OLD | NEW |
| (Empty) |
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 #include "content/common/gpu/media/vaapi_video_decode_accelerator.h" | |
6 | |
7 #include <string.h> | |
8 | |
9 #include <memory> | |
10 | |
11 #include "base/bind.h" | |
12 #include "base/logging.h" | |
13 #include "base/macros.h" | |
14 #include "base/metrics/histogram.h" | |
15 #include "base/stl_util.h" | |
16 #include "base/strings/string_util.h" | |
17 #include "base/synchronization/waitable_event.h" | |
18 #include "base/trace_event/trace_event.h" | |
19 #include "content/common/gpu/media/accelerated_video_decoder.h" | |
20 #include "content/common/gpu/media/h264_decoder.h" | |
21 #include "content/common/gpu/media/vaapi_picture.h" | |
22 #include "content/common/gpu/media/vp8_decoder.h" | |
23 #include "content/common/gpu/media/vp9_decoder.h" | |
24 #include "gpu/ipc/service/gpu_channel.h" | |
25 #include "media/base/bind_to_current_loop.h" | |
26 #include "media/video/picture.h" | |
27 #include "third_party/libva/va/va_dec_vp8.h" | |
28 #include "ui/gl/gl_bindings.h" | |
29 #include "ui/gl/gl_image.h" | |
30 | |
31 namespace content { | |
32 | |
33 namespace { | |
34 // UMA errors that the VaapiVideoDecodeAccelerator class reports. | |
35 enum VAVDADecoderFailure { | |
36 VAAPI_ERROR = 0, | |
37 // UMA requires that max must be greater than 1. | |
38 VAVDA_DECODER_FAILURES_MAX = 2, | |
39 }; | |
40 } | |
41 | |
42 static void ReportToUMA(VAVDADecoderFailure failure) { | |
43 UMA_HISTOGRAM_ENUMERATION("Media.VAVDA.DecoderFailure", failure, | |
44 VAVDA_DECODER_FAILURES_MAX); | |
45 } | |
46 | |
47 #define RETURN_AND_NOTIFY_ON_FAILURE(result, log, error_code, ret) \ | |
48 do { \ | |
49 if (!(result)) { \ | |
50 LOG(ERROR) << log; \ | |
51 NotifyError(error_code); \ | |
52 return ret; \ | |
53 } \ | |
54 } while (0) | |
55 | |
56 class VaapiVideoDecodeAccelerator::VaapiDecodeSurface | |
57 : public base::RefCountedThreadSafe<VaapiDecodeSurface> { | |
58 public: | |
59 VaapiDecodeSurface(int32_t bitstream_id, | |
60 const scoped_refptr<VASurface>& va_surface); | |
61 | |
62 int32_t bitstream_id() const { return bitstream_id_; } | |
63 scoped_refptr<VASurface> va_surface() { return va_surface_; } | |
64 | |
65 private: | |
66 friend class base::RefCountedThreadSafe<VaapiDecodeSurface>; | |
67 ~VaapiDecodeSurface(); | |
68 | |
69 int32_t bitstream_id_; | |
70 scoped_refptr<VASurface> va_surface_; | |
71 }; | |
72 | |
73 VaapiVideoDecodeAccelerator::VaapiDecodeSurface::VaapiDecodeSurface( | |
74 int32_t bitstream_id, | |
75 const scoped_refptr<VASurface>& va_surface) | |
76 : bitstream_id_(bitstream_id), va_surface_(va_surface) {} | |
77 | |
78 VaapiVideoDecodeAccelerator::VaapiDecodeSurface::~VaapiDecodeSurface() { | |
79 } | |
80 | |
81 class VaapiH264Picture : public H264Picture { | |
82 public: | |
83 VaapiH264Picture(const scoped_refptr< | |
84 VaapiVideoDecodeAccelerator::VaapiDecodeSurface>& dec_surface); | |
85 | |
86 VaapiH264Picture* AsVaapiH264Picture() override { return this; } | |
87 scoped_refptr<VaapiVideoDecodeAccelerator::VaapiDecodeSurface> dec_surface() { | |
88 return dec_surface_; | |
89 } | |
90 | |
91 private: | |
92 ~VaapiH264Picture() override; | |
93 | |
94 scoped_refptr<VaapiVideoDecodeAccelerator::VaapiDecodeSurface> dec_surface_; | |
95 | |
96 DISALLOW_COPY_AND_ASSIGN(VaapiH264Picture); | |
97 }; | |
98 | |
99 VaapiH264Picture::VaapiH264Picture(const scoped_refptr< | |
100 VaapiVideoDecodeAccelerator::VaapiDecodeSurface>& dec_surface) | |
101 : dec_surface_(dec_surface) { | |
102 } | |
103 | |
104 VaapiH264Picture::~VaapiH264Picture() { | |
105 } | |
106 | |
107 class VaapiVideoDecodeAccelerator::VaapiH264Accelerator | |
108 : public H264Decoder::H264Accelerator { | |
109 public: | |
110 VaapiH264Accelerator(VaapiVideoDecodeAccelerator* vaapi_dec, | |
111 VaapiWrapper* vaapi_wrapper); | |
112 ~VaapiH264Accelerator() override; | |
113 | |
114 // H264Decoder::H264Accelerator implementation. | |
115 scoped_refptr<H264Picture> CreateH264Picture() override; | |
116 | |
117 bool SubmitFrameMetadata(const media::H264SPS* sps, | |
118 const media::H264PPS* pps, | |
119 const H264DPB& dpb, | |
120 const H264Picture::Vector& ref_pic_listp0, | |
121 const H264Picture::Vector& ref_pic_listb0, | |
122 const H264Picture::Vector& ref_pic_listb1, | |
123 const scoped_refptr<H264Picture>& pic) override; | |
124 | |
125 bool SubmitSlice(const media::H264PPS* pps, | |
126 const media::H264SliceHeader* slice_hdr, | |
127 const H264Picture::Vector& ref_pic_list0, | |
128 const H264Picture::Vector& ref_pic_list1, | |
129 const scoped_refptr<H264Picture>& pic, | |
130 const uint8_t* data, | |
131 size_t size) override; | |
132 | |
133 bool SubmitDecode(const scoped_refptr<H264Picture>& pic) override; | |
134 bool OutputPicture(const scoped_refptr<H264Picture>& pic) override; | |
135 | |
136 void Reset() override; | |
137 | |
138 private: | |
139 scoped_refptr<VaapiDecodeSurface> H264PictureToVaapiDecodeSurface( | |
140 const scoped_refptr<H264Picture>& pic); | |
141 | |
142 void FillVAPicture(VAPictureH264* va_pic, scoped_refptr<H264Picture> pic); | |
143 int FillVARefFramesFromDPB(const H264DPB& dpb, | |
144 VAPictureH264* va_pics, | |
145 int num_pics); | |
146 | |
147 VaapiWrapper* vaapi_wrapper_; | |
148 VaapiVideoDecodeAccelerator* vaapi_dec_; | |
149 | |
150 DISALLOW_COPY_AND_ASSIGN(VaapiH264Accelerator); | |
151 }; | |
152 | |
153 class VaapiVP8Picture : public VP8Picture { | |
154 public: | |
155 VaapiVP8Picture(const scoped_refptr< | |
156 VaapiVideoDecodeAccelerator::VaapiDecodeSurface>& dec_surface); | |
157 | |
158 VaapiVP8Picture* AsVaapiVP8Picture() override { return this; } | |
159 scoped_refptr<VaapiVideoDecodeAccelerator::VaapiDecodeSurface> dec_surface() { | |
160 return dec_surface_; | |
161 } | |
162 | |
163 private: | |
164 ~VaapiVP8Picture() override; | |
165 | |
166 scoped_refptr<VaapiVideoDecodeAccelerator::VaapiDecodeSurface> dec_surface_; | |
167 | |
168 DISALLOW_COPY_AND_ASSIGN(VaapiVP8Picture); | |
169 }; | |
170 | |
171 VaapiVP8Picture::VaapiVP8Picture(const scoped_refptr< | |
172 VaapiVideoDecodeAccelerator::VaapiDecodeSurface>& dec_surface) | |
173 : dec_surface_(dec_surface) { | |
174 } | |
175 | |
176 VaapiVP8Picture::~VaapiVP8Picture() { | |
177 } | |
178 | |
179 class VaapiVideoDecodeAccelerator::VaapiVP8Accelerator | |
180 : public VP8Decoder::VP8Accelerator { | |
181 public: | |
182 VaapiVP8Accelerator(VaapiVideoDecodeAccelerator* vaapi_dec, | |
183 VaapiWrapper* vaapi_wrapper); | |
184 ~VaapiVP8Accelerator() override; | |
185 | |
186 // VP8Decoder::VP8Accelerator implementation. | |
187 scoped_refptr<VP8Picture> CreateVP8Picture() override; | |
188 | |
189 bool SubmitDecode(const scoped_refptr<VP8Picture>& pic, | |
190 const media::Vp8FrameHeader* frame_hdr, | |
191 const scoped_refptr<VP8Picture>& last_frame, | |
192 const scoped_refptr<VP8Picture>& golden_frame, | |
193 const scoped_refptr<VP8Picture>& alt_frame) override; | |
194 | |
195 bool OutputPicture(const scoped_refptr<VP8Picture>& pic) override; | |
196 | |
197 private: | |
198 scoped_refptr<VaapiDecodeSurface> VP8PictureToVaapiDecodeSurface( | |
199 const scoped_refptr<VP8Picture>& pic); | |
200 | |
201 VaapiWrapper* vaapi_wrapper_; | |
202 VaapiVideoDecodeAccelerator* vaapi_dec_; | |
203 | |
204 DISALLOW_COPY_AND_ASSIGN(VaapiVP8Accelerator); | |
205 }; | |
206 | |
207 class VaapiVP9Picture : public VP9Picture { | |
208 public: | |
209 VaapiVP9Picture( | |
210 const scoped_refptr<VaapiVideoDecodeAccelerator::VaapiDecodeSurface>& | |
211 dec_surface); | |
212 | |
213 VaapiVP9Picture* AsVaapiVP9Picture() override { return this; } | |
214 scoped_refptr<VaapiVideoDecodeAccelerator::VaapiDecodeSurface> dec_surface() { | |
215 return dec_surface_; | |
216 } | |
217 | |
218 private: | |
219 ~VaapiVP9Picture() override; | |
220 | |
221 scoped_refptr<VaapiVideoDecodeAccelerator::VaapiDecodeSurface> dec_surface_; | |
222 | |
223 DISALLOW_COPY_AND_ASSIGN(VaapiVP9Picture); | |
224 }; | |
225 | |
226 VaapiVP9Picture::VaapiVP9Picture( | |
227 const scoped_refptr<VaapiVideoDecodeAccelerator::VaapiDecodeSurface>& | |
228 dec_surface) | |
229 : dec_surface_(dec_surface) {} | |
230 | |
231 VaapiVP9Picture::~VaapiVP9Picture() {} | |
232 | |
233 class VaapiVideoDecodeAccelerator::VaapiVP9Accelerator | |
234 : public VP9Decoder::VP9Accelerator { | |
235 public: | |
236 VaapiVP9Accelerator(VaapiVideoDecodeAccelerator* vaapi_dec, | |
237 VaapiWrapper* vaapi_wrapper); | |
238 ~VaapiVP9Accelerator() override; | |
239 | |
240 // VP9Decoder::VP9Accelerator implementation. | |
241 scoped_refptr<VP9Picture> CreateVP9Picture() override; | |
242 | |
243 bool SubmitDecode( | |
244 const scoped_refptr<VP9Picture>& pic, | |
245 const media::Vp9Segmentation& seg, | |
246 const media::Vp9LoopFilter& lf, | |
247 const std::vector<scoped_refptr<VP9Picture>>& ref_pictures) override; | |
248 | |
249 bool OutputPicture(const scoped_refptr<VP9Picture>& pic) override; | |
250 | |
251 private: | |
252 scoped_refptr<VaapiDecodeSurface> VP9PictureToVaapiDecodeSurface( | |
253 const scoped_refptr<VP9Picture>& pic); | |
254 | |
255 VaapiWrapper* vaapi_wrapper_; | |
256 VaapiVideoDecodeAccelerator* vaapi_dec_; | |
257 | |
258 DISALLOW_COPY_AND_ASSIGN(VaapiVP9Accelerator); | |
259 }; | |
260 | |
261 VaapiVideoDecodeAccelerator::InputBuffer::InputBuffer() : id(0) {} | |
262 | |
263 VaapiVideoDecodeAccelerator::InputBuffer::~InputBuffer() { | |
264 } | |
265 | |
266 void VaapiVideoDecodeAccelerator::NotifyError(Error error) { | |
267 if (message_loop_ != base::MessageLoop::current()) { | |
268 DCHECK(decoder_thread_task_runner_->BelongsToCurrentThread()); | |
269 message_loop_->PostTask(FROM_HERE, base::Bind( | |
270 &VaapiVideoDecodeAccelerator::NotifyError, weak_this_, error)); | |
271 return; | |
272 } | |
273 | |
274 // Post Cleanup() as a task so we don't recursively acquire lock_. | |
275 message_loop_->PostTask(FROM_HERE, base::Bind( | |
276 &VaapiVideoDecodeAccelerator::Cleanup, weak_this_)); | |
277 | |
278 LOG(ERROR) << "Notifying of error " << error; | |
279 if (client_) { | |
280 client_->NotifyError(error); | |
281 client_ptr_factory_.reset(); | |
282 } | |
283 } | |
284 | |
285 VaapiPicture* VaapiVideoDecodeAccelerator::PictureById( | |
286 int32_t picture_buffer_id) { | |
287 Pictures::iterator it = pictures_.find(picture_buffer_id); | |
288 if (it == pictures_.end()) { | |
289 LOG(ERROR) << "Picture id " << picture_buffer_id << " does not exist"; | |
290 return NULL; | |
291 } | |
292 | |
293 return it->second.get(); | |
294 } | |
295 | |
296 VaapiVideoDecodeAccelerator::VaapiVideoDecodeAccelerator( | |
297 const MakeGLContextCurrentCallback& make_context_current_cb, | |
298 const BindGLImageCallback& bind_image_cb) | |
299 : state_(kUninitialized), | |
300 input_ready_(&lock_), | |
301 surfaces_available_(&lock_), | |
302 message_loop_(base::MessageLoop::current()), | |
303 decoder_thread_("VaapiDecoderThread"), | |
304 num_frames_at_client_(0), | |
305 num_stream_bufs_at_decoder_(0), | |
306 finish_flush_pending_(false), | |
307 awaiting_va_surfaces_recycle_(false), | |
308 requested_num_pics_(0), | |
309 make_context_current_cb_(make_context_current_cb), | |
310 bind_image_cb_(bind_image_cb), | |
311 weak_this_factory_(this) { | |
312 weak_this_ = weak_this_factory_.GetWeakPtr(); | |
313 va_surface_release_cb_ = media::BindToCurrentLoop( | |
314 base::Bind(&VaapiVideoDecodeAccelerator::RecycleVASurfaceID, weak_this_)); | |
315 } | |
316 | |
317 VaapiVideoDecodeAccelerator::~VaapiVideoDecodeAccelerator() { | |
318 DCHECK_EQ(message_loop_, base::MessageLoop::current()); | |
319 } | |
320 | |
321 bool VaapiVideoDecodeAccelerator::Initialize(const Config& config, | |
322 Client* client) { | |
323 DCHECK_EQ(message_loop_, base::MessageLoop::current()); | |
324 | |
325 if (make_context_current_cb_.is_null() || bind_image_cb_.is_null()) { | |
326 NOTREACHED() << "GL callbacks are required for this VDA"; | |
327 return false; | |
328 } | |
329 | |
330 if (config.is_encrypted) { | |
331 NOTREACHED() << "Encrypted streams are not supported for this VDA"; | |
332 return false; | |
333 } | |
334 | |
335 if (config.output_mode != Config::OutputMode::ALLOCATE) { | |
336 NOTREACHED() << "Only ALLOCATE OutputMode is supported by this VDA"; | |
337 return false; | |
338 } | |
339 | |
340 client_ptr_factory_.reset(new base::WeakPtrFactory<Client>(client)); | |
341 client_ = client_ptr_factory_->GetWeakPtr(); | |
342 | |
343 media::VideoCodecProfile profile = config.profile; | |
344 | |
345 base::AutoLock auto_lock(lock_); | |
346 DCHECK_EQ(state_, kUninitialized); | |
347 DVLOG(2) << "Initializing VAVDA, profile: " << profile; | |
348 | |
349 #if defined(USE_X11) | |
350 if (gfx::GetGLImplementation() != gfx::kGLImplementationDesktopGL) { | |
351 DVLOG(1) << "HW video decode acceleration not available without " | |
352 "DesktopGL (GLX)."; | |
353 return false; | |
354 } | |
355 #elif defined(USE_OZONE) | |
356 if (gfx::GetGLImplementation() != gfx::kGLImplementationEGLGLES2) { | |
357 DVLOG(1) << "HW video decode acceleration not available without " | |
358 << "EGLGLES2."; | |
359 return false; | |
360 } | |
361 #endif // USE_X11 | |
362 | |
363 vaapi_wrapper_ = VaapiWrapper::CreateForVideoCodec( | |
364 VaapiWrapper::kDecode, profile, base::Bind(&ReportToUMA, VAAPI_ERROR)); | |
365 | |
366 if (!vaapi_wrapper_.get()) { | |
367 DVLOG(1) << "Failed initializing VAAPI for profile " << profile; | |
368 return false; | |
369 } | |
370 | |
371 if (profile >= media::H264PROFILE_MIN && profile <= media::H264PROFILE_MAX) { | |
372 h264_accelerator_.reset( | |
373 new VaapiH264Accelerator(this, vaapi_wrapper_.get())); | |
374 decoder_.reset(new H264Decoder(h264_accelerator_.get())); | |
375 } else if (profile >= media::VP8PROFILE_MIN && | |
376 profile <= media::VP8PROFILE_MAX) { | |
377 vp8_accelerator_.reset(new VaapiVP8Accelerator(this, vaapi_wrapper_.get())); | |
378 decoder_.reset(new VP8Decoder(vp8_accelerator_.get())); | |
379 } else if (profile >= media::VP9PROFILE_MIN && | |
380 profile <= media::VP9PROFILE_MAX) { | |
381 vp9_accelerator_.reset(new VaapiVP9Accelerator(this, vaapi_wrapper_.get())); | |
382 decoder_.reset(new VP9Decoder(vp9_accelerator_.get())); | |
383 } else { | |
384 DLOG(ERROR) << "Unsupported profile " << profile; | |
385 return false; | |
386 } | |
387 | |
388 CHECK(decoder_thread_.Start()); | |
389 decoder_thread_task_runner_ = decoder_thread_.task_runner(); | |
390 | |
391 state_ = kIdle; | |
392 return true; | |
393 } | |
394 | |
395 void VaapiVideoDecodeAccelerator::OutputPicture( | |
396 const scoped_refptr<VASurface>& va_surface, | |
397 int32_t input_id, | |
398 VaapiPicture* picture) { | |
399 DCHECK_EQ(message_loop_, base::MessageLoop::current()); | |
400 | |
401 int32_t output_id = picture->picture_buffer_id(); | |
402 | |
403 TRACE_EVENT2("Video Decoder", "VAVDA::OutputSurface", | |
404 "input_id", input_id, | |
405 "output_id", output_id); | |
406 | |
407 DVLOG(3) << "Outputting VASurface " << va_surface->id() | |
408 << " into pixmap bound to picture buffer id " << output_id; | |
409 | |
410 RETURN_AND_NOTIFY_ON_FAILURE(picture->DownloadFromSurface(va_surface), | |
411 "Failed putting surface into pixmap", | |
412 PLATFORM_FAILURE, ); | |
413 | |
414 // Notify the client a picture is ready to be displayed. | |
415 ++num_frames_at_client_; | |
416 TRACE_COUNTER1("Video Decoder", "Textures at client", num_frames_at_client_); | |
417 DVLOG(4) << "Notifying output picture id " << output_id | |
418 << " for input "<< input_id << " is ready"; | |
419 // TODO(posciak): Use visible size from decoder here instead | |
420 // (crbug.com/402760). Passing (0, 0) results in the client using the | |
421 // visible size extracted from the container instead. | |
422 if (client_) | |
423 client_->PictureReady(media::Picture(output_id, input_id, | |
424 gfx::Rect(0, 0), | |
425 picture->AllowOverlay())); | |
426 } | |
427 | |
428 void VaapiVideoDecodeAccelerator::TryOutputSurface() { | |
429 DCHECK_EQ(message_loop_, base::MessageLoop::current()); | |
430 | |
431 // Handle Destroy() arriving while pictures are queued for output. | |
432 if (!client_) | |
433 return; | |
434 | |
435 if (pending_output_cbs_.empty() || output_buffers_.empty()) | |
436 return; | |
437 | |
438 OutputCB output_cb = pending_output_cbs_.front(); | |
439 pending_output_cbs_.pop(); | |
440 | |
441 VaapiPicture* picture = PictureById(output_buffers_.front()); | |
442 DCHECK(picture); | |
443 output_buffers_.pop(); | |
444 | |
445 output_cb.Run(picture); | |
446 | |
447 if (finish_flush_pending_ && pending_output_cbs_.empty()) | |
448 FinishFlush(); | |
449 } | |
450 | |
451 void VaapiVideoDecodeAccelerator::MapAndQueueNewInputBuffer( | |
452 const media::BitstreamBuffer& bitstream_buffer) { | |
453 DCHECK_EQ(message_loop_, base::MessageLoop::current()); | |
454 TRACE_EVENT1("Video Decoder", "MapAndQueueNewInputBuffer", "input_id", | |
455 bitstream_buffer.id()); | |
456 | |
457 DVLOG(4) << "Mapping new input buffer id: " << bitstream_buffer.id() | |
458 << " size: " << (int)bitstream_buffer.size(); | |
459 | |
460 std::unique_ptr<SharedMemoryRegion> shm( | |
461 new SharedMemoryRegion(bitstream_buffer, true)); | |
462 | |
463 // Skip empty buffers. | |
464 if (bitstream_buffer.size() == 0) { | |
465 if (client_) | |
466 client_->NotifyEndOfBitstreamBuffer(bitstream_buffer.id()); | |
467 return; | |
468 } | |
469 | |
470 RETURN_AND_NOTIFY_ON_FAILURE(shm->Map(), "Failed to map input buffer", | |
471 UNREADABLE_INPUT, ); | |
472 | |
473 base::AutoLock auto_lock(lock_); | |
474 | |
475 // Set up a new input buffer and queue it for later. | |
476 linked_ptr<InputBuffer> input_buffer(new InputBuffer()); | |
477 input_buffer->shm.reset(shm.release()); | |
478 input_buffer->id = bitstream_buffer.id(); | |
479 | |
480 ++num_stream_bufs_at_decoder_; | |
481 TRACE_COUNTER1("Video Decoder", "Stream buffers at decoder", | |
482 num_stream_bufs_at_decoder_); | |
483 | |
484 input_buffers_.push(input_buffer); | |
485 input_ready_.Signal(); | |
486 } | |
487 | |
488 bool VaapiVideoDecodeAccelerator::GetInputBuffer_Locked() { | |
489 DCHECK(decoder_thread_task_runner_->BelongsToCurrentThread()); | |
490 lock_.AssertAcquired(); | |
491 | |
492 if (curr_input_buffer_.get()) | |
493 return true; | |
494 | |
495 // Will only wait if it is expected that in current state new buffers will | |
496 // be queued from the client via Decode(). The state can change during wait. | |
497 while (input_buffers_.empty() && (state_ == kDecoding || state_ == kIdle)) { | |
498 input_ready_.Wait(); | |
499 } | |
500 | |
501 // We could have got woken up in a different state or never got to sleep | |
502 // due to current state; check for that. | |
503 switch (state_) { | |
504 case kFlushing: | |
505 // Here we are only interested in finishing up decoding buffers that are | |
506 // already queued up. Otherwise will stop decoding. | |
507 if (input_buffers_.empty()) | |
508 return false; | |
509 // else fallthrough | |
510 case kDecoding: | |
511 case kIdle: | |
512 DCHECK(!input_buffers_.empty()); | |
513 | |
514 curr_input_buffer_ = input_buffers_.front(); | |
515 input_buffers_.pop(); | |
516 | |
517 DVLOG(4) << "New current bitstream buffer, id: " << curr_input_buffer_->id | |
518 << " size: " << curr_input_buffer_->shm->size(); | |
519 | |
520 decoder_->SetStream( | |
521 static_cast<uint8_t*>(curr_input_buffer_->shm->memory()), | |
522 curr_input_buffer_->shm->size()); | |
523 return true; | |
524 | |
525 default: | |
526 // We got woken up due to being destroyed/reset, ignore any already | |
527 // queued inputs. | |
528 return false; | |
529 } | |
530 } | |
531 | |
532 void VaapiVideoDecodeAccelerator::ReturnCurrInputBuffer_Locked() { | |
533 lock_.AssertAcquired(); | |
534 DCHECK(decoder_thread_task_runner_->BelongsToCurrentThread()); | |
535 DCHECK(curr_input_buffer_.get()); | |
536 | |
537 int32_t id = curr_input_buffer_->id; | |
538 curr_input_buffer_.reset(); | |
539 DVLOG(4) << "End of input buffer " << id; | |
540 message_loop_->PostTask(FROM_HERE, base::Bind( | |
541 &Client::NotifyEndOfBitstreamBuffer, client_, id)); | |
542 | |
543 --num_stream_bufs_at_decoder_; | |
544 TRACE_COUNTER1("Video Decoder", "Stream buffers at decoder", | |
545 num_stream_bufs_at_decoder_); | |
546 } | |
547 | |
548 // TODO(posciak): refactor the whole class to remove sleeping in wait for | |
549 // surfaces, and reschedule DecodeTask instead. | |
550 bool VaapiVideoDecodeAccelerator::WaitForSurfaces_Locked() { | |
551 lock_.AssertAcquired(); | |
552 DCHECK(decoder_thread_task_runner_->BelongsToCurrentThread()); | |
553 | |
554 while (available_va_surfaces_.empty() && | |
555 (state_ == kDecoding || state_ == kFlushing || state_ == kIdle)) { | |
556 surfaces_available_.Wait(); | |
557 } | |
558 | |
559 if (state_ != kDecoding && state_ != kFlushing && state_ != kIdle) | |
560 return false; | |
561 | |
562 return true; | |
563 } | |
564 | |
565 void VaapiVideoDecodeAccelerator::DecodeTask() { | |
566 DCHECK(decoder_thread_task_runner_->BelongsToCurrentThread()); | |
567 TRACE_EVENT0("Video Decoder", "VAVDA::DecodeTask"); | |
568 base::AutoLock auto_lock(lock_); | |
569 | |
570 if (state_ != kDecoding) | |
571 return; | |
572 | |
573 // Main decode task. | |
574 DVLOG(4) << "Decode task"; | |
575 | |
576 // Try to decode what stream data is (still) in the decoder until we run out | |
577 // of it. | |
578 while (GetInputBuffer_Locked()) { | |
579 DCHECK(curr_input_buffer_.get()); | |
580 | |
581 AcceleratedVideoDecoder::DecodeResult res; | |
582 { | |
583 // We are OK releasing the lock here, as decoder never calls our methods | |
584 // directly and we will reacquire the lock before looking at state again. | |
585 // This is the main decode function of the decoder and while keeping | |
586 // the lock for its duration would be fine, it would defeat the purpose | |
587 // of having a separate decoder thread. | |
588 base::AutoUnlock auto_unlock(lock_); | |
589 res = decoder_->Decode(); | |
590 } | |
591 | |
592 switch (res) { | |
593 case AcceleratedVideoDecoder::kAllocateNewSurfaces: | |
594 DVLOG(1) << "Decoder requesting a new set of surfaces"; | |
595 message_loop_->PostTask(FROM_HERE, base::Bind( | |
596 &VaapiVideoDecodeAccelerator::InitiateSurfaceSetChange, weak_this_, | |
597 decoder_->GetRequiredNumOfPictures(), | |
598 decoder_->GetPicSize())); | |
599 // We'll get rescheduled once ProvidePictureBuffers() finishes. | |
600 return; | |
601 | |
602 case AcceleratedVideoDecoder::kRanOutOfStreamData: | |
603 ReturnCurrInputBuffer_Locked(); | |
604 break; | |
605 | |
606 case AcceleratedVideoDecoder::kRanOutOfSurfaces: | |
607 // No more output buffers in the decoder, try getting more or go to | |
608 // sleep waiting for them. | |
609 if (!WaitForSurfaces_Locked()) | |
610 return; | |
611 | |
612 break; | |
613 | |
614 case AcceleratedVideoDecoder::kDecodeError: | |
615 RETURN_AND_NOTIFY_ON_FAILURE(false, "Error decoding stream", | |
616 PLATFORM_FAILURE, ); | |
617 return; | |
618 } | |
619 } | |
620 } | |
621 | |
622 void VaapiVideoDecodeAccelerator::InitiateSurfaceSetChange(size_t num_pics, | |
623 gfx::Size size) { | |
624 DCHECK_EQ(message_loop_, base::MessageLoop::current()); | |
625 DCHECK(!awaiting_va_surfaces_recycle_); | |
626 | |
627 // At this point decoder has stopped running and has already posted onto our | |
628 // loop any remaining output request callbacks, which executed before we got | |
629 // here. Some of them might have been pended though, because we might not | |
630 // have had enough TFPictures to output surfaces to. Initiate a wait cycle, | |
631 // which will wait for client to return enough PictureBuffers to us, so that | |
632 // we can finish all pending output callbacks, releasing associated surfaces. | |
633 DVLOG(1) << "Initiating surface set change"; | |
634 awaiting_va_surfaces_recycle_ = true; | |
635 | |
636 requested_num_pics_ = num_pics; | |
637 requested_pic_size_ = size; | |
638 | |
639 TryFinishSurfaceSetChange(); | |
640 } | |
641 | |
642 void VaapiVideoDecodeAccelerator::TryFinishSurfaceSetChange() { | |
643 DCHECK_EQ(message_loop_, base::MessageLoop::current()); | |
644 | |
645 if (!awaiting_va_surfaces_recycle_) | |
646 return; | |
647 | |
648 if (!pending_output_cbs_.empty() || | |
649 pictures_.size() != available_va_surfaces_.size()) { | |
650 // Either: | |
651 // 1. Not all pending pending output callbacks have been executed yet. | |
652 // Wait for the client to return enough pictures and retry later. | |
653 // 2. The above happened and all surface release callbacks have been posted | |
654 // as the result, but not all have executed yet. Post ourselves after them | |
655 // to let them release surfaces. | |
656 DVLOG(2) << "Awaiting pending output/surface release callbacks to finish"; | |
657 message_loop_->PostTask(FROM_HERE, base::Bind( | |
658 &VaapiVideoDecodeAccelerator::TryFinishSurfaceSetChange, weak_this_)); | |
659 return; | |
660 } | |
661 | |
662 // All surfaces released, destroy them and dismiss all PictureBuffers. | |
663 awaiting_va_surfaces_recycle_ = false; | |
664 available_va_surfaces_.clear(); | |
665 vaapi_wrapper_->DestroySurfaces(); | |
666 | |
667 for (Pictures::iterator iter = pictures_.begin(); iter != pictures_.end(); | |
668 ++iter) { | |
669 DVLOG(2) << "Dismissing picture id: " << iter->first; | |
670 if (client_) | |
671 client_->DismissPictureBuffer(iter->first); | |
672 } | |
673 pictures_.clear(); | |
674 | |
675 // And ask for a new set as requested. | |
676 DVLOG(1) << "Requesting " << requested_num_pics_ << " pictures of size: " | |
677 << requested_pic_size_.ToString(); | |
678 | |
679 message_loop_->PostTask( | |
680 FROM_HERE, | |
681 base::Bind(&Client::ProvidePictureBuffers, client_, requested_num_pics_, | |
682 1, requested_pic_size_, VaapiPicture::GetGLTextureTarget())); | |
683 } | |
684 | |
685 void VaapiVideoDecodeAccelerator::Decode( | |
686 const media::BitstreamBuffer& bitstream_buffer) { | |
687 DCHECK_EQ(message_loop_, base::MessageLoop::current()); | |
688 | |
689 TRACE_EVENT1("Video Decoder", "VAVDA::Decode", "Buffer id", | |
690 bitstream_buffer.id()); | |
691 | |
692 if (bitstream_buffer.id() < 0) { | |
693 if (base::SharedMemory::IsHandleValid(bitstream_buffer.handle())) | |
694 base::SharedMemory::CloseHandle(bitstream_buffer.handle()); | |
695 LOG(ERROR) << "Invalid bitstream_buffer, id: " << bitstream_buffer.id(); | |
696 NotifyError(INVALID_ARGUMENT); | |
697 return; | |
698 } | |
699 | |
700 // We got a new input buffer from the client, map it and queue for later use. | |
701 MapAndQueueNewInputBuffer(bitstream_buffer); | |
702 | |
703 base::AutoLock auto_lock(lock_); | |
704 switch (state_) { | |
705 case kIdle: | |
706 state_ = kDecoding; | |
707 decoder_thread_task_runner_->PostTask( | |
708 FROM_HERE, base::Bind(&VaapiVideoDecodeAccelerator::DecodeTask, | |
709 base::Unretained(this))); | |
710 break; | |
711 | |
712 case kDecoding: | |
713 // Decoder already running, fallthrough. | |
714 case kResetting: | |
715 // When resetting, allow accumulating bitstream buffers, so that | |
716 // the client can queue after-seek-buffers while we are finishing with | |
717 // the before-seek one. | |
718 break; | |
719 | |
720 default: | |
721 RETURN_AND_NOTIFY_ON_FAILURE(false, | |
722 "Decode request from client in invalid state: " << state_, | |
723 PLATFORM_FAILURE, ); | |
724 break; | |
725 } | |
726 } | |
727 | |
728 void VaapiVideoDecodeAccelerator::RecycleVASurfaceID( | |
729 VASurfaceID va_surface_id) { | |
730 DCHECK_EQ(message_loop_, base::MessageLoop::current()); | |
731 base::AutoLock auto_lock(lock_); | |
732 | |
733 available_va_surfaces_.push_back(va_surface_id); | |
734 surfaces_available_.Signal(); | |
735 } | |
736 | |
737 void VaapiVideoDecodeAccelerator::AssignPictureBuffers( | |
738 const std::vector<media::PictureBuffer>& buffers) { | |
739 DCHECK_EQ(message_loop_, base::MessageLoop::current()); | |
740 | |
741 base::AutoLock auto_lock(lock_); | |
742 DCHECK(pictures_.empty()); | |
743 | |
744 while (!output_buffers_.empty()) | |
745 output_buffers_.pop(); | |
746 | |
747 RETURN_AND_NOTIFY_ON_FAILURE( | |
748 buffers.size() >= requested_num_pics_, | |
749 "Got an invalid number of picture buffers. (Got " << buffers.size() | |
750 << ", requested " << requested_num_pics_ << ")", INVALID_ARGUMENT, ); | |
751 DCHECK(requested_pic_size_ == buffers[0].size()); | |
752 | |
753 std::vector<VASurfaceID> va_surface_ids; | |
754 RETURN_AND_NOTIFY_ON_FAILURE( | |
755 vaapi_wrapper_->CreateSurfaces(VA_RT_FORMAT_YUV420, requested_pic_size_, | |
756 buffers.size(), &va_surface_ids), | |
757 "Failed creating VA Surfaces", PLATFORM_FAILURE, ); | |
758 DCHECK_EQ(va_surface_ids.size(), buffers.size()); | |
759 | |
760 for (size_t i = 0; i < buffers.size(); ++i) { | |
761 DCHECK_LE(1u, buffers[i].texture_ids().size()); | |
762 DVLOG(2) << "Assigning picture id: " << buffers[i].id() | |
763 << " to texture id: " << buffers[i].texture_ids()[0] | |
764 << " VASurfaceID: " << va_surface_ids[i]; | |
765 | |
766 linked_ptr<VaapiPicture> picture(VaapiPicture::CreatePicture( | |
767 vaapi_wrapper_, make_context_current_cb_, buffers[i].id(), | |
768 buffers[i].texture_ids()[0], requested_pic_size_)); | |
769 | |
770 scoped_refptr<gl::GLImage> image = picture->GetImageToBind(); | |
771 if (image && buffers[i].internal_texture_ids().size() > 0) { | |
772 RETURN_AND_NOTIFY_ON_FAILURE( | |
773 bind_image_cb_.Run(buffers[i].internal_texture_ids()[0], | |
774 VaapiPicture::GetGLTextureTarget(), image, true), | |
775 "Failed to bind image", PLATFORM_FAILURE, ); | |
776 } | |
777 | |
778 RETURN_AND_NOTIFY_ON_FAILURE( | |
779 picture.get(), "Failed assigning picture buffer to a texture.", | |
780 PLATFORM_FAILURE, ); | |
781 | |
782 bool inserted = | |
783 pictures_.insert(std::make_pair(buffers[i].id(), picture)).second; | |
784 DCHECK(inserted); | |
785 | |
786 output_buffers_.push(buffers[i].id()); | |
787 available_va_surfaces_.push_back(va_surface_ids[i]); | |
788 surfaces_available_.Signal(); | |
789 } | |
790 | |
791 state_ = kDecoding; | |
792 decoder_thread_task_runner_->PostTask( | |
793 FROM_HERE, base::Bind(&VaapiVideoDecodeAccelerator::DecodeTask, | |
794 base::Unretained(this))); | |
795 } | |
796 | |
797 void VaapiVideoDecodeAccelerator::ReusePictureBuffer( | |
798 int32_t picture_buffer_id) { | |
799 DCHECK_EQ(message_loop_, base::MessageLoop::current()); | |
800 TRACE_EVENT1("Video Decoder", "VAVDA::ReusePictureBuffer", "Picture id", | |
801 picture_buffer_id); | |
802 | |
803 --num_frames_at_client_; | |
804 TRACE_COUNTER1("Video Decoder", "Textures at client", num_frames_at_client_); | |
805 | |
806 output_buffers_.push(picture_buffer_id); | |
807 TryOutputSurface(); | |
808 } | |
809 | |
810 void VaapiVideoDecodeAccelerator::FlushTask() { | |
811 DCHECK(decoder_thread_task_runner_->BelongsToCurrentThread()); | |
812 DVLOG(1) << "Flush task"; | |
813 | |
814 // First flush all the pictures that haven't been outputted, notifying the | |
815 // client to output them. | |
816 bool res = decoder_->Flush(); | |
817 RETURN_AND_NOTIFY_ON_FAILURE(res, "Failed flushing the decoder.", | |
818 PLATFORM_FAILURE, ); | |
819 | |
820 // Put the decoder in idle state, ready to resume. | |
821 decoder_->Reset(); | |
822 | |
823 message_loop_->PostTask(FROM_HERE, base::Bind( | |
824 &VaapiVideoDecodeAccelerator::FinishFlush, weak_this_)); | |
825 } | |
826 | |
827 void VaapiVideoDecodeAccelerator::Flush() { | |
828 DCHECK_EQ(message_loop_, base::MessageLoop::current()); | |
829 DVLOG(1) << "Got flush request"; | |
830 | |
831 base::AutoLock auto_lock(lock_); | |
832 state_ = kFlushing; | |
833 // Queue a flush task after all existing decoding tasks to clean up. | |
834 decoder_thread_task_runner_->PostTask( | |
835 FROM_HERE, base::Bind(&VaapiVideoDecodeAccelerator::FlushTask, | |
836 base::Unretained(this))); | |
837 | |
838 input_ready_.Signal(); | |
839 surfaces_available_.Signal(); | |
840 } | |
841 | |
842 void VaapiVideoDecodeAccelerator::FinishFlush() { | |
843 DCHECK_EQ(message_loop_, base::MessageLoop::current()); | |
844 | |
845 finish_flush_pending_ = false; | |
846 | |
847 base::AutoLock auto_lock(lock_); | |
848 if (state_ != kFlushing) { | |
849 DCHECK_EQ(state_, kDestroying); | |
850 return; // We could've gotten destroyed already. | |
851 } | |
852 | |
853 // Still waiting for textures from client to finish outputting all pending | |
854 // frames. Try again later. | |
855 if (!pending_output_cbs_.empty()) { | |
856 finish_flush_pending_ = true; | |
857 return; | |
858 } | |
859 | |
860 state_ = kIdle; | |
861 | |
862 message_loop_->PostTask(FROM_HERE, base::Bind( | |
863 &Client::NotifyFlushDone, client_)); | |
864 | |
865 DVLOG(1) << "Flush finished"; | |
866 } | |
867 | |
868 void VaapiVideoDecodeAccelerator::ResetTask() { | |
869 DCHECK(decoder_thread_task_runner_->BelongsToCurrentThread()); | |
870 DVLOG(1) << "ResetTask"; | |
871 | |
872 // All the decoding tasks from before the reset request from client are done | |
873 // by now, as this task was scheduled after them and client is expected not | |
874 // to call Decode() after Reset() and before NotifyResetDone. | |
875 decoder_->Reset(); | |
876 | |
877 base::AutoLock auto_lock(lock_); | |
878 | |
879 // Return current input buffer, if present. | |
880 if (curr_input_buffer_.get()) | |
881 ReturnCurrInputBuffer_Locked(); | |
882 | |
883 // And let client know that we are done with reset. | |
884 message_loop_->PostTask(FROM_HERE, base::Bind( | |
885 &VaapiVideoDecodeAccelerator::FinishReset, weak_this_)); | |
886 } | |
887 | |
888 void VaapiVideoDecodeAccelerator::Reset() { | |
889 DCHECK_EQ(message_loop_, base::MessageLoop::current()); | |
890 DVLOG(1) << "Got reset request"; | |
891 | |
892 // This will make any new decode tasks exit early. | |
893 base::AutoLock auto_lock(lock_); | |
894 state_ = kResetting; | |
895 finish_flush_pending_ = false; | |
896 | |
897 // Drop all remaining input buffers, if present. | |
898 while (!input_buffers_.empty()) { | |
899 message_loop_->PostTask(FROM_HERE, base::Bind( | |
900 &Client::NotifyEndOfBitstreamBuffer, client_, | |
901 input_buffers_.front()->id)); | |
902 input_buffers_.pop(); | |
903 } | |
904 | |
905 decoder_thread_task_runner_->PostTask( | |
906 FROM_HERE, base::Bind(&VaapiVideoDecodeAccelerator::ResetTask, | |
907 base::Unretained(this))); | |
908 | |
909 input_ready_.Signal(); | |
910 surfaces_available_.Signal(); | |
911 } | |
912 | |
913 void VaapiVideoDecodeAccelerator::FinishReset() { | |
914 DCHECK_EQ(message_loop_, base::MessageLoop::current()); | |
915 DVLOG(1) << "FinishReset"; | |
916 base::AutoLock auto_lock(lock_); | |
917 | |
918 if (state_ != kResetting) { | |
919 DCHECK(state_ == kDestroying || state_ == kUninitialized) << state_; | |
920 return; // We could've gotten destroyed already. | |
921 } | |
922 | |
923 // Drop pending outputs. | |
924 while (!pending_output_cbs_.empty()) | |
925 pending_output_cbs_.pop(); | |
926 | |
927 if (awaiting_va_surfaces_recycle_) { | |
928 // Decoder requested a new surface set while we were waiting for it to | |
929 // finish the last DecodeTask, running at the time of Reset(). | |
930 // Let the surface set change finish first before resetting. | |
931 message_loop_->PostTask(FROM_HERE, base::Bind( | |
932 &VaapiVideoDecodeAccelerator::FinishReset, weak_this_)); | |
933 return; | |
934 } | |
935 | |
936 num_stream_bufs_at_decoder_ = 0; | |
937 state_ = kIdle; | |
938 | |
939 message_loop_->PostTask(FROM_HERE, base::Bind( | |
940 &Client::NotifyResetDone, client_)); | |
941 | |
942 // The client might have given us new buffers via Decode() while we were | |
943 // resetting and might be waiting for our move, and not call Decode() anymore | |
944 // until we return something. Post a DecodeTask() so that we won't | |
945 // sleep forever waiting for Decode() in that case. Having two of them | |
946 // in the pipe is harmless, the additional one will return as soon as it sees | |
947 // that we are back in kDecoding state. | |
948 if (!input_buffers_.empty()) { | |
949 state_ = kDecoding; | |
950 decoder_thread_task_runner_->PostTask( | |
951 FROM_HERE, base::Bind(&VaapiVideoDecodeAccelerator::DecodeTask, | |
952 base::Unretained(this))); | |
953 } | |
954 | |
955 DVLOG(1) << "Reset finished"; | |
956 } | |
957 | |
958 void VaapiVideoDecodeAccelerator::Cleanup() { | |
959 DCHECK_EQ(message_loop_, base::MessageLoop::current()); | |
960 | |
961 base::AutoLock auto_lock(lock_); | |
962 if (state_ == kUninitialized || state_ == kDestroying) | |
963 return; | |
964 | |
965 DVLOG(1) << "Destroying VAVDA"; | |
966 state_ = kDestroying; | |
967 | |
968 client_ptr_factory_.reset(); | |
969 weak_this_factory_.InvalidateWeakPtrs(); | |
970 | |
971 // Signal all potential waiters on the decoder_thread_, let them early-exit, | |
972 // as we've just moved to the kDestroying state, and wait for all tasks | |
973 // to finish. | |
974 input_ready_.Signal(); | |
975 surfaces_available_.Signal(); | |
976 { | |
977 base::AutoUnlock auto_unlock(lock_); | |
978 decoder_thread_.Stop(); | |
979 } | |
980 | |
981 state_ = kUninitialized; | |
982 } | |
983 | |
984 void VaapiVideoDecodeAccelerator::Destroy() { | |
985 DCHECK_EQ(message_loop_, base::MessageLoop::current()); | |
986 Cleanup(); | |
987 delete this; | |
988 } | |
989 | |
990 bool VaapiVideoDecodeAccelerator::TryToSetupDecodeOnSeparateThread( | |
991 const base::WeakPtr<Client>& decode_client, | |
992 const scoped_refptr<base::SingleThreadTaskRunner>& decode_task_runner) { | |
993 return false; | |
994 } | |
995 | |
996 bool VaapiVideoDecodeAccelerator::DecodeSurface( | |
997 const scoped_refptr<VaapiDecodeSurface>& dec_surface) { | |
998 if (!vaapi_wrapper_->ExecuteAndDestroyPendingBuffers( | |
999 dec_surface->va_surface()->id())) { | |
1000 DVLOG(1) << "Failed decoding picture"; | |
1001 return false; | |
1002 } | |
1003 | |
1004 return true; | |
1005 } | |
1006 | |
1007 void VaapiVideoDecodeAccelerator::SurfaceReady( | |
1008 const scoped_refptr<VaapiDecodeSurface>& dec_surface) { | |
1009 if (message_loop_ != base::MessageLoop::current()) { | |
1010 message_loop_->PostTask( | |
1011 FROM_HERE, base::Bind(&VaapiVideoDecodeAccelerator::SurfaceReady, | |
1012 weak_this_, dec_surface)); | |
1013 return; | |
1014 } | |
1015 | |
1016 DCHECK(!awaiting_va_surfaces_recycle_); | |
1017 | |
1018 { | |
1019 base::AutoLock auto_lock(lock_); | |
1020 // Drop any requests to output if we are resetting or being destroyed. | |
1021 if (state_ == kResetting || state_ == kDestroying) | |
1022 return; | |
1023 } | |
1024 | |
1025 pending_output_cbs_.push( | |
1026 base::Bind(&VaapiVideoDecodeAccelerator::OutputPicture, weak_this_, | |
1027 dec_surface->va_surface(), dec_surface->bitstream_id())); | |
1028 | |
1029 TryOutputSurface(); | |
1030 } | |
1031 | |
1032 scoped_refptr<VaapiVideoDecodeAccelerator::VaapiDecodeSurface> | |
1033 VaapiVideoDecodeAccelerator::CreateSurface() { | |
1034 DCHECK(decoder_thread_task_runner_->BelongsToCurrentThread()); | |
1035 base::AutoLock auto_lock(lock_); | |
1036 | |
1037 if (available_va_surfaces_.empty()) | |
1038 return nullptr; | |
1039 | |
1040 DCHECK(!awaiting_va_surfaces_recycle_); | |
1041 scoped_refptr<VASurface> va_surface(new VASurface( | |
1042 available_va_surfaces_.front(), requested_pic_size_, | |
1043 vaapi_wrapper_->va_surface_format(), va_surface_release_cb_)); | |
1044 available_va_surfaces_.pop_front(); | |
1045 | |
1046 scoped_refptr<VaapiDecodeSurface> dec_surface = | |
1047 new VaapiDecodeSurface(curr_input_buffer_->id, va_surface); | |
1048 | |
1049 return dec_surface; | |
1050 } | |
1051 | |
1052 VaapiVideoDecodeAccelerator::VaapiH264Accelerator::VaapiH264Accelerator( | |
1053 VaapiVideoDecodeAccelerator* vaapi_dec, | |
1054 VaapiWrapper* vaapi_wrapper) | |
1055 : vaapi_wrapper_(vaapi_wrapper), vaapi_dec_(vaapi_dec) { | |
1056 DCHECK(vaapi_wrapper_); | |
1057 DCHECK(vaapi_dec_); | |
1058 } | |
1059 | |
1060 VaapiVideoDecodeAccelerator::VaapiH264Accelerator::~VaapiH264Accelerator() { | |
1061 } | |
1062 | |
1063 scoped_refptr<H264Picture> | |
1064 VaapiVideoDecodeAccelerator::VaapiH264Accelerator::CreateH264Picture() { | |
1065 scoped_refptr<VaapiDecodeSurface> va_surface = vaapi_dec_->CreateSurface(); | |
1066 if (!va_surface) | |
1067 return nullptr; | |
1068 | |
1069 return new VaapiH264Picture(va_surface); | |
1070 } | |
1071 | |
1072 // Fill |va_pic| with default/neutral values. | |
1073 static void InitVAPicture(VAPictureH264* va_pic) { | |
1074 memset(va_pic, 0, sizeof(*va_pic)); | |
1075 va_pic->picture_id = VA_INVALID_ID; | |
1076 va_pic->flags = VA_PICTURE_H264_INVALID; | |
1077 } | |
1078 | |
1079 bool VaapiVideoDecodeAccelerator::VaapiH264Accelerator::SubmitFrameMetadata( | |
1080 const media::H264SPS* sps, | |
1081 const media::H264PPS* pps, | |
1082 const H264DPB& dpb, | |
1083 const H264Picture::Vector& ref_pic_listp0, | |
1084 const H264Picture::Vector& ref_pic_listb0, | |
1085 const H264Picture::Vector& ref_pic_listb1, | |
1086 const scoped_refptr<H264Picture>& pic) { | |
1087 VAPictureParameterBufferH264 pic_param; | |
1088 memset(&pic_param, 0, sizeof(pic_param)); | |
1089 | |
1090 #define FROM_SPS_TO_PP(a) pic_param.a = sps->a | |
1091 #define FROM_SPS_TO_PP2(a, b) pic_param.b = sps->a | |
1092 FROM_SPS_TO_PP2(pic_width_in_mbs_minus1, picture_width_in_mbs_minus1); | |
1093 // This assumes non-interlaced video | |
1094 FROM_SPS_TO_PP2(pic_height_in_map_units_minus1, picture_height_in_mbs_minus1); | |
1095 FROM_SPS_TO_PP(bit_depth_luma_minus8); | |
1096 FROM_SPS_TO_PP(bit_depth_chroma_minus8); | |
1097 #undef FROM_SPS_TO_PP | |
1098 #undef FROM_SPS_TO_PP2 | |
1099 | |
1100 #define FROM_SPS_TO_PP_SF(a) pic_param.seq_fields.bits.a = sps->a | |
1101 #define FROM_SPS_TO_PP_SF2(a, b) pic_param.seq_fields.bits.b = sps->a | |
1102 FROM_SPS_TO_PP_SF(chroma_format_idc); | |
1103 FROM_SPS_TO_PP_SF2(separate_colour_plane_flag, | |
1104 residual_colour_transform_flag); | |
1105 FROM_SPS_TO_PP_SF(gaps_in_frame_num_value_allowed_flag); | |
1106 FROM_SPS_TO_PP_SF(frame_mbs_only_flag); | |
1107 FROM_SPS_TO_PP_SF(mb_adaptive_frame_field_flag); | |
1108 FROM_SPS_TO_PP_SF(direct_8x8_inference_flag); | |
1109 pic_param.seq_fields.bits.MinLumaBiPredSize8x8 = (sps->level_idc >= 31); | |
1110 FROM_SPS_TO_PP_SF(log2_max_frame_num_minus4); | |
1111 FROM_SPS_TO_PP_SF(pic_order_cnt_type); | |
1112 FROM_SPS_TO_PP_SF(log2_max_pic_order_cnt_lsb_minus4); | |
1113 FROM_SPS_TO_PP_SF(delta_pic_order_always_zero_flag); | |
1114 #undef FROM_SPS_TO_PP_SF | |
1115 #undef FROM_SPS_TO_PP_SF2 | |
1116 | |
1117 #define FROM_PPS_TO_PP(a) pic_param.a = pps->a | |
1118 FROM_PPS_TO_PP(num_slice_groups_minus1); | |
1119 pic_param.slice_group_map_type = 0; | |
1120 pic_param.slice_group_change_rate_minus1 = 0; | |
1121 FROM_PPS_TO_PP(pic_init_qp_minus26); | |
1122 FROM_PPS_TO_PP(pic_init_qs_minus26); | |
1123 FROM_PPS_TO_PP(chroma_qp_index_offset); | |
1124 FROM_PPS_TO_PP(second_chroma_qp_index_offset); | |
1125 #undef FROM_PPS_TO_PP | |
1126 | |
1127 #define FROM_PPS_TO_PP_PF(a) pic_param.pic_fields.bits.a = pps->a | |
1128 #define FROM_PPS_TO_PP_PF2(a, b) pic_param.pic_fields.bits.b = pps->a | |
1129 FROM_PPS_TO_PP_PF(entropy_coding_mode_flag); | |
1130 FROM_PPS_TO_PP_PF(weighted_pred_flag); | |
1131 FROM_PPS_TO_PP_PF(weighted_bipred_idc); | |
1132 FROM_PPS_TO_PP_PF(transform_8x8_mode_flag); | |
1133 | |
1134 pic_param.pic_fields.bits.field_pic_flag = 0; | |
1135 FROM_PPS_TO_PP_PF(constrained_intra_pred_flag); | |
1136 FROM_PPS_TO_PP_PF2(bottom_field_pic_order_in_frame_present_flag, | |
1137 pic_order_present_flag); | |
1138 FROM_PPS_TO_PP_PF(deblocking_filter_control_present_flag); | |
1139 FROM_PPS_TO_PP_PF(redundant_pic_cnt_present_flag); | |
1140 pic_param.pic_fields.bits.reference_pic_flag = pic->ref; | |
1141 #undef FROM_PPS_TO_PP_PF | |
1142 #undef FROM_PPS_TO_PP_PF2 | |
1143 | |
1144 pic_param.frame_num = pic->frame_num; | |
1145 | |
1146 InitVAPicture(&pic_param.CurrPic); | |
1147 FillVAPicture(&pic_param.CurrPic, pic); | |
1148 | |
1149 // Init reference pictures' array. | |
1150 for (int i = 0; i < 16; ++i) | |
1151 InitVAPicture(&pic_param.ReferenceFrames[i]); | |
1152 | |
1153 // And fill it with picture info from DPB. | |
1154 FillVARefFramesFromDPB(dpb, pic_param.ReferenceFrames, | |
1155 arraysize(pic_param.ReferenceFrames)); | |
1156 | |
1157 pic_param.num_ref_frames = sps->max_num_ref_frames; | |
1158 | |
1159 if (!vaapi_wrapper_->SubmitBuffer(VAPictureParameterBufferType, | |
1160 sizeof(pic_param), | |
1161 &pic_param)) | |
1162 return false; | |
1163 | |
1164 VAIQMatrixBufferH264 iq_matrix_buf; | |
1165 memset(&iq_matrix_buf, 0, sizeof(iq_matrix_buf)); | |
1166 | |
1167 if (pps->pic_scaling_matrix_present_flag) { | |
1168 for (int i = 0; i < 6; ++i) { | |
1169 for (int j = 0; j < 16; ++j) | |
1170 iq_matrix_buf.ScalingList4x4[i][j] = pps->scaling_list4x4[i][j]; | |
1171 } | |
1172 | |
1173 for (int i = 0; i < 2; ++i) { | |
1174 for (int j = 0; j < 64; ++j) | |
1175 iq_matrix_buf.ScalingList8x8[i][j] = pps->scaling_list8x8[i][j]; | |
1176 } | |
1177 } else { | |
1178 for (int i = 0; i < 6; ++i) { | |
1179 for (int j = 0; j < 16; ++j) | |
1180 iq_matrix_buf.ScalingList4x4[i][j] = sps->scaling_list4x4[i][j]; | |
1181 } | |
1182 | |
1183 for (int i = 0; i < 2; ++i) { | |
1184 for (int j = 0; j < 64; ++j) | |
1185 iq_matrix_buf.ScalingList8x8[i][j] = sps->scaling_list8x8[i][j]; | |
1186 } | |
1187 } | |
1188 | |
1189 return vaapi_wrapper_->SubmitBuffer(VAIQMatrixBufferType, | |
1190 sizeof(iq_matrix_buf), | |
1191 &iq_matrix_buf); | |
1192 } | |
1193 | |
1194 bool VaapiVideoDecodeAccelerator::VaapiH264Accelerator::SubmitSlice( | |
1195 const media::H264PPS* pps, | |
1196 const media::H264SliceHeader* slice_hdr, | |
1197 const H264Picture::Vector& ref_pic_list0, | |
1198 const H264Picture::Vector& ref_pic_list1, | |
1199 const scoped_refptr<H264Picture>& pic, | |
1200 const uint8_t* data, | |
1201 size_t size) { | |
1202 VASliceParameterBufferH264 slice_param; | |
1203 memset(&slice_param, 0, sizeof(slice_param)); | |
1204 | |
1205 slice_param.slice_data_size = slice_hdr->nalu_size; | |
1206 slice_param.slice_data_offset = 0; | |
1207 slice_param.slice_data_flag = VA_SLICE_DATA_FLAG_ALL; | |
1208 slice_param.slice_data_bit_offset = slice_hdr->header_bit_size; | |
1209 | |
1210 #define SHDRToSP(a) slice_param.a = slice_hdr->a | |
1211 SHDRToSP(first_mb_in_slice); | |
1212 slice_param.slice_type = slice_hdr->slice_type % 5; | |
1213 SHDRToSP(direct_spatial_mv_pred_flag); | |
1214 | |
1215 // TODO posciak: make sure parser sets those even when override flags | |
1216 // in slice header is off. | |
1217 SHDRToSP(num_ref_idx_l0_active_minus1); | |
1218 SHDRToSP(num_ref_idx_l1_active_minus1); | |
1219 SHDRToSP(cabac_init_idc); | |
1220 SHDRToSP(slice_qp_delta); | |
1221 SHDRToSP(disable_deblocking_filter_idc); | |
1222 SHDRToSP(slice_alpha_c0_offset_div2); | |
1223 SHDRToSP(slice_beta_offset_div2); | |
1224 | |
1225 if (((slice_hdr->IsPSlice() || slice_hdr->IsSPSlice()) && | |
1226 pps->weighted_pred_flag) || | |
1227 (slice_hdr->IsBSlice() && pps->weighted_bipred_idc == 1)) { | |
1228 SHDRToSP(luma_log2_weight_denom); | |
1229 SHDRToSP(chroma_log2_weight_denom); | |
1230 | |
1231 SHDRToSP(luma_weight_l0_flag); | |
1232 SHDRToSP(luma_weight_l1_flag); | |
1233 | |
1234 SHDRToSP(chroma_weight_l0_flag); | |
1235 SHDRToSP(chroma_weight_l1_flag); | |
1236 | |
1237 for (int i = 0; i <= slice_param.num_ref_idx_l0_active_minus1; ++i) { | |
1238 slice_param.luma_weight_l0[i] = | |
1239 slice_hdr->pred_weight_table_l0.luma_weight[i]; | |
1240 slice_param.luma_offset_l0[i] = | |
1241 slice_hdr->pred_weight_table_l0.luma_offset[i]; | |
1242 | |
1243 for (int j = 0; j < 2; ++j) { | |
1244 slice_param.chroma_weight_l0[i][j] = | |
1245 slice_hdr->pred_weight_table_l0.chroma_weight[i][j]; | |
1246 slice_param.chroma_offset_l0[i][j] = | |
1247 slice_hdr->pred_weight_table_l0.chroma_offset[i][j]; | |
1248 } | |
1249 } | |
1250 | |
1251 if (slice_hdr->IsBSlice()) { | |
1252 for (int i = 0; i <= slice_param.num_ref_idx_l1_active_minus1; ++i) { | |
1253 slice_param.luma_weight_l1[i] = | |
1254 slice_hdr->pred_weight_table_l1.luma_weight[i]; | |
1255 slice_param.luma_offset_l1[i] = | |
1256 slice_hdr->pred_weight_table_l1.luma_offset[i]; | |
1257 | |
1258 for (int j = 0; j < 2; ++j) { | |
1259 slice_param.chroma_weight_l1[i][j] = | |
1260 slice_hdr->pred_weight_table_l1.chroma_weight[i][j]; | |
1261 slice_param.chroma_offset_l1[i][j] = | |
1262 slice_hdr->pred_weight_table_l1.chroma_offset[i][j]; | |
1263 } | |
1264 } | |
1265 } | |
1266 } | |
1267 | |
1268 static_assert( | |
1269 arraysize(slice_param.RefPicList0) == arraysize(slice_param.RefPicList1), | |
1270 "Invalid RefPicList sizes"); | |
1271 | |
1272 for (size_t i = 0; i < arraysize(slice_param.RefPicList0); ++i) { | |
1273 InitVAPicture(&slice_param.RefPicList0[i]); | |
1274 InitVAPicture(&slice_param.RefPicList1[i]); | |
1275 } | |
1276 | |
1277 for (size_t i = 0; | |
1278 i < ref_pic_list0.size() && i < arraysize(slice_param.RefPicList0); | |
1279 ++i) { | |
1280 if (ref_pic_list0[i]) | |
1281 FillVAPicture(&slice_param.RefPicList0[i], ref_pic_list0[i]); | |
1282 } | |
1283 for (size_t i = 0; | |
1284 i < ref_pic_list1.size() && i < arraysize(slice_param.RefPicList1); | |
1285 ++i) { | |
1286 if (ref_pic_list1[i]) | |
1287 FillVAPicture(&slice_param.RefPicList1[i], ref_pic_list1[i]); | |
1288 } | |
1289 | |
1290 if (!vaapi_wrapper_->SubmitBuffer(VASliceParameterBufferType, | |
1291 sizeof(slice_param), | |
1292 &slice_param)) | |
1293 return false; | |
1294 | |
1295 // Can't help it, blame libva... | |
1296 void* non_const_ptr = const_cast<uint8_t*>(data); | |
1297 return vaapi_wrapper_->SubmitBuffer(VASliceDataBufferType, size, | |
1298 non_const_ptr); | |
1299 } | |
1300 | |
1301 bool VaapiVideoDecodeAccelerator::VaapiH264Accelerator::SubmitDecode( | |
1302 const scoped_refptr<H264Picture>& pic) { | |
1303 DVLOG(4) << "Decoding POC " << pic->pic_order_cnt; | |
1304 scoped_refptr<VaapiDecodeSurface> dec_surface = | |
1305 H264PictureToVaapiDecodeSurface(pic); | |
1306 | |
1307 return vaapi_dec_->DecodeSurface(dec_surface); | |
1308 } | |
1309 | |
1310 bool VaapiVideoDecodeAccelerator::VaapiH264Accelerator::OutputPicture( | |
1311 const scoped_refptr<H264Picture>& pic) { | |
1312 scoped_refptr<VaapiDecodeSurface> dec_surface = | |
1313 H264PictureToVaapiDecodeSurface(pic); | |
1314 | |
1315 vaapi_dec_->SurfaceReady(dec_surface); | |
1316 | |
1317 return true; | |
1318 } | |
1319 | |
1320 void VaapiVideoDecodeAccelerator::VaapiH264Accelerator::Reset() { | |
1321 vaapi_wrapper_->DestroyPendingBuffers(); | |
1322 } | |
1323 | |
1324 scoped_refptr<VaapiVideoDecodeAccelerator::VaapiDecodeSurface> | |
1325 VaapiVideoDecodeAccelerator::VaapiH264Accelerator:: | |
1326 H264PictureToVaapiDecodeSurface(const scoped_refptr<H264Picture>& pic) { | |
1327 VaapiH264Picture* vaapi_pic = pic->AsVaapiH264Picture(); | |
1328 CHECK(vaapi_pic); | |
1329 return vaapi_pic->dec_surface(); | |
1330 } | |
1331 | |
1332 void VaapiVideoDecodeAccelerator::VaapiH264Accelerator::FillVAPicture( | |
1333 VAPictureH264* va_pic, | |
1334 scoped_refptr<H264Picture> pic) { | |
1335 VASurfaceID va_surface_id = VA_INVALID_SURFACE; | |
1336 | |
1337 if (!pic->nonexisting) { | |
1338 scoped_refptr<VaapiDecodeSurface> dec_surface = | |
1339 H264PictureToVaapiDecodeSurface(pic); | |
1340 va_surface_id = dec_surface->va_surface()->id(); | |
1341 } | |
1342 | |
1343 va_pic->picture_id = va_surface_id; | |
1344 va_pic->frame_idx = pic->frame_num; | |
1345 va_pic->flags = 0; | |
1346 | |
1347 switch (pic->field) { | |
1348 case H264Picture::FIELD_NONE: | |
1349 break; | |
1350 case H264Picture::FIELD_TOP: | |
1351 va_pic->flags |= VA_PICTURE_H264_TOP_FIELD; | |
1352 break; | |
1353 case H264Picture::FIELD_BOTTOM: | |
1354 va_pic->flags |= VA_PICTURE_H264_BOTTOM_FIELD; | |
1355 break; | |
1356 } | |
1357 | |
1358 if (pic->ref) { | |
1359 va_pic->flags |= pic->long_term ? VA_PICTURE_H264_LONG_TERM_REFERENCE | |
1360 : VA_PICTURE_H264_SHORT_TERM_REFERENCE; | |
1361 } | |
1362 | |
1363 va_pic->TopFieldOrderCnt = pic->top_field_order_cnt; | |
1364 va_pic->BottomFieldOrderCnt = pic->bottom_field_order_cnt; | |
1365 } | |
1366 | |
1367 int VaapiVideoDecodeAccelerator::VaapiH264Accelerator::FillVARefFramesFromDPB( | |
1368 const H264DPB& dpb, | |
1369 VAPictureH264* va_pics, | |
1370 int num_pics) { | |
1371 H264Picture::Vector::const_reverse_iterator rit; | |
1372 int i; | |
1373 | |
1374 // Return reference frames in reverse order of insertion. | |
1375 // Libva does not document this, but other implementations (e.g. mplayer) | |
1376 // do it this way as well. | |
1377 for (rit = dpb.rbegin(), i = 0; rit != dpb.rend() && i < num_pics; ++rit) { | |
1378 if ((*rit)->ref) | |
1379 FillVAPicture(&va_pics[i++], *rit); | |
1380 } | |
1381 | |
1382 return i; | |
1383 } | |
1384 | |
1385 VaapiVideoDecodeAccelerator::VaapiVP8Accelerator::VaapiVP8Accelerator( | |
1386 VaapiVideoDecodeAccelerator* vaapi_dec, | |
1387 VaapiWrapper* vaapi_wrapper) | |
1388 : vaapi_wrapper_(vaapi_wrapper), vaapi_dec_(vaapi_dec) { | |
1389 DCHECK(vaapi_wrapper_); | |
1390 DCHECK(vaapi_dec_); | |
1391 } | |
1392 | |
1393 VaapiVideoDecodeAccelerator::VaapiVP8Accelerator::~VaapiVP8Accelerator() { | |
1394 } | |
1395 | |
1396 scoped_refptr<VP8Picture> | |
1397 VaapiVideoDecodeAccelerator::VaapiVP8Accelerator::CreateVP8Picture() { | |
1398 scoped_refptr<VaapiDecodeSurface> va_surface = vaapi_dec_->CreateSurface(); | |
1399 if (!va_surface) | |
1400 return nullptr; | |
1401 | |
1402 return new VaapiVP8Picture(va_surface); | |
1403 } | |
1404 | |
1405 #define ARRAY_MEMCPY_CHECKED(to, from) \ | |
1406 do { \ | |
1407 static_assert(sizeof(to) == sizeof(from), \ | |
1408 #from " and " #to " arrays must be of same size"); \ | |
1409 memcpy(to, from, sizeof(to)); \ | |
1410 } while (0) | |
1411 | |
1412 bool VaapiVideoDecodeAccelerator::VaapiVP8Accelerator::SubmitDecode( | |
1413 const scoped_refptr<VP8Picture>& pic, | |
1414 const media::Vp8FrameHeader* frame_hdr, | |
1415 const scoped_refptr<VP8Picture>& last_frame, | |
1416 const scoped_refptr<VP8Picture>& golden_frame, | |
1417 const scoped_refptr<VP8Picture>& alt_frame) { | |
1418 VAIQMatrixBufferVP8 iq_matrix_buf; | |
1419 memset(&iq_matrix_buf, 0, sizeof(VAIQMatrixBufferVP8)); | |
1420 | |
1421 const media::Vp8SegmentationHeader& sgmnt_hdr = frame_hdr->segmentation_hdr; | |
1422 const media::Vp8QuantizationHeader& quant_hdr = frame_hdr->quantization_hdr; | |
1423 static_assert( | |
1424 arraysize(iq_matrix_buf.quantization_index) == media::kMaxMBSegments, | |
1425 "incorrect quantization matrix size"); | |
1426 for (size_t i = 0; i < media::kMaxMBSegments; ++i) { | |
1427 int q = quant_hdr.y_ac_qi; | |
1428 | |
1429 if (sgmnt_hdr.segmentation_enabled) { | |
1430 if (sgmnt_hdr.segment_feature_mode == | |
1431 media::Vp8SegmentationHeader::FEATURE_MODE_ABSOLUTE) | |
1432 q = sgmnt_hdr.quantizer_update_value[i]; | |
1433 else | |
1434 q += sgmnt_hdr.quantizer_update_value[i]; | |
1435 } | |
1436 | |
1437 #define CLAMP_Q(q) std::min(std::max(q, 0), 127) | |
1438 static_assert(arraysize(iq_matrix_buf.quantization_index[i]) == 6, | |
1439 "incorrect quantization matrix size"); | |
1440 iq_matrix_buf.quantization_index[i][0] = CLAMP_Q(q); | |
1441 iq_matrix_buf.quantization_index[i][1] = CLAMP_Q(q + quant_hdr.y_dc_delta); | |
1442 iq_matrix_buf.quantization_index[i][2] = CLAMP_Q(q + quant_hdr.y2_dc_delta); | |
1443 iq_matrix_buf.quantization_index[i][3] = CLAMP_Q(q + quant_hdr.y2_ac_delta); | |
1444 iq_matrix_buf.quantization_index[i][4] = CLAMP_Q(q + quant_hdr.uv_dc_delta); | |
1445 iq_matrix_buf.quantization_index[i][5] = CLAMP_Q(q + quant_hdr.uv_ac_delta); | |
1446 #undef CLAMP_Q | |
1447 } | |
1448 | |
1449 if (!vaapi_wrapper_->SubmitBuffer(VAIQMatrixBufferType, | |
1450 sizeof(VAIQMatrixBufferVP8), | |
1451 &iq_matrix_buf)) | |
1452 return false; | |
1453 | |
1454 VAProbabilityDataBufferVP8 prob_buf; | |
1455 memset(&prob_buf, 0, sizeof(VAProbabilityDataBufferVP8)); | |
1456 | |
1457 const media::Vp8EntropyHeader& entr_hdr = frame_hdr->entropy_hdr; | |
1458 ARRAY_MEMCPY_CHECKED(prob_buf.dct_coeff_probs, entr_hdr.coeff_probs); | |
1459 | |
1460 if (!vaapi_wrapper_->SubmitBuffer(VAProbabilityBufferType, | |
1461 sizeof(VAProbabilityDataBufferVP8), | |
1462 &prob_buf)) | |
1463 return false; | |
1464 | |
1465 VAPictureParameterBufferVP8 pic_param; | |
1466 memset(&pic_param, 0, sizeof(VAPictureParameterBufferVP8)); | |
1467 pic_param.frame_width = frame_hdr->width; | |
1468 pic_param.frame_height = frame_hdr->height; | |
1469 | |
1470 if (last_frame) { | |
1471 scoped_refptr<VaapiDecodeSurface> last_frame_surface = | |
1472 VP8PictureToVaapiDecodeSurface(last_frame); | |
1473 pic_param.last_ref_frame = last_frame_surface->va_surface()->id(); | |
1474 } else { | |
1475 pic_param.last_ref_frame = VA_INVALID_SURFACE; | |
1476 } | |
1477 | |
1478 if (golden_frame) { | |
1479 scoped_refptr<VaapiDecodeSurface> golden_frame_surface = | |
1480 VP8PictureToVaapiDecodeSurface(golden_frame); | |
1481 pic_param.golden_ref_frame = golden_frame_surface->va_surface()->id(); | |
1482 } else { | |
1483 pic_param.golden_ref_frame = VA_INVALID_SURFACE; | |
1484 } | |
1485 | |
1486 if (alt_frame) { | |
1487 scoped_refptr<VaapiDecodeSurface> alt_frame_surface = | |
1488 VP8PictureToVaapiDecodeSurface(alt_frame); | |
1489 pic_param.alt_ref_frame = alt_frame_surface->va_surface()->id(); | |
1490 } else { | |
1491 pic_param.alt_ref_frame = VA_INVALID_SURFACE; | |
1492 } | |
1493 | |
1494 pic_param.out_of_loop_frame = VA_INVALID_SURFACE; | |
1495 | |
1496 const media::Vp8LoopFilterHeader& lf_hdr = frame_hdr->loopfilter_hdr; | |
1497 | |
1498 #define FHDR_TO_PP_PF(a, b) pic_param.pic_fields.bits.a = (b) | |
1499 FHDR_TO_PP_PF(key_frame, frame_hdr->IsKeyframe() ? 0 : 1); | |
1500 FHDR_TO_PP_PF(version, frame_hdr->version); | |
1501 FHDR_TO_PP_PF(segmentation_enabled, sgmnt_hdr.segmentation_enabled); | |
1502 FHDR_TO_PP_PF(update_mb_segmentation_map, | |
1503 sgmnt_hdr.update_mb_segmentation_map); | |
1504 FHDR_TO_PP_PF(update_segment_feature_data, | |
1505 sgmnt_hdr.update_segment_feature_data); | |
1506 FHDR_TO_PP_PF(filter_type, lf_hdr.type); | |
1507 FHDR_TO_PP_PF(sharpness_level, lf_hdr.sharpness_level); | |
1508 FHDR_TO_PP_PF(loop_filter_adj_enable, lf_hdr.loop_filter_adj_enable); | |
1509 FHDR_TO_PP_PF(mode_ref_lf_delta_update, lf_hdr.mode_ref_lf_delta_update); | |
1510 FHDR_TO_PP_PF(sign_bias_golden, frame_hdr->sign_bias_golden); | |
1511 FHDR_TO_PP_PF(sign_bias_alternate, frame_hdr->sign_bias_alternate); | |
1512 FHDR_TO_PP_PF(mb_no_coeff_skip, frame_hdr->mb_no_skip_coeff); | |
1513 FHDR_TO_PP_PF(loop_filter_disable, lf_hdr.level == 0); | |
1514 #undef FHDR_TO_PP_PF | |
1515 | |
1516 ARRAY_MEMCPY_CHECKED(pic_param.mb_segment_tree_probs, sgmnt_hdr.segment_prob); | |
1517 | |
1518 static_assert(arraysize(sgmnt_hdr.lf_update_value) == | |
1519 arraysize(pic_param.loop_filter_level), | |
1520 "loop filter level arrays mismatch"); | |
1521 for (size_t i = 0; i < arraysize(sgmnt_hdr.lf_update_value); ++i) { | |
1522 int lf_level = lf_hdr.level; | |
1523 if (sgmnt_hdr.segmentation_enabled) { | |
1524 if (sgmnt_hdr.segment_feature_mode == | |
1525 media::Vp8SegmentationHeader::FEATURE_MODE_ABSOLUTE) | |
1526 lf_level = sgmnt_hdr.lf_update_value[i]; | |
1527 else | |
1528 lf_level += sgmnt_hdr.lf_update_value[i]; | |
1529 } | |
1530 | |
1531 // Clamp to [0..63] range. | |
1532 lf_level = std::min(std::max(lf_level, 0), 63); | |
1533 pic_param.loop_filter_level[i] = lf_level; | |
1534 } | |
1535 | |
1536 static_assert(arraysize(lf_hdr.ref_frame_delta) == | |
1537 arraysize(pic_param.loop_filter_deltas_ref_frame) && | |
1538 arraysize(lf_hdr.mb_mode_delta) == | |
1539 arraysize(pic_param.loop_filter_deltas_mode) && | |
1540 arraysize(lf_hdr.ref_frame_delta) == | |
1541 arraysize(lf_hdr.mb_mode_delta), | |
1542 "loop filter deltas arrays size mismatch"); | |
1543 for (size_t i = 0; i < arraysize(lf_hdr.ref_frame_delta); ++i) { | |
1544 pic_param.loop_filter_deltas_ref_frame[i] = lf_hdr.ref_frame_delta[i]; | |
1545 pic_param.loop_filter_deltas_mode[i] = lf_hdr.mb_mode_delta[i]; | |
1546 } | |
1547 | |
1548 #define FHDR_TO_PP(a) pic_param.a = frame_hdr->a | |
1549 FHDR_TO_PP(prob_skip_false); | |
1550 FHDR_TO_PP(prob_intra); | |
1551 FHDR_TO_PP(prob_last); | |
1552 FHDR_TO_PP(prob_gf); | |
1553 #undef FHDR_TO_PP | |
1554 | |
1555 ARRAY_MEMCPY_CHECKED(pic_param.y_mode_probs, entr_hdr.y_mode_probs); | |
1556 ARRAY_MEMCPY_CHECKED(pic_param.uv_mode_probs, entr_hdr.uv_mode_probs); | |
1557 ARRAY_MEMCPY_CHECKED(pic_param.mv_probs, entr_hdr.mv_probs); | |
1558 | |
1559 pic_param.bool_coder_ctx.range = frame_hdr->bool_dec_range; | |
1560 pic_param.bool_coder_ctx.value = frame_hdr->bool_dec_value; | |
1561 pic_param.bool_coder_ctx.count = frame_hdr->bool_dec_count; | |
1562 | |
1563 if (!vaapi_wrapper_->SubmitBuffer(VAPictureParameterBufferType, | |
1564 sizeof(pic_param), &pic_param)) | |
1565 return false; | |
1566 | |
1567 VASliceParameterBufferVP8 slice_param; | |
1568 memset(&slice_param, 0, sizeof(slice_param)); | |
1569 slice_param.slice_data_size = frame_hdr->frame_size; | |
1570 slice_param.slice_data_offset = frame_hdr->first_part_offset; | |
1571 slice_param.slice_data_flag = VA_SLICE_DATA_FLAG_ALL; | |
1572 slice_param.macroblock_offset = frame_hdr->macroblock_bit_offset; | |
1573 // Number of DCT partitions plus control partition. | |
1574 slice_param.num_of_partitions = frame_hdr->num_of_dct_partitions + 1; | |
1575 | |
1576 // Per VAAPI, this size only includes the size of the macroblock data in | |
1577 // the first partition (in bytes), so we have to subtract the header size. | |
1578 slice_param.partition_size[0] = | |
1579 frame_hdr->first_part_size - ((frame_hdr->macroblock_bit_offset + 7) / 8); | |
1580 | |
1581 for (size_t i = 0; i < frame_hdr->num_of_dct_partitions; ++i) | |
1582 slice_param.partition_size[i + 1] = frame_hdr->dct_partition_sizes[i]; | |
1583 | |
1584 if (!vaapi_wrapper_->SubmitBuffer(VASliceParameterBufferType, | |
1585 sizeof(VASliceParameterBufferVP8), | |
1586 &slice_param)) | |
1587 return false; | |
1588 | |
1589 void* non_const_ptr = const_cast<uint8_t*>(frame_hdr->data); | |
1590 if (!vaapi_wrapper_->SubmitBuffer(VASliceDataBufferType, | |
1591 frame_hdr->frame_size, | |
1592 non_const_ptr)) | |
1593 return false; | |
1594 | |
1595 scoped_refptr<VaapiDecodeSurface> dec_surface = | |
1596 VP8PictureToVaapiDecodeSurface(pic); | |
1597 | |
1598 return vaapi_dec_->DecodeSurface(dec_surface); | |
1599 } | |
1600 | |
1601 bool VaapiVideoDecodeAccelerator::VaapiVP8Accelerator::OutputPicture( | |
1602 const scoped_refptr<VP8Picture>& pic) { | |
1603 scoped_refptr<VaapiDecodeSurface> dec_surface = | |
1604 VP8PictureToVaapiDecodeSurface(pic); | |
1605 | |
1606 vaapi_dec_->SurfaceReady(dec_surface); | |
1607 return true; | |
1608 } | |
1609 | |
1610 scoped_refptr<VaapiVideoDecodeAccelerator::VaapiDecodeSurface> | |
1611 VaapiVideoDecodeAccelerator::VaapiVP8Accelerator:: | |
1612 VP8PictureToVaapiDecodeSurface(const scoped_refptr<VP8Picture>& pic) { | |
1613 VaapiVP8Picture* vaapi_pic = pic->AsVaapiVP8Picture(); | |
1614 CHECK(vaapi_pic); | |
1615 return vaapi_pic->dec_surface(); | |
1616 } | |
1617 | |
1618 VaapiVideoDecodeAccelerator::VaapiVP9Accelerator::VaapiVP9Accelerator( | |
1619 VaapiVideoDecodeAccelerator* vaapi_dec, | |
1620 VaapiWrapper* vaapi_wrapper) | |
1621 : vaapi_wrapper_(vaapi_wrapper), vaapi_dec_(vaapi_dec) { | |
1622 DCHECK(vaapi_wrapper_); | |
1623 DCHECK(vaapi_dec_); | |
1624 } | |
1625 | |
1626 VaapiVideoDecodeAccelerator::VaapiVP9Accelerator::~VaapiVP9Accelerator() {} | |
1627 | |
1628 scoped_refptr<VP9Picture> | |
1629 VaapiVideoDecodeAccelerator::VaapiVP9Accelerator::CreateVP9Picture() { | |
1630 scoped_refptr<VaapiDecodeSurface> va_surface = vaapi_dec_->CreateSurface(); | |
1631 if (!va_surface) | |
1632 return nullptr; | |
1633 | |
1634 return new VaapiVP9Picture(va_surface); | |
1635 } | |
1636 | |
1637 bool VaapiVideoDecodeAccelerator::VaapiVP9Accelerator::SubmitDecode( | |
1638 const scoped_refptr<VP9Picture>& pic, | |
1639 const media::Vp9Segmentation& seg, | |
1640 const media::Vp9LoopFilter& lf, | |
1641 const std::vector<scoped_refptr<VP9Picture>>& ref_pictures) { | |
1642 VADecPictureParameterBufferVP9 pic_param; | |
1643 memset(&pic_param, 0, sizeof(pic_param)); | |
1644 | |
1645 const media::Vp9FrameHeader* frame_hdr = pic->frame_hdr.get(); | |
1646 DCHECK(frame_hdr); | |
1647 | |
1648 if (frame_hdr->profile != 0) { | |
1649 DVLOG(1) << "Unsupported profile" << frame_hdr->profile; | |
1650 return false; | |
1651 } | |
1652 | |
1653 pic_param.frame_width = base::checked_cast<uint16_t>(frame_hdr->width); | |
1654 pic_param.frame_height = base::checked_cast<uint16_t>(frame_hdr->height); | |
1655 | |
1656 CHECK_EQ(ref_pictures.size(), arraysize(pic_param.reference_frames)); | |
1657 for (size_t i = 0; i < arraysize(pic_param.reference_frames); ++i) { | |
1658 VASurfaceID va_surface_id; | |
1659 if (ref_pictures[i]) { | |
1660 scoped_refptr<VaapiDecodeSurface> surface = | |
1661 VP9PictureToVaapiDecodeSurface(ref_pictures[i]); | |
1662 va_surface_id = surface->va_surface()->id(); | |
1663 } else { | |
1664 va_surface_id = VA_INVALID_SURFACE; | |
1665 } | |
1666 | |
1667 pic_param.reference_frames[i] = va_surface_id; | |
1668 } | |
1669 | |
1670 #define FHDR_TO_PP_PF1(a) pic_param.pic_fields.bits.a = frame_hdr->a | |
1671 #define FHDR_TO_PP_PF2(a, b) pic_param.pic_fields.bits.a = b | |
1672 FHDR_TO_PP_PF2(subsampling_x, frame_hdr->subsampling_x == 1); | |
1673 FHDR_TO_PP_PF2(subsampling_y, frame_hdr->subsampling_y == 1); | |
1674 FHDR_TO_PP_PF2(frame_type, frame_hdr->IsKeyframe() ? 0 : 1); | |
1675 FHDR_TO_PP_PF1(show_frame); | |
1676 FHDR_TO_PP_PF1(error_resilient_mode); | |
1677 FHDR_TO_PP_PF1(intra_only); | |
1678 FHDR_TO_PP_PF1(allow_high_precision_mv); | |
1679 FHDR_TO_PP_PF2(mcomp_filter_type, frame_hdr->interp_filter); | |
1680 FHDR_TO_PP_PF1(frame_parallel_decoding_mode); | |
1681 FHDR_TO_PP_PF2(reset_frame_context, frame_hdr->reset_context); | |
1682 FHDR_TO_PP_PF1(refresh_frame_context); | |
1683 FHDR_TO_PP_PF1(frame_context_idx); | |
1684 FHDR_TO_PP_PF2(segmentation_enabled, seg.enabled); | |
1685 FHDR_TO_PP_PF2(segmentation_temporal_update, seg.temporal_update); | |
1686 FHDR_TO_PP_PF2(segmentation_update_map, seg.update_map); | |
1687 FHDR_TO_PP_PF2(last_ref_frame, frame_hdr->frame_refs[0]); | |
1688 FHDR_TO_PP_PF2(last_ref_frame_sign_bias, frame_hdr->ref_sign_biases[0]); | |
1689 FHDR_TO_PP_PF2(golden_ref_frame, frame_hdr->frame_refs[1]); | |
1690 FHDR_TO_PP_PF2(golden_ref_frame_sign_bias, frame_hdr->ref_sign_biases[1]); | |
1691 FHDR_TO_PP_PF2(alt_ref_frame, frame_hdr->frame_refs[2]); | |
1692 FHDR_TO_PP_PF2(alt_ref_frame_sign_bias, frame_hdr->ref_sign_biases[2]); | |
1693 FHDR_TO_PP_PF2(lossless_flag, frame_hdr->quant_params.IsLossless()); | |
1694 #undef FHDR_TO_PP_PF2 | |
1695 #undef FHDR_TO_PP_PF1 | |
1696 | |
1697 pic_param.filter_level = lf.filter_level; | |
1698 pic_param.sharpness_level = lf.sharpness_level; | |
1699 pic_param.log2_tile_rows = frame_hdr->log2_tile_rows; | |
1700 pic_param.log2_tile_columns = frame_hdr->log2_tile_cols; | |
1701 pic_param.frame_header_length_in_bytes = frame_hdr->uncompressed_header_size; | |
1702 pic_param.first_partition_size = frame_hdr->first_partition_size; | |
1703 | |
1704 ARRAY_MEMCPY_CHECKED(pic_param.mb_segment_tree_probs, seg.tree_probs); | |
1705 ARRAY_MEMCPY_CHECKED(pic_param.segment_pred_probs, seg.pred_probs); | |
1706 | |
1707 pic_param.profile = frame_hdr->profile; | |
1708 | |
1709 if (!vaapi_wrapper_->SubmitBuffer(VAPictureParameterBufferType, | |
1710 sizeof(pic_param), &pic_param)) | |
1711 return false; | |
1712 | |
1713 VASliceParameterBufferVP9 slice_param; | |
1714 memset(&slice_param, 0, sizeof(slice_param)); | |
1715 slice_param.slice_data_size = frame_hdr->frame_size; | |
1716 slice_param.slice_data_offset = 0; | |
1717 slice_param.slice_data_flag = VA_SLICE_DATA_FLAG_ALL; | |
1718 | |
1719 static_assert(arraysize(media::Vp9Segmentation::feature_enabled) == | |
1720 arraysize(slice_param.seg_param), | |
1721 "seg_param array of incorrect size"); | |
1722 for (size_t i = 0; i < arraysize(slice_param.seg_param); ++i) { | |
1723 VASegmentParameterVP9& seg_param = slice_param.seg_param[i]; | |
1724 #define SEG_TO_SP_SF(a, b) seg_param.segment_flags.fields.a = b | |
1725 SEG_TO_SP_SF( | |
1726 segment_reference_enabled, | |
1727 seg.FeatureEnabled(i, media::Vp9Segmentation::SEG_LVL_REF_FRAME)); | |
1728 SEG_TO_SP_SF(segment_reference, | |
1729 seg.FeatureData(i, media::Vp9Segmentation::SEG_LVL_REF_FRAME)); | |
1730 SEG_TO_SP_SF(segment_reference_skipped, | |
1731 seg.FeatureEnabled(i, media::Vp9Segmentation::SEG_LVL_SKIP)); | |
1732 #undef SEG_TO_SP_SF | |
1733 | |
1734 ARRAY_MEMCPY_CHECKED(seg_param.filter_level, lf.lvl[i]); | |
1735 | |
1736 seg_param.luma_dc_quant_scale = seg.y_dequant[i][0]; | |
1737 seg_param.luma_ac_quant_scale = seg.y_dequant[i][1]; | |
1738 seg_param.chroma_dc_quant_scale = seg.uv_dequant[i][0]; | |
1739 seg_param.chroma_ac_quant_scale = seg.uv_dequant[i][1]; | |
1740 } | |
1741 | |
1742 if (!vaapi_wrapper_->SubmitBuffer(VASliceParameterBufferType, | |
1743 sizeof(slice_param), &slice_param)) | |
1744 return false; | |
1745 | |
1746 void* non_const_ptr = const_cast<uint8_t*>(frame_hdr->data); | |
1747 if (!vaapi_wrapper_->SubmitBuffer(VASliceDataBufferType, | |
1748 frame_hdr->frame_size, non_const_ptr)) | |
1749 return false; | |
1750 | |
1751 scoped_refptr<VaapiDecodeSurface> dec_surface = | |
1752 VP9PictureToVaapiDecodeSurface(pic); | |
1753 | |
1754 return vaapi_dec_->DecodeSurface(dec_surface); | |
1755 } | |
1756 | |
1757 bool VaapiVideoDecodeAccelerator::VaapiVP9Accelerator::OutputPicture( | |
1758 const scoped_refptr<VP9Picture>& pic) { | |
1759 scoped_refptr<VaapiDecodeSurface> dec_surface = | |
1760 VP9PictureToVaapiDecodeSurface(pic); | |
1761 | |
1762 vaapi_dec_->SurfaceReady(dec_surface); | |
1763 return true; | |
1764 } | |
1765 | |
1766 scoped_refptr<VaapiVideoDecodeAccelerator::VaapiDecodeSurface> | |
1767 VaapiVideoDecodeAccelerator::VaapiVP9Accelerator:: | |
1768 VP9PictureToVaapiDecodeSurface(const scoped_refptr<VP9Picture>& pic) { | |
1769 VaapiVP9Picture* vaapi_pic = pic->AsVaapiVP9Picture(); | |
1770 CHECK(vaapi_pic); | |
1771 return vaapi_pic->dec_surface(); | |
1772 } | |
1773 | |
1774 // static | |
1775 media::VideoDecodeAccelerator::SupportedProfiles | |
1776 VaapiVideoDecodeAccelerator::GetSupportedProfiles() { | |
1777 return VaapiWrapper::GetSupportedDecodeProfiles(); | |
1778 } | |
1779 | |
1780 } // namespace content | |
OLD | NEW |