Reland: Introduce GpuVideoDecodeAcceleratorFactory.

content/common/gpu/media/v4l2_slice_video_decode_accelerator.cc

 // Copyright 2015 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 <errno.h>
 #include <fcntl.h>
 #include <linux/videodev2.h>
 #include <poll.h>
 #include <string.h>
 #include <sys/eventfd.h>
 #include <sys/ioctl.h>
 #include <sys/mman.h>
 
 #include "base/bind.h"
 #include "base/bind_helpers.h"
 #include "base/callback.h"
 #include "base/callback_helpers.h"
 #include "base/command_line.h"
 #include "base/macros.h"
 #include "base/numerics/safe_conversions.h"
 #include "base/strings/stringprintf.h"
 #include "content/common/gpu/media/shared_memory_region.h"
 #include "content/common/gpu/media/v4l2_slice_video_decode_accelerator.h"
 #include "media/base/bind_to_current_loop.h"
 #include "media/base/media_switches.h"
+#include "ui/gl/gl_context.h"
 #include "ui/gl/scoped_binders.h"
 
 #define LOGF(level) LOG(level) << __FUNCTION__ << "(): "
 #define DVLOGF(level) DVLOG(level) << __FUNCTION__ << "(): "
 
 #define NOTIFY_ERROR(x)                        \
   do {                                         \
     LOG(ERROR) << "Setting error state:" << x; \
     SetErrorState(x);                          \
   } while (0)
@@ skipping 336 matching lines @@
     V4L2SliceVideoDecodeAccelerator::V4L2DecodeSurface>& dec_surface)
     : dec_surface_(dec_surface) {
 }
 
 V4L2VP8Picture::~V4L2VP8Picture() {
 }
 
 V4L2SliceVideoDecodeAccelerator::V4L2SliceVideoDecodeAccelerator(
     const scoped_refptr<V4L2Device>& device,
     EGLDisplay egl_display,
-    EGLContext egl_context,
-    const base::WeakPtr<Client>& io_client,
-    const base::Callback<bool(void)>& make_context_current,
-    const scoped_refptr<base::SingleThreadTaskRunner>& io_task_runner)
+    const GetGLContextCallback& get_gl_context_cb,
+    const MakeGLContextCurrentCallback& make_context_current_cb)
     : input_planes_count_(0),
       output_planes_count_(0),
       child_task_runner_(base::ThreadTaskRunnerHandle::Get()),
-      io_task_runner_(io_task_runner),
-      io_client_(io_client),
       device_(device),
       decoder_thread_("V4L2SliceVideoDecodeAcceleratorThread"),
       device_poll_thread_("V4L2SliceVideoDecodeAcceleratorDevicePollThread"),
       input_streamon_(false),
       input_buffer_queued_count_(0),
       output_streamon_(false),
       output_buffer_queued_count_(0),
       video_profile_(media::VIDEO_CODEC_PROFILE_UNKNOWN),
       output_format_fourcc_(0),
       state_(kUninitialized),
       decoder_flushing_(false),
       decoder_resetting_(false),
       surface_set_change_pending_(false),
       picture_clearing_count_(0),
-      make_context_current_(make_context_current),
       egl_display_(egl_display),
-      egl_context_(egl_context),
+      get_gl_context_cb_(get_gl_context_cb),
+      make_context_current_cb_(make_context_current_cb),
       weak_this_factory_(this) {
   weak_this_ = weak_this_factory_.GetWeakPtr();
 }
 
 V4L2SliceVideoDecodeAccelerator::~V4L2SliceVideoDecodeAccelerator() {
   DVLOGF(2);
 
   DCHECK(child_task_runner_->BelongsToCurrentThread());
   DCHECK(!decoder_thread_.IsRunning());
   DCHECK(!device_poll_thread_.IsRunning());
@@ skipping 15 matching lines @@
     client_ptr_factory_.reset();
   }
 }
 
 bool V4L2SliceVideoDecodeAccelerator::Initialize(const Config& config,
                                                  Client* client) {
   DVLOGF(3) << "profile: " << config.profile;
   DCHECK(child_task_runner_->BelongsToCurrentThread());
   DCHECK_EQ(state_, kUninitialized);
 
+  if (get_gl_context_cb_.is_null() || make_context_current_cb_.is_null()) {
+    NOTREACHED() << "GL callbacks are required for this VDA";
+    return false;
+  }
+
   if (config.is_encrypted) {
     NOTREACHED() << "Encrypted streams are not supported for this VDA";
     return false;
   }
 
   if (!device_->SupportsDecodeProfileForV4L2PixelFormats(
           config.profile, arraysize(supported_input_fourccs_),
           supported_input_fourccs_)) {
     DVLOGF(1) << "unsupported profile " << config.profile;
     return false;
   }
 
   client_ptr_factory_.reset(
       new base::WeakPtrFactory<VideoDecodeAccelerator::Client>(client));
   client_ = client_ptr_factory_->GetWeakPtr();
+  // If we haven't been set up to decode on separate thread via
+  // TryToSetupDecodeOnSeparateThread(), use the main thread/client for
+  // decode tasks.
+  if (!decode_task_runner_) {
+    decode_task_runner_ = child_task_runner_;
+    DCHECK(!decode_client_);
+    decode_client_ = client_;
+  }
 
   video_profile_ = config.profile;
 
   if (video_profile_ >= media::H264PROFILE_MIN &&
       video_profile_ <= media::H264PROFILE_MAX) {
     h264_accelerator_.reset(new V4L2H264Accelerator(this));
     decoder_.reset(new H264Decoder(h264_accelerator_.get()));
   } else if (video_profile_ >= media::VP8PROFILE_MIN &&
              video_profile_ <= media::VP8PROFILE_MAX) {
     vp8_accelerator_.reset(new V4L2VP8Accelerator(this));
     decoder_.reset(new VP8Decoder(vp8_accelerator_.get()));
   } else {
     NOTREACHED() << "Unsupported profile " << video_profile_;
     return false;
   }
 
   // TODO(posciak): This needs to be queried once supported.
   input_planes_count_ = 1;
   output_planes_count_ = 1;
 
   if (egl_display_ == EGL_NO_DISPLAY) {
     LOG(ERROR) << "Initialize(): could not get EGLDisplay";
     return false;
   }
 
   // We need the context to be initialized to query extensions.
-  if (!make_context_current_.Run()) {
+  if (!make_context_current_cb_.Run()) {
     LOG(ERROR) << "Initialize(): could not make context current";
     return false;
   }
 
   if (!gfx::g_driver_egl.ext.b_EGL_KHR_fence_sync) {
     LOG(ERROR) << "Initialize(): context does not have EGL_KHR_fence_sync";
     return false;
   }
 
   // Capabilities check.
@@ skipping 683 matching lines @@
     decoder_display_queue_.pop();
 
   DVLOGF(3) << "Device poll stopped";
   return true;
 }
 
 void V4L2SliceVideoDecodeAccelerator::Decode(
     const media::BitstreamBuffer& bitstream_buffer) {
   DVLOGF(3) << "input_id=" << bitstream_buffer.id()
             << ", size=" << bitstream_buffer.size();
-  DCHECK(io_task_runner_->BelongsToCurrentThread());
+  DCHECK(decode_task_runner_->BelongsToCurrentThread());
 
   if (bitstream_buffer.id() < 0) {
     LOG(ERROR) << "Invalid bitstream_buffer, id: " << bitstream_buffer.id();
     if (base::SharedMemory::IsHandleValid(bitstream_buffer.handle()))
       base::SharedMemory::CloseHandle(bitstream_buffer.handle());
     NOTIFY_ERROR(INVALID_ARGUMENT);
     return;
   }
 
   decoder_thread_task_runner_->PostTask(
       FROM_HERE, base::Bind(&V4L2SliceVideoDecodeAccelerator::DecodeTask,
                             base::Unretained(this), bitstream_buffer));
 }
 
 void V4L2SliceVideoDecodeAccelerator::DecodeTask(
     const media::BitstreamBuffer& bitstream_buffer) {
   DVLOGF(3) << "input_id=" << bitstream_buffer.id()
             << " size=" << bitstream_buffer.size();
   DCHECK(decoder_thread_task_runner_->BelongsToCurrentThread());
 
   scoped_ptr<BitstreamBufferRef> bitstream_record(new BitstreamBufferRef(
-      io_client_, io_task_runner_,
+      decode_client_, decode_task_runner_,
       new SharedMemoryRegion(bitstream_buffer, true), bitstream_buffer.id()));
   if (!bitstream_record->shm->Map()) {
     LOGF(ERROR) << "Could not map bitstream_buffer";
     NOTIFY_ERROR(UNREADABLE_INPUT);
     return;
   }
   DVLOGF(3) << "mapped at=" << bitstream_record->shm->memory();
 
   decoder_input_queue_.push(
       linked_ptr<BitstreamBufferRef>(bitstream_record.release()));
@@ skipping 260 matching lines @@
                             output_planes_count_));
 }
 
 void V4L2SliceVideoDecodeAccelerator::CreateEGLImages(
     const std::vector<media::PictureBuffer>& buffers,
     uint32_t output_format_fourcc,
     size_t output_planes_count) {
   DVLOGF(3);
   DCHECK(child_task_runner_->BelongsToCurrentThread());
 
-  if (!make_context_current_.Run()) {
-    DLOG(ERROR) << "could not make context current";
+  gfx::GLContext* gl_context = get_gl_context_cb_.Run();
+  if (!gl_context || !make_context_current_cb_.Run()) {
+    DLOG(ERROR) << "No GL context";
     NOTIFY_ERROR(PLATFORM_FAILURE);
     return;
   }
 
   gfx::ScopedTextureBinder bind_restore(GL_TEXTURE_EXTERNAL_OES, 0);
 
   std::vector<EGLImageKHR> egl_images;
   for (size_t i = 0; i < buffers.size(); ++i) {
     DCHECK_LE(1u, buffers[i].texture_ids().size());
+
     EGLImageKHR egl_image = device_->CreateEGLImage(
-        egl_display_, egl_context_, buffers[i].texture_ids()[0],
+        egl_display_, gl_context->GetHandle(), buffers[i].texture_ids()[0],
         buffers[i].size(), i, output_format_fourcc, output_planes_count);
     if (egl_image == EGL_NO_IMAGE_KHR) {
       LOGF(ERROR) << "Could not create EGLImageKHR";
       for (const auto& image_to_destroy : egl_images)
         device_->DestroyEGLImage(egl_display_, image_to_destroy);
 
       NOTIFY_ERROR(PLATFORM_FAILURE);
       return;
     }
 
@@ skipping 41 matching lines @@
   }
 
   ProcessPendingEventsIfNeeded();
 }
 
 void V4L2SliceVideoDecodeAccelerator::ReusePictureBuffer(
     int32_t picture_buffer_id) {
   DCHECK(child_task_runner_->BelongsToCurrentThread());
   DVLOGF(4) << "picture_buffer_id=" << picture_buffer_id;
 
-  if (!make_context_current_.Run()) {
+  if (!make_context_current_cb_.Run()) {
     LOGF(ERROR) << "could not make context current";
     NOTIFY_ERROR(PLATFORM_FAILURE);
     return;
   }
 
   EGLSyncKHR egl_sync =
       eglCreateSyncKHR(egl_display_, EGL_SYNC_FENCE_KHR, NULL);
   if (egl_sync == EGL_NO_SYNC_KHR) {
     LOGF(ERROR) << "eglCreateSyncKHR() failed";
     NOTIFY_ERROR(PLATFORM_FAILURE);
@@ skipping 55 matching lines @@
 
 void V4L2SliceVideoDecodeAccelerator::FlushTask() {
   DVLOGF(3);
   DCHECK(decoder_thread_task_runner_->BelongsToCurrentThread());
 
   if (!decoder_input_queue_.empty()) {
     // We are not done with pending inputs, so queue an empty buffer,
     // which - when reached - will trigger flush sequence.
     decoder_input_queue_.push(
         linked_ptr<BitstreamBufferRef>(new BitstreamBufferRef(
-            io_client_, io_task_runner_, nullptr, kFlushBufferId)));
+            decode_client_, decode_task_runner_, nullptr, kFlushBufferId)));
     return;
   }
 
   // No more inputs pending, so just finish flushing here.
   InitiateFlush();
 }
 
 void V4L2SliceVideoDecodeAccelerator::InitiateFlush() {
   DVLOGF(3);
   DCHECK(decoder_thread_task_runner_->BelongsToCurrentThread());
@@ skipping 890 matching lines @@
 }
 
 void V4L2SliceVideoDecodeAccelerator::SendPictureReady() {
   DVLOGF(3);
   DCHECK(decoder_thread_task_runner_->BelongsToCurrentThread());
   bool resetting_or_flushing = (decoder_resetting_ || decoder_flushing_);
   while (!pending_picture_ready_.empty()) {
     bool cleared = pending_picture_ready_.front().cleared;
     const media::Picture& picture = pending_picture_ready_.front().picture;
     if (cleared && picture_clearing_count_ == 0) {
-      DVLOGF(4) << "Posting picture ready to IO for: "
+      DVLOGF(4) << "Posting picture ready to decode task runner for: "
                 << picture.picture_buffer_id();
-      // This picture is cleared. Post it to IO thread to reduce latency. This
-      // should be the case after all pictures are cleared at the beginning.
-      io_task_runner_->PostTask(
-          FROM_HERE, base::Bind(&Client::PictureReady, io_client_, picture));
+      // This picture is cleared. It can be posted to a thread different than
+      // the main GPU thread to reduce latency. This should be the case after
+      // all pictures are cleared at the beginning.
+      decode_task_runner_->PostTask(
+          FROM_HERE,
+          base::Bind(&Client::PictureReady, decode_client_, picture));
       pending_picture_ready_.pop();
     } else if (!cleared || resetting_or_flushing) {
       DVLOGF(3) << "cleared=" << pending_picture_ready_.front().cleared
                 << ", decoder_resetting_=" << decoder_resetting_
                 << ", decoder_flushing_=" << decoder_flushing_
                 << ", picture_clearing_count_=" << picture_clearing_count_;
       DVLOGF(4) << "Posting picture ready to GPU for: "
                 << picture.picture_buffer_id();
       // If the picture is not cleared, post it to the child thread because it
       // has to be cleared in the child thread. A picture only needs to be
@@ skipping 17 matching lines @@
 }
 
 void V4L2SliceVideoDecodeAccelerator::PictureCleared() {
   DVLOGF(3) << "clearing count=" << picture_clearing_count_;
   DCHECK(decoder_thread_task_runner_->BelongsToCurrentThread());
   DCHECK_GT(picture_clearing_count_, 0);
   picture_clearing_count_--;
   SendPictureReady();
 }
 
-bool V4L2SliceVideoDecodeAccelerator::CanDecodeOnIOThread() {
+bool V4L2SliceVideoDecodeAccelerator::TryToSetupDecodeOnSeparateThread(
+    const base::WeakPtr<Client>& decode_client,
+    const scoped_refptr<base::SingleThreadTaskRunner>& decode_task_runner) {
+  decode_client_ = decode_client_;
+  decode_task_runner_ = decode_task_runner;
   return true;
 }
 
 // static
 media::VideoDecodeAccelerator::SupportedProfiles
 V4L2SliceVideoDecodeAccelerator::GetSupportedProfiles() {
   scoped_refptr<V4L2Device> device = V4L2Device::Create(V4L2Device::kDecoder);
   if (!device)
     return SupportedProfiles();
 
   return device->GetSupportedDecodeProfiles(arraysize(supported_input_fourccs_),
                                             supported_input_fourccs_);
 }
 
 }  // namespace content