Mac Overlays: Enable h264 overlays

Punch a hole from GpuVideoDecodeAccelerator to VTVideoDecodeAccelerator
via a callback to BindImage, following the model found in
VaapiVideoDecodeAccelerator. If successful, set allow_overlay on the
frame's resource.

BUG=527650

Review URL: https://codereview.chromium.org/1320893004

Cr-Commit-Position: refs/heads/master@{#347546}
(cherry picked from commit 7ed442bbaf3a5038ce6b9d712200557a0854f888)

Review URL: https://codereview.chromium.org/1307863010 .

Cr-Commit-Position: refs/branch-heads/2490@{#199}
Cr-Branched-From: 7790a3535f2a81a03685eca31a32cf69ae0c114f-refs/heads/master@{#344925}

4 files changed, 90 insertions, 23 deletions

diff --git a/content/browser/compositor/gpu_process_transport_factory.cc b/content/browser/compositor/gpu_process_transport_factory.cc
index 7bfccf7..a5b1433 100644
--- a/content/browser/compositor/gpu_process_transport_factory.cc
+++ b/content/browser/compositor/gpu_process_transport_factory.cc
@@ -316,12 +316,14 @@ void GpuProcessTransportFactory::EstablishedGpuChannel(
               CreateOverlayCandidateValidator(compositor->widget()), target,
               format, BrowserGpuMemoryBufferManager::current()));
     } else {
-      if (!surface) {
-        surface = make_scoped_ptr(new GpuBrowserCompositorOutputSurface(
-            context_provider, shared_worker_context_provider_,
-            compositor->vsync_manager(),
-            CreateOverlayCandidateValidator(compositor->widget())));
-      }
+      scoped_ptr<BrowserCompositorOverlayCandidateValidator> validator;
+#if !defined(OS_MACOSX)
+      // Overlays are only supported on surfaceless output surfaces on Mac.
+      validator = CreateOverlayCandidateValidator(compositor->widget());
+#endif
+      surface = make_scoped_ptr(new GpuBrowserCompositorOutputSurface(
+          context_provider, shared_worker_context_provider_,
+          compositor->vsync_manager(), validator.Pass()));
     }
   }
 
diff --git a/content/common/gpu/media/gpu_video_decode_accelerator.cc b/content/common/gpu/media/gpu_video_decode_accelerator.cc
index 9f78d23..a0321c5a 100644
--- a/content/common/gpu/media/gpu_video_decode_accelerator.cc
+++ b/content/common/gpu/media/gpu_video_decode_accelerator.cc
@@ -358,7 +358,9 @@ scoped_ptr<media::VideoDecodeAccelerator>
 GpuVideoDecodeAccelerator::CreateVTVDA() {
   scoped_ptr<media::VideoDecodeAccelerator> decoder;
 #if defined(OS_MACOSX)
-  decoder.reset(new VTVideoDecodeAccelerator(make_context_current_));
+  decoder.reset(new VTVideoDecodeAccelerator(
+      make_context_current_, base::Bind(&GpuVideoDecodeAccelerator::BindImage,
+                                        base::Unretained(this))));
 #endif
   return decoder.Pass();
 }
diff --git a/content/common/gpu/media/vt_video_decode_accelerator.cc b/content/common/gpu/media/vt_video_decode_accelerator.cc
index f7e5264..c98853d 100644
--- a/content/common/gpu/media/vt_video_decode_accelerator.cc
+++ b/content/common/gpu/media/vt_video_decode_accelerator.cc
@@ -21,6 +21,7 @@
 #include "content/public/common/content_switches.h"
 #include "media/base/limits.h"
 #include "ui/gl/gl_context.h"
+#include "ui/gl/gl_image_io_surface.h"
 #include "ui/gl/scoped_binders.h"
 
 using content_common_gpu_media::kModuleVt;
@@ -274,6 +275,16 @@ VTVideoDecodeAccelerator::Frame::Frame(int32_t bitstream_id)
 VTVideoDecodeAccelerator::Frame::~Frame() {
 }
 
+VTVideoDecodeAccelerator::PictureInfo::PictureInfo(uint32_t client_texture_id,
+                                                   uint32_t service_texture_id)
+    : client_texture_id(client_texture_id),
+      service_texture_id(service_texture_id) {}
+
+VTVideoDecodeAccelerator::PictureInfo::~PictureInfo() {
+  if (gl_image)
+    gl_image->Destroy(false);
+}
+
 bool VTVideoDecodeAccelerator::FrameOrder::operator()(
     const linked_ptr<Frame>& lhs,
     const linked_ptr<Frame>& rhs) const {
@@ -287,8 +298,11 @@ bool VTVideoDecodeAccelerator::FrameOrder::operator()(
 }
 
 VTVideoDecodeAccelerator::VTVideoDecodeAccelerator(
-    const base::Callback<bool(void)>& make_context_current)
+    const base::Callback<bool(void)>& make_context_current,
+    const base::Callback<void(uint32, uint32, scoped_refptr<gfx::GLImage>)>&
+        bind_image)
     : make_context_current_(make_context_current),
+      bind_image_(bind_image),
       client_(nullptr),
       state_(STATE_DECODING),
       format_(nullptr),
@@ -305,6 +319,7 @@ VTVideoDecodeAccelerator::VTVideoDecodeAccelerator(
 }
 
 VTVideoDecodeAccelerator::~VTVideoDecodeAccelerator() {
+  DCHECK(gpu_thread_checker_.CalledOnValidThread());
 }
 
 bool VTVideoDecodeAccelerator::Initialize(
@@ -810,10 +825,12 @@ void VTVideoDecodeAccelerator::AssignPictureBuffers(
   DCHECK(gpu_thread_checker_.CalledOnValidThread());
 
   for (const media::PictureBuffer& picture : pictures) {
-    DCHECK(!texture_ids_.count(picture.id()));
+    DCHECK(!picture_info_map_.count(picture.id()));
     assigned_picture_ids_.insert(picture.id());
     available_picture_ids_.push_back(picture.id());
-    texture_ids_[picture.id()] = picture.texture_id();
+    picture_info_map_.insert(picture.id(), make_scoped_ptr(new PictureInfo(
+                                               picture.internal_texture_id(),
+                                               picture.texture_id())));
   }
 
   // Pictures are not marked as uncleared until after this method returns, and
@@ -825,8 +842,13 @@ void VTVideoDecodeAccelerator::AssignPictureBuffers(
 
 void VTVideoDecodeAccelerator::ReusePictureBuffer(int32_t picture_id) {
   DCHECK(gpu_thread_checker_.CalledOnValidThread());
-  DCHECK_EQ(CFGetRetainCount(picture_bindings_[picture_id]), 1);
-  picture_bindings_.erase(picture_id);
+  DCHECK(picture_info_map_.count(picture_id));
+  PictureInfo* picture_info = picture_info_map_.find(picture_id)->second;
+  DCHECK_EQ(CFGetRetainCount(picture_info->cv_image), 1);
+  picture_info->cv_image.reset();
+  picture_info->gl_image->Destroy(false);
+  picture_info->gl_image = nullptr;
+
   if (assigned_picture_ids_.count(picture_id) != 0) {
     available_picture_ids_.push_back(picture_id);
     ProcessWorkQueues();
@@ -985,7 +1007,10 @@ bool VTVideoDecodeAccelerator::SendFrame(const Frame& frame) {
     return false;
 
   int32_t picture_id = available_picture_ids_.back();
-  IOSurfaceRef surface = CVPixelBufferGetIOSurface(frame.image.get());
+  DCHECK(picture_info_map_.count(picture_id));
+  PictureInfo* picture_info = picture_info_map_.find(picture_id)->second;
+  DCHECK(!picture_info->cv_image);
+  DCHECK(!picture_info->gl_image);
 
   if (!make_context_current_.Run()) {
     DLOG(ERROR) << "Failed to make GL context current";
@@ -993,9 +1018,10 @@ bool VTVideoDecodeAccelerator::SendFrame(const Frame& frame) {
     return false;
   }
 
+  IOSurfaceRef surface = CVPixelBufferGetIOSurface(frame.image.get());
   glEnable(GL_TEXTURE_RECTANGLE_ARB);
-  gfx::ScopedTextureBinder
-      texture_binder(GL_TEXTURE_RECTANGLE_ARB, texture_ids_[picture_id]);
+  gfx::ScopedTextureBinder texture_binder(GL_TEXTURE_RECTANGLE_ARB,
+                                          picture_info->service_texture_id);
   CGLContextObj cgl_context =
       static_cast<CGLContextObj>(gfx::GLContext::GetCurrent()->GetHandle());
   CGLError status = CGLTexImageIOSurface2D(
@@ -1014,15 +1040,30 @@ bool VTVideoDecodeAccelerator::SendFrame(const Frame& frame) {
     return false;
   }
 
+  bool allow_overlay = false;
+  scoped_refptr<gfx::GLImageIOSurface> gl_image(new gfx::GLImageIOSurface(
+      gfx::GenericSharedMemoryId(), frame.coded_size, GL_BGRA_EXT));
+  if (gl_image->Initialize(surface, gfx::BufferFormat::BGRA_8888)) {
+    allow_overlay = true;
+  } else {
+    gl_image = nullptr;
+  }
+  bind_image_.Run(picture_info->client_texture_id, GL_TEXTURE_RECTANGLE_ARB,
+                  gl_image);
+
+  // Assign the new image(s) to the the picture info.
+  picture_info->gl_image = gl_image;
+  picture_info->cv_image = frame.image;
   available_picture_ids_.pop_back();
-  picture_bindings_[picture_id] = frame.image;
+
   // TODO(sandersd): Currently, the size got from
   // CMVideoFormatDescriptionGetDimensions is visible size. We pass it to
   // GpuVideoDecoder so that GpuVideoDecoder can use correct visible size in
   // resolution changed. We should find the correct API to get the real
   // coded size and fix it.
   client_->PictureReady(media::Picture(picture_id, frame.bitstream_id,
-                                       gfx::Rect(frame.coded_size), false));
+                                       gfx::Rect(frame.coded_size),
+                                       allow_overlay));
   return true;
 }
 
diff --git a/content/common/gpu/media/vt_video_decode_accelerator.h b/content/common/gpu/media/vt_video_decode_accelerator.h
index 44f563b..0a38509 100644
--- a/content/common/gpu/media/vt_video_decode_accelerator.h
+++ b/content/common/gpu/media/vt_video_decode_accelerator.h
@@ -10,6 +10,7 @@
 #include <map>
 #include <queue>
 
+#include "base/containers/scoped_ptr_map.h"
 #include "base/mac/scoped_cftyperef.h"
 #include "base/memory/linked_ptr.h"
 #include "base/memory/weak_ptr.h"
@@ -22,6 +23,7 @@
 #include "media/video/video_decode_accelerator.h"
 #include "ui/gfx/geometry/size.h"
 #include "ui/gl/gl_context_cgl.h"
+#include "ui/gl/gl_image_io_surface.h"
 
 namespace content {
 
@@ -33,7 +35,9 @@ bool InitializeVideoToolbox();
 class VTVideoDecodeAccelerator : public media::VideoDecodeAccelerator {
  public:
   explicit VTVideoDecodeAccelerator(
-      const base::Callback<bool(void)>& make_context_current);
+      const base::Callback<bool(void)>& make_context_current,
+      const base::Callback<void(uint32, uint32, scoped_refptr<gfx::GLImage>)>&
+          bind_image);
   ~VTVideoDecodeAccelerator() override;
 
   // VideoDecodeAccelerator implementation.
@@ -113,6 +117,26 @@ class VTVideoDecodeAccelerator : public media::VideoDecodeAccelerator {
     linked_ptr<Frame> frame;
   };
 
+  struct PictureInfo {
+    PictureInfo(uint32_t client_texture_id, uint32_t service_texture_id);
+    ~PictureInfo();
+
+    // Image buffer, kept alive while they are bound to pictures.
+    base::ScopedCFTypeRef<CVImageBufferRef> cv_image;
+
+    // The GLImage representation of |cv_image|. This is kept around to ensure
+    // that Destroy is called on it before it hits its destructor (there is a
+    // DCHECK that requires this).
+    scoped_refptr<gfx::GLImageIOSurface> gl_image;
+
+    // Texture IDs for the image buffer.
+    const uint32_t client_texture_id;
+    const uint32_t service_texture_id;
+
+   private:
+    DISALLOW_COPY_AND_ASSIGN(PictureInfo);
+  };
+
   //
   // Methods for interacting with VideoToolbox. Run on |decoder_thread_|.
   //
@@ -163,6 +187,7 @@ class VTVideoDecodeAccelerator : public media::VideoDecodeAccelerator {
   // GPU thread state.
   //
   base::Callback<bool(void)> make_context_current_;
+  base::Callback<void(uint32, uint32, scoped_refptr<gfx::GLImage>)> bind_image_;
   media::VideoDecodeAccelerator::Client* client_;
   State state_;
 
@@ -200,15 +225,12 @@ class VTVideoDecodeAccelerator : public media::VideoDecodeAccelerator {
   // released immediately because we need the reuse event to free the binding.)
   std::set<int32_t> assigned_picture_ids_;
 
-  // Texture IDs of assigned pictures.
-  std::map<int32_t, uint32_t> texture_ids_;
+  // Texture IDs and image buffers of assigned pictures.
+  base::ScopedPtrMap<int32_t, scoped_ptr<PictureInfo>> picture_info_map_;
 
   // Pictures ready to be rendered to.
   std::vector<int32_t> available_picture_ids_;
 
-  // Image buffers kept alive while they are bound to pictures.
-  std::map<int32_t, base::ScopedCFTypeRef<CVImageBufferRef>> picture_bindings_;
-
   //
   // Decoder thread state.
   //