Add ArcGpuVideoDecodeAccelerator.

chrome/gpu/arc_gpu_video_decode_accelerator.cc

+// Copyright 2016 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 "base/callback_helpers.h"
+#include "base/logging.h"
+#include "base/run_loop.h"
+#include "chrome/gpu/arc_gpu_video_decode_accelerator.h"
+
+// TODO use Pawel's factory instead
+#include "content/public/common/create_vda.h"
+
+#undef DVLOG
+#define DVLOG VLOG
+
+namespace chromeos {
+namespace arc {
+
+ArcGpuVideoDecodeAccelerator::InputRecord::InputRecord(
+    int32_t bitstream_buffer_id,
+    uint32_t buffer_index,
+    int64_t timestamp)
+    : bitstream_buffer_id(bitstream_buffer_id),
+      buffer_index(buffer_index),
+      timestamp(timestamp) {}
+
+ArcGpuVideoDecodeAccelerator::InputBufferInfo::InputBufferInfo()
+    : offset(0), length(0) {}
+
+ArcGpuVideoDecodeAccelerator::InputBufferInfo::InputBufferInfo(
+    InputBufferInfo&& a) {
+  handle = std::move(a.handle);

[kcwu, 2016/03/09 07:58:21]

ArcGpuVideoDecodeAccelerator::InputBufferInfo::InputBufferInfo(
    InputBufferInfo&& a) = default;

[Owen Lin, 2016/03/14 08:46:59]

Removed.

+  offset = a.offset;
+  length = a.length;
+}
+
+ArcGpuVideoDecodeAccelerator::InputBufferInfo::~InputBufferInfo() {}
+
+ArcGpuVideoDecodeAccelerator::ArcGpuVideoDecodeAccelerator(
+    const scoped_refptr<base::SingleThreadTaskRunner>& io_task_runner)
+    : pending_eos_output_buffer_(false),
+      arc_client_(nullptr),
+      next_bitstream_buffer_id_(0),
+      io_task_runner_(io_task_runner),
+      output_buffer_size_(0) {}
+
+ArcGpuVideoDecodeAccelerator::~ArcGpuVideoDecodeAccelerator() {}
+
+static const size_t MAX_BUFFER_COUNT = 128;

[kcwu, 2016/03/09 07:58:21]

why 128? 
Is the same reason as kMaxNumberOfInputRecords?

[kcwu, 2016/03/10 07:27:26]

and s/MAX_BUFFER_COUNT/kMaxBufferCount/

[Owen Lin, 2016/03/14 08:46:59]

Done.

[Owen Lin, 2016/03/14 08:46:59]

Not the exactly same purpose but it is also arbitrary chosen. Comments are
added.

+
+bool ArcGpuVideoDecodeAccelerator::Initialize(
+    const Config& config,
+    ArcVideoAccelerator::Client* client) {
+  DVLOG(5) << "Initialize(device=" << config.device_type
+           << ", input_pixel_format=" << config.input_pixel_format
+           << ", num_input_buffers=" << config.num_input_buffers << ")";
+  DCHECK(thread_checker_.CalledOnValidThread());
+  if (config.device_type != DEVICE_DECODER)
+    return false;
+  DCHECK(client);
+  DCHECK(!arc_client_);
+  arc_client_ = client;
+
+  vda_.reset(content::CreateVDA(AsWeakPtr(), io_task_runner_));

[kcwu, 2016/03/09 07:58:21]

base::ThreadTaskRunnerHandle::Get().
Since you expect the same thread according to your thread_checker_ checks.

[Owen Lin, 2016/03/14 08:46:59]

io_task_runner_ is no longer required. Removed.

+  if (!vda_) {
+    DLOG(ERROR) << "Failed to create VDA.";
+    return false;
+  }
+
+  if (config.num_input_buffers > MAX_BUFFER_COUNT) {
+    DLOG(ERROR) << "Request too many buffers: " << config.num_input_buffers;
+    return false;
+  }
+  input_buffer_info_.resize(config.num_input_buffers);
+
+  media::VideoDecodeAccelerator::Config vda_config;
+  switch (config.input_pixel_format) {
+    case HAL_PIXEL_FORMAT_H264:
+      vda_config.profile = media::H264PROFILE_MAIN;
+      break;
+    case HAL_PIXEL_FORMAT_VP8:
+      vda_config.profile = media::VP8PROFILE_ANY;
+      break;
+    default:
+      DLOG(ERROR) << "Unsupported input format: " << config.input_pixel_format;
+      return false;
+  }
+  vda_config.output_mode =
+      media::VideoDecodeAccelerator::Config::OutputMode::IMPORT;
+  if (!vda_->Initialize(vda_config, this)) {
+    vda_.reset();
+    DLOG(ERROR) << "VDA::Initialize() failed.";
+    return false;
+  }
+  return true;
+}
+
+void ArcGpuVideoDecodeAccelerator::SetNumberOfOutputBuffers(size_t number) {
+  DVLOG(5) << "SetNumberOfOutputBuffers(" << number << ")";
+  DCHECK(thread_checker_.CalledOnValidThread());
+
+  if (number >= MAX_BUFFER_COUNT) {

[Owen Lin, 2016/03/14 08:46:59]

Done.

+    DLOG(ERROR) << "Too many buffers: " << number;
+    arc_client_->OnError(INVALID_ARGUMENT);
+    return;
+  }
+
+  std::vector<media::PictureBuffer> buffers;
+  for (int32_t id = 0, n = number; id < n; ++id) {
+    // TODO: Make sure the |coded_size| is what we want.
+    buffers.push_back(media::PictureBuffer(id, coded_size_, 0));
+  }
+  vda_->AssignPictureBuffers(buffers);
+
+  pending_import_buffer_.clear();
+  pending_import_buffer_.resize(number);
+}
+
+void ArcGpuVideoDecodeAccelerator::BindSharedMemory(PortType port,
+                                                    uint32_t index,
+                                                    int ashmem_fd,
+                                                    off_t offset,
+                                                    size_t length) {
+  DVLOG(5) << "ArcGVDA::BindSharedMemory, offset: " << offset
+           << ", length: " << length;
+  DCHECK(thread_checker_.CalledOnValidThread());
+
+  // Make sure we will close the file descriptor.
+  base::ScopedFD handle(ashmem_fd);
+  if (!ValidatePortAndIndex(port, index)) {
+    arc_client_->OnError(INVALID_ARGUMENT);
+    return;
+  }
+  if (port != PORT_INPUT) {

[kcwu, 2016/03/09 07:58:21]

How about move this check to line 130 (before ValidatePortAndIndex)?

[Owen Lin, 2016/03/14 08:46:59]

Done.

+    DLOG(ERROR) << "SharedBuffer is only supported for input";
+    arc_client_->OnError(INVALID_ARGUMENT);
+    return;
+  }
+  InputBufferInfo* input_info = &input_buffer_info_[index];
+  input_info->handle = std::move(handle);
+  input_info->offset = offset;
+  input_info->length = length;
+}
+
+void ArcGpuVideoDecodeAccelerator::BindDmabuf(PortType port,
+                                              uint32_t index,
+                                              int dmabuf_fd) {
+  DCHECK(thread_checker_.CalledOnValidThread());
+
+  // Make sure we will close the file descriptor.
+  base::ScopedFD handle(dmabuf_fd);
+  if (!ValidatePortAndIndex(port, index)) {
+    arc_client_->OnError(INVALID_ARGUMENT);
+    return;
+  }
+  if (port != PORT_OUTPUT) {

[kcwu, 2016/03/09 07:58:21]

How about move this check to line 152 (before ValidatePortAndIndex)?

[Owen Lin, 2016/03/14 08:46:59]

Done.

+    DLOG(ERROR) << "GraphicBuffer is only supported for input";
+    arc_client_->OnError(INVALID_ARGUMENT);
+    return;
+  }
+  pending_import_buffer_[index] = std::move(handle);
+}
+
+void ArcGpuVideoDecodeAccelerator::UseBuffer(PortType port,
+                                             uint32_t index,
+                                             const BufferMetadata& metadata) {
+  DVLOG(5) << "UseBuffer(port=" << port << ", index=" << index
+           << ", metadata=(bytes_used=" << metadata.bytes_used
+           << ", timestamp=" << metadata.timestamp << "))";

[kcwu, 2016/03/07 14:04:33]

extra )

[Owen Lin, 2016/03/14 08:46:59]

Done.

+  DCHECK(thread_checker_.CalledOnValidThread());
+  if (!ValidatePortAndIndex(port, index)) {
+    arc_client_->OnError(INVALID_ARGUMENT);
+    return;
+  }
+  switch (port) {
+    case PORT_INPUT: {
+      InputBufferInfo* input_info = &input_buffer_info_[index];
+      if (metadata.flags & BUFFER_FLAG_EOS) {
+        // TODO(owenlin): Ask VDA to return all output pictures so that we
+        // can output an EOS picture when Flush() is done.
+        vda_->Flush();

[kcwu, 2016/03/09 07:58:21]

I'm not sure. Should this move to line 196 or return here? Decode() after
Flush() looks weird to me.

[Owen Lin, 2016/03/14 08:46:59]

Move to line 196. We need to send the content to be decoded first and then call
flush.

+      }
+      int32_t bitstream_buffer_id = next_bitstream_buffer_id_;
+      // Mask against 30 bits, to avoid (undefined wraparound on signed integer)

[kcwu, 2016/03/09 07:58:21]

s/(undefined wraparound on signed integer)/(undefined) wraparound on signed
integer./

[Owen Lin, 2016/03/14 08:46:59]

Done.

+      next_bitstream_buffer_id_ = (next_bitstream_buffer_id_ + 1) & 0x3FFFFFFF;
+      SetInputRecord(bitstream_buffer_id, index, metadata.timestamp);
+      int dup_fd = HANDLE_EINTR(dup(input_info->handle.get()));
+      if (dup_fd < 0) {
+        DLOG(ERROR) << "dup() failed.";
+        arc_client_->OnError(PLATFORM_FAILURE);
+        return;
+      }
+      vda_->Decode(media::BitstreamBuffer(
+          bitstream_buffer_id, base::SharedMemoryHandle(dup_fd, true),
+          metadata.bytes_used, input_info->offset));
+      break;
+    }
+    case PORT_OUTPUT: {
+      SendEosIfNeededOrReusePicture(index);
+      break;
+    }
+    default:
+      NOTREACHED();
+  }
+}
+
+void ArcGpuVideoDecodeAccelerator::Reset() {
+  DCHECK(thread_checker_.CalledOnValidThread());
+  base::RunLoop loop;
+  reset_done_callback_ = loop.QuitClosure();
+  DCHECK(vda_);
+  vda_->Reset();
+  base::MessageLoop::ScopedNestableTaskAllower allow(
+      base::MessageLoop::current());
+  // Wait for the ResetDone callback.
+  loop.Run();
+}
+
+void ArcGpuVideoDecodeAccelerator::ProvidePictureBuffers(
+    size_t requested_num_of_buffers,
+    const gfx::Size& dimensions,
+    uint32_t texture_target) {
+  DVLOG(5) << "ProvidePictureBuffers("
+           << "requested_num_of_buffers=" << requested_num_of_buffers
+           << ", dimensions=" << dimensions.ToString() << ")";
+  DCHECK(thread_checker_.CalledOnValidThread());
+  coded_size_ = dimensions;
+  // TODO(owenlin): use VDA::GetOutputFormat() here and calculate correct
+  //                |image_size|.
+  VideoFormat video_format;
+  video_format.image_size = dimensions.GetArea() * 3 / 2;
+  output_buffer_size_ = video_format.image_size;
+  video_format.min_num_buffers = requested_num_of_buffers;
+  video_format.coded_width = dimensions.width();
+  video_format.coded_height = dimensions.height();
+  // TODO(owenlin): How to get visible size?
+  video_format.crop_top = 0;
+  video_format.crop_left = 0;
+  video_format.crop_width = dimensions.width();
+  video_format.crop_height = dimensions.height();
+  arc_client_->OnOutputFormatChanged(video_format);
+}
+
+void ArcGpuVideoDecodeAccelerator::DismissPictureBuffer(
+    int32_t picture_buffer) {
+  // no-op

[kcwu, 2016/03/09 07:58:21]

why can it be no-op?

[Owen Lin, 2016/03/14 08:46:59]

VDA has its owned FDs of the picture buffers and will close those FD after use.
The picture buffer on ARC is managed by the native window. We don't need to do
anything here.

+}
+
+void ArcGpuVideoDecodeAccelerator::PictureReady(const media::Picture& picture) {
+  DVLOG(5) << "PictureReady(picture_buffer_id=" << picture.picture_buffer_id()
+           << ", bitstream_buffer_id=" << picture.bitstream_buffer_id();
+  DCHECK(thread_checker_.CalledOnValidThread());
+
+  // Empty buffer, returned in Flushing.
+  if (picture.bitstream_buffer_id() == -1) {

[kcwu, 2016/03/09 07:58:21]

where is "-1" coming from?
I don't find it in this file and neither mentioned in ava.h and vda.h

[Owen Lin, 2016/03/14 08:46:59]

It would be defined in the new Flush(bool return_buffers); function.

+    buffers_pending_eos_.push(picture.picture_buffer_id());
+    return;
+  }
+  InputRecord* input_record = FindInputRecord(picture.bitstream_buffer_id());
+  if (input_record == nullptr) {
+    DLOG(ERROR) << "Cannot find for bitstream buffer id: "
+                << picture.bitstream_buffer_id();
+    arc_client_->OnError(PLATFORM_FAILURE);
+    return;
+  }
+
+  BufferMetadata metadata;
+  metadata.timestamp = input_record->timestamp;
+  metadata.bytes_used = output_buffer_size_;
+  arc_client_->OnBufferDone(PORT_OUTPUT, picture.picture_buffer_id(), metadata);
+}
+
+void ArcGpuVideoDecodeAccelerator::NotifyEndOfBitstreamBuffer(
+    int32_t bitstream_buffer_id) {
+  DVLOG(5) << "NotifyEndOfBitstreamBuffer(" << bitstream_buffer_id << ")";
+  DCHECK(thread_checker_.CalledOnValidThread());
+  InputRecord* input_record = FindInputRecord(bitstream_buffer_id);
+  if (input_record == nullptr) {
+    arc_client_->OnError(PLATFORM_FAILURE);
+    return;
+  }
+  arc_client_->OnBufferDone(PORT_INPUT, input_record->buffer_index,
+                            BufferMetadata());
+}
+
+void ArcGpuVideoDecodeAccelerator::NotifyFlushDone() {
+  DCHECK(thread_checker_.CalledOnValidThread());
+  pending_eos_output_buffer_ = true;
+  while (!buffers_pending_eos_.empty()) {
+    SendEosIfNeededOrReusePicture(buffers_pending_eos_.front());
+    buffers_pending_eos_.pop();
+  }
+}
+
+void ArcGpuVideoDecodeAccelerator::NotifyResetDone() {
+  DCHECK(thread_checker_.CalledOnValidThread());
+  base::ResetAndReturn(&reset_done_callback_).Run();
+}
+
+static ArcVideoAccelerator::Error ConvertErrorCode(
+    media::VideoDecodeAccelerator::Error error) {
+  switch (error) {
+    case media::VideoDecodeAccelerator::ILLEGAL_STATE:
+      return ArcVideoAccelerator::ILLEGAL_STATE;
+    case media::VideoDecodeAccelerator::INVALID_ARGUMENT:
+      return ArcVideoAccelerator::INVALID_ARGUMENT;
+    case media::VideoDecodeAccelerator::UNREADABLE_INPUT:
+      return ArcVideoAccelerator::UNREADABLE_INPUT;
+    case media::VideoDecodeAccelerator::PLATFORM_FAILURE:
+      return ArcVideoAccelerator::PLATFORM_FAILURE;
+    default:
+      DLOG(ERROR) << "Unknown error: " << error;
+      return ArcVideoAccelerator::PLATFORM_FAILURE;
+  }
+}
+
+void ArcGpuVideoDecodeAccelerator::NotifyError(
+    media::VideoDecodeAccelerator::Error error) {
+  DCHECK(thread_checker_.CalledOnValidThread());
+  arc_client_->OnError(ConvertErrorCode(error));
+}
+
+void ArcGpuVideoDecodeAccelerator::SendEosIfNeededOrReusePicture(
+    uint32_t index) {
+  if (pending_eos_output_buffer_) {
+    BufferMetadata metadata;
+    metadata.flags = BUFFER_FLAG_EOS;
+    arc_client_->OnBufferDone(PORT_OUTPUT, index, metadata);
+    pending_eos_output_buffer_ = false;
+  } else {
+    if (pending_import_buffer_[index].is_valid()) {
+      std::vector<base::ScopedFD> dmabuf_fds;
+      dmabuf_fds.push_back(std::move(pending_import_buffer_[index]));
+      vda_->ImportBufferForPicture(index, std::move(dmabuf_fds));
+      DCHECK(!pending_import_buffer_[index].is_valid());
+    } else {
+      vda_->ReusePictureBuffer(index);
+    }
+  }
+}
+
+void ArcGpuVideoDecodeAccelerator::SetInputRecord(int32_t bitstream_buffer_id,
+                                                  uint32_t buffer_index,
+                                                  int64_t timestamp) {
+  input_records_.push_front(
+      InputRecord(bitstream_buffer_id, buffer_index, timestamp));
+  // The same value copied from media::GpuVideoDecoder.
+  const size_t kMaxNumberOfInputRecords = 128;
+  if (input_records_.size() > kMaxNumberOfInputRecords)
+    input_records_.pop_back();
+}
+
+ArcGpuVideoDecodeAccelerator::InputRecord*
+ArcGpuVideoDecodeAccelerator::FindInputRecord(int32_t bitstream_buffer_id) {
+  for (auto& record : input_records_) {
+    if (record.bitstream_buffer_id == bitstream_buffer_id)
+      return &record;
+  }
+  return nullptr;
+}
+
+bool ArcGpuVideoDecodeAccelerator::ValidatePortAndIndex(PortType port,
+                                                        uint32_t index) {
+  switch (port) {
+    case PORT_INPUT:
+      if (index >= input_buffer_info_.size()) {
+        DLOG(ERROR) << "Invalid index: " << index;
+        return false;
+      }
+      return true;
+    case PORT_OUTPUT:
+      if (index >= pending_import_buffer_.size()) {
+        DLOG(ERROR) << "Invalid index: " << index;
+        return false;
+      }
+      return true;
+    default:
+      DLOG(ERROR) << "Invalid port: " << port;
+      return false;
+  }
+}
+
+}  // namespace arc
+}  // namespace chromeos