media: Add support for playback of VP8 Alpha video streams

BUG=147355
TEST=VP8 Alpha video streams play
TBR=sky

Review URL: https://chromiumcodereview.appspot.com/13886011

git-svn-id: svn://svn.chromium.org/chrome/trunk/src@195339 0039d316-1c4b-4281-b951-d872f2087c98

18 files changed, 614 insertions, 13 deletions

diff --git a/media/base/decoder_buffer.cc b/media/base/decoder_buffer.cc
index 03f9bbb..aec4521 100644
--- a/media/base/decoder_buffer.cc
+++ b/media/base/decoder_buffer.cc
@@ -10,12 +10,14 @@
 namespace media {
 
 DecoderBuffer::DecoderBuffer(int size)
-    : size_(size) {
+    : size_(size),
+      side_data_size_(0) {
   Initialize();
 }
 
 DecoderBuffer::DecoderBuffer(const uint8* data, int size)
-    : size_(size) {
+    : size_(size),
+      side_data_size_(0) {
   if (!data) {
     CHECK_EQ(size_, 0);
     return;
@@ -25,6 +27,20 @@ DecoderBuffer::DecoderBuffer(const uint8* data, int size)
   memcpy(data_.get(), data, size_);
 }
 
+DecoderBuffer::DecoderBuffer(const uint8* data, int size,
+                             const uint8* side_data, int side_data_size)
+    : size_(size),
+      side_data_size_(side_data_size) {
+  if (!data) {
+    CHECK_EQ(size_, 0);
+    return;
+  }
+
+  Initialize();
+  memcpy(data_.get(), data, size_);
+  memcpy(side_data_.get(), side_data, side_data_size_);
+}
+
 DecoderBuffer::~DecoderBuffer() {}
 
 void DecoderBuffer::Initialize() {
@@ -32,6 +48,11 @@ void DecoderBuffer::Initialize() {
   data_.reset(reinterpret_cast<uint8*>(
       base::AlignedAlloc(size_ + kPaddingSize, kAlignmentSize)));
   memset(data_.get() + size_, 0, kPaddingSize);
+  if (side_data_size_ > 0) {
+    side_data_.reset(reinterpret_cast<uint8*>(
+        base::AlignedAlloc(side_data_size_ + kPaddingSize, kAlignmentSize)));
+    memset(side_data_.get() + side_data_size_, 0, kPaddingSize);
+  }
 }
 
 // static
@@ -43,6 +64,18 @@ scoped_refptr<DecoderBuffer> DecoderBuffer::CopyFrom(const uint8* data,
 }
 
 // static
+scoped_refptr<DecoderBuffer> DecoderBuffer::CopyFrom(const uint8* data,
+                                                     int data_size,
+                                                     const uint8* side_data,
+                                                     int side_data_size) {
+  // If you hit this CHECK you likely have a bug in a demuxer. Go fix it.
+  CHECK(data);
+  CHECK(side_data);
+  return make_scoped_refptr(new DecoderBuffer(data, data_size,
+                                              side_data, side_data_size));
+}
+
+// static
 scoped_refptr<DecoderBuffer> DecoderBuffer::CreateEOSBuffer() {
   return make_scoped_refptr(new DecoderBuffer(NULL, 0));
 }
@@ -82,6 +115,16 @@ int DecoderBuffer::GetDataSize() const {
   return size_;
 }
 
+const uint8* DecoderBuffer::GetSideData() const {
+  DCHECK(!IsEndOfStream());
+  return side_data_.get();
+}
+
+int DecoderBuffer::GetSideDataSize() const {
+  DCHECK(!IsEndOfStream());
+  return side_data_size_;
+}
+
 const DecryptConfig* DecoderBuffer::GetDecryptConfig() const {
   DCHECK(!IsEndOfStream());
   return decrypt_config_.get();
@@ -105,6 +148,7 @@ std::string DecoderBuffer::AsHumanReadableString() {
   s << "timestamp: " << timestamp_.InMicroseconds()
     << " duration: " << duration_.InMicroseconds()
     << " size: " << size_
+    << " side_data_size: " << side_data_size_
     << " encrypted: " << (decrypt_config_ != NULL);
   return s.str();
 }
diff --git a/media/base/decoder_buffer.h b/media/base/decoder_buffer.h
index c23e88f..168ab2c 100644
--- a/media/base/decoder_buffer.h
+++ b/media/base/decoder_buffer.h
@@ -47,6 +47,13 @@ class MEDIA_EXPORT DecoderBuffer
   // padded and aligned as necessary.  |data| must not be NULL and |size| >= 0.
   static scoped_refptr<DecoderBuffer> CopyFrom(const uint8* data, int size);
 
+  // Create a DecoderBuffer whose |data_| is copied from |data| and |side_data_|
+  // is copied from |side_data|. Buffers will be padded and aligned as necessary
+  // Data pointers must not be NULL and sizes must be >= 0.
+  static scoped_refptr<DecoderBuffer> CopyFrom(const uint8* data, int size,
+                                               const uint8* side_data,
+                                               int side_data_size);
+
   // Create a DecoderBuffer indicating we've reached end of stream.
   //
   // Calling any method other than IsEndOfStream() on the resulting buffer
@@ -64,6 +71,9 @@ class MEDIA_EXPORT DecoderBuffer
 
   int GetDataSize() const;
 
+  const uint8* GetSideData() const;
+  int GetSideDataSize() const;
+
   const DecryptConfig* GetDecryptConfig() const;
   void SetDecryptConfig(scoped_ptr<DecryptConfig> decrypt_config);
 
@@ -80,6 +90,8 @@ class MEDIA_EXPORT DecoderBuffer
   // will be padded and aligned as necessary.  If |data| is NULL then |data_| is
   // set to NULL and |buffer_size_| to 0.
   DecoderBuffer(const uint8* data, int size);
+  DecoderBuffer(const uint8* data, int size,
+                const uint8* side_data, int side_data_size);
   virtual ~DecoderBuffer();
 
  private:
@@ -88,6 +100,8 @@ class MEDIA_EXPORT DecoderBuffer
 
   int size_;
   scoped_ptr<uint8, base::ScopedPtrAlignedFree> data_;
+  int side_data_size_;
+  scoped_ptr<uint8, base::ScopedPtrAlignedFree> side_data_;
   scoped_ptr<DecryptConfig> decrypt_config_;
 
   // Constructor helper method for memory allocations.
diff --git a/media/base/decoder_buffer_unittest.cc b/media/base/decoder_buffer_unittest.cc
index 32c38d0..7880a80 100644
--- a/media/base/decoder_buffer_unittest.cc
+++ b/media/base/decoder_buffer_unittest.cc
@@ -35,6 +35,17 @@ TEST(DecoderBufferTest, CopyFrom) {
   EXPECT_EQ(buffer2->GetDataSize(), kDataSize);
   EXPECT_EQ(0, memcmp(buffer2->GetData(), kData, kDataSize));
   EXPECT_FALSE(buffer2->IsEndOfStream());
+  scoped_refptr<DecoderBuffer> buffer3(DecoderBuffer::CopyFrom(
+      reinterpret_cast<const uint8*>(&kData), kDataSize,
+      reinterpret_cast<const uint8*>(&kData), kDataSize));
+  ASSERT_TRUE(buffer3);
+  EXPECT_NE(kData, buffer3->GetData());
+  EXPECT_EQ(buffer3->GetDataSize(), kDataSize);
+  EXPECT_EQ(0, memcmp(buffer3->GetData(), kData, kDataSize));
+  EXPECT_NE(kData, buffer3->GetSideData());
+  EXPECT_EQ(buffer3->GetSideDataSize(), kDataSize);
+  EXPECT_EQ(0, memcmp(buffer3->GetSideData(), kData, kDataSize));
+  EXPECT_FALSE(buffer3->IsEndOfStream());
 }
 
 #if !defined(OS_ANDROID)
diff --git a/media/base/media_switches.cc b/media/base/media_switches.cc
index 04e711e..7c1ab90 100644
--- a/media/base/media_switches.cc
+++ b/media/base/media_switches.cc
@@ -57,6 +57,9 @@ const char kEnableOpusPlayback[] = "enable-opus-playback";
 // Enables VP9 playback in media elements.
 const char kEnableVp9Playback[] = "enable-vp9-playback";
 
+// Enables VP8 Alpha playback in media elements.
+const char kEnableVp8AlphaPlayback[] = "enable-vp8-alpha-playback";
+
 #if defined(OS_WIN)
 const char kWaveOutBuffers[] = "waveout-buffers";
 #endif
diff --git a/media/base/media_switches.h b/media/base/media_switches.h
index ea32fae..e9b6ab9 100644
--- a/media/base/media_switches.h
+++ b/media/base/media_switches.h
@@ -44,6 +44,8 @@ MEDIA_EXPORT extern const char kEnableOpusPlayback[];
 
 MEDIA_EXPORT extern const char kEnableVp9Playback[];
 
+MEDIA_EXPORT extern const char kEnableVp8AlphaPlayback[];
+
 #if defined(OS_WIN)
 MEDIA_EXPORT extern const char kWaveOutBuffers[];
 #endif
diff --git a/media/base/simd/convert_yuv_to_rgb.h b/media/base/simd/convert_yuv_to_rgb.h
index d05f039..7db35b5 100644
--- a/media/base/simd/convert_yuv_to_rgb.h
+++ b/media/base/simd/convert_yuv_to_rgb.h
@@ -21,6 +21,19 @@ typedef void (*ConvertYUVToRGB32Proc)(const uint8*,
                                       int,
                                       YUVType);
 
+typedef void (*ConvertYUVAToARGBProc)(const uint8*,
+                                      const uint8*,
+                                      const uint8*,
+                                      const uint8*,
+                                      uint8*,
+                                      int,
+                                      int,
+                                      int,
+                                      int,
+                                      int,
+                                      int,
+                                      YUVType);
+
 void ConvertYUVToRGB32_C(const uint8* yplane,
                          const uint8* uplane,
                          const uint8* vplane,
@@ -32,6 +45,19 @@ void ConvertYUVToRGB32_C(const uint8* yplane,
                          int rgbstride,
                          YUVType yuv_type);
 
+void ConvertYUVAToARGB_C(const uint8* yplane,
+                         const uint8* uplane,
+                         const uint8* vplane,
+                         const uint8* aplane,
+                         uint8* rgbframe,
+                         int width,
+                         int height,
+                         int ystride,
+                         int uvstride,
+                         int avstride,
+                         int rgbstride,
+                         YUVType yuv_type);
+
 void ConvertYUVToRGB32_SSE(const uint8* yplane,
                            const uint8* uplane,
                            const uint8* vplane,
@@ -54,6 +80,19 @@ void ConvertYUVToRGB32_MMX(const uint8* yplane,
                            int rgbstride,
                            YUVType yuv_type);
 
+void ConvertYUVAToARGB_MMX(const uint8* yplane,
+                           const uint8* uplane,
+                           const uint8* vplane,
+                           const uint8* aplane,
+                           uint8* rgbframe,
+                           int width,
+                           int height,
+                           int ystride,
+                           int uvstride,
+                           int avstride,
+                           int rgbstride,
+                           YUVType yuv_type);
+
 }  // namespace media
 
 // Assembly functions are declared without namespace.
@@ -72,6 +111,13 @@ typedef void (*ConvertYUVToRGB32RowProc)(const uint8*,
                                           uint8*,
                                           ptrdiff_t);
 
+typedef void (*ConvertYUVAToARGBRowProc)(const uint8*,
+                                         const uint8*,
+                                         const uint8*,
+                                         const uint8*,
+                                         uint8*,
+                                         ptrdiff_t);
+
 typedef void (*ScaleYUVToRGB32RowProc)(const uint8*,
                                        const uint8*,
                                        const uint8*,
@@ -85,12 +131,26 @@ void ConvertYUVToRGB32Row_C(const uint8* yplane,
                             uint8* rgbframe,
                             ptrdiff_t width);
 
+void ConvertYUVAToARGBRow_C(const uint8* yplane,
+                            const uint8* uplane,
+                            const uint8* vplane,
+                            const uint8* aplane,
+                            uint8* rgbframe,
+                            ptrdiff_t width);
+
 void ConvertYUVToRGB32Row_MMX(const uint8* yplane,
                               const uint8* uplane,
                               const uint8* vplane,
                               uint8* rgbframe,
                               ptrdiff_t width);
 
+void ConvertYUVAToARGBRow_MMX(const uint8* yplane,
+                              const uint8* uplane,
+                              const uint8* vplane,
+                              const uint8* aplane,
+                              uint8* rgbframe,
+                              ptrdiff_t width);
+
 void ConvertYUVToRGB32Row_SSE(const uint8* yplane,
                               const uint8* uplane,
                               const uint8* vplane,
diff --git a/media/base/simd/convert_yuv_to_rgb_c.cc b/media/base/simd/convert_yuv_to_rgb_c.cc
index 2849cac..e09ed13 100644
--- a/media/base/simd/convert_yuv_to_rgb_c.cc
+++ b/media/base/simd/convert_yuv_to_rgb_c.cc
@@ -39,6 +39,34 @@ static inline void ConvertYUVToRGB32_C(uint8 y,
                                         (packuswb(a) << 24);
 }
 
+static inline void ConvertYUVAToARGB_C(uint8 y,
+                                       uint8 u,
+                                       uint8 v,
+                                       uint8 a,
+                                       uint8* rgb_buf) {
+  int b = kCoefficientsRgbY[256+u][0];
+  int g = kCoefficientsRgbY[256+u][1];
+  int r = kCoefficientsRgbY[256+u][2];
+
+  b = paddsw(b, kCoefficientsRgbY[512+v][0]);
+  g = paddsw(g, kCoefficientsRgbY[512+v][1]);
+  r = paddsw(r, kCoefficientsRgbY[512+v][2]);
+
+  b = paddsw(b, kCoefficientsRgbY[y][0]);
+  g = paddsw(g, kCoefficientsRgbY[y][1]);
+  r = paddsw(r, kCoefficientsRgbY[y][2]);
+
+  b >>= 6;
+  g >>= 6;
+  r >>= 6;
+
+  b = packuswb(b) * a >> 8;
+  g = packuswb(g) * a >> 8;
+  r = packuswb(r) * a >> 8;
+
+  *reinterpret_cast<uint32*>(rgb_buf) = b | (g << 8) | (r << 16) | (a << 24);
+}
+
 extern "C" {
 
 void ConvertYUVToRGB32Row_C(const uint8* y_buf,
@@ -59,6 +87,27 @@ void ConvertYUVToRGB32Row_C(const uint8* y_buf,
   }
 }
 
+void ConvertYUVAToARGBRow_C(const uint8* y_buf,
+                            const uint8* u_buf,
+                            const uint8* v_buf,
+                            const uint8* a_buf,
+                            uint8* rgba_buf,
+                            ptrdiff_t width) {
+  for (int x = 0; x < width; x += 2) {
+    uint8 u = u_buf[x >> 1];
+    uint8 v = v_buf[x >> 1];
+    uint8 y0 = y_buf[x];
+    uint8 a0 = a_buf[x];
+    ConvertYUVAToARGB_C(y0, u, v, a0, rgba_buf);
+    if ((x + 1) < width) {
+      uint8 y1 = y_buf[x + 1];
+      uint8 a1 = a_buf[x + 1];
+      ConvertYUVAToARGB_C(y1, u, v, a1, rgba_buf + 4);
+    }
+    rgba_buf += 8;  // Advance 2 pixels.
+  }
+}
+
 // 16.16 fixed point is used.  A shift by 16 isolates the integer.
 // A shift by 17 is used to further subsample the chrominence channels.
 // & 0xffff isolates the fixed point fraction.  >> 2 to get the upper 2 bits,
@@ -161,4 +210,33 @@ void ConvertYUVToRGB32_C(const uint8* yplane,
   }
 }
 
+void ConvertYUVAToARGB_C(const uint8* yplane,
+                         const uint8* uplane,
+                         const uint8* vplane,
+                         const uint8* aplane,
+                         uint8* rgbaframe,
+                         int width,
+                         int height,
+                         int ystride,
+                         int uvstride,
+                         int astride,
+                         int rgbastride,
+                         YUVType yuv_type) {
+  unsigned int y_shift = yuv_type;
+  for (int y = 0; y < height; y++) {
+    uint8* rgba_row = rgbaframe + y * rgbastride;
+    const uint8* y_ptr = yplane + y * ystride;
+    const uint8* u_ptr = uplane + (y >> y_shift) * uvstride;
+    const uint8* v_ptr = vplane + (y >> y_shift) * uvstride;
+    const uint8* a_ptr = aplane + y * astride;
+
+    ConvertYUVAToARGBRow_C(y_ptr,
+                           u_ptr,
+                           v_ptr,
+                           a_ptr,
+                           rgba_row,
+                           width);
+  }
+}
+
 }  // namespace media
diff --git a/media/base/simd/convert_yuv_to_rgb_x86.cc b/media/base/simd/convert_yuv_to_rgb_x86.cc
index 37b168d..d1d6e16 100644
--- a/media/base/simd/convert_yuv_to_rgb_x86.cc
+++ b/media/base/simd/convert_yuv_to_rgb_x86.cc
@@ -40,6 +40,37 @@ void ConvertYUVToRGB32_MMX(const uint8* yplane,
   EmptyRegisterState();
 }
 
+void ConvertYUVAToARGB_MMX(const uint8* yplane,
+                           const uint8* uplane,
+                           const uint8* vplane,
+                           const uint8* aplane,
+                           uint8* rgbframe,
+                           int width,
+                           int height,
+                           int ystride,
+                           int uvstride,
+                           int astride,
+                           int rgbstride,
+                           YUVType yuv_type) {
+  unsigned int y_shift = yuv_type;
+  for (int y = 0; y < height; ++y) {
+    uint8* rgb_row = rgbframe + y * rgbstride;
+    const uint8* y_ptr = yplane + y * ystride;
+    const uint8* u_ptr = uplane + (y >> y_shift) * uvstride;
+    const uint8* v_ptr = vplane + (y >> y_shift) * uvstride;
+    const uint8* a_ptr = aplane + y * astride;
+
+    ConvertYUVAToARGBRow_MMX(y_ptr,
+                             u_ptr,
+                             v_ptr,
+                             a_ptr,
+                             rgb_row,
+                             width);
+  }
+
+  EmptyRegisterState();
+}
+
 void ConvertYUVToRGB32_SSE(const uint8* yplane,
                            const uint8* uplane,
                            const uint8* vplane,
diff --git a/media/base/simd/convert_yuva_to_argb_mmx.asm b/media/base/simd/convert_yuva_to_argb_mmx.asm
new file mode 100644
index 0000000..b39315dc
--- /dev/null
+++ b/media/base/simd/convert_yuva_to_argb_mmx.asm
@@ -0,0 +1,23 @@
+; Copyright (c) 2011 The Chromium Authors. All rights reserved.
+; Use of this source code is governed by a BSD-style license that can be
+; found in the LICENSE file.
+
+%include "third_party/x86inc/x86inc.asm"
+
+;
+; This file uses MMX instructions.
+;
+  SECTION_TEXT
+  CPU       MMX
+
+; Use movq to save the output.
+%define MOVQ movq
+
+; extern "C" void ConvertYUVAToARGBRow_MMX(const uint8* y_buf,
+;                                          const uint8* u_buf,
+;                                          const uint8* v_buf,
+;                                          const uint8* a_buf,
+;                                          uint8* rgb_buf,
+;                                          ptrdiff_t width);
+%define SYMBOL ConvertYUVAToARGBRow_MMX
+%include "convert_yuva_to_argb_mmx.inc"
diff --git a/media/base/simd/convert_yuva_to_argb_mmx.inc b/media/base/simd/convert_yuva_to_argb_mmx.inc
new file mode 100644
index 0000000..621100e
--- /dev/null
+++ b/media/base/simd/convert_yuva_to_argb_mmx.inc
@@ -0,0 +1,174 @@
+; Copyright (c) 2011 The Chromium Authors. All rights reserved.
+; Use of this source code is governed by a BSD-style license that can be
+; found in the LICENSE file.
+
+  global    mangle(SYMBOL) PRIVATE
+  align     function_align
+
+; Non-PIC code is the fastest so use this if possible.
+%ifndef PIC
+mangle(SYMBOL):
+  %assign   stack_offset 0
+  PROLOGUE  6, 7, 3, Y, U, V, A, ARGB, WIDTH, TEMP
+  extern    mangle(kCoefficientsRgbY)
+  jmp       .convertend
+
+.convertloop:
+  movzx     TEMPd, BYTE [Uq]
+  movq      mm0, [mangle(kCoefficientsRgbY) + 2048 + 8 * TEMPq]
+  add       Uq, 1
+  movzx     TEMPd, BYTE [Vq]
+  paddsw    mm0, [mangle(kCoefficientsRgbY) + 4096 + 8 * TEMPq]
+  add       Vq, 1
+  movzx     TEMPd, BYTE [Yq]
+  movq      mm1, [mangle(kCoefficientsRgbY) + 8 * TEMPq]
+  movzx     TEMPd, BYTE [Yq + 1]
+  movq      mm2, [mangle(kCoefficientsRgbY) + 8 * TEMPq]
+  add       Yq, 2
+  paddsw    mm1, mm0
+  paddsw    mm2, mm0
+  psraw     mm1, 6
+  psraw     mm2, 6
+  packuswb  mm1, mm2
+
+  ; Multiply ARGB by alpha value.
+  movq      mm0, mm1
+  pxor      mm2, mm2
+  punpcklbw mm0, mm2
+  punpckhbw mm1, mm2
+  movzx     TEMPd, BYTE [Aq]
+  movq      mm2, [mangle(kCoefficientsRgbY) + 6144 + 8 * TEMPq]
+  pmullw    mm0, mm2
+  psrlw     mm0, 8
+  movzx     TEMPd, BYTE [Aq + 1]
+  movq      mm2, [mangle(kCoefficientsRgbY) + 6144 + 8 * TEMPq]
+  add       Aq, 2
+  pmullw    mm1, mm2
+  psrlw     mm1, 8
+  packuswb  mm0, mm1
+
+  MOVQ      [ARGBq], mm0
+  add       ARGBq, 8
+
+.convertend:
+  sub       WIDTHq, 2
+  jns       .convertloop
+
+  ; If number of pixels is odd then compute it.
+  and       WIDTHq, 1
+  jz        .convertdone
+
+  movzx     TEMPd, BYTE [Uq]
+  movq      mm0, [mangle(kCoefficientsRgbY) + 2048 + 8 * TEMPq]
+  movzx     TEMPd, BYTE [Vq]
+  paddsw    mm0, [mangle(kCoefficientsRgbY) + 4096 + 8 * TEMPq]
+  movzx     TEMPd, BYTE [Yq]
+  movq      mm1, [mangle(kCoefficientsRgbY) + 8 * TEMPq]
+  paddsw    mm1, mm0
+  psraw     mm1, 6
+  packuswb  mm1, mm1
+
+  ; Multiply ARGB by alpha value.
+  pxor      mm0, mm0
+  punpcklbw mm1, mm0
+  movzx     TEMPd, BYTE [Aq]
+  movq      mm0, [mangle(kCoefficientsRgbY) + 6144 + 8 * TEMPq]
+  pmullw    mm1, mm0
+  psrlw     mm1, 8
+  packuswb  mm1, mm1
+
+  movd      [ARGBq], mm1
+
+.convertdone:
+  RET
+%endif
+
+; With PIC code we need to load the address of mangle(kCoefficientsRgbY).
+; This code is slower than the above version.
+%ifdef PIC
+mangle(SYMBOL):
+  %assign   stack_offset 0
+  PROLOGUE  6, 7, 3, Y, U, V, A, ARGB, WIDTH, TEMP
+  extern    mangle(kCoefficientsRgbY)
+  PUSH      WIDTHq
+  DEFINE_ARGS Y, U, V, A, ARGB, TABLE, TEMP
+  LOAD_SYM  TABLEq, mangle(kCoefficientsRgbY)
+  jmp       .convertend
+
+.convertloop:
+  movzx     TEMPd, BYTE [Uq]
+  movq      mm0, [TABLEq + 2048 + 8 * TEMPq]
+  add       Uq, 1
+
+  movzx     TEMPd, BYTE [Vq]
+  paddsw    mm0, [TABLEq + 4096 + 8 * TEMPq]
+  add       Vq, 1
+
+  movzx     TEMPd, BYTE [Yq]
+  movq      mm1, [TABLEq + 8 * TEMPq]
+
+  movzx     TEMPd, BYTE [Yq + 1]
+  movq      mm2, [TABLEq + 8 * TEMPq]
+  add       Yq, 2
+
+  ; Add UV components to Y component.
+  paddsw    mm1, mm0
+  paddsw    mm2, mm0
+
+  ; Down shift and then pack.
+  psraw     mm1, 6
+  psraw     mm2, 6
+  packuswb  mm1, mm2
+
+  ; Unpack and multiply by alpha value, then repack high bytes of words.
+  movq      mm0, mm1
+  pxor      mm2, mm2
+  punpcklbw mm0, mm2
+  punpckhbw mm1, mm2
+  movzx     TEMPd, BYTE [Aq]
+  movq      mm2, [TABLEq + 6144 + 8 * TEMPq]
+  pmullw    mm0, mm2
+  psrlw     mm0, 8
+  movzx     TEMPd, BYTE [Aq + 1]
+  movq      mm2, [TABLEq + 6144 + 8 * TEMPq]
+  add       Aq, 2
+  pmullw    mm1, mm2
+  psrlw     mm1, 8
+  packuswb  mm0, mm1
+
+  MOVQ      [ARGBq], mm0
+  add       ARGBq, 8
+
+.convertend:
+  sub       dword [rsp], 2
+  jns       .convertloop
+
+  ; If number of pixels is odd then compute it.
+  and       dword [rsp], 1
+  jz        .convertdone
+
+  movzx     TEMPd, BYTE [Uq]
+  movq      mm0, [TABLEq + 2048 + 8 * TEMPq]
+  movzx     TEMPd, BYTE [Vq]
+  paddsw    mm0, [TABLEq + 4096 + 8 * TEMPq]
+  movzx     TEMPd, BYTE [Yq]
+  movq      mm1, [TABLEq + 8 * TEMPq]
+  paddsw    mm1, mm0
+  psraw     mm1, 6
+  packuswb  mm1, mm1
+
+  ; Multiply ARGB by alpha value.
+  pxor      mm0, mm0
+  punpcklbw mm1, mm0
+  movzx     TEMPd, BYTE [Aq]
+  movq      mm0, [TABLEq + 6144 + 8 * TEMPq]
+  pmullw    mm1, mm0
+  psrlw     mm1, 8
+  packuswb  mm1, mm1
+
+  movd      [ARGBq], mm1
+
+.convertdone:
+  POP       TABLEq
+  RET
+%endif
\ No newline at end of file
diff --git a/media/base/simd/yuv_to_rgb_table.cc b/media/base/simd/yuv_to_rgb_table.cc
index f998e85..3789969 100644
--- a/media/base/simd/yuv_to_rgb_table.cc
+++ b/media/base/simd/yuv_to_rgb_table.cc
@@ -6,6 +6,7 @@
 
 extern "C" {
 
+// Defines the R,G,B,A contributions from Y.
 #define RGBY(i) { \
   static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \
   static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \
@@ -13,6 +14,9 @@ extern "C" {
   0 \
 }
 
+// Defines the R,G,B,A contributions from U.
+// The contribution to A is the same for any value of U
+// causing the final A value to be 255 in every conversion.
 #define RGBU(i) { \
   static_cast<int16>(2.018 * 64 * (i - 128) + 0.5), \
   static_cast<int16>(-0.391 * 64 * (i - 128) + 0.5), \
@@ -20,6 +24,7 @@ extern "C" {
   static_cast<int16>(256 * 64 - 1) \
 }
 
+// Defines the R,G,B,A contributions from V.
 #define RGBV(i) { \
   0, \
   static_cast<int16>(-0.813 * 64 * (i - 128) + 0.5), \
@@ -27,7 +32,18 @@ extern "C" {
   0 \
 }
 
-SIMD_ALIGNED(int16 kCoefficientsRgbY[256 * 3][4]) = {
+// Used to define a set of multiplier words for each alpha level.
+#define ALPHA(i) { \
+  i, i, i, i \
+}
+
+// The following table defines the RGBA contributions
+// for each component of YUVA. The Y table is first followed
+// by the U, and V tables. The alpha multiplier table follows.
+// These tables are aligned and kept adjacent to optimize for
+// SIMD and cacheing.
+
+SIMD_ALIGNED(int16 kCoefficientsRgbY[256 * 4][4]) = {
   RGBY(0x00), RGBY(0x01), RGBY(0x02), RGBY(0x03),
   RGBY(0x04), RGBY(0x05), RGBY(0x06), RGBY(0x07),
   RGBY(0x08), RGBY(0x09), RGBY(0x0A), RGBY(0x0B),
@@ -224,10 +240,77 @@ SIMD_ALIGNED(int16 kCoefficientsRgbY[256 * 3][4]) = {
   RGBV(0xF4), RGBV(0xF5), RGBV(0xF6), RGBV(0xF7),
   RGBV(0xF8), RGBV(0xF9), RGBV(0xFA), RGBV(0xFB),
   RGBV(0xFC), RGBV(0xFD), RGBV(0xFE), RGBV(0xFF),
+
+  // Alpha multipliers for each  alpha level.
+  ALPHA(0x00), ALPHA(0x01), ALPHA(0x02), ALPHA(0x03),
+  ALPHA(0x04), ALPHA(0x05), ALPHA(0x06), ALPHA(0x07),
+  ALPHA(0x08), ALPHA(0x09), ALPHA(0x0A), ALPHA(0x0B),
+  ALPHA(0x0C), ALPHA(0x0D), ALPHA(0x0E), ALPHA(0x0F),
+  ALPHA(0x10), ALPHA(0x11), ALPHA(0x12), ALPHA(0x13),
+  ALPHA(0x14), ALPHA(0x15), ALPHA(0x16), ALPHA(0x17),
+  ALPHA(0x18), ALPHA(0x19), ALPHA(0x1A), ALPHA(0x1B),
+  ALPHA(0x1C), ALPHA(0x1D), ALPHA(0x1E), ALPHA(0x1F),
+  ALPHA(0x20), ALPHA(0x21), ALPHA(0x22), ALPHA(0x23),
+  ALPHA(0x24), ALPHA(0x25), ALPHA(0x26), ALPHA(0x27),
+  ALPHA(0x28), ALPHA(0x29), ALPHA(0x2A), ALPHA(0x2B),
+  ALPHA(0x2C), ALPHA(0x2D), ALPHA(0x2E), ALPHA(0x2F),
+  ALPHA(0x30), ALPHA(0x31), ALPHA(0x32), ALPHA(0x33),
+  ALPHA(0x34), ALPHA(0x35), ALPHA(0x36), ALPHA(0x37),
+  ALPHA(0x38), ALPHA(0x39), ALPHA(0x3A), ALPHA(0x3B),
+  ALPHA(0x3C), ALPHA(0x3D), ALPHA(0x3E), ALPHA(0x3F),
+  ALPHA(0x40), ALPHA(0x41), ALPHA(0x42), ALPHA(0x43),
+  ALPHA(0x44), ALPHA(0x45), ALPHA(0x46), ALPHA(0x47),
+  ALPHA(0x48), ALPHA(0x49), ALPHA(0x4A), ALPHA(0x4B),
+  ALPHA(0x4C), ALPHA(0x4D), ALPHA(0x4E), ALPHA(0x4F),
+  ALPHA(0x50), ALPHA(0x51), ALPHA(0x52), ALPHA(0x53),
+  ALPHA(0x54), ALPHA(0x55), ALPHA(0x56), ALPHA(0x57),
+  ALPHA(0x58), ALPHA(0x59), ALPHA(0x5A), ALPHA(0x5B),
+  ALPHA(0x5C), ALPHA(0x5D), ALPHA(0x5E), ALPHA(0x5F),
+  ALPHA(0x60), ALPHA(0x61), ALPHA(0x62), ALPHA(0x63),
+  ALPHA(0x64), ALPHA(0x65), ALPHA(0x66), ALPHA(0x67),
+  ALPHA(0x68), ALPHA(0x69), ALPHA(0x6A), ALPHA(0x6B),
+  ALPHA(0x6C), ALPHA(0x6D), ALPHA(0x6E), ALPHA(0x6F),
+  ALPHA(0x70), ALPHA(0x71), ALPHA(0x72), ALPHA(0x73),
+  ALPHA(0x74), ALPHA(0x75), ALPHA(0x76), ALPHA(0x77),
+  ALPHA(0x78), ALPHA(0x79), ALPHA(0x7A), ALPHA(0x7B),
+  ALPHA(0x7C), ALPHA(0x7D), ALPHA(0x7E), ALPHA(0x7F),
+  ALPHA(0x80), ALPHA(0x81), ALPHA(0x82), ALPHA(0x83),
+  ALPHA(0x84), ALPHA(0x85), ALPHA(0x86), ALPHA(0x87),
+  ALPHA(0x88), ALPHA(0x89), ALPHA(0x8A), ALPHA(0x8B),
+  ALPHA(0x8C), ALPHA(0x8D), ALPHA(0x8E), ALPHA(0x8F),
+  ALPHA(0x90), ALPHA(0x91), ALPHA(0x92), ALPHA(0x93),
+  ALPHA(0x94), ALPHA(0x95), ALPHA(0x96), ALPHA(0x97),
+  ALPHA(0x98), ALPHA(0x99), ALPHA(0x9A), ALPHA(0x9B),
+  ALPHA(0x9C), ALPHA(0x9D), ALPHA(0x9E), ALPHA(0x9F),
+  ALPHA(0xA0), ALPHA(0xA1), ALPHA(0xA2), ALPHA(0xA3),
+  ALPHA(0xA4), ALPHA(0xA5), ALPHA(0xA6), ALPHA(0xA7),
+  ALPHA(0xA8), ALPHA(0xA9), ALPHA(0xAA), ALPHA(0xAB),
+  ALPHA(0xAC), ALPHA(0xAD), ALPHA(0xAE), ALPHA(0xAF),
+  ALPHA(0xB0), ALPHA(0xB1), ALPHA(0xB2), ALPHA(0xB3),
+  ALPHA(0xB4), ALPHA(0xB5), ALPHA(0xB6), ALPHA(0xB7),
+  ALPHA(0xB8), ALPHA(0xB9), ALPHA(0xBA), ALPHA(0xBB),
+  ALPHA(0xBC), ALPHA(0xBD), ALPHA(0xBE), ALPHA(0xBF),
+  ALPHA(0xC0), ALPHA(0xC1), ALPHA(0xC2), ALPHA(0xC3),
+  ALPHA(0xC4), ALPHA(0xC5), ALPHA(0xC6), ALPHA(0xC7),
+  ALPHA(0xC8), ALPHA(0xC9), ALPHA(0xCA), ALPHA(0xCB),
+  ALPHA(0xCC), ALPHA(0xCD), ALPHA(0xCE), ALPHA(0xCF),
+  ALPHA(0xD0), ALPHA(0xD1), ALPHA(0xD2), ALPHA(0xD3),
+  ALPHA(0xD4), ALPHA(0xD5), ALPHA(0xD6), ALPHA(0xD7),
+  ALPHA(0xD8), ALPHA(0xD9), ALPHA(0xDA), ALPHA(0xDB),
+  ALPHA(0xDC), ALPHA(0xDD), ALPHA(0xDE), ALPHA(0xDF),
+  ALPHA(0xE0), ALPHA(0xE1), ALPHA(0xE2), ALPHA(0xE3),
+  ALPHA(0xE4), ALPHA(0xE5), ALPHA(0xE6), ALPHA(0xE7),
+  ALPHA(0xE8), ALPHA(0xE9), ALPHA(0xEA), ALPHA(0xEB),
+  ALPHA(0xEC), ALPHA(0xED), ALPHA(0xEE), ALPHA(0xEF),
+  ALPHA(0xF0), ALPHA(0xF1), ALPHA(0xF2), ALPHA(0xF3),
+  ALPHA(0xF4), ALPHA(0xF5), ALPHA(0xF6), ALPHA(0xF7),
+  ALPHA(0xF8), ALPHA(0xF9), ALPHA(0xFA), ALPHA(0xFB),
+  ALPHA(0xFC), ALPHA(0xFD), ALPHA(0xFE), ALPHA(0xFF),
 };
 
 #undef RGBY
 #undef RGBU
 #undef RGBV
+#undef ALPHA
 
 }  // extern "C"
diff --git a/media/base/simd/yuv_to_rgb_table.h b/media/base/simd/yuv_to_rgb_table.h
index 0c43a7a..3ed7bd9 100644
--- a/media/base/simd/yuv_to_rgb_table.h
+++ b/media/base/simd/yuv_to_rgb_table.h
@@ -19,7 +19,7 @@ extern "C" {
 #endif
 
 // Align the table to 16-bytes to allow faster reading.
-extern SIMD_ALIGNED(int16 kCoefficientsRgbY[768][4]);
+extern SIMD_ALIGNED(int16 kCoefficientsRgbY[256 * 4][4]);
 
 }  // extern "C"
 
diff --git a/media/base/video_frame.cc b/media/base/video_frame.cc
index 9c3b4b6..6d4997b 100644
--- a/media/base/video_frame.cc
+++ b/media/base/video_frame.cc
@@ -32,6 +32,7 @@ scoped_refptr<VideoFrame> VideoFrame::CreateFrame(
       frame->AllocateRGB(4u);
       break;
     case VideoFrame::YV12:
+    case VideoFrame::YV12A:
     case VideoFrame::YV16:
       frame->AllocateYUV();
       break;
@@ -167,6 +168,8 @@ size_t VideoFrame::NumPlanes(Format format) {
     case VideoFrame::YV12:
     case VideoFrame::YV16:
       return 3;
+    case VideoFrame::YV12A:
+      return 4;
     case VideoFrame::EMPTY:
     case VideoFrame::I420:
     case VideoFrame::INVALID:
@@ -204,7 +207,8 @@ void VideoFrame::AllocateRGB(size_t bytes_per_pixel) {
 }
 
 void VideoFrame::AllocateYUV() {
-  DCHECK(format_ == VideoFrame::YV12 || format_ == VideoFrame::YV16);
+  DCHECK(format_ == VideoFrame::YV12 || format_ == VideoFrame::YV16 ||
+         format_ == VideoFrame::YV12A);
   // Align Y rows at least at 16 byte boundaries.  The stride for both
   // YV12 and YV16 is 1/2 of the stride of Y.  For YV12, every row of bytes for
   // U and V applies to two rows of Y (one byte of UV for 4 bytes of Y), so in
@@ -213,7 +217,9 @@ void VideoFrame::AllocateYUV() {
   // YV16. We also round the height of the surface allocated to be an even
   // number to avoid any potential of faulting by code that attempts to access
   // the Y values of the final row, but assumes that the last row of U & V
-  // applies to a full two rows of Y.
+  // applies to a full two rows of Y. YV12A is the same as YV12, but with an
+  // additional alpha plane that has the same size and alignment as the Y plane.
+
   size_t y_stride = RoundUp(row_bytes(VideoFrame::kYPlane),
                             kFrameSizeAlignment);
   size_t uv_stride = RoundUp(row_bytes(VideoFrame::kUPlane),
@@ -222,9 +228,12 @@ void VideoFrame::AllocateYUV() {
   // and then the size needs to be a multiple of two macroblocks (vertically).
   // See libavcodec/utils.c:avcodec_align_dimensions2().
   size_t y_height = RoundUp(coded_size_.height(), kFrameSizeAlignment * 2);
-  size_t uv_height = format_ == VideoFrame::YV12 ? y_height / 2 : y_height;
+  size_t uv_height = (format_ == VideoFrame::YV12 ||
+                      format_ == VideoFrame::YV12A) ?
+                              y_height / 2 : y_height;
   size_t y_bytes = y_height * y_stride;
   size_t uv_bytes = uv_height * uv_stride;
+  size_t a_bytes = format_ == VideoFrame::YV12A ? y_bytes : 0;
 
   // The extra line of UV being allocated is because h264 chroma MC
   // overreads by one line in some cases, see libavcodec/utils.c:
@@ -232,7 +241,7 @@ void VideoFrame::AllocateYUV() {
   // put_h264_chroma_mc4_ssse3().
   uint8* data = reinterpret_cast<uint8*>(
       base::AlignedAlloc(
-          y_bytes + (uv_bytes * 2 + uv_stride) + kFrameSizePadding,
+          y_bytes + (uv_bytes * 2 + uv_stride) + a_bytes + kFrameSizePadding,
           kFrameAddressAlignment));
   no_longer_needed_cb_ = base::Bind(&ReleaseData, data);
   COMPILE_ASSERT(0 == VideoFrame::kYPlane, y_plane_data_must_be_index_0);
@@ -242,6 +251,10 @@ void VideoFrame::AllocateYUV() {
   strides_[VideoFrame::kYPlane] = y_stride;
   strides_[VideoFrame::kUPlane] = uv_stride;
   strides_[VideoFrame::kVPlane] = uv_stride;
+  if (format_ == YV12A) {
+    data_[VideoFrame::kAPlane] = data + y_bytes + (2 * uv_bytes);
+    strides_[VideoFrame::kAPlane] = y_stride;
+  }
 }
 
 VideoFrame::VideoFrame(VideoFrame::Format format,
@@ -285,7 +298,8 @@ int VideoFrame::row_bytes(size_t plane) const {
     // Planar, 8bpp.
     case YV12:
     case YV16:
-      if (plane == kYPlane)
+    case YV12A:
+      if (plane == kYPlane || plane == kAPlane)
         return width;
       return RoundUp(width, 2) / 2;
 
@@ -307,7 +321,8 @@ int VideoFrame::rows(size_t plane) const {
       return height;
 
     case YV12:
-      if (plane == kYPlane)
+    case YV12A:
+      if (plane == kYPlane || plane == kAPlane)
         return height;
       return RoundUp(height, 2) / 2;
 
diff --git a/media/base/video_frame.h b/media/base/video_frame.h
index 9a6f0a6..6cd5a0f 100644
--- a/media/base/video_frame.h
+++ b/media/base/video_frame.h
@@ -24,13 +24,14 @@ class MEDIA_EXPORT VideoFrame : public base::RefCountedThreadSafe<VideoFrame> {
   };
 
   enum {
-    kMaxPlanes = 3,
+    kMaxPlanes = 4,
 
     kRGBPlane = 0,
 
     kYPlane = 0,
     kUPlane = 1,
     kVPlane = 2,
+    kAPlane = 3,
   };
 
   // Surface formats roughly based on FOURCC labels, see:
@@ -47,6 +48,7 @@ class MEDIA_EXPORT VideoFrame : public base::RefCountedThreadSafe<VideoFrame> {
 #if defined(GOOGLE_TV)
     HOLE = 13,  // Hole frame.
 #endif
+    YV12A = 14,  // 20bpp YUVA planar 1x1 Y, 2x2 VU, 1x1 A samples.
   };
 
   // Creates a new frame in system memory with given parameters. Buffers for
diff --git a/media/base/video_util.cc b/media/base/video_util.cc
index 972ac17..e1de7bd 100644
--- a/media/base/video_util.cc
+++ b/media/base/video_util.cc
@@ -62,6 +62,16 @@ void CopyVPlane(const uint8* source, int stride, int rows, VideoFrame* frame) {
   CopyPlane(VideoFrame::kVPlane, source, stride, rows, frame);
 }
 
+void CopyAPlane(const uint8* source, int stride, int rows, VideoFrame* frame) {
+  CopyPlane(VideoFrame::kAPlane, source, stride, rows, frame);
+}
+
+void MakeOpaqueAPlane(int stride, int rows, VideoFrame* frame) {
+  int rows_to_clear = std::min(frame->rows(VideoFrame::kAPlane), rows);
+  memset(frame->data(VideoFrame::kAPlane), 255,
+         frame->stride(VideoFrame::kAPlane) * rows_to_clear);
+}
+
 void FillYUV(VideoFrame* frame, uint8 y, uint8 u, uint8 v) {
   // Fill the Y plane.
   uint8* y_plane = frame->data(VideoFrame::kYPlane);
diff --git a/media/base/video_util.h b/media/base/video_util.h
index 27156fa..05e5927 100644
--- a/media/base/video_util.h
+++ b/media/base/video_util.h
@@ -19,7 +19,7 @@ MEDIA_EXPORT gfx::Size GetNaturalSize(const gfx::Size& visible_size,
                                       int aspect_ratio_numerator,
                                       int aspect_ratio_denominator);
 
-// Copies a plane of YUV source into a VideoFrame object, taking into account
+// Copies a plane of YUV(A) source into a VideoFrame object, taking into account
 // source and destinations dimensions.
 //
 // NOTE: rows is *not* the same as height!
@@ -29,8 +29,14 @@ MEDIA_EXPORT void CopyUPlane(const uint8* source, int stride, int rows,
                              VideoFrame* frame);
 MEDIA_EXPORT void CopyVPlane(const uint8* source, int stride, int rows,
                              VideoFrame* frame);
+MEDIA_EXPORT void CopyAPlane(const uint8* source, int stride, int rows,
+                             VideoFrame* frame);
+
+// Sets alpha plane values to be completely opaque (all 255's).
+MEDIA_EXPORT void MakeOpaqueAPlane(int stride, int rows, VideoFrame* frame);
+
 // |plane| is one of VideoFrame::kYPlane, VideoFrame::kUPlane,
-// or VideoFrame::kVPlane.
+// VideoFrame::kVPlane or VideoFrame::kAPlane
 MEDIA_EXPORT void CopyPlane(size_t plane, const uint8* source, int stride,
                             int rows, VideoFrame* frame);
 
diff --git a/media/base/yuv_convert.cc b/media/base/yuv_convert.cc
index 85b0699..1d09a24 100644
--- a/media/base/yuv_convert.cc
+++ b/media/base/yuv_convert.cc
@@ -603,4 +603,34 @@ void ConvertYUVToRGB32(const uint8* yplane,
 #endif
 }
 
+void ConvertYUVAToARGB(const uint8* yplane,
+                       const uint8* uplane,
+                       const uint8* vplane,
+                       const uint8* aplane,
+                       uint8* rgbframe,
+                       int width,
+                       int height,
+                       int ystride,
+                       int uvstride,
+                       int astride,
+                       int rgbstride,
+                       YUVType yuv_type) {
+#if defined(ARCH_CPU_ARM_FAMILY) || defined(ARCH_CPU_MIPS_FAMILY)
+  ConvertYUVAToARGB_C(yplane, uplane, vplane, aplane, rgbframe,
+                      width, height, ystride, uvstride, astride, rgbstride,
+                      yuv_type);
+#else
+  static ConvertYUVAToARGBProc convert_proc = NULL;
+  if (!convert_proc) {
+    base::CPU cpu;
+    if (cpu.has_mmx())
+      convert_proc = &ConvertYUVAToARGB_MMX;
+    else
+      convert_proc = &ConvertYUVAToARGB_C;
+  }
+  convert_proc(yplane, uplane, vplane, aplane, rgbframe,
+               width, height, ystride, uvstride, astride, rgbstride, yuv_type);
+#endif
+}
+
 }  // namespace media
diff --git a/media/base/yuv_convert.h b/media/base/yuv_convert.h
index 30c07ab..0e53193 100644
--- a/media/base/yuv_convert.h
+++ b/media/base/yuv_convert.h
@@ -60,6 +60,21 @@ void ConvertYUVToRGB32(const uint8* yplane,
                        int rgbstride,
                        YUVType yuv_type);
 
+// Convert a frame of YUVA to 32 bit ARGB.
+// Pass in YV12A
+void ConvertYUVAToARGB(const uint8* yplane,
+                       const uint8* uplane,
+                       const uint8* vplane,
+                       const uint8* aplane,
+                       uint8* rgbframe,
+                       int width,
+                       int height,
+                       int ystride,
+                       int uvstride,
+                       int astride,
+                       int rgbstride,
+                       YUVType yuv_type);
+
 // Scale a frame of YUV to 32 bit ARGB.
 // Supports rotation and mirroring.
 void ScaleYUVToRGB32(const uint8* yplane,