OLD | NEW |
| (Empty) |
1 // Copyright 2015 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 #ifndef CONTENT_COMMON_GPU_MEDIA_V4L2_SLICE_VIDEO_DECODE_ACCELERATOR_H_ | |
6 #define CONTENT_COMMON_GPU_MEDIA_V4L2_SLICE_VIDEO_DECODE_ACCELERATOR_H_ | |
7 | |
8 #include <linux/videodev2.h> | |
9 #include <queue> | |
10 #include <vector> | |
11 | |
12 #include "base/memory/linked_ptr.h" | |
13 #include "base/memory/ref_counted.h" | |
14 #include "base/memory/scoped_ptr.h" | |
15 #include "base/memory/weak_ptr.h" | |
16 #include "base/synchronization/waitable_event.h" | |
17 #include "base/threading/thread.h" | |
18 #include "content/common/content_export.h" | |
19 #include "content/common/gpu/media/h264_decoder.h" | |
20 #include "content/common/gpu/media/v4l2_video_device.h" | |
21 #include "content/common/gpu/media/vp8_decoder.h" | |
22 #include "media/video/video_decode_accelerator.h" | |
23 | |
24 namespace content { | |
25 | |
26 // An implementation of VideoDecodeAccelerator that utilizes the V4L2 slice | |
27 // level codec API for decoding. The slice level API provides only a low-level | |
28 // decoding functionality and requires userspace to provide support for parsing | |
29 // the input stream and managing decoder state across frames. | |
30 class CONTENT_EXPORT V4L2SliceVideoDecodeAccelerator | |
31 : public media::VideoDecodeAccelerator { | |
32 public: | |
33 class V4L2DecodeSurface; | |
34 | |
35 V4L2SliceVideoDecodeAccelerator( | |
36 const scoped_refptr<V4L2Device>& device, | |
37 EGLDisplay egl_display, | |
38 EGLContext egl_context, | |
39 const base::WeakPtr<Client>& io_client_, | |
40 const base::Callback<bool(void)>& make_context_current, | |
41 const scoped_refptr<base::MessageLoopProxy>& io_message_loop_proxy); | |
42 virtual ~V4L2SliceVideoDecodeAccelerator() override; | |
43 | |
44 // media::VideoDecodeAccelerator implementation. | |
45 bool Initialize(media::VideoCodecProfile profile, | |
46 VideoDecodeAccelerator::Client* client) override; | |
47 void Decode(const media::BitstreamBuffer& bitstream_buffer) override; | |
48 void AssignPictureBuffers( | |
49 const std::vector<media::PictureBuffer>& buffers) override; | |
50 void ReusePictureBuffer(int32 picture_buffer_id) override; | |
51 void Flush() override; | |
52 void Reset() override; | |
53 void Destroy() override; | |
54 bool CanDecodeOnIOThread() override; | |
55 | |
56 private: | |
57 class V4L2H264Accelerator; | |
58 class V4L2VP8Accelerator; | |
59 | |
60 // Record for input buffers. | |
61 struct InputRecord { | |
62 InputRecord(); | |
63 int32 input_id; | |
64 void* address; | |
65 size_t length; | |
66 size_t bytes_used; | |
67 bool at_device; | |
68 }; | |
69 | |
70 // Record for output buffers. | |
71 struct OutputRecord { | |
72 OutputRecord(); | |
73 bool at_device; | |
74 bool at_client; | |
75 int32 picture_id; | |
76 EGLImageKHR egl_image; | |
77 EGLSyncKHR egl_sync; | |
78 bool cleared; | |
79 }; | |
80 | |
81 // See http://crbug.com/255116. | |
82 // Input bitstream buffer size for up to 1080p streams. | |
83 const size_t kInputBufferMaxSizeFor1080p = 1024 * 1024; | |
84 // Input bitstream buffer size for up to 4k streams. | |
85 const size_t kInputBufferMaxSizeFor4k = 4 * kInputBufferMaxSizeFor1080p; | |
86 const size_t kNumInputBuffers = 16; | |
87 | |
88 // | |
89 // Below methods are used by accelerator implementations. | |
90 // | |
91 // Append slice data in |data| of size |size| to pending hardware | |
92 // input buffer with |index|. This buffer will be submitted for decode | |
93 // on the next DecodeSurface(). Return true on success. | |
94 bool SubmitSlice(int index, const uint8_t* data, size_t size); | |
95 | |
96 // Submit controls in |ext_ctrls| to hardware. Return true on success. | |
97 bool SubmitExtControls(struct v4l2_ext_controls* ext_ctrls); | |
98 | |
99 // Decode of |dec_surface| is ready to be submitted and all codec-specific | |
100 // settings are set in hardware. | |
101 void DecodeSurface(const scoped_refptr<V4L2DecodeSurface>& dec_surface); | |
102 | |
103 // |dec_surface| is ready to be outputted once decode is finished. | |
104 // This can be called before decode is actually done in hardware, and this | |
105 // method is responsible for maintaining the ordering, i.e. the surfaces will | |
106 // be outputted in the same order as SurfaceReady calls. To do so, the | |
107 // surfaces are put on decoder_display_queue_ and sent to output in that | |
108 // order once all preceding surfaces are sent. | |
109 void SurfaceReady(const scoped_refptr<V4L2DecodeSurface>& dec_surface); | |
110 | |
111 // | |
112 // Internal methods of this class. | |
113 // | |
114 // Recycle V4L2 output buffer with |index|. Used as surface release callback. | |
115 void ReuseOutputBuffer(int index); | |
116 | |
117 // Queue a |dec_surface| to device for decoding. | |
118 void Enqueue(const scoped_refptr<V4L2DecodeSurface>& dec_surface); | |
119 | |
120 // Dequeue any V4L2 buffers available and process. | |
121 void Dequeue(); | |
122 | |
123 // V4L2 QBUF helpers. | |
124 bool EnqueueInputRecord(int index, uint32_t config_store); | |
125 bool EnqueueOutputRecord(int index); | |
126 | |
127 // Set input and output formats in hardware. | |
128 bool SetupFormats(); | |
129 | |
130 // Create input and output buffers. | |
131 bool CreateInputBuffers(); | |
132 bool CreateOutputBuffers(); | |
133 | |
134 // Destroy input buffers. | |
135 void DestroyInputBuffers(); | |
136 | |
137 // Destroy output buffers and release associated resources (textures, | |
138 // EGLImages). If |dismiss| is true, also dismissing the associated | |
139 // PictureBuffers. | |
140 bool DestroyOutputs(bool dismiss); | |
141 | |
142 // Used by DestroyOutputs. | |
143 bool DestroyOutputBuffers(); | |
144 | |
145 // Dismiss all |picture_buffer_ids| via Client::DismissPictureBuffer() | |
146 // and signal |done| after finishing. | |
147 void DismissPictures(std::vector<int32> picture_buffer_ids, | |
148 base::WaitableEvent* done); | |
149 | |
150 // Task to finish initialization on decoder_thread_. | |
151 void InitializeTask(); | |
152 | |
153 // Surface set change (resolution change) flow. | |
154 // If we have no surfaces allocated, just allocate them and return. | |
155 // Otherwise mark us as pending for surface set change. | |
156 void InitiateSurfaceSetChange(); | |
157 // If a surface set change is pending and we are ready, stop the device, | |
158 // destroy outputs, releasing resources and dismissing pictures as required, | |
159 // followed by allocating a new set for the new resolution/DPB size | |
160 // as provided by decoder. Finally, try to resume decoding. | |
161 void FinishSurfaceSetChangeIfNeeded(); | |
162 | |
163 void NotifyError(Error error); | |
164 void DestroyTask(); | |
165 | |
166 // Sets the state to kError and notifies client if needed. | |
167 void SetErrorState(Error error); | |
168 | |
169 // Flush flow when requested by client. | |
170 // When Flush() is called, it posts a FlushTask, which checks the input queue. | |
171 // If nothing is pending for decode on decoder_input_queue_, we call | |
172 // InitiateFlush() directly. Otherwise, we push a dummy BitstreamBufferRef | |
173 // onto the decoder_input_queue_ to schedule a flush. When we reach it later | |
174 // on, we call InitiateFlush() to perform it at the correct time. | |
175 void FlushTask(); | |
176 // Tell the decoder to flush all frames, reset it and mark us as scheduled | |
177 // for flush, so that we can finish it once all pending decodes are finished. | |
178 void InitiateFlush(); | |
179 // If all pending frames are decoded and we are waiting to flush, perform it. | |
180 // This will send all pending pictures to client and notify the client that | |
181 // flush is complete and puts us in a state ready to resume. | |
182 void FinishFlushIfNeeded(); | |
183 | |
184 // Reset flow when requested by client. | |
185 // Drop all inputs and reset the decoder and mark us as pending for reset. | |
186 void ResetTask(); | |
187 // If all pending frames are decoded and we are waiting to reset, perform it. | |
188 // This drops all pending outputs (client is not interested anymore), | |
189 // notifies the client we are done and puts us in a state ready to resume. | |
190 void FinishResetIfNeeded(); | |
191 | |
192 // Process pending events if any. | |
193 void ProcessPendingEventsIfNeeded(); | |
194 | |
195 // Performed on decoder_thread_ as a consequence of poll() on decoder_thread_ | |
196 // returning an event. | |
197 void ServiceDeviceTask(); | |
198 | |
199 // Schedule poll if we have any buffers queued and the poll thread | |
200 // is not stopped (on surface set change). | |
201 void SchedulePollIfNeeded(); | |
202 | |
203 // Attempt to start/stop device_poll_thread_. | |
204 bool StartDevicePoll(); | |
205 bool StopDevicePoll(bool keep_input_state); | |
206 | |
207 // Ran on device_poll_thread_ to wait for device events. | |
208 void DevicePollTask(bool poll_device); | |
209 | |
210 enum State { | |
211 // We are in this state until Initialize() returns successfully. | |
212 // We can't post errors to the client in this state yet. | |
213 kUninitialized, | |
214 // Initialize() returned successfully. | |
215 kInitialized, | |
216 // This state allows making progress decoding more input stream. | |
217 kDecoding, | |
218 // Transitional state when we are not decoding any more stream, but are | |
219 // performing flush, reset, resolution change or are destroying ourselves. | |
220 kIdle, | |
221 // Error state, set when sending NotifyError to client. | |
222 kError, | |
223 }; | |
224 | |
225 // Buffer id for flush buffer, queued by FlushTask(). | |
226 const int kFlushBufferId = -2; | |
227 | |
228 // Handler for Decode() on decoder_thread_. | |
229 void DecodeTask(const media::BitstreamBuffer& bitstream_buffer); | |
230 | |
231 // Schedule a new DecodeBufferTask if we are decoding. | |
232 void ScheduleDecodeBufferTaskIfNeeded(); | |
233 | |
234 // Main decoder loop. Keep decoding the current buffer in decoder_, asking | |
235 // for more stream via TrySetNewBistreamBuffer() if decoder_ requests so, | |
236 // and handle other returns from it appropriately. | |
237 void DecodeBufferTask(); | |
238 | |
239 // Check decoder_input_queue_ for any available buffers to decode and | |
240 // set the decoder_current_bitstream_buffer_ to the next buffer if one is | |
241 // available, taking it off the queue. Also set the current stream pointer | |
242 // in decoder_, and return true. | |
243 // Return false if no buffers are pending on decoder_input_queue_. | |
244 bool TrySetNewBistreamBuffer(); | |
245 | |
246 // Auto-destruction reference for EGLSync (for message-passing). | |
247 struct EGLSyncKHRRef; | |
248 void ReusePictureBufferTask(int32 picture_buffer_id, | |
249 scoped_ptr<EGLSyncKHRRef> egl_sync_ref); | |
250 | |
251 // Called to actually send |dec_surface| to the client, after it is decoded | |
252 // preserving the order in which it was scheduled via SurfaceReady(). | |
253 void OutputSurface(const scoped_refptr<V4L2DecodeSurface>& dec_surface); | |
254 | |
255 // Goes over the |decoder_display_queue_| and sends all buffers from the | |
256 // front of the queue that are already decoded to the client, in order. | |
257 void TryOutputSurfaces(); | |
258 | |
259 // Creates a new decode surface or returns nullptr if one is not available. | |
260 scoped_refptr<V4L2DecodeSurface> CreateSurface(); | |
261 | |
262 // Send decoded pictures to PictureReady. | |
263 void SendPictureReady(); | |
264 | |
265 // Callback that indicates a picture has been cleared. | |
266 void PictureCleared(); | |
267 | |
268 size_t input_planes_count_; | |
269 size_t output_planes_count_; | |
270 | |
271 // GPU Child thread message loop. | |
272 const scoped_refptr<base::MessageLoopProxy> child_message_loop_proxy_; | |
273 | |
274 // IO thread message loop. | |
275 scoped_refptr<base::MessageLoopProxy> io_message_loop_proxy_; | |
276 | |
277 // WeakPtr<> pointing to |this| for use in posting tasks from the decoder or | |
278 // device worker threads back to the child thread. | |
279 base::WeakPtr<V4L2SliceVideoDecodeAccelerator> weak_this_; | |
280 | |
281 // To expose client callbacks from VideoDecodeAccelerator. | |
282 // NOTE: all calls to these objects *MUST* be executed on | |
283 // child_message_loop_proxy_. | |
284 scoped_ptr<base::WeakPtrFactory<VideoDecodeAccelerator::Client>> | |
285 client_ptr_factory_; | |
286 base::WeakPtr<VideoDecodeAccelerator::Client> client_; | |
287 // Callbacks to |io_client_| must be executed on |io_message_loop_proxy_|. | |
288 base::WeakPtr<Client> io_client_; | |
289 | |
290 // V4L2 device in use. | |
291 scoped_refptr<V4L2Device> device_; | |
292 | |
293 // Thread to communicate with the device on. | |
294 base::Thread decoder_thread_; | |
295 scoped_refptr<base::MessageLoopProxy> decoder_thread_proxy_; | |
296 | |
297 // Thread used to poll the device for events. | |
298 base::Thread device_poll_thread_; | |
299 | |
300 // Input queue state. | |
301 bool input_streamon_; | |
302 // Number of input buffers enqueued to the device. | |
303 int input_buffer_queued_count_; | |
304 // Input buffers ready to use; LIFO since we don't care about ordering. | |
305 std::list<int> free_input_buffers_; | |
306 // Mapping of int index to an input buffer record. | |
307 std::vector<InputRecord> input_buffer_map_; | |
308 | |
309 // Output queue state. | |
310 bool output_streamon_; | |
311 // Number of output buffers enqueued to the device. | |
312 int output_buffer_queued_count_; | |
313 // Output buffers ready to use. | |
314 std::list<int> free_output_buffers_; | |
315 // Mapping of int index to an output buffer record. | |
316 std::vector<OutputRecord> output_buffer_map_; | |
317 | |
318 media::VideoCodecProfile video_profile_; | |
319 uint32_t output_format_fourcc_; | |
320 gfx::Size frame_buffer_size_; | |
321 size_t output_dpb_size_; | |
322 | |
323 struct BitstreamBufferRef; | |
324 // Input queue of stream buffers coming from the client. | |
325 std::queue<linked_ptr<BitstreamBufferRef>> decoder_input_queue_; | |
326 // BitstreamBuffer currently being processed. | |
327 scoped_ptr<BitstreamBufferRef> decoder_current_bitstream_buffer_; | |
328 | |
329 // Queue storing decode surfaces ready to be output as soon as they are | |
330 // decoded. The surfaces must be output in order they are queued. | |
331 std::queue<scoped_refptr<V4L2DecodeSurface>> decoder_display_queue_; | |
332 | |
333 // Decoder state. | |
334 State state_; | |
335 | |
336 // If any of these are true, we are waiting for the device to finish decoding | |
337 // all previously-queued frames, so we can finish the flush/reset/surface | |
338 // change flows. These can stack. | |
339 bool decoder_flushing_; | |
340 bool decoder_resetting_; | |
341 bool surface_set_change_pending_; | |
342 | |
343 // Hardware accelerators. | |
344 // TODO(posciak): Try to have a superclass here if possible. | |
345 scoped_ptr<V4L2H264Accelerator> h264_accelerator_; | |
346 scoped_ptr<V4L2VP8Accelerator> vp8_accelerator_; | |
347 | |
348 // Codec-specific software decoder in use. | |
349 scoped_ptr<AcceleratedVideoDecoder> decoder_; | |
350 | |
351 // Surfaces queued to device to keep references to them while decoded. | |
352 using V4L2DecodeSurfaceByOutputId = | |
353 std::map<int, scoped_refptr<V4L2DecodeSurface>>; | |
354 V4L2DecodeSurfaceByOutputId surfaces_at_device_; | |
355 | |
356 // Surfaces sent to client to keep references to them while displayed. | |
357 using V4L2DecodeSurfaceByPictureBufferId = | |
358 std::map<int32, scoped_refptr<V4L2DecodeSurface>>; | |
359 V4L2DecodeSurfaceByPictureBufferId surfaces_at_display_; | |
360 | |
361 // Record for decoded pictures that can be sent to PictureReady. | |
362 struct PictureRecord; | |
363 // Pictures that are ready but not sent to PictureReady yet. | |
364 std::queue<PictureRecord> pending_picture_ready_; | |
365 | |
366 // The number of pictures that are sent to PictureReady and will be cleared. | |
367 int picture_clearing_count_; | |
368 | |
369 // Used by the decoder thread to wait for AssignPictureBuffers to arrive | |
370 // to avoid races with potential Reset requests. | |
371 base::WaitableEvent pictures_assigned_; | |
372 | |
373 // Make the GL context current callback. | |
374 base::Callback<bool(void)> make_context_current_; | |
375 | |
376 // EGL state | |
377 EGLDisplay egl_display_; | |
378 EGLContext egl_context_; | |
379 | |
380 // The WeakPtrFactory for |weak_this_|. | |
381 base::WeakPtrFactory<V4L2SliceVideoDecodeAccelerator> weak_this_factory_; | |
382 | |
383 DISALLOW_COPY_AND_ASSIGN(V4L2SliceVideoDecodeAccelerator); | |
384 }; | |
385 | |
386 class V4L2H264Picture; | |
387 class V4L2VP8Picture; | |
388 | |
389 } // namespace content | |
390 | |
391 #endif // CONTENT_COMMON_GPU_MEDIA_V4L2_SLICE_VIDEO_DECODE_ACCELERATOR_H_ | |
OLD | NEW |