Line data Source code
1 : /*
2 : * Copyright (c) 2014 The WebRTC project authors. All Rights Reserved.
3 : *
4 : * Use of this source code is governed by a BSD-style license
5 : * that can be found in the LICENSE file in the root of the source
6 : * tree. An additional intellectual property rights grant can be found
7 : * in the file PATENTS. All contributing project authors may
8 : * be found in the AUTHORS file in the root of the source tree.
9 : *
10 : */
11 :
12 : #include "webrtc/modules/video_coding/codecs/vp9/vp9_impl.h"
13 :
14 : #include <stdlib.h>
15 : #include <string.h>
16 : #include <time.h>
17 : #include <vector>
18 :
19 : #include "vpx/vpx_encoder.h"
20 : #include "vpx/vpx_decoder.h"
21 : #include "vpx/vp8cx.h"
22 : #include "vpx/vp8dx.h"
23 :
24 : #include "webrtc/base/checks.h"
25 : #include "webrtc/base/timeutils.h"
26 : #include "webrtc/base/keep_ref_until_done.h"
27 : #include "webrtc/base/logging.h"
28 : #include "webrtc/base/trace_event.h"
29 : #include "webrtc/common_video/include/video_frame_buffer.h"
30 : #include "webrtc/common_video/libyuv/include/webrtc_libyuv.h"
31 : #include "webrtc/modules/include/module_common_types.h"
32 : #include "webrtc/modules/video_coding/codecs/vp9/screenshare_layers.h"
33 :
34 : namespace webrtc {
35 :
36 : // Only positive speeds, range for real-time coding currently is: 5 - 8.
37 : // Lower means slower/better quality, higher means fastest/lower quality.
38 0 : int GetCpuSpeed(int width, int height) {
39 : #if defined(WEBRTC_ARCH_ARM) || defined(WEBRTC_ARCH_ARM64) || defined(ANDROID)
40 : return 8;
41 : #else
42 : // For smaller resolutions, use lower speed setting (get some coding gain at
43 : // the cost of increased encoding complexity).
44 0 : if (width * height <= 352 * 288)
45 0 : return 5;
46 : else
47 0 : return 7;
48 : #endif
49 : }
50 :
51 0 : bool VP9Encoder::IsSupported() {
52 0 : return true;
53 : }
54 :
55 0 : VP9Encoder* VP9Encoder::Create() {
56 0 : return new VP9EncoderImpl();
57 : }
58 :
59 0 : void VP9EncoderImpl::EncoderOutputCodedPacketCallback(vpx_codec_cx_pkt* pkt,
60 : void* user_data) {
61 0 : VP9EncoderImpl* enc = static_cast<VP9EncoderImpl*>(user_data);
62 0 : enc->GetEncodedLayerFrame(pkt);
63 0 : }
64 :
65 0 : VP9EncoderImpl::VP9EncoderImpl()
66 : : encoded_image_(),
67 : encoded_complete_callback_(NULL),
68 : inited_(false),
69 : timestamp_(0),
70 : picture_id_(0),
71 : cpu_speed_(3),
72 : rc_max_intra_target_(0),
73 : encoder_(NULL),
74 : config_(NULL),
75 : raw_(NULL),
76 : input_image_(NULL),
77 : tl0_pic_idx_(0),
78 : frames_since_kf_(0),
79 : num_temporal_layers_(0),
80 : num_spatial_layers_(0),
81 : num_cores_(0),
82 : is_flexible_mode_(false),
83 : frames_encoded_(0),
84 : // Use two spatial when screensharing with flexible mode.
85 0 : spatial_layer_(new ScreenshareLayersVP9(2)) {
86 0 : memset(&codec_, 0, sizeof(codec_));
87 0 : memset(&svc_params_, 0, sizeof(vpx_svc_extra_cfg_t));
88 0 : uint32_t seed = rtc::Time32();
89 0 : srand(seed);
90 0 : }
91 :
92 0 : VP9EncoderImpl::~VP9EncoderImpl() {
93 0 : Release();
94 0 : }
95 :
96 0 : int VP9EncoderImpl::Release() {
97 0 : if (encoded_image_._buffer != NULL) {
98 0 : delete[] encoded_image_._buffer;
99 0 : encoded_image_._buffer = NULL;
100 : }
101 0 : if (encoder_ != NULL) {
102 0 : if (vpx_codec_destroy(encoder_)) {
103 0 : return WEBRTC_VIDEO_CODEC_MEMORY;
104 : }
105 0 : delete encoder_;
106 0 : encoder_ = NULL;
107 : }
108 0 : if (config_ != NULL) {
109 0 : delete config_;
110 0 : config_ = NULL;
111 : }
112 0 : if (raw_ != NULL) {
113 0 : vpx_img_free(raw_);
114 0 : raw_ = NULL;
115 : }
116 0 : inited_ = false;
117 0 : return WEBRTC_VIDEO_CODEC_OK;
118 : }
119 :
120 0 : bool VP9EncoderImpl::ExplicitlyConfiguredSpatialLayers() const {
121 : // We check target_bitrate_bps of the 0th layer to see if the spatial layers
122 : // (i.e. bitrates) were explicitly configured.
123 0 : return num_spatial_layers_ > 1 &&
124 0 : codec_.spatialLayers[0].target_bitrate_bps > 0;
125 : }
126 :
127 0 : bool VP9EncoderImpl::SetSvcRates() {
128 0 : uint8_t i = 0;
129 :
130 0 : if (ExplicitlyConfiguredSpatialLayers()) {
131 0 : if (num_temporal_layers_ > 1) {
132 0 : LOG(LS_ERROR) << "Multiple temporal layers when manually specifying "
133 0 : "spatial layers not implemented yet!";
134 0 : return false;
135 : }
136 0 : int total_bitrate_bps = 0;
137 0 : for (i = 0; i < num_spatial_layers_; ++i)
138 0 : total_bitrate_bps += codec_.spatialLayers[i].target_bitrate_bps;
139 : // If total bitrate differs now from what has been specified at the
140 : // beginning, update the bitrates in the same ratio as before.
141 0 : for (i = 0; i < num_spatial_layers_; ++i) {
142 0 : config_->ss_target_bitrate[i] = config_->layer_target_bitrate[i] =
143 0 : static_cast<int>(static_cast<int64_t>(config_->rc_target_bitrate) *
144 0 : codec_.spatialLayers[i].target_bitrate_bps /
145 : total_bitrate_bps);
146 : }
147 : } else {
148 0 : float rate_ratio[VPX_MAX_LAYERS] = {0};
149 0 : float total = 0;
150 :
151 0 : for (i = 0; i < num_spatial_layers_; ++i) {
152 0 : if (svc_params_.scaling_factor_num[i] <= 0 ||
153 0 : svc_params_.scaling_factor_den[i] <= 0) {
154 0 : LOG(LS_ERROR) << "Scaling factors not specified!";
155 0 : return false;
156 : }
157 0 : rate_ratio[i] =
158 0 : static_cast<float>(svc_params_.scaling_factor_num[i]) /
159 0 : svc_params_.scaling_factor_den[i];
160 0 : total += rate_ratio[i];
161 : }
162 :
163 0 : for (i = 0; i < num_spatial_layers_; ++i) {
164 0 : config_->ss_target_bitrate[i] = static_cast<unsigned int>(
165 0 : config_->rc_target_bitrate * rate_ratio[i] / total);
166 0 : if (num_temporal_layers_ == 1) {
167 0 : config_->layer_target_bitrate[i] = config_->ss_target_bitrate[i];
168 0 : } else if (num_temporal_layers_ == 2) {
169 0 : config_->layer_target_bitrate[i * num_temporal_layers_] =
170 0 : config_->ss_target_bitrate[i] * 2 / 3;
171 0 : config_->layer_target_bitrate[i * num_temporal_layers_ + 1] =
172 0 : config_->ss_target_bitrate[i];
173 0 : } else if (num_temporal_layers_ == 3) {
174 0 : config_->layer_target_bitrate[i * num_temporal_layers_] =
175 0 : config_->ss_target_bitrate[i] / 2;
176 0 : config_->layer_target_bitrate[i * num_temporal_layers_ + 1] =
177 0 : config_->layer_target_bitrate[i * num_temporal_layers_] +
178 0 : (config_->ss_target_bitrate[i] / 4);
179 0 : config_->layer_target_bitrate[i * num_temporal_layers_ + 2] =
180 0 : config_->ss_target_bitrate[i];
181 : } else {
182 0 : LOG(LS_ERROR) << "Unsupported number of temporal layers: "
183 0 : << num_temporal_layers_;
184 0 : return false;
185 : }
186 : }
187 : }
188 :
189 : // For now, temporal layers only supported when having one spatial layer.
190 0 : if (num_spatial_layers_ == 1) {
191 0 : for (i = 0; i < num_temporal_layers_; ++i) {
192 0 : config_->ts_target_bitrate[i] = config_->layer_target_bitrate[i];
193 : }
194 : }
195 :
196 0 : return true;
197 : }
198 :
199 0 : int VP9EncoderImpl::SetRateAllocation(
200 : const BitrateAllocation& bitrate_allocation,
201 : uint32_t frame_rate) {
202 0 : if (!inited_) {
203 0 : return WEBRTC_VIDEO_CODEC_UNINITIALIZED;
204 : }
205 0 : if (encoder_->err) {
206 0 : return WEBRTC_VIDEO_CODEC_ERROR;
207 : }
208 0 : if (frame_rate < 1) {
209 0 : return WEBRTC_VIDEO_CODEC_ERR_PARAMETER;
210 : }
211 : // Update bit rate
212 0 : if (codec_.maxBitrate > 0 &&
213 0 : bitrate_allocation.get_sum_kbps() > codec_.maxBitrate) {
214 0 : return WEBRTC_VIDEO_CODEC_ERR_PARAMETER;
215 : }
216 :
217 : // TODO(sprang): Actually use BitrateAllocation layer info.
218 0 : config_->rc_target_bitrate = bitrate_allocation.get_sum_kbps();
219 0 : codec_.maxFramerate = frame_rate;
220 0 : spatial_layer_->ConfigureBitrate(bitrate_allocation.get_sum_kbps(), 0);
221 :
222 0 : if (!SetSvcRates()) {
223 0 : return WEBRTC_VIDEO_CODEC_ERR_PARAMETER;
224 : }
225 :
226 : // Update encoder context
227 0 : if (vpx_codec_enc_config_set(encoder_, config_)) {
228 0 : return WEBRTC_VIDEO_CODEC_ERROR;
229 : }
230 0 : return WEBRTC_VIDEO_CODEC_OK;
231 : }
232 :
233 0 : int VP9EncoderImpl::InitEncode(const VideoCodec* inst,
234 : int number_of_cores,
235 : size_t /*max_payload_size*/) {
236 0 : if (inst == NULL) {
237 0 : return WEBRTC_VIDEO_CODEC_ERR_PARAMETER;
238 : }
239 0 : if (inst->maxFramerate < 1) {
240 0 : return WEBRTC_VIDEO_CODEC_ERR_PARAMETER;
241 : }
242 : // Allow zero to represent an unspecified maxBitRate
243 0 : if (inst->maxBitrate > 0 && inst->startBitrate > inst->maxBitrate) {
244 0 : return WEBRTC_VIDEO_CODEC_ERR_PARAMETER;
245 : }
246 0 : if (inst->width < 1 || inst->height < 1) {
247 0 : return WEBRTC_VIDEO_CODEC_ERR_PARAMETER;
248 : }
249 0 : if (number_of_cores < 1 || number_of_cores > UINT8_MAX) {
250 0 : return WEBRTC_VIDEO_CODEC_ERR_PARAMETER;
251 : }
252 0 : if (inst->VP9().numberOfTemporalLayers > 3) {
253 0 : return WEBRTC_VIDEO_CODEC_ERR_PARAMETER;
254 : }
255 : // libvpx currently supports only one or two spatial layers.
256 0 : if (inst->VP9().numberOfSpatialLayers > 2) {
257 0 : return WEBRTC_VIDEO_CODEC_ERR_PARAMETER;
258 : }
259 :
260 0 : int ret_val = Release();
261 0 : if (ret_val < 0) {
262 0 : return ret_val;
263 : }
264 0 : if (encoder_ == NULL) {
265 0 : encoder_ = new vpx_codec_ctx_t;
266 : // Only randomize pid/tl0 the first time the encoder is initialized
267 : // in order to not make random jumps mid-stream.
268 0 : picture_id_ = static_cast<uint16_t>(rand()) & 0x7FFF; // NOLINT
269 0 : tl0_pic_idx_ = static_cast<uint8_t>(rand()); // NOLINT
270 : }
271 0 : if (config_ == NULL) {
272 0 : config_ = new vpx_codec_enc_cfg_t;
273 : }
274 0 : timestamp_ = 0;
275 0 : if (&codec_ != inst) {
276 0 : codec_ = *inst;
277 : }
278 :
279 0 : num_cores_ = number_of_cores;
280 0 : num_spatial_layers_ = inst->VP9().numberOfSpatialLayers;
281 0 : num_temporal_layers_ = inst->VP9().numberOfTemporalLayers;
282 0 : if (num_temporal_layers_ == 0)
283 0 : num_temporal_layers_ = 1;
284 :
285 : // Allocate memory for encoded image
286 0 : if (encoded_image_._buffer != NULL) {
287 0 : delete[] encoded_image_._buffer;
288 : }
289 0 : encoded_image_._size = CalcBufferSize(kI420, codec_.width, codec_.height);
290 0 : encoded_image_._buffer = new uint8_t[encoded_image_._size];
291 0 : encoded_image_._completeFrame = true;
292 : // Creating a wrapper to the image - setting image data to NULL. Actual
293 : // pointer will be set in encode. Setting align to 1, as it is meaningless
294 : // (actual memory is not allocated).
295 0 : raw_ = vpx_img_wrap(NULL, VPX_IMG_FMT_I420, codec_.width, codec_.height, 1,
296 : NULL);
297 : // Populate encoder configuration with default values.
298 0 : if (vpx_codec_enc_config_default(vpx_codec_vp9_cx(), config_, 0)) {
299 0 : return WEBRTC_VIDEO_CODEC_ERROR;
300 : }
301 0 : config_->g_w = codec_.width;
302 0 : config_->g_h = codec_.height;
303 0 : config_->rc_target_bitrate = inst->startBitrate; // in kbit/s
304 0 : config_->g_error_resilient = 1;
305 : // Setting the time base of the codec.
306 0 : config_->g_timebase.num = 1;
307 0 : config_->g_timebase.den = 90000;
308 0 : config_->g_lag_in_frames = 0; // 0- no frame lagging
309 0 : config_->g_threads = 1;
310 : // Rate control settings.
311 0 : config_->rc_dropframe_thresh = inst->VP9().frameDroppingOn ? 30 : 0;
312 0 : config_->rc_end_usage = VPX_CBR;
313 0 : config_->g_pass = VPX_RC_ONE_PASS;
314 0 : config_->rc_min_quantizer = 2;
315 0 : config_->rc_max_quantizer = 52;
316 0 : config_->rc_undershoot_pct = 50;
317 0 : config_->rc_overshoot_pct = 50;
318 0 : config_->rc_buf_initial_sz = 500;
319 0 : config_->rc_buf_optimal_sz = 600;
320 0 : config_->rc_buf_sz = 1000;
321 : // Set the maximum target size of any key-frame.
322 0 : rc_max_intra_target_ = MaxIntraTarget(config_->rc_buf_optimal_sz);
323 0 : if (inst->VP9().keyFrameInterval > 0) {
324 0 : config_->kf_mode = VPX_KF_AUTO;
325 0 : config_->kf_max_dist = inst->VP9().keyFrameInterval;
326 : // Needs to be set (in svc mode) to get correct periodic key frame interval
327 : // (will have no effect in non-svc).
328 0 : config_->kf_min_dist = config_->kf_max_dist;
329 : } else {
330 0 : config_->kf_mode = VPX_KF_DISABLED;
331 : }
332 0 : config_->rc_resize_allowed = inst->VP9().automaticResizeOn ? 1 : 0;
333 : // Determine number of threads based on the image size and #cores.
334 0 : config_->g_threads =
335 0 : NumberOfThreads(config_->g_w, config_->g_h, number_of_cores);
336 :
337 0 : cpu_speed_ = GetCpuSpeed(config_->g_w, config_->g_h);
338 :
339 : // TODO(asapersson): Check configuration of temporal switch up and increase
340 : // pattern length.
341 0 : is_flexible_mode_ = inst->VP9().flexibleMode;
342 0 : if (is_flexible_mode_) {
343 0 : config_->temporal_layering_mode = VP9E_TEMPORAL_LAYERING_MODE_BYPASS;
344 0 : config_->ts_number_layers = num_temporal_layers_;
345 0 : if (codec_.mode == kScreensharing)
346 0 : spatial_layer_->ConfigureBitrate(inst->startBitrate, 0);
347 0 : } else if (num_temporal_layers_ == 1) {
348 0 : gof_.SetGofInfoVP9(kTemporalStructureMode1);
349 0 : config_->temporal_layering_mode = VP9E_TEMPORAL_LAYERING_MODE_NOLAYERING;
350 0 : config_->ts_number_layers = 1;
351 0 : config_->ts_rate_decimator[0] = 1;
352 0 : config_->ts_periodicity = 1;
353 0 : config_->ts_layer_id[0] = 0;
354 0 : } else if (num_temporal_layers_ == 2) {
355 0 : gof_.SetGofInfoVP9(kTemporalStructureMode2);
356 0 : config_->temporal_layering_mode = VP9E_TEMPORAL_LAYERING_MODE_0101;
357 0 : config_->ts_number_layers = 2;
358 0 : config_->ts_rate_decimator[0] = 2;
359 0 : config_->ts_rate_decimator[1] = 1;
360 0 : config_->ts_periodicity = 2;
361 0 : config_->ts_layer_id[0] = 0;
362 0 : config_->ts_layer_id[1] = 1;
363 0 : } else if (num_temporal_layers_ == 3) {
364 0 : gof_.SetGofInfoVP9(kTemporalStructureMode3);
365 0 : config_->temporal_layering_mode = VP9E_TEMPORAL_LAYERING_MODE_0212;
366 0 : config_->ts_number_layers = 3;
367 0 : config_->ts_rate_decimator[0] = 4;
368 0 : config_->ts_rate_decimator[1] = 2;
369 0 : config_->ts_rate_decimator[2] = 1;
370 0 : config_->ts_periodicity = 4;
371 0 : config_->ts_layer_id[0] = 0;
372 0 : config_->ts_layer_id[1] = 2;
373 0 : config_->ts_layer_id[2] = 1;
374 0 : config_->ts_layer_id[3] = 2;
375 : } else {
376 0 : return WEBRTC_VIDEO_CODEC_ERR_PARAMETER;
377 : }
378 :
379 0 : return InitAndSetControlSettings(inst);
380 : }
381 :
382 0 : int VP9EncoderImpl::NumberOfThreads(int width,
383 : int height,
384 : int number_of_cores) {
385 : // Keep the number of encoder threads equal to the possible number of column
386 : // tiles, which is (1, 2, 4, 8). See comments below for VP9E_SET_TILE_COLUMNS.
387 0 : if (width * height >= 1280 * 720 && number_of_cores > 4) {
388 0 : return 4;
389 0 : } else if (width * height >= 640 * 480 && number_of_cores > 2) {
390 0 : return 2;
391 : } else {
392 : // 1 thread less than VGA.
393 0 : return 1;
394 : }
395 : }
396 :
397 0 : int VP9EncoderImpl::InitAndSetControlSettings(const VideoCodec* inst) {
398 : // Set QP-min/max per spatial and temporal layer.
399 0 : int tot_num_layers = num_spatial_layers_ * num_temporal_layers_;
400 0 : for (int i = 0; i < tot_num_layers; ++i) {
401 0 : svc_params_.max_quantizers[i] = config_->rc_max_quantizer;
402 0 : svc_params_.min_quantizers[i] = config_->rc_min_quantizer;
403 : }
404 0 : config_->ss_number_layers = num_spatial_layers_;
405 0 : if (ExplicitlyConfiguredSpatialLayers()) {
406 0 : for (int i = 0; i < num_spatial_layers_; ++i) {
407 0 : const auto& layer = codec_.spatialLayers[i];
408 0 : svc_params_.scaling_factor_num[i] = layer.scaling_factor_num;
409 0 : svc_params_.scaling_factor_den[i] = layer.scaling_factor_den;
410 : }
411 : } else {
412 0 : int scaling_factor_num = 256;
413 0 : for (int i = num_spatial_layers_ - 1; i >= 0; --i) {
414 : // 1:2 scaling in each dimension.
415 0 : svc_params_.scaling_factor_num[i] = scaling_factor_num;
416 0 : svc_params_.scaling_factor_den[i] = 256;
417 0 : if (codec_.mode != kScreensharing)
418 0 : scaling_factor_num /= 2;
419 : }
420 : }
421 :
422 0 : if (!SetSvcRates()) {
423 0 : return WEBRTC_VIDEO_CODEC_ERR_PARAMETER;
424 : }
425 :
426 0 : if (vpx_codec_enc_init(encoder_, vpx_codec_vp9_cx(), config_, 0)) {
427 0 : return WEBRTC_VIDEO_CODEC_UNINITIALIZED;
428 : }
429 0 : vpx_codec_control(encoder_, VP8E_SET_CPUUSED, cpu_speed_);
430 0 : vpx_codec_control(encoder_, VP8E_SET_MAX_INTRA_BITRATE_PCT,
431 0 : rc_max_intra_target_);
432 0 : vpx_codec_control(encoder_, VP9E_SET_AQ_MODE,
433 0 : inst->VP9().adaptiveQpMode ? 3 : 0);
434 :
435 0 : vpx_codec_control(
436 : encoder_, VP9E_SET_SVC,
437 0 : (num_temporal_layers_ > 1 || num_spatial_layers_ > 1) ? 1 : 0);
438 0 : if (num_temporal_layers_ > 1 || num_spatial_layers_ > 1) {
439 0 : vpx_codec_control(encoder_, VP9E_SET_SVC_PARAMETERS,
440 0 : &svc_params_);
441 : }
442 : // Register callback for getting each spatial layer.
443 : vpx_codec_priv_output_cx_pkt_cb_pair_t cbp = {
444 : VP9EncoderImpl::EncoderOutputCodedPacketCallback,
445 0 : reinterpret_cast<void*>(this)};
446 0 : vpx_codec_control(encoder_, VP9E_REGISTER_CX_CALLBACK,
447 0 : reinterpret_cast<void*>(&cbp));
448 :
449 : // Control function to set the number of column tiles in encoding a frame, in
450 : // log2 unit: e.g., 0 = 1 tile column, 1 = 2 tile columns, 2 = 4 tile columns.
451 : // The number tile columns will be capped by the encoder based on image size
452 : // (minimum width of tile column is 256 pixels, maximum is 4096).
453 0 : vpx_codec_control(encoder_, VP9E_SET_TILE_COLUMNS, (config_->g_threads >> 1));
454 : #if !defined(WEBRTC_ARCH_ARM) && !defined(WEBRTC_ARCH_ARM64) && \
455 : !defined(ANDROID)
456 : // Note denoiser is still off by default until further testing/optimization,
457 : // i.e., VP9().denoisingOn == 0.
458 0 : vpx_codec_control(encoder_, VP9E_SET_NOISE_SENSITIVITY,
459 0 : inst->VP9().denoisingOn ? 1 : 0);
460 : #endif
461 0 : if (codec_.mode == kScreensharing) {
462 : // Adjust internal parameters to screen content.
463 0 : vpx_codec_control(encoder_, VP9E_SET_TUNE_CONTENT, 1);
464 : }
465 : // Enable encoder skip of static/low content blocks.
466 0 : vpx_codec_control(encoder_, VP8E_SET_STATIC_THRESHOLD, 1);
467 0 : inited_ = true;
468 0 : return WEBRTC_VIDEO_CODEC_OK;
469 : }
470 :
471 0 : uint32_t VP9EncoderImpl::MaxIntraTarget(uint32_t optimal_buffer_size) {
472 : // Set max to the optimal buffer level (normalized by target BR),
473 : // and scaled by a scale_par.
474 : // Max target size = scale_par * optimal_buffer_size * targetBR[Kbps].
475 : // This value is presented in percentage of perFrameBw:
476 : // perFrameBw = targetBR[Kbps] * 1000 / framerate.
477 : // The target in % is as follows:
478 0 : float scale_par = 0.5;
479 : uint32_t target_pct =
480 0 : optimal_buffer_size * scale_par * codec_.maxFramerate / 10;
481 : // Don't go below 3 times the per frame bandwidth.
482 0 : const uint32_t min_intra_size = 300;
483 0 : return (target_pct < min_intra_size) ? min_intra_size : target_pct;
484 : }
485 :
486 0 : int VP9EncoderImpl::Encode(const VideoFrame& input_image,
487 : const CodecSpecificInfo* codec_specific_info,
488 : const std::vector<FrameType>* frame_types) {
489 0 : if (!inited_) {
490 0 : return WEBRTC_VIDEO_CODEC_UNINITIALIZED;
491 : }
492 0 : if (encoded_complete_callback_ == NULL) {
493 0 : return WEBRTC_VIDEO_CODEC_UNINITIALIZED;
494 : }
495 0 : FrameType frame_type = kVideoFrameDelta;
496 : // We only support one stream at the moment.
497 0 : if (frame_types && frame_types->size() > 0) {
498 0 : frame_type = (*frame_types)[0];
499 : }
500 0 : if (input_image.width() != codec_.width ||
501 0 : input_image.height() != codec_.height) {
502 0 : int ret = UpdateCodecFrameSize(input_image);
503 0 : if (ret < 0) {
504 0 : return ret;
505 : }
506 : }
507 0 : RTC_DCHECK_EQ(input_image.width(), raw_->d_w);
508 0 : RTC_DCHECK_EQ(input_image.height(), raw_->d_h);
509 :
510 : // Set input image for use in the callback.
511 : // This was necessary since you need some information from input_image.
512 : // You can save only the necessary information (such as timestamp) instead of
513 : // doing this.
514 0 : input_image_ = &input_image;
515 :
516 : // Image in vpx_image_t format.
517 : // Input image is const. VPX's raw image is not defined as const.
518 0 : raw_->planes[VPX_PLANE_Y] =
519 0 : const_cast<uint8_t*>(input_image.video_frame_buffer()->DataY());
520 0 : raw_->planes[VPX_PLANE_U] =
521 0 : const_cast<uint8_t*>(input_image.video_frame_buffer()->DataU());
522 0 : raw_->planes[VPX_PLANE_V] =
523 0 : const_cast<uint8_t*>(input_image.video_frame_buffer()->DataV());
524 0 : raw_->stride[VPX_PLANE_Y] = input_image.video_frame_buffer()->StrideY();
525 0 : raw_->stride[VPX_PLANE_U] = input_image.video_frame_buffer()->StrideU();
526 0 : raw_->stride[VPX_PLANE_V] = input_image.video_frame_buffer()->StrideV();
527 :
528 0 : vpx_enc_frame_flags_t flags = 0;
529 0 : bool send_keyframe = (frame_type == kVideoFrameKey);
530 0 : if (send_keyframe) {
531 : // Key frame request from caller.
532 0 : flags = VPX_EFLAG_FORCE_KF;
533 : }
534 :
535 0 : if (is_flexible_mode_) {
536 0 : SuperFrameRefSettings settings;
537 :
538 : // These structs are copied when calling vpx_codec_control,
539 : // therefore it is ok for them to go out of scope.
540 : vpx_svc_ref_frame_config enc_layer_conf;
541 : vpx_svc_layer_id layer_id;
542 :
543 0 : if (codec_.mode == kRealtimeVideo) {
544 : // Real time video not yet implemented in flexible mode.
545 0 : RTC_NOTREACHED();
546 : } else {
547 : settings = spatial_layer_->GetSuperFrameSettings(input_image.timestamp(),
548 0 : send_keyframe);
549 : }
550 0 : enc_layer_conf = GenerateRefsAndFlags(settings);
551 0 : layer_id.temporal_layer_id = 0;
552 0 : layer_id.spatial_layer_id = settings.start_layer;
553 0 : vpx_codec_control(encoder_, VP9E_SET_SVC_LAYER_ID, &layer_id);
554 0 : vpx_codec_control(encoder_, VP9E_SET_SVC_REF_FRAME_CONFIG, &enc_layer_conf);
555 : }
556 :
557 0 : assert(codec_.maxFramerate > 0);
558 0 : uint32_t duration = 90000 / codec_.maxFramerate;
559 0 : if (vpx_codec_encode(encoder_, raw_, timestamp_, duration, flags,
560 : VPX_DL_REALTIME)) {
561 0 : return WEBRTC_VIDEO_CODEC_ERROR;
562 : }
563 0 : timestamp_ += duration;
564 :
565 0 : return WEBRTC_VIDEO_CODEC_OK;
566 : }
567 :
568 0 : int VP9EncoderImpl::UpdateCodecFrameSize(
569 : const VideoFrame& input_image) {
570 0 : fprintf(stderr, "Reconfiging VP( from %dx%d to %dx%d\n",
571 0 : codec_.width, codec_.height, input_image.width(), input_image.height());
572 : // Preserve latest bitrate/framerate setting
573 0 : uint32_t old_bitrate_kbit = config_->rc_target_bitrate;
574 0 : uint32_t old_framerate = codec_.maxFramerate;
575 :
576 0 : codec_.width = input_image.width();
577 0 : codec_.height = input_image.height();
578 :
579 0 : vpx_img_free(raw_);
580 0 : raw_ = vpx_img_wrap(NULL, VPX_IMG_FMT_I420, codec_.width, codec_.height,
581 : 1, NULL);
582 : // Update encoder context for new frame size.
583 0 : config_->g_w = codec_.width;
584 0 : config_->g_h = codec_.height;
585 :
586 : // Determine number of threads based on the image size and #cores.
587 0 : config_->g_threads = NumberOfThreads(codec_.width, codec_.height,
588 0 : num_cores_);
589 : // Update the cpu_speed setting for resolution change.
590 0 : cpu_speed_ = GetCpuSpeed(codec_.width, codec_.height);
591 :
592 : // NOTE: We would like to do this the same way vp8 does it
593 : // (with vpx_codec_enc_config_set()), but that causes asserts
594 : // in AQ 3 (cyclic); and in AQ 0 it works, but on a resize to smaller
595 : // than 1/2 x 1/2 original it asserts in convolve(). Given these
596 : // bugs in trying to do it the "right" way, we basically re-do
597 : // the initialization.
598 0 : vpx_codec_destroy(encoder_); // clean up old state
599 0 : int result = InitAndSetControlSettings(&codec_);
600 0 : if (result == WEBRTC_VIDEO_CODEC_OK) {
601 0 : return SetRates(old_bitrate_kbit, old_framerate);
602 : }
603 0 : return result;
604 : }
605 :
606 0 : void VP9EncoderImpl::PopulateCodecSpecific(CodecSpecificInfo* codec_specific,
607 : const vpx_codec_cx_pkt& pkt,
608 : uint32_t timestamp) {
609 0 : assert(codec_specific != NULL);
610 0 : codec_specific->codecType = kVideoCodecVP9;
611 0 : codec_specific->codec_name = ImplementationName();
612 0 : CodecSpecificInfoVP9* vp9_info = &(codec_specific->codecSpecific.VP9);
613 : // TODO(asapersson): Set correct value.
614 0 : vp9_info->inter_pic_predicted =
615 0 : (pkt.data.frame.flags & VPX_FRAME_IS_KEY) ? false : true;
616 0 : vp9_info->flexible_mode = codec_.VP9()->flexibleMode;
617 0 : vp9_info->ss_data_available =
618 0 : ((pkt.data.frame.flags & VPX_FRAME_IS_KEY) && !codec_.VP9()->flexibleMode)
619 0 : ? true
620 : : false;
621 :
622 0 : vpx_svc_layer_id_t layer_id = {0};
623 0 : vpx_codec_control(encoder_, VP9E_GET_SVC_LAYER_ID, &layer_id);
624 :
625 0 : assert(num_temporal_layers_ > 0);
626 0 : assert(num_spatial_layers_ > 0);
627 0 : if (num_temporal_layers_ == 1) {
628 0 : assert(layer_id.temporal_layer_id == 0);
629 0 : vp9_info->temporal_idx = kNoTemporalIdx;
630 : } else {
631 0 : vp9_info->temporal_idx = layer_id.temporal_layer_id;
632 : }
633 0 : if (num_spatial_layers_ == 1) {
634 0 : assert(layer_id.spatial_layer_id == 0);
635 0 : vp9_info->spatial_idx = kNoSpatialIdx;
636 : } else {
637 0 : vp9_info->spatial_idx = layer_id.spatial_layer_id;
638 : }
639 0 : if (layer_id.spatial_layer_id != 0) {
640 0 : vp9_info->ss_data_available = false;
641 : }
642 :
643 : // TODO(asapersson): this info has to be obtained from the encoder.
644 0 : vp9_info->temporal_up_switch = false;
645 :
646 0 : bool is_first_frame = false;
647 0 : if (is_flexible_mode_) {
648 0 : is_first_frame =
649 0 : layer_id.spatial_layer_id == spatial_layer_->GetStartLayer();
650 : } else {
651 0 : is_first_frame = layer_id.spatial_layer_id == 0;
652 : }
653 :
654 0 : if (is_first_frame) {
655 0 : picture_id_ = (picture_id_ + 1) & 0x7FFF;
656 : // TODO(asapersson): this info has to be obtained from the encoder.
657 0 : vp9_info->inter_layer_predicted = false;
658 0 : ++frames_since_kf_;
659 : } else {
660 : // TODO(asapersson): this info has to be obtained from the encoder.
661 0 : vp9_info->inter_layer_predicted = true;
662 : }
663 :
664 0 : if (pkt.data.frame.flags & VPX_FRAME_IS_KEY) {
665 0 : frames_since_kf_ = 0;
666 : }
667 :
668 0 : vp9_info->picture_id = picture_id_;
669 :
670 0 : if (!vp9_info->flexible_mode) {
671 0 : if (layer_id.temporal_layer_id == 0 && layer_id.spatial_layer_id == 0) {
672 0 : tl0_pic_idx_++;
673 : }
674 0 : vp9_info->tl0_pic_idx = tl0_pic_idx_;
675 : }
676 :
677 : // Always populate this, so that the packetizer can properly set the marker
678 : // bit.
679 0 : vp9_info->num_spatial_layers = num_spatial_layers_;
680 :
681 0 : vp9_info->num_ref_pics = 0;
682 0 : if (vp9_info->flexible_mode) {
683 0 : vp9_info->gof_idx = kNoGofIdx;
684 0 : vp9_info->num_ref_pics = num_ref_pics_[layer_id.spatial_layer_id];
685 0 : for (int i = 0; i < num_ref_pics_[layer_id.spatial_layer_id]; ++i) {
686 0 : vp9_info->p_diff[i] = p_diff_[layer_id.spatial_layer_id][i];
687 : }
688 : } else {
689 0 : vp9_info->gof_idx =
690 0 : static_cast<uint8_t>(frames_since_kf_ % gof_.num_frames_in_gof);
691 0 : vp9_info->temporal_up_switch = gof_.temporal_up_switch[vp9_info->gof_idx];
692 : }
693 :
694 0 : if (vp9_info->ss_data_available) {
695 0 : vp9_info->spatial_layer_resolution_present = true;
696 0 : for (size_t i = 0; i < vp9_info->num_spatial_layers; ++i) {
697 0 : vp9_info->width[i] = codec_.width *
698 0 : svc_params_.scaling_factor_num[i] /
699 0 : svc_params_.scaling_factor_den[i];
700 0 : vp9_info->height[i] = codec_.height *
701 0 : svc_params_.scaling_factor_num[i] /
702 0 : svc_params_.scaling_factor_den[i];
703 : }
704 0 : if (!vp9_info->flexible_mode) {
705 0 : vp9_info->gof.CopyGofInfoVP9(gof_);
706 : }
707 : }
708 0 : }
709 :
710 0 : int VP9EncoderImpl::GetEncodedLayerFrame(const vpx_codec_cx_pkt* pkt) {
711 0 : RTC_DCHECK_EQ(pkt->kind, VPX_CODEC_CX_FRAME_PKT);
712 :
713 0 : if (pkt->data.frame.sz > encoded_image_._size) {
714 0 : delete[] encoded_image_._buffer;
715 0 : encoded_image_._size = pkt->data.frame.sz;
716 0 : encoded_image_._buffer = new uint8_t[encoded_image_._size];
717 : }
718 0 : memcpy(encoded_image_._buffer, pkt->data.frame.buf, pkt->data.frame.sz);
719 0 : encoded_image_._length = pkt->data.frame.sz;
720 :
721 : // No data partitioning in VP9, so 1 partition only.
722 0 : int part_idx = 0;
723 0 : RTPFragmentationHeader frag_info;
724 0 : frag_info.VerifyAndAllocateFragmentationHeader(1);
725 0 : frag_info.fragmentationOffset[part_idx] = 0;
726 0 : frag_info.fragmentationLength[part_idx] = pkt->data.frame.sz;
727 0 : frag_info.fragmentationPlType[part_idx] = 0;
728 0 : frag_info.fragmentationTimeDiff[part_idx] = 0;
729 :
730 0 : vpx_svc_layer_id_t layer_id = {0};
731 0 : vpx_codec_control(encoder_, VP9E_GET_SVC_LAYER_ID, &layer_id);
732 0 : if (is_flexible_mode_ && codec_.mode == kScreensharing)
733 0 : spatial_layer_->LayerFrameEncoded(
734 0 : static_cast<unsigned int>(encoded_image_._length),
735 0 : layer_id.spatial_layer_id);
736 :
737 : // End of frame.
738 : // Check if encoded frame is a key frame.
739 0 : encoded_image_._frameType = kVideoFrameDelta;
740 0 : if (pkt->data.frame.flags & VPX_FRAME_IS_KEY) {
741 0 : encoded_image_._frameType = kVideoFrameKey;
742 : }
743 0 : RTC_DCHECK_LE(encoded_image_._length, encoded_image_._size);
744 :
745 0 : CodecSpecificInfo codec_specific;
746 0 : PopulateCodecSpecific(&codec_specific, *pkt, input_image_->timestamp());
747 :
748 0 : if (encoded_image_._length > 0) {
749 0 : TRACE_COUNTER1("webrtc", "EncodedFrameSize", encoded_image_._length);
750 0 : encoded_image_._timeStamp = input_image_->timestamp();
751 0 : encoded_image_.capture_time_ms_ = input_image_->render_time_ms();
752 0 : encoded_image_.rotation_ = input_image_->rotation();
753 0 : encoded_image_._encodedHeight = raw_->d_h;
754 0 : encoded_image_._encodedWidth = raw_->d_w;
755 0 : int qp = -1;
756 0 : vpx_codec_control(encoder_, VP8E_GET_LAST_QUANTIZER, &qp);
757 0 : encoded_image_.qp_ = qp;
758 0 : encoded_complete_callback_->OnEncodedImage(encoded_image_, &codec_specific,
759 0 : &frag_info);
760 : }
761 0 : return WEBRTC_VIDEO_CODEC_OK;
762 : }
763 :
764 0 : vpx_svc_ref_frame_config VP9EncoderImpl::GenerateRefsAndFlags(
765 : const SuperFrameRefSettings& settings) {
766 : static const vpx_enc_frame_flags_t kAllFlags =
767 : VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_LAST |
768 : VP8_EFLAG_NO_UPD_LAST | VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_GF;
769 0 : vpx_svc_ref_frame_config sf_conf = {};
770 0 : if (settings.is_keyframe) {
771 : // Used later on to make sure we don't make any invalid references.
772 0 : memset(buffer_updated_at_frame_, -1, sizeof(buffer_updated_at_frame_));
773 0 : for (int layer = settings.start_layer; layer <= settings.stop_layer;
774 : ++layer) {
775 0 : num_ref_pics_[layer] = 0;
776 0 : buffer_updated_at_frame_[settings.layer[layer].upd_buf] = frames_encoded_;
777 : // When encoding a keyframe only the alt_fb_idx is used
778 : // to specify which layer ends up in which buffer.
779 0 : sf_conf.alt_fb_idx[layer] = settings.layer[layer].upd_buf;
780 : }
781 : } else {
782 0 : for (int layer_idx = settings.start_layer; layer_idx <= settings.stop_layer;
783 : ++layer_idx) {
784 0 : vpx_enc_frame_flags_t layer_flags = kAllFlags;
785 0 : num_ref_pics_[layer_idx] = 0;
786 0 : int8_t refs[3] = {settings.layer[layer_idx].ref_buf1,
787 0 : settings.layer[layer_idx].ref_buf2,
788 0 : settings.layer[layer_idx].ref_buf3};
789 :
790 0 : for (unsigned int ref_idx = 0; ref_idx < kMaxVp9RefPics; ++ref_idx) {
791 0 : if (refs[ref_idx] == -1)
792 0 : continue;
793 :
794 0 : RTC_DCHECK_GE(refs[ref_idx], 0);
795 0 : RTC_DCHECK_LE(refs[ref_idx], 7);
796 : // Easier to remove flags from all flags rather than having to
797 : // build the flags from 0.
798 0 : switch (num_ref_pics_[layer_idx]) {
799 : case 0: {
800 0 : sf_conf.lst_fb_idx[layer_idx] = refs[ref_idx];
801 0 : layer_flags &= ~VP8_EFLAG_NO_REF_LAST;
802 0 : break;
803 : }
804 : case 1: {
805 0 : sf_conf.gld_fb_idx[layer_idx] = refs[ref_idx];
806 0 : layer_flags &= ~VP8_EFLAG_NO_REF_GF;
807 0 : break;
808 : }
809 : case 2: {
810 0 : sf_conf.alt_fb_idx[layer_idx] = refs[ref_idx];
811 0 : layer_flags &= ~VP8_EFLAG_NO_REF_ARF;
812 0 : break;
813 : }
814 : }
815 : // Make sure we don't reference a buffer that hasn't been
816 : // used at all or hasn't been used since a keyframe.
817 0 : RTC_DCHECK_NE(buffer_updated_at_frame_[refs[ref_idx]], -1);
818 :
819 0 : p_diff_[layer_idx][num_ref_pics_[layer_idx]] =
820 0 : frames_encoded_ - buffer_updated_at_frame_[refs[ref_idx]];
821 0 : num_ref_pics_[layer_idx]++;
822 : }
823 :
824 0 : bool upd_buf_same_as_a_ref = false;
825 0 : if (settings.layer[layer_idx].upd_buf != -1) {
826 0 : for (unsigned int ref_idx = 0; ref_idx < kMaxVp9RefPics; ++ref_idx) {
827 0 : if (settings.layer[layer_idx].upd_buf == refs[ref_idx]) {
828 0 : switch (ref_idx) {
829 : case 0: {
830 0 : layer_flags &= ~VP8_EFLAG_NO_UPD_LAST;
831 0 : break;
832 : }
833 : case 1: {
834 0 : layer_flags &= ~VP8_EFLAG_NO_UPD_GF;
835 0 : break;
836 : }
837 : case 2: {
838 0 : layer_flags &= ~VP8_EFLAG_NO_UPD_ARF;
839 0 : break;
840 : }
841 : }
842 0 : upd_buf_same_as_a_ref = true;
843 0 : break;
844 : }
845 : }
846 0 : if (!upd_buf_same_as_a_ref) {
847 : // If we have three references and a buffer is specified to be
848 : // updated, then that buffer must be the same as one of the
849 : // three references.
850 0 : RTC_CHECK_LT(num_ref_pics_[layer_idx], kMaxVp9RefPics);
851 :
852 0 : sf_conf.alt_fb_idx[layer_idx] = settings.layer[layer_idx].upd_buf;
853 0 : layer_flags ^= VP8_EFLAG_NO_UPD_ARF;
854 : }
855 :
856 0 : int updated_buffer = settings.layer[layer_idx].upd_buf;
857 0 : buffer_updated_at_frame_[updated_buffer] = frames_encoded_;
858 0 : sf_conf.frame_flags[layer_idx] = layer_flags;
859 : }
860 : }
861 : }
862 0 : ++frames_encoded_;
863 0 : return sf_conf;
864 : }
865 :
866 0 : int VP9EncoderImpl::SetChannelParameters(uint32_t packet_loss, int64_t rtt) {
867 0 : return WEBRTC_VIDEO_CODEC_OK;
868 : }
869 :
870 0 : int VP9EncoderImpl::RegisterEncodeCompleteCallback(
871 : EncodedImageCallback* callback) {
872 0 : encoded_complete_callback_ = callback;
873 0 : return WEBRTC_VIDEO_CODEC_OK;
874 : }
875 :
876 0 : const char* VP9EncoderImpl::ImplementationName() const {
877 0 : return "libvpx";
878 : }
879 :
880 0 : bool VP9Decoder::IsSupported() {
881 0 : return true;
882 : }
883 :
884 0 : VP9Decoder* VP9Decoder::Create() {
885 0 : return new VP9DecoderImpl();
886 : }
887 :
888 0 : VP9DecoderImpl::VP9DecoderImpl()
889 : : decode_complete_callback_(NULL),
890 : inited_(false),
891 : decoder_(NULL),
892 0 : key_frame_required_(true) {
893 0 : memset(&codec_, 0, sizeof(codec_));
894 0 : }
895 :
896 0 : VP9DecoderImpl::~VP9DecoderImpl() {
897 0 : inited_ = true; // in order to do the actual release
898 0 : Release();
899 0 : int num_buffers_in_use = frame_buffer_pool_.GetNumBuffersInUse();
900 0 : if (num_buffers_in_use > 0) {
901 : // The frame buffers are reference counted and frames are exposed after
902 : // decoding. There may be valid usage cases where previous frames are still
903 : // referenced after ~VP9DecoderImpl that is not a leak.
904 0 : LOG(LS_INFO) << num_buffers_in_use << " Vp9FrameBuffers are still "
905 0 : << "referenced during ~VP9DecoderImpl.";
906 : }
907 0 : }
908 :
909 0 : int VP9DecoderImpl::InitDecode(const VideoCodec* inst, int number_of_cores) {
910 0 : if (inst == NULL) {
911 0 : return WEBRTC_VIDEO_CODEC_ERR_PARAMETER;
912 : }
913 0 : int ret_val = Release();
914 0 : if (ret_val < 0) {
915 0 : return ret_val;
916 : }
917 0 : if (decoder_ == NULL) {
918 0 : decoder_ = new vpx_codec_ctx_t;
919 : }
920 : vpx_codec_dec_cfg_t cfg;
921 : // Setting number of threads to a constant value (1)
922 0 : cfg.threads = 1;
923 0 : cfg.h = cfg.w = 0; // set after decode
924 0 : vpx_codec_flags_t flags = 0;
925 0 : if (vpx_codec_dec_init(decoder_, vpx_codec_vp9_dx(), &cfg, flags)) {
926 0 : return WEBRTC_VIDEO_CODEC_MEMORY;
927 : }
928 0 : if (&codec_ != inst) {
929 : // Save VideoCodec instance for later; mainly for duplicating the decoder.
930 0 : codec_ = *inst;
931 : }
932 :
933 0 : if (!frame_buffer_pool_.InitializeVpxUsePool(decoder_)) {
934 0 : return WEBRTC_VIDEO_CODEC_MEMORY;
935 : }
936 :
937 0 : inited_ = true;
938 : // Always start with a complete key frame.
939 0 : key_frame_required_ = true;
940 0 : return WEBRTC_VIDEO_CODEC_OK;
941 : }
942 :
943 0 : int VP9DecoderImpl::Decode(const EncodedImage& input_image,
944 : bool missing_frames,
945 : const RTPFragmentationHeader* fragmentation,
946 : const CodecSpecificInfo* codec_specific_info,
947 : int64_t /*render_time_ms*/) {
948 0 : if (!inited_) {
949 0 : return WEBRTC_VIDEO_CODEC_UNINITIALIZED;
950 : }
951 0 : if (decode_complete_callback_ == NULL) {
952 0 : return WEBRTC_VIDEO_CODEC_UNINITIALIZED;
953 : }
954 : // Always start with a complete key frame.
955 0 : if (key_frame_required_) {
956 0 : if (input_image._frameType != kVideoFrameKey)
957 0 : return WEBRTC_VIDEO_CODEC_ERROR;
958 : // We have a key frame - is it complete?
959 0 : if (input_image._completeFrame) {
960 0 : key_frame_required_ = false;
961 : } else {
962 0 : return WEBRTC_VIDEO_CODEC_ERROR;
963 : }
964 : }
965 0 : vpx_codec_iter_t iter = NULL;
966 : vpx_image_t* img;
967 0 : uint8_t* buffer = input_image._buffer;
968 0 : if (input_image._length == 0) {
969 0 : buffer = NULL; // Triggers full frame concealment.
970 : }
971 : // During decode libvpx may get and release buffers from |frame_buffer_pool_|.
972 : // In practice libvpx keeps a few (~3-4) buffers alive at a time.
973 0 : if (vpx_codec_decode(decoder_, buffer,
974 0 : static_cast<unsigned int>(input_image._length), 0,
975 : VPX_DL_REALTIME)) {
976 0 : return WEBRTC_VIDEO_CODEC_ERROR;
977 : }
978 : // |img->fb_priv| contains the image data, a reference counted Vp9FrameBuffer.
979 : // It may be released by libvpx during future vpx_codec_decode or
980 : // vpx_codec_destroy calls.
981 0 : img = vpx_codec_get_frame(decoder_, &iter);
982 0 : int ret = ReturnFrame(img, input_image._timeStamp, input_image.ntp_time_ms_);
983 0 : if (ret != 0) {
984 0 : return ret;
985 : }
986 0 : return WEBRTC_VIDEO_CODEC_OK;
987 : }
988 :
989 0 : int VP9DecoderImpl::ReturnFrame(const vpx_image_t* img,
990 : uint32_t timestamp,
991 : int64_t ntp_time_ms) {
992 0 : if (img == NULL) {
993 : // Decoder OK and NULL image => No show frame.
994 0 : return WEBRTC_VIDEO_CODEC_NO_OUTPUT;
995 : }
996 :
997 : // This buffer contains all of |img|'s image data, a reference counted
998 : // Vp9FrameBuffer. (libvpx is done with the buffers after a few
999 : // vpx_codec_decode calls or vpx_codec_destroy).
1000 : Vp9FrameBufferPool::Vp9FrameBuffer* img_buffer =
1001 0 : static_cast<Vp9FrameBufferPool::Vp9FrameBuffer*>(img->fb_priv);
1002 : // The buffer can be used directly by the VideoFrame (without copy) by
1003 : // using a WrappedI420Buffer.
1004 : rtc::scoped_refptr<WrappedI420Buffer> img_wrapped_buffer(
1005 : new rtc::RefCountedObject<webrtc::WrappedI420Buffer>(
1006 : img->d_w, img->d_h, img->planes[VPX_PLANE_Y],
1007 : img->stride[VPX_PLANE_Y], img->planes[VPX_PLANE_U],
1008 : img->stride[VPX_PLANE_U], img->planes[VPX_PLANE_V],
1009 : img->stride[VPX_PLANE_V],
1010 : // WrappedI420Buffer's mechanism for allowing the release of its frame
1011 : // buffer is through a callback function. This is where we should
1012 : // release |img_buffer|.
1013 0 : rtc::KeepRefUntilDone(img_buffer)));
1014 :
1015 : VideoFrame decoded_image(img_wrapped_buffer, timestamp,
1016 0 : 0 /* render_time_ms */, webrtc::kVideoRotation_0);
1017 0 : decoded_image.set_ntp_time_ms(ntp_time_ms);
1018 0 : int ret = decode_complete_callback_->Decoded(decoded_image);
1019 0 : if (ret != 0)
1020 0 : return ret;
1021 0 : return WEBRTC_VIDEO_CODEC_OK;
1022 : }
1023 :
1024 0 : int VP9DecoderImpl::RegisterDecodeCompleteCallback(
1025 : DecodedImageCallback* callback) {
1026 0 : decode_complete_callback_ = callback;
1027 0 : return WEBRTC_VIDEO_CODEC_OK;
1028 : }
1029 :
1030 0 : int VP9DecoderImpl::Release() {
1031 0 : if (decoder_ != NULL) {
1032 : // When a codec is destroyed libvpx will release any buffers of
1033 : // |frame_buffer_pool_| it is currently using.
1034 0 : if (vpx_codec_destroy(decoder_)) {
1035 0 : return WEBRTC_VIDEO_CODEC_MEMORY;
1036 : }
1037 0 : delete decoder_;
1038 0 : decoder_ = NULL;
1039 : }
1040 : // Releases buffers from the pool. Any buffers not in use are deleted. Buffers
1041 : // still referenced externally are deleted once fully released, not returning
1042 : // to the pool.
1043 0 : frame_buffer_pool_.ClearPool();
1044 0 : inited_ = false;
1045 0 : return WEBRTC_VIDEO_CODEC_OK;
1046 : }
1047 :
1048 0 : const char* VP9DecoderImpl::ImplementationName() const {
1049 0 : return "libvpx";
1050 : }
1051 :
1052 : } // namespace webrtc
|