Initial implementation of the DXVA 2.0 H.264 hardware decoder for pepper for Windows. The decodin...

content/common/gpu/media/dxva_video_decode_accelerator.cc

+// 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 "content/common/gpu/media/dxva_video_decode_accelerator.h"
+
+#include <ks.h>
+#include <codecapi.h>
+#include <d3dx9tex.h>
+#include <d3d9types.h>
+#include <dxva.h>
+#include <dxva2api.h>
+#include <mfapi.h>
+#include <mferror.h>
+#include <wmcodecdsp.h>
+
+#include "base/lazy_instance.h"
+#include "base/logging.h"
+#include "base/memory/scoped_handle.h"
+#include "base/memory/scoped_ptr.h"
+#include "base/shared_memory.h"
+#include "base/win/scoped_com_initializer.h"
+#include "base/time.h"
+#include "media/video/video_decode_accelerator.h"
+#include "third_party/angle/include/EGL/egl.h"
+#include "third_party/angle/include/GLES2/gl2.h"
+#include "third_party/angle/include/GLES2/gl2ext.h"
+
+namespace {
+
+static const int kNumPictureBuffers = 5;
+static const int kNumInputs = 100;
+
+IMFSample* CreateEmptySample() {
+  HRESULT hr = E_FAIL;
+  base::win::ScopedComPtr<IMFSample> sample;
+  hr = MFCreateSample(sample.Receive());
+  if (FAILED(hr)) {
+    NOTREACHED() << "Unable to create an empty sample";
+    return NULL;
+  }
+  return sample.Detach();
+}
+
+// Creates a Media Foundation sample with one buffer of length |buffer_length|
+// on a |align|-byte boundary. Alignment must be a perfect power of 2 or 0.
+// If |align| is 0, then no alignment is specified.
+IMFSample* CreateEmptySampleWithBuffer(int buffer_length, int align) {
+  CHECK_GT(buffer_length, 0);
+  base::win::ScopedComPtr<IMFSample> sample;
+  sample.Attach(CreateEmptySample());
+  if (!sample.get())
+    return NULL;
+  base::win::ScopedComPtr<IMFMediaBuffer> buffer;
+  HRESULT hr = E_FAIL;
+  if (align == 0) {
+    // Note that MFCreateMemoryBuffer is same as MFCreateAlignedMemoryBuffer
+    // with the align argument being 0.
+    hr = MFCreateMemoryBuffer(buffer_length, buffer.Receive());
+  } else {
+    hr = MFCreateAlignedMemoryBuffer(buffer_length,
+                                     align - 1,
+                                     buffer.Receive());
+  }
+  if (FAILED(hr)) {
+    NOTREACHED() << "Unable to create an empty buffer";
+    return NULL;
+  }
+  hr = sample->AddBuffer(buffer.get());
+  if (FAILED(hr)) {
+    NOTREACHED() << "Failed to add empty buffer to sample";
+    return NULL;
+  }
+  return sample.Detach();
+}
+
+// Creates a Media Foundation sample with one buffer containing a copy of the
+// given Annex B stream data.
+// If duration and sample time are not known, provide 0.
+// |min_size| specifies the minimum size of the buffer (might be required by
+// the decoder for input). The times here should be given in 100ns units.
+// |alignment| specifies the buffer in the sample to be aligned. If no
+// alignment is required, provide 0 or 1.
+static IMFSample* CreateInputSample(const uint8* stream, int size,
+                                    int64 timestamp, int64 duration,
+                                    int min_size, int alignment) {
+  CHECK(stream);
+  CHECK_GT(size, 0);
+  base::win::ScopedComPtr<IMFSample> sample;
+  sample.Attach(CreateEmptySampleWithBuffer(std::max(min_size, size),
+                                            alignment));
+  if (!sample.get()) {
+    NOTREACHED() << "Failed to create empty buffer for input";
+    return NULL;
+  }
+  HRESULT hr = E_FAIL;
+  if (duration > 0) {
+    hr = sample->SetSampleDuration(duration);
+    if (FAILED(hr)) {
+      NOTREACHED() << "Failed to set sample duration";
+      return NULL;
+    }
+  }
+  if (timestamp > 0) {
+    hr = sample->SetSampleTime(timestamp);
+    if (FAILED(hr)) {
+      NOTREACHED() << "Failed to set sample time";
+      return NULL;
+    }
+  }
+  base::win::ScopedComPtr<IMFMediaBuffer> buffer;
+  hr = sample->GetBufferByIndex(0, buffer.Receive());
+  if (FAILED(hr)) {
+    NOTREACHED() << "Failed to get buffer in sample";
+    return NULL;
+  }
+  DWORD max_length = 0, current_length = 0;
+  uint8* destination = NULL;
+  hr = buffer->Lock(&destination, &max_length, &current_length);
+  if (FAILED(hr)) {
+    NOTREACHED() << "Failed to lock buffer";
+    return NULL;
+  }
+  CHECK_EQ(current_length, 0u);
+  CHECK_GE(static_cast<int>(max_length), size);
+  memcpy(destination, stream, size);
+  CHECK(SUCCEEDED(buffer->Unlock()));
+  hr = buffer->SetCurrentLength(size);
+  if (FAILED(hr)) {
+    NOTREACHED() << "Failed to set current length to " << size;
+    return NULL;
+  }
+  hr = sample->SetUINT32(MFSampleExtension_CleanPoint, TRUE);
+  if (FAILED(hr)) {
+    NOTREACHED() << "Failed to mark sample as key sample";
+    return NULL;
+  }
+  return sample.Detach();
+}
+
+}  // namespace
+
+DXVAVideoDecodeAccelerator::DXVAVideoDecodeAccelerator(
+    media::VideoDecodeAccelerator::Client* client,
+    base::ProcessHandle renderer_process)
+    : client_(client),
+      message_loop_(MessageLoop::current()),
+      surface_width_(0),
+      surface_height_(0),
+      state_(kUninitialized),
+      input_stream_info_(),
+      output_stream_info_(),
+      pictures_requested_(false),
+      renderer_process_(renderer_process) {
+  input_buffer_frame_times_.reserve(kNumInputs);
+
+#if !defined(NDEBUG)
+  decode_start_time_ = 0;
+  inputs_before_decode_ = 0;
+#endif  // !defined(NDEBUG)
+}
+
+DXVAVideoDecodeAccelerator::~DXVAVideoDecodeAccelerator() {
+  DCHECK_EQ(message_loop_, MessageLoop::current());
+  client_ = NULL;
+  message_loop_ = NULL;
+}
+
+bool DXVAVideoDecodeAccelerator::Initialize(Profile profile) {
+  DCHECK_EQ(message_loop_, MessageLoop::current());
+
+  if (state_ != kUninitialized) {
+    NOTREACHED() << "Initialize: invalid state: "
+                 << state_;
+    return false;
+  }
+
+  HRESULT hr = MFStartup(MF_VERSION, MFSTARTUP_FULL);
+  if (FAILED(hr)) {
+    NOTREACHED() << "MFStartup failed. Error:"
+                 << std::hex << std::showbase << hr;
+    return false;
+  }
+  if (!CreateD3DDevManager())
+    return false;
+  if (!InitDecoder())
+    return false;
+  if (!GetStreamsInfoAndBufferReqs())
+    return false;
+  if (SendMFTMessage(MFT_MESSAGE_NOTIFY_BEGIN_STREAMING, 0)) {
+    state_ = DXVAVideoDecodeAccelerator::kNormal;
+    client_->NotifyInitializeDone();
+    return true;
+  }
+  return false;
+}
+
+void DXVAVideoDecodeAccelerator::Decode(
+    const media::BitstreamBuffer& bitstream_buffer) {
+  DCHECK_EQ(message_loop_, MessageLoop::current());
+  if (state_ == DXVAVideoDecodeAccelerator::kUninitialized) {
+    NOTREACHED() << "ConsumeVideoSample: invalid state";
+    return;
+  }
+
+  HANDLE shared_memory_handle = NULL;
+  if (!::DuplicateHandle(renderer_process_,
+                         bitstream_buffer.handle(),
+                         ::GetCurrentProcess(),
+                         &shared_memory_handle,
+                         0,
+                         FALSE,
+                         DUPLICATE_SAME_ACCESS)) {
+    NOTREACHED() << "Failed to open duplicate shared mem handle";
+    return;
+  }
+
+  base::SharedMemory shm(shared_memory_handle, true);
+  if (!shm.Map(bitstream_buffer.size())) {
+    NOTREACHED() << "Failed in SharedMemory::Map()";
+    return;
+  }
+
+  base::Time::Exploded exploded;
+  base::Time::Now().LocalExplode(&exploded);
+  base::win::ScopedComPtr<IMFSample> sample;
+  sample.Attach(CreateInputSample(reinterpret_cast<const uint8*>(shm.memory()),
+                                  bitstream_buffer.size(),
+                                  exploded.second * 10000000,
+                                  400000,
+                                  input_stream_info_.cbSize,
+                                  input_stream_info_.cbAlignment));
+  if (!sample.get()) {
+    NOTREACHED() << "Failed to create an input sample";
+    return;
+  } else {
+#if !defined(NDEBUG)
+    inputs_before_decode_++;
+    if (!decode_start_time_)
+      decode_start_time_ = ::GetTickCount();
+#endif  // !defined(NDEBUG)
+    SendMFTMessage(MFT_MESSAGE_NOTIFY_START_OF_STREAM, 0);
+
+    if (FAILED(decoder_->ProcessInput(0, sample.get(), 0))) {
+      NOTREACHED() << "Failed to process input";
+      return;
+    }
+  }
+  if (state_ != DXVAVideoDecodeAccelerator::kEosDrain) {
+    if (!SendMFTMessage(MFT_MESSAGE_NOTIFY_END_OF_STREAM, 0)) {
+      NOTREACHED() << "Failed to send eos message to MFT";
+    } else {
+      state_ = DXVAVideoDecodeAccelerator::kEosDrain;
+    }
+  }
+  input_buffer_frame_times_.push_back(bitstream_buffer.id());
+  DoDecode();
+  client_->NotifyEndOfBitstreamBuffer(bitstream_buffer.id());
+}
+
+void DXVAVideoDecodeAccelerator::AssignPictureBuffers(
+    const std::vector<media::PictureBuffer>& buffers) {
+  DCHECK_EQ(message_loop_, MessageLoop::current());
+  // Copy the picture buffers provided by the client to the available list,
+  // and mark these buffers as available for use.
+  for (size_t buffer_index = 0; buffer_index < buffers.size(); ++buffer_index) {
+    DXVAPictureBuffer picture_buffer;
+    picture_buffer.available = true;
+    picture_buffer.picture_buffer = buffers[buffer_index];
+
+    DCHECK(available_pictures_.find(buffers[buffer_index].id()) ==
+           available_pictures_.end());
+    available_pictures_[buffers[buffer_index].id()] = picture_buffer;
+  }
+  int buffer_index = 0;
+  PendingOutputSamples::iterator sample_index =
+      pending_output_samples_.begin();
+  HRESULT hr = E_FAIL;
+
+  ProcessPendingSamples();
+}
+
+void DXVAVideoDecodeAccelerator::ReusePictureBuffer(
+    int32 picture_buffer_id) {
+  DCHECK_EQ(message_loop_, MessageLoop::current());
+  DCHECK(available_pictures_.find(picture_buffer_id) !=
+         available_pictures_.end());
+  available_pictures_[picture_buffer_id].available = true;
+
+  PendingOutputSamples::iterator sample_index =
+      pending_output_samples_.begin();
+
+  if (sample_index != pending_output_samples_.end()) {
+    const PendingSampleInfo& sample_info = *sample_index;
+    CopyOutputSampleDataToPictureBuffer(
+        sample_info.surface,
+        available_pictures_[picture_buffer_id].picture_buffer,
+        sample_info.input_buffer_id);
+    available_pictures_[picture_buffer_id].available = false;
+    pending_output_samples_.erase(sample_index);
+  }
+}
+
+void DXVAVideoDecodeAccelerator::Flush() {
+  DCHECK_EQ(message_loop_, MessageLoop::current());
+
+  LOG(INFO) << "DXVAVideoDecodeAccelerator::Flush";
+  if (state_ == DXVAVideoDecodeAccelerator::kUninitialized) {
+    NOTREACHED() << "ConsumeVideoSample: invalid state";
+    client_->NotifyFlushDone();
+    return;
+  }
+
+  if (state_ == DXVAVideoDecodeAccelerator::kStopped) {
+    LOG(INFO) << "No data available from the decoder";
+    client_->NotifyFlushDone();
+    return;
+  }
+
+  state_ = DXVAVideoDecodeAccelerator::kEosDrain;
+  if (!SendMFTMessage(MFT_MESSAGE_COMMAND_DRAIN, 0)) {
+    LOG(WARNING) << "Failed to send drain message";
+    state_ = DXVAVideoDecodeAccelerator::kStopped;
+    client_->NotifyFlushDone();
+    return;
+  }
+
+  // As per MSDN docs after the client sends this message, it calls
+  // IMFTransform::ProcessOutput in a loop, until ProcessOutput returns the
+  // error code MF_E_TRANSFORM_NEED_MORE_INPUT. The DoDecode function sets the
+  // state to DXVAVideoDecodeAccelerator::kStopped when the decoder returns
+  // MF_E_TRANSFORM_NEED_MORE_INPUT.
+  while (state_ != DXVAVideoDecodeAccelerator::kStopped) {
+    DoDecode();
+  }
+  client_->NotifyFlushDone();
+}
+
+void DXVAVideoDecodeAccelerator::Reset() {
+  DCHECK_EQ(message_loop_, MessageLoop::current());
+  LOG(INFO) << "DXVAVideoDecodeAccelerator::Reset";
+  if (state_ != kNormal && state_ != kStopped) {
+    NOTREACHED() << "Reset: invalid state";
+    client_->NotifyResetDone();
+    return;
+  }
+
+  state_ = DXVAVideoDecodeAccelerator::kResetting;
+  if (!SendMFTMessage(MFT_MESSAGE_COMMAND_FLUSH, 0)) {
+    LOG(WARNING) << "DXVAVideoDecodeAccelerator::Flush failed to send message";
+    client_->NotifyResetDone();
+    return;
+  }
+  state_ = DXVAVideoDecodeAccelerator::kNormal;
+  client_->NotifyResetDone();
+  input_buffer_frame_times_.clear();
+}
+
+void DXVAVideoDecodeAccelerator::Destroy() {
+  DCHECK_EQ(message_loop_, MessageLoop::current());
+  OutputBuffers::iterator index;
+  for (index = available_pictures_.begin(); index != available_pictures_.end();
+       ++index) {
+    client_->DismissPictureBuffer(index->second.picture_buffer.id());
+  }
+  available_pictures_.clear();
+  pending_output_samples_.clear();
+  input_buffer_frame_times_.clear();
+}
+
+bool DXVAVideoDecodeAccelerator::CreateD3DDevManager() {
+  d3d9_.Attach(Direct3DCreate9(D3D_SDK_VERSION));

[apatrick_chromium, 2011/12/13 02:00:13]

I don't think the IDirect3D9 needs to outlive the scope of this function.

[ananta, 2011/12/13 02:51:18]

Done.

+  if (d3d9_.get() == NULL) {
+    NOTREACHED() << "Failed to create D3D9";
+    return false;
+  }
+
+  D3DPRESENT_PARAMETERS present_params = {0};
+  present_params.BackBufferWidth = 0;

[apatrick_chromium, 2011/12/13 02:00:13]

0 by 0 might not work reliably. maybe 1 by 1 would be safer.

[ananta, 2011/12/13 02:51:18]

Done.

+  present_params.BackBufferHeight = 0;
+  present_params.BackBufferFormat = D3DFMT_UNKNOWN;

[apatrick_chromium, 2011/12/13 02:00:13]

and i'd use a valid format as well.

[ananta, 2011/12/13 02:51:18]

As per msdn  D3DFMT_UNKNOWN can be specified for the BackBufferFormat while in
windowed mode. This tells the runtime to use the current display-mode format and
eliminates the need to call GetDisplayMode.

+  present_params.BackBufferCount = 1;
+  present_params.SwapEffect = D3DSWAPEFFECT_DISCARD;
+  present_params.hDeviceWindow = GetShellWindow();
+  present_params.Windowed = TRUE;
+  present_params.Flags = D3DPRESENTFLAG_VIDEO;
+  present_params.FullScreen_RefreshRateInHz = 0;
+  present_params.PresentationInterval = 0;
+
+  HRESULT hr = d3d9_->CreateDevice(D3DADAPTER_DEFAULT,

[apatrick_chromium, 2011/12/13 02:00:13]

This will need to be a D3D9Ex device if we are going to share textures with
ANGLE.

[ananta, 2011/12/13 02:51:18]

Done.

+                                   D3DDEVTYPE_HAL,
+                                   GetShellWindow(),
+                                   (D3DCREATE_HARDWARE_VERTEXPROCESSING |

[apatrick_chromium, 2011/12/13 02:00:13]

Older Intel cards do not support hardware vertex processing.

[ananta, 2011/12/13 02:51:18]

Changed to mixed.

+                                   D3DCREATE_MULTITHREADED),
+                                   &present_params,
+                                   device_.Receive());
+  if (FAILED(hr)) {
+    NOTREACHED() << "Failed to create D3D Device";
+    return false;
+  }
+
+  UINT dev_manager_reset_token = 0;
+  hr = DXVA2CreateDirect3DDeviceManager9(&dev_manager_reset_token,

[apatrick_chromium, 2011/12/13 02:00:13]

I think D3D retains a pointer to this and modifies its value every time the
device is reset. This could corrupt the stack.

[ananta, 2011/12/13 02:51:18]

Done.

+                                         device_manager_.Receive());
+  if (FAILED(hr)) {
+    NOTREACHED() << "Couldn't create D3D Device manager";
+    return false;
+  }
+
+  hr = device_manager_->ResetDevice(device_.get(),
+                                    dev_manager_reset_token);
+  if (FAILED(hr)) {
+    NOTREACHED() << "Failed to set device to device manager";
+    return false;
+  }
+  return true;
+}
+
+
+bool DXVAVideoDecodeAccelerator::InitDecoder() {
+  HRESULT hr = CoCreateInstance(__uuidof(CMSH264DecoderMFT),
+                                NULL,
+                                CLSCTX_INPROC_SERVER,
+                                __uuidof(IMFTransform),
+                                reinterpret_cast<void**>(decoder_.Receive()));
+  if (FAILED(hr) || !decoder_.get()) {
+    NOTREACHED() << "CoCreateInstance failed "
+                 << std::hex << std::showbase << hr;
+    return false;
+  }
+
+  if (!CheckDecoderDxvaSupport())
+    return false;
+  hr = decoder_->ProcessMessage(
+            MFT_MESSAGE_SET_D3D_MANAGER,
+            reinterpret_cast<ULONG_PTR>(device_manager_.get()));
+  if (FAILED(hr)) {
+    NOTREACHED() << "Failed to pass D3D9 device to decoder "
+                 << std::hex << hr;
+    return false;
+  }
+  return SetDecoderMediaTypes();
+}
+
+bool DXVAVideoDecodeAccelerator::CheckDecoderDxvaSupport() {
+  base::win::ScopedComPtr<IMFAttributes> attributes;
+  HRESULT hr = decoder_->GetAttributes(attributes.Receive());
+  if (FAILED(hr)) {
+    NOTREACHED() << "Unlock: Failed to get attributes, hr = "
+                 << std::hex << std::showbase << hr;
+    return false;
+  }
+  UINT32 dxva = 0;
+  hr = attributes->GetUINT32(MF_SA_D3D_AWARE, &dxva);
+  if (FAILED(hr) || !dxva) {
+    NOTREACHED() << "Failed to get DXVA attr. Error:"
+                 << std::hex << std::showbase << hr
+                 << " .This might not be the right decoder.";
+    return false;
+  }
+
+  hr = attributes->SetUINT32(CODECAPI_AVDecVideoAcceleration_H264, TRUE);
+  DCHECK(SUCCEEDED(hr));
+  return true;
+}
+
+bool DXVAVideoDecodeAccelerator::SetDecoderMediaTypes() {
+  MFT_REGISTER_TYPE_INFO* input_types = NULL;
+  MFT_REGISTER_TYPE_INFO* output_types = NULL;
+  uint32 num_inputs = 0;
+  uint32 num_outputs = 0;
+  HRESULT hr = MFTGetInfo(__uuidof(CMSH264DecoderMFT), NULL, &input_types,
+                          &num_inputs, &output_types, &num_outputs, NULL);
+  if (!SetDecoderInputMediaType())
+    return false;
+  return SetDecoderOutputMediaType(MFVideoFormat_NV12);
+}
+
+bool DXVAVideoDecodeAccelerator::SetDecoderInputMediaType() {
+  base::win::ScopedComPtr<IMFMediaType> media_type;
+  HRESULT hr = MFCreateMediaType(media_type.Receive());
+  if (FAILED(hr)) {
+    NOTREACHED() << "Failed to create empty media type object";
+    return false;
+  }
+
+  hr = media_type->SetGUID(MF_MT_MAJOR_TYPE, MFMediaType_Video);
+  if (FAILED(hr)) {
+    NOTREACHED() << "SetGUID for major type failed";
+    return false;
+  }
+
+  hr = media_type->SetGUID(MF_MT_SUBTYPE, MFVideoFormat_H264);
+  if (FAILED(hr)) {
+    NOTREACHED() << "SetGUID for subtype failed";
+    return false;
+  }
+
+  hr = decoder_->SetInputType(0, media_type.get(), 0);  // No flags
+  if (FAILED(hr)) {
+    NOTREACHED() << "Failed to set decoder's input type";
+    return false;
+  }
+  return true;
+}
+
+bool DXVAVideoDecodeAccelerator::SetDecoderOutputMediaType(
+    const GUID& subtype) {
+  DWORD i = 0;
+  IMFMediaType* out_media_type = NULL;

[apatrick_chromium, 2011/12/13 02:00:13]

nit: Why can't this be ScopedComPtr?

[ananta, 2011/12/13 02:51:18]

Done.

+  bool found = false;
+  while (SUCCEEDED(decoder_->GetOutputAvailableType(0, i, &out_media_type))) {
+    GUID out_subtype = {0};
+    HRESULT hr = out_media_type->GetGUID(MF_MT_SUBTYPE, &out_subtype);
+    if (FAILED(hr)) {
+      NOTREACHED() << "Failed to GetGUID() on GetOutputAvailableType() "
+                   << i;
+      out_media_type->Release();
+      continue;
+    }
+    if (out_subtype == subtype) {
+      hr = decoder_->SetOutputType(0, out_media_type, 0);  // No flags
+      hr = MFGetAttributeSize(out_media_type, MF_MT_FRAME_SIZE,
+              reinterpret_cast<UINT32*>(&surface_width_),
+              reinterpret_cast<UINT32*>(&surface_height_));
+      if (FAILED(hr)) {
+        NOTREACHED() << "Failed to SetOutputType to |subtype| or obtain "
+                     << "width/height " << std::hex << hr;
+      }
+      out_media_type->Release();
+      return true;
+    }
+    i++;
+    out_media_type->Release();
+  }
+  return false;
+}
+
+bool DXVAVideoDecodeAccelerator::SendMFTMessage(MFT_MESSAGE_TYPE msg,
+                                                int32 param) {
+  HRESULT hr = decoder_->ProcessMessage(msg, param);
+  return SUCCEEDED(hr);
+}
+
+// Gets the minimum buffer sizes for input and output samples.
+// The MFT will not allocate buffer for neither input nor output, so we have

[apatrick_chromium, 2011/12/13 02:00:13]

nit: double negative

[ananta, 2011/12/13 02:51:18]

Done.

+// to do it ourselves and make sure they're the correct size.
+// Exception is when dxva is enabled, the decoder will allocate output.
+bool DXVAVideoDecodeAccelerator::GetStreamsInfoAndBufferReqs() {
+  HRESULT hr = decoder_->GetInputStreamInfo(0, &input_stream_info_);
+  if (FAILED(hr)) {
+    LOG(ERROR) << "Failed to get input stream info";
+    return false;
+  }
+  LOG(INFO) << "Input stream info: ";

[apatrick_chromium, 2011/12/13 02:00:13]

would VLOG be more appropriate?

[ananta, 2011/12/13 02:51:18]

Done.

+  LOG(INFO) << "Max latency: " << input_stream_info_.hnsMaxLatency;
+  // There should be three flags, one for requiring a whole frame be in a
+  // single sample, one for requiring there be one buffer only in a single
+  // sample, and one that specifies a fixed sample size. (as in cbSize)
+  LOG(INFO) << "Flags: "
+            << std::hex << std::showbase << input_stream_info_.dwFlags;
+  CHECK_EQ(input_stream_info_.dwFlags, 0x7u);
+  LOG(INFO) << "Min buffer size: " << input_stream_info_.cbSize;
+  LOG(INFO) << "Max lookahead: " << input_stream_info_.cbMaxLookahead;
+  LOG(INFO) << "Alignment: " << input_stream_info_.cbAlignment;
+
+  hr = decoder_->GetOutputStreamInfo(0, &output_stream_info_);
+  if (FAILED(hr)) {
+    LOG(ERROR) << "Failed to get output stream info";
+    return false;
+  }
+  LOG(INFO) << "Output stream info: ";
+  // The flags here should be the same and mean the same thing, except when
+  // DXVA is enabled, there is an extra 0x100 flag meaning decoder will
+  // allocate its own sample.
+  LOG(INFO) << "Flags: "
+           << std::hex << std::showbase << output_stream_info_.dwFlags;
+  CHECK_EQ(output_stream_info_.dwFlags, 0x107u);
+  LOG(INFO) << "Min buffer size: " << output_stream_info_.cbSize;
+  LOG(INFO) << "Alignment: " << output_stream_info_.cbAlignment;
+  return true;
+}
+
+bool DXVAVideoDecodeAccelerator::DoDecode() {
+  if (state_ != kNormal && state_ != kEosDrain) {
+    NOTREACHED() << "DoDecode: not in normal or drain state";
+    return false;
+  }
+  // scoped_refptr<VideoFrame> frame;

[apatrick_chromium, 2011/12/13 02:00:13]

nit: is this old code that can be deleted?

[ananta, 2011/12/13 02:51:18]

Done.

+  base::win::ScopedComPtr<IMFSample> output_sample;
+  MFT_OUTPUT_DATA_BUFFER output_data_buffer;
+  DWORD status = 0;
+
+  memset(&output_data_buffer, 0, sizeof(output_data_buffer));
+  output_data_buffer.pSample = output_sample;
+  status = 0;
+
+  HRESULT hr = decoder_->ProcessOutput(0,  // No flags
+                                       1,  // # of out streams to pull from
+                                       &output_data_buffer,
+                                       &status);
+  IMFCollection* events = output_data_buffer.pEvents;
+  if (events != NULL) {
+    LOG(INFO) << "Got events from ProcessOuput, but discarding";
+    events->Release();
+  }
+  if (FAILED(hr)) {
+    if (hr == MF_E_TRANSFORM_STREAM_CHANGE) {
+      hr = SetDecoderOutputMediaType(MFVideoFormat_NV12);
+      if (SUCCEEDED(hr)) {
+        return true;
+      } else {
+        NOTREACHED() << "Failed to set decoder output media type";
+        return false;
+      }
+    } else if (hr == MF_E_TRANSFORM_NEED_MORE_INPUT) {
+      // No more output from the decoder. Notify EOS and stop playback.
+      state_ = DXVAVideoDecodeAccelerator::kStopped;
+      return false;
+    } else {
+      NOTREACHED() << "Unhandled error in DoDecode()";
+      state_ = DXVAVideoDecodeAccelerator::kStopped;
+      return false;
+    }
+  }
+
+#if !defined(NDEBUG)
+  DLOG(INFO) << "Number of input packets before successful decode: "
+             << inputs_before_decode_;
+  inputs_before_decode_ = 0;
+  uint32 end_decode = GetTickCount();
+  DLOG(INFO) << "Total time for decode: "
+             << end_decode - decode_start_time_;
+  decode_start_time_ = 0;
+#endif  // !defined(NDEBUG)
+  if (!ProcessOutputSample(output_data_buffer.pSample)) {
+    state_ = DXVAVideoDecodeAccelerator::kStopped;
+    return false;
+  }
+
+  input_buffer_frame_times_.clear();
+  state_ = DXVAVideoDecodeAccelerator::kNormal;
+  return true;
+}
+
+bool DXVAVideoDecodeAccelerator::ProcessOutputSample(IMFSample* sample) {
+  if (!sample) {
+    NOTREACHED() << "ProcessOutput succeeded, but did not get a sample back";
+    return false;
+  }
+  base::win::ScopedComPtr<IMFSample> output_sample;
+  output_sample.Attach(sample);
+
+  if (!input_buffer_frame_times_.size()) {
+    DLOG(INFO) << "In input buffers left to process output";
+    return false;
+  }
+  int32 input_buffer_id =
+      input_buffer_frame_times_[input_buffer_frame_times_.size() - 1];
+  input_buffer_frame_times_.clear();
+
+  base::win::ScopedComPtr<IMFMediaBuffer> output_buffer;
+  HRESULT hr = sample->GetBufferByIndex(0, output_buffer.Receive());
+  if (FAILED(hr)) {
+    NOTREACHED() << "Failed to get buffer from sample";
+    return false;
+  }
+  base::win::ScopedComPtr<IDirect3DSurface9> surface;
+  hr = MFGetService(output_buffer, MR_BUFFER_SERVICE,
+                    IID_PPV_ARGS(surface.Receive()));
+  if (FAILED(hr)) {
+    NOTREACHED() << "Failed to get surface from buffer";
+    return false;
+  }
+
+  D3DSURFACE_DESC surface_desc;
+  hr = surface->GetDesc(&surface_desc);
+  if (FAILED(hr)) {
+    NOTREACHED() << "Failed to get surface description";
+    return false;
+  }
+
+#if !defined(NDEBUG)
+  DWORD start_surface_time = GetTickCount();
+#endif  // !defined(NDEBUG)
+
+  // TODO(ananta)
+  // The code below may not be necessary once we have an ANGLE extension which
+  // allows us to pass the Direct 3D surface directly for rendering.
+
+  // The decoded bits in the source direct 3d surface are in the YUV
+  // format. Angle does not support that. As a workaround we create an
+  // offscreen surface in the RGB format and copy the source surface
+  // to this surface.
+  base::win::ScopedComPtr<IDirect3DSurface9> dest_surface;
+  hr = device_->CreateOffscreenPlainSurface(surface_desc.Width,
+                                            surface_desc.Height,
+                                            D3DFMT_A8R8G8B8,
+                                            D3DPOOL_DEFAULT,
+                                            dest_surface.Receive(),
+                                            NULL);
+  if (FAILED(hr)) {
+    NOTREACHED() << "Failed to create offscreen surface";
+    return false;
+  }
+
+  hr = D3DXLoadSurfaceFromSurface(dest_surface, NULL, NULL, surface, NULL,
+                                  NULL, 0, 0);
+  if (FAILED(hr)) {
+    NOTREACHED() << "Failed to copy source surface to dest.";
+    return false;
+  }
+
+#if !defined(NDEBUG)
+  DWORD end_surface_time = GetTickCount();
+  DLOG(INFO) << "Time to create and copy new surface is "
+             << end_surface_time - start_surface_time;
+#endif  // !defined(NDEBUG)
+
+  PendingSampleInfo sample_info;
+
+  sample_info.input_buffer_id = input_buffer_id;
+  sample_info.surface = dest_surface;
+  pending_output_samples_.push_back(sample_info);
+
+  // If we have available picture buffers to copy the output data then use the
+  // first one and then flag it as not being available for use.
+  if (available_pictures_.size()) {
+    ProcessPendingSamples();
+    return true;
+  }
+
+  if (pictures_requested_) {
+    DLOG(INFO) << "Waiting for picture slots from the client.";
+    return true;
+  }
+  // Go ahead and request picture buffers.
+  client_->ProvidePictureBuffers(
+      kNumPictureBuffers, gfx::Size(surface_desc.Width, surface_desc.Height));
+  pictures_requested_ = true;
+  return true;
+}
+
+bool DXVAVideoDecodeAccelerator::CopyOutputSampleDataToPictureBuffer(
+    IDirect3DSurface9* dest_surface, media::PictureBuffer picture_buffer,
+    int32 input_buffer_id) {
+  DCHECK(dest_surface);
+
+  // Get the currently loaded bitmap from the DC.
+  HDC hdc = NULL;
+
+  HRESULT hr = dest_surface->GetDC(&hdc);
+  if (FAILED(hr)) {
+    NOTREACHED() << "Failed to get HDC for dest offscreen surface";
+    return false;
+  }
+
+  HBITMAP bitmap =
+      reinterpret_cast<HBITMAP>(GetCurrentObject(hdc, OBJ_BITMAP));
+  if (!bitmap) {
+    NOTREACHED() << "Failed to get bitmap from DC";
+    return false;
+  }
+  // TODO(ananta)
+  // The code below may not be necessary once we have an ANGLE extension which
+  // allows us to pass the Direct 3D surface directly for rendering.
+
+  // The Device dependent bitmap is upside down for OpenGL. We convert the
+  // bitmap to a DIB and render it on the texture instead.
+  BITMAP bitmap_basic_info = {0};
+  GetObject(bitmap, sizeof(BITMAP), &bitmap_basic_info);
+
+  BITMAPINFO bitmap_info = {0};
+  bitmap_info.bmiHeader.biSize = sizeof(BITMAPINFOHEADER);
+  bitmap_info.bmiHeader.biWidth = bitmap_basic_info.bmWidth;
+  bitmap_info.bmiHeader.biHeight = bitmap_basic_info.bmHeight;
+  bitmap_info.bmiHeader.biPlanes = 1;
+  bitmap_info.bmiHeader.biBitCount = bitmap_basic_info.bmBitsPixel;
+  bitmap_info.bmiHeader.biCompression = BI_RGB;
+  bitmap_info.bmiHeader.biSizeImage = 0;
+  bitmap_info.bmiHeader.biClrUsed = 0;
+
+  int ret = GetDIBits(hdc, bitmap, 0, 0, NULL, &bitmap_info, DIB_RGB_COLORS);
+  if (bitmap_info.bmiHeader.biSizeImage <= 0) {
+    NOTREACHED() << "Failed to read bitmap size";
+    return false;
+  }
+  scoped_ptr<char> bits(new char[bitmap_info.bmiHeader.biSizeImage]);
+  ret = GetDIBits(hdc, bitmap, 0, bitmap_basic_info.bmHeight, bits.get(),
+                  &bitmap_info, DIB_RGB_COLORS);
+  DCHECK(ret > 0);
+
+  D3DSURFACE_DESC surface_desc;
+  hr = dest_surface->GetDesc(&surface_desc);
+  DCHECK(SUCCEEDED(hr));
+
+  glBindTexture(GL_TEXTURE_2D, picture_buffer.texture_id());
+  glTexImage2D(GL_TEXTURE_2D, 0, GL_BGRA_EXT, surface_desc.Width,
+               surface_desc.Height, 0, GL_BGRA_EXT, GL_UNSIGNED_BYTE,
+               (GLvoid*) bits.get());
+  glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+  glBindTexture(GL_TEXTURE_2D, 0);
+  dest_surface->ReleaseDC(hdc);
+
+  media::Picture output_picture(picture_buffer.id(), input_buffer_id);
+  client_->PictureReady(output_picture);
+  return true;
+}
+
+void DXVAVideoDecodeAccelerator::ProcessPendingSamples() {
+  PendingOutputSamples::iterator sample_index =
+      pending_output_samples_.begin();
+  HRESULT hr = E_FAIL;
+
+  while (sample_index != pending_output_samples_.end()) {
+    const PendingSampleInfo& sample_info = *sample_index;
+    OutputBuffers::iterator index;
+    for (index = available_pictures_.begin();
+         index != available_pictures_.end();
+         ++index) {
+      if (index->second.available) {
+        CopyOutputSampleDataToPictureBuffer(sample_info.surface,
+                                            index->second.picture_buffer,
+                                            sample_info.input_buffer_id);
+        index->second.available = false;
+        sample_index = pending_output_samples_.erase(sample_index);
+        break;
+      }
+    }
+    if (index == available_pictures_.end()) {
+      DLOG(INFO) << "No available picture slots for output";
+      break;
+    }
+  }
+}