dest0_v],       %[src1],               %[src0]         \n\t"
>+      "psraw      %[dest0_v],       %[dest0_v],            %[eight]        \n\t"
>+
>+      "gsldrc1    %[src0],          0x08(%[src_argb4444])                  \n\t"
>+      "gsldlc1    %[src0],          0x0f(%[src_argb4444])                  \n\t"
>+      "gsldrc1    %[src1],          0x08(%[next_argb4444])                 \n\t"
>+      "gsldlc1    %[src1],          0x0f(%[next_argb4444])                 \n\t"
>+      "psrlh      %[dest1_u],       %[src0],               %[eight]        \n\t"
>+      "and        %[b0],            %[src0],               %[c0]           \n\t"
>+      "and        %[src0],          %[src0],               %[c1]           \n\t"
>+      "psrlh      %[g0],            %[src0],               %[four]         \n\t"
>+      "and        %[r0],            %[dest1_u],            %[c0]           \n\t"
>+      "psrlh      %[src0],          %[src1],               %[eight]        \n\t"
>+      "and        %[dest1_u],       %[src1],               %[c0]           \n\t"
>+      "and        %[src1],          %[src1],               %[c1]           \n\t"
>+      "psrlh      %[dest1_v],       %[src1],               %[four]         \n\t"
>+      "and        %[src0],          %[src0],               %[c0]           \n\t"
>+      "paddh      %[b0],            %[b0],                 %[dest1_u]      \n\t"
>+      "paddh      %[g0],            %[g0],                 %[dest1_v]      \n\t"
>+      "paddh      %[r0],            %[r0],                 %[src0]         \n\t"
>+      "punpcklhw  %[src0],          %[b0],                 %[r0]           \n\t"
>+      "punpckhhw  %[src1],          %[b0],                 %[r0]           \n\t"
>+      "punpcklwd  %[dest1_u],       %[src0],               %[src1]         \n\t"
>+      "punpckhwd  %[dest1_v],       %[src0],               %[src1]         \n\t"
>+      "paddh      %[src0],          %[dest1_u],            %[dest1_v]      \n\t"
>+      "psrlh      %[b0],            %[src0],               %[four]         \n\t"
>+      "psllh      %[r0],            %[src0],               %[two]          \n\t"
>+      "or         %[b0],            %[b0],                 %[r0]           \n\t"
>+      "psrlh      %[r0],            %[g0],                 %[four]         \n\t"
>+      "psllh      %[g0],            %[g0],                 %[two]          \n\t"
>+      "or         %[g0],            %[g0],                 %[r0]           \n\t"
>+      "punpcklhw  %[src0],          %[g0],                 %[value]        \n\t"
>+      "punpckhhw  %[src1],          %[g0],                 %[value]        \n\t"
>+      "punpcklwd  %[dest1_u],       %[src0],               %[src1]         \n\t"
>+      "punpckhwd  %[dest1_v],       %[src0],               %[src1]         \n\t"
>+      "paddh      %[g0],            %[dest1_u],            %[dest1_v]      \n\t"
>+      "punpcklhw  %[src0],          %[b0],                 %[g0]           \n\t"
>+      "punpckhhw  %[src1],          %[b0],                 %[g0]           \n\t"
>+
>+      "pmaddhw    %[dest1_v],       %[src0],               %[mask_v]       \n\t"
>+      "pshufh     %[dest1_u],       %[src0],               %[mask]         \n\t"
>+      "pmaddhw    %[dest1_u],       %[dest1_u],            %[mask_u]       \n\t"
>+      "pmaddhw    %[g0],            %[src1],               %[mask_v]       \n\t"
>+      "pshufh     %[b0],            %[src1],               %[mask]         \n\t"
>+      "pmaddhw    %[b0],            %[b0],                 %[mask_u]       \n\t"
>+
>+      "punpcklwd  %[src0],          %[dest1_u],            %[b0]           \n\t"
>+      "punpckhwd  %[src1],          %[dest1_u],            %[b0]           \n\t"
>+      "psubw      %[dest1_u],       %[src0],               %[src1]         \n\t"
>+      "psraw      %[dest1_u],       %[dest1_u],            %[eight]        \n\t"
>+      "punpcklwd  %[src0],          %[dest1_v],            %[g0]           \n\t"
>+      "punpckhwd  %[src1],          %[dest1_v],            %[g0]           \n\t"
>+      "psubw      %[dest1_v],       %[src1],               %[src0]         \n\t"
>+      "psraw      %[dest1_v],       %[dest1_v],            %[eight]        \n\t"
>+
>+      "gsldrc1    %[src0],          0x10(%[src_argb4444])                  \n\t"
>+      "gsldlc1    %[src0],          0x17(%[src_argb4444])                  \n\t"
>+      "gsldrc1    %[src1],          0x10(%[next_argb4444])                 \n\t"
>+      "gsldlc1    %[src1],          0x17(%[next_argb4444])                 \n\t"
>+      "psrlh      %[dest2_u],       %[src0],               %[eight]        \n\t"
>+      "and        %[b0],            %[src0],               %[c0]           \n\t"
>+      "and        %[src0],          %[src0],               %[c1]           \n\t"
>+      "psrlh      %[g0],            %[src0],               %[four]         \n\t"
>+      "and        %[r0],            %[dest2_u],            %[c0]           \n\t"
>+      "psrlh      %[src0],          %[src1],               %[eight]        \n\t"
>+      "and        %[dest2_u],       %[src1],               %[c0]           \n\t"
>+      "and        %[src1],          %[src1],               %[c1]           \n\t"
>+      "psrlh      %[dest2_v],       %[src1],               %[four]         \n\t"
>+      "and        %[src0],          %[src0],               %[c0]           \n\t"
>+      "paddh      %[b0],            %[b0],                 %[dest2_u]      \n\t"
>+      "paddh      %[g0],            %[g0],                 %[dest2_v]      \n\t"
>+      "paddh      %[r0],            %[r0],                 %[src0]         \n\t"
>+      "punpcklhw  %[src0],          %[b0],                 %[r0]           \n\t"
>+      "punpckhhw  %[src1],          %[b0],                 %[r0]           \n\t"
>+      "punpcklwd  %[dest2_u],       %[src0],               %[src1]         \n\t"
>+      "punpckhwd  %[dest2_v],       %[src0],               %[src1]         \n\t"
>+      "paddh      %[src0],          %[dest2_u],            %[dest2_v]      \n\t"
>+      "psrlh      %[b0],            %[src0],               %[four]         \n\t"
>+      "psllh      %[r0],            %[src0],               %[two]          \n\t"
>+      "or         %[b0],            %[b0],                 %[r0]           \n\t"
>+      "psrlh      %[r0],            %[g0],                 %[four]         \n\t"
>+      "psllh      %[g0],            %[g0],                 %[two]          \n\t"
>+      "or         %[g0],            %[g0],                 %[r0]           \n\t"
>+      "punpcklhw  %[src0],          %[g0],                 %[value]        \n\t"
>+      "punpckhhw  %[src1],          %[g0],                 %[value]        \n\t"
>+      "punpcklwd  %[dest2_u],       %[src0],               %[src1]         \n\t"
>+      "punpckhwd  %[dest2_v],       %[src0],               %[src1]         \n\t"
>+      "paddh      %[g0],            %[dest2_u],            %[dest2_v]      \n\t"
>+      "punpcklhw  %[src0],          %[b0],                 %[g0]           \n\t"
>+      "punpckhhw  %[src1],          %[b0],                 %[g0]           \n\t"
>+
>+      "pmaddhw    %[dest2_v],       %[src0],               %[mask_v]       \n\t"
>+      "pshufh     %[dest2_u],       %[src0],               %[mask]         \n\t"
>+      "pmaddhw    %[dest2_u],       %[dest2_u],            %[mask_u]       \n\t"
>+      "pmaddhw    %[g0],            %[src1],               %[mask_v]       \n\t"
>+      "pshufh     %[b0],            %[src1],               %[mask]         \n\t"
>+      "pmaddhw    %[b0],            %[b0],                 %[mask_u]       \n\t"
>+
>+      "punpcklwd  %[src0],          %[dest2_u],            %[b0]           \n\t"
>+      "punpckhwd  %[src1],          %[dest2_u],            %[b0]           \n\t"
>+      "psubw      %[dest2_u],       %[src0],               %[src1]         \n\t"
>+      "psraw      %[dest2_u],       %[dest2_u],            %[eight]        \n\t"
>+      "punpcklwd  %[src0],          %[dest2_v],            %[g0]           \n\t"
>+      "punpckhwd  %[src1],          %[dest2_v],            %[g0]           \n\t"
>+      "psubw      %[dest2_v],       %[src1],               %[src0]         \n\t"
>+      "psraw      %[dest2_v],       %[dest2_v],            %[eight]        \n\t"
>+
>+      "gsldrc1    %[src0],          0x18(%[src_argb4444])                  \n\t"
>+      "gsldlc1    %[src0],          0x1f(%[src_argb4444])                  \n\t"
>+      "gsldrc1    %[src1],          0x18(%[next_argb4444])                 \n\t"
>+      "gsldlc1    %[src1],          0x1f(%[next_argb4444])                 \n\t"
>+      "psrlh      %[dest3_u],       %[src0],               %[eight]        \n\t"
>+      "and        %[b0],            %[src0],               %[c0]           \n\t"
>+      "and        %[src0],          %[src0],               %[c1]           \n\t"
>+      "psrlh      %[g0],            %[src0],               %[four]         \n\t"
>+      "and        %[r0],            %[dest3_u],            %[c0]           \n\t"
>+      "psrlh      %[src0],          %[src1],               %[eight]        \n\t"
>+      "and        %[dest3_u],       %[src1],               %[c0]           \n\t"
>+      "and        %[src1],          %[src1],               %[c1]           \n\t"
>+      "psrlh      %[dest3_v],       %[src1],               %[four]         \n\t"
>+      "and        %[src0],          %[src0],               %[c0]           \n\t"
>+      "paddh      %[b0],            %[b0],                 %[dest3_u]      \n\t"
>+      "paddh      %[g0],            %[g0],                 %[dest3_v]      \n\t"
>+      "paddh      %[r0],            %[r0],                 %[src0]         \n\t"
>+      "punpcklhw  %[src0],          %[b0],                 %[r0]           \n\t"
>+      "punpckhhw  %[src1],          %[b0],                 %[r0]           \n\t"
>+      "punpcklwd  %[dest3_u],       %[src0],               %[src1]         \n\t"
>+      "punpckhwd  %[dest3_v],       %[src0],               %[src1]         \n\t"
>+      "paddh      %[src0],          %[dest3_u],            %[dest3_v]      \n\t"
>+      "psrlh      %[b0],            %[src0],               %[four]         \n\t"
>+      "psllh      %[r0],            %[src0],               %[two]          \n\t"
>+      "or         %[b0],            %[b0],                 %[r0]           \n\t"
>+      "psrlh      %[r0],            %[g0],                 %[four]         \n\t"
>+      "psllh      %[g0],            %[g0],                 %[two]          \n\t"
>+      "or         %[g0],            %[g0],                 %[r0]           \n\t"
>+      "punpcklhw  %[src0],          %[g0],                 %[value]        \n\t"
>+      "punpckhhw  %[src1],          %[g0],                 %[value]        \n\t"
>+      "punpcklwd  %[dest3_u],       %[src0],               %[src1]         \n\t"
>+      "punpckhwd  %[dest3_v],       %[src0],               %[src1]         \n\t"
>+      "paddh      %[g0],            %[dest3_u],            %[dest3_v]      \n\t"
>+      "punpcklhw  %[src0],          %[b0],                 %[g0]           \n\t"
>+      "punpckhhw  %[src1],          %[b0],                 %[g0]           \n\t"
>+
>+      "pmaddhw    %[dest3_v],       %[src0],               %[mask_v]       \n\t"
>+      "pshufh     %[dest3_u],       %[src0],               %[mask]         \n\t"
>+      "pmaddhw    %[dest3_u],       %[dest3_u],            %[mask_u]       \n\t"
>+      "pmaddhw    %[g0],            %[src1],               %[mask_v]       \n\t"
>+      "pshufh     %[b0],            %[src1],               %[mask]         \n\t"
>+      "pmaddhw    %[b0],            %[b0],                 %[mask_u]       \n\t"
>+
>+      "punpcklwd  %[src0],          %[dest3_u],            %[b0]           \n\t"
>+      "punpckhwd  %[src1],          %[dest3_u],            %[b0]           \n\t"
>+      "psubw      %[dest3_u],       %[src0],               %[src1]         \n\t"
>+      "psraw      %[dest3_u],       %[dest3_u],            %[eight]        \n\t"
>+      "punpcklwd  %[src0],          %[dest3_v],            %[g0]           \n\t"
>+      "punpckhwd  %[src1],          %[dest3_v],            %[g0]           \n\t"
>+      "psubw      %[dest3_v],       %[src1],               %[src0]         \n\t"
>+      "psraw      %[dest3_v],       %[dest3_v],            %[eight]        \n\t"
>+
>+      "packsswh   %[src0],          %[dest0_u],            %[dest1_u]      \n\t"
>+      "packsswh   %[src1],          %[dest2_u],            %[dest3_u]      \n\t"
>+      "packushb   %[dest0_u],       %[src0],               %[src1]         \n\t"
>+      "gssdlc1    %[dest0_u],       0x07(%[dst_u])                         \n\t"
>+      "gssdrc1    %[dest0_u],       0x00(%[dst_u])                         \n\t"
>+      "packsswh   %[src0],          %[dest0_v],            %[dest1_v]      \n\t"
>+      "packsswh   %[src1],          %[dest2_v],            %[dest3_v]      \n\t"
>+      "packushb   %[dest0_v],       %[src0],               %[src1]         \n\t"
>+      "gssdlc1    %[dest0_v],       0x07(%[dst_v])                         \n\t"
>+      "gssdrc1    %[dest0_v],       0x00(%[dst_v])                         \n\t"
>+
>+      "daddiu    %[src_argb4444],   %[src_argb4444],       0x20            \n\t"
>+      "daddiu    %[next_argb4444],  %[next_argb4444],      0x20            \n\t"
>+      "daddiu    %[dst_u],          %[dst_u],              0x08            \n\t"
>+      "daddiu    %[dst_v],          %[dst_v],              0x08            \n\t"
>+      "daddiu    %[width],          %[width],             -0x10            \n\t"
>+      "bgtz      %[width],          1b                                     \n\t"
>+      : [src0] "=&f"(ftmp[0]), [src1] "=&f"(ftmp[1]), [b0] "=&f"(ftmp[2]),
>+        [g0] "=&f"(ftmp[3]), [r0] "=&f"(ftmp[4]), [dest0_u] "=&f"(ftmp[5]),
>+        [dest1_u] "=&f"(ftmp[6]), [dest2_u] "=&f"(ftmp[7]),
>+        [dest3_u] "=&f"(ftmp[8]), [dest0_v] "=&f"(ftmp[9]),
>+        [dest1_v] "=&f"(ftmp[10]), [dest2_v] "=&f"(ftmp[11]),
>+        [dest3_v] "=&f"(ftmp[12])
>+      : [src_argb4444] "r"(src_argb4444),
>+        [next_argb4444] "r"(src_stride_argb4444), [dst_u] "r"(dst_u),
>+        [dst_v] "r"(dst_v), [width] "r"(width), [value] "f"(value),
>+        [c0] "f"(c0), [c1] "f"(c1), [mask] "f"(mask), [mask_u] "f"(mask_u),
>+        [mask_v] "f"(mask_v), [eight] "f"(0x08), [four] "f"(0x04),
>+        [two] "f"(0x02)
>+      : "memory");
>+}
>+
>+void ARGBToUV444Row_MMI(const uint8_t* src_argb,
>+                        uint8_t* dst_u,
>+                        uint8_t* dst_v,
>+                        int width) {
>+  uint64_t ftmp[12];
>+  const uint64_t value = 0x4040;
>+  const uint64_t mask_u = 0x0026004a00700002;
>+  const uint64_t mask_v = 0x00020070005e0012;
>+
>+  __asm__ volatile(
>+      "1:                                                               \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src_argb])                    \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src_argb])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "dsll       %[dest0_u],      %[src_lo],         %[sixteen]        \n\t"
>+      "pinsrh_0   %[dest0_u],      %[dest0_u],        %[value]          \n\t"
>+      "pinsrh_3   %[dest0_v],      %[src_lo],         %[value]          \n\t"
>+      "pmaddhw    %[dest0_u],      %[dest0_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest0_v],      %[dest0_v],        %[mask_v]         \n\t"
>+
>+      "dsll       %[src_lo],       %[src_hi],         %[sixteen]        \n\t"
>+      "pinsrh_0   %[src_lo],       %[src_lo],         %[value]          \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest0_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest0_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest0_u],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest0_u],      %[dest0_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest0_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest0_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest0_v],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest0_v],      %[dest0_v],        %[eight]          \n\t"
>+
>+      "gsldrc1    %[src0],         0x08(%[src_argb])                    \n\t"
>+      "gsldlc1    %[src0],         0x0f(%[src_argb])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "dsll       %[dest1_u],      %[src_lo],         %[sixteen]        \n\t"
>+      "pinsrh_0   %[dest1_u],      %[dest1_u],        %[value]          \n\t"
>+      "pinsrh_3   %[dest1_v],      %[src_lo],         %[value]          \n\t"
>+      "pmaddhw    %[dest1_u],      %[dest1_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest1_v],      %[dest1_v],        %[mask_v]         \n\t"
>+      "dsll       %[src_lo],       %[src_hi],         %[sixteen]        \n\t"
>+      "pinsrh_0   %[src_lo],       %[src_lo],         %[value]          \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest1_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest1_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest1_u],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest1_u],      %[dest1_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest1_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest1_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest1_v],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest1_v],      %[dest1_v],        %[eight]          \n\t"
>+
>+      "gsldrc1    %[src0],         0x10(%[src_argb])                    \n\t"
>+      "gsldlc1    %[src0],         0x17(%[src_argb])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "dsll       %[dest2_u],      %[src_lo],         %[sixteen]        \n\t"
>+      "pinsrh_0   %[dest2_u],      %[dest2_u],        %[value]          \n\t"
>+      "pinsrh_3   %[dest2_v],      %[src_lo],         %[value]          \n\t"
>+      "pmaddhw    %[dest2_u],      %[dest2_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest2_v],      %[dest2_v],        %[mask_v]         \n\t"
>+      "dsll       %[src_lo],       %[src_hi],         %[sixteen]        \n\t"
>+      "pinsrh_0   %[src_lo],       %[src_lo],         %[value]          \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest2_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest2_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest2_u],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest2_u],      %[dest2_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest2_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest2_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest2_v],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest2_v],      %[dest2_v],        %[eight]          \n\t"
>+
>+      "gsldrc1    %[src0],         0x18(%[src_argb])                    \n\t"
>+      "gsldlc1    %[src0],         0x1f(%[src_argb])                    \n\t"
>+      "punpcklbh  %[src_lo],       %[src0],           %[zero]           \n\t"
>+      "punpckhbh  %[src_hi],       %[src0],           %[zero]           \n\t"
>+      "dsll       %[dest3_u],      %[src_lo],         %[sixteen]        \n\t"
>+      "pinsrh_0   %[dest3_u],      %[dest3_u],        %[value]          \n\t"
>+      "pinsrh_3   %[dest3_v],      %[src_lo],         %[value]          \n\t"
>+      "pmaddhw    %[dest3_u],      %[dest3_u],        %[mask_u]         \n\t"
>+      "pmaddhw    %[dest3_v],      %[dest3_v],        %[mask_v]         \n\t"
>+      "dsll       %[src_lo],       %[src_hi],         %[sixteen]        \n\t"
>+      "pinsrh_0   %[src_lo],       %[src_lo],         %[value]          \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[value]          \n\t"
>+      "pmaddhw    %[src_lo],       %[src_lo],         %[mask_u]         \n\t"
>+      "pmaddhw    %[src_hi],       %[src_hi],         %[mask_v]         \n\t"
>+
>+      "punpcklwd  %[src0],         %[dest3_u],        %[src_lo]         \n\t"
>+      "punpckhwd  %[src1],         %[dest3_u],        %[src_lo]         \n\t"
>+      "psubw      %[dest3_u],      %[src0],           %[src1]           \n\t"
>+      "psraw      %[dest3_u],      %[dest3_u],        %[eight]          \n\t"
>+      "punpcklwd  %[src0],         %[dest3_v],        %[src_hi]         \n\t"
>+      "punpckhwd  %[src1],         %[dest3_v],        %[src_hi]         \n\t"
>+      "psubw      %[dest3_v],      %[src1],           %[src0]           \n\t"
>+      "psraw      %[dest3_v],      %[dest3_v],        %[eight]          \n\t"
>+
>+      "packsswh   %[src0],         %[dest0_u],        %[dest1_u]        \n\t"
>+      "packsswh   %[src1],         %[dest2_u],        %[dest3_u]        \n\t"
>+      "packushb   %[dest0_u],      %[src0],           %[src1]           \n\t"
>+      "gssdlc1    %[dest0_u],      0x07(%[dst_u])                       \n\t"
>+      "gssdrc1    %[dest0_u],      0x00(%[dst_u])                       \n\t"
>+
>+      "packsswh   %[src0],         %[dest0_v],        %[dest1_v]        \n\t"
>+      "packsswh   %[src1],         %[dest2_v],        %[dest3_v]        \n\t"
>+      "packushb   %[dest0_v],      %[src0],           %[src1]           \n\t"
>+      "gssdlc1    %[dest0_v],      0x07(%[dst_v])                       \n\t"
>+      "gssdrc1    %[dest0_v],      0x00(%[dst_v])                       \n\t"
>+
>+      "daddiu     %[src_argb],     %[src_argb],       0x20              \n\t"
>+      "daddiu     %[dst_u],        %[dst_u],          0x08              \n\t"
>+      "daddiu     %[dst_v],        %[dst_v],          0x08              \n\t"
>+      "daddi      %[width],        %[width],         -0x08              \n\t"
>+      "bgtz       %[width],        1b                                   \n\t"
>+      : [src0] "=&f"(ftmp[0]), [src1] "=&f"(ftmp[1]), [src_lo] "=&f"(ftmp[2]),
>+        [src_hi] "=&f"(ftmp[3]), [dest0_u] "=&f"(ftmp[4]),
>+        [dest0_v] "=&f"(ftmp[5]), [dest1_u] "=&f"(ftmp[6]),
>+        [dest1_v] "=&f"(ftmp[7]), [dest2_u] "=&f"(ftmp[8]),
>+        [dest2_v] "=&f"(ftmp[9]), [dest3_u] "=&f"(ftmp[10]),
>+        [dest3_v] "=&f"(ftmp[11])
>+      : [src_argb] "r"(src_argb), [dst_u] "r"(dst_u), [dst_v] "r"(dst_v),
>+        [width] "r"(width), [mask_u] "f"(mask_u), [mask_v] "f"(mask_v),
>+        [value] "f"(value), [zero] "f"(0x00), [sixteen] "f"(0x10),
>+        [eight] "f"(0x08)
>+      : "memory");
>+}
>+
>+void ARGBGrayRow_MMI(const uint8_t* src_argb, uint8_t* dst_argb, int width) {
>+  uint64_t src, src_lo, src_hi, src37, dest, dest_lo, dest_hi;
>+  uint64_t tmp0, tmp1;
>+  const uint64_t mask0 = 0x0;
>+  const uint64_t mask1 = 0x01;
>+  const uint64_t mask2 = 0x00400026004B000FULL;
>+  const uint64_t mask3 = 0xFF000000FF000000ULL;
>+  const uint64_t mask4 = ~mask3;
>+  const uint64_t shift = 0x07;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[src],          0x07(%[src_ptr])                 \n\t"
>+      "gsldrc1    %[src],          0x00(%[src_ptr])                 \n\t"
>+
>+      "and        %[src37],        %[src],            %[mask3]      \n\t"
>+
>+      "punpcklbh  %[src_lo],       %[src],            %[mask0]      \n\t"
>+      "pinsrh_3   %[src_lo],       %[src_lo],         %[mask1]      \n\t"
>+      "pmaddhw    %[dest_lo],      %[src_lo],         %[mask2]      \n\t"
>+      "punpcklwd  %[tmp0],         %[dest_lo],        %[dest_lo]    \n\t"
>+      "punpckhwd  %[tmp1],         %[dest_lo],        %[dest_lo]    \n\t"
>+      "paddw      %[dest_lo],      %[tmp0],           %[tmp1]       \n\t"
>+      "psrlw      %[dest_lo],      %[dest_lo],        %[shift]      \n\t"
>+      "packsswh   %[dest_lo],      %[dest_lo],        %[dest_lo]    \n\t"
>+
>+      "punpckhbh  %[src_hi],       %[src],            %[mask0]      \n\t"
>+      "pinsrh_3   %[src_hi],       %[src_hi],         %[mask1]      \n\t"
>+      "pmaddhw    %[dest_hi],      %[src_hi],         %[mask2]      \n\t"
>+      "punpcklwd  %[tmp0],         %[dest_hi],        %[dest_hi]    \n\t"
>+      "punpckhwd  %[tmp1],         %[dest_hi],        %[dest_hi]    \n\t"
>+      "paddw      %[dest_hi],      %[tmp0],           %[tmp1]       \n\t"
>+      "psrlw      %[dest_hi],      %[dest_hi],        %[shift]      \n\t"
>+      "packsswh   %[dest_hi],      %[dest_hi],        %[dest_hi]    \n\t"
>+
>+      "packushb   %[dest],         %[dest_lo],        %[dest_hi]    \n\t"
>+      "and        %[dest],         %[dest],           %[mask4]      \n\t"
>+      "or         %[dest],         %[dest],           %[src37]      \n\t"
>+
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x08          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x02          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [dest_hi] "=&f"(dest_hi), [dest_lo] "=&f"(dest_lo),
>+        [src_hi] "=&f"(src_hi), [src_lo] "=&f"(src_lo), [tmp0] "=&f"(tmp0),
>+        [tmp1] "=&f"(tmp1), [src] "=&f"(src), [dest] "=&f"(dest),
>+        [src37] "=&f"(src37)
>+      : [src_ptr] "r"(src_argb), [dst_ptr] "r"(dst_argb), [width] "r"(width),
>+        [shift] "f"(shift), [mask0] "f"(mask0), [mask1] "f"(mask1),
>+        [mask2] "f"(mask2), [mask3] "f"(mask3), [mask4] "f"(mask4)
>+      : "memory");
>+}
>+
>+// Convert a row of image to Sepia tone.
>+void ARGBSepiaRow_MMI(uint8_t* dst_argb, int width) {
>+  uint64_t dest, dest_lo, dest_hi, dest37, dest0, dest1, dest2;
>+  uint64_t tmp0, tmp1;
>+  const uint64_t mask0 = 0x0;
>+  const uint64_t mask1 = 0x002300440011ULL;
>+  const uint64_t mask2 = 0x002D00580016ULL;
>+  const uint64_t mask3 = 0x003200620018ULL;
>+  const uint64_t mask4 = 0xFF000000FF000000ULL;
>+  const uint64_t shift = 0x07;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gsldrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "and        %[dest37],       %[dest],           %[mask4]      \n\t"
>+
>+      "punpcklbh  %[dest_lo],      %[dest],           %[mask0]      \n\t"
>+      "pmaddhw    %[dest0],        %[dest_lo],        %[mask1]      \n\t"
>+      "pmaddhw    %[dest1],        %[dest_lo],        %[mask2]      \n\t"
>+      "pmaddhw    %[dest2],        %[dest_lo],        %[mask3]      \n\t"
>+      "punpcklwd  %[tmp0],         %[dest0],          %[dest1]      \n\t"
>+      "punpckhwd  %[tmp1],         %[dest0],          %[dest1]      \n\t"
>+      "paddw      %[dest0],        %[tmp0],           %[tmp1]       \n\t"
>+      "psrlw      %[dest0],        %[dest0],          %[shift]      \n\t"
>+      "punpcklwd  %[tmp0],         %[dest2],          %[mask0]      \n\t"
>+      "punpckhwd  %[tmp1],         %[dest2],          %[mask0]      \n\t"
>+      "paddw      %[dest1],        %[tmp0],           %[tmp1]       \n\t"
>+      "psrlw      %[dest1],        %[dest1],          %[shift]      \n\t"
>+      "packsswh   %[dest_lo],      %[dest0],          %[dest1]      \n\t"
>+
>+      "punpckhbh  %[dest_hi],      %[dest],           %[mask0]      \n\t"
>+      "pmaddhw    %[dest0],        %[dest_hi],        %[mask1]      \n\t"
>+      "pmaddhw    %[dest1],        %[dest_hi],        %[mask2]      \n\t"
>+      "pmaddhw    %[dest2],        %[dest_hi],        %[mask3]      \n\t"
>+      "punpcklwd  %[tmp0],         %[dest0],          %[dest1]      \n\t"
>+      "punpckhwd  %[tmp1],         %[dest0],          %[dest1]      \n\t"
>+      "paddw      %[dest0],        %[tmp0],           %[tmp1]       \n\t"
>+      "psrlw      %[dest0],        %[dest0],          %[shift]      \n\t"
>+      "punpcklwd  %[tmp0],         %[dest2],          %[mask0]      \n\t"
>+      "punpckhwd  %[tmp1],         %[dest2],          %[mask0]      \n\t"
>+      "paddw      %[dest1],        %[tmp0],           %[tmp1]       \n\t"
>+      "psrlw      %[dest1],        %[dest1],          %[shift]      \n\t"
>+      "packsswh   %[dest_hi],      %[dest0],          %[dest1]      \n\t"
>+
>+      "packushb   %[dest],         %[dest_lo],        %[dest_hi]    \n\t"
>+      "or         %[dest],         %[dest],           %[dest37]     \n\t"
>+
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x02          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [dest_hi] "=&f"(dest_hi), [dest_lo] "=&f"(dest_lo),
>+        [dest0] "=&f"(dest0), [dest1] "=&f"(dest1), [dest2] "=&f"(dest2),
>+        [dest37] "=&f"(dest37), [tmp0] "=&f"(tmp0), [tmp1] "=&f"(tmp1),
>+        [dest] "=&f"(dest)
>+      : [dst_ptr] "r"(dst_argb), [width] "r"(width), [mask0] "f"(mask0),
>+        [mask1] "f"(mask1), [mask2] "f"(mask2), [mask3] "f"(mask3),
>+        [mask4] "f"(mask4), [shift] "f"(shift)
>+      : "memory");
>+}
>+
>+// Apply color matrix to a row of image. Matrix is signed.
>+// TODO(fbarchard): Consider adding rounding (+32).
>+void ARGBColorMatrixRow_MMI(const uint8_t* src_argb,
>+                            uint8_t* dst_argb,
>+                            const int8_t* matrix_argb,
>+                            int width) {
>+  uint64_t src, src_hi, src_lo, dest, dest_lo, dest_hi, dest0, dest1, dest2,
>+      dest3;
>+  uint64_t matrix, matrix_hi, matrix_lo;
>+  uint64_t tmp0, tmp1;
>+  const uint64_t shift0 = 0x06;
>+  const uint64_t shift1 = 0x08;
>+  const uint64_t mask0 = 0x0;
>+  const uint64_t mask1 = 0x08;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[src],          0x07(%[src_ptr])                 \n\t"
>+      "gsldrc1    %[src],          0x00(%[src_ptr])                 \n\t"
>+
>+      "punpcklbh  %[src_lo],       %[src],            %[mask0]      \n\t"
>+
>+      "gsldlc1    %[matrix],       0x07(%[matrix_ptr])              \n\t"
>+      "gsldrc1    %[matrix],       0x00(%[matrix_ptr])              \n\t"
>+      "punpcklbh  %[matrix_lo],    %[matrix],         %[mask0]      \n\t"
>+      "psllh      %[matrix_lo],    %[matrix_lo],      %[shift1]     \n\t"
>+      "psrah      %[matrix_lo],    %[matrix_lo],      %[shift1]     \n\t"
>+      "punpckhbh  %[matrix_hi],    %[matrix],         %[mask0]      \n\t"
>+      "psllh      %[matrix_hi],    %[matrix_hi],      %[shift1]     \n\t"
>+      "psrah      %[matrix_hi],    %[matrix_hi],      %[shift1]     \n\t"
>+      "pmaddhw    %[dest_lo],      %[src_lo],         %[matrix_lo]  \n\t"
>+      "pmaddhw    %[dest_hi],      %[src_lo],         %[matrix_hi]  \n\t"
>+      "punpcklwd  %[tmp0],         %[dest_lo],        %[dest_hi]    \n\t"
>+      "punpckhwd  %[tmp1],         %[dest_lo],        %[dest_hi]    \n\t"
>+      "paddw      %[dest0],        %[tmp0],           %[tmp1]       \n\t"
>+      "psraw      %[dest0],        %[dest0],          %[shift0]     \n\t"
>+
>+      "gsldlc1    %[matrix],       0x0f(%[matrix_ptr])              \n\t"
>+      "gsldrc1    %[matrix],       0x08(%[matrix_ptr])              \n\t"
>+      "punpcklbh  %[matrix_lo],    %[matrix],         %[mask0]      \n\t"
>+      "psllh      %[matrix_lo],    %[matrix_lo],      %[shift1]     \n\t"
>+      "psrah      %[matrix_lo],    %[matrix_lo],      %[shift1]     \n\t"
>+      "punpckhbh  %[matrix_hi],    %[matrix],         %[mask0]      \n\t"
>+      "psllh      %[matrix_hi],    %[matrix_hi],      %[shift1]     \n\t"
>+      "psrah      %[matrix_hi],    %[matrix_hi],      %[shift1]     \n\t"
>+      "pmaddhw    %[dest_lo],      %[src_lo],         %[matrix_lo]  \n\t"
>+      "pmaddhw    %[dest_hi],      %[src_lo],         %[matrix_hi]  \n\t"
>+      "punpcklwd  %[tmp0],         %[dest_lo],        %[dest_hi]    \n\t"
>+      "punpckhwd  %[tmp1],         %[dest_lo],        %[dest_hi]    \n\t"
>+      "paddw      %[dest1],        %[tmp0],           %[tmp1]       \n\t"
>+      "psraw      %[dest1],        %[dest1],          %[shift0]     \n\t"
>+
>+      "punpckhbh  %[src_hi],       %[src],            %[mask0]      \n\t"
>+
>+      "gsldlc1    %[matrix],       0x07(%[matrix_ptr])              \n\t"
>+      "gsldrc1    %[matrix],       0x00(%[matrix_ptr])              \n\t"
>+      "punpcklbh  %[matrix_lo],    %[matrix],         %[mask0]      \n\t"
>+      "psllh      %[matrix_lo],    %[matrix_lo],      %[shift1]     \n\t"
>+      "psrah      %[matrix_lo],    %[matrix_lo],      %[shift1]     \n\t"
>+      "punpckhbh  %[matrix_hi],    %[matrix],         %[mask0]      \n\t"
>+      "psllh      %[matrix_hi],    %[matrix_hi],      %[shift1]     \n\t"
>+      "psrah      %[matrix_hi],    %[matrix_hi],      %[shift1]     \n\t"
>+      "pmaddhw    %[dest_lo],      %[src_hi],         %[matrix_lo]  \n\t"
>+      "pmaddhw    %[dest_hi],      %[src_hi],         %[matrix_hi]  \n\t"
>+      "punpcklwd  %[tmp0],         %[dest_lo],        %[dest_hi]    \n\t"
>+      "punpckhwd  %[tmp1],         %[dest_lo],        %[dest_hi]    \n\t"
>+      "paddw      %[dest2],        %[tmp0],           %[tmp1]       \n\t"
>+      "psraw      %[dest2],        %[dest2],          %[shift0]     \n\t"
>+
>+      "gsldlc1    %[matrix],       0x0f(%[matrix_ptr])              \n\t"
>+      "gsldrc1    %[matrix],       0x08(%[matrix_ptr])              \n\t"
>+      "punpcklbh  %[matrix_lo],    %[matrix],         %[mask0]      \n\t"
>+      "psllh      %[matrix_lo],    %[matrix_lo],      %[shift1]     \n\t"
>+      "psrah      %[matrix_lo],    %[matrix_lo],      %[shift1]     \n\t"
>+      "punpckhbh  %[matrix_hi],    %[matrix],         %[mask0]      \n\t"
>+      "psllh      %[matrix_hi],    %[matrix_hi],      %[shift1]     \n\t"
>+      "psrah      %[matrix_hi],    %[matrix_hi],      %[shift1]     \n\t"
>+      "pmaddhw    %[dest_lo],      %[src_hi],         %[matrix_lo]  \n\t"
>+      "pmaddhw    %[dest_hi],      %[src_hi],         %[matrix_hi]  \n\t"
>+      "punpcklwd  %[tmp0],         %[dest_lo],        %[dest_hi]    \n\t"
>+      "punpckhwd  %[tmp1],         %[dest_lo],        %[dest_hi]    \n\t"
>+      "paddw      %[dest3],        %[tmp0],           %[tmp1]       \n\t"
>+      "psraw      %[dest3],        %[dest3],          %[shift0]     \n\t"
>+
>+      "packsswh   %[tmp0],         %[dest0],          %[dest1]      \n\t"
>+      "packsswh   %[tmp1],         %[dest2],          %[dest3]      \n\t"
>+      "packushb   %[dest],         %[tmp0],           %[tmp1]       \n\t"
>+
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x08          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x02          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src_hi] "=&f"(src_hi), [src_lo] "=&f"(src_lo),
>+        [dest_hi] "=&f"(dest_hi), [dest_lo] "=&f"(dest_lo),
>+        [dest0] "=&f"(dest0), [dest1] "=&f"(dest1), [dest2] "=&f"(dest2),
>+        [dest3] "=&f"(dest3), [src] "=&f"(src), [dest] "=&f"(dest),
>+        [tmp0] "=&f"(tmp0), [tmp1] "=&f"(tmp1), [matrix_hi] "=&f"(matrix_hi),
>+        [matrix_lo] "=&f"(matrix_lo), [matrix] "=&f"(matrix)
>+      : [src_ptr] "r"(src_argb), [matrix_ptr] "r"(matrix_argb),
>+        [dst_ptr] "r"(dst_argb), [width] "r"(width), [shift0] "f"(shift0),
>+        [shift1] "f"(shift1), [mask0] "f"(mask0), [mask1] "f"(mask1)
>+      : "memory");
>+}
>+
>+void ARGBShadeRow_MMI(const uint8_t* src_argb,
>+                      uint8_t* dst_argb,
>+                      int width,
>+                      uint32_t value) {
>+  uint64_t src, src_hi, src_lo, dest, dest_lo, dest_hi;
>+  const uint64_t shift = 0x08;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[src],          0x07(%[src_ptr])                 \n\t"
>+      "gsldrc1    %[src],          0x00(%[src_ptr])                 \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[src]        \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[src]        \n\t"
>+
>+      "punpcklbh  %[value],        %[value],          %[value]      \n\t"
>+
>+      "pmulhuh    %[dest_lo],      %[src_lo],         %[value]      \n\t"
>+      "psrlh      %[dest_lo],      %[dest_lo],        %[shift]      \n\t"
>+      "pmulhuh    %[dest_hi],      %[src_hi],         %[value]      \n\t"
>+      "psrlh      %[dest_hi],      %[dest_hi],        %[shift]      \n\t"
>+      "packushb   %[dest],         %[dest_lo],        %[dest_hi]    \n\t"
>+
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x08          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x02          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src_hi] "=&f"(src_hi), [src_lo] "=&f"(src_lo),
>+        [dest_hi] "=&f"(dest_hi), [dest_lo] "=&f"(dest_lo), [src] "=&f"(src),
>+        [dest] "=&f"(dest)
>+      : [src_ptr] "r"(src_argb), [dst_ptr] "r"(dst_argb), [width] "r"(width),
>+        [value] "f"(value), [shift] "f"(shift)
>+      : "memory");
>+}
>+
>+void ARGBMultiplyRow_MMI(const uint8_t* src_argb0,
>+                         const uint8_t* src_argb1,
>+                         uint8_t* dst_argb,
>+                         int width) {
>+  uint64_t src0, src0_hi, src0_lo, src1, src1_hi, src1_lo;
>+  uint64_t dest, dest_lo, dest_hi;
>+  const uint64_t mask = 0x0;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src0_ptr])                \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src0_ptr])                \n\t"
>+      "punpcklbh  %[src0_lo],      %[src0],           %[src0]       \n\t"
>+      "punpckhbh  %[src0_hi],      %[src0],           %[src0]       \n\t"
>+
>+      "gsldlc1    %[src1],         0x07(%[src1_ptr])                \n\t"
>+      "gsldrc1    %[src1],         0x00(%[src1_ptr])                \n\t"
>+      "punpcklbh  %[src1_lo],      %[src1],           %[mask]       \n\t"
>+      "punpckhbh  %[src1_hi],      %[src1],           %[mask]       \n\t"
>+
>+      "pmulhuh    %[dest_lo],      %[src0_lo],        %[src1_lo]    \n\t"
>+      "pmulhuh    %[dest_hi],      %[src0_hi],        %[src1_hi]    \n\t"
>+      "packushb   %[dest],         %[dest_lo],        %[dest_hi]    \n\t"
>+
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src0_ptr],     %[src0_ptr],       0x08          \n\t"
>+      "daddiu     %[src1_ptr],     %[src1_ptr],       0x08          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x02          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src0_hi] "=&f"(src0_hi), [src0_lo] "=&f"(src0_lo),
>+        [src1_hi] "=&f"(src1_hi), [src1_lo] "=&f"(src1_lo),
>+        [dest_hi] "=&f"(dest_hi), [dest_lo] "=&f"(dest_lo), [src0] "=&f"(src0),
>+        [src1] "=&f"(src1), [dest] "=&f"(dest)
>+      : [src0_ptr] "r"(src_argb0), [src1_ptr] "r"(src_argb1),
>+        [dst_ptr] "r"(dst_argb), [width] "r"(width), [mask] "f"(mask)
>+      : "memory");
>+}
>+
>+void ARGBAddRow_MMI(const uint8_t* src_argb0,
>+                    const uint8_t* src_argb1,
>+                    uint8_t* dst_argb,
>+                    int width) {
>+  uint64_t src0, src1, dest;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src0_ptr])                \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src0_ptr])                \n\t"
>+      "gsldlc1    %[src1],         0x07(%[src1_ptr])                \n\t"
>+      "gsldrc1    %[src1],         0x00(%[src1_ptr])                \n\t"
>+      "paddusb    %[dest],         %[src0],           %[src1]       \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src0_ptr],     %[src0_ptr],       0x08          \n\t"
>+      "daddiu     %[src1_ptr],     %[src1_ptr],       0x08          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x02          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src0] "=&f"(src0), [src1] "=&f"(src1), [dest] "=&f"(dest)
>+      : [src0_ptr] "r"(src_argb0), [src1_ptr] "r"(src_argb1),
>+        [dst_ptr] "r"(dst_argb), [width] "r"(width)
>+      : "memory");
>+}
>+
>+void ARGBSubtractRow_MMI(const uint8_t* src_argb0,
>+                         const uint8_t* src_argb1,
>+                         uint8_t* dst_argb,
>+                         int width) {
>+  uint64_t src0, src1, dest;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src0_ptr])                \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src0_ptr])                \n\t"
>+      "gsldlc1    %[src1],         0x07(%[src1_ptr])                \n\t"
>+      "gsldrc1    %[src1],         0x00(%[src1_ptr])                \n\t"
>+      "psubusb    %[dest],         %[src0],           %[src1]       \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src0_ptr],     %[src0_ptr],       0x08          \n\t"
>+      "daddiu     %[src1_ptr],     %[src1_ptr],       0x08          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x02          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src0] "=&f"(src0), [src1] "=&f"(src1), [dest] "=&f"(dest)
>+      : [src0_ptr] "r"(src_argb0), [src1_ptr] "r"(src_argb1),
>+        [dst_ptr] "r"(dst_argb), [width] "r"(width)
>+      : "memory");
>+}
>+
>+// Sobel functions which mimics SSSE3.
>+void SobelXRow_MMI(const uint8_t* src_y0,
>+                   const uint8_t* src_y1,
>+                   const uint8_t* src_y2,
>+                   uint8_t* dst_sobelx,
>+                   int width) {
>+  uint64_t y00 = 0, y10 = 0, y20 = 0;
>+  uint64_t y02 = 0, y12 = 0, y22 = 0;
>+  uint64_t zero = 0x0;
>+  uint64_t sobel = 0x0;
>+  __asm__ volatile(
>+      "1:	                                         \n\t"
>+      "gsldlc1   %[y00],        0x07(%[src_y0])          \n\t"  // a=src_y0[i]
>+      "gsldrc1   %[y00],        0x00(%[src_y0])          \n\t"
>+      "gsldlc1   %[y02],        0x09(%[src_y0])          \n\t"  // a_sub=src_y0[i+2]
>+      "gsldrc1   %[y02],        0x02(%[src_y0])          \n\t"
>+
>+      "gsldlc1   %[y10],        0x07(%[src_y1])          \n\t"  // b=src_y1[i]
>+      "gsldrc1   %[y10],        0x00(%[src_y1])          \n\t"
>+      "gsldlc1   %[y12],        0x09(%[src_y1])          \n\t"  // b_sub=src_y1[i+2]
>+      "gsldrc1   %[y12],        0x02(%[src_y1])          \n\t"
>+
>+      "gsldlc1   %[y20],        0x07(%[src_y2])          \n\t"  // c=src_y2[i]
>+      "gsldrc1   %[y20],        0x00(%[src_y2])          \n\t"
>+      "gsldlc1   %[y22],        0x09(%[src_y2])          \n\t"  // c_sub=src_y2[i+2]
>+      "gsldrc1   %[y22],        0x02(%[src_y2])          \n\t"
>+
>+      "punpcklbh %[y00],        %[y00],          %[zero] \n\t"
>+      "punpcklbh %[y10],        %[y10],          %[zero] \n\t"
>+      "punpcklbh %[y20],        %[y20],          %[zero] \n\t"
>+
>+      "punpcklbh %[y02],        %[y02],          %[zero] \n\t"
>+      "punpcklbh %[y12],        %[y12],          %[zero] \n\t"
>+      "punpcklbh %[y22],        %[y22],          %[zero] \n\t"
>+
>+      "paddh     %[y00],        %[y00],          %[y10]  \n\t"  // a+b
>+      "paddh     %[y20],        %[y20],          %[y10]  \n\t"  // c+b
>+      "paddh     %[y00],        %[y00],          %[y20]  \n\t"  // a+2b+c
>+
>+      "paddh     %[y02],        %[y02],          %[y12]  \n\t"  // a_sub+b_sub
>+      "paddh     %[y22],        %[y22],          %[y12]  \n\t"  // c_sub+b_sub
>+      "paddh     %[y02],        %[y02],          %[y22]  \n\t"  // a_sub+2b_sub+c_sub
>+
>+      "pmaxsh    %[y10],        %[y00],          %[y02]  \n\t"
>+      "pminsh    %[y20],        %[y00],          %[y02]  \n\t"
>+      "psubh     %[sobel],      %[y10],          %[y20]  \n\t"  // Abs
>+
>+      "gsldlc1   %[y00],        0x0B(%[src_y0])          \n\t"
>+      "gsldrc1   %[y00],        0x04(%[src_y0])          \n\t"
>+      "gsldlc1   %[y02],        0x0D(%[src_y0])          \n\t"
>+      "gsldrc1   %[y02],        0x06(%[src_y0])          \n\t"
>+
>+      "gsldlc1   %[y10],        0x0B(%[src_y1])          \n\t"
>+      "gsldrc1   %[y10],        0x04(%[src_y1])          \n\t"
>+      "gsldlc1   %[y12],        0x0D(%[src_y1])          \n\t"
>+      "gsldrc1   %[y12],        0x06(%[src_y1])          \n\t"
>+
>+      "gsldlc1   %[y20],        0x0B(%[src_y2])          \n\t"
>+      "gsldrc1   %[y20],        0x04(%[src_y2])          \n\t"
>+      "gsldlc1   %[y22],        0x0D(%[src_y2])          \n\t"
>+      "gsldrc1   %[y22],        0x06(%[src_y2])          \n\t"
>+
>+      "punpcklbh %[y00],        %[y00],          %[zero] \n\t"
>+      "punpcklbh %[y10],        %[y10],          %[zero] \n\t"
>+      "punpcklbh %[y20],        %[y20],          %[zero] \n\t"
>+
>+      "punpcklbh %[y02],        %[y02],          %[zero] \n\t"
>+      "punpcklbh %[y12],        %[y12],          %[zero] \n\t"
>+      "punpcklbh %[y22],        %[y22],          %[zero] \n\t"
>+
>+      "paddh     %[y00],        %[y00],          %[y10]  \n\t"
>+      "paddh     %[y20],        %[y20],          %[y10]  \n\t"
>+      "paddh     %[y00],        %[y00],          %[y20]  \n\t"
>+
>+      "paddh     %[y02],        %[y02],          %[y12]  \n\t"
>+      "paddh     %[y22],        %[y22],          %[y12]  \n\t"
>+      "paddh     %[y02],        %[y02],          %[y22]  \n\t"
>+
>+      "pmaxsh    %[y10],        %[y00],          %[y02]  \n\t"
>+      "pminsh    %[y20],        %[y00],          %[y02]  \n\t"
>+      "psubh     %[y00],        %[y10],          %[y20]  \n\t"
>+
>+      "packushb  %[sobel],      %[sobel],        %[y00]  \n\t"  // clamp255
>+      "gssdrc1   %[sobel],      0(%[dst_sobelx])         \n\t"
>+      "gssdlc1   %[sobel],      7(%[dst_sobelx])         \n\t"
>+
>+      "daddiu    %[src_y0],     %[src_y0],      8        \n\t"
>+      "daddiu    %[src_y1],     %[src_y1],      8        \n\t"
>+      "daddiu    %[src_y2],     %[src_y2],      8        \n\t"
>+      "daddiu    %[dst_sobelx], %[dst_sobelx],  8        \n\t"
>+      "daddiu    %[width],      %[width],      -8        \n\t"
>+      "bgtz      %[width],      1b                       \n\t"
>+      "nop                                               \n\t"
>+      : [sobel] "=&f"(sobel), [y00] "=&f"(y00), [y10] "=&f"(y10),
>+        [y20] "=&f"(y20), [y02] "=&f"(y02), [y12] "=&f"(y12), [y22] "=&f"(y22)
>+      : [src_y0] "r"(src_y0), [src_y1] "r"(src_y1), [src_y2] "r"(src_y2),
>+        [dst_sobelx] "r"(dst_sobelx), [width] "r"(width), [zero] "f"(zero)
>+      : "memory");
>+}
>+
>+void SobelYRow_MMI(const uint8_t* src_y0,
>+                   const uint8_t* src_y1,
>+                   uint8_t* dst_sobely,
>+                   int width) {
>+  uint64_t y00 = 0, y01 = 0, y02 = 0;
>+  uint64_t y10 = 0, y11 = 0, y12 = 0;
>+  uint64_t zero = 0x0;
>+  uint64_t sobel = 0x0;
>+  __asm__ volatile(
>+      "1:	                                        \n\t"
>+      "gsldlc1   %[y00],        0x07(%[src_y0])         \n\t"  // a=src_y0[i]
>+      "gsldrc1   %[y00],        0x00(%[src_y0])         \n\t"
>+      "gsldlc1   %[y01],        0x08(%[src_y0])         \n\t"  // b=src_y0[i+1]
>+      "gsldrc1   %[y01],        0x01(%[src_y0])         \n\t"
>+      "gsldlc1   %[y02],        0x09(%[src_y0])         \n\t"  // c=src_y0[i+2]
>+      "gsldrc1   %[y02],        0x02(%[src_y0])         \n\t"
>+
>+      "gsldlc1   %[y10],        0x07(%[src_y1])         \n\t"  // a_sub=src_y1[i]
>+      "gsldrc1   %[y10],        0x00(%[src_y1])         \n\t"
>+      "gsldlc1   %[y11],        0x08(%[src_y1])         \n\t"  // b_sub=src_y1[i+1]
>+      "gsldrc1   %[y11],        0x01(%[src_y1])         \n\t"
>+      "gsldlc1   %[y12],        0x09(%[src_y1])         \n\t"  // c_sub=src_y1[i+2]
>+      "gsldrc1   %[y12],        0x02(%[src_y1])         \n\t"
>+
>+      "punpcklbh %[y00],        %[y00],         %[zero] \n\t"
>+      "punpcklbh %[y01],        %[y01],         %[zero] \n\t"
>+      "punpcklbh %[y02],        %[y02],         %[zero] \n\t"
>+
>+      "punpcklbh %[y10],        %[y10],         %[zero] \n\t"
>+      "punpcklbh %[y11],        %[y11],         %[zero] \n\t"
>+      "punpcklbh %[y12],        %[y12],         %[zero] \n\t"
>+
>+      "paddh     %[y00],        %[y00],         %[y01]  \n\t"  // a+b
>+      "paddh     %[y02],        %[y02],         %[y01]  \n\t"  // c+b
>+      "paddh     %[y00],        %[y00],         %[y02]  \n\t"  // a+2b+c
>+
>+      "paddh     %[y10],        %[y10],         %[y11]  \n\t"  // a_sub+b_sub
>+      "paddh     %[y12],        %[y12],         %[y11]  \n\t"  // c_sub+b_sub
>+      "paddh     %[y10],        %[y10],         %[y12]  \n\t"  // a_sub+2b_sub+c_sub
>+
>+      "pmaxsh    %[y02],        %[y00],         %[y10]  \n\t"
>+      "pminsh    %[y12],        %[y00],         %[y10]  \n\t"
>+      "psubh     %[sobel],      %[y02],         %[y12]  \n\t"  // Abs
>+
>+      "gsldlc1   %[y00],        0x0B(%[src_y0])         \n\t"
>+      "gsldrc1   %[y00],        0x04(%[src_y0])         \n\t"
>+      "gsldlc1   %[y01],        0x0C(%[src_y0])         \n\t"
>+      "gsldrc1   %[y01],        0x05(%[src_y0])         \n\t"
>+      "gsldlc1   %[y02],        0x0D(%[src_y0])         \n\t"
>+      "gsldrc1   %[y02],        0x06(%[src_y0])         \n\t"
>+
>+      "gsldlc1   %[y10],        0x0B(%[src_y1])         \n\t"
>+      "gsldrc1   %[y10],        0x04(%[src_y1])         \n\t"
>+      "gsldlc1   %[y11],        0x0C(%[src_y1])         \n\t"
>+      "gsldrc1   %[y11],        0x05(%[src_y1])         \n\t"
>+      "gsldlc1   %[y12],        0x0D(%[src_y1])         \n\t"
>+      "gsldrc1   %[y12],        0x06(%[src_y1])         \n\t"
>+
>+      "punpcklbh %[y00],        %[y00],         %[zero] \n\t"
>+      "punpcklbh %[y01],        %[y01],         %[zero] \n\t"
>+      "punpcklbh %[y02],        %[y02],         %[zero] \n\t"
>+
>+      "punpcklbh %[y10],        %[y10],         %[zero] \n\t"
>+      "punpcklbh %[y11],        %[y11],         %[zero] \n\t"
>+      "punpcklbh %[y12],        %[y12],         %[zero] \n\t"
>+
>+      "paddh     %[y00],        %[y00],         %[y01]  \n\t"
>+      "paddh     %[y02],        %[y02],         %[y01]  \n\t"
>+      "paddh     %[y00],        %[y00],         %[y02]  \n\t"
>+
>+      "paddh     %[y10],        %[y10],         %[y11]  \n\t"
>+      "paddh     %[y12],        %[y12],         %[y11]  \n\t"
>+      "paddh     %[y10],        %[y10],         %[y12]  \n\t"
>+
>+      "pmaxsh    %[y02],        %[y00],         %[y10]  \n\t"
>+      "pminsh    %[y12],        %[y00],         %[y10]  \n\t"
>+      "psubh     %[y00],        %[y02],         %[y12]  \n\t"
>+
>+      "packushb  %[sobel],      %[sobel],       %[y00]  \n\t"  // clamp255
>+      "gssdrc1   %[sobel],      0(%[dst_sobely])        \n\t"
>+      "gssdlc1   %[sobel],      7(%[dst_sobely])        \n\t"
>+
>+      "daddiu    %[src_y0],     %[src_y0],      8       \n\t"
>+      "daddiu    %[src_y1],     %[src_y1],      8       \n\t"
>+      "daddiu    %[dst_sobely], %[dst_sobely],  8       \n\t"
>+      "daddiu    %[width],      %[width],      -8       \n\t"
>+      "bgtz      %[width],      1b                      \n\t"
>+      "nop                                              \n\t"
>+      : [sobel] "=&f"(sobel), [y00] "=&f"(y00), [y01] "=&f"(y01),
>+        [y02] "=&f"(y02), [y10] "=&f"(y10), [y11] "=&f"(y11), [y12] "=&f"(y12)
>+      : [src_y0] "r"(src_y0), [src_y1] "r"(src_y1),
>+        [dst_sobely] "r"(dst_sobely), [width] "r"(width), [zero] "f"(zero)
>+      : "memory");
>+}
>+
>+void SobelRow_MMI(const uint8_t* src_sobelx,
>+                  const uint8_t* src_sobely,
>+                  uint8_t* dst_argb,
>+                  int width) {
>+  double temp[3];
>+  uint64_t c1 = 0xff000000ff000000;
>+  __asm__ volatile(
>+      "1:	                                          \n\t"
>+      "gsldlc1   %[t0],         0x07(%[src_sobelx])       \n\t"  // a=src_sobelx[i]
>+      "gsldrc1   %[t0],         0x00(%[src_sobelx])       \n\t"
>+      "gsldlc1   %[t1],         0x07(%[src_sobely])       \n\t"  // b=src_sobely[i]
>+      "gsldrc1   %[t1],         0x00(%[src_sobely])       \n\t"
>+      // s7 s6 s5 s4 s3 s2 s1 s0 = a+b
>+      "paddusb   %[t2] ,        %[t0],              %[t1] \n\t"
>+
>+      // s3 s2 s1 s0->s3 s3 s2 s2 s1 s1 s0 s0
>+      "punpcklbh %[t0],         %[t2],              %[t2] \n\t"
>+
>+      // s1 s1 s0 s0->s1 s2 s1 s1 s0 s0 s0 s0
>+      "punpcklbh %[t1],         %[t0],              %[t0] \n\t"
>+      "or        %[t1],         %[t1],              %[c1] \n\t"
>+      // 255 s1 s1 s1 s55 s0 s0 s0
>+      "gssdrc1   %[t1],         0x00(%[dst_argb])	  \n\t"
>+      "gssdlc1   %[t1],         0x07(%[dst_argb])         \n\t"
>+
>+      // s3 s3 s2 s2->s3 s3 s3 s3 s2 s2 s2 s2
>+      "punpckhbh %[t1],         %[t0],              %[t0] \n\t"
>+      "or        %[t1],         %[t1],              %[c1] \n\t"
>+      // 255 s3 s3 s3 255 s2 s2 s2
>+      "gssdrc1   %[t1],         0x08(%[dst_argb])	  \n\t"
>+      "gssdlc1   %[t1],         0x0f(%[dst_argb])         \n\t"
>+
>+      // s7 s6 s5 s4->s7 s7 s6 s6 s5 s5 s4 s4
>+      "punpckhbh %[t0],         %[t2],              %[t2] \n\t"
>+
>+      // s5 s5 s4 s4->s5 s5 s5 s5 s4 s4 s4 s4
>+      "punpcklbh %[t1],         %[t0],              %[t0] \n\t"
>+      "or        %[t1],         %[t1],              %[c1] \n\t"
>+      "gssdrc1   %[t1],         0x10(%[dst_argb])	  \n\t"
>+      "gssdlc1   %[t1],         0x17(%[dst_argb])         \n\t"
>+
>+      // s7 s7 s6 s6->s7 s7 s7 s7 s6 s6 s6 s6
>+      "punpckhbh %[t1],         %[t0],              %[t0] \n\t"
>+      "or        %[t1],         %[t1],              %[c1] \n\t"
>+      "gssdrc1   %[t1],         0x18(%[dst_argb])	  \n\t"
>+      "gssdlc1   %[t1],         0x1f(%[dst_argb])         \n\t"
>+
>+      "daddiu    %[dst_argb],   %[dst_argb],        32    \n\t"
>+      "daddiu    %[src_sobelx], %[src_sobelx],      8     \n\t"
>+      "daddiu    %[src_sobely], %[src_sobely],      8     \n\t"
>+      "daddiu    %[width],      %[width],          -8     \n\t"
>+      "bgtz      %[width],      1b                        \n\t"
>+      "nop                                                \n\t"
>+      : [t0] "=&f"(temp[0]), [t1] "=&f"(temp[1]), [t2] "=&f"(temp[2])
>+      : [src_sobelx] "r"(src_sobelx), [src_sobely] "r"(src_sobely),
>+        [dst_argb] "r"(dst_argb), [width] "r"(width), [c1] "f"(c1)
>+      : "memory");
>+}
>+
>+void SobelToPlaneRow_MMI(const uint8_t* src_sobelx,
>+                         const uint8_t* src_sobely,
>+                         uint8_t* dst_y,
>+                         int width) {
>+  uint64_t tr = 0;
>+  uint64_t tb = 0;
>+  __asm__ volatile(
>+      "1:	                                       \n\t"
>+      "gsldrc1 %[tr],         0x0(%[src_sobelx])       \n\t"
>+      "gsldlc1 %[tr],         0x7(%[src_sobelx])       \n\t"  // r=src_sobelx[i]
>+      "gsldrc1 %[tb],         0x0(%[src_sobely])       \n\t"
>+      "gsldlc1 %[tb],         0x7(%[src_sobely])       \n\t"  // b=src_sobely[i]
>+      "paddusb %[tr],         %[tr],             %[tb] \n\t"  // g
>+      "gssdrc1 %[tr],         0x0(%[dst_y])	       \n\t"
>+      "gssdlc1 %[tr],         0x7(%[dst_y])            \n\t"
>+
>+      "daddiu  %[dst_y],      %[dst_y],          8     \n\t"
>+      "daddiu  %[src_sobelx], %[src_sobelx],     8     \n\t"
>+      "daddiu  %[src_sobely], %[src_sobely],     8     \n\t"
>+      "daddiu  %[width],      %[width],         -8     \n\t"
>+      "bgtz    %[width],      1b                       \n\t"
>+      "nop                                             \n\t"
>+      : [tr] "=&f"(tr), [tb] "=&f"(tb)
>+      : [src_sobelx] "r"(src_sobelx), [src_sobely] "r"(src_sobely),
>+        [dst_y] "r"(dst_y), [width] "r"(width)
>+      : "memory");
>+}
>+
>+void SobelXYRow_MMI(const uint8_t* src_sobelx,
>+                    const uint8_t* src_sobely,
>+                    uint8_t* dst_argb,
>+                    int width) {
>+  uint64_t temp[3];
>+  uint64_t result = 0;
>+  uint64_t gb = 0;
>+  uint64_t cr = 0;
>+  uint64_t c1 = 0xffffffffffffffff;
>+  __asm__ volatile(
>+      "1:	                                          \n\t"
>+      "gsldlc1   %[tr],         0x07(%[src_sobelx])       \n\t"  // r=src_sobelx[i]
>+      "gsldrc1   %[tr],         0x00(%[src_sobelx])       \n\t"
>+      "gsldlc1   %[tb],         0x07(%[src_sobely])       \n\t"  // b=src_sobely[i]
>+      "gsldrc1   %[tb],         0x00(%[src_sobely])       \n\t"
>+      "paddusb   %[tg] ,        %[tr],              %[tb] \n\t"  // g
>+
>+      // g3 b3 g2 b2 g1 b1 g0 b0
>+      "punpcklbh %[gb],         %[tb],              %[tg] \n\t"
>+      // c3 r3 r2 r2 c1 r1 c0 r0
>+      "punpcklbh %[cr],         %[tr],              %[c1] \n\t"
>+      // c1 r1 g1 b1 c0 r0 g0 b0
>+      "punpcklhw %[result],     %[gb],              %[cr] \n\t"
>+      "gssdrc1   %[result],     0x00(%[dst_argb])	  \n\t"
>+      "gssdlc1   %[result],     0x07(%[dst_argb])         \n\t"
>+      // c3 r3 g3 b3 c2 r2 g2 b2
>+      "punpckhhw %[result],     %[gb],              %[cr] \n\t"
>+      "gssdrc1   %[result],     0x08(%[dst_argb])	  \n\t"
>+      "gssdlc1   %[result],     0x0f(%[dst_argb])         \n\t"
>+
>+      // g7 b7 g6 b6 g5 b5 g4 b4
>+      "punpckhbh %[gb],         %[tb],              %[tg] \n\t"
>+      // c7 r7 c6 r6 c5 r5 c4 r4
>+      "punpckhbh %[cr],         %[tr],              %[c1] \n\t"
>+      // c5 r5 g5 b5 c4 r4 g4 b4
>+      "punpcklhw %[result],     %[gb],              %[cr] \n\t"
>+      "gssdrc1   %[result],     0x10(%[dst_argb])	  \n\t"
>+      "gssdlc1   %[result],     0x17(%[dst_argb])         \n\t"
>+      // c7 r7 g7 b7 c6 r6 g6 b6
>+      "punpckhhw %[result],     %[gb],              %[cr] \n\t"
>+      "gssdrc1   %[result],     0x18(%[dst_argb])	  \n\t"
>+      "gssdlc1   %[result],     0x1f(%[dst_argb])         \n\t"
>+
>+      "daddiu    %[dst_argb],   %[dst_argb],        32    \n\t"
>+      "daddiu    %[src_sobelx], %[src_sobelx],      8     \n\t"
>+      "daddiu    %[src_sobely], %[src_sobely],      8     \n\t"
>+      "daddiu    %[width],      %[width],          -8     \n\t"
>+      "bgtz      %[width],      1b                        \n\t"
>+      "nop                                                \n\t"
>+      : [tr] "=&f"(temp[0]), [tb] "=&f"(temp[1]), [tg] "=&f"(temp[2]),
>+        [gb] "=&f"(gb), [cr] "=&f"(cr), [result] "=&f"(result)
>+      : [src_sobelx] "r"(src_sobelx), [src_sobely] "r"(src_sobely),
>+        [dst_argb] "r"(dst_argb), [width] "r"(width), [c1] "f"(c1)
>+      : "memory");
>+}
>+
>+void J400ToARGBRow_MMI(const uint8_t* src_y, uint8_t* dst_argb, int width) {
>+  // Copy a Y to RGB.
>+  uint64_t src, dest;
>+  const uint64_t mask0 = 0x00ffffff00ffffffULL;
>+  const uint64_t mask1 = ~mask0;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gslwlc1    %[src],          0x03(%[src_ptr])                 \n\t"
>+      "gslwrc1    %[src],          0x00(%[src_ptr])                 \n\t"
>+      "punpcklbh  %[src],          %[src],            %[src]        \n\t"
>+      "punpcklhw  %[dest],         %[src],            %[src]        \n\t"
>+      "and        %[dest],         %[dest],           %[mask0]      \n\t"
>+      "or         %[dest],         %[dest],           %[mask1]      \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+
>+      "punpckhhw  %[dest],         %[src],            %[src]        \n\t"
>+      "and        %[dest],         %[dest],           %[mask0]      \n\t"
>+      "or         %[dest],         %[dest],           %[mask1]      \n\t"
>+      "gssdrc1    %[dest],         0x08(%[dst_ptr])                 \n\t"
>+      "gssdlc1    %[dest],         0x0f(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x04          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x10          \n\t"
>+      "daddi      %[width],        %[width],         -0x04          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src] "=&f"(src), [dest] "=&f"(dest)
>+      : [src_ptr] "r"(src_y), [dst_ptr] "r"(dst_argb), [mask0] "f"(mask0),
>+        [mask1] "f"(mask1), [width] "r"(width)
>+      : "memory");
>+}
>+
>+void I400ToARGBRow_MMI(const uint8_t* src_y, uint8_t* rgb_buf, int width) {
>+  uint64_t src, src_lo, src_hi, dest, dest_lo, dest_hi;
>+  const uint64_t mask0 = 0x0;
>+  const uint64_t mask1 = 0x55;
>+  const uint64_t mask2 = 0xAA;
>+  const uint64_t mask3 = 0xFF;
>+  const uint64_t mask4 = 0x4A354A354A354A35ULL;
>+  const uint64_t mask5 = 0x0488048804880488ULL;
>+  const uint64_t shift0 = 0x08;
>+  const uint64_t shift1 = 0x06;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[src],          0x07(%[src_ptr])                 \n\t"
>+      "gsldrc1    %[src],          0x00(%[src_ptr])                 \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[mask0]      \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[mask0]      \n\t"
>+
>+      "pshufh     %[src],          %[src_lo],         %[mask0]      \n\t"
>+      "psllh      %[dest_lo],      %[src],            %[shift0]     \n\t"
>+      "paddush    %[dest_lo],      %[dest_lo],        %[src]        \n\t"
>+      "pmulhuh    %[dest_lo],      %[dest_lo],        %[mask4]      \n\t"
>+      "psubh      %[dest_lo],      %[dest_lo],        %[mask5]      \n\t"
>+      "psrah      %[dest_lo],      %[dest_lo],        %[shift1]     \n\t"
>+      "pinsrh_3   %[dest_lo],      %[dest_lo],        %[mask3]      \n\t"
>+      "pshufh     %[src],          %[src_lo],         %[mask1]      \n\t"
>+      "psllh      %[dest_hi],      %[src],            %[shift0]     \n\t"
>+      "paddush    %[dest_hi],      %[dest_hi],        %[src]        \n\t"
>+      "pmulhuh    %[dest_hi],      %[dest_hi],        %[mask4]      \n\t"
>+      "psubh      %[dest_hi],      %[dest_hi],        %[mask5]      \n\t"
>+      "psrah      %[dest_hi],      %[dest_hi],        %[shift1]     \n\t"
>+      "pinsrh_3   %[dest_hi],      %[dest_hi],        %[mask3]      \n\t"
>+      "packushb   %[dest],         %[dest_lo],        %[dest_hi]    \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "pshufh     %[src],          %[src_lo],         %[mask2]      \n\t"
>+      "psllh      %[dest_lo],      %[src],            %[shift0]     \n\t"
>+      "paddush    %[dest_lo],      %[dest_lo],        %[src]        \n\t"
>+      "pmulhuh    %[dest_lo],      %[dest_lo],        %[mask4]      \n\t"
>+      "psubh      %[dest_lo],      %[dest_lo],        %[mask5]      \n\t"
>+      "psrah      %[dest_lo],      %[dest_lo],        %[shift1]     \n\t"
>+      "pinsrh_3   %[dest_lo],      %[dest_lo],        %[mask3]      \n\t"
>+      "pshufh     %[src],          %[src_lo],         %[mask3]      \n\t"
>+      "psllh      %[dest_hi],      %[src],            %[shift0]     \n\t"
>+      "paddush    %[dest_hi],      %[dest_hi],        %[src]        \n\t"
>+      "pmulhuh    %[dest_hi],      %[dest_hi],        %[mask4]      \n\t"
>+      "psubh      %[dest_hi],      %[dest_hi],        %[mask5]      \n\t"
>+      "psrah      %[dest_hi],      %[dest_hi],        %[shift1]     \n\t"
>+      "pinsrh_3   %[dest_hi],      %[dest_hi],        %[mask3]      \n\t"
>+      "packushb   %[dest],         %[dest_lo],        %[dest_hi]    \n\t"
>+      "gssdlc1    %[dest],         0x0f(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x08(%[dst_ptr])                 \n\t"
>+
>+      "pshufh     %[src],          %[src_hi],         %[mask0]      \n\t"
>+      "psllh      %[dest_lo],      %[src],            %[shift0]     \n\t"
>+      "paddush    %[dest_lo],      %[dest_lo],        %[src]        \n\t"
>+      "pmulhuh    %[dest_lo],      %[dest_lo],        %[mask4]      \n\t"
>+      "psubh      %[dest_lo],      %[dest_lo],        %[mask5]      \n\t"
>+      "psrah      %[dest_lo],      %[dest_lo],        %[shift1]     \n\t"
>+      "pinsrh_3   %[dest_lo],      %[dest_lo],        %[mask3]      \n\t"
>+      "pshufh     %[src],          %[src_hi],         %[mask1]      \n\t"
>+      "psllh      %[dest_hi],      %[src],            %[shift0]     \n\t"
>+      "paddush    %[dest_hi],      %[dest_hi],        %[src]        \n\t"
>+      "pmulhuh    %[dest_hi],      %[dest_hi],        %[mask4]      \n\t"
>+      "psubh      %[dest_hi],      %[dest_hi],        %[mask5]      \n\t"
>+      "psrah      %[dest_hi],      %[dest_hi],        %[shift1]     \n\t"
>+      "pinsrh_3   %[dest_hi],      %[dest_hi],        %[mask3]      \n\t"
>+      "packushb   %[dest],         %[dest_lo],        %[dest_hi]    \n\t"
>+      "gssdlc1    %[dest],         0x17(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x10(%[dst_ptr])                 \n\t"
>+
>+      "pshufh     %[src],          %[src_hi],         %[mask2]      \n\t"
>+      "psllh      %[dest_lo],      %[src],            %[shift0]     \n\t"
>+      "paddush    %[dest_lo],      %[dest_lo],        %[src]        \n\t"
>+      "pmulhuh    %[dest_lo],      %[dest_lo],        %[mask4]      \n\t"
>+      "psubh      %[dest_lo],      %[dest_lo],        %[mask5]      \n\t"
>+      "psrah      %[dest_lo],      %[dest_lo],        %[shift1]     \n\t"
>+      "pinsrh_3   %[dest_lo],      %[dest_lo],        %[mask3]      \n\t"
>+      "pshufh     %[src],          %[src_hi],         %[mask3]      \n\t"
>+      "psllh      %[dest_hi],      %[src],            %[shift0]     \n\t"
>+      "paddush    %[dest_hi],      %[dest_hi],        %[src]        \n\t"
>+      "pmulhuh    %[dest_hi],      %[dest_hi],        %[mask4]      \n\t"
>+      "psubh      %[dest_hi],      %[dest_hi],        %[mask5]      \n\t"
>+      "psrah      %[dest_hi],      %[dest_hi],        %[shift1]     \n\t"
>+      "pinsrh_3   %[dest_hi],      %[dest_hi],        %[mask3]      \n\t"
>+      "packushb   %[dest],         %[dest_lo],        %[dest_hi]    \n\t"
>+      "gssdlc1    %[dest],         0x1f(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x18(%[dst_ptr])                 \n\t"
>+
>+      "daddi      %[src_ptr],      %[src_ptr],        0x08          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x20          \n\t"
>+      "daddi      %[width],        %[width],         -0x08          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src] "=&f"(src), [dest] "=&f"(dest), [src_hi] "=&f"(src_hi),
>+        [src_lo] "=&f"(src_lo), [dest_hi] "=&f"(dest_hi),
>+        [dest_lo] "=&f"(dest_lo)
>+      : [src_ptr] "r"(src_y), [dst_ptr] "r"(rgb_buf), [mask0] "f"(mask0),
>+        [mask1] "f"(mask1), [mask2] "f"(mask2), [mask3] "f"(mask3),
>+        [mask4] "f"(mask4), [mask5] "f"(mask5), [shift0] "f"(shift0),
>+        [shift1] "f"(shift1), [width] "r"(width)
>+      : "memory");
>+}
>+
>+void MirrorRow_MMI(const uint8_t* src, uint8_t* dst, int width) {
>+  uint64_t source, src0, src1, dest;
>+  const uint64_t mask0 = 0x0;
>+  const uint64_t mask1 = 0x1b;
>+
>+  src += width - 1;
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[source],       0(%[src_ptr])                    \n\t"
>+      "gsldrc1    %[source],       -7(%[src_ptr])                   \n\t"
>+      "punpcklbh  %[src0],         %[source],         %[mask0]      \n\t"
>+      "pshufh     %[src0],         %[src0],           %[mask1]      \n\t"
>+      "punpckhbh  %[src1],         %[source],         %[mask0]      \n\t"
>+      "pshufh     %[src1],         %[src1],           %[mask1]      \n\t"
>+      "packushb   %[dest],         %[src1],           %[src0]       \n\t"
>+
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddi      %[src_ptr],      %[src_ptr],       -0x08          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x08          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [source] "=&f"(source), [dest] "=&f"(dest), [src0] "=&f"(src0),
>+        [src1] "=&f"(src1)
>+      : [src_ptr] "r"(src), [dst_ptr] "r"(dst), [mask0] "f"(mask0),
>+        [mask1] "f"(mask1), [width] "r"(width)
>+      : "memory");
>+}
>+
>+void MirrorUVRow_MMI(const uint8_t* src_uv,
>+                     uint8_t* dst_u,
>+                     uint8_t* dst_v,
>+                     int width) {
>+  uint64_t src0, src1, dest0, dest1;
>+  const uint64_t mask0 = 0x00ff00ff00ff00ffULL;
>+  const uint64_t mask1 = 0x1b;
>+  const uint64_t shift = 0x08;
>+
>+  src_uv += (width - 1) << 1;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[src0],         1(%[src_ptr])                    \n\t"
>+      "gsldrc1    %[src0],         -6(%[src_ptr])                   \n\t"
>+      "gsldlc1    %[src1],         -7(%[src_ptr])                   \n\t"
>+      "gsldrc1    %[src1],         -14(%[src_ptr])                  \n\t"
>+
>+      "and        %[dest0],        %[src0],           %[mask0]      \n\t"
>+      "pshufh     %[dest0],        %[dest0],          %[mask1]      \n\t"
>+      "and        %[dest1],        %[src1],           %[mask0]      \n\t"
>+      "pshufh     %[dest1],        %[dest1],          %[mask1]      \n\t"
>+      "packushb   %[dest0],        %[dest0],          %[dest1]      \n\t"
>+      "gssdlc1    %[dest0],        0x07(%[dstu_ptr])                \n\t"
>+      "gssdrc1    %[dest0],        0x00(%[dstu_ptr])                \n\t"
>+
>+      "psrlh      %[dest0],        %[src0],           %[shift]      \n\t"
>+      "pshufh     %[dest0],        %[dest0],          %[mask1]      \n\t"
>+      "psrlh      %[dest1],        %[src1],           %[shift]      \n\t"
>+      "pshufh     %[dest1],        %[dest1],          %[mask1]      \n\t"
>+      "packushb   %[dest0],        %[dest0],          %[dest1]      \n\t"
>+      "gssdlc1    %[dest0],        0x07(%[dstv_ptr])                \n\t"
>+      "gssdrc1    %[dest0],        0x00(%[dstv_ptr])                \n\t"
>+
>+      "daddi      %[src_ptr],      %[src_ptr],       -0x10          \n\t"
>+      "daddiu     %[dstu_ptr],     %[dstu_ptr],       0x08          \n\t"
>+      "daddiu     %[dstv_ptr],     %[dstv_ptr],       0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x08          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [dest0] "=&f"(dest0), [dest1] "=&f"(dest1), [src0] "=&f"(src0),
>+        [src1] "=&f"(src1)
>+      : [src_ptr] "r"(src_uv), [dstu_ptr] "r"(dst_u), [dstv_ptr] "r"(dst_v),
>+        [width] "r"(width), [mask0] "f"(mask0), [mask1] "f"(mask1),
>+        [shift] "f"(shift)
>+      : "memory");
>+}
>+
>+void ARGBMirrorRow_MMI(const uint8_t* src, uint8_t* dst, int width) {
>+  src += (width - 1) * 4;
>+  uint64_t temp = 0x0;
>+  uint64_t shuff = 0x4e;  // 01 00 11 10
>+  __asm__ volatile(
>+      "1:                                      \n\t"
>+      "gsldlc1 %[temp],  3(%[src])     	       \n\t"
>+      "gsldrc1 %[temp], -4(%[src])     	       \n\t"
>+      "pshufh  %[temp],  %[temp],    %[shuff]  \n\t"
>+      "gssdrc1 %[temp],  0x0(%[dst])           \n\t"
>+      "gssdlc1 %[temp],  0x7(%[dst])           \n\t"
>+
>+      "daddiu  %[src],   %[src],    -0x08      \n\t"
>+      "daddiu  %[dst],   %[dst],     0x08      \n\t"
>+      "daddiu  %[width], %[width],  -0x02      \n\t"
>+      "bnez    %[width], 1b                    \n\t"
>+      : [temp] "=&f"(temp)
>+      : [src] "r"(src), [dst] "r"(dst), [width] "r"(width), [shuff] "f"(shuff)
>+      : "memory");
>+}
>+
>+void SplitUVRow_MMI(const uint8_t* src_uv,
>+                    uint8_t* dst_u,
>+                    uint8_t* dst_v,
>+                    int width) {
>+  uint64_t c0 = 0x00ff00ff00ff00ff;
>+  uint64_t temp[4];
>+  uint64_t shift = 0x08;
>+  __asm__ volatile(
>+      "1:	                                    \n\t"
>+      "gsldrc1  %[t0],     0x00(%[src_uv])          \n\t"
>+      "gsldlc1  %[t0],     0x07(%[src_uv])          \n\t"
>+      "gsldrc1  %[t1],     0x08(%[src_uv])          \n\t"
>+      "gsldlc1  %[t1],     0x0f(%[src_uv])          \n\t"
>+
>+      "and      %[t2],     %[t0],          %[c0]    \n\t"
>+      "and      %[t3],     %[t1],          %[c0]    \n\t"
>+      "packushb %[t2],     %[t2],          %[t3]    \n\t"
>+      "gssdrc1  %[t2],     0x0(%[dst_u])	    \n\t"
>+      "gssdlc1  %[t2],     0x7(%[dst_u])            \n\t"
>+
>+      "psrlh    %[t2],     %[t0],          %[shift] \n\t"
>+      "psrlh    %[t3],     %[t1],          %[shift] \n\t"
>+      "packushb %[t2],     %[t2],          %[t3]    \n\t"
>+      "gssdrc1  %[t2],     0x0(%[dst_v])            \n\t"
>+      "gssdlc1  %[t2],     0x7(%[dst_v])            \n\t"
>+
>+      "daddiu   %[src_uv], %[src_uv],      16       \n\t"
>+      "daddiu   %[dst_u],  %[dst_u],       8        \n\t"
>+      "daddiu   %[dst_v],  %[dst_v],       8        \n\t"
>+      "daddiu   %[width],  %[width],      -8        \n\t"
>+      "bgtz     %[width],  1b                       \n\t"
>+      "nop                                          \n\t"
>+      : [t0] "=&f"(temp[0]), [t1] "=&f"(temp[1]), [t2] "=&f"(temp[2]),
>+        [t3] "=&f"(temp[3])
>+      : [src_uv] "r"(src_uv), [dst_u] "r"(dst_u), [dst_v] "r"(dst_v),
>+        [width] "r"(width), [c0] "f"(c0), [shift] "f"(shift)
>+      : "memory");
>+}
>+
>+void MergeUVRow_MMI(const uint8_t* src_u,
>+                    const uint8_t* src_v,
>+                    uint8_t* dst_uv,
>+                    int width) {
>+  uint64_t temp[3];
>+  __asm__ volatile(
>+      "1:	                                 \n\t"
>+      "gsldrc1   %[t0],     0x0(%[src_u])        \n\t"
>+      "gsldlc1   %[t0],     0x7(%[src_u])        \n\t"
>+      "gsldrc1   %[t1],     0x0(%[src_v])        \n\t"
>+      "gsldlc1   %[t1],     0x7(%[src_v])        \n\t"
>+      "punpcklbh %[t2],     %[t0],         %[t1] \n\t"
>+      "gssdrc1   %[t2],     0x0(%[dst_uv])	 \n\t"
>+      "gssdlc1   %[t2],     0x7(%[dst_uv])       \n\t"
>+      "punpckhbh %[t2],     %[t0],         %[t1] \n\t"
>+      "gssdrc1   %[t2],     0x8(%[dst_uv])	 \n\t"
>+      "gssdlc1   %[t2],     0xf(%[dst_uv])       \n\t"
>+
>+      "daddiu    %[src_u],  %[src_u],      8     \n\t"
>+      "daddiu    %[src_v],  %[src_v],      8     \n\t"
>+      "daddiu    %[dst_uv], %[dst_uv],     16    \n\t"
>+      "daddiu    %[width],  %[width],     -8     \n\t"
>+      "bgtz      %[width],  1b                   \n\t"
>+      "nop                                       \n\t"
>+      : [t0] "=&f"(temp[0]), [t1] "=&f"(temp[1]), [t2] "=&f"(temp[2])
>+      : [dst_uv] "r"(dst_uv), [src_u] "r"(src_u), [src_v] "r"(src_v),
>+        [width] "r"(width)
>+      : "memory");
>+}
>+
>+void SplitRGBRow_MMI(const uint8_t* src_rgb,
>+                     uint8_t* dst_r,
>+                     uint8_t* dst_g,
>+                     uint8_t* dst_b,
>+                     int width) {
>+  uint64_t src[4];
>+  uint64_t dest_hi, dest_lo, dest;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gslwlc1    %[src0],         0x03(%[src_ptr])                 \n\t"
>+      "gslwrc1    %[src0],         0x00(%[src_ptr])                 \n\t"
>+      "gslwlc1    %[src1],         0x06(%[src_ptr])                 \n\t"
>+      "gslwrc1    %[src1],         0x03(%[src_ptr])                 \n\t"
>+      "punpcklbh  %[dest_lo],      %[src0],           %[src1]       \n\t"
>+      "gslwlc1    %[src2],         0x09(%[src_ptr])                 \n\t"
>+      "gslwrc1    %[src2],         0x06(%[src_ptr])                 \n\t"
>+      "gslwlc1    %[src3],         0x0c(%[src_ptr])                 \n\t"
>+      "gslwrc1    %[src3],         0x09(%[src_ptr])                 \n\t"
>+      "punpcklbh  %[dest_hi],      %[src2],           %[src3]       \n\t"
>+
>+      "punpcklhw  %[dest],         %[dest_lo],        %[dest_hi]    \n\t"
>+      "gsswlc1    %[dest],         0x03(%[dstr_ptr])                \n\t"
>+      "gsswrc1    %[dest],         0x00(%[dstr_ptr])                \n\t"
>+      "punpckhwd  %[dest],         %[dest],           %[dest]       \n\t"
>+      "gsswlc1    %[dest],         0x03(%[dstg_ptr])                \n\t"
>+      "gsswrc1    %[dest],         0x00(%[dstg_ptr])                \n\t"
>+      "punpckhhw  %[dest],         %[dest_lo],        %[dest_hi]    \n\t"
>+      "gsswlc1    %[dest],         0x03(%[dstb_ptr])                \n\t"
>+      "gsswrc1    %[dest],         0x00(%[dstb_ptr])                \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x0c          \n\t"
>+      "daddiu     %[dstr_ptr],     %[dstr_ptr],       0x04          \n\t"
>+      "daddiu     %[dstg_ptr],     %[dstg_ptr],       0x04          \n\t"
>+      "daddiu     %[dstb_ptr],     %[dstb_ptr],       0x04          \n\t"
>+      "daddi      %[width],        %[width],         -0x04          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src0] "=&f"(src[0]), [src1] "=&f"(src[1]), [src2] "=&f"(src[2]),
>+        [src3] "=&f"(src[3]), [dest_hi] "=&f"(dest_hi),
>+        [dest_lo] "=&f"(dest_lo), [dest] "=&f"(dest)
>+      : [src_ptr] "r"(src_rgb), [dstr_ptr] "r"(dst_r), [dstg_ptr] "r"(dst_g),
>+        [dstb_ptr] "r"(dst_b), [width] "r"(width)
>+      : "memory");
>+}
>+
>+void MergeRGBRow_MMI(const uint8_t* src_r,
>+                     const uint8_t* src_g,
>+                     const uint8_t* src_b,
>+                     uint8_t* dst_rgb,
>+                     int width) {
>+  uint64_t srcr, srcg, srcb, dest;
>+  uint64_t srcrg_hi, srcrg_lo, srcbz_hi, srcbz_lo;
>+  const uint64_t temp = 0x0;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[srcr],         0x07(%[srcr_ptr])                \n\t"
>+      "gsldrc1    %[srcr],         0x00(%[srcr_ptr])                \n\t"
>+      "gsldlc1    %[srcg],         0x07(%[srcg_ptr])                \n\t"
>+      "gsldrc1    %[srcg],         0x00(%[srcg_ptr])                \n\t"
>+      "punpcklbh  %[srcrg_lo],     %[srcr],           %[srcg]       \n\t"
>+      "punpckhbh  %[srcrg_hi],     %[srcr],           %[srcg]       \n\t"
>+
>+      "gsldlc1    %[srcb],         0x07(%[srcb_ptr])                \n\t"
>+      "gsldrc1    %[srcb],         0x00(%[srcb_ptr])                \n\t"
>+      "punpcklbh  %[srcbz_lo],     %[srcb],           %[temp]       \n\t"
>+      "punpckhbh  %[srcbz_hi],     %[srcb],           %[temp]       \n\t"
>+
>+      "punpcklhw  %[dest],         %[srcrg_lo],       %[srcbz_lo]   \n\t"
>+      "gsswlc1    %[dest],         0x03(%[dst_ptr])                 \n\t"
>+      "gsswrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+      "punpckhwd  %[dest],         %[dest],           %[dest]       \n\t"
>+      "gsswlc1    %[dest],         0x06(%[dst_ptr])                 \n\t"
>+      "gsswrc1    %[dest],         0x03(%[dst_ptr])                 \n\t"
>+      "punpckhhw  %[dest],         %[srcrg_lo],       %[srcbz_lo]   \n\t"
>+      "gsswlc1    %[dest],         0x09(%[dst_ptr])                 \n\t"
>+      "gsswrc1    %[dest],         0x06(%[dst_ptr])                 \n\t"
>+      "punpckhwd  %[dest],         %[dest],           %[dest]       \n\t"
>+      "gsswlc1    %[dest],         0x0c(%[dst_ptr])                 \n\t"
>+      "gsswrc1    %[dest],         0x09(%[dst_ptr])                 \n\t"
>+      "punpcklhw  %[dest],         %[srcrg_hi],       %[srcbz_hi]   \n\t"
>+      "gsswlc1    %[dest],         0x0f(%[dst_ptr])                 \n\t"
>+      "gsswrc1    %[dest],         0x0c(%[dst_ptr])                 \n\t"
>+      "punpckhwd  %[dest],         %[dest],           %[dest]       \n\t"
>+      "gsswlc1    %[dest],         0x12(%[dst_ptr])                 \n\t"
>+      "gsswrc1    %[dest],         0x0f(%[dst_ptr])                 \n\t"
>+      "punpckhhw  %[dest],         %[srcrg_hi],       %[srcbz_hi]   \n\t"
>+      "gsswlc1    %[dest],         0x15(%[dst_ptr])                 \n\t"
>+      "gsswrc1    %[dest],         0x12(%[dst_ptr])                 \n\t"
>+      "punpckhwd  %[dest],         %[dest],           %[dest]       \n\t"
>+      "gsswlc1    %[dest],         0x18(%[dst_ptr])                 \n\t"
>+      "gsswrc1    %[dest],         0x15(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[srcr_ptr],     %[srcr_ptr],       0x08          \n\t"
>+      "daddiu     %[srcg_ptr],     %[srcg_ptr],       0x08          \n\t"
>+      "daddiu     %[srcb_ptr],     %[srcb_ptr],       0x08          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x18          \n\t"
>+      "daddi      %[width],        %[width],         -0x08          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [srcr] "=&f"(srcr), [srcg] "=&f"(srcg), [srcb] "=&f"(srcb),
>+        [dest] "=&f"(dest), [srcrg_hi] "=&f"(srcrg_hi),
>+        [srcrg_lo] "=&f"(srcrg_lo), [srcbz_hi] "=&f"(srcbz_hi),
>+        [srcbz_lo] "=&f"(srcbz_lo)
>+      : [srcr_ptr] "r"(src_r), [srcg_ptr] "r"(src_g), [srcb_ptr] "r"(src_b),
>+        [dst_ptr] "r"(dst_rgb), [width] "r"(width), [temp] "f"(temp)
>+      : "memory");
>+}
>+
>+// Filter 2 rows of YUY2 UV's (422) into U and V (420).
>+void YUY2ToUVRow_MMI(const uint8_t* src_yuy2,
>+                     int src_stride_yuy2,
>+                     uint8_t* dst_u,
>+                     uint8_t* dst_v,
>+                     int width) {
>+  uint64_t c0 = 0xff00ff00ff00ff00;
>+  uint64_t c1 = 0x00ff00ff00ff00ff;
>+  uint64_t temp[3];
>+  uint64_t data[4];
>+  uint64_t shift = 0x08;
>+  uint64_t src_stride = 0x0;
>+  __asm__ volatile(
>+      "1:	                                                     \n\t"
>+      "gsldrc1  %[t0],         0x00(%[src_yuy2])                     \n\t"
>+      "gsldlc1  %[t0],         0x07(%[src_yuy2])                     \n\t"
>+      "daddu    %[src_stride], %[src_yuy2],       %[src_stride_yuy2] \n\t"
>+      "gsldrc1  %[t1],         0x00(%[src_stride])                   \n\t"
>+      "gsldlc1  %[t1],         0x07(%[src_stride])                   \n\t"
>+      "pavgb    %[t0],         %[t0],             %[t1]              \n\t"
>+
>+      "gsldrc1  %[t2],         0x08(%[src_yuy2])                     \n\t"
>+      "gsldlc1  %[t2],         0x0f(%[src_yuy2])                     \n\t"
>+      "gsldrc1  %[t1],         0x08(%[src_stride])                   \n\t"
>+      "gsldlc1  %[t1],         0x0f(%[src_stride])                   \n\t"
>+      "pavgb    %[t1],         %[t2],             %[t1]              \n\t"
>+
>+      "and      %[t0],         %[t0],             %[c0]              \n\t"
>+      "and      %[t1],         %[t1],             %[c0]              \n\t"
>+      "psrlh    %[t0],         %[t0],             %[shift]           \n\t"
>+      "psrlh    %[t1],         %[t1],             %[shift]           \n\t"
>+      "packushb %[t0],         %[t0],             %[t1]              \n\t"
>+      "mov.s    %[t1],         %[t0]                                 \n\t"
>+      "and      %[d0],         %[t0],             %[c1]              \n\t"
>+      "psrlh    %[d1],         %[t1],             %[shift]           \n\t"
>+
>+      "gsldrc1  %[t0],         0x10(%[src_yuy2])                     \n\t"
>+      "gsldlc1  %[t0],         0x17(%[src_yuy2])                     \n\t"
>+      "gsldrc1  %[t1],         0x10(%[src_stride])                   \n\t"
>+      "gsldlc1  %[t1],         0x17(%[src_stride])                   \n\t"
>+      "pavgb    %[t0],         %[t0],              %[t1]             \n\t"
>+
>+      "gsldrc1  %[t2],         0x18(%[src_yuy2])                     \n\t"
>+      "gsldlc1  %[t2],         0x1f(%[src_yuy2])                     \n\t"
>+      "gsldrc1  %[t1],         0x18(%[src_stride])                   \n\t"
>+      "gsldlc1  %[t1],         0x1f(%[src_stride])                   \n\t"
>+      "pavgb    %[t1],         %[t2],              %[t1]             \n\t"
>+
>+      "and      %[t0],         %[t0],              %[c0]             \n\t"
>+      "and      %[t1],         %[t1],              %[c0]             \n\t"
>+      "psrlh    %[t0],         %[t0],              %[shift]          \n\t"
>+      "psrlh    %[t1],         %[t1],              %[shift]          \n\t"
>+      "packushb %[t0],         %[t0],              %[t1]             \n\t"
>+      "mov.s    %[t1],         %[t0]                                 \n\t"
>+      "and      %[d2],         %[t0],              %[c1]             \n\t"
>+      "psrlh    %[d3],         %[t1],              %[shift]          \n\t"
>+
>+      "packushb %[d0],         %[d0],              %[d2]             \n\t"
>+      "packushb %[d1],         %[d1],              %[d3]             \n\t"
>+      "gssdrc1  %[d0],         0x0(%[dst_u])	                     \n\t"
>+      "gssdlc1  %[d0],         0x7(%[dst_u])                         \n\t"
>+      "gssdrc1  %[d1],         0x0(%[dst_v])	                     \n\t"
>+      "gssdlc1  %[d1],         0x7(%[dst_v])                         \n\t"
>+      "daddiu   %[src_yuy2],   %[src_yuy2],        32                \n\t"
>+      "daddiu   %[dst_u],      %[dst_u],           8                 \n\t"
>+      "daddiu   %[dst_v],      %[dst_v],           8                 \n\t"
>+      "daddiu   %[width],      %[width],          -16                \n\t"
>+      "bgtz     %[width],      1b                                    \n\t"
>+      "nop                                                           \n\t"
>+      : [t0] "=&f"(temp[0]), [t1] "=&f"(temp[1]), [t2] "=&f"(temp[2]),
>+        [d0] "=&f"(data[0]), [d1] "=&f"(data[1]), [d2] "=&f"(data[2]),
>+        [d3] "=&f"(data[3]), [src_stride] "=&r"(src_stride)
>+      : [src_yuy2] "r"(src_yuy2), [src_stride_yuy2] "r"(src_stride_yuy2),
>+        [dst_u] "r"(dst_u), [dst_v] "r"(dst_v), [width] "r"(width),
>+        [c0] "f"(c0), [c1] "f"(c1), [shift] "f"(shift)
>+      : "memory");
>+}
>+
>+// Copy row of YUY2 UV's (422) into U and V (422).
>+void YUY2ToUV422Row_MMI(const uint8_t* src_yuy2,
>+                        uint8_t* dst_u,
>+                        uint8_t* dst_v,
>+                        int width) {
>+  uint64_t c0 = 0xff00ff00ff00ff00;
>+  uint64_t c1 = 0x00ff00ff00ff00ff;
>+  uint64_t temp[2];
>+  uint64_t data[4];
>+  uint64_t shift = 0x08;
>+  __asm__ volatile(
>+      "1:	                                        \n\t"
>+      "gsldrc1  %[t0],       0x00(%[src_yuy2])          \n\t"
>+      "gsldlc1  %[t0],       0x07(%[src_yuy2])          \n\t"
>+      "gsldrc1  %[t1],       0x08(%[src_yuy2])          \n\t"
>+      "gsldlc1  %[t1],       0x0f(%[src_yuy2])          \n\t"
>+      "and      %[t0],       %[t0],            %[c0]    \n\t"
>+      "and      %[t1],       %[t1],            %[c0]    \n\t"
>+      "psrlh    %[t0],       %[t0],            %[shift] \n\t"
>+      "psrlh    %[t1],       %[t1],            %[shift] \n\t"
>+      "packushb %[t0],       %[t0],            %[t1]    \n\t"
>+      "mov.s    %[t1],       %[t0]                      \n\t"
>+      "and      %[d0],       %[t0],            %[c1]    \n\t"
>+      "psrlh    %[d1],       %[t1],            %[shift] \n\t"
>+
>+      "gsldrc1  %[t0],       0x10(%[src_yuy2])          \n\t"
>+      "gsldlc1  %[t0],       0x17(%[src_yuy2])          \n\t"
>+      "gsldrc1  %[t1],       0x18(%[src_yuy2])          \n\t"
>+      "gsldlc1  %[t1],       0x1f(%[src_yuy2])          \n\t"
>+      "and      %[t0],       %[t0],            %[c0]    \n\t"
>+      "and      %[t1],       %[t1],            %[c0]    \n\t"
>+      "psrlh    %[t0],       %[t0],            %[shift] \n\t"
>+      "psrlh    %[t1],       %[t1],            %[shift] \n\t"
>+      "packushb %[t0],       %[t0],            %[t1]    \n\t"
>+      "mov.s    %[t1],       %[t0]                      \n\t"
>+      "and      %[d2],       %[t0],            %[c1]    \n\t"
>+      "psrlh    %[d3],       %[t1],            %[shift] \n\t"
>+
>+      "packushb %[d0],       %[d0],            %[d2]    \n\t"
>+      "packushb %[d1],       %[d1],            %[d3]    \n\t"
>+      "gssdrc1  %[d0],       0x0(%[dst_u])	        \n\t"
>+      "gssdlc1  %[d0],       0x7(%[dst_u])              \n\t"
>+      "gssdrc1  %[d1],       0x0(%[dst_v])	        \n\t"
>+      "gssdlc1  %[d1],       0x7(%[dst_v])              \n\t"
>+      "daddiu   %[src_yuy2], %[src_yuy2],      32       \n\t"
>+      "daddiu   %[dst_u],    %[dst_u],         8        \n\t"
>+      "daddiu   %[dst_v],    %[dst_v],         8        \n\t"
>+      "daddiu   %[width],    %[width],        -16       \n\t"
>+      "bgtz     %[width],    1b                         \n\t"
>+      "nop                                              \n\t"
>+      : [t0] "=&f"(temp[0]), [t1] "=&f"(temp[1]), [d0] "=&f"(data[0]),
>+        [d1] "=&f"(data[1]), [d2] "=&f"(data[2]), [d3] "=&f"(data[3])
>+      : [src_yuy2] "r"(src_yuy2), [dst_u] "r"(dst_u), [dst_v] "r"(dst_v),
>+        [width] "r"(width), [c0] "f"(c0), [c1] "f"(c1), [shift] "f"(shift)
>+      : "memory");
>+}
>+
>+// Copy row of YUY2 Y's (422) into Y (420/422).
>+void YUY2ToYRow_MMI(const uint8_t* src_yuy2, uint8_t* dst_y, int width) {
>+  uint64_t c0 = 0x00ff00ff00ff00ff;
>+  uint64_t temp[2];
>+  __asm__ volatile(
>+      "1:	                                     \n\t"
>+      "gsldrc1  %[t0],       0x00(%[src_yuy2])       \n\t"
>+      "gsldlc1  %[t0],       0x07(%[src_yuy2])       \n\t"
>+      "gsldrc1  %[t1],       0x08(%[src_yuy2])       \n\t"
>+      "gsldlc1  %[t1],       0x0f(%[src_yuy2])       \n\t"
>+      "and      %[t0],       %[t0],            %[c0] \n\t"
>+      "and      %[t1],       %[t1],            %[c0] \n\t"
>+      "packushb %[t0],       %[t0],            %[t1] \n\t"
>+      "gssdrc1  %[t0],       0x0(%[dst_y])	     \n\t"
>+      "gssdlc1  %[t0],       0x7(%[dst_y])           \n\t"
>+      "daddiu   %[src_yuy2], %[src_yuy2],      16    \n\t"
>+      "daddiu   %[dst_y],    %[dst_y],         8     \n\t"
>+      "daddiu   %[width],    %[width],        -8     \n\t"
>+      "bgtz     %[width],    1b                      \n\t"
>+      "nop                                           \n\t"
>+      : [t0] "=&f"(temp[0]), [t1] "=&f"(temp[1])
>+      : [src_yuy2] "r"(src_yuy2), [dst_y] "r"(dst_y), [width] "r"(width),
>+        [c0] "f"(c0)
>+      : "memory");
>+}
>+
>+// Filter 2 rows of UYVY UV's (422) into U and V (420).
>+void UYVYToUVRow_MMI(const uint8_t* src_uyvy,
>+                     int src_stride_uyvy,
>+                     uint8_t* dst_u,
>+                     uint8_t* dst_v,
>+                     int width) {
>+  // Output a row of UV values.
>+  uint64_t c0 = 0x00ff00ff00ff00ff;
>+  uint64_t temp[3];
>+  uint64_t data[4];
>+  uint64_t shift = 0x08;
>+  uint64_t src_stride = 0x0;
>+  __asm__ volatile(
>+      "1:	                                                      \n\t"
>+      "gsldrc1  %[t0],         0x00(%[src_uyvy])                      \n\t"
>+      "gsldlc1  %[t0],         0x07(%[src_uyvy])                      \n\t"
>+      "daddu    %[src_stride], %[src_uyvy],        %[src_stride_uyvy] \n\t"
>+      "gsldrc1  %[t1],         0x00(%[src_stride])                    \n\t"
>+      "gsldlc1  %[t1],         0x07(%[src_stride])                    \n\t"
>+      "pavgb    %[t0],         %[t0],              %[t1]              \n\t"
>+
>+      "gsldrc1  %[t2],         0x08(%[src_uyvy])                      \n\t"
>+      "gsldlc1  %[t2],         0x0f(%[src_uyvy])                      \n\t"
>+      "gsldrc1  %[t1],         0x08(%[src_stride])                    \n\t"
>+      "gsldlc1  %[t1],         0x0f(%[src_stride])                    \n\t"
>+      "pavgb    %[t1],         %[t2],              %[t1]              \n\t"
>+
>+      "and      %[t0],         %[t0],              %[c0]              \n\t"
>+      "and      %[t1],         %[t1],              %[c0]              \n\t"
>+      "packushb %[t0],         %[t0],              %[t1]              \n\t"
>+      "mov.s    %[t1],         %[t0]                                  \n\t"
>+      "and      %[d0],         %[t0],              %[c0]              \n\t"
>+      "psrlh    %[d1],         %[t1],              %[shift]           \n\t"
>+
>+      "gsldrc1  %[t0],         0x10(%[src_uyvy])                      \n\t"
>+      "gsldlc1  %[t0],         0x17(%[src_uyvy])                      \n\t"
>+      "gsldrc1  %[t1],         0x10(%[src_stride])                    \n\t"
>+      "gsldlc1  %[t1],         0x17(%[src_stride])                    \n\t"
>+      "pavgb    %[t0],         %[t0],              %[t1]              \n\t"
>+
>+      "gsldrc1  %[t2],         0x18(%[src_uyvy])                      \n\t"
>+      "gsldlc1  %[t2],         0x1f(%[src_uyvy])                      \n\t"
>+      "gsldrc1  %[t1],         0x18(%[src_stride])                    \n\t"
>+      "gsldlc1  %[t1],         0x1f(%[src_stride])                    \n\t"
>+      "pavgb    %[t1],         %[t2],              %[t1]              \n\t"
>+
>+      "and      %[t0],         %[t0],              %[c0]              \n\t"
>+      "and      %[t1],         %[t1],              %[c0]              \n\t"
>+      "packushb %[t0],         %[t0],              %[t1]              \n\t"
>+      "mov.s    %[t1],         %[t0]                                  \n\t"
>+      "and      %[d2],         %[t0],              %[c0]              \n\t"
>+      "psrlh    %[d3],         %[t1],              %[shift]           \n\t"
>+
>+      "packushb %[d0],         %[d0],              %[d2]              \n\t"
>+      "packushb %[d1],         %[d1],              %[d3]              \n\t"
>+      "gssdrc1  %[d0],         0x0(%[dst_u])	                      \n\t"
>+      "gssdlc1  %[d0],         0x7(%[dst_u])                          \n\t"
>+      "gssdrc1  %[d1],         0x0(%[dst_v])	                      \n\t"
>+      "gssdlc1  %[d1],         0x7(%[dst_v])                          \n\t"
>+      "daddiu   %[src_uyvy],   %[src_uyvy],        32                 \n\t"
>+      "daddiu   %[dst_u],      %[dst_u],           8                  \n\t"
>+      "daddiu   %[dst_v],      %[dst_v],           8                  \n\t"
>+      "daddiu   %[width],      %[width],          -16                 \n\t"
>+      "bgtz     %[width],      1b                                     \n\t"
>+      "nop                                                            \n\t"
>+      : [t0] "=&f"(temp[0]), [t1] "=&f"(temp[1]), [t2] "=&f"(temp[2]),
>+        [d0] "=&f"(data[0]), [d1] "=&f"(data[1]), [d2] "=&f"(data[2]),
>+        [d3] "=&f"(data[3]), [src_stride] "=&r"(src_stride)
>+      : [src_uyvy] "r"(src_uyvy), [src_stride_uyvy] "r"(src_stride_uyvy),
>+        [dst_u] "r"(dst_u), [dst_v] "r"(dst_v), [width] "r"(width),
>+        [c0] "f"(c0), [shift] "f"(shift)
>+      : "memory");
>+}
>+
>+// Copy row of UYVY UV's (422) into U and V (422).
>+void UYVYToUV422Row_MMI(const uint8_t* src_uyvy,
>+                        uint8_t* dst_u,
>+                        uint8_t* dst_v,
>+                        int width) {
>+  // Output a row of UV values.
>+  uint64_t c0 = 0x00ff00ff00ff00ff;
>+  uint64_t temp[2];
>+  uint64_t data[4];
>+  uint64_t shift = 0x08;
>+  __asm__ volatile(
>+      "1:	                                        \n\t"
>+      "gsldrc1  %[t0],       0x00(%[src_uyvy])          \n\t"
>+      "gsldlc1  %[t0],       0x07(%[src_uyvy])          \n\t"
>+      "gsldrc1  %[t1],       0x08(%[src_uyvy])          \n\t"
>+      "gsldlc1  %[t1],       0x0f(%[src_uyvy])          \n\t"
>+      "and      %[t0],       %[t0],            %[c0]    \n\t"
>+      "and      %[t1],       %[t1],            %[c0]    \n\t"
>+      "packushb %[t0],       %[t0],            %[t1]    \n\t"
>+      "mov.s    %[t1],       %[t0]                      \n\t"
>+      "and      %[d0],       %[t0],            %[c0]    \n\t"
>+      "psrlh    %[d1],       %[t1],            %[shift] \n\t"
>+
>+      "gsldrc1  %[t0],       0x10(%[src_uyvy])          \n\t"
>+      "gsldlc1  %[t0],       0x17(%[src_uyvy])          \n\t"
>+      "gsldrc1  %[t1],       0x18(%[src_uyvy])          \n\t"
>+      "gsldlc1  %[t1],       0x1f(%[src_uyvy])          \n\t"
>+      "and      %[t0],       %[t0],            %[c0]    \n\t"
>+      "and      %[t1],       %[t1],            %[c0]    \n\t"
>+      "packushb %[t0],       %[t0],            %[t1]    \n\t"
>+      "mov.s    %[t1],       %[t0]                      \n\t"
>+      "and      %[d2],       %[t0],            %[c0]    \n\t"
>+      "psrlh    %[d3],       %[t1],            %[shift] \n\t"
>+
>+      "packushb %[d0],       %[d0],            %[d2]    \n\t"
>+      "packushb %[d1],       %[d1],            %[d3]    \n\t"
>+      "gssdrc1  %[d0],       0x0(%[dst_u])	        \n\t"
>+      "gssdlc1  %[d0],       0x7(%[dst_u])              \n\t"
>+      "gssdrc1  %[d1],       0x0(%[dst_v])	        \n\t"
>+      "gssdlc1  %[d1],       0x7(%[dst_v])              \n\t"
>+      "daddiu   %[src_uyvy], %[src_uyvy],      32       \n\t"
>+      "daddiu   %[dst_u],    %[dst_u],         8        \n\t"
>+      "daddiu   %[dst_v],    %[dst_v],         8        \n\t"
>+      "daddiu   %[width],    %[width],        -16       \n\t"
>+      "bgtz     %[width],    1b                         \n\t"
>+      "nop                                              \n\t"
>+      : [t0] "=&f"(temp[0]), [t1] "=&f"(temp[1]), [d0] "=&f"(data[0]),
>+        [d1] "=&f"(data[1]), [d2] "=&f"(data[2]), [d3] "=&f"(data[3])
>+      : [src_uyvy] "r"(src_uyvy), [dst_u] "r"(dst_u), [dst_v] "r"(dst_v),
>+        [width] "r"(width), [c0] "f"(c0), [shift] "f"(shift)
>+      : "memory");
>+}
>+
>+// Copy row of UYVY Y's (422) into Y (420/422).
>+void UYVYToYRow_MMI(const uint8_t* src_uyvy, uint8_t* dst_y, int width) {
>+  // Output a row of Y values.
>+  uint64_t c0 = 0x00ff00ff00ff00ff;
>+  uint64_t shift = 0x08;
>+  uint64_t temp[2];
>+  __asm__ volatile(
>+      "1:	                                        \n\t"
>+      "gsldrc1  %[t0],       0x00(%[src_uyvy])          \n\t"
>+      "gsldlc1  %[t0],       0x07(%[src_uyvy])          \n\t"
>+      "gsldrc1  %[t1],       0x08(%[src_uyvy])          \n\t"
>+      "gsldlc1  %[t1],       0x0f(%[src_uyvy])          \n\t"
>+      "dsrl     %[t0],       %[t0],            %[shift] \n\t"
>+      "dsrl     %[t1],       %[t1],            %[shift] \n\t"
>+      "and      %[t0],       %[t0],            %[c0]    \n\t"
>+      "and      %[t1],       %[t1],            %[c0]    \n\t"
>+      "and      %[t1],       %[t1],            %[c0]    \n\t"
>+      "packushb %[t0],       %[t0],            %[t1]    \n\t"
>+      "gssdrc1  %[t0],       0x0(%[dst_y])	        \n\t"
>+      "gssdlc1  %[t0],       0x7(%[dst_y])              \n\t"
>+      "daddiu   %[src_uyvy], %[src_uyvy],      16       \n\t"
>+      "daddiu   %[dst_y],    %[dst_y],         8        \n\t"
>+      "daddiu   %[width],    %[width],        -8        \n\t"
>+      "bgtz     %[width],    1b                         \n\t"
>+      "nop                                              \n\t"
>+      : [t0] "=&f"(temp[0]), [t1] "=&f"(temp[1])
>+      : [src_uyvy] "r"(src_uyvy), [dst_y] "r"(dst_y), [width] "r"(width),
>+        [c0] "f"(c0), [shift] "f"(shift)
>+      : "memory");
>+}
>+
>+// Blend src_argb0 over src_argb1 and store to dst_argb.
>+// dst_argb may be src_argb0 or src_argb1.
>+// This code mimics the SSSE3 version for better testability.
>+void ARGBBlendRow_MMI(const uint8_t* src_argb0,
>+                      const uint8_t* src_argb1,
>+                      uint8_t* dst_argb,
>+                      int width) {
>+  uint64_t src0, src1, dest, alpha, src0_hi, src0_lo, src1_hi, src1_lo, dest_hi,
>+      dest_lo;
>+  const uint64_t mask0 = 0x0;
>+  const uint64_t mask1 = 0x00FFFFFF00FFFFFFULL;
>+  const uint64_t mask2 = 0x00FF00FF00FF00FFULL;
>+  const uint64_t mask3 = 0xFF;
>+  const uint64_t mask4 = ~mask1;
>+  const uint64_t shift = 0x08;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src0_ptr])                \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src0_ptr])                \n\t"
>+      "punpcklbh  %[src0_lo],      %[src0],           %[mask0]      \n\t"
>+
>+      "gsldlc1    %[src1],         0x07(%[src1_ptr])                \n\t"
>+      "gsldrc1    %[src1],         0x00(%[src1_ptr])                \n\t"
>+      "punpcklbh  %[src1_lo],      %[src1],           %[mask0]      \n\t"
>+
>+      "psubush    %[alpha],        %[mask2],          %[src0_lo]    \n\t"
>+      "pshufh     %[alpha],        %[alpha],          %[mask3]      \n\t"
>+      "pmullh     %[dest_lo],      %[src1_lo],        %[alpha]      \n\t"
>+      "psrlh      %[dest_lo],      %[dest_lo],        %[shift]      \n\t"
>+      "paddush    %[dest_lo],      %[dest_lo],        %[src0_lo]    \n\t"
>+
>+      "punpckhbh  %[src0_hi],      %[src0],           %[mask0]      \n\t"
>+      "punpckhbh  %[src1_hi],      %[src1],           %[mask0]      \n\t"
>+
>+      "psubush    %[alpha],        %[mask2],          %[src0_hi]    \n\t"
>+      "pshufh     %[alpha],        %[alpha],          %[mask3]      \n\t"
>+      "pmullh     %[dest_hi],      %[src1_hi],        %[alpha]      \n\t"
>+      "psrlh      %[dest_hi],      %[dest_hi],        %[shift]      \n\t"
>+      "paddush    %[dest_hi],      %[dest_hi],        %[src0_hi]    \n\t"
>+
>+      "packushb   %[dest],         %[dest_lo],        %[dest_hi]    \n\t"
>+      "and        %[dest],         %[dest],           %[mask1]      \n\t"
>+      "or         %[dest],         %[dest],           %[mask4]      \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src0_ptr],     %[src0_ptr],       0x08          \n\t"
>+      "daddiu     %[src1_ptr],     %[src1_ptr],       0x08          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x02          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src0] "=&f"(src0), [src1] "=&f"(src1), [alpha] "=&f"(alpha),
>+        [dest] "=&f"(dest), [src0_hi] "=&f"(src0_hi), [src0_lo] "=&f"(src0_lo),
>+        [src1_hi] "=&f"(src1_hi), [src1_lo] "=&f"(src1_lo),
>+        [dest_hi] "=&f"(dest_hi), [dest_lo] "=&f"(dest_lo)
>+      : [src0_ptr] "r"(src_argb0), [src1_ptr] "r"(src_argb1),
>+        [dst_ptr] "r"(dst_argb), [mask0] "f"(mask0), [mask1] "f"(mask1),
>+        [mask2] "f"(mask2), [mask3] "f"(mask3), [mask4] "f"(mask4),
>+        [shift] "f"(shift), [width] "r"(width)
>+      : "memory");
>+}
>+
>+void BlendPlaneRow_MMI(const uint8_t* src0,
>+                       const uint8_t* src1,
>+                       const uint8_t* alpha,
>+                       uint8_t* dst,
>+                       int width) {
>+  uint64_t source0, source1, dest, alph;
>+  uint64_t src0_hi, src0_lo, src1_hi, src1_lo, alpha_hi, alpha_lo, dest_hi,
>+      dest_lo;
>+  uint64_t alpha_rev, alpha_rev_lo, alpha_rev_hi;
>+  const uint64_t mask0 = 0x0;
>+  const uint64_t mask1 = 0xFFFFFFFFFFFFFFFFULL;
>+  const uint64_t mask2 = 0x00FF00FF00FF00FFULL;
>+  const uint64_t shift = 0x08;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src0_ptr])                \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src0_ptr])                \n\t"
>+      "punpcklbh  %[src0_lo],      %[src0],           %[mask0]      \n\t"
>+      "punpckhbh  %[src0_hi],      %[src0],           %[mask0]      \n\t"
>+
>+      "gsldlc1    %[src1],         0x07(%[src1_ptr])                \n\t"
>+      "gsldrc1    %[src1],         0x00(%[src1_ptr])                \n\t"
>+      "punpcklbh  %[src1_lo],      %[src1],           %[mask0]      \n\t"
>+      "punpckhbh  %[src1_hi],      %[src1],           %[mask0]      \n\t"
>+
>+      "gsldlc1    %[alpha],        0x07(%[alpha_ptr])               \n\t"
>+      "gsldrc1    %[alpha],        0x00(%[alpha_ptr])               \n\t"
>+      "psubusb    %[alpha_r],      %[mask1],          %[alpha]      \n\t"
>+      "punpcklbh  %[alpha_lo],     %[alpha],          %[mask0]      \n\t"
>+      "punpckhbh  %[alpha_hi],     %[alpha],          %[mask0]      \n\t"
>+      "punpcklbh  %[alpha_rlo],    %[alpha_r],        %[mask0]      \n\t"
>+      "punpckhbh  %[alpha_rhi],    %[alpha_r],        %[mask0]      \n\t"
>+
>+      "pmullh     %[dest_lo],      %[src0_lo],        %[alpha_lo]   \n\t"
>+      "pmullh     %[dest],         %[src1_lo],        %[alpha_rlo]  \n\t"
>+      "paddush    %[dest_lo],      %[dest_lo],        %[dest]       \n\t"
>+      "paddush    %[dest_lo],      %[dest_lo],        %[mask2]      \n\t"
>+      "psrlh      %[dest_lo],      %[dest_lo],        %[shift]      \n\t"
>+
>+      "pmullh     %[dest_hi],      %[src0_hi],        %[alpha_hi]   \n\t"
>+      "pmullh     %[dest],         %[src1_hi],        %[alpha_rhi]  \n\t"
>+      "paddush    %[dest_hi],      %[dest_hi],        %[dest]       \n\t"
>+      "paddush    %[dest_hi],      %[dest_hi],        %[mask2]      \n\t"
>+      "psrlh      %[dest_hi],      %[dest_hi],        %[shift]      \n\t"
>+
>+      "packushb   %[dest],         %[dest_lo],        %[dest_hi]    \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src0_ptr],     %[src0_ptr],       0x08          \n\t"
>+      "daddiu     %[src1_ptr],     %[src1_ptr],       0x08          \n\t"
>+      "daddiu     %[alpha_ptr],    %[alpha_ptr],      0x08          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x08          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src0] "=&f"(source0), [src1] "=&f"(source1), [alpha] "=&f"(alph),
>+        [dest] "=&f"(dest), [src0_hi] "=&f"(src0_hi), [src0_lo] "=&f"(src0_lo),
>+        [src1_hi] "=&f"(src1_hi), [src1_lo] "=&f"(src1_lo),
>+        [alpha_hi] "=&f"(alpha_hi), [alpha_lo] "=&f"(alpha_lo),
>+        [dest_hi] "=&f"(dest_hi), [dest_lo] "=&f"(dest_lo),
>+        [alpha_rlo] "=&f"(alpha_rev_lo), [alpha_rhi] "=&f"(alpha_rev_hi),
>+        [alpha_r] "=&f"(alpha_rev)
>+      : [src0_ptr] "r"(src0), [src1_ptr] "r"(src1), [alpha_ptr] "r"(alpha),
>+        [dst_ptr] "r"(dst), [mask0] "f"(mask0), [mask1] "f"(mask1),
>+        [mask2] "f"(mask2), [shift] "f"(shift), [width] "r"(width)
>+      : "memory");
>+}
>+
>+// Multiply source RGB by alpha and store to destination.
>+// This code mimics the SSSE3 version for better testability.
>+void ARGBAttenuateRow_MMI(const uint8_t* src_argb,
>+                          uint8_t* dst_argb,
>+                          int width) {
>+  uint64_t src, src_hi, src_lo, dest, dest_hi, dest_lo, alpha;
>+  const uint64_t mask0 = 0xFF;
>+  const uint64_t mask1 = 0xFF000000FF000000ULL;
>+  const uint64_t mask2 = ~mask1;
>+  const uint64_t shift = 0x08;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[src],          0x07(%[src_ptr])                 \n\t"
>+      "gsldrc1    %[src],          0x00(%[src_ptr])                 \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[src]        \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[src]        \n\t"
>+
>+      "pshufh     %[alpha],        %[src_lo],         %[mask0]      \n\t"
>+      "pmulhuh    %[dest_lo],      %[alpha],          %[src_lo]     \n\t"
>+      "psrlh      %[dest_lo],      %[dest_lo],        %[shift]      \n\t"
>+      "pshufh     %[alpha],        %[src_hi],         %[mask0]      \n\t"
>+      "pmulhuh    %[dest_hi],      %[alpha],          %[src_hi]     \n\t"
>+      "psrlh      %[dest_hi],      %[dest_hi],        %[shift]      \n\t"
>+
>+      "packushb   %[dest],         %[dest_lo],        %[dest_hi]    \n\t"
>+      "and        %[dest],         %[dest],           %[mask2]      \n\t"
>+      "and        %[src],          %[src],            %[mask1]      \n\t"
>+      "or         %[dest],         %[dest],           %[src]        \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x08          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x02          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src] "=&f"(src), [dest] "=&f"(dest), [src_hi] "=&f"(src_hi),
>+        [src_lo] "=&f"(src_lo), [dest_hi] "=&f"(dest_hi),
>+        [dest_lo] "=&f"(dest_lo), [alpha] "=&f"(alpha)
>+      : [src_ptr] "r"(src_argb), [dst_ptr] "r"(dst_argb), [mask0] "f"(mask0),
>+        [mask1] "f"(mask1), [mask2] "f"(mask2), [shift] "f"(shift),
>+        [width] "r"(width)
>+      : "memory");
>+}
>+
>+void ComputeCumulativeSumRow_MMI(const uint8_t* row,
>+                                 int32_t* cumsum,
>+                                 const int32_t* previous_cumsum,
>+                                 int width) {
>+  int64_t row_sum[2] = {0, 0};
>+  uint64_t src, dest0, dest1, presrc0, presrc1, dest;
>+  const uint64_t mask = 0x0;
>+
>+  __asm__ volatile(
>+      "xor        %[row_sum0],     %[row_sum0],       %[row_sum0]   \n\t"
>+      "xor        %[row_sum1],     %[row_sum1],       %[row_sum1]   \n\t"
>+
>+      "1:                                                           \n\t"
>+      "gslwlc1    %[src],          0x03(%[row_ptr])                 \n\t"
>+      "gslwrc1    %[src],          0x00(%[row_ptr])                 \n\t"
>+
>+      "punpcklbh  %[src],          %[src],            %[mask]       \n\t"
>+      "punpcklhw  %[dest0],        %[src],            %[mask]       \n\t"
>+      "punpckhhw  %[dest1],        %[src],            %[mask]       \n\t"
>+
>+      "paddw      %[row_sum0],     %[row_sum0],       %[dest0]      \n\t"
>+      "paddw      %[row_sum1],     %[row_sum1],       %[dest1]      \n\t"
>+
>+      "gsldlc1    %[presrc0],      0x07(%[pre_ptr])                 \n\t"
>+      "gsldrc1    %[presrc0],      0x00(%[pre_ptr])                 \n\t"
>+      "gsldlc1    %[presrc1],      0x0f(%[pre_ptr])                 \n\t"
>+      "gsldrc1    %[presrc1],      0x08(%[pre_ptr])                 \n\t"
>+
>+      "paddw      %[dest0],        %[row_sum0],       %[presrc0]    \n\t"
>+      "paddw      %[dest1],        %[row_sum1],       %[presrc1]    \n\t"
>+
>+      "gssdlc1    %[dest0],        0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest0],        0x00(%[dst_ptr])                 \n\t"
>+      "gssdlc1    %[dest1],        0x0f(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest1],        0x08(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[row_ptr],      %[row_ptr],        0x04          \n\t"
>+      "daddiu     %[pre_ptr],      %[pre_ptr],        0x10          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x10          \n\t"
>+      "daddi      %[width],        %[width],         -0x01          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src] "=&f"(src), [dest] "=&f"(dest), [dest0] "=&f"(dest0),
>+        [dest1] "=&f"(dest1), [row_sum0] "+&f"(row_sum[0]),
>+        [row_sum1] "+&f"(row_sum[1]), [presrc0] "=&f"(presrc0),
>+        [presrc1] "=&f"(presrc1)
>+      : [row_ptr] "r"(row), [pre_ptr] "r"(previous_cumsum),
>+        [dst_ptr] "r"(cumsum), [width] "r"(width), [mask] "f"(mask)
>+      : "memory");
>+}
>+
>+// C version 2x2 -> 2x1.
>+void InterpolateRow_MMI(uint8_t* dst_ptr,
>+                        const uint8_t* src_ptr,
>+                        ptrdiff_t src_stride,
>+                        int width,
>+                        int source_y_fraction) {
>+  if (source_y_fraction == 0) {
>+    __asm__ volatile(
>+        "1:	                              \n\t"
>+        "ld     $t0,        0x0(%[src_ptr])   \n\t"
>+        "sd     $t0,        0x0(%[dst_ptr])   \n\t"
>+        "daddiu %[src_ptr], %[src_ptr],     8 \n\t"
>+        "daddiu %[dst_ptr], %[dst_ptr],     8 \n\t"
>+        "daddiu %[width],   %[width],      -8 \n\t"
>+        "bgtz   %[width],   1b                \n\t"
>+        "nop                                  \n\t"
>+        :
>+        : [dst_ptr] "r"(dst_ptr), [src_ptr] "r"(src_ptr), [width] "r"(width)
>+        : "memory");
>+    return;
>+  }
>+  if (source_y_fraction == 128) {
>+    uint64_t uv = 0x0;
>+    uint64_t uv_stride = 0x0;
>+    __asm__ volatile(
>+        "1:	                                            \n\t"
>+        "gsldrc1 %[uv],        0x0(%[src_ptr])              \n\t"
>+        "gsldlc1 %[uv],        0x7(%[src_ptr])              \n\t"
>+        "daddu   $t0,          %[src_ptr],     %[stride]    \n\t"
>+        "gsldrc1 %[uv_stride], 0x0($t0)                     \n\t"
>+        "gsldlc1 %[uv_stride], 0x7($t0)                     \n\t"
>+
>+        "pavgb   %[uv],        %[uv],          %[uv_stride] \n\t"
>+        "gssdrc1 %[uv],        0x0(%[dst_ptr])              \n\t"
>+        "gssdlc1 %[uv],        0x7(%[dst_ptr])              \n\t"
>+
>+        "daddiu  %[src_ptr],   %[src_ptr],     8            \n\t"
>+        "daddiu  %[dst_ptr],   %[dst_ptr],     8            \n\t"
>+        "daddiu  %[width],     %[width],      -8            \n\t"
>+        "bgtz    %[width],     1b                           \n\t"
>+        "nop                                                \n\t"
>+        : [uv] "=&f"(uv), [uv_stride] "=&f"(uv_stride)
>+        : [src_ptr] "r"(src_ptr), [dst_ptr] "r"(dst_ptr), [width] "r"(width),
>+          [stride] "r"((int64_t)src_stride)
>+        : "memory");
>+    return;
>+  }
>+  const uint8_t* src_ptr1 = src_ptr + src_stride;
>+  uint64_t temp;
>+  uint64_t data[4];
>+  uint64_t zero = 0x0;
>+  uint64_t c0 = 0x0080008000800080;
>+  uint64_t fy0 = 0x0100010001000100;
>+  uint64_t shift = 0x8;
>+  __asm__ volatile(
>+      "pshufh    %[fy1],      %[fy1],          %[zero]  \n\t"
>+      "psubh     %[fy0],      %[fy0],          %[fy1]   \n\t"
>+      "1:	                                        \n\t"
>+      "gsldrc1   %[t0],       0x0(%[src_ptr])           \n\t"
>+      "gsldlc1   %[t0],       0x7(%[src_ptr])           \n\t"
>+      "punpcklbh %[d0],       %[t0],           %[zero]  \n\t"
>+      "punpckhbh %[d1],       %[t0],           %[zero]  \n\t"
>+      "gsldrc1   %[t0],       0x0(%[src_ptr1])          \n\t"
>+      "gsldlc1   %[t0],       0x7(%[src_ptr1])          \n\t"
>+      "punpcklbh %[d2],       %[t0],           %[zero]  \n\t"
>+      "punpckhbh %[d3],       %[t0],           %[zero]  \n\t"
>+
>+      "pmullh    %[d0],       %[d0],           %[fy0]   \n\t"
>+      "pmullh    %[d2],       %[d2],           %[fy1]   \n\t"
>+      "paddh     %[d0],       %[d0],           %[d2]    \n\t"
>+      "paddh     %[d0],       %[d0],           %[c0]    \n\t"
>+      "psrlh     %[d0],       %[d0],           %[shift] \n\t"
>+
>+      "pmullh    %[d1],       %[d1],           %[fy0]   \n\t"
>+      "pmullh    %[d3],       %[d3],           %[fy1]   \n\t"
>+      "paddh     %[d1],       %[d1],           %[d3]    \n\t"
>+      "paddh     %[d1],       %[d1],           %[c0]    \n\t"
>+      "psrlh     %[d1],       %[d1],           %[shift] \n\t"
>+
>+      "packushb  %[d0],       %[d0],           %[d1]    \n\t"
>+      "gssdrc1   %[d0],       0x0(%[dst_ptr])           \n\t"
>+      "gssdlc1   %[d0],       0x7(%[dst_ptr])           \n\t"
>+      "daddiu    %[src_ptr],  %[src_ptr],      8        \n\t"
>+      "daddiu    %[src_ptr1], %[src_ptr1],     8        \n\t"
>+      "daddiu    %[dst_ptr],  %[dst_ptr],      8        \n\t"
>+      "daddiu    %[width],    %[width],       -8        \n\t"
>+      "bgtz      %[width],    1b                        \n\t"
>+      "nop                                              \n\t"
>+      : [t0] "=&f"(temp), [d0] "=&f"(data[0]), [d1] "=&f"(data[1]),
>+        [d2] "=&f"(data[2]), [d3] "=&f"(data[3])
>+      : [src_ptr] "r"(src_ptr), [src_ptr1] "r"(src_ptr1),
>+        [dst_ptr] "r"(dst_ptr), [width] "r"(width),
>+        [fy1] "f"(source_y_fraction), [fy0] "f"(fy0), [c0] "f"(c0),
>+        [shift] "f"(shift), [zero] "f"(zero)
>+      : "memory");
>+}
>+
>+// Use first 4 shuffler values to reorder ARGB channels.
>+void ARGBShuffleRow_MMI(const uint8_t* src_argb,
>+                        uint8_t* dst_argb,
>+                        const uint8_t* shuffler,
>+                        int width) {
>+  uint64_t source, dest0, dest1, dest;
>+  const uint64_t mask0 = 0x0;
>+  const uint64_t mask1 = (shuffler[0] & 0x03) | ((shuffler[1] & 0x03) << 2) |
>+                         ((shuffler[2] & 0x03) << 4) |
>+                         ((shuffler[3] & 0x03) << 6);
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[src],          0x07(%[src_ptr])                 \n\t"
>+      "gsldrc1    %[src],          0x00(%[src_ptr])                 \n\t"
>+
>+      "punpcklbh  %[dest0],        %[src],            %[mask0]      \n\t"
>+      "pshufh     %[dest0],        %[dest0],          %[mask1]      \n\t"
>+      "punpckhbh  %[dest1],        %[src],            %[mask0]      \n\t"
>+      "pshufh     %[dest1],        %[dest1],          %[mask1]      \n\t"
>+      "packushb   %[dest],         %[dest0],          %[dest1]      \n\t"
>+
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x08          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x02          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src] "=&f"(source), [dest] "=&f"(dest), [dest0] "=&f"(dest0),
>+        [dest1] "=&f"(dest1)
>+      : [src_ptr] "r"(src_argb), [dst_ptr] "r"(dst_argb), [mask0] "f"(mask0),
>+        [mask1] "f"(mask1), [width] "r"(width)
>+      : "memory");
>+}
>+
>+void I422ToYUY2Row_MMI(const uint8_t* src_y,
>+                       const uint8_t* src_u,
>+                       const uint8_t* src_v,
>+                       uint8_t* dst_frame,
>+                       int width) {
>+  uint64_t temp[3];
>+  uint64_t vu = 0x0;
>+  __asm__ volatile(
>+      "1:	                                        \n\t"
>+      "gsldlc1   %[ty],        0x7(%[src_y])            \n\t"  // r=src_sobelx[i]
>+      "gsldrc1   %[ty],        0x0(%[src_y])            \n\t"  // r=src_sobelx[i]
>+      "gslwlc1   %[tu],        0x3(%[src_u])            \n\t"  // b=src_sobely[i]
>+      "gslwrc1   %[tu],        0x0(%[src_u])            \n\t"  // b=src_sobely[i]
>+      "gslwlc1   %[tv],        0x3(%[src_v])            \n\t"  // b=src_sobely[i]
>+      "gslwrc1   %[tv],        0x0(%[src_v])            \n\t"  // b=src_sobely[i]
>+      "punpcklbh %[vu],        %[tu],             %[tv]	\n\t"  // g
>+      "punpcklbh %[tu],        %[ty],             %[vu]	\n\t"  // g
>+      "gssdlc1   %[tu],        0x7(%[dst_frame])        \n\t"
>+      "gssdrc1   %[tu],        0x0(%[dst_frame])        \n\t"
>+      "punpckhbh %[tu],        %[ty],             %[vu]	\n\t"  // g
>+      "gssdlc1   %[tu],        0x0F(%[dst_frame])       \n\t"
>+      "gssdrc1   %[tu],        0x08(%[dst_frame])       \n\t"
>+      "daddiu    %[src_y],     %[src_y],          8     \n\t"
>+      "daddiu    %[src_u],     %[src_u],          4     \n\t"
>+      "daddiu    %[src_v],     %[src_v],          4     \n\t"
>+      "daddiu    %[dst_frame], %[dst_frame],      16    \n\t"
>+      "daddiu    %[width],     %[width],         -8     \n\t"
>+      "bgtz      %[width],     1b                       \n\t"
>+      "nop                                              \n\t"
>+      : [ty] "=&f"(temp[1]), [tu] "=&f"(temp[1]), [tv] "=&f"(temp[1]),
>+        [vu] "=&f"(vu)
>+      : [src_y] "r"(src_y), [src_u] "r"(src_u), [src_v] "r"(src_v),
>+        [dst_frame] "r"(dst_frame), [width] "r"(width)
>+      : "memory");
>+}
>+
>+void I422ToUYVYRow_MMI(const uint8_t* src_y,
>+                       const uint8_t* src_u,
>+                       const uint8_t* src_v,
>+                       uint8_t* dst_frame,
>+                       int width) {
>+  uint64_t temp[3];
>+  uint64_t vu = 0x0;
>+  __asm__ volatile(
>+      "1:	                                        \n\t"
>+      "gsldlc1   %[ty],        0x7(%[src_y])            \n\t"  // r=src_sobelx[i]
>+      "gsldrc1   %[ty],        0x0(%[src_y])            \n\t"  // r=src_sobelx[i]
>+      "gslwlc1   %[tu],        0x3(%[src_u])            \n\t"  // b=src_sobely[i]
>+      "gslwrc1   %[tu],        0x0(%[src_u])            \n\t"  // b=src_sobely[i]
>+      "gslwlc1   %[tv],        0x3(%[src_v])            \n\t"  // b=src_sobely[i]
>+      "gslwrc1   %[tv],        0x0(%[src_v])            \n\t"  // b=src_sobely[i]
>+      "punpcklbh %[vu],        %[tu],             %[tv]	\n\t"  // g
>+      "punpcklbh %[tu],        %[vu],             %[ty]	\n\t"  // g
>+      "gssdlc1   %[tu],        0x7(%[dst_frame])        \n\t"
>+      "gssdrc1   %[tu],        0x0(%[dst_frame])        \n\t"
>+      "punpckhbh %[tu],        %[vu],             %[ty]	\n\t"  // g
>+      "gssdlc1   %[tu],        0x0F(%[dst_frame])       \n\t"
>+      "gssdrc1   %[tu],        0x08(%[dst_frame])       \n\t"
>+      "daddiu    %[src_y],     %[src_y],          8     \n\t"
>+      "daddiu    %[src_u],     %[src_u],          4     \n\t"
>+      "daddiu    %[src_v],     %[src_v],          4     \n\t"
>+      "daddiu    %[dst_frame], %[dst_frame],      16    \n\t"
>+      "daddiu    %[width],     %[width],         -8     \n\t"
>+      "bgtz      %[width],     1b                       \n\t"
>+      "nop                                              \n\t"
>+      : [ty] "=&f"(temp[1]), [tu] "=&f"(temp[1]), [tv] "=&f"(temp[1]),
>+        [vu] "=&f"(vu)
>+      : [src_y] "r"(src_y), [src_u] "r"(src_u), [src_v] "r"(src_v),
>+        [dst_frame] "r"(dst_frame), [width] "r"(width)
>+      : "memory");
>+}
>+
>+void ARGBCopyAlphaRow_MMI(const uint8_t* src, uint8_t* dst, int width) {
>+  uint64_t source, dest;
>+  const uint64_t mask0 = 0xff000000ff000000ULL;
>+  const uint64_t mask1 = ~mask0;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[src],          0x07(%[src_ptr])                 \n\t"
>+      "gsldrc1    %[src],          0x00(%[src_ptr])                 \n\t"
>+      "gsldlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gsldrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "and        %[src],          %[src],            %[mask0]      \n\t"
>+      "and        %[dest],         %[dest],           %[mask1]      \n\t"
>+      "or         %[dest],         %[src],            %[dest]       \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x08          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x02          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src] "=&f"(source), [dest] "=&f"(dest)
>+      : [src_ptr] "r"(src), [dst_ptr] "r"(dst), [mask0] "f"(mask0),
>+        [mask1] "f"(mask1), [width] "r"(width)
>+      : "memory");
>+}
>+
>+void ARGBExtractAlphaRow_MMI(const uint8_t* src_argb,
>+                             uint8_t* dst_a,
>+                             int width) {
>+  uint64_t src, dest0, dest1, dest_lo, dest_hi, dest;
>+  const uint64_t mask = 0xff000000ff000000ULL;
>+  const uint64_t shift = 0x18;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[src],          0x07(%[src_ptr])                 \n\t"
>+      "gsldrc1    %[src],          0x00(%[src_ptr])                 \n\t"
>+      "and        %[dest0],        %[src],            %[mask]       \n\t"
>+      "psrlw      %[dest0],        %[dest0],          %[shift]      \n\t"
>+      "gsldlc1    %[src],          0x0f(%[src_ptr])                 \n\t"
>+      "gsldrc1    %[src],          0x08(%[src_ptr])                 \n\t"
>+      "and        %[dest1],        %[src],            %[mask]       \n\t"
>+      "psrlw      %[dest1],        %[dest1],          %[shift]      \n\t"
>+      "packsswh   %[dest_lo],      %[dest0],          %[dest1]      \n\t"
>+
>+      "gsldlc1    %[src],          0x17(%[src_ptr])                 \n\t"
>+      "gsldrc1    %[src],          0x10(%[src_ptr])                 \n\t"
>+      "and        %[dest0],        %[src],            %[mask]       \n\t"
>+      "psrlw      %[dest0],        %[dest0],          %[shift]      \n\t"
>+      "gsldlc1    %[src],          0x1f(%[src_ptr])                 \n\t"
>+      "gsldrc1    %[src],          0x18(%[src_ptr])                 \n\t"
>+      "and        %[dest1],        %[src],            %[mask]       \n\t"
>+      "psrlw      %[dest1],        %[dest1],          %[shift]      \n\t"
>+      "packsswh   %[dest_hi],      %[dest0],          %[dest1]      \n\t"
>+
>+      "packushb   %[dest],         %[dest_lo],        %[dest_hi]    \n\t"
>+
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x20          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x08          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src] "=&f"(src), [dest] "=&f"(dest), [dest0] "=&f"(dest0),
>+        [dest1] "=&f"(dest1), [dest_lo] "=&f"(dest_lo), [dest_hi] "=&f"(dest_hi)
>+      : [src_ptr] "r"(src_argb), [dst_ptr] "r"(dst_a), [mask] "f"(mask),
>+        [shift] "f"(shift), [width] "r"(width)
>+      : "memory");
>+}
>+
>+void ARGBCopyYToAlphaRow_MMI(const uint8_t* src, uint8_t* dst, int width) {
>+  uint64_t source, dest0, dest1, dest;
>+  const uint64_t mask0 = 0x0;
>+  const uint64_t mask1 = 0x00ffffff00ffffffULL;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[src],          0x07(%[src_ptr])                 \n\t"
>+      "gsldrc1    %[src],          0x00(%[src_ptr])                 \n\t"
>+
>+      "punpcklbh  %[dest0],        %[mask0],          %[src]        \n\t"
>+      "punpcklhw  %[dest1],        %[mask0],          %[dest0]      \n\t"
>+      "gsldlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gsldrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+      "and        %[dest],         %[dest],           %[mask1]      \n\t"
>+      "or         %[dest],         %[dest],           %[dest1]      \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+      "punpckhhw  %[dest1],        %[mask0],          %[dest0]      \n\t"
>+      "gsldlc1    %[dest],         0x0f(%[dst_ptr])                 \n\t"
>+      "gsldrc1    %[dest],         0x08(%[dst_ptr])                 \n\t"
>+      "and        %[dest],         %[dest],           %[mask1]      \n\t"
>+      "or         %[dest],         %[dest],           %[dest1]      \n\t"
>+      "gssdlc1    %[dest],         0x0f(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x08(%[dst_ptr])                 \n\t"
>+
>+      "punpckhbh  %[dest0],        %[mask0],          %[src]        \n\t"
>+      "punpcklhw  %[dest1],        %[mask0],          %[dest0]      \n\t"
>+      "gsldlc1    %[dest],         0x17(%[dst_ptr])                 \n\t"
>+      "gsldrc1    %[dest],         0x10(%[dst_ptr])                 \n\t"
>+      "and        %[dest],         %[dest],           %[mask1]      \n\t"
>+      "or         %[dest],         %[dest],           %[dest1]      \n\t"
>+      "gssdlc1    %[dest],         0x17(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x10(%[dst_ptr])                 \n\t"
>+      "punpckhhw  %[dest1],        %[mask0],          %[dest0]      \n\t"
>+      "gsldlc1    %[dest],         0x1f(%[dst_ptr])                 \n\t"
>+      "gsldrc1    %[dest],         0x18(%[dst_ptr])                 \n\t"
>+      "and        %[dest],         %[dest],           %[mask1]      \n\t"
>+      "or         %[dest],         %[dest],           %[dest1]      \n\t"
>+      "gssdlc1    %[dest],         0x1f(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x18(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x08          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x20          \n\t"
>+      "daddi      %[width],        %[width],         -0x08          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src] "=&f"(source), [dest] "=&f"(dest), [dest0] "=&f"(dest0),
>+        [dest1] "=&f"(dest1)
>+      : [src_ptr] "r"(src), [dst_ptr] "r"(dst), [mask0] "f"(mask0),
>+        [mask1] "f"(mask1), [width] "r"(width)
>+      : "memory");
>+}
>+
>+#endif  // !defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)
>+
>+#ifdef __cplusplus
>+}  // extern "C"
>+}  // namespace libyuv
>+#endif
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/row_msa.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/row_msa.cc
>index 66666cefcd98c4f2314bf27ac142d34f1864cfcf..5c0239a37f0f646d4fd8ebbd42cff08d9ee16d4c 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/row_msa.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/row_msa.cc
>@@ -163,14 +163,14 @@ extern "C" {
>     v8u16 reg0_m, reg1_m, reg2_m, reg3_m, reg4_m, reg5_m, reg6_m, reg7_m; \
>     v8u16 reg8_m, reg9_m;                                                 \
>                                                                           \
>-    src0_m = (v16u8)__msa_ld_b((v16i8*)s, 0);                             \
>-    src1_m = (v16u8)__msa_ld_b((v16i8*)s, 16);                            \
>-    src2_m = (v16u8)__msa_ld_b((v16i8*)s, 32);                            \
>-    src3_m = (v16u8)__msa_ld_b((v16i8*)s, 48);                            \
>-    src4_m = (v16u8)__msa_ld_b((v16i8*)t, 0);                             \
>-    src5_m = (v16u8)__msa_ld_b((v16i8*)t, 16);                            \
>-    src6_m = (v16u8)__msa_ld_b((v16i8*)t, 32);                            \
>-    src7_m = (v16u8)__msa_ld_b((v16i8*)t, 48);                            \
>+    src0_m = (v16u8)__msa_ld_b((void*)s, 0);                              \
>+    src1_m = (v16u8)__msa_ld_b((void*)s, 16);                             \
>+    src2_m = (v16u8)__msa_ld_b((void*)s, 32);                             \
>+    src3_m = (v16u8)__msa_ld_b((void*)s, 48);                             \
>+    src4_m = (v16u8)__msa_ld_b((void*)t, 0);                              \
>+    src5_m = (v16u8)__msa_ld_b((void*)t, 16);                             \
>+    src6_m = (v16u8)__msa_ld_b((void*)t, 32);                             \
>+    src7_m = (v16u8)__msa_ld_b((void*)t, 48);                             \
>     vec0_m = (v16u8)__msa_ilvr_b((v16i8)src0_m, (v16i8)src4_m);           \
>     vec1_m = (v16u8)__msa_ilvr_b((v16i8)src1_m, (v16i8)src5_m);           \
>     vec2_m = (v16u8)__msa_ilvr_b((v16i8)src2_m, (v16i8)src6_m);           \
>@@ -201,14 +201,14 @@ extern "C" {
>     reg1_m = (v8u16)__msa_srai_h((v8i16)reg1_m, 2);                       \
>     argb0 = (v16u8)__msa_pckev_b((v16i8)reg9_m, (v16i8)reg8_m);           \
>     argb1 = (v16u8)__msa_pckev_b((v16i8)reg1_m, (v16i8)reg0_m);           \
>-    src0_m = (v16u8)__msa_ld_b((v16i8*)s, 64);                            \
>-    src1_m = (v16u8)__msa_ld_b((v16i8*)s, 80);                            \
>-    src2_m = (v16u8)__msa_ld_b((v16i8*)s, 96);                            \
>-    src3_m = (v16u8)__msa_ld_b((v16i8*)s, 112);                           \
>-    src4_m = (v16u8)__msa_ld_b((v16i8*)t, 64);                            \
>-    src5_m = (v16u8)__msa_ld_b((v16i8*)t, 80);                            \
>-    src6_m = (v16u8)__msa_ld_b((v16i8*)t, 96);                            \
>-    src7_m = (v16u8)__msa_ld_b((v16i8*)t, 112);                           \
>+    src0_m = (v16u8)__msa_ld_b((void*)s, 64);                             \
>+    src1_m = (v16u8)__msa_ld_b((void*)s, 80);                             \
>+    src2_m = (v16u8)__msa_ld_b((void*)s, 96);                             \
>+    src3_m = (v16u8)__msa_ld_b((void*)s, 112);                            \
>+    src4_m = (v16u8)__msa_ld_b((void*)t, 64);                             \
>+    src5_m = (v16u8)__msa_ld_b((void*)t, 80);                             \
>+    src6_m = (v16u8)__msa_ld_b((void*)t, 96);                             \
>+    src7_m = (v16u8)__msa_ld_b((void*)t, 112);                            \
>     vec2_m = (v16u8)__msa_ilvr_b((v16i8)src0_m, (v16i8)src4_m);           \
>     vec3_m = (v16u8)__msa_ilvr_b((v16i8)src1_m, (v16i8)src5_m);           \
>     vec4_m = (v16u8)__msa_ilvr_b((v16i8)src2_m, (v16i8)src6_m);           \
>@@ -942,10 +942,10 @@ void ARGBToRGB24Row_MSA(const uint8_t* src_argb, uint8_t* dst_rgb, int width) {
>                      21, 22, 24, 25, 26, 28, 29, 30};
> 
>   for (x = 0; x < width; x += 16) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_argb, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_argb, 16);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)src_argb, 32);
>-    src3 = (v16u8)__msa_ld_b((v16i8*)src_argb, 48);
>+    src0 = (v16u8)__msa_ld_b((void*)src_argb, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_argb, 16);
>+    src2 = (v16u8)__msa_ld_b((void*)src_argb, 32);
>+    src3 = (v16u8)__msa_ld_b((void*)src_argb, 48);
>     dst0 = (v16u8)__msa_vshf_b(shuffler0, (v16i8)src1, (v16i8)src0);
>     dst1 = (v16u8)__msa_vshf_b(shuffler1, (v16i8)src2, (v16i8)src1);
>     dst2 = (v16u8)__msa_vshf_b(shuffler2, (v16i8)src3, (v16i8)src2);
>@@ -966,10 +966,10 @@ void ARGBToRAWRow_MSA(const uint8_t* src_argb, uint8_t* dst_rgb, int width) {
>                      21, 20, 26, 25, 24, 30, 29, 28};
> 
>   for (x = 0; x < width; x += 16) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_argb, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_argb, 16);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)src_argb, 32);
>-    src3 = (v16u8)__msa_ld_b((v16i8*)src_argb, 48);
>+    src0 = (v16u8)__msa_ld_b((void*)src_argb, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_argb, 16);
>+    src2 = (v16u8)__msa_ld_b((void*)src_argb, 32);
>+    src3 = (v16u8)__msa_ld_b((void*)src_argb, 48);
>     dst0 = (v16u8)__msa_vshf_b(shuffler0, (v16i8)src1, (v16i8)src0);
>     dst1 = (v16u8)__msa_vshf_b(shuffler1, (v16i8)src2, (v16i8)src1);
>     dst2 = (v16u8)__msa_vshf_b(shuffler2, (v16i8)src3, (v16i8)src2);
>@@ -987,8 +987,8 @@ void ARGBToRGB565Row_MSA(const uint8_t* src_argb, uint8_t* dst_rgb, int width) {
>   v16i8 zero = {0};
> 
>   for (x = 0; x < width; x += 8) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_argb, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_argb, 16);
>+    src0 = (v16u8)__msa_ld_b((void*)src_argb, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_argb, 16);
>     vec0 = (v16u8)__msa_srai_b((v16i8)src0, 3);
>     vec1 = (v16u8)__msa_slli_b((v16i8)src0, 3);
>     vec2 = (v16u8)__msa_srai_b((v16i8)src0, 5);
>@@ -1023,8 +1023,8 @@ void ARGBToARGB1555Row_MSA(const uint8_t* src_argb,
>   v16i8 zero = {0};
> 
>   for (x = 0; x < width; x += 8) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_argb, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_argb, 16);
>+    src0 = (v16u8)__msa_ld_b((void*)src_argb, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_argb, 16);
>     vec0 = (v16u8)__msa_srai_b((v16i8)src0, 3);
>     vec1 = (v16u8)__msa_slli_b((v16i8)src0, 2);
>     vec2 = (v16u8)__msa_srai_b((v16i8)vec0, 3);
>@@ -1066,8 +1066,8 @@ void ARGBToARGB4444Row_MSA(const uint8_t* src_argb,
>   v16i8 zero = {0};
> 
>   for (x = 0; x < width; x += 8) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_argb, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_argb, 16);
>+    src0 = (v16u8)__msa_ld_b((void*)src_argb, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_argb, 16);
>     vec0 = (v16u8)__msa_srai_b((v16i8)src0, 4);
>     vec1 = (v16u8)__msa_srai_b((v16i8)src1, 4);
>     src0 = (v16u8)__msa_sldi_b(zero, (v16i8)src0, 1);
>@@ -1098,10 +1098,10 @@ void ARGBToUV444Row_MSA(const uint8_t* src_argb,
>   v16i8 zero = {0};
> 
>   for (x = width; x > 0; x -= 16) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_argb, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_argb, 16);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)src_argb, 32);
>-    src3 = (v16u8)__msa_ld_b((v16i8*)src_argb, 48);
>+    src0 = (v16u8)__msa_ld_b((void*)src_argb, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_argb, 16);
>+    src2 = (v16u8)__msa_ld_b((void*)src_argb, 32);
>+    src3 = (v16u8)__msa_ld_b((void*)src_argb, 48);
>     reg0 = (v16u8)__msa_pckev_b((v16i8)src1, (v16i8)src0);
>     reg1 = (v16u8)__msa_pckev_b((v16i8)src3, (v16i8)src2);
>     reg2 = (v16u8)__msa_pckod_b((v16i8)src1, (v16i8)src0);
>@@ -1164,8 +1164,8 @@ void ARGBMultiplyRow_MSA(const uint8_t* src_argb0,
>   v8i16 zero = {0};
> 
>   for (x = 0; x < width; x += 4) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_argb1, 0);
>+    src0 = (v16u8)__msa_ld_b((void*)src_argb0, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_argb1, 0);
>     vec0 = (v8u16)__msa_ilvr_b((v16i8)src0, (v16i8)src0);
>     vec1 = (v8u16)__msa_ilvl_b((v16i8)src0, (v16i8)src0);
>     vec2 = (v8u16)__msa_ilvr_b((v16i8)zero, (v16i8)src1);
>@@ -1200,10 +1200,10 @@ void ARGBAddRow_MSA(const uint8_t* src_argb0,
>   v16u8 src0, src1, src2, src3, dst0, dst1;
> 
>   for (x = 0; x < width; x += 8) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 16);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)src_argb1, 0);
>-    src3 = (v16u8)__msa_ld_b((v16i8*)src_argb1, 16);
>+    src0 = (v16u8)__msa_ld_b((void*)src_argb0, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_argb0, 16);
>+    src2 = (v16u8)__msa_ld_b((void*)src_argb1, 0);
>+    src3 = (v16u8)__msa_ld_b((void*)src_argb1, 16);
>     dst0 = __msa_adds_u_b(src0, src2);
>     dst1 = __msa_adds_u_b(src1, src3);
>     ST_UB2(dst0, dst1, dst_argb, 16);
>@@ -1221,10 +1221,10 @@ void ARGBSubtractRow_MSA(const uint8_t* src_argb0,
>   v16u8 src0, src1, src2, src3, dst0, dst1;
> 
>   for (x = 0; x < width; x += 8) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 16);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)src_argb1, 0);
>-    src3 = (v16u8)__msa_ld_b((v16i8*)src_argb1, 16);
>+    src0 = (v16u8)__msa_ld_b((void*)src_argb0, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_argb0, 16);
>+    src2 = (v16u8)__msa_ld_b((void*)src_argb1, 0);
>+    src3 = (v16u8)__msa_ld_b((void*)src_argb1, 16);
>     dst0 = __msa_subs_u_b(src0, src2);
>     dst1 = __msa_subs_u_b(src1, src3);
>     ST_UB2(dst0, dst1, dst_argb, 16);
>@@ -1245,8 +1245,8 @@ void ARGBAttenuateRow_MSA(const uint8_t* src_argb,
>   v16u8 mask = {0, 0, 0, 255, 0, 0, 0, 255, 0, 0, 0, 255, 0, 0, 0, 255};
> 
>   for (x = 0; x < width; x += 8) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_argb, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_argb, 16);
>+    src0 = (v16u8)__msa_ld_b((void*)src_argb, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_argb, 16);
>     vec0 = (v8u16)__msa_ilvr_b((v16i8)src0, (v16i8)src0);
>     vec1 = (v8u16)__msa_ilvl_b((v16i8)src0, (v16i8)src0);
>     vec2 = (v8u16)__msa_ilvr_b((v16i8)src1, (v16i8)src1);
>@@ -1316,8 +1316,8 @@ void ARGBToRGB565DitherRow_MSA(const uint8_t* src_argb,
>   vec_d0 = (v8i16)__msa_ilvr_b(zero, (v16i8)vec_d0);
> 
>   for (x = 0; x < width; x += 8) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_argb, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_argb, 16);
>+    src0 = (v16u8)__msa_ld_b((void*)src_argb, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_argb, 16);
>     vec0 = (v16u8)__msa_pckev_b((v16i8)src1, (v16i8)src0);
>     vec1 = (v16u8)__msa_pckod_b((v16i8)src1, (v16i8)src0);
>     reg0 = (v8i16)__msa_ilvev_b(zero, (v16i8)vec0);
>@@ -1514,8 +1514,8 @@ void ARGB1555ToARGBRow_MSA(const uint8_t* src_argb1555,
>   v8u16 const_0x1F = (v8u16)__msa_ldi_h(0x1F);
> 
>   for (x = 0; x < width; x += 16) {
>-    src0 = (v8u16)__msa_ld_h((v8u16*)src_argb1555, 0);
>-    src1 = (v8u16)__msa_ld_h((v8u16*)src_argb1555, 16);
>+    src0 = (v8u16)__msa_ld_h((void*)src_argb1555, 0);
>+    src1 = (v8u16)__msa_ld_h((void*)src_argb1555, 16);
>     vec0 = src0 & const_0x1F;
>     vec1 = src1 & const_0x1F;
>     src0 = (v8u16)__msa_srli_h((v8i16)src0, 5);
>@@ -1566,8 +1566,8 @@ void RGB565ToARGBRow_MSA(const uint8_t* src_rgb565,
>   v8u16 const_0xF800 = (v8u16)__msa_fill_h(0xF800);
> 
>   for (x = 0; x < width; x += 16) {
>-    src0 = (v8u16)__msa_ld_h((v8u16*)src_rgb565, 0);
>-    src1 = (v8u16)__msa_ld_h((v8u16*)src_rgb565, 16);
>+    src0 = (v8u16)__msa_ld_h((void*)src_rgb565, 0);
>+    src1 = (v8u16)__msa_ld_h((void*)src_rgb565, 16);
>     vec0 = src0 & const_0x1F;
>     vec1 = src0 & const_0x7E0;
>     vec2 = src0 & const_0xF800;
>@@ -1611,9 +1611,9 @@ void RGB24ToARGBRow_MSA(const uint8_t* src_rgb24,
>   v16i8 shuffler = {0, 1, 2, 16, 3, 4, 5, 17, 6, 7, 8, 18, 9, 10, 11, 19};
> 
>   for (x = 0; x < width; x += 16) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_rgb24, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_rgb24, 16);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)src_rgb24, 32);
>+    src0 = (v16u8)__msa_ld_b((void*)src_rgb24, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_rgb24, 16);
>+    src2 = (v16u8)__msa_ld_b((void*)src_rgb24, 32);
>     vec0 = (v16u8)__msa_sldi_b((v16i8)src1, (v16i8)src0, 12);
>     vec1 = (v16u8)__msa_sldi_b((v16i8)src2, (v16i8)src1, 8);
>     vec2 = (v16u8)__msa_sldi_b((v16i8)src2, (v16i8)src2, 4);
>@@ -1636,9 +1636,9 @@ void RAWToARGBRow_MSA(const uint8_t* src_raw, uint8_t* dst_argb, int width) {
>   v16i8 mask = {2, 1, 0, 16, 5, 4, 3, 17, 8, 7, 6, 18, 11, 10, 9, 19};
> 
>   for (x = 0; x < width; x += 16) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_raw, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_raw, 16);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)src_raw, 32);
>+    src0 = (v16u8)__msa_ld_b((void*)src_raw, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_raw, 16);
>+    src2 = (v16u8)__msa_ld_b((void*)src_raw, 32);
>     vec0 = (v16u8)__msa_sldi_b((v16i8)src1, (v16i8)src0, 12);
>     vec1 = (v16u8)__msa_sldi_b((v16i8)src2, (v16i8)src1, 8);
>     vec2 = (v16u8)__msa_sldi_b((v16i8)src2, (v16i8)src2, 4);
>@@ -1666,8 +1666,8 @@ void ARGB1555ToYRow_MSA(const uint8_t* src_argb1555,
>   v8u16 const_0x1080 = (v8u16)__msa_fill_h(0x1080);
> 
>   for (x = 0; x < width; x += 16) {
>-    src0 = (v8u16)__msa_ld_b((v8i16*)src_argb1555, 0);
>-    src1 = (v8u16)__msa_ld_b((v8i16*)src_argb1555, 16);
>+    src0 = (v8u16)__msa_ld_b((void*)src_argb1555, 0);
>+    src1 = (v8u16)__msa_ld_b((void*)src_argb1555, 16);
>     vec0 = src0 & const_0x1F;
>     vec1 = src1 & const_0x1F;
>     src0 = (v8u16)__msa_srai_h((v8i16)src0, 5);
>@@ -1725,8 +1725,8 @@ void RGB565ToYRow_MSA(const uint8_t* src_rgb565, uint8_t* dst_y, int width) {
>   v8u16 const_0xF800 = (v8u16)__msa_fill_h(0xF800);
> 
>   for (x = 0; x < width; x += 16) {
>-    src0 = (v8u16)__msa_ld_b((v8i16*)src_rgb565, 0);
>-    src1 = (v8u16)__msa_ld_b((v8i16*)src_rgb565, 16);
>+    src0 = (v8u16)__msa_ld_b((void*)src_rgb565, 0);
>+    src1 = (v8u16)__msa_ld_b((void*)src_rgb565, 16);
>     vec0 = src0 & const_0x1F;
>     vec1 = src0 & const_0x7E0;
>     vec2 = src0 & const_0xF800;
>@@ -1789,9 +1789,9 @@ void RGB24ToYRow_MSA(const uint8_t* src_argb0, uint8_t* dst_y, int width) {
>   v16i8 zero = {0};
> 
>   for (x = 0; x < width; x += 16) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 16);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 32);
>+    src0 = (v16u8)__msa_ld_b((void*)src_argb0, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_argb0, 16);
>+    src2 = (v16u8)__msa_ld_b((void*)src_argb0, 32);
>     reg0 = (v16u8)__msa_vshf_b(mask0, zero, (v16i8)src0);
>     reg1 = (v16u8)__msa_vshf_b(mask1, (v16i8)src1, (v16i8)src0);
>     reg2 = (v16u8)__msa_vshf_b(mask2, (v16i8)src2, (v16i8)src1);
>@@ -1830,9 +1830,9 @@ void RAWToYRow_MSA(const uint8_t* src_argb0, uint8_t* dst_y, int width) {
>   v16i8 zero = {0};
> 
>   for (x = 0; x < width; x += 16) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 16);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 32);
>+    src0 = (v16u8)__msa_ld_b((void*)src_argb0, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_argb0, 16);
>+    src2 = (v16u8)__msa_ld_b((void*)src_argb0, 32);
>     reg0 = (v16u8)__msa_vshf_b(mask0, zero, (v16i8)src0);
>     reg1 = (v16u8)__msa_vshf_b(mask1, (v16i8)src1, (v16i8)src0);
>     reg2 = (v16u8)__msa_vshf_b(mask2, (v16i8)src2, (v16i8)src1);
>@@ -1877,10 +1877,10 @@ void ARGB1555ToUVRow_MSA(const uint8_t* src_argb1555,
>   v8u16 const_0x1F = (v8u16)__msa_ldi_h(0x1F);
> 
>   for (x = 0; x < width; x += 16) {
>-    src0 = (v8u16)__msa_ld_b((v8i16*)s, 0);
>-    src1 = (v8u16)__msa_ld_b((v8i16*)s, 16);
>-    src2 = (v8u16)__msa_ld_b((v8i16*)t, 0);
>-    src3 = (v8u16)__msa_ld_b((v8i16*)t, 16);
>+    src0 = (v8u16)__msa_ld_b((void*)s, 0);
>+    src1 = (v8u16)__msa_ld_b((void*)s, 16);
>+    src2 = (v8u16)__msa_ld_b((void*)t, 0);
>+    src3 = (v8u16)__msa_ld_b((void*)t, 16);
>     vec0 = src0 & const_0x1F;
>     vec1 = src1 & const_0x1F;
>     vec0 += src2 & const_0x1F;
>@@ -1959,10 +1959,10 @@ void RGB565ToUVRow_MSA(const uint8_t* src_rgb565,
>   v8u16 const_0x3F = (v8u16)__msa_fill_h(0x3F);
> 
>   for (x = 0; x < width; x += 16) {
>-    src0 = (v8u16)__msa_ld_b((v8i16*)s, 0);
>-    src1 = (v8u16)__msa_ld_b((v8i16*)s, 16);
>-    src2 = (v8u16)__msa_ld_b((v8i16*)t, 0);
>-    src3 = (v8u16)__msa_ld_b((v8i16*)t, 16);
>+    src0 = (v8u16)__msa_ld_b((void*)s, 0);
>+    src1 = (v8u16)__msa_ld_b((void*)s, 16);
>+    src2 = (v8u16)__msa_ld_b((void*)t, 0);
>+    src3 = (v8u16)__msa_ld_b((void*)t, 16);
>     vec0 = src0 & const_0x1F;
>     vec1 = src1 & const_0x1F;
>     vec0 += src2 & const_0x1F;
>@@ -2041,12 +2041,12 @@ void RGB24ToUVRow_MSA(const uint8_t* src_rgb0,
>   v16i8 zero = {0};
> 
>   for (x = 0; x < width; x += 16) {
>-    inp0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
>-    inp1 = (v16u8)__msa_ld_b((v16i8*)s, 16);
>-    inp2 = (v16u8)__msa_ld_b((v16i8*)s, 32);
>-    inp3 = (v16u8)__msa_ld_b((v16i8*)t, 0);
>-    inp4 = (v16u8)__msa_ld_b((v16i8*)t, 16);
>-    inp5 = (v16u8)__msa_ld_b((v16i8*)t, 32);
>+    inp0 = (v16u8)__msa_ld_b((void*)s, 0);
>+    inp1 = (v16u8)__msa_ld_b((void*)s, 16);
>+    inp2 = (v16u8)__msa_ld_b((void*)s, 32);
>+    inp3 = (v16u8)__msa_ld_b((void*)t, 0);
>+    inp4 = (v16u8)__msa_ld_b((void*)t, 16);
>+    inp5 = (v16u8)__msa_ld_b((void*)t, 32);
>     src1 = (v16u8)__msa_sldi_b((v16i8)inp1, (v16i8)inp0, 12);
>     src5 = (v16u8)__msa_sldi_b((v16i8)inp4, (v16i8)inp3, 12);
>     src2 = (v16u8)__msa_sldi_b((v16i8)inp2, (v16i8)inp1, 8);
>@@ -2146,12 +2146,12 @@ void RAWToUVRow_MSA(const uint8_t* src_rgb0,
>   v16i8 zero = {0};
> 
>   for (x = 0; x < width; x += 16) {
>-    inp0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
>-    inp1 = (v16u8)__msa_ld_b((v16i8*)s, 16);
>-    inp2 = (v16u8)__msa_ld_b((v16i8*)s, 32);
>-    inp3 = (v16u8)__msa_ld_b((v16i8*)t, 0);
>-    inp4 = (v16u8)__msa_ld_b((v16i8*)t, 16);
>-    inp5 = (v16u8)__msa_ld_b((v16i8*)t, 32);
>+    inp0 = (v16u8)__msa_ld_b((void*)s, 0);
>+    inp1 = (v16u8)__msa_ld_b((void*)s, 16);
>+    inp2 = (v16u8)__msa_ld_b((void*)s, 32);
>+    inp3 = (v16u8)__msa_ld_b((void*)t, 0);
>+    inp4 = (v16u8)__msa_ld_b((void*)t, 16);
>+    inp5 = (v16u8)__msa_ld_b((void*)t, 32);
>     src1 = (v16u8)__msa_sldi_b((v16i8)inp1, (v16i8)inp0, 12);
>     src5 = (v16u8)__msa_sldi_b((v16i8)inp4, (v16i8)inp3, 12);
>     src2 = (v16u8)__msa_sldi_b((v16i8)inp2, (v16i8)inp1, 8);
>@@ -2353,8 +2353,8 @@ void SobelRow_MSA(const uint8_t* src_sobelx,
>   v16u8 alpha = (v16u8)__msa_ldi_b(ALPHA_VAL);
> 
>   for (x = 0; x < width; x += 16) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_sobelx, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_sobely, 0);
>+    src0 = (v16u8)__msa_ld_b((void*)src_sobelx, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_sobely, 0);
>     vec0 = __msa_adds_u_b(src0, src1);
>     dst0 = (v16u8)__msa_vshf_b(mask0, (v16i8)alpha, (v16i8)vec0);
>     dst1 = (v16u8)__msa_vshf_b(mask1, (v16i8)alpha, (v16i8)vec0);
>@@ -2375,10 +2375,10 @@ void SobelToPlaneRow_MSA(const uint8_t* src_sobelx,
>   v16u8 src0, src1, src2, src3, dst0, dst1;
> 
>   for (x = 0; x < width; x += 32) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_sobelx, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_sobelx, 16);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)src_sobely, 0);
>-    src3 = (v16u8)__msa_ld_b((v16i8*)src_sobely, 16);
>+    src0 = (v16u8)__msa_ld_b((void*)src_sobelx, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_sobelx, 16);
>+    src2 = (v16u8)__msa_ld_b((void*)src_sobely, 0);
>+    src3 = (v16u8)__msa_ld_b((void*)src_sobely, 16);
>     dst0 = __msa_adds_u_b(src0, src2);
>     dst1 = __msa_adds_u_b(src1, src3);
>     ST_UB2(dst0, dst1, dst_y, 16);
>@@ -2398,8 +2398,8 @@ void SobelXYRow_MSA(const uint8_t* src_sobelx,
>   v16u8 alpha = (v16u8)__msa_ldi_b(ALPHA_VAL);
> 
>   for (x = 0; x < width; x += 16) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_sobelx, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_sobely, 0);
>+    src0 = (v16u8)__msa_ld_b((void*)src_sobelx, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_sobely, 0);
>     vec0 = __msa_adds_u_b(src0, src1);
>     vec1 = (v16u8)__msa_ilvr_b((v16i8)src0, (v16i8)src1);
>     vec2 = (v16u8)__msa_ilvl_b((v16i8)src0, (v16i8)src1);
>@@ -2424,10 +2424,10 @@ void ARGBToYJRow_MSA(const uint8_t* src_argb0, uint8_t* dst_y, int width) {
>   v8u16 const_0x40 = (v8u16)__msa_fill_h(0x40);
> 
>   for (x = 0; x < width; x += 16) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 16);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 32);
>-    src3 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 48);
>+    src0 = (v16u8)__msa_ld_b((void*)src_argb0, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_argb0, 16);
>+    src2 = (v16u8)__msa_ld_b((void*)src_argb0, 32);
>+    src3 = (v16u8)__msa_ld_b((void*)src_argb0, 48);
>     ARGBTOY(src0, src1, src2, src3, const_0x4B0F, const_0x26, const_0x40, 7,
>             dst0);
>     ST_UB(dst0, dst_y);
>@@ -2444,10 +2444,10 @@ void BGRAToYRow_MSA(const uint8_t* src_argb0, uint8_t* dst_y, int width) {
>   v8u16 const_0x1080 = (v8u16)__msa_fill_h(0x1080);
> 
>   for (x = 0; x < width; x += 16) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 16);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 32);
>-    src3 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 48);
>+    src0 = (v16u8)__msa_ld_b((void*)src_argb0, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_argb0, 16);
>+    src2 = (v16u8)__msa_ld_b((void*)src_argb0, 32);
>+    src3 = (v16u8)__msa_ld_b((void*)src_argb0, 48);
>     ARGBTOY(src0, src1, src2, src3, const_0x4200, const_0x1981, const_0x1080, 8,
>             dst0);
>     ST_UB(dst0, dst_y);
>@@ -2464,10 +2464,10 @@ void ABGRToYRow_MSA(const uint8_t* src_argb0, uint8_t* dst_y, int width) {
>   v8u16 const_0x1080 = (v8u16)__msa_fill_h(0x1080);
> 
>   for (x = 0; x < width; x += 16) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 16);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 32);
>-    src3 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 48);
>+    src0 = (v16u8)__msa_ld_b((void*)src_argb0, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_argb0, 16);
>+    src2 = (v16u8)__msa_ld_b((void*)src_argb0, 32);
>+    src3 = (v16u8)__msa_ld_b((void*)src_argb0, 48);
>     ARGBTOY(src0, src1, src2, src3, const_0x8142, const_0x19, const_0x1080, 8,
>             dst0);
>     ST_UB(dst0, dst_y);
>@@ -2484,10 +2484,10 @@ void RGBAToYRow_MSA(const uint8_t* src_argb0, uint8_t* dst_y, int width) {
>   v8u16 const_0x1080 = (v8u16)__msa_fill_h(0x1080);
> 
>   for (x = 0; x < width; x += 16) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 16);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 32);
>-    src3 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 48);
>+    src0 = (v16u8)__msa_ld_b((void*)src_argb0, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_argb0, 16);
>+    src2 = (v16u8)__msa_ld_b((void*)src_argb0, 32);
>+    src3 = (v16u8)__msa_ld_b((void*)src_argb0, 48);
>     ARGBTOY(src0, src1, src2, src3, const_0x1900, const_0x4281, const_0x1080, 8,
>             dst0);
>     ST_UB(dst0, dst_y);
>@@ -2518,14 +2518,14 @@ void ARGBToUVJRow_MSA(const uint8_t* src_rgb0,
>   v8u16 const_0x8080 = (v8u16)__msa_fill_h(0x8080);
> 
>   for (x = 0; x < width; x += 32) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)s, 16);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)s, 32);
>-    src3 = (v16u8)__msa_ld_b((v16i8*)s, 48);
>-    src4 = (v16u8)__msa_ld_b((v16i8*)t, 0);
>-    src5 = (v16u8)__msa_ld_b((v16i8*)t, 16);
>-    src6 = (v16u8)__msa_ld_b((v16i8*)t, 32);
>-    src7 = (v16u8)__msa_ld_b((v16i8*)t, 48);
>+    src0 = (v16u8)__msa_ld_b((void*)s, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)s, 16);
>+    src2 = (v16u8)__msa_ld_b((void*)s, 32);
>+    src3 = (v16u8)__msa_ld_b((void*)s, 48);
>+    src4 = (v16u8)__msa_ld_b((void*)t, 0);
>+    src5 = (v16u8)__msa_ld_b((void*)t, 16);
>+    src6 = (v16u8)__msa_ld_b((void*)t, 32);
>+    src7 = (v16u8)__msa_ld_b((void*)t, 48);
>     src0 = __msa_aver_u_b(src0, src4);
>     src1 = __msa_aver_u_b(src1, src5);
>     src2 = __msa_aver_u_b(src2, src6);
>@@ -2536,14 +2536,14 @@ void ARGBToUVJRow_MSA(const uint8_t* src_rgb0,
>     src7 = (v16u8)__msa_pckod_w((v4i32)src3, (v4i32)src2);
>     vec0 = __msa_aver_u_b(src4, src6);
>     vec1 = __msa_aver_u_b(src5, src7);
>-    src0 = (v16u8)__msa_ld_b((v16i8*)s, 64);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)s, 80);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)s, 96);
>-    src3 = (v16u8)__msa_ld_b((v16i8*)s, 112);
>-    src4 = (v16u8)__msa_ld_b((v16i8*)t, 64);
>-    src5 = (v16u8)__msa_ld_b((v16i8*)t, 80);
>-    src6 = (v16u8)__msa_ld_b((v16i8*)t, 96);
>-    src7 = (v16u8)__msa_ld_b((v16i8*)t, 112);
>+    src0 = (v16u8)__msa_ld_b((void*)s, 64);
>+    src1 = (v16u8)__msa_ld_b((void*)s, 80);
>+    src2 = (v16u8)__msa_ld_b((void*)s, 96);
>+    src3 = (v16u8)__msa_ld_b((void*)s, 112);
>+    src4 = (v16u8)__msa_ld_b((void*)t, 64);
>+    src5 = (v16u8)__msa_ld_b((void*)t, 80);
>+    src6 = (v16u8)__msa_ld_b((void*)t, 96);
>+    src7 = (v16u8)__msa_ld_b((void*)t, 112);
>     src0 = __msa_aver_u_b(src0, src4);
>     src1 = __msa_aver_u_b(src1, src5);
>     src2 = __msa_aver_u_b(src2, src6);
>@@ -2746,7 +2746,7 @@ void I400ToARGBRow_MSA(const uint8_t* src_y, uint8_t* dst_argb, int width) {
>   v8i16 zero = {0};
> 
>   for (x = 0; x < width; x += 16) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_y, 0);
>+    src0 = (v16u8)__msa_ld_b((void*)src_y, 0);
>     vec0 = (v8i16)__msa_ilvr_b((v16i8)src0, (v16i8)src0);
>     vec1 = (v8i16)__msa_ilvl_b((v16i8)src0, (v16i8)src0);
>     reg0 = (v4i32)__msa_ilvr_h(zero, vec0);
>@@ -2792,7 +2792,7 @@ void J400ToARGBRow_MSA(const uint8_t* src_y, uint8_t* dst_argb, int width) {
>   v16u8 alpha = (v16u8)__msa_ldi_b(ALPHA_VAL);
> 
>   for (x = 0; x < width; x += 16) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_y, 0);
>+    src0 = (v16u8)__msa_ld_b((void*)src_y, 0);
>     vec0 = (v16u8)__msa_ilvr_b((v16i8)src0, (v16i8)src0);
>     vec1 = (v16u8)__msa_ilvl_b((v16i8)src0, (v16i8)src0);
>     vec2 = (v16u8)__msa_ilvr_b((v16i8)alpha, (v16i8)src0);
>@@ -2824,7 +2824,7 @@ void YUY2ToARGBRow_MSA(const uint8_t* src_yuy2,
>   vec_ugvg = (v4i32)__msa_ilvev_h((v8i16)vec_vg, (v8i16)vec_ug);
> 
>   for (x = 0; x < width; x += 8) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_yuy2, 0);
>+    src0 = (v16u8)__msa_ld_b((void*)src_yuy2, 0);
>     src1 = (v16u8)__msa_pckev_b((v16i8)src0, (v16i8)src0);
>     src2 = (v16u8)__msa_pckod_b((v16i8)src0, (v16i8)src0);
>     YUVTORGB(src1, src2, vec_ubvr, vec_ugvg, vec_bb, vec_bg, vec_br, vec_yg,
>@@ -2852,7 +2852,7 @@ void UYVYToARGBRow_MSA(const uint8_t* src_uyvy,
>   vec_ugvg = (v4i32)__msa_ilvev_h((v8i16)vec_vg, (v8i16)vec_ug);
> 
>   for (x = 0; x < width; x += 8) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_uyvy, 0);
>+    src0 = (v16u8)__msa_ld_b((void*)src_uyvy, 0);
>     src1 = (v16u8)__msa_pckod_b((v16i8)src0, (v16i8)src0);
>     src2 = (v16u8)__msa_pckev_b((v16i8)src0, (v16i8)src0);
>     YUVTORGB(src1, src2, vec_ubvr, vec_ugvg, vec_bb, vec_bg, vec_br, vec_yg,
>@@ -2884,10 +2884,10 @@ void InterpolateRow_MSA(uint8_t* dst_ptr,
> 
>   if (128 == y1_fraction) {
>     for (x = 0; x < width; x += 32) {
>-      src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
>-      src1 = (v16u8)__msa_ld_b((v16i8*)s, 16);
>-      src2 = (v16u8)__msa_ld_b((v16i8*)t, 0);
>-      src3 = (v16u8)__msa_ld_b((v16i8*)t, 16);
>+      src0 = (v16u8)__msa_ld_b((void*)s, 0);
>+      src1 = (v16u8)__msa_ld_b((void*)s, 16);
>+      src2 = (v16u8)__msa_ld_b((void*)t, 0);
>+      src3 = (v16u8)__msa_ld_b((void*)t, 16);
>       dst0 = __msa_aver_u_b(src0, src2);
>       dst1 = __msa_aver_u_b(src1, src3);
>       ST_UB2(dst0, dst1, dst_ptr, 16);
>@@ -2902,10 +2902,10 @@ void InterpolateRow_MSA(uint8_t* dst_ptr,
>   y_frac = (v8u16)__msa_fill_h(y_fractions);
> 
>   for (x = 0; x < width; x += 32) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)s, 16);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)t, 0);
>-    src3 = (v16u8)__msa_ld_b((v16i8*)t, 16);
>+    src0 = (v16u8)__msa_ld_b((void*)s, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)s, 16);
>+    src2 = (v16u8)__msa_ld_b((void*)t, 0);
>+    src3 = (v16u8)__msa_ld_b((void*)t, 16);
>     vec0 = (v8u16)__msa_ilvr_b((v16i8)src2, (v16i8)src0);
>     vec1 = (v8u16)__msa_ilvl_b((v16i8)src2, (v16i8)src0);
>     vec2 = (v8u16)__msa_ilvr_b((v16i8)src3, (v16i8)src1);
>@@ -2947,9 +2947,9 @@ void RAWToRGB24Row_MSA(const uint8_t* src_raw, uint8_t* dst_rgb24, int width) {
>                      24, 23, 28, 27, 26, 31, 30, 29};
> 
>   for (x = 0; x < width; x += 16) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_raw, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_raw, 16);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)src_raw, 32);
>+    src0 = (v16u8)__msa_ld_b((void*)src_raw, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_raw, 16);
>+    src2 = (v16u8)__msa_ld_b((void*)src_raw, 32);
>     src3 = (v16u8)__msa_sldi_b((v16i8)src1, (v16i8)src0, 8);
>     src4 = (v16u8)__msa_sldi_b((v16i8)src2, (v16i8)src1, 8);
>     dst0 = (v16u8)__msa_vshf_b(shuffler0, (v16i8)src1, (v16i8)src0);
>@@ -2970,8 +2970,8 @@ void MergeUVRow_MSA(const uint8_t* src_u,
>   v16u8 src0, src1, dst0, dst1;
> 
>   for (x = 0; x < width; x += 16) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_u, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_v, 0);
>+    src0 = (v16u8)__msa_ld_b((void*)src_u, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_v, 0);
>     dst0 = (v16u8)__msa_ilvr_b((v16i8)src1, (v16i8)src0);
>     dst1 = (v16u8)__msa_ilvl_b((v16i8)src1, (v16i8)src0);
>     ST_UB2(dst0, dst1, dst_uv, 16);
>@@ -2988,10 +2988,10 @@ void ARGBExtractAlphaRow_MSA(const uint8_t* src_argb,
>   v16u8 src0, src1, src2, src3, vec0, vec1, dst0;
> 
>   for (i = 0; i < width; i += 16) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_argb, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_argb, 16);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)src_argb, 32);
>-    src3 = (v16u8)__msa_ld_b((v16i8*)src_argb, 48);
>+    src0 = (v16u8)__msa_ld_b((void*)src_argb, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_argb, 16);
>+    src2 = (v16u8)__msa_ld_b((void*)src_argb, 32);
>+    src3 = (v16u8)__msa_ld_b((void*)src_argb, 48);
>     vec0 = (v16u8)__msa_pckod_b((v16i8)src1, (v16i8)src0);
>     vec1 = (v16u8)__msa_pckod_b((v16i8)src3, (v16i8)src2);
>     dst0 = (v16u8)__msa_pckod_b((v16i8)vec1, (v16i8)vec0);
>@@ -3015,10 +3015,10 @@ void ARGBBlendRow_MSA(const uint8_t* src_argb0,
>   v16i8 zero = {0};
> 
>   for (x = 0; x < width; x += 8) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_argb0, 16);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)src_argb1, 0);
>-    src3 = (v16u8)__msa_ld_b((v16i8*)src_argb1, 16);
>+    src0 = (v16u8)__msa_ld_b((void*)src_argb0, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_argb0, 16);
>+    src2 = (v16u8)__msa_ld_b((void*)src_argb1, 0);
>+    src3 = (v16u8)__msa_ld_b((void*)src_argb1, 16);
>     vec0 = (v8u16)__msa_ilvr_b(zero, (v16i8)src0);
>     vec1 = (v8u16)__msa_ilvl_b(zero, (v16i8)src0);
>     vec2 = (v8u16)__msa_ilvr_b(zero, (v16i8)src1);
>@@ -3083,10 +3083,10 @@ void ARGBQuantizeRow_MSA(uint8_t* dst_argb,
>   v16i8 zero = {0};
> 
>   for (x = 0; x < width; x += 8) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)dst_argb, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)dst_argb, 16);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)dst_argb, 32);
>-    src3 = (v16u8)__msa_ld_b((v16i8*)dst_argb, 48);
>+    src0 = (v16u8)__msa_ld_b((void*)dst_argb, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)dst_argb, 16);
>+    src2 = (v16u8)__msa_ld_b((void*)dst_argb, 32);
>+    src3 = (v16u8)__msa_ld_b((void*)dst_argb, 48);
>     vec0 = (v8i16)__msa_ilvr_b(zero, (v16i8)src0);
>     vec1 = (v8i16)__msa_ilvl_b(zero, (v16i8)src0);
>     vec2 = (v8i16)__msa_ilvr_b(zero, (v16i8)src1);
>@@ -3186,13 +3186,13 @@ void ARGBColorMatrixRow_MSA(const uint8_t* src_argb,
>   v16i8 zero = {0};
>   v8i16 max = __msa_ldi_h(255);
> 
>-  src0 = __msa_ld_b((v16i8*)matrix_argb, 0);
>+  src0 = __msa_ld_b((void*)matrix_argb, 0);
>   vec0 = (v8i16)__msa_ilvr_b(zero, src0);
>   vec1 = (v8i16)__msa_ilvl_b(zero, src0);
> 
>   for (x = 0; x < width; x += 8) {
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_argb, 0);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)src_argb, 16);
>+    src1 = (v16u8)__msa_ld_b((void*)src_argb, 0);
>+    src2 = (v16u8)__msa_ld_b((void*)src_argb, 16);
>     vec2 = (v8i16)__msa_ilvr_b(zero, (v16i8)src1);
>     vec3 = (v8i16)__msa_ilvl_b(zero, (v16i8)src1);
>     vec4 = (v8i16)__msa_ilvr_b(zero, (v16i8)src2);
>@@ -3289,10 +3289,10 @@ void SplitUVRow_MSA(const uint8_t* src_uv,
>   v16u8 src0, src1, src2, src3, dst0, dst1, dst2, dst3;
> 
>   for (x = 0; x < width; x += 32) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_uv, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_uv, 16);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)src_uv, 32);
>-    src3 = (v16u8)__msa_ld_b((v16i8*)src_uv, 48);
>+    src0 = (v16u8)__msa_ld_b((void*)src_uv, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_uv, 16);
>+    src2 = (v16u8)__msa_ld_b((void*)src_uv, 32);
>+    src3 = (v16u8)__msa_ld_b((void*)src_uv, 48);
>     dst0 = (v16u8)__msa_pckev_b((v16i8)src1, (v16i8)src0);
>     dst1 = (v16u8)__msa_pckev_b((v16i8)src3, (v16i8)src2);
>     dst2 = (v16u8)__msa_pckod_b((v16i8)src1, (v16i8)src0);
>@@ -3329,10 +3329,10 @@ void MirrorUVRow_MSA(const uint8_t* src_uv,
> 
>   for (x = 0; x < width; x += 32) {
>     src_uv -= 64;
>-    src2 = (v16u8)__msa_ld_b((v16i8*)src_uv, 0);
>-    src3 = (v16u8)__msa_ld_b((v16i8*)src_uv, 16);
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_uv, 32);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_uv, 48);
>+    src2 = (v16u8)__msa_ld_b((void*)src_uv, 0);
>+    src3 = (v16u8)__msa_ld_b((void*)src_uv, 16);
>+    src0 = (v16u8)__msa_ld_b((void*)src_uv, 32);
>+    src1 = (v16u8)__msa_ld_b((void*)src_uv, 48);
>     dst0 = (v16u8)__msa_vshf_b(mask1, (v16i8)src1, (v16i8)src0);
>     dst1 = (v16u8)__msa_vshf_b(mask1, (v16i8)src3, (v16i8)src2);
>     dst2 = (v16u8)__msa_vshf_b(mask0, (v16i8)src1, (v16i8)src0);
>@@ -3359,12 +3359,12 @@ void SobelXRow_MSA(const uint8_t* src_y0,
>   v8i16 max = __msa_ldi_h(255);
> 
>   for (x = 0; x < width; x += 16) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_y0, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_y0, 16);
>-    src2 = (v16u8)__msa_ld_b((v16i8*)src_y1, 0);
>-    src3 = (v16u8)__msa_ld_b((v16i8*)src_y1, 16);
>-    src4 = (v16u8)__msa_ld_b((v16i8*)src_y2, 0);
>-    src5 = (v16u8)__msa_ld_b((v16i8*)src_y2, 16);
>+    src0 = (v16u8)__msa_ld_b((void*)src_y0, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_y0, 16);
>+    src2 = (v16u8)__msa_ld_b((void*)src_y1, 0);
>+    src3 = (v16u8)__msa_ld_b((void*)src_y1, 16);
>+    src4 = (v16u8)__msa_ld_b((void*)src_y2, 0);
>+    src5 = (v16u8)__msa_ld_b((void*)src_y2, 16);
>     vec0 = (v8i16)__msa_vshf_b(mask0, (v16i8)src1, (v16i8)src0);
>     vec1 = (v8i16)__msa_vshf_b(mask1, (v16i8)src1, (v16i8)src0);
>     vec2 = (v8i16)__msa_vshf_b(mask0, (v16i8)src3, (v16i8)src2);
>@@ -3409,8 +3409,8 @@ void SobelYRow_MSA(const uint8_t* src_y0,
>   v8i16 max = __msa_ldi_h(255);
> 
>   for (x = 0; x < width; x += 16) {
>-    src0 = (v16u8)__msa_ld_b((v16i8*)src_y0, 0);
>-    src1 = (v16u8)__msa_ld_b((v16i8*)src_y1, 0);
>+    src0 = (v16u8)__msa_ld_b((void*)src_y0, 0);
>+    src1 = (v16u8)__msa_ld_b((void*)src_y1, 0);
>     vec0 = (v8i16)__msa_ilvr_b((v16i8)zero, (v16i8)src0);
>     vec1 = (v8i16)__msa_ilvl_b((v16i8)zero, (v16i8)src0);
>     vec2 = (v8i16)__msa_ilvr_b((v16i8)zero, (v16i8)src1);
>@@ -3457,10 +3457,10 @@ void HalfFloatRow_MSA(const uint16_t* src,
>   mult_vec = (v4f32)__msa_splati_w((v4i32)mult_vec, 0);
> 
>   for (i = 0; i < width; i += 32) {
>-    src0 = (v8u16)__msa_ld_h((v8i16*)src, 0);
>-    src1 = (v8u16)__msa_ld_h((v8i16*)src, 16);
>-    src2 = (v8u16)__msa_ld_h((v8i16*)src, 32);
>-    src3 = (v8u16)__msa_ld_h((v8i16*)src, 48);
>+    src0 = (v8u16)__msa_ld_h((void*)src, 0);
>+    src1 = (v8u16)__msa_ld_h((void*)src, 16);
>+    src2 = (v8u16)__msa_ld_h((void*)src, 32);
>+    src3 = (v8u16)__msa_ld_h((void*)src, 48);
>     vec0 = (v4u32)__msa_ilvr_h(zero, (v8i16)src0);
>     vec1 = (v4u32)__msa_ilvl_h(zero, (v8i16)src0);
>     vec2 = (v4u32)__msa_ilvr_h(zero, (v8i16)src1);
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/row_neon.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/row_neon.cc
>index 8b6c19520722988b59ba061e1b0ec5a9d889c66d..ff87e74c62c9f909b096d84ae53edf33bad3ffe0 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/row_neon.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/row_neon.cc
>@@ -2604,12 +2604,15 @@ void SobelYRow_NEON(const uint8_t* src_y0,
>       );
> }
> 
>+// %y passes a float as a scalar vector for vector * scalar multiply.
>+// the regoster must be d0 to d15 and indexed with [0] or [1] to access
>+// the float in the first or second float of the d-reg
>+
> void HalfFloat1Row_NEON(const uint16_t* src,
>                         uint16_t* dst,
>                         float /*unused*/,
>                         int width) {
>   asm volatile(
>-      "vdup.32    q0, %3                         \n"
> 
>       "1:                                        \n"
>       "vld1.8     {q1}, [%0]!                    \n"  // load 8 shorts
>@@ -2618,8 +2621,8 @@ void HalfFloat1Row_NEON(const uint16_t* src,
>       "vmovl.u16  q3, d3                         \n"
>       "vcvt.f32.u32  q2, q2                      \n"  // 8 floats
>       "vcvt.f32.u32  q3, q3                      \n"
>-      "vmul.f32   q2, q2, q0                     \n"  // adjust exponent
>-      "vmul.f32   q3, q3, q0                     \n"
>+      "vmul.f32   q2, q2, %y3                    \n"  // adjust exponent
>+      "vmul.f32   q3, q3, %y3                    \n"
>       "vqshrn.u32 d2, q2, #13                    \n"  // isolate halffloat
>       "vqshrn.u32 d3, q3, #13                    \n"
>       "vst1.8     {q1}, [%1]!                    \n"
>@@ -2627,17 +2630,15 @@ void HalfFloat1Row_NEON(const uint16_t* src,
>       : "+r"(src),              // %0
>         "+r"(dst),              // %1
>         "+r"(width)             // %2
>-      : "r"(1.9259299444e-34f)  // %3
>-      : "cc", "memory", "q0", "q1", "q2", "q3");
>+      : "w"(1.9259299444e-34f)  // %3
>+      : "cc", "memory", "q1", "q2", "q3");
> }
> 
>-// TODO(fbarchard): multiply by element.
> void HalfFloatRow_NEON(const uint16_t* src,
>                        uint16_t* dst,
>                        float scale,
>                        int width) {
>   asm volatile(
>-      "vdup.32    q0, %3                         \n"
> 
>       "1:                                        \n"
>       "vld1.8     {q1}, [%0]!                    \n"  // load 8 shorts
>@@ -2646,8 +2647,8 @@ void HalfFloatRow_NEON(const uint16_t* src,
>       "vmovl.u16  q3, d3                         \n"
>       "vcvt.f32.u32  q2, q2                      \n"  // 8 floats
>       "vcvt.f32.u32  q3, q3                      \n"
>-      "vmul.f32   q2, q2, q0                     \n"  // adjust exponent
>-      "vmul.f32   q3, q3, q0                     \n"
>+      "vmul.f32   q2, q2, %y3                    \n"  // adjust exponent
>+      "vmul.f32   q3, q3, %y3                    \n"
>       "vqshrn.u32 d2, q2, #13                    \n"  // isolate halffloat
>       "vqshrn.u32 d3, q3, #13                    \n"
>       "vst1.8     {q1}, [%1]!                    \n"
>@@ -2655,8 +2656,8 @@ void HalfFloatRow_NEON(const uint16_t* src,
>       : "+r"(src),                      // %0
>         "+r"(dst),                      // %1
>         "+r"(width)                     // %2
>-      : "r"(scale * 1.9259299444e-34f)  // %3
>-      : "cc", "memory", "q0", "q1", "q2", "q3");
>+      : "w"(scale * 1.9259299444e-34f)  // %3
>+      : "cc", "memory", "q1", "q2", "q3");
> }
> 
> void ByteToFloatRow_NEON(const uint8_t* src,
>@@ -2664,7 +2665,6 @@ void ByteToFloatRow_NEON(const uint8_t* src,
>                          float scale,
>                          int width) {
>   asm volatile(
>-      "vdup.32    q0, %3                         \n"
> 
>       "1:                                        \n"
>       "vld1.8     {d2}, [%0]!                    \n"  // load 8 bytes
>@@ -2674,15 +2674,15 @@ void ByteToFloatRow_NEON(const uint8_t* src,
>       "vmovl.u16  q3, d3                         \n"
>       "vcvt.f32.u32  q2, q2                      \n"  // 8 floats
>       "vcvt.f32.u32  q3, q3                      \n"
>-      "vmul.f32   q2, q2, d0[0]                  \n"  // scale
>-      "vmul.f32   q3, q3, d0[0]                  \n"
>+      "vmul.f32   q2, q2, %y3                    \n"  // scale
>+      "vmul.f32   q3, q3, %y3                    \n"
>       "vst1.8     {q2, q3}, [%1]!                \n"  // store 8 floats
>       "bgt        1b                             \n"
>       : "+r"(src),   // %0
>         "+r"(dst),   // %1
>         "+r"(width)  // %2
>-      : "r"(scale)   // %3
>-      : "cc", "memory", "q0", "q1", "q2", "q3");
>+      : "w"(scale)   // %3
>+      : "cc", "memory", "q1", "q2", "q3");
> }
> 
> #endif  // !defined(LIBYUV_DISABLE_NEON) && defined(__ARM_NEON__)..
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale.cc
>index 2cfa1c6cb1c4b8a4be72bb590372aa7b54ccd7ee..a8db93fde48bb2ea0d9f44ccf61eea74767a9946 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale.cc
>@@ -118,6 +118,21 @@ static void ScalePlaneDown2(int src_width,
>     }
>   }
> #endif
>+#if defined(HAS_SCALEROWDOWN2_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ScaleRowDown2 =
>+        filtering == kFilterNone
>+            ? ScaleRowDown2_Any_MMI
>+            : (filtering == kFilterLinear ? ScaleRowDown2Linear_Any_MMI
>+                                          : ScaleRowDown2Box_Any_MMI);
>+    if (IS_ALIGNED(dst_width, 8)) {
>+      ScaleRowDown2 = filtering == kFilterNone ? ScaleRowDown2_MMI
>+                                               : (filtering == kFilterLinear
>+                                                      ? ScaleRowDown2Linear_MMI
>+                                                      : ScaleRowDown2Box_MMI);
>+    }
>+  }
>+#endif
> 
>   if (filtering == kFilterLinear) {
>     src_stride = 0;
>@@ -169,6 +184,14 @@ static void ScalePlaneDown2_16(int src_width,
>                                           : ScaleRowDown2Box_16_SSE2);
>   }
> #endif
>+#if defined(HAS_SCALEROWDOWN2_16_MMI)
>+  if (TestCpuFlag(kCpuHasMMI) && IS_ALIGNED(dst_width, 4)) {
>+    ScaleRowDown2 = filtering == kFilterNone ? ScaleRowDown2_16_MMI
>+                                             : (filtering == kFilterLinear
>+                                                    ? ScaleRowDown2Linear_16_MMI
>+                                                    : ScaleRowDown2Box_16_MMI);
>+  }
>+#endif
> 
>   if (filtering == kFilterLinear) {
>     src_stride = 0;
>@@ -241,6 +264,15 @@ static void ScalePlaneDown4(int src_width,
>     }
>   }
> #endif
>+#if defined(HAS_SCALEROWDOWN4_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ScaleRowDown4 =
>+        filtering ? ScaleRowDown4Box_Any_MMI : ScaleRowDown4_Any_MMI;
>+    if (IS_ALIGNED(dst_width, 8)) {
>+      ScaleRowDown4 = filtering ? ScaleRowDown4Box_MMI : ScaleRowDown4_MMI;
>+    }
>+  }
>+#endif
> 
>   if (filtering == kFilterLinear) {
>     src_stride = 0;
>@@ -284,6 +316,11 @@ static void ScalePlaneDown4_16(int src_width,
>         filtering ? ScaleRowDown4Box_16_SSE2 : ScaleRowDown4_16_SSE2;
>   }
> #endif
>+#if defined(HAS_SCALEROWDOWN4_16_MMI)
>+  if (TestCpuFlag(kCpuHasMMI) && IS_ALIGNED(dst_width, 8)) {
>+    ScaleRowDown4 = filtering ? ScaleRowDown4Box_16_MMI : ScaleRowDown4_16_MMI;
>+  }
>+#endif
> 
>   if (filtering == kFilterLinear) {
>     src_stride = 0;
>@@ -849,6 +886,14 @@ static void ScalePlaneBox(int src_width,
>       }
>     }
> #endif
>+#if defined(HAS_SCALEADDROW_MMI)
>+    if (TestCpuFlag(kCpuHasMMI)) {
>+      ScaleAddRow = ScaleAddRow_Any_MMI;
>+      if (IS_ALIGNED(src_width, 8)) {
>+        ScaleAddRow = ScaleAddRow_MMI;
>+      }
>+    }
>+#endif
> 
>     for (j = 0; j < dst_height; ++j) {
>       int boxheight;
>@@ -904,6 +949,11 @@ static void ScalePlaneBox_16(int src_width,
>     }
> #endif
> 
>+#if defined(HAS_SCALEADDROW_16_MMI)
>+    if (TestCpuFlag(kCpuHasMMI) && IS_ALIGNED(src_width, 4)) {
>+      ScaleAddRow = ScaleAddRow_16_MMI;
>+    }
>+#endif
>     for (j = 0; j < dst_height; ++j) {
>       int boxheight;
>       int iy = y >> 16;
>@@ -988,6 +1038,14 @@ void ScalePlaneBilinearDown(int src_width,
>     }
>   }
> #endif
>+#if defined(HAS_INTERPOLATEROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    InterpolateRow = InterpolateRow_Any_MMI;
>+    if (IS_ALIGNED(src_width, 16)) {
>+      InterpolateRow = InterpolateRow_MMI;
>+    }
>+  }
>+#endif
> 
> #if defined(HAS_SCALEFILTERCOLS_SSSE3)
>   if (TestCpuFlag(kCpuHasSSSE3) && src_width < 32768) {
>@@ -1206,6 +1264,11 @@ void ScalePlaneBilinearUp(int src_width,
>     if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8)) {
>       ScaleFilterCols = ScaleColsUp2_SSE2;
>     }
>+#endif
>+#if defined(HAS_SCALECOLS_MMI)
>+    if (TestCpuFlag(kCpuHasMMI) && IS_ALIGNED(dst_width, 8)) {
>+      ScaleFilterCols = ScaleColsUp2_MMI;
>+    }
> #endif
>   }
> 
>@@ -1333,6 +1396,11 @@ void ScalePlaneBilinearUp_16(int src_width,
>     if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8)) {
>       ScaleFilterCols = ScaleColsUp2_16_SSE2;
>     }
>+#endif
>+#if defined(HAS_SCALECOLS_16_MMI)
>+    if (TestCpuFlag(kCpuHasMMI) && IS_ALIGNED(dst_width, 8)) {
>+      ScaleFilterCols = ScaleColsUp2_16_MMI;
>+    }
> #endif
>   }
> 
>@@ -1418,6 +1486,11 @@ static void ScalePlaneSimple(int src_width,
>     if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8)) {
>       ScaleCols = ScaleColsUp2_SSE2;
>     }
>+#endif
>+#if defined(HAS_SCALECOLS_MMI)
>+    if (TestCpuFlag(kCpuHasMMI) && IS_ALIGNED(dst_width, 8)) {
>+      ScaleCols = ScaleColsUp2_MMI;
>+    }
> #endif
>   }
> 
>@@ -1454,6 +1527,11 @@ static void ScalePlaneSimple_16(int src_width,
>     if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8)) {
>       ScaleCols = ScaleColsUp2_16_SSE2;
>     }
>+#endif
>+#if defined(HAS_SCALECOLS_16_MMI)
>+    if (TestCpuFlag(kCpuHasMMI) && IS_ALIGNED(dst_width, 8)) {
>+      ScaleCols = ScaleColsUp2_16_MMI;
>+    }
> #endif
>   }
> 
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale_any.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale_any.cc
>index 53ad136404995333e572364b70076334c5eb8d69..17831372cf0a02632524138a30d9dd004a255813 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale_any.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale_any.cc
>@@ -8,6 +8,8 @@
>  *  be found in the AUTHORS file in the root of the source tree.
>  */
> 
>+#include <string.h>  // For memset/memcpy
>+
> #include "libyuv/scale.h"
> #include "libyuv/scale_row.h"
> 
>@@ -42,6 +44,9 @@ CANY(ScaleARGBCols_Any_NEON, ScaleARGBCols_NEON, ScaleARGBCols_C, 4, 7)
> #ifdef HAS_SCALEARGBCOLS_MSA
> CANY(ScaleARGBCols_Any_MSA, ScaleARGBCols_MSA, ScaleARGBCols_C, 4, 3)
> #endif
>+#ifdef HAS_SCALEARGBCOLS_MMI
>+CANY(ScaleARGBCols_Any_MMI, ScaleARGBCols_MMI, ScaleARGBCols_C, 4, 0)
>+#endif
> #ifdef HAS_SCALEARGBFILTERCOLS_NEON
> CANY(ScaleARGBFilterCols_Any_NEON,
>      ScaleARGBFilterCols_NEON,
>@@ -165,6 +170,27 @@ SDANY(ScaleRowDown2Box_Any_MSA,
>       1,
>       31)
> #endif
>+#ifdef HAS_SCALEROWDOWN2_MMI
>+SDANY(ScaleRowDown2_Any_MMI, ScaleRowDown2_MMI, ScaleRowDown2_C, 2, 1, 7)
>+SDANY(ScaleRowDown2Linear_Any_MMI,
>+      ScaleRowDown2Linear_MMI,
>+      ScaleRowDown2Linear_C,
>+      2,
>+      1,
>+      7)
>+SDANY(ScaleRowDown2Box_Any_MMI,
>+      ScaleRowDown2Box_MMI,
>+      ScaleRowDown2Box_C,
>+      2,
>+      1,
>+      7)
>+SDODD(ScaleRowDown2Box_Odd_MMI,
>+      ScaleRowDown2Box_MMI,
>+      ScaleRowDown2Box_Odd_C,
>+      2,
>+      1,
>+      7)
>+#endif
> #ifdef HAS_SCALEROWDOWN4_SSSE3
> SDANY(ScaleRowDown4_Any_SSSE3, ScaleRowDown4_SSSE3, ScaleRowDown4_C, 4, 1, 7)
> SDANY(ScaleRowDown4Box_Any_SSSE3,
>@@ -201,6 +227,15 @@ SDANY(ScaleRowDown4Box_Any_MSA,
>       1,
>       15)
> #endif
>+#ifdef HAS_SCALEROWDOWN4_MMI
>+SDANY(ScaleRowDown4_Any_MMI, ScaleRowDown4_MMI, ScaleRowDown4_C, 4, 1, 7)
>+SDANY(ScaleRowDown4Box_Any_MMI,
>+      ScaleRowDown4Box_MMI,
>+      ScaleRowDown4Box_C,
>+      4,
>+      1,
>+      7)
>+#endif
> #ifdef HAS_SCALEROWDOWN34_SSSE3
> SDANY(ScaleRowDown34_Any_SSSE3,
>       ScaleRowDown34_SSSE3,
>@@ -382,6 +417,26 @@ SDANY(ScaleARGBRowDown2Box_Any_MSA,
>       4,
>       3)
> #endif
>+#ifdef HAS_SCALEARGBROWDOWN2_MMI
>+SDANY(ScaleARGBRowDown2_Any_MMI,
>+      ScaleARGBRowDown2_MMI,
>+      ScaleARGBRowDown2_C,
>+      2,
>+      4,
>+      1)
>+SDANY(ScaleARGBRowDown2Linear_Any_MMI,
>+      ScaleARGBRowDown2Linear_MMI,
>+      ScaleARGBRowDown2Linear_C,
>+      2,
>+      4,
>+      1)
>+SDANY(ScaleARGBRowDown2Box_Any_MMI,
>+      ScaleARGBRowDown2Box_MMI,
>+      ScaleARGBRowDown2Box_C,
>+      2,
>+      4,
>+      1)
>+#endif
> #undef SDANY
> 
> // Scale down by even scale factor.
>@@ -433,6 +488,57 @@ SDAANY(ScaleARGBRowDownEvenBox_Any_MSA,
>        4,
>        3)
> #endif
>+#ifdef HAS_SCALEARGBROWDOWNEVEN_MMI
>+SDAANY(ScaleARGBRowDownEven_Any_MMI,
>+       ScaleARGBRowDownEven_MMI,
>+       ScaleARGBRowDownEven_C,
>+       4,
>+       1)
>+SDAANY(ScaleARGBRowDownEvenBox_Any_MMI,
>+       ScaleARGBRowDownEvenBox_MMI,
>+       ScaleARGBRowDownEvenBox_C,
>+       4,
>+       1)
>+#endif
>+
>+#ifdef SASIMDONLY
>+// This also works and uses memcpy and SIMD instead of C, but is slower on ARM
>+
>+// Add rows box filter scale down.  Using macro from row_any
>+#define SAROW(NAMEANY, ANY_SIMD, SBPP, BPP, MASK)                      \
>+  void NAMEANY(const uint8_t* src_ptr, uint16_t* dst_ptr, int width) { \
>+    SIMD_ALIGNED(uint16_t dst_temp[32]);                               \
>+    SIMD_ALIGNED(uint8_t src_temp[32]);                                \
>+    memset(dst_temp, 0, 32 * 2); /* for msan */                        \
>+    int r = width & MASK;                                              \
>+    int n = width & ~MASK;                                             \
>+    if (n > 0) {                                                       \
>+      ANY_SIMD(src_ptr, dst_ptr, n);                                   \
>+    }                                                                  \
>+    memcpy(src_temp, src_ptr + n * SBPP, r * SBPP);                    \
>+    memcpy(dst_temp, dst_ptr + n * BPP, r * BPP);                      \
>+    ANY_SIMD(src_temp, dst_temp, MASK + 1);                            \
>+    memcpy(dst_ptr + n * BPP, dst_temp, r * BPP);                      \
>+  }
>+
>+#ifdef HAS_SCALEADDROW_SSE2
>+SAROW(ScaleAddRow_Any_SSE2, ScaleAddRow_SSE2, 1, 2, 15)
>+#endif
>+#ifdef HAS_SCALEADDROW_AVX2
>+SAROW(ScaleAddRow_Any_AVX2, ScaleAddRow_AVX2, 1, 2, 31)
>+#endif
>+#ifdef HAS_SCALEADDROW_NEON
>+SAROW(ScaleAddRow_Any_NEON, ScaleAddRow_NEON, 1, 2, 15)
>+#endif
>+#ifdef HAS_SCALEADDROW_MSA
>+SAROW(ScaleAddRow_Any_MSA, ScaleAddRow_MSA, 1, 2, 15)
>+#endif
>+#ifdef HAS_SCALEADDROW_MMI
>+SAROW(ScaleAddRow_Any_MMI, ScaleAddRow_MMI, 1, 2, 7)
>+#endif
>+#undef SAANY
>+
>+#else
> 
> // Add rows box filter scale down.
> #define SAANY(NAMEANY, SCALEADDROW_SIMD, SCALEADDROW_C, MASK)              \
>@@ -456,8 +562,13 @@ SAANY(ScaleAddRow_Any_NEON, ScaleAddRow_NEON, ScaleAddRow_C, 15)
> #ifdef HAS_SCALEADDROW_MSA
> SAANY(ScaleAddRow_Any_MSA, ScaleAddRow_MSA, ScaleAddRow_C, 15)
> #endif
>+#ifdef HAS_SCALEADDROW_MMI
>+SAANY(ScaleAddRow_Any_MMI, ScaleAddRow_MMI, ScaleAddRow_C, 7)
>+#endif
> #undef SAANY
> 
>+#endif  // SASIMDONLY
>+
> #ifdef __cplusplus
> }  // extern "C"
> }  // namespace libyuv
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale_argb.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale_argb.cc
>index 53a22e8b41ea1273fc2ae15215dbf91aa409aec6..beef380a8e61dba28300f88e59cec2c234690a10 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale_argb.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale_argb.cc
>@@ -111,6 +111,22 @@ static void ScaleARGBDown2(int src_width,
>     }
>   }
> #endif
>+#if defined(HAS_SCALEARGBROWDOWN2_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ScaleARGBRowDown2 =
>+        filtering == kFilterNone
>+            ? ScaleARGBRowDown2_Any_MMI
>+            : (filtering == kFilterLinear ? ScaleARGBRowDown2Linear_Any_MMI
>+                                          : ScaleARGBRowDown2Box_Any_MMI);
>+    if (IS_ALIGNED(dst_width, 2)) {
>+      ScaleARGBRowDown2 =
>+          filtering == kFilterNone
>+              ? ScaleARGBRowDown2_MMI
>+              : (filtering == kFilterLinear ? ScaleARGBRowDown2Linear_MMI
>+                                            : ScaleARGBRowDown2Box_MMI);
>+    }
>+  }
>+#endif
> 
>   if (filtering == kFilterLinear) {
>     src_stride = 0;
>@@ -237,6 +253,16 @@ static void ScaleARGBDownEven(int src_width,
>     }
>   }
> #endif
>+#if defined(HAS_SCALEARGBROWDOWNEVEN_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ScaleARGBRowDownEven = filtering ? ScaleARGBRowDownEvenBox_Any_MMI
>+                                     : ScaleARGBRowDownEven_Any_MMI;
>+    if (IS_ALIGNED(dst_width, 2)) {
>+      ScaleARGBRowDownEven =
>+          filtering ? ScaleARGBRowDownEvenBox_MMI : ScaleARGBRowDownEven_MMI;
>+    }
>+  }
>+#endif
> 
>   if (filtering == kFilterLinear) {
>     src_stride = 0;
>@@ -417,6 +443,14 @@ static void ScaleARGBBilinearUp(int src_width,
>       InterpolateRow = InterpolateRow_MSA;
>     }
>   }
>+#endif
>+#if defined(HAS_INTERPOLATEROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    InterpolateRow = InterpolateRow_Any_MMI;
>+    if (IS_ALIGNED(dst_width, 2)) {
>+      InterpolateRow = InterpolateRow_MMI;
>+    }
>+  }
> #endif
>   if (src_width >= 32768) {
>     ScaleARGBFilterCols =
>@@ -463,6 +497,14 @@ static void ScaleARGBBilinearUp(int src_width,
>       ScaleARGBFilterCols = ScaleARGBCols_MSA;
>     }
>   }
>+#endif
>+#if defined(HAS_SCALEARGBCOLS_MMI)
>+  if (!filtering && TestCpuFlag(kCpuHasMMI)) {
>+    ScaleARGBFilterCols = ScaleARGBCols_Any_MMI;
>+    if (IS_ALIGNED(dst_width, 1)) {
>+      ScaleARGBFilterCols = ScaleARGBCols_MMI;
>+    }
>+  }
> #endif
>   if (!filtering && src_width * 2 == dst_width && x < 0x8000) {
>     ScaleARGBFilterCols = ScaleARGBColsUp2_C;
>@@ -470,6 +512,11 @@ static void ScaleARGBBilinearUp(int src_width,
>     if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8)) {
>       ScaleARGBFilterCols = ScaleARGBColsUp2_SSE2;
>     }
>+#endif
>+#if defined(HAS_SCALEARGBCOLSUP2_MMI)
>+    if (TestCpuFlag(kCpuHasMMI) && IS_ALIGNED(dst_width, 4)) {
>+      ScaleARGBFilterCols = ScaleARGBColsUp2_MMI;
>+    }
> #endif
>   }
> 
>@@ -665,6 +712,14 @@ static void ScaleYUVToARGBBilinearUp(int src_width,
>       ScaleARGBFilterCols = ScaleARGBCols_MSA;
>     }
>   }
>+#endif
>+#if defined(HAS_SCALEARGBCOLS_MMI)
>+  if (!filtering && TestCpuFlag(kCpuHasMMI)) {
>+    ScaleARGBFilterCols = ScaleARGBCols_Any_MMI;
>+    if (IS_ALIGNED(dst_width, 1)) {
>+      ScaleARGBFilterCols = ScaleARGBCols_MMI;
>+    }
>+  }
> #endif
>   if (!filtering && src_width * 2 == dst_width && x < 0x8000) {
>     ScaleARGBFilterCols = ScaleARGBColsUp2_C;
>@@ -672,6 +727,11 @@ static void ScaleYUVToARGBBilinearUp(int src_width,
>     if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8)) {
>       ScaleARGBFilterCols = ScaleARGBColsUp2_SSE2;
>     }
>+#endif
>+#if defined(HAS_SCALEARGBCOLSUP2_MMI)
>+    if (TestCpuFlag(kCpuHasMMI) && IS_ALIGNED(dst_width, 4)) {
>+      ScaleARGBFilterCols = ScaleARGBColsUp2_MMI;
>+    }
> #endif
>   }
> 
>@@ -796,6 +856,14 @@ static void ScaleARGBSimple(int src_width,
>       ScaleARGBCols = ScaleARGBCols_MSA;
>     }
>   }
>+#endif
>+#if defined(HAS_SCALEARGBCOLS_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    ScaleARGBCols = ScaleARGBCols_Any_MMI;
>+    if (IS_ALIGNED(dst_width, 1)) {
>+      ScaleARGBCols = ScaleARGBCols_MMI;
>+    }
>+  }
> #endif
>   if (src_width * 2 == dst_width && x < 0x8000) {
>     ScaleARGBCols = ScaleARGBColsUp2_C;
>@@ -803,6 +871,11 @@ static void ScaleARGBSimple(int src_width,
>     if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8)) {
>       ScaleARGBCols = ScaleARGBColsUp2_SSE2;
>     }
>+#endif
>+#if defined(HAS_SCALEARGBCOLSUP2_MMI)
>+    if (TestCpuFlag(kCpuHasMMI) && IS_ALIGNED(dst_width, 4)) {
>+      ScaleARGBCols = ScaleARGBColsUp2_MMI;
>+    }
> #endif
>   }
> 
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale_common.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale_common.cc
>index b28d7da41fce9b6230aa3ccf6c110bdebe072e10..6369027175f1605e550ad83b57828fcb3dcf3b24 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale_common.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale_common.cc
>@@ -542,7 +542,9 @@ void ScaleFilterCols64_C(uint8_t* dst_ptr,
> 
> // Same as 8 bit arm blender but return is cast to uint16_t
> #define BLENDER(a, b, f) \
>-  (uint16_t)((int)(a) + ((((int)((f)) * ((int)(b) - (int)(a))) + 0x8000) >> 16))
>+  (uint16_t)(            \
>+      (int)(a) +         \
>+      (int)((((int64_t)((f)) * ((int64_t)(b) - (int)(a))) + 0x8000) >> 16))
> 
> void ScaleFilterCols_16_C(uint16_t* dst_ptr,
>                           const uint16_t* src_ptr,
>@@ -1072,6 +1074,14 @@ void ScalePlaneVertical(int src_height,
>       InterpolateRow = InterpolateRow_MSA;
>     }
>   }
>+#endif
>+#if defined(HAS_INTERPOLATEROW_MMI)
>+  if (TestCpuFlag(kCpuHasMMI)) {
>+    InterpolateRow = InterpolateRow_Any_MMI;
>+    if (IS_ALIGNED(dst_width_bytes, 8)) {
>+      InterpolateRow = InterpolateRow_MMI;
>+    }
>+  }
> #endif
>   for (j = 0; j < dst_height; ++j) {
>     int yi;
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale_mmi.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale_mmi.cc
>new file mode 100644
>index 0000000000000000000000000000000000000000..e12c6bb79ba839a94ced53faf5eee317151452dd
>--- /dev/null
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale_mmi.cc
>@@ -0,0 +1,1109 @@
>+/*
>+ *  Copyright 2013 The LibYuv Project Authors. All rights reserved.
>+ *
>+ *  Use of this source code is governed by a BSD-style license
>+ *  that can be found in the LICENSE file in the root of the source
>+ *  tree. An additional intellectual property rights grant can be found
>+ *  in the file PATENTS. All contributing project authors may
>+ *  be found in the AUTHORS file in the root of the source tree.
>+ */
>+
>+#include "libyuv/scale.h"
>+
>+#include <assert.h>
>+#include <string.h>
>+
>+#include "libyuv/cpu_id.h"
>+#include "libyuv/planar_functions.h"  // For CopyARGB
>+#include "libyuv/row.h"
>+#include "libyuv/scale_row.h"
>+
>+#ifdef __cplusplus
>+namespace libyuv {
>+extern "C" {
>+#endif
>+
>+// This module is for Mips MMI.
>+#if !defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)
>+
>+// CPU agnostic row functions
>+void ScaleRowDown2_MMI(const uint8_t* src_ptr,
>+                       ptrdiff_t src_stride,
>+                       uint8_t* dst,
>+                       int dst_width) {
>+  (void)src_stride;
>+
>+  uint64_t src0, src1, dest;
>+  const uint64_t shift = 0x8ULL;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src_ptr])                 \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src_ptr])                 \n\t"
>+      "psrlh      %[src0],         %[src0],           %[shift]      \n\t"
>+
>+      "gsldrc1    %[src1],         0x08(%[src_ptr])                 \n\t"
>+      "gsldlc1    %[src1],         0x0f(%[src_ptr])                 \n\t"
>+      "psrlh      %[src1],         %[src1],           %[shift]      \n\t"
>+
>+      "packushb   %[dest],         %[src0],           %[src1]       \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x10          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x08          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src0] "=&f"(src0), [src1] "=&f"(src1), [dest] "=&f"(dest)
>+      : [src_ptr] "r"(src_ptr), [dst_ptr] "r"(dst), [width] "r"(dst_width),
>+        [shift] "f"(shift)
>+      : "memory");
>+}
>+
>+void ScaleRowDown2Linear_MMI(const uint8_t* src_ptr,
>+                             ptrdiff_t src_stride,
>+                             uint8_t* dst,
>+                             int dst_width) {
>+  (void)src_stride;
>+
>+  uint64_t src0, src1;
>+  uint64_t dest, dest0, dest1;
>+
>+  const uint64_t mask = 0x00ff00ff00ff00ffULL;
>+  const uint64_t shift = 0x8ULL;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldrc1    %[src0],          0x00(%[src_ptr])                \n\t"
>+      "gsldlc1    %[src0],          0x07(%[src_ptr])                \n\t"
>+      "and        %[dest0],         %[src0],          %[mask]       \n\t"
>+      "gsldrc1    %[src1],          0x08(%[src_ptr])                \n\t"
>+      "gsldlc1    %[src1],          0x0f(%[src_ptr])                \n\t"
>+      "and        %[dest1],         %[src1],          %[mask]       \n\t"
>+      "packushb   %[dest0],         %[dest0],         %[dest1]      \n\t"
>+
>+      "psrlh      %[src0],          %[src0],          %[shift]      \n\t"
>+      "psrlh      %[src1],          %[src1],          %[shift]      \n\t"
>+      "packushb   %[dest1],         %[src0],          %[src1]       \n\t"
>+
>+      "pavgb      %[dest],          %[dest0],         %[dest1]      \n\t"
>+      "gssdlc1    %[dest],          0x07(%[dst_ptr])                \n\t"
>+      "gssdrc1    %[dest],          0x00(%[dst_ptr])                \n\t"
>+
>+      "daddiu     %[src_ptr],       %[src_ptr],        0x10         \n\t"
>+      "daddiu     %[dst_ptr],       %[dst_ptr],        0x08         \n\t"
>+      "daddi      %[width],         %[width],         -0x08         \n\t"
>+      "bnez       %[width],         1b                              \n\t"
>+      : [src0] "=&f"(src0), [src1] "=&f"(src1), [dest0] "=&f"(dest0),
>+        [dest1] "=&f"(dest1), [dest] "=&f"(dest)
>+      : [src_ptr] "r"(src_ptr), [dst_ptr] "r"(dst), [mask] "f"(mask),
>+        [shift] "f"(shift), [width] "r"(dst_width)
>+      : "memory");
>+}
>+
>+void ScaleRowDown2Box_MMI(const uint8_t* src_ptr,
>+                          ptrdiff_t src_stride,
>+                          uint8_t* dst,
>+                          int dst_width) {
>+  const uint8_t* s = src_ptr;
>+  const uint8_t* t = src_ptr + src_stride;
>+
>+  uint64_t s0, s1, t0, t1;
>+  uint64_t dest, dest0, dest1;
>+
>+  const uint64_t ph = 0x0002000200020002ULL;
>+  const uint64_t mask = 0x00ff00ff00ff00ffULL;
>+  const uint64_t shift0 = 0x2ULL;
>+  const uint64_t shift1 = 0x8ULL;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldrc1    %[s0],            0x00(%[s])                      \n\t"
>+      "gsldlc1    %[s0],            0x07(%[s])                      \n\t"
>+      "psrlh      %[s1],            %[s0],            %[shift1]     \n\t"
>+      "and        %[s0],            %[s0],            %[mask]       \n\t"
>+
>+      "gsldrc1    %[t0],            0x00(%[t])                      \n\t"
>+      "gsldlc1    %[t0],            0x07(%[t])                      \n\t"
>+      "psrlh      %[t1],            %[t0],            %[shift1]     \n\t"
>+      "and        %[t0],            %[t0],            %[mask]       \n\t"
>+
>+      "paddh      %[dest0],         %[s0],            %[s1]         \n\t"
>+      "paddh      %[dest0],         %[dest0],         %[t0]         \n\t"
>+      "paddh      %[dest0],         %[dest0],         %[t1]         \n\t"
>+      "paddh      %[dest0],         %[dest0],         %[ph]         \n\t"
>+      "psrlh      %[dest0],         %[dest0],         %[shift0]     \n\t"
>+
>+      "gsldrc1    %[s0],            0x08(%[s])                      \n\t"
>+      "gsldlc1    %[s0],            0x0f(%[s])                      \n\t"
>+      "psrlh      %[s1],            %[s0],            %[shift1]     \n\t"
>+      "and        %[s0],            %[s0],            %[mask]       \n\t"
>+
>+      "gsldrc1    %[t0],            0x08(%[t])                      \n\t"
>+      "gsldlc1    %[t0],            0x0f(%[t])                      \n\t"
>+      "psrlh      %[t1],            %[t0],            %[shift1]     \n\t"
>+      "and        %[t0],            %[t0],            %[mask]       \n\t"
>+
>+      "paddh      %[dest1],         %[s0],            %[s1]         \n\t"
>+      "paddh      %[dest1],         %[dest1],         %[t0]         \n\t"
>+      "paddh      %[dest1],         %[dest1],         %[t1]         \n\t"
>+      "paddh      %[dest1],         %[dest1],         %[ph]         \n\t"
>+      "psrlh      %[dest1],         %[dest1],         %[shift0]     \n\t"
>+
>+      "packushb   %[dest],          %[dest0],         %[dest1]      \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[s],            %[s],              0x10          \n\t"
>+      "daddiu     %[t],            %[t],              0x10          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x08          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [s0] "=&f"(s0), [s1] "=&f"(s1), [t0] "=&f"(t0), [t1] "=&f"(t1),
>+        [dest0] "=&f"(dest0), [dest1] "=&f"(dest1), [dest] "=&f"(dest)
>+      : [s] "r"(s), [t] "r"(t), [dst_ptr] "r"(dst), [width] "r"(dst_width),
>+        [shift0] "f"(shift0), [shift1] "f"(shift1), [ph] "f"(ph),
>+        [mask] "f"(mask)
>+      : "memory");
>+}
>+
>+void ScaleARGBRowDown2_MMI(const uint8_t* src_argb,
>+                           ptrdiff_t src_stride,
>+                           uint8_t* dst_argb,
>+                           int dst_width) {
>+  (void)src_stride;
>+
>+  const uint32_t* src = (const uint32_t*)(src_argb);
>+  uint32_t* dst = (uint32_t*)(dst_argb);
>+
>+  uint64_t src0, src1, dest;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src_ptr])                 \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src_ptr])                 \n\t"
>+      "gsldrc1    %[src1],         0x08(%[src_ptr])                 \n\t"
>+      "gsldlc1    %[src1],         0x0f(%[src_ptr])                 \n\t"
>+      "punpckhwd  %[dest],         %[src0],           %[src1]       \n\t"
>+
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x10          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x02          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src0] "=&f"(src0), [src1] "=&f"(src1), [dest] "=&f"(dest)
>+      : [src_ptr] "r"(src), [dst_ptr] "r"(dst), [width] "r"(dst_width)
>+      : "memory");
>+}
>+
>+void ScaleARGBRowDown2Linear_MMI(const uint8_t* src_argb,
>+                                 ptrdiff_t src_stride,
>+                                 uint8_t* dst_argb,
>+                                 int dst_width) {
>+  (void)src_stride;
>+
>+  uint64_t src0, src1;
>+  uint64_t dest, dest_hi, dest_lo;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "lwc1       %[src0],         0x00(%[src_ptr])                 \n\t"
>+      "lwc1       %[src1],         0x08(%[src_ptr])                 \n\t"
>+      "punpcklwd  %[dest_lo],      %[src0],           %[src1]       \n\t"
>+      "lwc1       %[src0],         0x04(%[src_ptr])                 \n\t"
>+      "lwc1       %[src1],         0x0c(%[src_ptr])                 \n\t"
>+      "punpcklwd  %[dest_hi],      %[src0],           %[src1]       \n\t"
>+
>+      "pavgb      %[dest],         %[dest_lo],        %[dest_hi]    \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x10          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x02          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src0] "=&f"(src0), [src1] "=&f"(src1), [dest_hi] "=&f"(dest_hi),
>+        [dest_lo] "=&f"(dest_lo), [dest] "=&f"(dest)
>+      : [src_ptr] "r"(src_argb), [dst_ptr] "r"(dst_argb), [width] "r"(dst_width)
>+      : "memory");
>+}
>+
>+void ScaleARGBRowDown2Box_MMI(const uint8_t* src_argb,
>+                              ptrdiff_t src_stride,
>+                              uint8_t* dst_argb,
>+                              int dst_width) {
>+  const uint8_t* s = src_argb;
>+  const uint8_t* t = src_argb + src_stride;
>+
>+  uint64_t s0, s_hi, s_lo;
>+  uint64_t t0, t_hi, t_lo;
>+  uint64_t dest, dest_hi, dest_lo;
>+
>+  const uint64_t mask = 0x0ULL;
>+  const uint64_t ph = 0x0002000200020002ULL;
>+  const uint64_t shfit = 0x2ULL;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldrc1    %[s0],            0x00(%[s])                      \n\t"
>+      "gsldlc1    %[s0],            0x07(%[s])                      \n\t"
>+      "punpcklbh  %[s_lo],          %[s0],           %[mask]        \n\t"
>+      "punpckhbh  %[s_hi],          %[s0],           %[mask]        \n\t"
>+      "paddh      %[dest_lo],       %[s_lo],         %[s_hi]        \n\t"
>+
>+      "gsldrc1    %[t0],            0x00(%[t])                      \n\t"
>+      "gsldlc1    %[t0],            0x07(%[t])                      \n\t"
>+      "punpcklbh  %[t_lo],          %[t0],           %[mask]        \n\t"
>+      "punpckhbh  %[t_hi],          %[t0],           %[mask]        \n\t"
>+      "paddh      %[dest_lo],       %[dest_lo],      %[t_lo]        \n\t"
>+      "paddh      %[dest_lo],       %[dest_lo],      %[t_hi]        \n\t"
>+
>+      "paddh      %[dest_lo],      %[dest_lo],       %[ph]          \n\t"
>+      "psrlh      %[dest_lo],      %[dest_lo],       %[shfit]       \n\t"
>+
>+      "gsldrc1    %[s0],            0x08(%[s])                      \n\t"
>+      "gsldlc1    %[s0],            0x0f(%[s])                      \n\t"
>+      "punpcklbh  %[s_lo],          %[s0],           %[mask]        \n\t"
>+      "punpckhbh  %[s_hi],          %[s0],           %[mask]        \n\t"
>+      "paddh      %[dest_hi],       %[s_lo],         %[s_hi]        \n\t"
>+
>+      "gsldrc1    %[t0],            0x08(%[t])                      \n\t"
>+      "gsldlc1    %[t0],            0x0f(%[t])                      \n\t"
>+      "punpcklbh  %[t_lo],          %[t0],           %[mask]        \n\t"
>+      "punpckhbh  %[t_hi],          %[t0],           %[mask]        \n\t"
>+      "paddh      %[dest_hi],       %[dest_hi],      %[t_lo]        \n\t"
>+      "paddh      %[dest_hi],       %[dest_hi],      %[t_hi]        \n\t"
>+
>+      "paddh      %[dest_hi],      %[dest_hi],       %[ph]          \n\t"
>+      "psrlh      %[dest_hi],      %[dest_hi],       %[shfit]       \n\t"
>+
>+      "packushb   %[dest],         %[dest_lo],       %[dest_hi]     \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[s],            %[s],              0x10          \n\t"
>+      "daddiu     %[t],            %[t],              0x10          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x02          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [s0] "=&f"(s0), [t0] "=&f"(t0), [dest_hi] "=&f"(dest_hi),
>+        [dest_lo] "=&f"(dest_lo), [s_hi] "=&f"(s_hi), [s_lo] "=&f"(s_lo),
>+        [t_hi] "=&f"(t_hi), [t_lo] "=&f"(t_lo), [dest] "=&f"(dest)
>+      : [s] "r"(s), [t] "r"(t), [dst_ptr] "r"(dst_argb), [width] "r"(dst_width),
>+        [mask] "f"(mask), [ph] "f"(ph), [shfit] "f"(shfit)
>+      : "memory");
>+}
>+
>+void ScaleRowDown2_16_MMI(const uint16_t* src_ptr,
>+                          ptrdiff_t src_stride,
>+                          uint16_t* dst,
>+                          int dst_width) {
>+  (void)src_stride;
>+
>+  uint64_t src0, src1, dest;
>+  const uint64_t shift = 0x10ULL;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src_ptr])                 \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src_ptr])                 \n\t"
>+      "psrlw      %[src0],         %[src0],           %[shift]      \n\t"
>+
>+      "gsldrc1    %[src1],         0x08(%[src_ptr])                 \n\t"
>+      "gsldlc1    %[src1],         0x0f(%[src_ptr])                 \n\t"
>+      "psrlw      %[src1],         %[src1],           %[shift]      \n\t"
>+
>+      "packsswh   %[dest],         %[src0],           %[src1]       \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x10          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x04          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src0] "=&f"(src0), [src1] "=&f"(src1), [dest] "=&f"(dest)
>+      : [src_ptr] "r"(src_ptr), [dst_ptr] "r"(dst), [width] "r"(dst_width),
>+        [shift] "f"(shift)
>+      : "memory");
>+}
>+
>+void ScaleRowDown2Linear_16_MMI(const uint16_t* src_ptr,
>+                                ptrdiff_t src_stride,
>+                                uint16_t* dst,
>+                                int dst_width) {
>+  (void)src_stride;
>+
>+  uint64_t src0, src1;
>+  uint64_t dest, dest_hi, dest_lo;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src_ptr])                 \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src_ptr])                 \n\t"
>+      "gsldrc1    %[src1],         0x08(%[src_ptr])                 \n\t"
>+      "gsldlc1    %[src1],         0x0f(%[src_ptr])                 \n\t"
>+      "punpcklhw  %[dest_lo],      %[src0],           %[src1]       \n\t"
>+      "punpckhhw  %[dest_hi],      %[src0],           %[src1]       \n\t"
>+
>+      "punpcklhw  %[src0],         %[dest_lo],        %[dest_hi]    \n\t"
>+      "punpckhhw  %[src1],         %[dest_lo],        %[dest_hi]    \n\t"
>+
>+      "pavgh      %[dest],         %[src0],           %[src1]       \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x10          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x04          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src0] "=&f"(src0), [src1] "=&f"(src1), [dest_hi] "=&f"(dest_hi),
>+        [dest_lo] "=&f"(dest_lo), [dest] "=&f"(dest)
>+      : [src_ptr] "r"(src_ptr), [dst_ptr] "r"(dst), [width] "r"(dst_width)
>+      : "memory");
>+}
>+
>+void ScaleRowDown2Box_16_MMI(const uint16_t* src_ptr,
>+                             ptrdiff_t src_stride,
>+                             uint16_t* dst,
>+                             int dst_width) {
>+  const uint16_t* s = src_ptr;
>+  const uint16_t* t = src_ptr + src_stride;
>+
>+  uint64_t s0, s1, s_hi, s_lo;
>+  uint64_t t0, t1, t_hi, t_lo;
>+  uint64_t dest, dest0, dest1;
>+
>+  const uint64_t ph = 0x0000000200000002ULL;
>+  const uint64_t mask = 0x0000ffff0000ffffULL;
>+  const uint64_t shift0 = 0x10ULL;
>+  const uint64_t shift1 = 0x2ULL;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldrc1    %[s0],            0x00(%[s])                      \n\t"
>+      "gsldlc1    %[s0],            0x07(%[s])                      \n\t"
>+      "psrlw      %[s1],            %[s0],            %[shift0]     \n\t"
>+      "and        %[s0],            %[s0],            %[mask]       \n\t"
>+
>+      "gsldrc1    %[t0],            0x00(%[t])                      \n\t"
>+      "gsldlc1    %[t0],            0x07(%[t])                      \n\t"
>+      "psrlw      %[t1],            %[t0],            %[shift0]     \n\t"
>+      "and        %[t0],            %[t0],            %[mask]       \n\t"
>+
>+      "paddw      %[dest0],         %[s0],            %[s1]         \n\t"
>+      "paddw      %[dest0],         %[dest0],         %[t0]         \n\t"
>+      "paddw      %[dest0],         %[dest0],         %[t1]         \n\t"
>+      "paddw      %[dest0],         %[dest0],         %[ph]         \n\t"
>+      "psrlw      %[dest0],         %[dest0],         %[shift1]     \n\t"
>+
>+      "gsldrc1    %[s0],            0x08(%[s])                      \n\t"
>+      "gsldlc1    %[s0],            0x0f(%[s])                      \n\t"
>+      "psrlw      %[s1],            %[s0],            %[shift0]     \n\t"
>+      "and        %[s0],            %[s0],            %[mask]       \n\t"
>+
>+      "gsldrc1    %[t0],            0x08(%[t])                      \n\t"
>+      "gsldlc1    %[t0],            0x0f(%[t])                      \n\t"
>+      "psrlw      %[t1],            %[t0],            %[shift0]     \n\t"
>+      "and        %[t0],            %[t0],            %[mask]       \n\t"
>+
>+      "paddw      %[dest1],         %[s0],            %[s1]         \n\t"
>+      "paddw      %[dest1],         %[dest1],         %[t0]         \n\t"
>+      "paddw      %[dest1],         %[dest1],         %[t1]         \n\t"
>+      "paddw      %[dest1],         %[dest1],         %[ph]         \n\t"
>+      "psrlw      %[dest1],         %[dest1],         %[shift1]     \n\t"
>+
>+      "packsswh   %[dest],          %[dest0],         %[dest1]      \n\t"
>+      "gssdlc1    %[dest],          0x07(%[dst_ptr])                \n\t"
>+      "gssdrc1    %[dest],          0x00(%[dst_ptr])                \n\t"
>+
>+      "daddiu     %[s],             %[s],              0x10         \n\t"
>+      "daddiu     %[t],             %[t],              0x10         \n\t"
>+      "daddiu     %[dst_ptr],       %[dst_ptr],        0x08         \n\t"
>+      "daddi      %[width],         %[width],         -0x04         \n\t"
>+      "bnez       %[width],         1b                              \n\t"
>+      : [s0] "=&f"(s0), [s1] "=&f"(s1), [t0] "=&f"(t0), [t1] "=&f"(t1),
>+        [s_hi] "=&f"(s_hi), [s_lo] "=&f"(s_lo), [t_hi] "=&f"(t_hi),
>+        [t_lo] "=&f"(t_lo), [dest0] "=&f"(dest0), [dest1] "=&f"(dest1),
>+        [dest] "=&f"(dest)
>+      : [s] "r"(s), [t] "r"(t), [dst_ptr] "r"(dst), [width] "r"(dst_width),
>+        [shift0] "f"(shift0), [shift1] "f"(shift1), [ph] "f"(ph),
>+        [mask] "f"(mask)
>+      : "memory");
>+}
>+
>+void ScaleRowDown4_MMI(const uint8_t* src_ptr,
>+                       ptrdiff_t src_stride,
>+                       uint8_t* dst,
>+                       int dst_width) {
>+  (void)src_stride;
>+
>+  uint64_t src0, src1;
>+  uint64_t dest, dest_hi, dest_lo;
>+
>+  const uint64_t shift = 0x10ULL;
>+  const uint64_t mask = 0x000000ff000000ffULL;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src_ptr])                 \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src_ptr])                 \n\t"
>+      "psrlw      %[src0],         %[src0],           %[shift]      \n\t"
>+      "and        %[src0],         %[src0],           %[mask]       \n\t"
>+      "gsldrc1    %[src1],         0x08(%[src_ptr])                 \n\t"
>+      "gsldlc1    %[src1],         0x0f(%[src_ptr])                 \n\t"
>+      "psrlw      %[src1],         %[src1],           %[shift]      \n\t"
>+      "and        %[src1],         %[src1],           %[mask]       \n\t"
>+      "packsswh   %[dest_lo],      %[src0],           %[src1]       \n\t"
>+
>+      "gsldrc1    %[src0],         0x10(%[src_ptr])                 \n\t"
>+      "gsldlc1    %[src0],         0x17(%[src_ptr])                 \n\t"
>+      "psrlw      %[src0],         %[src0],           %[shift]      \n\t"
>+      "and        %[src0],         %[src0],           %[mask]       \n\t"
>+      "gsldrc1    %[src1],         0x18(%[src_ptr])                 \n\t"
>+      "gsldlc1    %[src1],         0x1f(%[src_ptr])                 \n\t"
>+      "psrlw      %[src1],         %[src1],           %[shift]      \n\t"
>+      "and        %[src1],         %[src1],           %[mask]       \n\t"
>+      "packsswh   %[dest_hi],      %[src0],           %[src1]       \n\t"
>+
>+      "packushb   %[dest],         %[dest_lo],         %[dest_hi]   \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x20          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x08          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src0] "=&f"(src0), [src1] "=&f"(src1), [dest_hi] "=&f"(dest_hi),
>+        [dest_lo] "=&f"(dest_lo), [dest] "=&f"(dest)
>+      : [src_ptr] "r"(src_ptr), [dst_ptr] "r"(dst), [width] "r"(dst_width),
>+        [shift] "f"(shift), [mask] "f"(mask)
>+      : "memory");
>+}
>+
>+void ScaleRowDown4_16_MMI(const uint16_t* src_ptr,
>+                          ptrdiff_t src_stride,
>+                          uint16_t* dst,
>+                          int dst_width) {
>+  (void)src_stride;
>+
>+  uint64_t src0, src1;
>+  uint64_t dest, dest_hi, dest_lo;
>+
>+  const uint64_t mask = 0x0ULL;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldrc1    %[src0],         0x00(%[src_ptr])                 \n\t"
>+      "gsldlc1    %[src0],         0x07(%[src_ptr])                 \n\t"
>+      "gsldrc1    %[src1],         0x08(%[src_ptr])                 \n\t"
>+      "gsldlc1    %[src1],         0x0f(%[src_ptr])                 \n\t"
>+      "punpckhhw  %[dest_lo],      %[src0],           %[src1]       \n\t"
>+      "punpcklhw  %[dest_lo],      %[dest_lo],        %[mask]       \n\t"
>+
>+      "gsldrc1    %[src0],         0x10(%[src_ptr])                 \n\t"
>+      "gsldlc1    %[src0],         0x17(%[src_ptr])                 \n\t"
>+      "gsldrc1    %[src1],         0x18(%[src_ptr])                 \n\t"
>+      "gsldlc1    %[src1],         0x1f(%[src_ptr])                 \n\t"
>+      "punpckhhw  %[dest_hi],      %[src0],           %[src1]       \n\t"
>+      "punpcklhw  %[dest_hi],      %[dest_hi],        %[mask]       \n\t"
>+
>+      "packushb   %[dest],         %[dest_lo],        %[dest_hi]    \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x20          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x04          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src0] "=&f"(src0), [src1] "=&f"(src1), [dest_hi] "=&f"(dest_hi),
>+        [dest_lo] "=&f"(dest_lo), [dest] "=&f"(dest)
>+      : [src_ptr] "r"(src_ptr), [dst_ptr] "r"(dst), [width] "r"(dst_width),
>+        [mask] "f"(mask)
>+      : "memory");
>+}
>+
>+#define DO_SCALEROWDOWN4BOX_PUNPCKADD()                              \
>+  "punpcklbh  %[src_lo],       %[src],           %[mask0]      \n\t" \
>+  "punpckhbh  %[src_hi],       %[src],           %[mask0]      \n\t" \
>+  "paddh      %[dest_lo],      %[dest_lo],       %[src_lo]     \n\t" \
>+  "paddh      %[dest_hi],      %[dest_hi],       %[src_hi]     \n\t"
>+
>+#define DO_SCALEROWDOWN4BOX_LOOP(reg)                                \
>+  "ldc1       %[src],          0x00(%[src0_ptr])               \n\t" \
>+  "punpcklbh  %[dest_lo],      %[src],           %[mask0]      \n\t" \
>+  "punpckhbh  %[dest_hi],      %[src],           %[mask0]      \n\t" \
>+                                                                     \
>+  "ldc1       %[src],          0x00(%[src1_ptr])               \n\t" \
>+  DO_SCALEROWDOWN4BOX_PUNPCKADD()                                    \
>+                                                                     \
>+  "ldc1       %[src],          0x00(%[src2_ptr])               \n\t" \
>+  DO_SCALEROWDOWN4BOX_PUNPCKADD()                                    \
>+                                                                     \
>+  "ldc1       %[src],          0x00(%[src3_ptr])               \n\t" \
>+  DO_SCALEROWDOWN4BOX_PUNPCKADD()                                    \
>+                                                                     \
>+  "pmaddhw    %[dest_lo],      %[dest_lo],       %[mask1]      \n\t" \
>+  "pmaddhw    %[dest_hi],      %[dest_hi],       %[mask1]      \n\t" \
>+  "packsswh   " #reg   ",      %[dest_lo],       %[dest_hi]    \n\t" \
>+  "pmaddhw    " #reg   ",      " #reg   ",       %[mask1]      \n\t" \
>+  "paddh      " #reg   ",      " #reg   ",       %[ph]         \n\t" \
>+  "psrlh      " #reg   ",      " #reg   ",       %[shift]      \n\t" \
>+                                                                     \
>+  "daddiu     %[src0_ptr],     %[src0_ptr],      0x08          \n\t" \
>+  "daddiu     %[src1_ptr],     %[src1_ptr],      0x08          \n\t" \
>+  "daddiu     %[src2_ptr],     %[src2_ptr],      0x08          \n\t" \
>+  "daddiu     %[src3_ptr],     %[src3_ptr],      0x08          \n\t"
>+
>+/* LibYUVScaleTest.ScaleDownBy4_Box */
>+void ScaleRowDown4Box_MMI(const uint8_t* src_ptr,
>+                          ptrdiff_t src_stride,
>+                          uint8_t* dst,
>+                          int dst_width) {
>+  const uint8_t* src0_ptr = src_ptr;
>+  const uint8_t* src1_ptr = src_ptr + src_stride;
>+  const uint8_t* src2_ptr = src_ptr + src_stride * 2;
>+  const uint8_t* src3_ptr = src_ptr + src_stride * 3;
>+
>+  uint64_t src, src_hi, src_lo;
>+  uint64_t dest, dest_hi, dest_lo, dest0, dest1, dest2, dest3;
>+
>+  const uint64_t mask0 = 0x0ULL;
>+  const uint64_t mask1 = 0x0001000100010001ULL;
>+  const uint64_t ph = 0x0008000800080008ULL;
>+  const uint64_t shift = 0x4ULL;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+
>+      DO_SCALEROWDOWN4BOX_LOOP(%[dest0])
>+      DO_SCALEROWDOWN4BOX_LOOP(%[dest1])
>+      DO_SCALEROWDOWN4BOX_LOOP(%[dest2])
>+      DO_SCALEROWDOWN4BOX_LOOP(%[dest3])
>+
>+      "packsswh   %[dest_lo],      %[dest0],          %[dest1]      \n\t"
>+      "packsswh   %[dest_hi],      %[dest2],          %[dest3]      \n\t"
>+
>+      "packushb   %[dest],         %[dest_lo],        %[dest_hi]    \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x08          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src_hi] "=&f"(src_hi), [src_lo] "=&f"(src_lo),
>+        [dest_hi] "=&f"(dest_hi), [dest_lo] "=&f"(dest_lo),
>+        [dest0] "=&f"(dest0), [dest1] "=&f"(dest1), [dest2] "=&f"(dest2),
>+        [dest3] "=&f"(dest3), [src] "=&f"(src), [dest] "=&f"(dest)
>+      : [src0_ptr] "r"(src0_ptr), [src1_ptr] "r"(src1_ptr),
>+        [src2_ptr] "r"(src2_ptr), [src3_ptr] "r"(src3_ptr), [dst_ptr] "r"(dst),
>+        [width] "r"(dst_width), [shift] "f"(shift), [mask0] "f"(mask0),
>+        [ph] "f"(ph), [mask1] "f"(mask1)
>+      : "memory");
>+}
>+
>+#define DO_SCALEROWDOWN4BOX_16_PUNPCKADD()                            \
>+  "punpcklbh  %[src_lo],       %[src],            %[mask0]      \n\t" \
>+  "punpckhbh  %[src_hi],       %[src],            %[mask0]      \n\t" \
>+  "paddh      %[dest_lo],      %[dest_lo],        %[src_lo]     \n\t" \
>+  "paddh      %[dest_hi],      %[dest_hi],        %[src_hi]     \n\t"
>+
>+#define DO_SCALEROWDOWN4BOX_16_LOOP(reg)                              \
>+  "ldc1       %[src],          0x00(%[src0_ptr])                \n\t" \
>+  "punpcklbh  %[dest_lo],      %[src],            %[mask0]      \n\t" \
>+  "punpckhbh  %[dest_hi],      %[src],            %[mask0]      \n\t" \
>+                                                                      \
>+  "ldc1       %[src],          0x00(%[src1_ptr])                \n\t" \
>+  DO_SCALEROWDOWN4BOX_16_PUNPCKADD()                                  \
>+                                                                      \
>+  "ldc1       %[src],          0x00(%[src2_ptr])                \n\t" \
>+  DO_SCALEROWDOWN4BOX_16_PUNPCKADD()                                  \
>+                                                                      \
>+  "ldc1       %[src],          0x00(%[src3_ptr])                \n\t" \
>+  DO_SCALEROWDOWN4BOX_16_PUNPCKADD()                                  \
>+                                                                      \
>+  "paddw      %[dest],         %[dest_lo],        %[dest_hi]    \n\t" \
>+  "punpckhwd  %[dest_hi],      %[dest],           %[dest]       \n\t" \
>+  "paddw      %[dest],         %[dest_hi],        %[dest]       \n\t" \
>+  "paddw      %[dest],         %[dest],           %[ph]         \n\t" \
>+  "psraw      %[dest],         %[dest],           %[shift]      \n\t" \
>+  "and        " #reg ",        %[dest],           %[mask1]      \n\t" \
>+                                                                      \
>+  "daddiu     %[src0_ptr],     %[src0_ptr],       0x08          \n\t" \
>+  "daddiu     %[src1_ptr],     %[src1_ptr],       0x08          \n\t" \
>+  "daddiu     %[src2_ptr],     %[src2_ptr],       0x08          \n\t" \
>+  "daddiu     %[src3_ptr],     %[src3_ptr],       0x08          \n\t"
>+
>+/* LibYUVScaleTest.ScaleDownBy4_Box_16 */
>+void ScaleRowDown4Box_16_MMI(const uint16_t* src_ptr,
>+                             ptrdiff_t src_stride,
>+                             uint16_t* dst,
>+                             int dst_width) {
>+  const uint16_t* src0_ptr = src_ptr;
>+  const uint16_t* src1_ptr = src_ptr + src_stride;
>+  const uint16_t* src2_ptr = src_ptr + src_stride * 2;
>+  const uint16_t* src3_ptr = src_ptr + src_stride * 3;
>+
>+  uint64_t src, src_hi, src_lo;
>+  uint64_t dest, dest_hi, dest_lo, dest0, dest1, dest2, dest3;
>+
>+  const uint64_t mask0 = 0x0ULL;
>+  const uint64_t mask1 = 0x00000000ffffffffULL;
>+  const uint64_t ph = 0x0000000800000008ULL;
>+  const uint64_t shift = 0x04ULL;
>+
>+  __asm__ volatile(
>+      "1:                                                        \n\t"
>+
>+      DO_SCALEROWDOWN4BOX_16_LOOP(%[dest0])
>+      DO_SCALEROWDOWN4BOX_16_LOOP(%[dest1])
>+      DO_SCALEROWDOWN4BOX_16_LOOP(%[dest2])
>+      DO_SCALEROWDOWN4BOX_16_LOOP(%[dest3])
>+      "punpcklwd  %[dest_lo],      %[dest0],          %[dest1]   \n\t"
>+      "punpcklwd  %[dest_hi],      %[dest2],          %[dest3]   \n\t"
>+
>+      "packushb   %[dest],         %[dest_lo],        %[dest_hi] \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])              \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])              \n\t"
>+
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08       \n\t"
>+      "daddi      %[width],        %[width],         -0x04       \n\t"
>+      "bnez       %[width],        1b                            \n\t"
>+      : [src_hi] "=&f"(src_hi), [src_lo] "=&f"(src_lo),
>+        [dest_hi] "=&f"(dest_hi), [dest_lo] "=&f"(dest_lo),
>+        [dest0] "=&f"(dest0), [dest1] "=&f"(dest1), [dest2] "=&f"(dest2),
>+        [dest3] "=&f"(dest3), [src] "=&f"(src), [dest] "=&f"(dest)
>+      : [src0_ptr] "r"(src0_ptr), [src1_ptr] "r"(src1_ptr),
>+        [src2_ptr] "r"(src2_ptr), [src3_ptr] "r"(src3_ptr), [dst_ptr] "r"(dst),
>+        [width] "r"(dst_width), [shift] "f"(shift), [mask0] "f"(mask0),
>+        [ph] "f"(ph), [mask1] "f"(mask1)
>+      : "memory");
>+}
>+
>+// Scales a single row of pixels up by 2x using point sampling.
>+void ScaleColsUp2_MMI(uint8_t* dst_ptr,
>+                      const uint8_t* src_ptr,
>+                      int dst_width,
>+                      int x,
>+                      int dx) {
>+  uint64_t src, dest;
>+
>+  (void)x;
>+  (void)dx;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "lwc1       %[src],          0x00(%[src_ptr])                 \n\t"
>+
>+      "punpcklbh  %[dest],         %[src],            %[src]        \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x04          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x08          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src] "=&f"(src), [dest] "=&f"(dest)
>+      : [src_ptr] "r"(src_ptr), [dst_ptr] "r"(dst_ptr), [width] "r"(dst_width)
>+      : "memory");
>+}
>+
>+void ScaleColsUp2_16_MMI(uint16_t* dst_ptr,
>+                         const uint16_t* src_ptr,
>+                         int dst_width,
>+                         int x,
>+                         int dx) {
>+  uint64_t src, dest;
>+
>+  (void)x;
>+  (void)dx;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldrc1    %[src],          0x00(%[src_ptr])                 \n\t"
>+      "gsldlc1    %[src],          0x07(%[src_ptr])                 \n\t"
>+
>+      "punpcklhw  %[dest],         %[src],            %[src]        \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "punpckhhw  %[dest],         %[src],            %[src]        \n\t"
>+      "gssdlc1    %[dest],         0x0f(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x08(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x08          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x10          \n\t"
>+      "daddi      %[width],        %[width],         -0x08          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src] "=&f"(src), [dest] "=&f"(dest)
>+      : [src_ptr] "r"(src_ptr), [dst_ptr] "r"(dst_ptr), [width] "r"(dst_width)
>+      : "memory");
>+}
>+
>+void ScaleAddRow_MMI(const uint8_t* src_ptr, uint16_t* dst_ptr, int src_width) {
>+  uint64_t src, src_hi, src_lo, dest0, dest1;
>+  const uint64_t mask = 0x0ULL;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldlc1    %[src],          0x07(%[src_ptr])                 \n\t"
>+      "gsldrc1    %[src],          0x00(%[src_ptr])                 \n\t"
>+      "punpcklbh  %[src_lo],       %[src],            %[mask]       \n\t"
>+      "punpckhbh  %[src_hi],       %[src],            %[mask]       \n\t"
>+
>+      "gsldrc1    %[dest0],        0x00(%[dst_ptr])                 \n\t"
>+      "gsldlc1    %[dest0],        0x07(%[dst_ptr])                 \n\t"
>+      "paddush    %[dest0],        %[dest0],          %[src_lo]     \n\t"
>+      "gsldrc1    %[dest1],        0x08(%[dst_ptr])                 \n\t"
>+      "gsldlc1    %[dest1],        0x0f(%[dst_ptr])                 \n\t"
>+      "paddush    %[dest1],        %[dest1],          %[src_hi]     \n\t"
>+
>+      "gssdlc1    %[dest0],        0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest0],        0x00(%[dst_ptr])                 \n\t"
>+      "gssdlc1    %[dest1],        0x0f(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest1],        0x08(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x08          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x10          \n\t"
>+      "daddi      %[width],        %[width],         -0x08          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [dest0] "=&f"(dest0), [dest1] "=&f"(dest1), [src_hi] "=&f"(src_hi),
>+        [src_lo] "=&f"(src_lo), [src] "=&f"(src)
>+      : [src_ptr] "r"(src_ptr), [dst_ptr] "r"(dst_ptr), [width] "r"(src_width),
>+        [mask] "f"(mask)
>+      : "memory");
>+}
>+
>+void ScaleAddRow_16_MMI(const uint16_t* src_ptr,
>+                        uint32_t* dst_ptr,
>+                        int src_width) {
>+  uint64_t src, src_hi, src_lo, dest0, dest1;
>+  const uint64_t mask = 0x0ULL;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldrc1    %[src],          0x00(%[src_ptr])                 \n\t"
>+      "gsldlc1    %[src],          0x07(%[src_ptr])                 \n\t"
>+      "punpcklhw  %[src_lo],       %[src],            %[mask]       \n\t"
>+      "punpckhhw  %[src_hi],       %[src],            %[mask]       \n\t"
>+
>+      "gsldrc1    %[dest0],        0x00(%[dst_ptr])                 \n\t"
>+      "gsldlc1    %[dest0],        0x07(%[dst_ptr])                 \n\t"
>+      "paddw      %[dest0],        %[dest0],          %[src_lo]     \n\t"
>+      "gssdlc1    %[dest0],        0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest0],        0x00(%[dst_ptr])                 \n\t"
>+
>+      "gsldrc1    %[dest1],        0x08(%[dst_ptr])                 \n\t"
>+      "gsldlc1    %[dest1],        0x0f(%[dst_ptr])                 \n\t"
>+      "paddw      %[dest1],        %[dest1],          %[src_hi]     \n\t"
>+      "gssdlc1    %[dest1],        0x0f(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest1],        0x08(%[dst_ptr])                 \n\t"
>+
>+      "daddiu     %[src_ptr],      %[src_ptr],        0x08          \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x10          \n\t"
>+      "daddi      %[width],        %[width],         -0x04          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [dest0] "=&f"(dest0), [dest1] "=&f"(dest1), [src_hi] "=&f"(src_hi),
>+        [src_lo] "=&f"(src_lo), [src] "=&f"(src)
>+      : [src_ptr] "r"(src_ptr), [dst_ptr] "r"(dst_ptr), [width] "r"(src_width),
>+        [mask] "f"(mask)
>+      : "memory");
>+}
>+
>+void ScaleARGBRowDownEven_MMI(const uint8_t* src_argb,
>+                              ptrdiff_t src_stride,
>+                              int src_stepx,
>+                              uint8_t* dst_argb,
>+                              int dst_width) {
>+  (void)src_stride;
>+
>+  uint64_t src0, src1, dest;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "lwc1       %[src0],          0x00(%[src_ptr])                \n\t"
>+      "dadd       %[src_ptr],       %[src_ptr],       %[src_stepx_4]\n\t"
>+      "lwc1       %[src1],          0x00(%[src_ptr])                \n\t"
>+      "punpcklwd  %[dest],          %[src0],          %[src1]       \n\t"
>+
>+      "gssdlc1    %[dest],          0x07(%[dst_ptr])                \n\t"
>+      "gssdrc1    %[dest],          0x00(%[dst_ptr])                \n\t"
>+
>+      "dadd       %[src_ptr],       %[src_ptr],       %[src_stepx_4]\n\t"
>+      "daddiu     %[dst_ptr],       %[dst_ptr],       0x08          \n\t"
>+      "daddi      %[width],         %[width],        -0x02          \n\t"
>+      "bnez       %[width],         1b                              \n\t"
>+      : [src0] "=&f"(src0), [src1] "=&f"(src1), [dest] "=&f"(dest)
>+      : [src_ptr] "r"(src_argb), [dst_ptr] "r"(dst_argb),
>+        [src_stepx_4] "r"(src_stepx << 2), [width] "r"(dst_width)
>+      : "memory");
>+}
>+
>+void ScaleARGBRowDownEvenBox_MMI(const uint8_t* src_argb,
>+                                 ptrdiff_t src_stride,
>+                                 int src_stepx,
>+                                 uint8_t* dst_argb,
>+                                 int dst_width) {
>+  const uint8_t* src0_ptr = src_argb;
>+  const uint8_t* src1_ptr = src_argb + src_stride;
>+
>+  uint64_t src0, src1, src_hi, src_lo;
>+  uint64_t dest, dest_hi, dest_lo, dest0, dest1;
>+
>+  const uint64_t mask = 0x0ULL;
>+  const uint64_t ph = 0x0002000200020002ULL;
>+  const uint64_t shift = 0x2ULL;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+
>+      "lwc1       %[src0],         0x00(%[src0_ptr])                \n\t"
>+      "punpcklbh  %[dest_lo],      %[src0],          %[mask]        \n\t"
>+      "lwc1       %[src0],         0x04(%[src0_ptr])                \n\t"
>+      "punpcklbh  %[dest_hi],      %[src0],          %[mask]        \n\t"
>+
>+      "lwc1       %[src1],         0x00(%[src1_ptr])                \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],          %[mask]        \n\t"
>+      "lwc1       %[src1],         0x04(%[src1_ptr])                \n\t"
>+      "punpcklbh  %[src_hi],       %[src1],          %[mask]        \n\t"
>+      "paddh      %[dest_lo],      %[dest_lo],       %[src_lo]      \n\t"
>+      "paddh      %[dest_hi],      %[dest_hi],       %[src_hi]      \n\t"
>+      "paddh      %[dest0],        %[dest_hi],       %[dest_lo]     \n\t"
>+      "paddh      %[dest0],        %[dest0],         %[ph]          \n\t"
>+      "psrlh      %[dest0],        %[dest0],         %[shift]       \n\t"
>+
>+      "dadd       %[src0_ptr],     %[src0_ptr],      %[src_stepx_4] \n\t"
>+      "dadd       %[src1_ptr],     %[src1_ptr],      %[src_stepx_4] \n\t"
>+
>+      "lwc1       %[src0],         0x00(%[src0_ptr])                \n\t"
>+      "punpcklbh  %[dest_lo],      %[src0],          %[mask]        \n\t"
>+      "lwc1       %[src0],         0x04(%[src0_ptr])                \n\t"
>+      "punpcklbh  %[dest_hi],      %[src0],          %[mask]        \n\t"
>+
>+      "lwc1       %[src1],         0x00(%[src1_ptr])                \n\t"
>+      "punpcklbh  %[src_lo],       %[src1],          %[mask]        \n\t"
>+      "lwc1       %[src1],         0x04(%[src1_ptr])                \n\t"
>+      "punpcklbh  %[src_hi],       %[src1],          %[mask]        \n\t"
>+      "paddh      %[dest_lo],      %[dest_lo],       %[src_lo]      \n\t"
>+      "paddh      %[dest_hi],      %[dest_hi],       %[src_hi]      \n\t"
>+      "paddh      %[dest1],        %[dest_hi],       %[dest_lo]     \n\t"
>+      "paddh      %[dest1],        %[dest1],         %[ph]          \n\t"
>+      "psrlh      %[dest1],        %[dest1],         %[shift]       \n\t"
>+
>+      "packushb   %[dest],         %[dest0],          %[dest1]      \n\t"
>+      "gssdlc1    %[dest],         0x07(%[dst_ptr])                 \n\t"
>+      "gssdrc1    %[dest],         0x00(%[dst_ptr])                 \n\t"
>+
>+      "dadd       %[src0_ptr],     %[src0_ptr],      %[src_stepx_4] \n\t"
>+      "dadd       %[src1_ptr],     %[src1_ptr],      %[src_stepx_4] \n\t"
>+      "daddiu     %[dst_ptr],      %[dst_ptr],        0x08          \n\t"
>+      "daddi      %[width],        %[width],         -0x02          \n\t"
>+      "bnez       %[width],        1b                               \n\t"
>+      : [src_hi] "=&f"(src_hi), [src_lo] "=&f"(src_lo),
>+        [dest_hi] "=&f"(dest_hi), [dest_lo] "=&f"(dest_lo),
>+        [dest0] "=&f"(dest0), [dest1] "=&f"(dest1), [src0] "=&f"(src0),
>+        [src1] "=&f"(src1), [dest] "=&f"(dest)
>+      : [src0_ptr] "r"(src0_ptr), [src1_ptr] "r"(src1_ptr),
>+        [dst_ptr] "r"(dst_argb), [width] "r"(dst_width),
>+        [src_stepx_4] "r"(src_stepx << 2), [shift] "f"(shift), [mask] "f"(mask),
>+        [ph] "f"(ph)
>+      : "memory");
>+}
>+
>+// Scales a single row of pixels using point sampling.
>+void ScaleARGBCols_MMI(uint8_t* dst_argb,
>+                       const uint8_t* src_argb,
>+                       int dst_width,
>+                       int x,
>+                       int dx) {
>+  const uint32_t* src = (const uint32_t*)(src_argb);
>+  uint32_t* dst = (uint32_t*)(dst_argb);
>+
>+  const uint32_t* src_tmp;
>+
>+  uint64_t dest, offset;
>+
>+  const uint64_t shift0 = 16;
>+  const uint64_t shift1 = 2;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "srav       %[offset],        %[x],             %[shift0]     \n\t"
>+      "sllv       %[offset],        %[offset],        %[shift1]     \n\t"
>+      "dadd       %[src_tmp],       %[src_ptr],       %[offset]     \n\t"
>+      "lwc1       %[dest],          0x00(%[src_tmp])                \n\t"
>+      "swc1       %[dest],          0x00(%[dst_ptr])                \n\t"
>+
>+      "dadd       %[x],             %[x],             %[dx]         \n\t"
>+
>+      "daddiu     %[dst_ptr],       %[dst_ptr],       0x04          \n\t"
>+      "daddi      %[width],         %[width],        -0x01          \n\t"
>+      "bnez       %[width],         1b                              \n\t"
>+      : [dest] "=&f"(dest), [offset] "=&r"(offset), [src_tmp] "=&r"(src_tmp)
>+      : [src_ptr] "r"(src), [dst_ptr] "r"(dst), [width] "r"(dst_width),
>+        [dx] "r"(dx), [x] "r"(x), [shift0] "r"(shift0), [shift1] "r"(shift1)
>+      : "memory");
>+}
>+
>+// Scales a single row of pixels up by 2x using point sampling.
>+void ScaleARGBColsUp2_MMI(uint8_t* dst_argb,
>+                          const uint8_t* src_argb,
>+                          int dst_width,
>+                          int x,
>+                          int dx) {
>+  uint64_t src, dest0, dest1;
>+  (void)x;
>+  (void)dx;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldrc1    %[src],           0x00(%[src_ptr])                \n\t"
>+      "gsldlc1    %[src],           0x07(%[src_ptr])                \n\t"
>+      "punpcklwd  %[dest0],         %[src],           %[src]        \n\t"
>+      "gssdlc1    %[dest0],         0x07(%[dst_ptr])                \n\t"
>+      "gssdrc1    %[dest0],         0x00(%[dst_ptr])                \n\t"
>+      "punpckhwd  %[dest1],         %[src],           %[src]        \n\t"
>+      "gssdlc1    %[dest1],         0x0f(%[dst_ptr])                \n\t"
>+      "gssdrc1    %[dest1],         0x08(%[dst_ptr])                \n\t"
>+
>+      "daddiu     %[src_ptr],       %[src_ptr],       0x08          \n\t"
>+      "daddiu     %[dst_ptr],       %[dst_ptr],       0x10          \n\t"
>+      "daddi      %[width],         %[width],        -0x04          \n\t"
>+      "bnez       %[width],         1b                              \n\t"
>+      : [dest0] "=&f"(dest0), [dest1] "=&f"(dest1), [src] "=&f"(src)
>+      : [src_ptr] "r"(src_argb), [dst_ptr] "r"(dst_argb), [width] "r"(dst_width)
>+      : "memory");
>+}
>+
>+// Divide num by div and return as 16.16 fixed point result.
>+/* LibYUVBaseTest.TestFixedDiv */
>+int FixedDiv_MIPS(int num, int div) {
>+  int quotient = 0;
>+  const int shift = 16;
>+
>+  asm(
>+      "dsll    %[num],     %[num],     %[shift]    \n\t"
>+      "ddiv    %[num],     %[div]                  \t\n"
>+      "mflo    %[quo]                              \t\n"
>+      : [quo] "+&r"(quotient)
>+      : [num] "r"(num), [div] "r"(div), [shift] "r"(shift));
>+
>+  return quotient;
>+}
>+
>+// Divide num by div and return as 16.16 fixed point result.
>+/* LibYUVScaleTest.ARGBScaleTo320x240_Linear */
>+int FixedDiv1_MIPS(int num, int div) {
>+  int quotient = 0;
>+  const int shift = 16;
>+  const int val1 = 1;
>+  const int64_t val11 = 0x00010001ULL;
>+
>+  asm(
>+      "dsll    %[num],     %[num],     %[shift]    \n\t"
>+      "dsub    %[num],     %[num],     %[val11]    \n\t"
>+      "dsub    %[div],     %[div],     %[val1]     \n\t"
>+      "ddiv    %[num],     %[div]                  \t\n"
>+      "mflo    %[quo]                              \t\n"
>+      : [quo] "+&r"(quotient)
>+      : [num] "r"(num), [div] "r"(div), [val1] "r"(val1), [val11] "r"(val11),
>+        [shift] "r"(shift));
>+
>+  return quotient;
>+}
>+
>+// Read 8x2 upsample with filtering and write 16x1.
>+// actually reads an extra pixel, so 9x2.
>+void ScaleRowUp2_16_MMI(const uint16_t* src_ptr,
>+                        ptrdiff_t src_stride,
>+                        uint16_t* dst,
>+                        int dst_width) {
>+  const uint16_t* src2_ptr = src_ptr + src_stride;
>+
>+  uint64_t src0, src1;
>+  uint64_t dest, dest04, dest15, dest26, dest37;
>+  uint64_t tmp0, tmp1, tmp2, tmp3;
>+
>+  const uint64_t mask0 = 0x0003000900030009ULL;
>+  const uint64_t mask1 = 0x0001000300010003ULL;
>+  const uint64_t mask2 = 0x0009000300090003ULL;
>+  const uint64_t mask3 = 0x0003000100030001ULL;
>+  const uint64_t ph = 0x0000000800000008ULL;
>+  const uint64_t shift = 4;
>+
>+  __asm__ volatile(
>+      "1:                                                           \n\t"
>+      "gsldrc1    %[src0],          0x00(%[src1_ptr])               \n\t"
>+      "gsldlc1    %[src0],          0x07(%[src1_ptr])               \n\t"
>+      "pmaddhw    %[dest04],        %[src0],          %[mask0]      \n\t"
>+      "gsldrc1    %[src1],          0x00(%[src2_ptr])               \n\t"
>+      "gsldlc1    %[src1],          0x07(%[src2_ptr])               \n\t"
>+      "pmaddhw    %[dest],          %[src1],          %[mask1]      \n\t"
>+      "paddw      %[dest04],        %[dest04],        %[dest]       \n\t"
>+      "paddw      %[dest04],        %[dest04],        %[ph]         \n\t"
>+      "psrlw      %[dest04],        %[dest04],        %[shift]      \n\t"
>+
>+      "pmaddhw    %[dest15],        %[src0],          %[mask2]      \n\t"
>+      "pmaddhw    %[dest],          %[src1],          %[mask3]      \n\t"
>+      "paddw      %[dest15],        %[dest15],        %[dest]       \n\t"
>+      "paddw      %[dest15],        %[dest15],        %[ph]         \n\t"
>+      "psrlw      %[dest15],        %[dest15],        %[shift]      \n\t"
>+
>+      "gsldrc1    %[src0],          0x02(%[src1_ptr])               \n\t"
>+      "gsldlc1    %[src0],          0x09(%[src1_ptr])               \n\t"
>+      "pmaddhw    %[dest26],        %[src0],          %[mask0]      \n\t"
>+      "gsldrc1    %[src1],          0x02(%[src2_ptr])               \n\t"
>+      "gsldlc1    %[src1],          0x09(%[src2_ptr])               \n\t"
>+      "pmaddhw    %[dest],          %[src1],          %[mask1]      \n\t"
>+      "paddw      %[dest26],        %[dest26],        %[dest]       \n\t"
>+      "paddw      %[dest26],        %[dest26],        %[ph]         \n\t"
>+      "psrlw      %[dest26],        %[dest26],        %[shift]      \n\t"
>+
>+      "pmaddhw    %[dest37],        %[src0],          %[mask2]      \n\t"
>+      "pmaddhw    %[dest],          %[src1],          %[mask3]      \n\t"
>+      "paddw      %[dest37],        %[dest37],        %[dest]       \n\t"
>+      "paddw      %[dest37],        %[dest37],        %[ph]         \n\t"
>+      "psrlw      %[dest37],        %[dest37],        %[shift]      \n\t"
>+
>+      /* tmp0 = ( 00 04 02 06 ) */
>+      "packsswh   %[tmp0],          %[dest04],        %[dest26]     \n\t"
>+      /* tmp1 = ( 01 05 03 07 ) */
>+      "packsswh   %[tmp1],          %[dest15],        %[dest37]     \n\t"
>+
>+      /* tmp2 = ( 00 01 04 05 )*/
>+      "punpcklhw  %[tmp2],          %[tmp0],          %[tmp1]       \n\t"
>+      /* tmp3 = ( 02 03 06 07 )*/
>+      "punpckhhw  %[tmp3],          %[tmp0],          %[tmp1]       \n\t"
>+
>+      /* ( 00 01 02 03 ) */
>+      "punpcklwd  %[dest],          %[tmp2],          %[tmp3]       \n\t"
>+      "gssdlc1    %[dest],          0x07(%[dst_ptr])                \n\t"
>+      "gssdrc1    %[dest],          0x00(%[dst_ptr])                \n\t"
>+
>+      /* ( 04 05 06 07 ) */
>+      "punpckhwd  %[dest],          %[tmp2],          %[tmp3]       \n\t"
>+      "gssdlc1    %[dest],          0x0f(%[dst_ptr])                \n\t"
>+      "gssdrc1    %[dest],          0x08(%[dst_ptr])                \n\t"
>+
>+      "daddiu     %[src1_ptr],      %[src1_ptr],      0x08          \n\t"
>+      "daddiu     %[src2_ptr],      %[src2_ptr],      0x08          \n\t"
>+      "daddiu     %[dst_ptr],       %[dst_ptr],       0x10          \n\t"
>+      "daddi      %[width],         %[width],        -0x08          \n\t"
>+      "bnez       %[width],         1b                              \n\t"
>+      : [src0] "=&f"(src0), [src1] "=&f"(src1), [dest04] "=&f"(dest04),
>+        [dest15] "=&f"(dest15), [dest26] "=&f"(dest26), [dest37] "=&f"(dest37),
>+        [tmp0] "=&f"(tmp0), [tmp1] "=&f"(tmp1), [tmp2] "=&f"(tmp2),
>+        [tmp3] "=&f"(tmp3), [dest] "=&f"(dest)
>+      : [src1_ptr] "r"(src_ptr), [src2_ptr] "r"(src2_ptr), [dst_ptr] "r"(dst),
>+        [width] "r"(dst_width), [mask0] "f"(mask0), [mask1] "f"(mask1),
>+        [mask2] "f"(mask2), [mask3] "f"(mask3), [shift] "f"(shift), [ph] "f"(ph)
>+      : "memory");
>+}
>+
>+#endif  // !defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)
>+
>+#ifdef __cplusplus
>+}  // extern "C"
>+}  // namespace libyuv
>+#endif
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale_neon.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale_neon.cc
>index 459a2995dfe7530eaac460864436387d2130224c..46f5ba4cd4fddb1fab08e3347c5c857f66b8b03f 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale_neon.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale_neon.cc
>@@ -504,37 +504,25 @@ void ScaleRowDown38_2_Box_NEON(const uint8_t* src_ptr,
>       : "q0", "q1", "q2", "q3", "q13", "q14", "memory", "cc");
> }
> 
>-void ScaleAddRows_NEON(const uint8_t* src_ptr,
>-                       ptrdiff_t src_stride,
>-                       uint16_t* dst_ptr,
>-                       int src_width,
>-                       int src_height) {
>-  const uint8_t* src_tmp;
>+// Add a row of bytes to a row of shorts.  Used for box filter.
>+// Reads 16 bytes and accumulates to 16 shorts at a time.
>+void ScaleAddRow_NEON(const uint8_t* src_ptr,
>+                      uint16_t* dst_ptr,
>+                      int src_width) {
>   asm volatile(
>       "1:                                        \n"
>-      "mov       %0, %1                          \n"
>-      "mov       r12, %5                         \n"
>-      "veor      q2, q2, q2                      \n"
>-      "veor      q3, q3, q3                      \n"
>-      "2:                                        \n"
>-      // load 16 pixels into q0
>-      "vld1.8     {q0}, [%0], %3                 \n"
>-      "vaddw.u8   q3, q3, d1                     \n"
>-      "vaddw.u8   q2, q2, d0                     \n"
>-      "subs       r12, r12, #1                   \n"
>-      "bgt        2b                             \n"
>-      "vst1.16    {q2, q3}, [%2]!                \n"  // store pixels
>-      "add        %1, %1, #16                    \n"
>-      "subs       %4, %4, #16                    \n"  // 16 processed per loop
>+      "vld1.16    {q1, q2}, [%1]                 \n"  // load accumulator
>+      "vld1.8     {q0}, [%0]!                    \n"  // load 16 bytes
>+      "vaddw.u8   q2, q2, d1                     \n"  // add
>+      "vaddw.u8   q1, q1, d0                     \n"
>+      "vst1.16    {q1, q2}, [%1]!                \n"  // store accumulator
>+      "subs       %2, %2, #16                    \n"  // 16 processed per loop
>       "bgt        1b                             \n"
>-      : "=&r"(src_tmp),    // %0
>-        "+r"(src_ptr),     // %1
>-        "+r"(dst_ptr),     // %2
>-        "+r"(src_stride),  // %3
>-        "+r"(src_width),   // %4
>-        "+r"(src_height)   // %5
>+      : "+r"(src_ptr),   // %0
>+        "+r"(dst_ptr),   // %1
>+        "+r"(src_width)  // %2
>       :
>-      : "memory", "cc", "r12", "q0", "q1", "q2", "q3"  // Clobber List
>+      : "memory", "cc", "q0", "q1", "q2"  // Clobber List
>       );
> }
> 
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale_neon64.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale_neon64.cc
>index 494a9cfbfbe3c383e5ebf872c098a832fb82b238..f4aed5fc92f4f043c3f597b7b11b07f155d76283 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale_neon64.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/source/scale_neon64.cc
>@@ -515,37 +515,25 @@ void ScaleRowDown38_2_Box_NEON(const uint8_t* src_ptr,
>         "v19", "v30", "v31", "memory", "cc");
> }
> 
>-void ScaleAddRows_NEON(const uint8_t* src_ptr,
>-                       ptrdiff_t src_stride,
>-                       uint16_t* dst_ptr,
>-                       int src_width,
>-                       int src_height) {
>-  const uint8_t* src_tmp;
>+// Add a row of bytes to a row of shorts.  Used for box filter.
>+// Reads 16 bytes and accumulates to 16 shorts at a time.
>+void ScaleAddRow_NEON(const uint8_t* src_ptr,
>+                      uint16_t* dst_ptr,
>+                      int src_width) {
>   asm volatile(
>       "1:                                        \n"
>-      "mov       %0, %1                          \n"
>-      "mov       w12, %w5                        \n"
>-      "eor       v2.16b, v2.16b, v2.16b          \n"
>-      "eor       v3.16b, v3.16b, v3.16b          \n"
>-      "2:                                        \n"
>-      // load 16 pixels into q0
>-      "ld1       {v0.16b}, [%0], %3              \n"
>-      "uaddw2    v3.8h, v3.8h, v0.16b            \n"
>-      "uaddw     v2.8h, v2.8h, v0.8b             \n"
>-      "subs      w12, w12, #1                    \n"
>-      "b.gt      2b                              \n"
>-      "st1      {v2.8h, v3.8h}, [%2], #32        \n"  // store pixels
>-      "add      %1, %1, #16                      \n"
>-      "subs     %w4, %w4, #16                    \n"  // 16 processed per loop
>+      "ld1      {v1.8h, v2.8h}, [%1]             \n"  // load accumulator
>+      "ld1      {v0.16b}, [%0], #16              \n"  // load 16 bytes
>+      "uaddw2   v2.8h, v2.8h, v0.16b             \n"  // add
>+      "uaddw    v1.8h, v1.8h, v0.8b              \n"
>+      "st1      {v1.8h, v2.8h}, [%1], #32        \n"  // store accumulator
>+      "subs     %w2, %w2, #16                    \n"  // 16 processed per loop
>       "b.gt     1b                               \n"
>-      : "=&r"(src_tmp),    // %0
>-        "+r"(src_ptr),     // %1
>-        "+r"(dst_ptr),     // %2
>-        "+r"(src_stride),  // %3
>-        "+r"(src_width),   // %4
>-        "+r"(src_height)   // %5
>+      : "+r"(src_ptr),   // %0
>+        "+r"(dst_ptr),   // %1
>+        "+r"(src_width)  // %2
>       :
>-      : "memory", "cc", "w12", "v0", "v1", "v2", "v3"  // Clobber List
>+      : "memory", "cc", "v0", "v1", "v2"  // Clobber List
>       );
> }
> 
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/tools_libyuv/autoroller/roll_deps.py b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/tools_libyuv/autoroller/roll_deps.py
>index 09ddc40e730ff9c4fb93e8d4bcac5e4492b9a071..ea8f69f18825b5a4d92e917a8ee3c642c5b0976f 100755
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/tools_libyuv/autoroller/roll_deps.py
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/tools_libyuv/autoroller/roll_deps.py
>@@ -8,7 +8,7 @@
> # be found in the AUTHORS file in the root of the source tree.
> 
> # This is a modified copy of the script in
>-# https://chromium.googlesource.com/external/webrtc/+/master/tools-webrtc/autoroller/roll_deps.py
>+# https://webrtc.googlesource.com/src/+/master/tools_webrtc/autoroller/roll_deps.py
> # customized for libyuv.
> 
> 
>@@ -22,7 +22,7 @@ import os
> import re
> import subprocess
> import sys
>-import urllib
>+import urllib2
> 
> 
> # Skip these dependencies (list without solution name prefix).
>@@ -90,7 +90,7 @@ def ParseCommitPosition(commit_message):
>   for line in reversed(commit_message.splitlines()):
>     m = COMMIT_POSITION_RE.match(line.strip())
>     if m:
>-      return m.group(1)
>+      return int(m.group(1))
>   logging.error('Failed to parse commit position id from:\n%s\n',
>                 commit_message)
>   sys.exit(-1)
>@@ -109,7 +109,7 @@ def _RunCommand(command, working_dir=None, ignore_exit_code=False,
>   logging.debug('CMD: %s CWD: %s', ' '.join(command), working_dir)
>   env = os.environ.copy()
>   if extra_env:
>-    assert all(type(value) == str for value in extra_env.values())
>+    assert all(isinstance(value, str) for value in extra_env.values())
>     logging.debug('extra env: %s', extra_env)
>     env.update(extra_env)
>   p = subprocess.Popen(command, stdout=subprocess.PIPE,
>@@ -169,7 +169,7 @@ def ReadRemoteCrCommit(revision):
> 
> def ReadUrlContent(url):
>   """Connect to a remote host and read the contents. Returns a list of lines."""
>-  conn = urllib.urlopen(url)
>+  conn = urllib2.urlopen(url)
>   try:
>     return conn.readlines()
>   except IOError as e:
>@@ -210,6 +210,8 @@ def BuildDepsentryDict(deps_dict):
>     for path, deps_url_spec in deps_subdict.iteritems():
>       # The deps url is either an URL and a condition, or just the URL.
>       if isinstance(deps_url_spec, dict):
>+        if deps_url_spec.get('dep_type') == 'cipd':
>+          continue
>         deps_url = deps_url_spec['url']
>       else:
>         deps_url = deps_url_spec
>@@ -345,11 +347,9 @@ def UpdateDepsFile(deps_filename, old_cr_revision, new_cr_revision,
>           'contains all platforms in the target_os list, i.e.\n'
>           'target_os = ["android", "unix", "mac", "ios", "win"];\n'
>           'Then run "gclient sync" again.' % local_dep_dir)
>-    _, stderr = _RunCommand(
>-      ['roll-dep-svn', '--no-verify-revision', dep.path, dep.new_rev],
>-      working_dir=CHECKOUT_SRC_DIR, ignore_exit_code=True)
>-    if stderr:
>-      logging.warning('roll-dep-svn: %s', stderr)
>+    _RunCommand(
>+      ['gclient', 'setdep', '--revision', '%s@%s' % (dep.path, dep.new_rev)],
>+      working_dir=CHECKOUT_SRC_DIR)
> 
> 
> def _IsTreeClean():
>@@ -397,20 +397,36 @@ def _LocalCommit(commit_msg, dry_run):
>     _RunCommand(['git', 'commit', '-m', commit_msg])
> 
> 
>-def _UploadCL(dry_run, rietveld_email=None):
>-  logging.info('Uploading CL...')
>-  if not dry_run:
>-    cmd = ['git', 'cl', 'upload', '-f']
>-    if rietveld_email:
>-      cmd.append('--email=%s' % rietveld_email)
>-    _RunCommand(cmd, extra_env={'EDITOR': 'true'})
>+def ChooseCQMode(skip_cq, cq_over, current_commit_pos, new_commit_pos):
>+  if skip_cq:
>+    return 0
>+  if (new_commit_pos - current_commit_pos) < cq_over:
>+    return 1
>+  return 2
> 
> 
>-def _SendToCQ(dry_run, skip_cq):
>-  logging.info('Sending the CL to the CQ...')
>-  if not dry_run and not skip_cq:
>-    _RunCommand(['git', 'cl', 'set_commit'])
>-    logging.info('Sent the CL to the CQ.')
>+def _UploadCL(commit_queue_mode):
>+  """Upload the committed changes as a changelist to Gerrit.
>+
>+  commit_queue_mode:
>+    - 2: Submit to commit queue.
>+    - 1: Run trybots but do not submit to CQ.
>+    - 0: Skip CQ, upload only.
>+  """
>+  cmd = ['git', 'cl', 'upload', '--force', '--bypass-hooks', '--send-mail']
>+  if commit_queue_mode >= 2:
>+    logging.info('Sending the CL to the CQ...')
>+    cmd.extend(['--use-commit-queue'])
>+  elif commit_queue_mode >= 1:
>+    logging.info('Starting CQ dry run...')
>+    cmd.extend(['--cq-dry-run'])
>+  extra_env = {
>+      'EDITOR': 'true',
>+      'SKIP_GCE_AUTH_FOR_GIT': '1',
>+  }
>+  stdout, stderr = _RunCommand(cmd, extra_env=extra_env)
>+  logging.debug('Output from "git cl upload":\nstdout:\n%s\n\nstderr:\n%s',
>+      stdout, stderr)
> 
> 
> def main():
>@@ -420,10 +436,6 @@ def main():
>   p.add_argument('-r', '--revision',
>                  help=('Chromium Git revision to roll to. Defaults to the '
>                        'Chromium HEAD revision if omitted.'))
>-  p.add_argument('-u', '--rietveld-email',
>-                 help=('E-mail address to use for creating the CL at Rietveld'
>-                       'If omitted a previously cached one will be used or an '
>-                       'error will be thrown during upload.'))
>   p.add_argument('--dry-run', action='store_true', default=False,
>                  help=('Calculate changes and modify DEPS, but don\'t create '
>                        'any local branch, commit, upload CL or send any '
>@@ -432,8 +444,12 @@ def main():
>                  default=False,
>                  help=('Ignore if the current branch is not master or if there '
>                        'are uncommitted changes (default: %(default)s).'))
>-  p.add_argument('--skip-cq', action='store_true', default=False,
>-                 help='Skip sending the CL to the CQ (default: %(default)s)')
>+  grp = p.add_mutually_exclusive_group()
>+  grp.add_argument('--skip-cq', action='store_true', default=False,
>+                   help='Skip sending the CL to the CQ (default: %(default)s)')
>+  grp.add_argument('--cq-over', type=int, default=1,
>+                   help=('Commit queue dry run if the revision difference '
>+                         'is below this number (default: %(default)s)'))
>   p.add_argument('-v', '--verbose', action='store_true', default=False,
>                  help='Be extra verbose in printing of log messages.')
>   opts = p.parse_args()
>@@ -478,8 +494,11 @@ def main():
>   _CreateRollBranch(opts.dry_run)
>   UpdateDepsFile(deps_filename, current_cr_rev, new_cr_rev, changed_deps)
>   _LocalCommit(commit_msg, opts.dry_run)
>-  _UploadCL(opts.dry_run, opts.rietveld_email)
>-  _SendToCQ(opts.dry_run, opts.skip_cq)
>+  commit_queue_mode = ChooseCQMode(opts.skip_cq, opts.cq_over,
>+                                   current_commit_pos, new_commit_pos)
>+  logging.info('Uploading CL...')
>+  if not opts.dry_run:
>+    _UploadCL(commit_queue_mode)
>   return 0
> 
> 
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/tools_libyuv/valgrind/libyuv_tests.bat b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/tools_libyuv/valgrind/libyuv_tests.bat
>index e37f09eb25f3e154863e1c10a6ef466ebd9dcfa1..5fceca67620cf85d2e0e47a05aa7261e76412500 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/tools_libyuv/valgrind/libyuv_tests.bat
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/tools_libyuv/valgrind/libyuv_tests.bat
>@@ -1,79 +1,79 @@
>-@echo off
>-:: Copyright (c) 2012 The LibYuv Project Authors. All rights reserved.
>-::
>-:: Use of this source code is governed by a BSD-style license
>-:: that can be found in the LICENSE file in the root of the source
>-:: tree. An additional intellectual property rights grant can be found
>-:: in the file PATENTS.  All contributing project authors may
>-:: be found in the AUTHORS file in the root of the source tree.
>-
>-:: This script is a copy of chrome_tests.bat with the following changes:
>-:: - Invokes libyuv_tests.py instead of chrome_tests.py
>-:: - Chromium's Valgrind scripts directory is added to the PYTHONPATH to make
>-::   it possible to execute the Python scripts properly.
>-
>-:: TODO(timurrrr): batch files 'export' all the variables to the parent shell
>-set THISDIR=%~dp0
>-set TOOL_NAME="unknown"
>-
>-:: Get the tool name and put it into TOOL_NAME {{{1
>-:: NB: SHIFT command doesn't modify %*
>-:PARSE_ARGS_LOOP
>-  if %1 == () GOTO:TOOLNAME_NOT_FOUND
>-  if %1 == --tool GOTO:TOOLNAME_FOUND
>-  SHIFT
>-  goto :PARSE_ARGS_LOOP
>-
>-:TOOLNAME_NOT_FOUND
>-echo "Please specify a tool (tsan or drmemory) by using --tool flag"
>-exit /B 1
>-
>-:TOOLNAME_FOUND
>-SHIFT
>-set TOOL_NAME=%1
>-:: }}}
>-if "%TOOL_NAME%" == "drmemory"          GOTO :SETUP_DRMEMORY
>-if "%TOOL_NAME%" == "drmemory_light"    GOTO :SETUP_DRMEMORY
>-if "%TOOL_NAME%" == "drmemory_full"     GOTO :SETUP_DRMEMORY
>-if "%TOOL_NAME%" == "drmemory_pattern"  GOTO :SETUP_DRMEMORY
>-if "%TOOL_NAME%" == "tsan"     GOTO :SETUP_TSAN
>-echo "Unknown tool: `%TOOL_NAME%`! Only tsan and drmemory are supported."
>-exit /B 1
>-
>-:SETUP_DRMEMORY
>-if NOT "%DRMEMORY_COMMAND%"=="" GOTO :RUN_TESTS
>-:: Set up DRMEMORY_COMMAND to invoke Dr. Memory {{{1
>-set DRMEMORY_PATH=%THISDIR%..\..\third_party\drmemory
>-set DRMEMORY_SFX=%DRMEMORY_PATH%\drmemory-windows-sfx.exe
>-if EXIST %DRMEMORY_SFX% GOTO DRMEMORY_BINARY_OK
>-echo "Can't find Dr. Memory executables."
>-echo "See http://www.chromium.org/developers/how-tos/using-valgrind/dr-memory"
>-echo "for the instructions on how to get them."
>-exit /B 1
>-
>-:DRMEMORY_BINARY_OK
>-%DRMEMORY_SFX% -o%DRMEMORY_PATH%\unpacked -y
>-set DRMEMORY_COMMAND=%DRMEMORY_PATH%\unpacked\bin\drmemory.exe
>-:: }}}
>-goto :RUN_TESTS
>-
>-:SETUP_TSAN
>-:: Set up PIN_COMMAND to invoke TSan {{{1
>-set TSAN_PATH=%THISDIR%..\..\third_party\tsan
>-set TSAN_SFX=%TSAN_PATH%\tsan-x86-windows-sfx.exe
>-if EXIST %TSAN_SFX% GOTO TSAN_BINARY_OK
>-echo "Can't find ThreadSanitizer executables."
>-echo "See http://www.chromium.org/developers/how-tos/using-valgrind/threadsanitizer/threadsanitizer-on-windows"
>-echo "for the instructions on how to get them."
>-exit /B 1
>-
>-:TSAN_BINARY_OK
>-%TSAN_SFX% -o%TSAN_PATH%\unpacked -y
>-set PIN_COMMAND=%TSAN_PATH%\unpacked\tsan-x86-windows\tsan.bat
>-:: }}}
>-goto :RUN_TESTS
>-
>-:RUN_TESTS
>-set PYTHONPATH=%THISDIR%..\python\google;%THISDIR%..\valgrind
>-set RUNNING_ON_VALGRIND=yes
>-python %THISDIR%libyuv_tests.py %*
>+@echo off
>+:: Copyright (c) 2012 The LibYuv Project Authors. All rights reserved.
>+::
>+:: Use of this source code is governed by a BSD-style license
>+:: that can be found in the LICENSE file in the root of the source
>+:: tree. An additional intellectual property rights grant can be found
>+:: in the file PATENTS.  All contributing project authors may
>+:: be found in the AUTHORS file in the root of the source tree.
>+
>+:: This script is a copy of chrome_tests.bat with the following changes:
>+:: - Invokes libyuv_tests.py instead of chrome_tests.py
>+:: - Chromium's Valgrind scripts directory is added to the PYTHONPATH to make
>+::   it possible to execute the Python scripts properly.
>+
>+:: TODO(timurrrr): batch files 'export' all the variables to the parent shell
>+set THISDIR=%~dp0
>+set TOOL_NAME="unknown"
>+
>+:: Get the tool name and put it into TOOL_NAME {{{1
>+:: NB: SHIFT command doesn't modify %*
>+:PARSE_ARGS_LOOP
>+  if %1 == () GOTO:TOOLNAME_NOT_FOUND
>+  if %1 == --tool GOTO:TOOLNAME_FOUND
>+  SHIFT
>+  goto :PARSE_ARGS_LOOP
>+
>+:TOOLNAME_NOT_FOUND
>+echo "Please specify a tool (tsan or drmemory) by using --tool flag"
>+exit /B 1
>+
>+:TOOLNAME_FOUND
>+SHIFT
>+set TOOL_NAME=%1
>+:: }}}
>+if "%TOOL_NAME%" == "drmemory"          GOTO :SETUP_DRMEMORY
>+if "%TOOL_NAME%" == "drmemory_light"    GOTO :SETUP_DRMEMORY
>+if "%TOOL_NAME%" == "drmemory_full"     GOTO :SETUP_DRMEMORY
>+if "%TOOL_NAME%" == "drmemory_pattern"  GOTO :SETUP_DRMEMORY
>+if "%TOOL_NAME%" == "tsan"     GOTO :SETUP_TSAN
>+echo "Unknown tool: `%TOOL_NAME%`! Only tsan and drmemory are supported."
>+exit /B 1
>+
>+:SETUP_DRMEMORY
>+if NOT "%DRMEMORY_COMMAND%"=="" GOTO :RUN_TESTS
>+:: Set up DRMEMORY_COMMAND to invoke Dr. Memory {{{1
>+set DRMEMORY_PATH=%THISDIR%..\..\third_party\drmemory
>+set DRMEMORY_SFX=%DRMEMORY_PATH%\drmemory-windows-sfx.exe
>+if EXIST %DRMEMORY_SFX% GOTO DRMEMORY_BINARY_OK
>+echo "Can't find Dr. Memory executables."
>+echo "See http://www.chromium.org/developers/how-tos/using-valgrind/dr-memory"
>+echo "for the instructions on how to get them."
>+exit /B 1
>+
>+:DRMEMORY_BINARY_OK
>+%DRMEMORY_SFX% -o%DRMEMORY_PATH%\unpacked -y
>+set DRMEMORY_COMMAND=%DRMEMORY_PATH%\unpacked\bin\drmemory.exe
>+:: }}}
>+goto :RUN_TESTS
>+
>+:SETUP_TSAN
>+:: Set up PIN_COMMAND to invoke TSan {{{1
>+set TSAN_PATH=%THISDIR%..\..\third_party\tsan
>+set TSAN_SFX=%TSAN_PATH%\tsan-x86-windows-sfx.exe
>+if EXIST %TSAN_SFX% GOTO TSAN_BINARY_OK
>+echo "Can't find ThreadSanitizer executables."
>+echo "See http://www.chromium.org/developers/how-tos/using-valgrind/threadsanitizer/threadsanitizer-on-windows"
>+echo "for the instructions on how to get them."
>+exit /B 1
>+
>+:TSAN_BINARY_OK
>+%TSAN_SFX% -o%TSAN_PATH%\unpacked -y
>+set PIN_COMMAND=%TSAN_PATH%\unpacked\tsan-x86-windows\tsan.bat
>+:: }}}
>+goto :RUN_TESTS
>+
>+:RUN_TESTS
>+set PYTHONPATH=%THISDIR%..\python\google;%THISDIR%..\valgrind
>+set RUNNING_ON_VALGRIND=yes
>+python %THISDIR%libyuv_tests.py %*
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/convert_test.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/convert_test.cc
>index 39281ae08076d7d5ed2d209ab36998f269817fbd..d97b4fc723c740b0f7cd141c4d77b892c7d44b39 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/convert_test.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/convert_test.cc
>@@ -281,6 +281,23 @@ TESTAPLANARTOP(Android420, I420, 1, 0, 0, 2, 2, I420, 2, 2)
> TESTAPLANARTOP(Android420, NV12, 2, 0, 1, 2, 2, I420, 2, 2)
> TESTAPLANARTOP(Android420, NV21, 2, 1, 0, 2, 2, I420, 2, 2)
> 
>+// wrapper to keep API the same
>+int I400ToNV21(const uint8_t* src_y,
>+               int src_stride_y,
>+               const uint8_t* /* src_u */,
>+               int /* src_stride_u */,
>+               const uint8_t* /* src_v */,
>+               int /* src_stride_v */,
>+               uint8_t* dst_y,
>+               int dst_stride_y,
>+               uint8_t* dst_vu,
>+               int dst_stride_vu,
>+               int width,
>+               int height) {
>+  return I400ToNV21(src_y, src_stride_y, dst_y, dst_stride_y, dst_vu,
>+                    dst_stride_vu, width, height);
>+}
>+
> #define TESTPLANARTOBPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y,         \
>                         FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, W1280, N, NEG, OFF) \
>   TEST_F(LibYUVConvertTest, SRC_FMT_PLANAR##To##FMT_PLANAR##N) {              \
>@@ -374,6 +391,9 @@ TESTAPLANARTOP(Android420, NV21, 2, 1, 0, 2, 2, I420, 2, 2)
> 
> TESTPLANARTOBP(I420, 2, 2, NV12, 2, 2)
> TESTPLANARTOBP(I420, 2, 2, NV21, 2, 2)
>+TESTPLANARTOBP(I422, 2, 1, NV21, 2, 2)
>+TESTPLANARTOBP(I444, 1, 1, NV21, 2, 2)
>+TESTPLANARTOBP(I400, 2, 2, NV21, 2, 2)
> 
> #define TESTBIPLANARTOPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y,         \
>                          FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, W1280, N, NEG, OFF, \
>@@ -566,6 +586,8 @@ TESTPLANARTOB(I420, 2, 2, RGB24, 3, 3, 1)
> TESTPLANARTOB(H420, 2, 2, RAW, 3, 3, 1)
> TESTPLANARTOB(H420, 2, 2, RGB24, 3, 3, 1)
> TESTPLANARTOB(I420, 2, 2, RGB565, 2, 2, 1)
>+TESTPLANARTOB(J420, 2, 2, RGB565, 2, 2, 1)
>+TESTPLANARTOB(H420, 2, 2, RGB565, 2, 2, 1)
> TESTPLANARTOB(I420, 2, 2, ARGB1555, 2, 2, 1)
> TESTPLANARTOB(I420, 2, 2, ARGB4444, 2, 2, 1)
> TESTPLANARTOB(I422, 2, 1, ARGB, 4, 4, 1)
>@@ -734,6 +756,8 @@ TESTBIPLANARTOB(NV12, 2, 2, ABGR, 4, 2)
> TESTBIPLANARTOB(NV21, 2, 2, ABGR, 4, 2)
> TESTBIPLANARTOB(NV12, 2, 2, RGB24, 3, 2)
> TESTBIPLANARTOB(NV21, 2, 2, RGB24, 3, 2)
>+TESTBIPLANARTOB(NV12, 2, 2, RAW, 3, 2)
>+TESTBIPLANARTOB(NV21, 2, 2, RAW, 3, 2)
> TESTBIPLANARTOB(NV12, 2, 2, RGB565, 2, 9)
> 
> #ifdef DO_THREE_PLANES
>@@ -1022,15 +1046,9 @@ TESTATOBIPLANAR(UYVY, 2, 4, NV12, 2, 2)
>       MaskCpuFlags(benchmark_cpu_info_);                                   \
>       FMT_A##To##FMT_B(src_argb, kStrideA, dst_argb_opt, kStrideB, kWidth, \
>                        kHeight);                                           \
>-      int max_diff = 0;                                                    \
>       for (int i = 0; i < kStrideB * kHeightB; ++i) {                      \
>-        int abs_diff = abs(static_cast<int>(dst_argb_c[i]) -               \
>-                           static_cast<int>(dst_argb_opt[i]));             \
>-        if (abs_diff > max_diff) {                                         \
>-          max_diff = abs_diff;                                             \
>-        }                                                                  \
>+        EXPECT_NEAR(dst_argb_c[i], dst_argb_opt[i], DIFF);                 \
>       }                                                                    \
>-      EXPECT_LE(max_diff, DIFF);                                           \
>       free_aligned_buffer_page_end(src_argb);                              \
>       free_aligned_buffer_page_end(dst_argb_c);                            \
>       free_aligned_buffer_page_end(dst_argb_opt);                          \
>@@ -1050,6 +1068,7 @@ TESTATOBIPLANAR(UYVY, 2, 4, NV12, 2, 2)
>   TESTATOBRANDOM(FMT_A, BPP_A, STRIDE_A, HEIGHT_A, FMT_B, BPP_B, STRIDE_B, \
>                  HEIGHT_B, DIFF)
> 
>+// TODO(fbarchard): make ARM version of C code that matches NEON.
> TESTATOB(ARGB, 4, 4, 1, ARGB, 4, 4, 1, 0)
> TESTATOB(ARGB, 4, 4, 1, BGRA, 4, 4, 1, 0)
> TESTATOB(ARGB, 4, 4, 1, ABGR, 4, 4, 1, 0)
>@@ -1361,72 +1380,613 @@ TEST_F(LibYUVConvertTest, FuzzJpeg) {
>   }
> }
> 
>-TEST_F(LibYUVConvertTest, MJPGToI420) {
>-  const int kOff = 10;
>-  const int kMinJpeg = 64;
>-  const int kImageSize = benchmark_width_ * benchmark_height_ >= kMinJpeg
>-                             ? benchmark_width_ * benchmark_height_
>-                             : kMinJpeg;
>-  const int kSize = kImageSize + kOff;
>-  align_buffer_page_end(orig_pixels, kSize);
>-  align_buffer_page_end(dst_y_opt, benchmark_width_ * benchmark_height_);
>-  align_buffer_page_end(dst_u_opt, SUBSAMPLE(benchmark_width_, 2) *
>-                                       SUBSAMPLE(benchmark_height_, 2));
>-  align_buffer_page_end(dst_v_opt, SUBSAMPLE(benchmark_width_, 2) *
>-                                       SUBSAMPLE(benchmark_height_, 2));
>+// Test data created in GIMP.  In export jpeg, disable thumbnails etc,
>+// choose a subsampling, and use low quality (50) to keep size small.
>+// Generated with xxd -i test.jpg
>+// test 0 is J400
>+static const uint8_t kTest0Jpg[] = {
>+    0xff, 0xd8, 0xff, 0xe0, 0x00, 0x10, 0x4a, 0x46, 0x49, 0x46, 0x00, 0x01,
>+    0x01, 0x01, 0x00, 0x48, 0x00, 0x48, 0x00, 0x00, 0xff, 0xdb, 0x00, 0x43,
>+    0x00, 0x10, 0x0b, 0x0c, 0x0e, 0x0c, 0x0a, 0x10, 0x0e, 0x0d, 0x0e, 0x12,
>+    0x11, 0x10, 0x13, 0x18, 0x28, 0x1a, 0x18, 0x16, 0x16, 0x18, 0x31, 0x23,
>+    0x25, 0x1d, 0x28, 0x3a, 0x33, 0x3d, 0x3c, 0x39, 0x33, 0x38, 0x37, 0x40,
>+    0x48, 0x5c, 0x4e, 0x40, 0x44, 0x57, 0x45, 0x37, 0x38, 0x50, 0x6d, 0x51,
>+    0x57, 0x5f, 0x62, 0x67, 0x68, 0x67, 0x3e, 0x4d, 0x71, 0x79, 0x70, 0x64,
>+    0x78, 0x5c, 0x65, 0x67, 0x63, 0xff, 0xc2, 0x00, 0x0b, 0x08, 0x00, 0x10,
>+    0x00, 0x20, 0x01, 0x01, 0x11, 0x00, 0xff, 0xc4, 0x00, 0x17, 0x00, 0x01,
>+    0x01, 0x01, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x03, 0x04, 0x01, 0x02, 0xff, 0xda, 0x00, 0x08, 0x01,
>+    0x01, 0x00, 0x00, 0x00, 0x01, 0x43, 0x7e, 0xa7, 0x97, 0x57, 0xff, 0xc4,
>+    0x00, 0x1b, 0x10, 0x00, 0x03, 0x00, 0x02, 0x03, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x02, 0x11, 0x00, 0x03,
>+    0x10, 0x12, 0x13, 0xff, 0xda, 0x00, 0x08, 0x01, 0x01, 0x00, 0x01, 0x05,
>+    0x02, 0x3b, 0xc0, 0x6f, 0x66, 0x76, 0x56, 0x23, 0x87, 0x99, 0x0d, 0x26,
>+    0x62, 0xf6, 0xbf, 0xff, 0xc4, 0x00, 0x1e, 0x10, 0x00, 0x02, 0x01, 0x03,
>+    0x05, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x11, 0x21, 0x02, 0x12, 0x32, 0x10, 0x31, 0x71, 0x81, 0xa1, 0xff,
>+    0xda, 0x00, 0x08, 0x01, 0x01, 0x00, 0x06, 0x3f, 0x02, 0x4b, 0xb3, 0x28,
>+    0x32, 0xd2, 0xed, 0xf9, 0x1d, 0x3e, 0x13, 0x51, 0x73, 0x83, 0xff, 0xc4,
>+    0x00, 0x1c, 0x10, 0x01, 0x01, 0x01, 0x00, 0x02, 0x03, 0x01, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x11, 0x00, 0x21, 0x51,
>+    0x31, 0x61, 0x81, 0xf0, 0xff, 0xda, 0x00, 0x08, 0x01, 0x01, 0x00, 0x01,
>+    0x3f, 0x21, 0x65, 0x6e, 0x31, 0x86, 0x28, 0xf9, 0x30, 0xdc, 0x27, 0xdb,
>+    0xa9, 0x01, 0xf3, 0xde, 0x02, 0xa0, 0xed, 0x1e, 0x34, 0x68, 0x23, 0xf9,
>+    0xc6, 0x48, 0x5d, 0x7a, 0x35, 0x02, 0xf5, 0x6f, 0xff, 0xda, 0x00, 0x08,
>+    0x01, 0x01, 0x00, 0x00, 0x00, 0x10, 0x35, 0xff, 0xc4, 0x00, 0x1f, 0x10,
>+    0x01, 0x00, 0x02, 0x01, 0x04, 0x03, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x11, 0x31, 0x41, 0x61, 0x71, 0x91,
>+    0x21, 0x81, 0xd1, 0xb1, 0xff, 0xda, 0x00, 0x08, 0x01, 0x01, 0x00, 0x01,
>+    0x3f, 0x10, 0x0b, 0x30, 0xe9, 0x58, 0xbe, 0x1a, 0xfd, 0x88, 0xab, 0x8b,
>+    0x34, 0x74, 0x80, 0x4b, 0xb5, 0xd5, 0xab, 0xcd, 0x46, 0x96, 0x2e, 0xec,
>+    0xbd, 0xaa, 0x78, 0x47, 0x5c, 0x47, 0xa7, 0x30, 0x49, 0xad, 0x88, 0x7c,
>+    0x40, 0x74, 0x30, 0xff, 0x00, 0x23, 0x1d, 0x03, 0x0b, 0xb7, 0xd4, 0xff,
>+    0xd9};
>+static const size_t kTest0JpgLen = 421;
>+
>+// test 1 is J444
>+static const uint8_t kTest1Jpg[] = {
>+    0xff, 0xd8, 0xff, 0xe0, 0x00, 0x10, 0x4a, 0x46, 0x49, 0x46, 0x00, 0x01,
>+    0x01, 0x01, 0x00, 0x48, 0x00, 0x48, 0x00, 0x00, 0xff, 0xdb, 0x00, 0x43,
>+    0x00, 0x10, 0x0b, 0x0c, 0x0e, 0x0c, 0x0a, 0x10, 0x0e, 0x0d, 0x0e, 0x12,
>+    0x11, 0x10, 0x13, 0x18, 0x28, 0x1a, 0x18, 0x16, 0x16, 0x18, 0x31, 0x23,
>+    0x25, 0x1d, 0x28, 0x3a, 0x33, 0x3d, 0x3c, 0x39, 0x33, 0x38, 0x37, 0x40,
>+    0x48, 0x5c, 0x4e, 0x40, 0x44, 0x57, 0x45, 0x37, 0x38, 0x50, 0x6d, 0x51,
>+    0x57, 0x5f, 0x62, 0x67, 0x68, 0x67, 0x3e, 0x4d, 0x71, 0x79, 0x70, 0x64,
>+    0x78, 0x5c, 0x65, 0x67, 0x63, 0xff, 0xdb, 0x00, 0x43, 0x01, 0x11, 0x12,
>+    0x12, 0x18, 0x15, 0x18, 0x2f, 0x1a, 0x1a, 0x2f, 0x63, 0x42, 0x38, 0x42,
>+    0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63,
>+    0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63,
>+    0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63,
>+    0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63,
>+    0x63, 0x63, 0xff, 0xc2, 0x00, 0x11, 0x08, 0x00, 0x10, 0x00, 0x20, 0x03,
>+    0x01, 0x11, 0x00, 0x02, 0x11, 0x01, 0x03, 0x11, 0x01, 0xff, 0xc4, 0x00,
>+    0x17, 0x00, 0x01, 0x01, 0x01, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x04, 0x01, 0x02, 0xff, 0xc4,
>+    0x00, 0x16, 0x01, 0x01, 0x01, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x01, 0x03, 0xff, 0xda,
>+    0x00, 0x0c, 0x03, 0x01, 0x00, 0x02, 0x10, 0x03, 0x10, 0x00, 0x00, 0x01,
>+    0x40, 0x8f, 0x26, 0xe8, 0xf4, 0xcc, 0xf9, 0x69, 0x2b, 0x1b, 0x2a, 0xcb,
>+    0xff, 0xc4, 0x00, 0x1b, 0x10, 0x00, 0x03, 0x00, 0x02, 0x03, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x02, 0x11,
>+    0x00, 0x03, 0x10, 0x12, 0x13, 0xff, 0xda, 0x00, 0x08, 0x01, 0x01, 0x00,
>+    0x01, 0x05, 0x02, 0x3b, 0x80, 0x6f, 0x56, 0x76, 0x56, 0x23, 0x87, 0x99,
>+    0x0d, 0x26, 0x62, 0xf6, 0xbf, 0xff, 0xc4, 0x00, 0x19, 0x11, 0x01, 0x00,
>+    0x03, 0x01, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x01, 0x00, 0x10, 0x11, 0x02, 0x12, 0xff, 0xda, 0x00, 0x08,
>+    0x01, 0x03, 0x01, 0x01, 0x3f, 0x01, 0xf1, 0x00, 0x27, 0x45, 0xbb, 0x31,
>+    0xaf, 0xff, 0xc4, 0x00, 0x1a, 0x11, 0x00, 0x02, 0x03, 0x01, 0x01, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01,
>+    0x02, 0x10, 0x11, 0x41, 0x12, 0xff, 0xda, 0x00, 0x08, 0x01, 0x02, 0x01,
>+    0x01, 0x3f, 0x01, 0xf6, 0x4b, 0x5f, 0x48, 0xb3, 0x69, 0x63, 0x35, 0x72,
>+    0xbf, 0xff, 0xc4, 0x00, 0x1e, 0x10, 0x00, 0x02, 0x01, 0x03, 0x05, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x11,
>+    0x21, 0x02, 0x12, 0x32, 0x10, 0x31, 0x71, 0x81, 0xa1, 0xff, 0xda, 0x00,
>+    0x08, 0x01, 0x01, 0x00, 0x06, 0x3f, 0x02, 0x4b, 0xb3, 0x28, 0x32, 0xd2,
>+    0xed, 0xf9, 0x1d, 0x3e, 0x13, 0x51, 0x73, 0x83, 0xff, 0xc4, 0x00, 0x1c,
>+    0x10, 0x01, 0x01, 0x01, 0x00, 0x02, 0x03, 0x01, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x11, 0x00, 0x21, 0x51, 0x31, 0x61,
>+    0x81, 0xf0, 0xff, 0xda, 0x00, 0x08, 0x01, 0x01, 0x00, 0x01, 0x3f, 0x21,
>+    0x75, 0x6e, 0x31, 0x94, 0x28, 0xf9, 0x30, 0xdc, 0x27, 0xdb, 0xa9, 0x01,
>+    0xf3, 0xde, 0x02, 0xa0, 0xed, 0x1e, 0x34, 0x68, 0x23, 0xf9, 0xc6, 0x48,
>+    0x5d, 0x7a, 0x35, 0x02, 0xf5, 0x6f, 0xff, 0xda, 0x00, 0x0c, 0x03, 0x01,
>+    0x00, 0x02, 0x00, 0x03, 0x00, 0x00, 0x00, 0x10, 0x26, 0x61, 0xd4, 0xff,
>+    0xc4, 0x00, 0x1a, 0x11, 0x00, 0x03, 0x01, 0x00, 0x03, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x11, 0x21,
>+    0x31, 0x41, 0x51, 0xff, 0xda, 0x00, 0x08, 0x01, 0x03, 0x01, 0x01, 0x3f,
>+    0x10, 0x54, 0xa8, 0xbf, 0x50, 0x87, 0xb0, 0x9d, 0x8b, 0xc4, 0x6a, 0x26,
>+    0x6b, 0x2a, 0x9c, 0x1f, 0xff, 0xc4, 0x00, 0x18, 0x11, 0x01, 0x01, 0x01,
>+    0x01, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x01, 0x00, 0x11, 0x21, 0x51, 0xff, 0xda, 0x00, 0x08, 0x01, 0x02,
>+    0x01, 0x01, 0x3f, 0x10, 0x70, 0xe1, 0x3e, 0xd1, 0x8e, 0x0d, 0xe1, 0xb5,
>+    0xd5, 0x91, 0x76, 0x43, 0x82, 0x45, 0x4c, 0x7b, 0x7f, 0xff, 0xc4, 0x00,
>+    0x1f, 0x10, 0x01, 0x00, 0x02, 0x01, 0x04, 0x03, 0x01, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x11, 0x31, 0x41, 0x61,
>+    0x71, 0x91, 0x21, 0x81, 0xd1, 0xb1, 0xff, 0xda, 0x00, 0x08, 0x01, 0x01,
>+    0x00, 0x01, 0x3f, 0x10, 0x1b, 0x30, 0xe9, 0x58, 0xbe, 0x1a, 0xfd, 0x8a,
>+    0xeb, 0x8b, 0x34, 0x74, 0x80, 0x4b, 0xb5, 0xd5, 0xab, 0xcd, 0x46, 0x96,
>+    0x2e, 0xec, 0xbd, 0xaa, 0x78, 0x47, 0x5c, 0x47, 0xa7, 0x30, 0x49, 0xad,
>+    0x88, 0x7c, 0x40, 0x74, 0x30, 0xff, 0x00, 0x23, 0x1d, 0x03, 0x0b, 0xb7,
>+    0xd4, 0xff, 0xd9};
>+static const size_t kTest1JpgLen = 735;
>+
>+// test 2 is J420
>+static const uint8_t kTest2Jpg[] = {
>+    0xff, 0xd8, 0xff, 0xe0, 0x00, 0x10, 0x4a, 0x46, 0x49, 0x46, 0x00, 0x01,
>+    0x01, 0x01, 0x00, 0x48, 0x00, 0x48, 0x00, 0x00, 0xff, 0xdb, 0x00, 0x43,
>+    0x00, 0x10, 0x0b, 0x0c, 0x0e, 0x0c, 0x0a, 0x10, 0x0e, 0x0d, 0x0e, 0x12,
>+    0x11, 0x10, 0x13, 0x18, 0x28, 0x1a, 0x18, 0x16, 0x16, 0x18, 0x31, 0x23,
>+    0x25, 0x1d, 0x28, 0x3a, 0x33, 0x3d, 0x3c, 0x39, 0x33, 0x38, 0x37, 0x40,
>+    0x48, 0x5c, 0x4e, 0x40, 0x44, 0x57, 0x45, 0x37, 0x38, 0x50, 0x6d, 0x51,
>+    0x57, 0x5f, 0x62, 0x67, 0x68, 0x67, 0x3e, 0x4d, 0x71, 0x79, 0x70, 0x64,
>+    0x78, 0x5c, 0x65, 0x67, 0x63, 0xff, 0xdb, 0x00, 0x43, 0x01, 0x11, 0x12,
>+    0x12, 0x18, 0x15, 0x18, 0x2f, 0x1a, 0x1a, 0x2f, 0x63, 0x42, 0x38, 0x42,
>+    0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63,
>+    0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63,
>+    0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63,
>+    0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63,
>+    0x63, 0x63, 0xff, 0xc2, 0x00, 0x11, 0x08, 0x00, 0x10, 0x00, 0x20, 0x03,
>+    0x01, 0x22, 0x00, 0x02, 0x11, 0x01, 0x03, 0x11, 0x01, 0xff, 0xc4, 0x00,
>+    0x18, 0x00, 0x00, 0x02, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x05, 0x01, 0x02, 0x04, 0xff,
>+    0xc4, 0x00, 0x16, 0x01, 0x01, 0x01, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x01, 0x02, 0xff,
>+    0xda, 0x00, 0x0c, 0x03, 0x01, 0x00, 0x02, 0x10, 0x03, 0x10, 0x00, 0x00,
>+    0x01, 0x20, 0xe7, 0x28, 0xa3, 0x0b, 0x2e, 0x2d, 0xcf, 0xff, 0xc4, 0x00,
>+    0x1b, 0x10, 0x00, 0x03, 0x00, 0x02, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x02, 0x11, 0x00, 0x03, 0x10,
>+    0x12, 0x13, 0xff, 0xda, 0x00, 0x08, 0x01, 0x01, 0x00, 0x01, 0x05, 0x02,
>+    0x3b, 0x80, 0x6f, 0x56, 0x76, 0x56, 0x23, 0x87, 0x99, 0x0d, 0x26, 0x62,
>+    0xf6, 0xbf, 0xff, 0xc4, 0x00, 0x17, 0x11, 0x01, 0x00, 0x03, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x01, 0x11, 0x21, 0xff, 0xda, 0x00, 0x08, 0x01, 0x03, 0x01, 0x01, 0x3f,
>+    0x01, 0xc8, 0x53, 0xff, 0xc4, 0x00, 0x16, 0x11, 0x01, 0x01, 0x01, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x11, 0x32, 0xff, 0xda, 0x00, 0x08, 0x01, 0x02, 0x01, 0x01, 0x3f,
>+    0x01, 0xd2, 0xc7, 0xff, 0xc4, 0x00, 0x1e, 0x10, 0x00, 0x02, 0x01, 0x03,
>+    0x05, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x11, 0x21, 0x02, 0x12, 0x32, 0x10, 0x31, 0x71, 0x81, 0xa1, 0xff,
>+    0xda, 0x00, 0x08, 0x01, 0x01, 0x00, 0x06, 0x3f, 0x02, 0x4b, 0xb3, 0x28,
>+    0x32, 0xd2, 0xed, 0xf9, 0x1d, 0x3e, 0x13, 0x51, 0x73, 0x83, 0xff, 0xc4,
>+    0x00, 0x1c, 0x10, 0x01, 0x01, 0x01, 0x00, 0x02, 0x03, 0x01, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x11, 0x00, 0x21, 0x51,
>+    0x31, 0x61, 0x81, 0xf0, 0xff, 0xda, 0x00, 0x08, 0x01, 0x01, 0x00, 0x01,
>+    0x3f, 0x21, 0x75, 0x6e, 0x31, 0x94, 0x28, 0xf9, 0x30, 0xdc, 0x27, 0xdb,
>+    0xa9, 0x01, 0xf3, 0xde, 0x02, 0xa0, 0xed, 0x1e, 0x34, 0x68, 0x23, 0xf9,
>+    0xc6, 0x48, 0x5d, 0x7a, 0x35, 0x02, 0xf5, 0x6f, 0xff, 0xda, 0x00, 0x0c,
>+    0x03, 0x01, 0x00, 0x02, 0x00, 0x03, 0x00, 0x00, 0x00, 0x10, 0x13, 0x5f,
>+    0xff, 0xc4, 0x00, 0x17, 0x11, 0x01, 0x01, 0x01, 0x01, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x11,
>+    0x21, 0xff, 0xda, 0x00, 0x08, 0x01, 0x03, 0x01, 0x01, 0x3f, 0x10, 0x0e,
>+    0xa1, 0x3a, 0x76, 0xff, 0xc4, 0x00, 0x17, 0x11, 0x01, 0x01, 0x01, 0x01,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x01, 0x00, 0x21, 0x11, 0xff, 0xda, 0x00, 0x08, 0x01, 0x02, 0x01, 0x01,
>+    0x3f, 0x10, 0x57, 0x0b, 0x08, 0x70, 0xdb, 0xff, 0xc4, 0x00, 0x1f, 0x10,
>+    0x01, 0x00, 0x02, 0x01, 0x04, 0x03, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x11, 0x31, 0x41, 0x61, 0x71, 0x91,
>+    0x21, 0x81, 0xd1, 0xb1, 0xff, 0xda, 0x00, 0x08, 0x01, 0x01, 0x00, 0x01,
>+    0x3f, 0x10, 0x1b, 0x30, 0xe9, 0x58, 0xbe, 0x1a, 0xfd, 0x8a, 0xeb, 0x8b,
>+    0x34, 0x74, 0x80, 0x4b, 0xb5, 0xd5, 0xab, 0xcd, 0x46, 0x96, 0x2e, 0xec,
>+    0xbd, 0xaa, 0x78, 0x47, 0x5c, 0x47, 0xa7, 0x30, 0x49, 0xad, 0x88, 0x7c,
>+    0x40, 0x74, 0x30, 0xff, 0x00, 0x23, 0x1d, 0x03, 0x0b, 0xb7, 0xd4, 0xff,
>+    0xd9};
>+static const size_t kTest2JpgLen = 685;
>+
>+// test 3 is J422
>+static const uint8_t kTest3Jpg[] = {
>+    0xff, 0xd8, 0xff, 0xe0, 0x00, 0x10, 0x4a, 0x46, 0x49, 0x46, 0x00, 0x01,
>+    0x01, 0x01, 0x00, 0x48, 0x00, 0x48, 0x00, 0x00, 0xff, 0xdb, 0x00, 0x43,
>+    0x00, 0x10, 0x0b, 0x0c, 0x0e, 0x0c, 0x0a, 0x10, 0x0e, 0x0d, 0x0e, 0x12,
>+    0x11, 0x10, 0x13, 0x18, 0x28, 0x1a, 0x18, 0x16, 0x16, 0x18, 0x31, 0x23,
>+    0x25, 0x1d, 0x28, 0x3a, 0x33, 0x3d, 0x3c, 0x39, 0x33, 0x38, 0x37, 0x40,
>+    0x48, 0x5c, 0x4e, 0x40, 0x44, 0x57, 0x45, 0x37, 0x38, 0x50, 0x6d, 0x51,
>+    0x57, 0x5f, 0x62, 0x67, 0x68, 0x67, 0x3e, 0x4d, 0x71, 0x79, 0x70, 0x64,
>+    0x78, 0x5c, 0x65, 0x67, 0x63, 0xff, 0xdb, 0x00, 0x43, 0x01, 0x11, 0x12,
>+    0x12, 0x18, 0x15, 0x18, 0x2f, 0x1a, 0x1a, 0x2f, 0x63, 0x42, 0x38, 0x42,
>+    0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63,
>+    0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63,
>+    0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63,
>+    0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63,
>+    0x63, 0x63, 0xff, 0xc2, 0x00, 0x11, 0x08, 0x00, 0x10, 0x00, 0x20, 0x03,
>+    0x01, 0x21, 0x00, 0x02, 0x11, 0x01, 0x03, 0x11, 0x01, 0xff, 0xc4, 0x00,
>+    0x17, 0x00, 0x01, 0x01, 0x01, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x04, 0x01, 0x02, 0xff, 0xc4,
>+    0x00, 0x17, 0x01, 0x00, 0x03, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x02, 0x03, 0x00, 0xff,
>+    0xda, 0x00, 0x0c, 0x03, 0x01, 0x00, 0x02, 0x10, 0x03, 0x10, 0x00, 0x00,
>+    0x01, 0x43, 0x8d, 0x1f, 0xa2, 0xb3, 0xca, 0x1b, 0x57, 0x0f, 0xff, 0xc4,
>+    0x00, 0x1b, 0x10, 0x00, 0x03, 0x00, 0x02, 0x03, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x02, 0x11, 0x00, 0x03,
>+    0x10, 0x12, 0x13, 0xff, 0xda, 0x00, 0x08, 0x01, 0x01, 0x00, 0x01, 0x05,
>+    0x02, 0x3b, 0x80, 0x6f, 0x56, 0x76, 0x56, 0x23, 0x87, 0x99, 0x0d, 0x26,
>+    0x62, 0xf6, 0xbf, 0xff, 0xc4, 0x00, 0x19, 0x11, 0x00, 0x02, 0x03, 0x01,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x01, 0x02, 0x10, 0x11, 0x21, 0xff, 0xda, 0x00, 0x08, 0x01, 0x03,
>+    0x01, 0x01, 0x3f, 0x01, 0x51, 0xce, 0x8c, 0x75, 0xff, 0xc4, 0x00, 0x18,
>+    0x11, 0x00, 0x03, 0x01, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x02, 0x61, 0x21, 0xff, 0xda,
>+    0x00, 0x08, 0x01, 0x02, 0x01, 0x01, 0x3f, 0x01, 0xa6, 0xd9, 0x2f, 0x84,
>+    0xe8, 0xf0, 0xff, 0xc4, 0x00, 0x1e, 0x10, 0x00, 0x02, 0x01, 0x03, 0x05,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x11, 0x21, 0x02, 0x12, 0x32, 0x10, 0x31, 0x71, 0x81, 0xa1, 0xff, 0xda,
>+    0x00, 0x08, 0x01, 0x01, 0x00, 0x06, 0x3f, 0x02, 0x4b, 0xb3, 0x28, 0x32,
>+    0xd2, 0xed, 0xf9, 0x1d, 0x3e, 0x13, 0x51, 0x73, 0x83, 0xff, 0xc4, 0x00,
>+    0x1c, 0x10, 0x01, 0x01, 0x01, 0x00, 0x02, 0x03, 0x01, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x11, 0x00, 0x21, 0x51, 0x31,
>+    0x61, 0x81, 0xf0, 0xff, 0xda, 0x00, 0x08, 0x01, 0x01, 0x00, 0x01, 0x3f,
>+    0x21, 0x75, 0x6e, 0x31, 0x94, 0x28, 0xf9, 0x30, 0xdc, 0x27, 0xdb, 0xa9,
>+    0x01, 0xf3, 0xde, 0x02, 0xa0, 0xed, 0x1e, 0x34, 0x68, 0x23, 0xf9, 0xc6,
>+    0x48, 0x5d, 0x7a, 0x35, 0x02, 0xf5, 0x6f, 0xff, 0xda, 0x00, 0x0c, 0x03,
>+    0x01, 0x00, 0x02, 0x00, 0x03, 0x00, 0x00, 0x00, 0x10, 0x2e, 0x45, 0xff,
>+    0xc4, 0x00, 0x18, 0x11, 0x00, 0x03, 0x01, 0x01, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x11, 0x21,
>+    0x31, 0xff, 0xda, 0x00, 0x08, 0x01, 0x03, 0x01, 0x01, 0x3f, 0x10, 0x53,
>+    0x50, 0xba, 0x54, 0xc1, 0x67, 0x4f, 0xff, 0xc4, 0x00, 0x18, 0x11, 0x00,
>+    0x03, 0x01, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x01, 0x11, 0x21, 0x00, 0x10, 0xff, 0xda, 0x00, 0x08,
>+    0x01, 0x02, 0x01, 0x01, 0x3f, 0x10, 0x18, 0x81, 0x5c, 0x04, 0x1a, 0xca,
>+    0x91, 0xbf, 0xff, 0xc4, 0x00, 0x1f, 0x10, 0x01, 0x00, 0x02, 0x01, 0x04,
>+    0x03, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01,
>+    0x00, 0x11, 0x31, 0x41, 0x61, 0x71, 0x91, 0x21, 0x81, 0xd1, 0xb1, 0xff,
>+    0xda, 0x00, 0x08, 0x01, 0x01, 0x00, 0x01, 0x3f, 0x10, 0x1b, 0x30, 0xe9,
>+    0x58, 0xbe, 0x1a, 0xfd, 0x8a, 0xeb, 0x8b, 0x34, 0x74, 0x80, 0x4b, 0xb5,
>+    0xd5, 0xab, 0xcd, 0x46, 0x96, 0x2e, 0xec, 0xbd, 0xaa, 0x78, 0x47, 0x5c,
>+    0x47, 0xa7, 0x30, 0x49, 0xad, 0x88, 0x7c, 0x40, 0x74, 0x30, 0xff, 0x00,
>+    0x23, 0x1d, 0x03, 0x0b, 0xb7, 0xd4, 0xff, 0xd9};
>+static const size_t kTest3JpgLen = 704;
>+
>+// test 4 is J422 vertical - not supported
>+static const uint8_t kTest4Jpg[] = {
>+    0xff, 0xd8, 0xff, 0xe0, 0x00, 0x10, 0x4a, 0x46, 0x49, 0x46, 0x00, 0x01,
>+    0x01, 0x01, 0x00, 0x48, 0x00, 0x48, 0x00, 0x00, 0xff, 0xdb, 0x00, 0x43,
>+    0x00, 0x10, 0x0b, 0x0c, 0x0e, 0x0c, 0x0a, 0x10, 0x0e, 0x0d, 0x0e, 0x12,
>+    0x11, 0x10, 0x13, 0x18, 0x28, 0x1a, 0x18, 0x16, 0x16, 0x18, 0x31, 0x23,
>+    0x25, 0x1d, 0x28, 0x3a, 0x33, 0x3d, 0x3c, 0x39, 0x33, 0x38, 0x37, 0x40,
>+    0x48, 0x5c, 0x4e, 0x40, 0x44, 0x57, 0x45, 0x37, 0x38, 0x50, 0x6d, 0x51,
>+    0x57, 0x5f, 0x62, 0x67, 0x68, 0x67, 0x3e, 0x4d, 0x71, 0x79, 0x70, 0x64,
>+    0x78, 0x5c, 0x65, 0x67, 0x63, 0xff, 0xdb, 0x00, 0x43, 0x01, 0x11, 0x12,
>+    0x12, 0x18, 0x15, 0x18, 0x2f, 0x1a, 0x1a, 0x2f, 0x63, 0x42, 0x38, 0x42,
>+    0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63,
>+    0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63,
>+    0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63,
>+    0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63, 0x63,
>+    0x63, 0x63, 0xff, 0xc2, 0x00, 0x11, 0x08, 0x00, 0x10, 0x00, 0x20, 0x03,
>+    0x01, 0x12, 0x00, 0x02, 0x11, 0x01, 0x03, 0x11, 0x01, 0xff, 0xc4, 0x00,
>+    0x18, 0x00, 0x00, 0x02, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x04, 0x05, 0x01, 0x02, 0x03, 0xff,
>+    0xc4, 0x00, 0x16, 0x01, 0x01, 0x01, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x03, 0xff,
>+    0xda, 0x00, 0x0c, 0x03, 0x01, 0x00, 0x02, 0x10, 0x03, 0x10, 0x00, 0x00,
>+    0x01, 0xd2, 0x98, 0xe9, 0x03, 0x0c, 0x00, 0x46, 0x21, 0xd9, 0xff, 0xc4,
>+    0x00, 0x1b, 0x10, 0x00, 0x03, 0x00, 0x02, 0x03, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x02, 0x11, 0x00, 0x03,
>+    0x10, 0x12, 0x13, 0xff, 0xda, 0x00, 0x08, 0x01, 0x01, 0x00, 0x01, 0x05,
>+    0x02, 0x3b, 0x80, 0x6f, 0x56, 0x76, 0x56, 0x23, 0x87, 0x99, 0x0d, 0x26,
>+    0x62, 0xf6, 0xbf, 0xff, 0xc4, 0x00, 0x17, 0x11, 0x01, 0x01, 0x01, 0x01,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x11, 0x01, 0x21, 0xff, 0xda, 0x00, 0x08, 0x01, 0x03, 0x01, 0x01,
>+    0x3f, 0x01, 0x98, 0xb1, 0xbd, 0x47, 0xff, 0xc4, 0x00, 0x18, 0x11, 0x00,
>+    0x03, 0x01, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x01, 0x12, 0x11, 0x21, 0xff, 0xda, 0x00, 0x08,
>+    0x01, 0x02, 0x01, 0x01, 0x3f, 0x01, 0xb6, 0x35, 0xa2, 0xe1, 0x47, 0xff,
>+    0xc4, 0x00, 0x1e, 0x10, 0x00, 0x02, 0x01, 0x03, 0x05, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x11, 0x21, 0x02,
>+    0x12, 0x32, 0x10, 0x31, 0x71, 0x81, 0xa1, 0xff, 0xda, 0x00, 0x08, 0x01,
>+    0x01, 0x00, 0x06, 0x3f, 0x02, 0x4b, 0xb3, 0x28, 0x32, 0xd2, 0xed, 0xf9,
>+    0x1d, 0x3e, 0x13, 0x51, 0x73, 0x83, 0xff, 0xc4, 0x00, 0x1c, 0x10, 0x01,
>+    0x01, 0x01, 0x00, 0x02, 0x03, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x01, 0x11, 0x00, 0x21, 0x51, 0x31, 0x61, 0x81, 0xf0,
>+    0xff, 0xda, 0x00, 0x08, 0x01, 0x01, 0x00, 0x01, 0x3f, 0x21, 0x75, 0x6e,
>+    0x31, 0x94, 0x28, 0xf9, 0x30, 0xdc, 0x27, 0xdb, 0xa9, 0x01, 0xf3, 0xde,
>+    0x02, 0xa0, 0xed, 0x1e, 0x34, 0x68, 0x23, 0xf9, 0xc6, 0x48, 0x5d, 0x7a,
>+    0x35, 0x02, 0xf5, 0x6f, 0xff, 0xda, 0x00, 0x0c, 0x03, 0x01, 0x00, 0x02,
>+    0x00, 0x03, 0x00, 0x00, 0x00, 0x10, 0x24, 0xaf, 0xff, 0xc4, 0x00, 0x19,
>+    0x11, 0x00, 0x03, 0x01, 0x01, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x11, 0x51, 0x21, 0x31, 0xff,
>+    0xda, 0x00, 0x08, 0x01, 0x03, 0x01, 0x01, 0x3f, 0x10, 0x59, 0x11, 0xca,
>+    0x42, 0x60, 0x9f, 0x69, 0xff, 0xc4, 0x00, 0x19, 0x11, 0x00, 0x02, 0x03,
>+    0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>+    0x00, 0x00, 0x01, 0x11, 0x21, 0x31, 0x61, 0xff, 0xda, 0x00, 0x08, 0x01,
>+    0x02, 0x01, 0x01, 0x3f, 0x10, 0xb0, 0xd7, 0x27, 0x51, 0xb6, 0x41, 0xff,
>+    0xc4, 0x00, 0x1f, 0x10, 0x01, 0x00, 0x02, 0x01, 0x04, 0x03, 0x01, 0x00,
>+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x11, 0x31,
>+    0x41, 0x61, 0x71, 0x91, 0x21, 0x81, 0xd1, 0xb1, 0xff, 0xda, 0x00, 0x08,
>+    0x01, 0x01, 0x00, 0x01, 0x3f, 0x10, 0x1b, 0x30, 0xe9, 0x58, 0xbe, 0x1a,
>+    0xfd, 0x8a, 0xeb, 0x8b, 0x34, 0x74, 0x80, 0x4b, 0xb5, 0xd5, 0xab, 0xcd,
>+    0x46, 0x96, 0x2e, 0xec, 0xbd, 0xaa, 0x78, 0x47, 0x5c, 0x47, 0xa7, 0x30,
>+    0x49, 0xad, 0x88, 0x7c, 0x40, 0x74, 0x30, 0xff, 0x00, 0x23, 0x1d, 0x03,
>+    0x0b, 0xb7, 0xd4, 0xff, 0xd9};
>+static const size_t kTest4JpgLen = 701;
>+
>+TEST_F(LibYUVConvertTest, TestMJPGSize) {
>+  int width = 0;
>+  int height = 0;
>+  int ret = MJPGSize(kTest2Jpg, kTest2JpgLen, &width, &height);
>+  EXPECT_EQ(0, ret);
>+
>+  printf("test jpeg size %d x %d\n", width, height);
>+}
> 
>-  // EOI, SOI to make MJPG appear valid.
>-  memset(orig_pixels, 0, kSize);
>-  orig_pixels[0] = 0xff;
>-  orig_pixels[1] = 0xd8;  // SOI.
>-  orig_pixels[kSize - kOff + 0] = 0xff;
>-  orig_pixels[kSize - kOff + 1] = 0xd9;  // EOI.
>+TEST_F(LibYUVConvertTest, TestMJPGToI420) {
>+  int width = 0;
>+  int height = 0;
>+  int ret = MJPGSize(kTest2Jpg, kTest2JpgLen, &width, &height);
>+  EXPECT_EQ(0, ret);
>+
>+  int half_width = (width + 1) / 2;
>+  int half_height = (height + 1) / 2;
>+  int benchmark_iterations = benchmark_iterations_ * benchmark_width_ *
>+                             benchmark_height_ / (width * height);
>+
>+  align_buffer_page_end(dst_y, width * height);
>+  align_buffer_page_end(dst_u, half_width * half_height);
>+  align_buffer_page_end(dst_v, half_width * half_height);
>+  for (int times = 0; times < benchmark_iterations; ++times) {
>+    ret = MJPGToI420(kTest2Jpg, kTest2JpgLen, dst_y, width, dst_u, half_width,
>+                     dst_v, half_width, width, height, width, height);
>+  }
>+  // Expect sucesss
>+  EXPECT_EQ(0, ret);
>+
>+  // Test result matches known hash value.
>+  uint32_t dst_y_hash = HashDjb2(dst_y, width * height, 5381);
>+  uint32_t dst_u_hash = HashDjb2(dst_u, half_width * half_height, 5381);
>+  uint32_t dst_v_hash = HashDjb2(dst_v, half_width * half_height, 5381);
>+  EXPECT_EQ(dst_y_hash, 2682851208u);
>+  EXPECT_EQ(dst_u_hash, 2501859930u);
>+  EXPECT_EQ(dst_v_hash, 2126459123u);
> 
>-  for (int times = 0; times < benchmark_iterations_; ++times) {
>-    int ret =
>-        MJPGToI420(orig_pixels, kSize, dst_y_opt, benchmark_width_, dst_u_opt,
>-                   SUBSAMPLE(benchmark_width_, 2), dst_v_opt,
>-                   SUBSAMPLE(benchmark_width_, 2), benchmark_width_,
>-                   benchmark_height_, benchmark_width_, benchmark_height_);
>-    // Expect failure because image is not really valid.
>-    EXPECT_EQ(1, ret);
>-  }
>-
>-  free_aligned_buffer_page_end(dst_y_opt);
>-  free_aligned_buffer_page_end(dst_u_opt);
>-  free_aligned_buffer_page_end(dst_v_opt);
>-  free_aligned_buffer_page_end(orig_pixels);
>+  free_aligned_buffer_page_end(dst_y);
>+  free_aligned_buffer_page_end(dst_u);
>+  free_aligned_buffer_page_end(dst_v);
> }
> 
>-TEST_F(LibYUVConvertTest, MJPGToARGB) {
>-  const int kOff = 10;
>-  const int kMinJpeg = 64;
>-  const int kImageSize = benchmark_width_ * benchmark_height_ >= kMinJpeg
>-                             ? benchmark_width_ * benchmark_height_
>-                             : kMinJpeg;
>-  const int kSize = kImageSize + kOff;
>-  align_buffer_page_end(orig_pixels, kSize);
>-  align_buffer_page_end(dst_argb_opt, benchmark_width_ * benchmark_height_ * 4);
>+TEST_F(LibYUVConvertTest, TestMJPGToI420_NV21) {
>+  int width = 0;
>+  int height = 0;
>+  int ret = MJPGSize(kTest2Jpg, kTest2JpgLen, &width, &height);
>+  EXPECT_EQ(0, ret);
> 
>-  // EOI, SOI to make MJPG appear valid.
>-  memset(orig_pixels, 0, kSize);
>-  orig_pixels[0] = 0xff;
>-  orig_pixels[1] = 0xd8;  // SOI.
>-  orig_pixels[kSize - kOff + 0] = 0xff;
>-  orig_pixels[kSize - kOff + 1] = 0xd9;  // EOI.
>+  int half_width = (width + 1) / 2;
>+  int half_height = (height + 1) / 2;
>+  int benchmark_iterations = benchmark_iterations_ * benchmark_width_ *
>+                             benchmark_height_ / (width * height);
> 
>-  for (int times = 0; times < benchmark_iterations_; ++times) {
>-    int ret = MJPGToARGB(orig_pixels, kSize, dst_argb_opt, benchmark_width_ * 4,
>-                         benchmark_width_, benchmark_height_, benchmark_width_,
>-                         benchmark_height_);
>-    // Expect failure because image is not really valid.
>-    EXPECT_EQ(1, ret);
>+  // Convert to NV21
>+  align_buffer_page_end(dst_y, width * height);
>+  align_buffer_page_end(dst_vu, half_width * half_height * 2);
>+
>+  for (int times = 0; times < benchmark_iterations; ++times) {
>+    ret = MJPGToNV21(kTest2Jpg, kTest2JpgLen, dst_y, width, dst_vu,
>+                     half_width * 2, width, height, width, height);
>   }
>+  // Expect sucesss
>+  EXPECT_EQ(0, ret);
> 
>-  free_aligned_buffer_page_end(dst_argb_opt);
>-  free_aligned_buffer_page_end(orig_pixels);
>+  // Convert to I420
>+  align_buffer_page_end(dst2_y, width * height);
>+  align_buffer_page_end(dst2_u, half_width * half_height);
>+  align_buffer_page_end(dst2_v, half_width * half_height);
>+  for (int times = 0; times < benchmark_iterations; ++times) {
>+    ret = MJPGToI420(kTest2Jpg, kTest2JpgLen, dst2_y, width, dst2_u, half_width,
>+                     dst2_v, half_width, width, height, width, height);
>+  }
>+  // Expect sucesss
>+  EXPECT_EQ(0, ret);
>+
>+  // Convert I420 to NV21
>+  align_buffer_page_end(dst3_y, width * height);
>+  align_buffer_page_end(dst3_vu, half_width * half_height * 2);
>+
>+  I420ToNV21(dst2_y, width, dst2_u, half_width, dst2_v, half_width, dst3_y,
>+             width, dst3_vu, half_width * 2, width, height);
>+
>+  for (int i = 0; i < width * height; ++i) {
>+    EXPECT_EQ(dst_y[i], dst3_y[i]);
>+  }
>+  for (int i = 0; i < half_width * half_height * 2; ++i) {
>+    EXPECT_EQ(dst_vu[i], dst3_vu[i]);
>+    EXPECT_EQ(dst_vu[i], dst3_vu[i]);
>+  }
>+
>+  free_aligned_buffer_page_end(dst3_y);
>+  free_aligned_buffer_page_end(dst3_vu);
>+
>+  free_aligned_buffer_page_end(dst2_y);
>+  free_aligned_buffer_page_end(dst2_u);
>+  free_aligned_buffer_page_end(dst2_v);
>+
>+  free_aligned_buffer_page_end(dst_y);
>+  free_aligned_buffer_page_end(dst_vu);
> }
> 
>+TEST_F(LibYUVConvertTest, TestMJPGToNV21_420) {
>+  int width = 0;
>+  int height = 0;
>+  int ret = MJPGSize(kTest2Jpg, kTest2JpgLen, &width, &height);
>+  EXPECT_EQ(0, ret);
>+
>+  int half_width = (width + 1) / 2;
>+  int half_height = (height + 1) / 2;
>+  int benchmark_iterations = benchmark_iterations_ * benchmark_width_ *
>+                             benchmark_height_ / (width * height);
>+
>+  align_buffer_page_end(dst_y, width * height);
>+  align_buffer_page_end(dst_uv, half_width * half_height * 2);
>+  for (int times = 0; times < benchmark_iterations; ++times) {
>+    ret = MJPGToNV21(kTest2Jpg, kTest2JpgLen, dst_y, width, dst_uv,
>+                     half_width * 2, width, height, width, height);
>+  }
>+  // Expect sucesss
>+  EXPECT_EQ(0, ret);
>+
>+  // Test result matches known hash value.
>+  uint32_t dst_y_hash = HashDjb2(dst_y, width * height, 5381);
>+  uint32_t dst_uv_hash = HashDjb2(dst_uv, half_width * half_height * 2, 5381);
>+  EXPECT_EQ(dst_y_hash, 2682851208u);
>+  EXPECT_EQ(dst_uv_hash, 1069662856u);
>+
>+  free_aligned_buffer_page_end(dst_y);
>+  free_aligned_buffer_page_end(dst_uv);
>+}
>+
>+TEST_F(LibYUVConvertTest, TestMJPGToNV21_422) {
>+  int width = 0;
>+  int height = 0;
>+  int ret = MJPGSize(kTest3Jpg, kTest3JpgLen, &width, &height);
>+  EXPECT_EQ(0, ret);
>+
>+  int half_width = (width + 1) / 2;
>+  int half_height = (height + 1) / 2;
>+  int benchmark_iterations = benchmark_iterations_ * benchmark_width_ *
>+                             benchmark_height_ / (width * height);
>+
>+  align_buffer_page_end(dst_y, width * height);
>+  align_buffer_page_end(dst_uv, half_width * half_height * 2);
>+  for (int times = 0; times < benchmark_iterations; ++times) {
>+    ret = MJPGToNV21(kTest3Jpg, kTest3JpgLen, dst_y, width, dst_uv,
>+                     half_width * 2, width, height, width, height);
>+  }
>+  // Expect sucesss
>+  EXPECT_EQ(0, ret);
>+
>+  // Test result matches known hash value.
>+  uint32_t dst_y_hash = HashDjb2(dst_y, width * height, 5381);
>+  uint32_t dst_uv_hash = HashDjb2(dst_uv, half_width * half_height * 2, 5381);
>+  EXPECT_EQ(dst_y_hash, 2682851208u);
>+  EXPECT_EQ(dst_uv_hash, 3543430771u);
>+
>+  free_aligned_buffer_page_end(dst_y);
>+  free_aligned_buffer_page_end(dst_uv);
>+}
>+
>+TEST_F(LibYUVConvertTest, TestMJPGToNV21_400) {
>+  int width = 0;
>+  int height = 0;
>+  int ret = MJPGSize(kTest0Jpg, kTest0JpgLen, &width, &height);
>+  EXPECT_EQ(0, ret);
>+
>+  int half_width = (width + 1) / 2;
>+  int half_height = (height + 1) / 2;
>+  int benchmark_iterations = benchmark_iterations_ * benchmark_width_ *
>+                             benchmark_height_ / (width * height);
>+
>+  align_buffer_page_end(dst_y, width * height);
>+  align_buffer_page_end(dst_uv, half_width * half_height * 2);
>+  for (int times = 0; times < benchmark_iterations; ++times) {
>+    ret = MJPGToNV21(kTest0Jpg, kTest0JpgLen, dst_y, width, dst_uv,
>+                     half_width * 2, width, height, width, height);
>+  }
>+  // Expect sucesss
>+  EXPECT_EQ(0, ret);
>+
>+  // Test result matches known hash value.
>+  uint32_t dst_y_hash = HashDjb2(dst_y, width * height, 5381);
>+  uint32_t dst_uv_hash = HashDjb2(dst_uv, half_width * half_height * 2, 5381);
>+  EXPECT_EQ(dst_y_hash, 330644005u);
>+  EXPECT_EQ(dst_uv_hash, 135214341u);
>+
>+  free_aligned_buffer_page_end(dst_y);
>+  free_aligned_buffer_page_end(dst_uv);
>+}
>+
>+TEST_F(LibYUVConvertTest, TestMJPGToNV21_444) {
>+  int width = 0;
>+  int height = 0;
>+  int ret = MJPGSize(kTest1Jpg, kTest1JpgLen, &width, &height);
>+  EXPECT_EQ(0, ret);
>+
>+  int half_width = (width + 1) / 2;
>+  int half_height = (height + 1) / 2;
>+  int benchmark_iterations = benchmark_iterations_ * benchmark_width_ *
>+                             benchmark_height_ / (width * height);
>+
>+  align_buffer_page_end(dst_y, width * height);
>+  align_buffer_page_end(dst_uv, half_width * half_height * 2);
>+  for (int times = 0; times < benchmark_iterations; ++times) {
>+    ret = MJPGToNV21(kTest1Jpg, kTest1JpgLen, dst_y, width, dst_uv,
>+                     half_width * 2, width, height, width, height);
>+  }
>+  // Expect sucesss
>+  EXPECT_EQ(0, ret);
>+
>+  // Test result matches known hash value.
>+  uint32_t dst_y_hash = HashDjb2(dst_y, width * height, 5381);
>+  uint32_t dst_uv_hash = HashDjb2(dst_uv, half_width * half_height * 2, 5381);
>+  EXPECT_EQ(dst_y_hash, 2682851208u);
>+  EXPECT_EQ(dst_uv_hash, 506143297u);
>+
>+  free_aligned_buffer_page_end(dst_y);
>+  free_aligned_buffer_page_end(dst_uv);
>+}
>+
>+TEST_F(LibYUVConvertTest, TestMJPGToARGB) {
>+  int width = 0;
>+  int height = 0;
>+  int ret = MJPGSize(kTest3Jpg, kTest3JpgLen, &width, &height);
>+  EXPECT_EQ(0, ret);
>+
>+  int benchmark_iterations = benchmark_iterations_ * benchmark_width_ *
>+                             benchmark_height_ / (width * height);
>+
>+  align_buffer_page_end(dst_argb, width * height * 4);
>+  for (int times = 0; times < benchmark_iterations; ++times) {
>+    ret = MJPGToARGB(kTest3Jpg, kTest3JpgLen, dst_argb, width * 4, width,
>+                     height, width, height);
>+  }
>+  // Expect sucesss
>+  EXPECT_EQ(0, ret);
>+
>+  // Test result matches known hash value.
>+  uint32_t dst_argb_hash = HashDjb2(dst_argb, width * height, 5381);
>+  EXPECT_EQ(dst_argb_hash, 2355976473u);
>+
>+  free_aligned_buffer_page_end(dst_argb);
>+}
>+
>+static int ShowJPegInfo(const uint8_t* sample, size_t sample_size) {
>+  MJpegDecoder mjpeg_decoder;
>+  LIBYUV_BOOL ret = mjpeg_decoder.LoadFrame(sample, sample_size);
>+
>+  int width = mjpeg_decoder.GetWidth();
>+  int height = mjpeg_decoder.GetHeight();
>+
>+  // YUV420
>+  if (mjpeg_decoder.GetColorSpace() == MJpegDecoder::kColorSpaceYCbCr &&
>+      mjpeg_decoder.GetNumComponents() == 3 &&
>+      mjpeg_decoder.GetVertSampFactor(0) == 2 &&
>+      mjpeg_decoder.GetHorizSampFactor(0) == 2 &&
>+      mjpeg_decoder.GetVertSampFactor(1) == 1 &&
>+      mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
>+      mjpeg_decoder.GetVertSampFactor(2) == 1 &&
>+      mjpeg_decoder.GetHorizSampFactor(2) == 1) {
>+    printf("JPeg is J420, %dx%d %d bytes\n", width, height,
>+           static_cast<int>(sample_size));
>+    // YUV422
>+  } else if (mjpeg_decoder.GetColorSpace() == MJpegDecoder::kColorSpaceYCbCr &&
>+             mjpeg_decoder.GetNumComponents() == 3 &&
>+             mjpeg_decoder.GetVertSampFactor(0) == 1 &&
>+             mjpeg_decoder.GetHorizSampFactor(0) == 2 &&
>+             mjpeg_decoder.GetVertSampFactor(1) == 1 &&
>+             mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
>+             mjpeg_decoder.GetVertSampFactor(2) == 1 &&
>+             mjpeg_decoder.GetHorizSampFactor(2) == 1) {
>+    printf("JPeg is J422, %dx%d %d bytes\n", width, height,
>+           static_cast<int>(sample_size));
>+    // YUV444
>+  } else if (mjpeg_decoder.GetColorSpace() == MJpegDecoder::kColorSpaceYCbCr &&
>+             mjpeg_decoder.GetNumComponents() == 3 &&
>+             mjpeg_decoder.GetVertSampFactor(0) == 1 &&
>+             mjpeg_decoder.GetHorizSampFactor(0) == 1 &&
>+             mjpeg_decoder.GetVertSampFactor(1) == 1 &&
>+             mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
>+             mjpeg_decoder.GetVertSampFactor(2) == 1 &&
>+             mjpeg_decoder.GetHorizSampFactor(2) == 1) {
>+    printf("JPeg is J444, %dx%d %d bytes\n", width, height,
>+           static_cast<int>(sample_size));
>+    // YUV400
>+  } else if (mjpeg_decoder.GetColorSpace() ==
>+                 MJpegDecoder::kColorSpaceGrayscale &&
>+             mjpeg_decoder.GetNumComponents() == 1 &&
>+             mjpeg_decoder.GetVertSampFactor(0) == 1 &&
>+             mjpeg_decoder.GetHorizSampFactor(0) == 1) {
>+    printf("JPeg is J400, %dx%d %d bytes\n", width, height,
>+           static_cast<int>(sample_size));
>+  } else {
>+    // Unknown colorspace.
>+    printf("JPeg is Unknown colorspace.\n");
>+  }
>+  mjpeg_decoder.UnloadFrame();
>+  return ret;
>+}
>+
>+TEST_F(LibYUVConvertTest, TestMJPGInfo) {
>+  EXPECT_EQ(1, ShowJPegInfo(kTest0Jpg, kTest0JpgLen));
>+  EXPECT_EQ(1, ShowJPegInfo(kTest1Jpg, kTest1JpgLen));
>+  EXPECT_EQ(1, ShowJPegInfo(kTest2Jpg, kTest2JpgLen));
>+  EXPECT_EQ(1, ShowJPegInfo(kTest3Jpg, kTest3JpgLen));
>+  EXPECT_EQ(1,
>+            ShowJPegInfo(kTest4Jpg, kTest4JpgLen));  // Valid but unsupported.
>+}
> #endif  // HAVE_JPEG
> 
> TEST_F(LibYUVConvertTest, NV12Crop) {
>@@ -1509,6 +2069,78 @@ TEST_F(LibYUVConvertTest, NV12Crop) {
>   free_aligned_buffer_page_end(src_y);
> }
> 
>+TEST_F(LibYUVConvertTest, I420CropOddY) {
>+  const int SUBSAMP_X = 2;
>+  const int SUBSAMP_Y = 2;
>+  const int kWidth = benchmark_width_;
>+  const int kHeight = benchmark_height_;
>+  const int crop_y = 1;
>+  const int kDestWidth = benchmark_width_;
>+  const int kDestHeight = benchmark_height_ - crop_y * 2;
>+  const int kStrideU = SUBSAMPLE(kWidth, SUBSAMP_X);
>+  const int kStrideV = SUBSAMPLE(kWidth, SUBSAMP_X);
>+  const int sample_size = kWidth * kHeight +
>+                          kStrideU * SUBSAMPLE(kHeight, SUBSAMP_Y) +
>+                          kStrideV * SUBSAMPLE(kHeight, SUBSAMP_Y);
>+  align_buffer_page_end(src_y, sample_size);
>+  uint8_t* src_u = src_y + kWidth * kHeight;
>+  uint8_t* src_v = src_u + kStrideU * SUBSAMPLE(kHeight, SUBSAMP_Y);
>+
>+  align_buffer_page_end(dst_y, kDestWidth * kDestHeight);
>+  align_buffer_page_end(dst_u, SUBSAMPLE(kDestWidth, SUBSAMP_X) *
>+                                   SUBSAMPLE(kDestHeight, SUBSAMP_Y));
>+  align_buffer_page_end(dst_v, SUBSAMPLE(kDestWidth, SUBSAMP_X) *
>+                                   SUBSAMPLE(kDestHeight, SUBSAMP_Y));
>+
>+  for (int i = 0; i < kHeight * kWidth; ++i) {
>+    src_y[i] = (fastrand() & 0xff);
>+  }
>+  for (int i = 0; i < SUBSAMPLE(kHeight, SUBSAMP_Y) * kStrideU; ++i) {
>+    src_u[i] = (fastrand() & 0xff);
>+  }
>+  for (int i = 0; i < SUBSAMPLE(kHeight, SUBSAMP_Y) * kStrideV; ++i) {
>+    src_v[i] = (fastrand() & 0xff);
>+  }
>+  memset(dst_y, 1, kDestWidth * kDestHeight);
>+  memset(dst_u, 2,
>+         SUBSAMPLE(kDestWidth, SUBSAMP_X) * SUBSAMPLE(kDestHeight, SUBSAMP_Y));
>+  memset(dst_v, 3,
>+         SUBSAMPLE(kDestWidth, SUBSAMP_X) * SUBSAMPLE(kDestHeight, SUBSAMP_Y));
>+
>+  MaskCpuFlags(benchmark_cpu_info_);
>+  for (int i = 0; i < benchmark_iterations_; ++i) {
>+    ConvertToI420(src_y, sample_size, dst_y, kDestWidth, dst_u,
>+                  SUBSAMPLE(kDestWidth, SUBSAMP_X), dst_v,
>+                  SUBSAMPLE(kDestWidth, SUBSAMP_X), 0, crop_y, kWidth, kHeight,
>+                  kDestWidth, kDestHeight, libyuv::kRotate0,
>+                  libyuv::FOURCC_I420);
>+  }
>+
>+  for (int i = 0; i < kDestHeight; ++i) {
>+    for (int j = 0; j < kDestWidth; ++j) {
>+      EXPECT_EQ(src_y[crop_y * kWidth + i * kWidth + j],
>+                dst_y[i * kDestWidth + j]);
>+    }
>+  }
>+  for (int i = 0; i < SUBSAMPLE(kDestHeight, SUBSAMP_Y); ++i) {
>+    for (int j = 0; j < SUBSAMPLE(kDestWidth, SUBSAMP_X); ++j) {
>+      EXPECT_EQ(src_u[(crop_y / 2 + i) * kStrideU + j],
>+                dst_u[i * SUBSAMPLE(kDestWidth, SUBSAMP_X) + j]);
>+    }
>+  }
>+  for (int i = 0; i < SUBSAMPLE(kDestHeight, SUBSAMP_Y); ++i) {
>+    for (int j = 0; j < SUBSAMPLE(kDestWidth, SUBSAMP_X); ++j) {
>+      EXPECT_EQ(src_v[(crop_y / 2 + i) * kStrideV + j],
>+                dst_v[i * SUBSAMPLE(kDestWidth, SUBSAMP_X) + j]);
>+    }
>+  }
>+
>+  free_aligned_buffer_page_end(dst_y);
>+  free_aligned_buffer_page_end(dst_u);
>+  free_aligned_buffer_page_end(dst_v);
>+  free_aligned_buffer_page_end(src_y);
>+}
>+
> TEST_F(LibYUVConvertTest, TestYToARGB) {
>   uint8_t y[32];
>   uint8_t expectedg[32];
>@@ -2458,4 +3090,28 @@ TEST_F(LibYUVConvertTest, TestH420ToAR30) {
>   free_aligned_buffer_page_end(ar30_pixels);
> }
> 
>+// Test RGB24 to ARGB and back to RGB24
>+TEST_F(LibYUVConvertTest, TestARGBToRGB24) {
>+  const int kSize = 256;
>+  align_buffer_page_end(orig_rgb24, kSize * 3);
>+  align_buffer_page_end(argb_pixels, kSize * 4);
>+  align_buffer_page_end(dest_rgb24, kSize * 3);
>+
>+  // Test grey scale
>+  for (int i = 0; i < kSize * 3; ++i) {
>+    orig_rgb24[i] = i;
>+  }
>+
>+  RGB24ToARGB(orig_rgb24, 0, argb_pixels, 0, kSize, 1);
>+  ARGBToRGB24(argb_pixels, 0, dest_rgb24, 0, kSize, 1);
>+
>+  for (int i = 0; i < kSize * 3; ++i) {
>+    EXPECT_EQ(orig_rgb24[i], dest_rgb24[i]);
>+  }
>+
>+  free_aligned_buffer_page_end(orig_rgb24);
>+  free_aligned_buffer_page_end(argb_pixels);
>+  free_aligned_buffer_page_end(dest_rgb24);
>+}
>+
> }  // namespace libyuv
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/cpu_test.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/cpu_test.cc
>index a8fb4b4ac01f16fc7525c39a20c66478ea0fccaa..a7991d2bab10102a6c986a2fdbafd97ab8ca1125 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/cpu_test.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/cpu_test.cc
>@@ -20,12 +20,12 @@ namespace libyuv {
> 
> TEST_F(LibYUVBaseTest, TestCpuHas) {
>   int cpu_flags = TestCpuFlag(-1);
>-  printf("Cpu Flags %x\n", cpu_flags);
>+  printf("Cpu Flags %d\n", cpu_flags);
> #if defined(__arm__) || defined(__aarch64__)
>   int has_arm = TestCpuFlag(kCpuHasARM);
>-  printf("Has ARM %x\n", has_arm);
>+  printf("Has ARM %d\n", has_arm);
>   int has_neon = TestCpuFlag(kCpuHasNEON);
>-  printf("Has NEON %x\n", has_neon);
>+  printf("Has NEON %d\n", has_neon);
> #endif
>   int has_x86 = TestCpuFlag(kCpuHasX86);
>   int has_sse2 = TestCpuFlag(kCpuHasSSE2);
>@@ -44,29 +44,31 @@ TEST_F(LibYUVBaseTest, TestCpuHas) {
>   int has_avx512vbmi2 = TestCpuFlag(kCpuHasAVX512VBMI2);
>   int has_avx512vbitalg = TestCpuFlag(kCpuHasAVX512VBITALG);
>   int has_avx512vpopcntdq = TestCpuFlag(kCpuHasAVX512VPOPCNTDQ);
>-  printf("Has X86 %x\n", has_x86);
>-  printf("Has SSE2 %x\n", has_sse2);
>-  printf("Has SSSE3 %x\n", has_ssse3);
>-  printf("Has SSE4.1 %x\n", has_sse41);
>-  printf("Has SSE4.2 %x\n", has_sse42);
>-  printf("Has AVX %x\n", has_avx);
>-  printf("Has AVX2 %x\n", has_avx2);
>-  printf("Has ERMS %x\n", has_erms);
>-  printf("Has FMA3 %x\n", has_fma3);
>-  printf("Has F16C %x\n", has_f16c);
>-  printf("Has GFNI %x\n", has_gfni);
>-  printf("Has AVX512BW %x\n", has_avx512bw);
>-  printf("Has AVX512VL %x\n", has_avx512vl);
>-  printf("Has AVX512VBMI %x\n", has_avx512vbmi);
>-  printf("Has AVX512VBMI2 %x\n", has_avx512vbmi2);
>-  printf("Has AVX512VBITALG %x\n", has_avx512vbitalg);
>-  printf("Has AVX512VPOPCNTDQ %x\n", has_avx512vpopcntdq);
>+  printf("Has X86 %d\n", has_x86);
>+  printf("Has SSE2 %d\n", has_sse2);
>+  printf("Has SSSE3 %d\n", has_ssse3);
>+  printf("Has SSE41 %d\n", has_sse41);
>+  printf("Has SSE42 %d\n", has_sse42);
>+  printf("Has AVX %d\n", has_avx);
>+  printf("Has AVX2 %d\n", has_avx2);
>+  printf("Has ERMS %d\n", has_erms);
>+  printf("Has FMA3 %d\n", has_fma3);
>+  printf("Has F16C %d\n", has_f16c);
>+  printf("Has GFNI %d\n", has_gfni);
>+  printf("Has AVX512BW %d\n", has_avx512bw);
>+  printf("Has AVX512VL %d\n", has_avx512vl);
>+  printf("Has AVX512VBMI %d\n", has_avx512vbmi);
>+  printf("Has AVX512VBMI2 %d\n", has_avx512vbmi2);
>+  printf("Has AVX512VBITALG %d\n", has_avx512vbitalg);
>+  printf("Has AVX512VPOPCNTDQ %d\n", has_avx512vpopcntdq);
> 
> #if defined(__mips__)
>   int has_mips = TestCpuFlag(kCpuHasMIPS);
>-  printf("Has MIPS %x\n", has_mips);
>+  printf("Has MIPS %d\n", has_mips);
>   int has_msa = TestCpuFlag(kCpuHasMSA);
>-  printf("Has MSA %x\n", has_msa);
>+  printf("Has MSA %d\n", has_msa);
>+  int has_mmi = TestCpuFlag(kCpuHasMMI);
>+  printf("Has MMI %d\n", has_mmi);
> #endif
> }
> 
>@@ -158,4 +160,27 @@ TEST_F(LibYUVBaseTest, TestLinuxNeon) {
> #endif
> }
> 
>+TEST_F(LibYUVBaseTest, TestSetCpuFlags) {
>+  // Reset any masked flags that may have been set so auto init is enabled.
>+  MaskCpuFlags(0);
>+
>+  int original_cpu_flags = TestCpuFlag(-1);
>+
>+  // Test setting different CPU configurations.
>+  int cpu_flags = kCpuHasARM | kCpuHasNEON | kCpuInitialized;
>+  SetCpuFlags(cpu_flags);
>+  EXPECT_EQ(cpu_flags, TestCpuFlag(-1));
>+
>+  cpu_flags = kCpuHasX86 | kCpuInitialized;
>+  SetCpuFlags(cpu_flags);
>+  EXPECT_EQ(cpu_flags, TestCpuFlag(-1));
>+
>+  // Test that setting 0 turns auto-init back on.
>+  SetCpuFlags(0);
>+  EXPECT_EQ(original_cpu_flags, TestCpuFlag(-1));
>+
>+  // Restore the CPU flag mask.
>+  MaskCpuFlags(benchmark_cpu_info_);
>+}
>+
> }  // namespace libyuv
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/scale_argb_test.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/scale_argb_test.cc
>index a1be85b8d8ab49807ec4e9270a7f2d5156671ac7..94aef60e22632fc3a42863e97d511ed9240d4149 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/scale_argb_test.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/scale_argb_test.cc
>@@ -303,10 +303,10 @@ TEST_FACTOR(3, 1, 3)
> 
> TEST_SCALETO(ARGBScale, 1, 1)
> TEST_SCALETO(ARGBScale, 320, 240)
>-TEST_SCALETO(ARGBScale, 352, 288)
> TEST_SCALETO(ARGBScale, 569, 480)
> TEST_SCALETO(ARGBScale, 640, 360)
> TEST_SCALETO(ARGBScale, 1280, 720)
>+TEST_SCALETO(ARGBScale, 1920, 1080)
> #undef TEST_SCALETO1
> #undef TEST_SCALETO
> 
>@@ -319,12 +319,12 @@ int YUVToARGBScaleReference2(const uint8_t* src_y,
>                              int src_stride_u,
>                              const uint8_t* src_v,
>                              int src_stride_v,
>-                             uint32 /* src_fourcc */,
>+                             uint32_t /* src_fourcc */,
>                              int src_width,
>                              int src_height,
>                              uint8_t* dst_argb,
>                              int dst_stride_argb,
>-                             uint32 /* dst_fourcc */,
>+                             uint32_t /* dst_fourcc */,
>                              int dst_width,
>                              int dst_height,
>                              int clip_x,
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/scale_test.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/scale_test.cc
>index 08b6cffaa26b7863371e691df89f8be135c7470b..d97d54a8830fb2620c6f19c55dcd16c364b58122 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/scale_test.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/scale_test.cc
>@@ -336,10 +336,10 @@ TEST_FACTOR(3, 1, 3, 0)
> 
> TEST_SCALETO(Scale, 1, 1)
> TEST_SCALETO(Scale, 320, 240)
>-TEST_SCALETO(Scale, 352, 288)
> TEST_SCALETO(Scale, 569, 480)
> TEST_SCALETO(Scale, 640, 360)
> TEST_SCALETO(Scale, 1280, 720)
>+TEST_SCALETO(Scale, 1920, 1080)
> #undef TEST_SCALETO1
> #undef TEST_SCALETO
> 
>@@ -437,6 +437,10 @@ extern "C" void ScaleRowUp2_16_NEON(const uint16_t* src_ptr,
>                                     ptrdiff_t src_stride,
>                                     uint16_t* dst,
>                                     int dst_width);
>+extern "C" void ScaleRowUp2_16_MMI(const uint16_t* src_ptr,
>+                                   ptrdiff_t src_stride,
>+                                   uint16_t* dst,
>+                                   int dst_width);
> extern "C" void ScaleRowUp2_16_C(const uint16_t* src_ptr,
>                                  ptrdiff_t src_stride,
>                                  uint16_t* dst,
>@@ -463,6 +467,13 @@ TEST_F(LibYUVScaleTest, TestScaleRowUp2_16) {
>     } else {
>       ScaleRowUp2_16_C(&orig_pixels[0], 640, &dst_pixels_opt[0], 1280);
>     }
>+#elif !defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)
>+    int has_mmi = TestCpuFlag(kCpuHasMMI);
>+    if (has_mmi) {
>+      ScaleRowUp2_16_MMI(&orig_pixels[0], 640, &dst_pixels_opt[0], 1280);
>+    } else {
>+      ScaleRowUp2_16_C(&orig_pixels[0], 640, &dst_pixels_opt[0], 1280);
>+    }
> #else
>     ScaleRowUp2_16_C(&orig_pixels[0], 640, &dst_pixels_opt[0], 1280);
> #endif
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/testdata/juno.txt b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/testdata/juno.txt
>index c275be7425f29568f0d18e7b956305d64527ae10..dd465272b8837bdc6588a533e66fbc9672d58de4 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/testdata/juno.txt
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/testdata/juno.txt
>@@ -1,15 +1,15 @@
>-Processor       : AArch64 Processor rev 0 (aarch64)
>-processor       : 0
>-processor       : 1
>-processor       : 2
>-processor       : 3
>-processor       : 4
>-processor       : 5
>-Features        : fp asimd evtstrm aes pmull sha1 sha2 crc32
>-CPU implementer : 0x41
>-CPU architecture: AArch64
>-CPU variant     : 0x0
>-CPU part        : 0xd07
>-CPU revision    : 0
>-
>-Hardware        : Juno
>+Processor       : AArch64 Processor rev 0 (aarch64)
>+processor       : 0
>+processor       : 1
>+processor       : 2
>+processor       : 3
>+processor       : 4
>+processor       : 5
>+Features        : fp asimd evtstrm aes pmull sha1 sha2 crc32
>+CPU implementer : 0x41
>+CPU architecture: AArch64
>+CPU variant     : 0x0
>+CPU part        : 0xd07
>+CPU revision    : 0
>+
>+Hardware        : Juno
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/testdata/test0.jpg b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/testdata/test0.jpg
>new file mode 100644
>index 0000000000000000000000000000000000000000..f4461a813f14ad79395cd844ae26fa8fddcb18a6
>GIT binary patch
>literal 421
>zcmex=<NpH&0WUXCHwH#VMg|WcWcYuZ!I?pTn}?5wOMs7;Pe@QeSVBWeLQG7;P+3)0
>z!^+s!#?sir+`%Kp&%q_!)!ZT=H!wUtDLo_I&bP3#Af+NEH9h(NAqH*^1_1^IMn*w~
>z|3?_aLFO|uAOU6;MyCI_7&yS9jLvn-r-uX8NeeJAGcYkDsbgdkWMCE$5{9Z}WM#5G
>zke^l-rrbW0S1sw=exN2f0R|>UW>zFk41$VGLPi3Hg^dfL`q=E5yf<qYU3&Xd)=oIE
>zxEZKTMgZg_CT4`QfNoU`G)!##0JF?qF*VPyP2;D*9rfEQ89(1+TJToRBt!Y<F^||P
>zQ>L%^a1RKW0u7WG02;^0f@~mzprK=8;Y7v8iyPr)3UC{|jMyjjw_|m;NlAnE)~l<}
>jx=qu2vv*a6dyM;X1JAV`H4Y^P{~45JnYp)L`F|4th$m{S
>
>literal 0
>HcmV?d00001
>
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/testdata/test1.jpg b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/testdata/test1.jpg
>new file mode 100644
>index 0000000000000000000000000000000000000000..a0210e9d4dbaa6ff6d83d4c353d56f32dcf777a5
>GIT binary patch
>literal 735
>zcmex=<NpH&0WUXCHwH#VMg|WcWcYuZ!I?pTn}?5wOMs7;Pe@QeSVBWeLQG7;P+3)0
>z!^+s!#?sir+`%Kp&%q_!)!ZT=H!wUtDLo_I&bP3#Af+NEH9Z+*2&15okc6m&zLb=H
>zvXg~VGRffoAqGJX1_1^IW=25<CP7AKLB{__7{ozNWn@4C%q)ydKnXE)aRw$v=Kr@C
>zc$gU(m;{&w7#J8G`qf^1IrB49TUzThP=~Yt12Y2?Gm=q^Oo9x|0z$$-Z5%*L7+IOD
>z8}h@-!j#)*@~S0$+Yi(vDagpcjN$-B1_41PA+Q#pggxU&26flnhU<Z9rGV~6Rm;H0
>zBp~PrQOpEV{LMSwV{>M*X%Wb5IRTKHS&<yXAgIVBWF%l%*tihtCN_H}@68%Um)`!A
>zwG$33ZU$<T5deh<&?dM>V1OwG8YVV=fO*7Tu{6(cipEcaJL<PrGJd|twBW6rNrv*z
>zV;-?prc7V+f$G5F1Pl-c1_8CiD-hQMjYJL}21Y?eL&re4R|P^=><?()Ft__imRh#f
>z9C@H72|-XaAY1|rb)X+0T0kBaD0pahv5)uR)~gfCoSR&Is_TK8<OP7vXJkS40E3{R
>zV`AY%#m0*p;eHU1Hh3AaPwH>i>u!^h2Jfv`SD$s8ruSy=stWfQ_vHqjYddNjN(}xp
>ND9bW)Z@=>YCIC;)v-|)6
>
>literal 0
>HcmV?d00001
>
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/testdata/test2.jpg b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/testdata/test2.jpg
>new file mode 100644
>index 0000000000000000000000000000000000000000..816ca767b1a1a6f5a6f8124bdaee488ba1caf511
>GIT binary patch
>literal 685
>zcmex=<NpH&0WUXCHwH#VMg|WcWcYuZ!I?pTn}?5wOMs7;Pe@QeSVBWeLQG7;P+3)0
>z!^+s!#?sir+`%Kp&%q_!)!ZT=H!wUtDLo_I&bP3#Af+NEH9Z+*2&15okc6m&zLb=H
>zvXg~VGRffoAqGJX1_1^IW=16jCP7AKLB{__7$g`Nn3x%m0W&Kj6AMsA4Cn_&WC;dl
>zMyCI_7<iZ&8JGl^1sE6@6`pG>=GN0a4^%BJz`%@Z5+jo!1G9jTFi;Bz&>%)uChLa$
>zu(B}a_L;nDN#FJZHHix{0(GGPMnOfe24+S^d&U#NAa@9&yMsZ{2&{$)q~_9bpc**=
>z1|~*kR%DX|6`6#L1Plus7eXDtX3ylkS;Oek+n=&_!hyxjKy5MtAfuUp9)bZzK?cP@
>z!^Fl9Fjw0vmgX5w(fDa_NB#Cn#?SYd7QB@+$x!}z%p<nSl<8|eP#rix!2S>rj)(XP
>z&BcriNWK!_TWD2=Lz$u=++zaa+#ChBf$HQ1fKFs&L3SjAprK=8;Y7v8iyPtY6p%J}
>z8L?05Z`bQ?ladDStyfo{b(^O5X78#B_Zau(2A*p>Y8*-o{xc}cGIMXg^8Y3P^0b`M
>
>literal 0
>HcmV?d00001
>
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/testdata/test3.jpg b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/testdata/test3.jpg
>new file mode 100644
>index 0000000000000000000000000000000000000000..792d91dc8db88b7c16ff8eb00e22b5118ec9f3d3
>GIT binary patch
>literal 704
>zcmex=<NpH&0WUXCHwH#VMg|WcWcYuZ!I?pTn}?5wOMs7;Pe@QeSVBWeLQG7;P+3)0
>z!^+s!#?sir+`%Kp&%q_!)!ZT=H!wUtDLo_I&bP3#Af+NEH9Z+*2&15okc6m&zLb=H
>zvXg~VGRffoAqGJX1_1^IW=2H@CP7AKLB{__7{ozNWn@4C%q)ydAPGhWW)yJ-MkZ#4
>z|F;-;fWk}y%mNGyjLyCCi#DH<4(A7|mlj}PW&o-{07fQ324(>vVW1ukpcRa)Ox6wg
>zVP#>;?K64RlD_Q+YLXNLYC*R}Ku{5^g_)7jo-y!TPbpBL1kfl{X9As*2vNrbQn&1;
>ze#?swK$UU=3`~s7tO#c?FbFC#2^k3(7B((~I)%-i$$PVg(WSRPW$lCmi<^PkWCTFL
>zfb1AXK?cP@!^Fl9F!$RlmgX5w(fDa_NB#Cn#?SYd7QB@+$x!}z%p<nSl<8|eP#rjI
>zfZ@QvAfV@p!`Xt0hH&o*1PAO2IhgK`RVkyOB7*=}DaeNc5{)q|Ql}<@f=ylk=u}1)
>zq>y7|5Hxg5ES#v=cyS}#y#mq(FC+Fz{q1_)ZBo+Uz4hwqvu@M$-t1jf;U43@+`w~f
>UM~y>?!G8v2S!V9-SN`7w0A}!{tN;K2
>
>literal 0
>HcmV?d00001
>
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/testdata/test4.jpg b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/testdata/test4.jpg
>new file mode 100644
>index 0000000000000000000000000000000000000000..1ef41668facb24eadd6a7d07c9f58f57a4f2bf61
>GIT binary patch
>literal 701
>zcmex=<NpH&0WUXCHwH#VMg|WcWcYuZ!I?pTn}?5wOMs7;Pe@QeSVBWeLQG7;P+3)0
>z!^+s!#?sir+`%Kp&%q_!)!ZT=H!wUtDLo_I&bP3#Af+NEH9Z+*2&15okc6m&zLb=H
>zvXg~VGRffoAqGJX1_1^IW=0_fCP7AKLB{__7$g`Nn3x%m0ShZ56Ejdo4Cn_&WC;ca
>zM&|#w7<iZ&8JGl^1%TX3GhQ<DFt{n+1S*#nU|>cyijhf>fmuLE7^sH>Xc8kUlXXLW
>zSXr2I`%GT7q;LCyn#2V$ED>Z>1Z!YsWVC0Tv2m|E$dN#U(A>f(BnVN*1X8!nbkReQ
>zLOB5jCPrpfWLFC+G6@+87#21zgt~*xp2>T&hS8<BKV|KN1B;u1+GGSkUS$G$3<ekl
>z859Ez6B|Fk9B;2!nrApg<EOzL_1h~MKi^|o@K(+wL;2@1kJu_xrmy)xb>M&j1_A?v
>zfXaHHsght<Bg_O!FbW1L8p6FN5Gi=dDPevlL@l!QAZIWNDjFt&)q=bzu;IFT;5J8~
>zQh5QOQyE#19m^nS=$KeIQL*vjM!0(gqzzt1?34Q2^}5@nq``aZ)zxR+rs=)eyQ;!H
>a#(lYg=h}`MhZ2MT49c?1+}p4GzX<>te530C
>
>literal 0
>HcmV?d00001
>
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/unit_test.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/unit_test.cc
>index 20aadb44e2faf809f938515b538f6868ed14a108..a1ae7ea3c7c27c9460f7f65c8d0b4f05b878b25a 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/unit_test.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/unit_test.cc
>@@ -38,6 +38,99 @@ static const int32_t FLAGS_libyuv_flags = 0;
> static const int32_t FLAGS_libyuv_cpu_info = 0;
> #endif
> 
>+// Test environment variable for disabling CPU features. Any non-zero value
>+// to disable. Zero ignored to make it easy to set the variable on/off.
>+#if !defined(__native_client__) && !defined(_M_ARM)
>+static LIBYUV_BOOL TestEnv(const char* name) {
>+  const char* var = getenv(name);
>+  if (var) {
>+    if (var[0] != '0') {
>+      return LIBYUV_TRUE;
>+    }
>+  }
>+  return LIBYUV_FALSE;
>+}
>+#else  // nacl does not support getenv().
>+static LIBYUV_BOOL TestEnv(const char*) {
>+  return LIBYUV_FALSE;
>+}
>+#endif
>+
>+int TestCpuEnv(int cpu_info) {
>+#if defined(__arm__) || defined(__aarch64__)
>+  if (TestEnv("LIBYUV_DISABLE_NEON")) {
>+    cpu_info &= ~libyuv::kCpuHasNEON;
>+  }
>+#endif
>+#if defined(__mips__) && defined(__linux__)
>+  if (TestEnv("LIBYUV_DISABLE_MSA")) {
>+    cpu_info &= ~libyuv::kCpuHasMSA;
>+  }
>+  if (TestEnv("LIBYUV_DISABLE_MMI")) {
>+    cpu_info &= ~libyuv::kCpuHasMMI;
>+  }
>+#endif
>+#if !defined(__pnacl__) && !defined(__CLR_VER) &&                   \
>+    (defined(__x86_64__) || defined(_M_X64) || defined(__i386__) || \
>+     defined(_M_IX86))
>+  if (TestEnv("LIBYUV_DISABLE_X86")) {
>+    cpu_info &= ~libyuv::kCpuHasX86;
>+  }
>+  if (TestEnv("LIBYUV_DISABLE_SSE2")) {
>+    cpu_info &= ~libyuv::kCpuHasSSE2;
>+  }
>+  if (TestEnv("LIBYUV_DISABLE_SSSE3")) {
>+    cpu_info &= ~libyuv::kCpuHasSSSE3;
>+  }
>+  if (TestEnv("LIBYUV_DISABLE_SSE41")) {
>+    cpu_info &= ~libyuv::kCpuHasSSE41;
>+  }
>+  if (TestEnv("LIBYUV_DISABLE_SSE42")) {
>+    cpu_info &= ~libyuv::kCpuHasSSE42;
>+  }
>+  if (TestEnv("LIBYUV_DISABLE_AVX")) {
>+    cpu_info &= ~libyuv::kCpuHasAVX;
>+  }
>+  if (TestEnv("LIBYUV_DISABLE_AVX2")) {
>+    cpu_info &= ~libyuv::kCpuHasAVX2;
>+  }
>+  if (TestEnv("LIBYUV_DISABLE_ERMS")) {
>+    cpu_info &= ~libyuv::kCpuHasERMS;
>+  }
>+  if (TestEnv("LIBYUV_DISABLE_FMA3")) {
>+    cpu_info &= ~libyuv::kCpuHasFMA3;
>+  }
>+  if (TestEnv("LIBYUV_DISABLE_F16C")) {
>+    cpu_info &= ~libyuv::kCpuHasF16C;
>+  }
>+  if (TestEnv("LIBYUV_DISABLE_AVX512BW")) {
>+    cpu_info &= ~libyuv::kCpuHasAVX512BW;
>+  }
>+  if (TestEnv("LIBYUV_DISABLE_AVX512VL")) {
>+    cpu_info &= ~libyuv::kCpuHasAVX512VL;
>+  }
>+  if (TestEnv("LIBYUV_DISABLE_AVX512VBMI")) {
>+    cpu_info &= ~libyuv::kCpuHasAVX512VBMI;
>+  }
>+  if (TestEnv("LIBYUV_DISABLE_AVX512VBMI2")) {
>+    cpu_info &= ~libyuv::kCpuHasAVX512VBMI2;
>+  }
>+  if (TestEnv("LIBYUV_DISABLE_AVX512VBITALG")) {
>+    cpu_info &= ~libyuv::kCpuHasAVX512VBITALG;
>+  }
>+  if (TestEnv("LIBYUV_DISABLE_AVX512VPOPCNTDQ")) {
>+    cpu_info &= ~libyuv::kCpuHasAVX512VPOPCNTDQ;
>+  }
>+  if (TestEnv("LIBYUV_DISABLE_GFNI")) {
>+    cpu_info &= ~libyuv::kCpuHasGFNI;
>+  }
>+#endif
>+  if (TestEnv("LIBYUV_DISABLE_ASM")) {
>+    cpu_info = libyuv::kCpuInitialized;
>+  }
>+  return cpu_info;
>+}
>+
> // For quicker unittests, default is 128 x 72.  But when benchmarking,
> // default to 720p.  Allow size to specify.
> // Set flags to -1 for benchmarking to avoid slower C code.
>@@ -87,6 +180,8 @@ LibYUVConvertTest::LibYUVConvertTest()
>   if (FLAGS_libyuv_cpu_info) {
>     benchmark_cpu_info_ = FLAGS_libyuv_cpu_info;
>   }
>+  disable_cpu_flags_ = TestCpuEnv(disable_cpu_flags_);
>+  benchmark_cpu_info_ = TestCpuEnv(benchmark_cpu_info_);
>   libyuv::MaskCpuFlags(benchmark_cpu_info_);
>   benchmark_pixels_div1280_ =
>       static_cast<int>((static_cast<double>(Abs(benchmark_width_)) *
>@@ -141,6 +236,8 @@ LibYUVColorTest::LibYUVColorTest()
>   if (FLAGS_libyuv_cpu_info) {
>     benchmark_cpu_info_ = FLAGS_libyuv_cpu_info;
>   }
>+  disable_cpu_flags_ = TestCpuEnv(disable_cpu_flags_);
>+  benchmark_cpu_info_ = TestCpuEnv(benchmark_cpu_info_);
>   libyuv::MaskCpuFlags(benchmark_cpu_info_);
>   benchmark_pixels_div1280_ =
>       static_cast<int>((static_cast<double>(Abs(benchmark_width_)) *
>@@ -195,6 +292,8 @@ LibYUVScaleTest::LibYUVScaleTest()
>   if (FLAGS_libyuv_cpu_info) {
>     benchmark_cpu_info_ = FLAGS_libyuv_cpu_info;
>   }
>+  disable_cpu_flags_ = TestCpuEnv(disable_cpu_flags_);
>+  benchmark_cpu_info_ = TestCpuEnv(benchmark_cpu_info_);
>   libyuv::MaskCpuFlags(benchmark_cpu_info_);
>   benchmark_pixels_div1280_ =
>       static_cast<int>((static_cast<double>(Abs(benchmark_width_)) *
>@@ -249,6 +348,8 @@ LibYUVRotateTest::LibYUVRotateTest()
>   if (FLAGS_libyuv_cpu_info) {
>     benchmark_cpu_info_ = FLAGS_libyuv_cpu_info;
>   }
>+  disable_cpu_flags_ = TestCpuEnv(disable_cpu_flags_);
>+  benchmark_cpu_info_ = TestCpuEnv(benchmark_cpu_info_);
>   libyuv::MaskCpuFlags(benchmark_cpu_info_);
>   benchmark_pixels_div1280_ =
>       static_cast<int>((static_cast<double>(Abs(benchmark_width_)) *
>@@ -303,6 +404,8 @@ LibYUVPlanarTest::LibYUVPlanarTest()
>   if (FLAGS_libyuv_cpu_info) {
>     benchmark_cpu_info_ = FLAGS_libyuv_cpu_info;
>   }
>+  disable_cpu_flags_ = TestCpuEnv(disable_cpu_flags_);
>+  benchmark_cpu_info_ = TestCpuEnv(benchmark_cpu_info_);
>   libyuv::MaskCpuFlags(benchmark_cpu_info_);
>   benchmark_pixels_div1280_ =
>       static_cast<int>((static_cast<double>(Abs(benchmark_width_)) *
>@@ -357,6 +460,8 @@ LibYUVBaseTest::LibYUVBaseTest()
>   if (FLAGS_libyuv_cpu_info) {
>     benchmark_cpu_info_ = FLAGS_libyuv_cpu_info;
>   }
>+  disable_cpu_flags_ = TestCpuEnv(disable_cpu_flags_);
>+  benchmark_cpu_info_ = TestCpuEnv(benchmark_cpu_info_);
>   libyuv::MaskCpuFlags(benchmark_cpu_info_);
>   benchmark_pixels_div1280_ =
>       static_cast<int>((static_cast<double>(Abs(benchmark_width_)) *
>@@ -411,6 +516,8 @@ LibYUVCompareTest::LibYUVCompareTest()
>   if (FLAGS_libyuv_cpu_info) {
>     benchmark_cpu_info_ = FLAGS_libyuv_cpu_info;
>   }
>+  disable_cpu_flags_ = TestCpuEnv(disable_cpu_flags_);
>+  benchmark_cpu_info_ = TestCpuEnv(benchmark_cpu_info_);
>   libyuv::MaskCpuFlags(benchmark_cpu_info_);
>   benchmark_pixels_div1280_ =
>       static_cast<int>((static_cast<double>(Abs(benchmark_width_)) *
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/video_common_test.cc b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/video_common_test.cc
>index 4d89586e76f225c04736d27964b49152130b104f..a84206a2adbf426f4d5da140a728821ed752a756 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/video_common_test.cc
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/unit_test/video_common_test.cc
>@@ -79,6 +79,7 @@ TEST_F(LibYUVBaseTest, TestFourCC) {
>   EXPECT_TRUE(TestValidFourCC(FOURCC_RGBO, FOURCC_BPP_RGBO));
>   EXPECT_TRUE(TestValidFourCC(FOURCC_R444, FOURCC_BPP_R444));
>   EXPECT_TRUE(TestValidFourCC(FOURCC_H420, FOURCC_BPP_H420));
>+  EXPECT_TRUE(TestValidFourCC(FOURCC_H422, FOURCC_BPP_H422));
>   EXPECT_TRUE(TestValidFourCC(FOURCC_H010, FOURCC_BPP_H010));
>   EXPECT_TRUE(TestValidFourCC(FOURCC_MJPG, FOURCC_BPP_MJPG));
>   EXPECT_TRUE(TestValidFourCC(FOURCC_YV12, FOURCC_BPP_YV12));
>diff --git a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/util/cpuid.c b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/util/cpuid.c
>index 59c65d60e0fe46d42c2fdb7554b1efc77f03315a..84c0602287bf5477dc53f7c4eb0a1e9d77ab1d74 100644
>--- a/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/util/cpuid.c
>+++ b/Source/ThirdParty/libwebrtc/Source/third_party/libyuv/util/cpuid.c
>@@ -71,6 +71,8 @@ int main(int argc, const char* argv[]) {
>   if (has_mips) {
>     int has_msa = TestCpuFlag(kCpuHasMSA);
>     printf("Has MSA %x\n", has_msa);
>+    int has_mmi = TestCpuFlag(kCpuHasMMI);
>+    printf("Has MMI %x\n", has_mmi);
>   }
>   if (has_x86) {
>     int has_sse2 = TestCpuFlag(kCpuHasSSE2);

Actions: View | Formatted Diff | Diff

Attachments on bug 192864: 357696