Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(285)

Side by Side Diff: content/common/gpu/media/v4l2_slice_video_decode_accelerator.h

Issue 1882373004: Migrate content/common/gpu/media code to media/gpu (Closed) Base URL: https://chromium.googlesource.com/chromium/src.git@master
Patch Set: Fix several more bot-identified build issues Created 4 years, 8 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View unified diff | Download patch
OLDNEW
(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 <stddef.h>
10 #include <stdint.h>
11
12 #include <memory>
13 #include <queue>
14 #include <vector>
15
16 #include "base/macros.h"
17 #include "base/memory/linked_ptr.h"
18 #include "base/memory/ref_counted.h"
19 #include "base/memory/weak_ptr.h"
20 #include "base/synchronization/waitable_event.h"
21 #include "base/threading/thread.h"
22 #include "content/common/content_export.h"
23 #include "content/common/gpu/media/gpu_video_decode_accelerator_helpers.h"
24 #include "content/common/gpu/media/h264_decoder.h"
25 #include "content/common/gpu/media/v4l2_device.h"
26 #include "content/common/gpu/media/vp8_decoder.h"
27 #include "media/video/video_decode_accelerator.h"
28
29 namespace content {
30
31 // An implementation of VideoDecodeAccelerator that utilizes the V4L2 slice
32 // level codec API for decoding. The slice level API provides only a low-level
33 // decoding functionality and requires userspace to provide support for parsing
34 // the input stream and managing decoder state across frames.
35 class CONTENT_EXPORT V4L2SliceVideoDecodeAccelerator
36 : public media::VideoDecodeAccelerator {
37 public:
38 class V4L2DecodeSurface;
39
40 V4L2SliceVideoDecodeAccelerator(
41 const scoped_refptr<V4L2Device>& device,
42 EGLDisplay egl_display,
43 const GetGLContextCallback& get_gl_context_cb,
44 const MakeGLContextCurrentCallback& make_context_current_cb);
45 ~V4L2SliceVideoDecodeAccelerator() override;
46
47 // media::VideoDecodeAccelerator implementation.
48 bool Initialize(const Config& config, Client* client) override;
49 void Decode(const media::BitstreamBuffer& bitstream_buffer) override;
50 void AssignPictureBuffers(
51 const std::vector<media::PictureBuffer>& buffers) override;
52 void ReusePictureBuffer(int32_t picture_buffer_id) override;
53 void Flush() override;
54 void Reset() override;
55 void Destroy() override;
56 bool TryToSetupDecodeOnSeparateThread(
57 const base::WeakPtr<Client>& decode_client,
58 const scoped_refptr<base::SingleThreadTaskRunner>& decode_task_runner)
59 override;
60
61 static media::VideoDecodeAccelerator::SupportedProfiles
62 GetSupportedProfiles();
63
64 private:
65 class V4L2H264Accelerator;
66 class V4L2VP8Accelerator;
67
68 // Record for input buffers.
69 struct InputRecord {
70 InputRecord();
71 int32_t input_id;
72 void* address;
73 size_t length;
74 size_t bytes_used;
75 bool at_device;
76 };
77
78 // Record for output buffers.
79 struct OutputRecord {
80 OutputRecord();
81 bool at_device;
82 bool at_client;
83 int32_t picture_id;
84 EGLImageKHR egl_image;
85 EGLSyncKHR egl_sync;
86 bool cleared;
87 };
88
89 // See http://crbug.com/255116.
90 // Input bitstream buffer size for up to 1080p streams.
91 const size_t kInputBufferMaxSizeFor1080p = 1024 * 1024;
92 // Input bitstream buffer size for up to 4k streams.
93 const size_t kInputBufferMaxSizeFor4k = 4 * kInputBufferMaxSizeFor1080p;
94 const size_t kNumInputBuffers = 16;
95
96 // Input format V4L2 fourccs this class supports.
97 static const uint32_t supported_input_fourccs_[];
98
99 //
100 // Below methods are used by accelerator implementations.
101 //
102 // Append slice data in |data| of size |size| to pending hardware
103 // input buffer with |index|. This buffer will be submitted for decode
104 // on the next DecodeSurface(). Return true on success.
105 bool SubmitSlice(int index, const uint8_t* data, size_t size);
106
107 // Submit controls in |ext_ctrls| to hardware. Return true on success.
108 bool SubmitExtControls(struct v4l2_ext_controls* ext_ctrls);
109
110 // Decode of |dec_surface| is ready to be submitted and all codec-specific
111 // settings are set in hardware.
112 void DecodeSurface(const scoped_refptr<V4L2DecodeSurface>& dec_surface);
113
114 // |dec_surface| is ready to be outputted once decode is finished.
115 // This can be called before decode is actually done in hardware, and this
116 // method is responsible for maintaining the ordering, i.e. the surfaces will
117 // be outputted in the same order as SurfaceReady calls. To do so, the
118 // surfaces are put on decoder_display_queue_ and sent to output in that
119 // order once all preceding surfaces are sent.
120 void SurfaceReady(const scoped_refptr<V4L2DecodeSurface>& dec_surface);
121
122 //
123 // Internal methods of this class.
124 //
125 // Recycle a V4L2 input buffer with |index| after dequeuing from device.
126 void ReuseInputBuffer(int index);
127
128 // Recycle V4L2 output buffer with |index|. Used as surface release callback.
129 void ReuseOutputBuffer(int index);
130
131 // Queue a |dec_surface| to device for decoding.
132 void Enqueue(const scoped_refptr<V4L2DecodeSurface>& dec_surface);
133
134 // Dequeue any V4L2 buffers available and process.
135 void Dequeue();
136
137 // V4L2 QBUF helpers.
138 bool EnqueueInputRecord(int index, uint32_t config_store);
139 bool EnqueueOutputRecord(int index);
140
141 // Set input and output formats in hardware.
142 bool SetupFormats();
143
144 // Create input and output buffers.
145 bool CreateInputBuffers();
146 bool CreateOutputBuffers();
147
148 // Destroy input buffers.
149 void DestroyInputBuffers();
150
151 // Destroy output buffers and release associated resources (textures,
152 // EGLImages). If |dismiss| is true, also dismissing the associated
153 // PictureBuffers.
154 bool DestroyOutputs(bool dismiss);
155
156 // Used by DestroyOutputs.
157 bool DestroyOutputBuffers();
158
159 // Dismiss all |picture_buffer_ids| via Client::DismissPictureBuffer()
160 // and signal |done| after finishing.
161 void DismissPictures(std::vector<int32_t> picture_buffer_ids,
162 base::WaitableEvent* done);
163
164 // Task to finish initialization on decoder_thread_.
165 void InitializeTask();
166
167 // Surface set change (resolution change) flow.
168 // If we have no surfaces allocated, just allocate them and return.
169 // Otherwise mark us as pending for surface set change.
170 void InitiateSurfaceSetChange();
171 // If a surface set change is pending and we are ready, stop the device,
172 // destroy outputs, releasing resources and dismissing pictures as required,
173 // followed by allocating a new set for the new resolution/DPB size
174 // as provided by decoder. Finally, try to resume decoding.
175 void FinishSurfaceSetChangeIfNeeded();
176
177 void NotifyError(Error error);
178 void DestroyTask();
179
180 // Sets the state to kError and notifies client if needed.
181 void SetErrorState(Error error);
182
183 // Flush flow when requested by client.
184 // When Flush() is called, it posts a FlushTask, which checks the input queue.
185 // If nothing is pending for decode on decoder_input_queue_, we call
186 // InitiateFlush() directly. Otherwise, we push a dummy BitstreamBufferRef
187 // onto the decoder_input_queue_ to schedule a flush. When we reach it later
188 // on, we call InitiateFlush() to perform it at the correct time.
189 void FlushTask();
190 // Tell the decoder to flush all frames, reset it and mark us as scheduled
191 // for flush, so that we can finish it once all pending decodes are finished.
192 void InitiateFlush();
193 // If all pending frames are decoded and we are waiting to flush, perform it.
194 // This will send all pending pictures to client and notify the client that
195 // flush is complete and puts us in a state ready to resume.
196 void FinishFlushIfNeeded();
197
198 // Reset flow when requested by client.
199 // Drop all inputs and reset the decoder and mark us as pending for reset.
200 void ResetTask();
201 // If all pending frames are decoded and we are waiting to reset, perform it.
202 // This drops all pending outputs (client is not interested anymore),
203 // notifies the client we are done and puts us in a state ready to resume.
204 void FinishResetIfNeeded();
205
206 // Process pending events if any.
207 void ProcessPendingEventsIfNeeded();
208
209 // Performed on decoder_thread_ as a consequence of poll() on decoder_thread_
210 // returning an event.
211 void ServiceDeviceTask();
212
213 // Schedule poll if we have any buffers queued and the poll thread
214 // is not stopped (on surface set change).
215 void SchedulePollIfNeeded();
216
217 // Attempt to start/stop device_poll_thread_.
218 bool StartDevicePoll();
219 bool StopDevicePoll(bool keep_input_state);
220
221 // Ran on device_poll_thread_ to wait for device events.
222 void DevicePollTask(bool poll_device);
223
224 enum State {
225 // We are in this state until Initialize() returns successfully.
226 // We can't post errors to the client in this state yet.
227 kUninitialized,
228 // Initialize() returned successfully.
229 kInitialized,
230 // This state allows making progress decoding more input stream.
231 kDecoding,
232 // Transitional state when we are not decoding any more stream, but are
233 // performing flush, reset, resolution change or are destroying ourselves.
234 kIdle,
235 // Error state, set when sending NotifyError to client.
236 kError,
237 };
238
239 // Buffer id for flush buffer, queued by FlushTask().
240 const int kFlushBufferId = -2;
241
242 // Handler for Decode() on decoder_thread_.
243 void DecodeTask(const media::BitstreamBuffer& bitstream_buffer);
244
245 // Schedule a new DecodeBufferTask if we are decoding.
246 void ScheduleDecodeBufferTaskIfNeeded();
247
248 // Main decoder loop. Keep decoding the current buffer in decoder_, asking
249 // for more stream via TrySetNewBistreamBuffer() if decoder_ requests so,
250 // and handle other returns from it appropriately.
251 void DecodeBufferTask();
252
253 // Check decoder_input_queue_ for any available buffers to decode and
254 // set the decoder_current_bitstream_buffer_ to the next buffer if one is
255 // available, taking it off the queue. Also set the current stream pointer
256 // in decoder_, and return true.
257 // Return false if no buffers are pending on decoder_input_queue_.
258 bool TrySetNewBistreamBuffer();
259
260 // Auto-destruction reference for EGLSync (for message-passing).
261 struct EGLSyncKHRRef;
262 void ReusePictureBufferTask(int32_t picture_buffer_id,
263 std::unique_ptr<EGLSyncKHRRef> egl_sync_ref);
264
265 // Called to actually send |dec_surface| to the client, after it is decoded
266 // preserving the order in which it was scheduled via SurfaceReady().
267 void OutputSurface(const scoped_refptr<V4L2DecodeSurface>& dec_surface);
268
269 // Goes over the |decoder_display_queue_| and sends all buffers from the
270 // front of the queue that are already decoded to the client, in order.
271 void TryOutputSurfaces();
272
273 // Creates a new decode surface or returns nullptr if one is not available.
274 scoped_refptr<V4L2DecodeSurface> CreateSurface();
275
276 // Send decoded pictures to PictureReady.
277 void SendPictureReady();
278
279 // Callback that indicates a picture has been cleared.
280 void PictureCleared();
281
282 size_t input_planes_count_;
283 size_t output_planes_count_;
284
285 // GPU Child thread task runner.
286 const scoped_refptr<base::SingleThreadTaskRunner> child_task_runner_;
287
288 // Task runner Decode() and PictureReady() run on.
289 scoped_refptr<base::SingleThreadTaskRunner> decode_task_runner_;
290
291 // WeakPtr<> pointing to |this| for use in posting tasks from the decoder or
292 // device worker threads back to the child thread.
293 base::WeakPtr<V4L2SliceVideoDecodeAccelerator> weak_this_;
294
295 // To expose client callbacks from VideoDecodeAccelerator.
296 // NOTE: all calls to these objects *MUST* be executed on
297 // child_task_runner_.
298 std::unique_ptr<base::WeakPtrFactory<VideoDecodeAccelerator::Client>>
299 client_ptr_factory_;
300 base::WeakPtr<VideoDecodeAccelerator::Client> client_;
301 // Callbacks to |decode_client_| must be executed on |decode_task_runner_|.
302 base::WeakPtr<Client> decode_client_;
303
304 // V4L2 device in use.
305 scoped_refptr<V4L2Device> device_;
306
307 // Thread to communicate with the device on.
308 base::Thread decoder_thread_;
309 scoped_refptr<base::SingleThreadTaskRunner> decoder_thread_task_runner_;
310
311 // Thread used to poll the device for events.
312 base::Thread device_poll_thread_;
313
314 // Input queue state.
315 bool input_streamon_;
316 // Number of input buffers enqueued to the device.
317 int input_buffer_queued_count_;
318 // Input buffers ready to use; LIFO since we don't care about ordering.
319 std::list<int> free_input_buffers_;
320 // Mapping of int index to an input buffer record.
321 std::vector<InputRecord> input_buffer_map_;
322
323 // Output queue state.
324 bool output_streamon_;
325 // Number of output buffers enqueued to the device.
326 int output_buffer_queued_count_;
327 // Output buffers ready to use.
328 std::list<int> free_output_buffers_;
329 // Mapping of int index to an output buffer record.
330 std::vector<OutputRecord> output_buffer_map_;
331
332 media::VideoCodecProfile video_profile_;
333 uint32_t output_format_fourcc_;
334 gfx::Size visible_size_;
335 gfx::Size coded_size_;
336
337 struct BitstreamBufferRef;
338 // Input queue of stream buffers coming from the client.
339 std::queue<linked_ptr<BitstreamBufferRef>> decoder_input_queue_;
340 // BitstreamBuffer currently being processed.
341 std::unique_ptr<BitstreamBufferRef> decoder_current_bitstream_buffer_;
342
343 // Queue storing decode surfaces ready to be output as soon as they are
344 // decoded. The surfaces must be output in order they are queued.
345 std::queue<scoped_refptr<V4L2DecodeSurface>> decoder_display_queue_;
346
347 // Decoder state.
348 State state_;
349
350 // If any of these are true, we are waiting for the device to finish decoding
351 // all previously-queued frames, so we can finish the flush/reset/surface
352 // change flows. These can stack.
353 bool decoder_flushing_;
354 bool decoder_resetting_;
355 bool surface_set_change_pending_;
356
357 // Hardware accelerators.
358 // TODO(posciak): Try to have a superclass here if possible.
359 std::unique_ptr<V4L2H264Accelerator> h264_accelerator_;
360 std::unique_ptr<V4L2VP8Accelerator> vp8_accelerator_;
361
362 // Codec-specific software decoder in use.
363 std::unique_ptr<AcceleratedVideoDecoder> decoder_;
364
365 // Surfaces queued to device to keep references to them while decoded.
366 using V4L2DecodeSurfaceByOutputId =
367 std::map<int, scoped_refptr<V4L2DecodeSurface>>;
368 V4L2DecodeSurfaceByOutputId surfaces_at_device_;
369
370 // Surfaces sent to client to keep references to them while displayed.
371 using V4L2DecodeSurfaceByPictureBufferId =
372 std::map<int32_t, scoped_refptr<V4L2DecodeSurface>>;
373 V4L2DecodeSurfaceByPictureBufferId surfaces_at_display_;
374
375 // Record for decoded pictures that can be sent to PictureReady.
376 struct PictureRecord;
377 // Pictures that are ready but not sent to PictureReady yet.
378 std::queue<PictureRecord> pending_picture_ready_;
379
380 // The number of pictures that are sent to PictureReady and will be cleared.
381 int picture_clearing_count_;
382
383 // Used by the decoder thread to wait for AssignPictureBuffers to arrive
384 // to avoid races with potential Reset requests.
385 base::WaitableEvent pictures_assigned_;
386
387 // EGL state
388 EGLDisplay egl_display_;
389
390 // Callback to get current GLContext.
391 GetGLContextCallback get_gl_context_cb_;
392 // Callback to set the correct gl context.
393 MakeGLContextCurrentCallback make_context_current_cb_;
394
395 // The WeakPtrFactory for |weak_this_|.
396 base::WeakPtrFactory<V4L2SliceVideoDecodeAccelerator> weak_this_factory_;
397
398 DISALLOW_COPY_AND_ASSIGN(V4L2SliceVideoDecodeAccelerator);
399 };
400
401 class V4L2H264Picture;
402 class V4L2VP8Picture;
403
404 } // namespace content
405
406 #endif // CONTENT_COMMON_GPU_MEDIA_V4L2_SLICE_VIDEO_DECODE_ACCELERATOR_H_
OLDNEW

Powered by Google App Engine
This is Rietveld 408576698