VAVDA: Redesign stage 1.

content/common/gpu/media/vaapi_video_decode_accelerator.h

 // Copyright (c) 2012 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.
 //
 // This file contains an implementation of VideoDecoderAccelerator
 // that utilizes hardware video decoder present on Intel CPUs.
 
 #ifndef CONTENT_COMMON_GPU_MEDIA_VAAPI_VIDEO_DECODE_ACCELERATOR_H_
 #define CONTENT_COMMON_GPU_MEDIA_VAAPI_VIDEO_DECODE_ACCELERATOR_H_
 
+#include <map>
 #include <queue>
 #include <utility>
 #include <vector>
 
 #include "base/logging.h"
-#include "base/memory/ref_counted.h"
+#include "base/memory/linked_ptr.h"
 #include "base/memory/weak_ptr.h"
 #include "base/message_loop.h"
 #include "base/shared_memory.h"
 #include "base/synchronization/condition_variable.h"
 #include "base/synchronization/lock.h"
 #include "base/threading/non_thread_safe.h"
 #include "base/threading/thread.h"
 #include "content/common/content_export.h"
+#include "content/common/gpu/media/vaapi_delegate.h"
 #include "content/common/gpu/media/vaapi_h264_decoder.h"
 #include "media/base/bitstream_buffer.h"
 #include "media/video/picture.h"
 #include "media/video/video_decode_accelerator.h"
 #include "ui/gl/gl_bindings.h"
 
 namespace content {
 
 // Class to provide video decode acceleration for Intel systems with hardware
 // support for it, and on which libva is available.
@@ skipping 15 matching lines @@
   // media::VideoDecodeAccelerator implementation.
   virtual bool Initialize(media::VideoCodecProfile profile) OVERRIDE;
   virtual void Decode(const media::BitstreamBuffer& bitstream_buffer) OVERRIDE;
   virtual void AssignPictureBuffers(
       const std::vector<media::PictureBuffer>& buffers) OVERRIDE;
   virtual void ReusePictureBuffer(int32 picture_buffer_id) OVERRIDE;
   virtual void Flush() OVERRIDE;
   virtual void Reset() OVERRIDE;
   virtual void Destroy() OVERRIDE;
 
-  // Do any necessary initialization before the sandbox is enabled.
-  static void PreSandboxInitialization();
-
 private:
   // Notify the client that |output_id| is ready for displaying.
   void NotifyPictureReady(int32 input_id, int32 output_id);
 
   // Notify the client that an error has occurred and decoding cannot continue.
   void NotifyError(Error error);
 
   // Map the received input buffer into this process' address space and
   // queue it for decode.
   void MapAndQueueNewInputBuffer(
       const media::BitstreamBuffer& bitstream_buffer);
 
   // Get a new input buffer from the queue and set it up in decoder. This will
   // sleep if no input buffers are available. Return true if a new buffer has
   // been set up, false if an early exit has been requested (due to initiated
   // reset/flush/destroy).
   bool GetInputBuffer_Locked();
 
   // Signal the client that the current buffer has been read and can be
   // returned. Will also release the mapping.
   void ReturnCurrInputBuffer_Locked();
 
-  // Get and set up one or more output buffers in the decoder. This will sleep
+  // Pass one or more output buffers to the decoder. This will sleep
   // if no buffers are available. Return true if buffers have been set up or
   // false if an early exit has been requested (due to initiated
   // reset/flush/destroy).
-  bool GetOutputBuffers_Locked();
+  bool GetOutputSurfaces_Locked();

[Ami GONE FROM CHROMIUM, 2013/05/17 23:19:15]

function name is crying out for improvement

[Pawel Osciak, 2013/05/21 22:32:35]

Done.

 
-  // Initial decode task: get the decoder to the point in the stream from which
-  // it can start/continue decoding. Does not require output buffers and does
-  // not produce output frames. Called either when starting with a new stream
-  // or when playback is to be resumed following a seek.
+  // Get the decoder to the point in the stream from which it can start/continue
+  // decoding. Called either when starting to decode a new stream or when
+  // playback is to be resumed following a seek.
   void InitialDecodeTask();
 
-  // Decoding task. Will continue decoding given input buffers and sleep
-  // waiting for input/output as needed. Will exit if a reset/flush/destroy
-  // is requested.
+  // Continue decoding given input buffers and sleep waiting for input/output
+  // as needed. Will exit if a reset/flush/destroy is requested.
   void DecodeTask();
 
   // Scheduled after receiving a flush request and executed after the current
   // decoding task finishes decoding pending inputs. Makes the decoder return
   // all remaining output pictures and puts it in an idle state, ready
   // to resume if needed and schedules a FinishFlush.
   void FlushTask();
 
   // Scheduled by the FlushTask after decoder is flushed to put VAVDA into idle
   // state and notify the client that flushing has been finished.
   void FinishFlush();
 
   // Scheduled after receiving a reset request and executed after the current
   // decoding task finishes decoding the current frame. Puts the decoder into
   // an idle state, ready to resume if needed, discarding decoded but not yet
   // outputted pictures (decoder keeps ownership of their associated picture
   // buffers). Schedules a FinishReset afterwards.
   void ResetTask();
 
   // Scheduled by ResetTask after it's done putting VAVDA into an idle state.
   // Drops remaining input buffers and notifies the client that reset has been
   // finished.
   void FinishReset();
 
   // Helper for Destroy(), doing all the actual work except for deleting self.
   void Cleanup();
 
-  // Lazily initialize static data after sandbox is enabled.  Return false on
-  // init failure.
-  static bool PostSandboxInitialization();
+  // Get a usable framebuffer configuration for use in binding textures
+  // or return false on failure.
+  bool InitializeFBConfig();
+
+  // Callback for the decoder to execute when it wants us to output given
+  // |va_surface|.
+  void SurfaceReady(int32 input_id, const scoped_refptr<VASurface>& va_surface);
+
+  // Represents a texture bound to an X Pixmap for output purposes.
+  class TFPPicture;
+
+  // Put contents of the |va_surface| into given |tfp_picture| for output
+  // and release the surface afterwards.
+  void SyncSurfaceToPicture(VASurfaceID va_surface_id, TFPPicture* tfp_picture);

[Ami GONE FROM CHROMIUM, 2013/05/17 23:19:15]

This CL finaly makes it possible for this madness to stop!
(this method can now be inlined into OutputPicture).

[Pawel Osciak, 2013/05/21 22:32:35]

Yes! 
Done.

+
+  // Callback to be executed once we have a |va_surface| to be output and
+  // an available |tfp_picture| to use for output.
+  void OutputPicture(const scoped_refptr<VASurface>& va_surface,
+                     int32 input_id,
+                     TFPPicture* tfp_picture);
+
+  // Try to OutputPicture() if we have both a ready surface and picture.
+  void TryOutputSurface();
+
+  // Called when a VASurface is no longer in use by the decoder or is not being
+  // synced/waiting to be synced to a picture. Returns it to available surfaces
+  // pool.
+  void RecycleVASurfaceID(VASurfaceID va_surface_id);
 
   // Client-provided X/GLX state.
   Display* x_display_;
   GLXContext glx_context_;
   base::Callback<bool(void)> make_context_current_;
+  GLXFBConfig fb_config_;
 
   // VAVDA state.
   enum State {
     // Initialize() not called yet or failed.
     kUninitialized,
     // Initialize() succeeded, no initial decode and no pictures requested.
     kInitialized,
     // Initial decode finished, requested pictures and waiting for them.
     kPicturesRequested,
     // Everything initialized, pictures received and assigned, in decoding.
     kDecoding,
     // Resetting, waiting for decoder to finish current task and cleanup.
     kResetting,
     // Flushing, waiting for decoder to finish current task and cleanup.
     kFlushing,
     // Idle, decoder in state ready to resume decoding.
     kIdle,
     // Destroying, waiting for the decoder to finish current task.
     kDestroying,
   };
 
   State state_;

[Ami GONE FROM CHROMIUM, 2013/05/17 23:19:15]

should go under lock_

[Pawel Osciak, 2013/05/21 22:32:35]

Done.

 
-  // Protects input and output buffer queues and state_.
+  // Protects input buffer and surface queues and state_.
   base::Lock lock_;
 
   // An input buffer awaiting consumption, provided by the client.
   struct InputBuffer {
     InputBuffer();
     ~InputBuffer();
 
     int32 id;
     size_t size;
     scoped_ptr<base::SharedMemory> shm;
   };
 
   // Queue for incoming input buffers.
   typedef std::queue<linked_ptr<InputBuffer> > InputBuffers;
   InputBuffers input_buffers_;
   // Signalled when input buffers are queued onto the input_buffers_ queue.
   base::ConditionVariable input_ready_;
 
   // Current input buffer at decoder.
   linked_ptr<InputBuffer> curr_input_buffer_;
 
-  // Queue for incoming input buffers.
+  // Queue for incoming output buffers (texture ids).
   typedef std::queue<int32> OutputBuffers;
   OutputBuffers output_buffers_;
-  // Signalled when output buffers are queued onto the output_buffers_ queue.
-  base::ConditionVariable output_ready_;
+
+  typedef std::map<int32, linked_ptr<TFPPicture> > TFPPictures;
+  // All allocated TFPPictures, regardless of their current state. TFPPictures
+  // are allocated once and destroyed at the end of decode.
+  TFPPictures tfp_pictures_;
+  std::list<TFPPicture*> available_tfp_pictures_;
+
+  // Return a TFPPicture associated with given client-provided id.
+  TFPPicture* TFPPictureById(int32 picture_buffer_id);
+
+  // Number/resolution of output picture buffers.
+  size_t num_pics_;
+  gfx::Size pic_size_;

[Ami GONE FROM CHROMIUM, 2013/05/17 23:19:15]

You can drop both of these in favor of simply calling the decoder_ methods
twice.

[Pawel Osciak, 2013/05/21 22:32:35]

This would result in calling the decoder from ChildThread in
AssignPictureBuffers, which I didn't want to do.

+
+  // VA Surfaces no longer in use that can be passed back to the decoder for
+  // reuse, once it requests them.
+  std::list<VASurfaceID> available_va_surfaces_;
+  // Signalled when output surfaces are queued onto the available_va_surfaces_
+  // queue.
+  base::ConditionVariable surfaces_available_;
+
+  // Pending output requests from the decoder. When it indicates that we should
+  // output a surface and we have an available TFPPicture (i.e. texture) ready
+  // to use, we'll execute the callback passing the TFPPicture. The callback
+  // will put the contents of the surface into the picture and return it to
+  // the client, releasing the surface as well.
+  // If we don't have any available TFPPictures at the time when the decoder
+  // requests output, we'll store the request on pending_output_cbs_ queue for
+  // later and run it once the client gives us more textures
+  // via ReusePictureBuffer().
+  typedef base::Callback<void(TFPPicture*)> OutputCB;
+  std::queue<OutputCB> pending_output_cbs_;
 
   // ChildThread's message loop
   base::MessageLoop* message_loop_;
 
   // WeakPtr<> pointing to |this| for use in posting tasks from the decoder
   // thread back to the ChildThread.  Because the decoder thread is a member of
   // this class, any task running on the decoder thread is guaranteed that this
   // object is still alive.  As a result, tasks posted from ChildThread to
   // decoder thread should use base::Unretained(this), and tasks posted from the
   // decoder thread to the ChildThread should use |weak_this_|.
   base::WeakPtr<VaapiVideoDecodeAccelerator> weak_this_;
 
   // To expose client callbacks from VideoDecodeAccelerator.
   // NOTE: all calls to these objects *MUST* be executed on message_loop_.
   base::WeakPtrFactory<Client> client_ptr_factory_;
   base::WeakPtr<Client> client_;
 
   base::Thread decoder_thread_;
-  VaapiH264Decoder decoder_;
+  scoped_ptr<VaapiH264Decoder> decoder_;

[Ami GONE FROM CHROMIUM, 2013/05/17 23:19:15]

this should go above thread_ so that it's easy to see that the decoder can't be
deleted before the thread is stopped/joined/deleted.

[Pawel Osciak, 2013/05/21 22:32:35]

Done.

 
   int num_frames_at_client_;
   int num_stream_bufs_at_decoder_;
 
-  // Posted onto ChildThread by the decoder to submit a GPU job to decode
-  // and put the decoded picture into output buffer. Takes ownership of
-  // the queues' memory.
-  void SubmitDecode(int32 output_id,
-                    scoped_ptr<std::queue<VABufferID> > va_bufs,
-                    scoped_ptr<std::queue<VABufferID> > slice_bufs);
+  scoped_refptr<VaapiDelegate> vaapi_delegate_;
 
   DISALLOW_COPY_AND_ASSIGN(VaapiVideoDecodeAccelerator);
 };
 
 }  // namespace content
 
 #endif  // CONTENT_COMMON_GPU_MEDIA_VAAPI_VIDEO_DECODE_ACCELERATOR_H_