1 // Copyright (c) 2012 The Chromium Authors. All rights reserved. 2 // Use of this source code is governed by a BSD-style license that can be 3 // found in the LICENSE file. 4 // 5 // This file contains an implementation of VideoDecoderAccelerator 6 // that utilizes hardware video decoder present on Intel CPUs. 7 8 #ifndef CONTENT_COMMON_GPU_MEDIA_VAAPI_VIDEO_DECODE_ACCELERATOR_H_ 9 #define CONTENT_COMMON_GPU_MEDIA_VAAPI_VIDEO_DECODE_ACCELERATOR_H_ 10 11 #include <map> 12 #include <queue> 13 #include <utility> 14 #include <vector> 15 16 #include "base/logging.h" 17 #include "base/memory/linked_ptr.h" 18 #include "base/memory/shared_memory.h" 19 #include "base/memory/weak_ptr.h" 20 #include "base/message_loop/message_loop.h" 21 #include "base/synchronization/condition_variable.h" 22 #include "base/synchronization/lock.h" 23 #include "base/threading/non_thread_safe.h" 24 #include "base/threading/thread.h" 25 #include "content/common/content_export.h" 26 #include "content/common/gpu/media/vaapi_h264_decoder.h" 27 #include "content/common/gpu/media/vaapi_wrapper.h" 28 #include "media/base/bitstream_buffer.h" 29 #include "media/video/picture.h" 30 #include "media/video/video_decode_accelerator.h" 31 #include "ui/gl/gl_bindings.h" 32 33 namespace content { 34 35 // Class to provide video decode acceleration for Intel systems with hardware 36 // support for it, and on which libva is available. 37 // Decoding tasks are performed in a separate decoding thread. 38 // 39 // Threading/life-cycle: this object is created & destroyed on the GPU 40 // ChildThread. A few methods on it are called on the decoder thread which is 41 // stopped during |this->Destroy()|, so any tasks posted to the decoder thread 42 // can assume |*this| is still alive. See |weak_this_| below for more details. 43 class CONTENT_EXPORT VaapiVideoDecodeAccelerator : 44 public media::VideoDecodeAccelerator { 45 public: 46 VaapiVideoDecodeAccelerator( 47 Display* x_display, GLXContext glx_context, 48 Client* client, 49 const base::Callback<bool(void)>& make_context_current); 50 virtual ~VaapiVideoDecodeAccelerator(); 51 52 // media::VideoDecodeAccelerator implementation. 53 virtual bool Initialize(media::VideoCodecProfile profile) OVERRIDE; 54 virtual void Decode(const media::BitstreamBuffer& bitstream_buffer) OVERRIDE; 55 virtual void AssignPictureBuffers( 56 const std::vector<media::PictureBuffer>& buffers) OVERRIDE; 57 virtual void ReusePictureBuffer(int32 picture_buffer_id) OVERRIDE; 58 virtual void Flush() OVERRIDE; 59 virtual void Reset() OVERRIDE; 60 virtual void Destroy() OVERRIDE; 61 62 private: 63 // Notify the client that |output_id| is ready for displaying. 64 void NotifyPictureReady(int32 input_id, int32 output_id); 65 66 // Notify the client that an error has occurred and decoding cannot continue. 67 void NotifyError(Error error); 68 69 // Map the received input buffer into this process' address space and 70 // queue it for decode. 71 void MapAndQueueNewInputBuffer( 72 const media::BitstreamBuffer& bitstream_buffer); 73 74 // Get a new input buffer from the queue and set it up in decoder. This will 75 // sleep if no input buffers are available. Return true if a new buffer has 76 // been set up, false if an early exit has been requested (due to initiated 77 // reset/flush/destroy). 78 bool GetInputBuffer_Locked(); 79 80 // Signal the client that the current buffer has been read and can be 81 // returned. Will also release the mapping. 82 void ReturnCurrInputBuffer_Locked(); 83 84 // Pass one or more output buffers to the decoder. This will sleep 85 // if no buffers are available. Return true if buffers have been set up or 86 // false if an early exit has been requested (due to initiated 87 // reset/flush/destroy). 88 bool FeedDecoderWithOutputSurfaces_Locked(); 89 90 // Continue decoding given input buffers and sleep waiting for input/output 91 // as needed. Will exit if a new set of surfaces or reset/flush/destroy 92 // is requested. 93 void DecodeTask(); 94 95 // Scheduled after receiving a flush request and executed after the current 96 // decoding task finishes decoding pending inputs. Makes the decoder return 97 // all remaining output pictures and puts it in an idle state, ready 98 // to resume if needed and schedules a FinishFlush. 99 void FlushTask(); 100 101 // Scheduled by the FlushTask after decoder is flushed to put VAVDA into idle 102 // state and notify the client that flushing has been finished. 103 void FinishFlush(); 104 105 // Scheduled after receiving a reset request and executed after the current 106 // decoding task finishes decoding the current frame. Puts the decoder into 107 // an idle state, ready to resume if needed, discarding decoded but not yet 108 // outputted pictures (decoder keeps ownership of their associated picture 109 // buffers). Schedules a FinishReset afterwards. 110 void ResetTask(); 111 112 // Scheduled by ResetTask after it's done putting VAVDA into an idle state. 113 // Drops remaining input buffers and notifies the client that reset has been 114 // finished. 115 void FinishReset(); 116 117 // Helper for Destroy(), doing all the actual work except for deleting self. 118 void Cleanup(); 119 120 // Get a usable framebuffer configuration for use in binding textures 121 // or return false on failure. 122 bool InitializeFBConfig(); 123 124 // Callback for the decoder to execute when it wants us to output given 125 // |va_surface|. 126 void SurfaceReady(int32 input_id, const scoped_refptr<VASurface>& va_surface); 127 128 // Represents a texture bound to an X Pixmap for output purposes. 129 class TFPPicture; 130 131 // Callback to be executed once we have a |va_surface| to be output and 132 // an available |tfp_picture| to use for output. 133 // Puts contents of |va_surface| into given |tfp_picture|, releases the 134 // surface and passes the resulting picture to client for output. 135 void OutputPicture(const scoped_refptr<VASurface>& va_surface, 136 int32 input_id, 137 TFPPicture* tfp_picture); 138 139 // Try to OutputPicture() if we have both a ready surface and picture. 140 void TryOutputSurface(); 141 142 // Called when a VASurface is no longer in use by the decoder or is not being 143 // synced/waiting to be synced to a picture. Returns it to available surfaces 144 // pool. 145 void RecycleVASurfaceID(VASurfaceID va_surface_id); 146 147 // Initiate wait cycle for surfaces to be released before we release them 148 // and allocate new ones, as requested by the decoder. 149 void InitiateSurfaceSetChange(size_t num_pics, gfx::Size size); 150 // Check if the surfaces have been released or post ourselves for later. 151 void TryFinishSurfaceSetChange(); 152 153 // Client-provided X/GLX state. 154 Display* x_display_; 155 GLXContext glx_context_; 156 base::Callback<bool(void)> make_context_current_; 157 GLXFBConfig fb_config_; 158 159 // VAVDA state. 160 enum State { 161 // Initialize() not called yet or failed. 162 kUninitialized, 163 // DecodeTask running. 164 kDecoding, 165 // Resetting, waiting for decoder to finish current task and cleanup. 166 kResetting, 167 // Flushing, waiting for decoder to finish current task and cleanup. 168 kFlushing, 169 // Idle, decoder in state ready to start/resume decoding. 170 kIdle, 171 // Destroying, waiting for the decoder to finish current task. 172 kDestroying, 173 }; 174 175 // Protects input buffer and surface queues and state_. 176 base::Lock lock_; 177 State state_; 178 179 // An input buffer awaiting consumption, provided by the client. 180 struct InputBuffer { 181 InputBuffer(); 182 ~InputBuffer(); 183 184 int32 id; 185 size_t size; 186 scoped_ptr<base::SharedMemory> shm; 187 }; 188 189 // Queue for incoming input buffers. 190 typedef std::queue<linked_ptr<InputBuffer> > InputBuffers; 191 InputBuffers input_buffers_; 192 // Signalled when input buffers are queued onto the input_buffers_ queue. 193 base::ConditionVariable input_ready_; 194 195 // Current input buffer at decoder. 196 linked_ptr<InputBuffer> curr_input_buffer_; 197 198 // Queue for incoming output buffers (texture ids). 199 typedef std::queue<int32> OutputBuffers; 200 OutputBuffers output_buffers_; 201 202 typedef std::map<int32, linked_ptr<TFPPicture> > TFPPictures; 203 // All allocated TFPPictures, regardless of their current state. TFPPictures 204 // are allocated once and destroyed at the end of decode. 205 TFPPictures tfp_pictures_; 206 207 // Return a TFPPicture associated with given client-provided id. 208 TFPPicture* TFPPictureById(int32 picture_buffer_id); 209 210 // VA Surfaces no longer in use that can be passed back to the decoder for 211 // reuse, once it requests them. 212 std::list<VASurfaceID> available_va_surfaces_; 213 // Signalled when output surfaces are queued onto the available_va_surfaces_ 214 // queue. 215 base::ConditionVariable surfaces_available_; 216 217 // Pending output requests from the decoder. When it indicates that we should 218 // output a surface and we have an available TFPPicture (i.e. texture) ready 219 // to use, we'll execute the callback passing the TFPPicture. The callback 220 // will put the contents of the surface into the picture and return it to 221 // the client, releasing the surface as well. 222 // If we don't have any available TFPPictures at the time when the decoder 223 // requests output, we'll store the request on pending_output_cbs_ queue for 224 // later and run it once the client gives us more textures 225 // via ReusePictureBuffer(). 226 typedef base::Callback<void(TFPPicture*)> OutputCB; 227 std::queue<OutputCB> pending_output_cbs_; 228 229 // ChildThread's message loop 230 base::MessageLoop* message_loop_; 231 232 // WeakPtr<> pointing to |this| for use in posting tasks from the decoder 233 // thread back to the ChildThread. Because the decoder thread is a member of 234 // this class, any task running on the decoder thread is guaranteed that this 235 // object is still alive. As a result, tasks posted from ChildThread to 236 // decoder thread should use base::Unretained(this), and tasks posted from the 237 // decoder thread to the ChildThread should use |weak_this_|. 238 base::WeakPtr<VaapiVideoDecodeAccelerator> weak_this_; 239 240 // To expose client callbacks from VideoDecodeAccelerator. 241 // NOTE: all calls to these objects *MUST* be executed on message_loop_. 242 base::WeakPtrFactory<Client> client_ptr_factory_; 243 base::WeakPtr<Client> client_; 244 245 scoped_ptr<VaapiWrapper> vaapi_wrapper_; 246 247 // Comes after vaapi_wrapper_ to ensure its destructor is executed before 248 // vaapi_wrapper_ is destroyed. 249 scoped_ptr<VaapiH264Decoder> decoder_; 250 base::Thread decoder_thread_; 251 252 int num_frames_at_client_; 253 int num_stream_bufs_at_decoder_; 254 255 // Whether we are waiting for any pending_output_cbs_ to be run before 256 // NotifyingFlushDone. 257 bool finish_flush_pending_; 258 259 // Decoder requested a new surface set and we are waiting for all the surfaces 260 // to be returned before we can free them. 261 bool awaiting_va_surfaces_recycle_; 262 263 // Last requested number/resolution of output picture buffers. 264 size_t requested_num_pics_; 265 gfx::Size requested_pic_size_; 266 267 DISALLOW_COPY_AND_ASSIGN(VaapiVideoDecodeAccelerator); 268 }; 269 270 } // namespace content 271 272 #endif // CONTENT_COMMON_GPU_MEDIA_VAAPI_VIDEO_DECODE_ACCELERATOR_H_ 273