1 /*
2  *  Copyright (c) 2015 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 #include <stdio.h>
11 
12 #include <algorithm>
13 #include <deque>
14 #include <map>
15 #include <sstream>
16 #include <string>
17 #include <vector>
18 
19 #include "testing/gtest/include/gtest/gtest.h"
20 
21 #include "webrtc/base/checks.h"
22 #include "webrtc/base/event.h"
23 #include "webrtc/base/format_macros.h"
24 #include "webrtc/base/scoped_ptr.h"
25 #include "webrtc/call.h"
26 #include "webrtc/common_video/libyuv/include/webrtc_libyuv.h"
27 #include "webrtc/modules/rtp_rtcp/include/rtp_header_parser.h"
28 #include "webrtc/modules/rtp_rtcp/source/rtp_utility.h"
29 #include "webrtc/system_wrappers/include/cpu_info.h"
30 #include "webrtc/test/layer_filtering_transport.h"
31 #include "webrtc/test/run_loop.h"
32 #include "webrtc/test/statistics.h"
33 #include "webrtc/test/testsupport/fileutils.h"
34 #include "webrtc/test/video_renderer.h"
35 #include "webrtc/video/video_quality_test.h"
36 
37 namespace webrtc {
38 
39 static const int kSendStatsPollingIntervalMs = 1000;
40 static const int kPayloadTypeVP8 = 123;
41 static const int kPayloadTypeVP9 = 124;
42 
43 class VideoAnalyzer : public PacketReceiver,
44                       public Transport,
45                       public VideoRenderer,
46                       public VideoCaptureInput,
47                       public EncodedFrameObserver,
48                       public EncodingTimeObserver {
49  public:
VideoAnalyzer(test::LayerFilteringTransport * transport,const std::string & test_label,double avg_psnr_threshold,double avg_ssim_threshold,int duration_frames,FILE * graph_data_output_file,const std::string & graph_title,uint32_t ssrc_to_analyze)50   VideoAnalyzer(test::LayerFilteringTransport* transport,
51                 const std::string& test_label,
52                 double avg_psnr_threshold,
53                 double avg_ssim_threshold,
54                 int duration_frames,
55                 FILE* graph_data_output_file,
56                 const std::string& graph_title,
57                 uint32_t ssrc_to_analyze)
58       : input_(nullptr),
59         transport_(transport),
60         receiver_(nullptr),
61         send_stream_(nullptr),
62         test_label_(test_label),
63         graph_data_output_file_(graph_data_output_file),
64         graph_title_(graph_title),
65         ssrc_to_analyze_(ssrc_to_analyze),
66         frames_to_process_(duration_frames),
67         frames_recorded_(0),
68         frames_processed_(0),
69         dropped_frames_(0),
70         last_render_time_(0),
71         rtp_timestamp_delta_(0),
72         avg_psnr_threshold_(avg_psnr_threshold),
73         avg_ssim_threshold_(avg_ssim_threshold),
74         stats_polling_thread_(&PollStatsThread, this, "StatsPoller"),
75         comparison_available_event_(false, false),
76         done_(false, false) {
77     // Create thread pool for CPU-expensive PSNR/SSIM calculations.
78 
79     // Try to use about as many threads as cores, but leave kMinCoresLeft alone,
80     // so that we don't accidentally starve "real" worker threads (codec etc).
81     // Also, don't allocate more than kMaxComparisonThreads, even if there are
82     // spare cores.
83 
84     uint32_t num_cores = CpuInfo::DetectNumberOfCores();
85     RTC_DCHECK_GE(num_cores, 1u);
86     static const uint32_t kMinCoresLeft = 4;
87     static const uint32_t kMaxComparisonThreads = 8;
88 
89     if (num_cores <= kMinCoresLeft) {
90       num_cores = 1;
91     } else {
92       num_cores -= kMinCoresLeft;
93       num_cores = std::min(num_cores, kMaxComparisonThreads);
94     }
95 
96     for (uint32_t i = 0; i < num_cores; ++i) {
97       rtc::PlatformThread* thread =
98           new rtc::PlatformThread(&FrameComparisonThread, this, "Analyzer");
99       thread->Start();
100       comparison_thread_pool_.push_back(thread);
101     }
102   }
103 
~VideoAnalyzer()104   ~VideoAnalyzer() {
105     for (rtc::PlatformThread* thread : comparison_thread_pool_) {
106       thread->Stop();
107       delete thread;
108     }
109   }
110 
SetReceiver(PacketReceiver * receiver)111   virtual void SetReceiver(PacketReceiver* receiver) { receiver_ = receiver; }
112 
DeliverPacket(MediaType media_type,const uint8_t * packet,size_t length,const PacketTime & packet_time)113   DeliveryStatus DeliverPacket(MediaType media_type,
114                                const uint8_t* packet,
115                                size_t length,
116                                const PacketTime& packet_time) override {
117     RtpUtility::RtpHeaderParser parser(packet, length);
118     RTPHeader header;
119     parser.Parse(&header);
120     {
121       rtc::CritScope lock(&crit_);
122       recv_times_[header.timestamp - rtp_timestamp_delta_] =
123           Clock::GetRealTimeClock()->CurrentNtpInMilliseconds();
124     }
125 
126     return receiver_->DeliverPacket(media_type, packet, length, packet_time);
127   }
128 
129   // EncodingTimeObserver.
OnReportEncodedTime(int64_t ntp_time_ms,int encode_time_ms)130   void OnReportEncodedTime(int64_t ntp_time_ms, int encode_time_ms) override {
131     rtc::CritScope crit(&comparison_lock_);
132     samples_encode_time_ms_[ntp_time_ms] = encode_time_ms;
133   }
134 
IncomingCapturedFrame(const VideoFrame & video_frame)135   void IncomingCapturedFrame(const VideoFrame& video_frame) override {
136     VideoFrame copy = video_frame;
137     copy.set_timestamp(copy.ntp_time_ms() * 90);
138 
139     {
140       rtc::CritScope lock(&crit_);
141       if (first_send_frame_.IsZeroSize() && rtp_timestamp_delta_ == 0)
142         first_send_frame_ = copy;
143 
144       frames_.push_back(copy);
145     }
146 
147     input_->IncomingCapturedFrame(video_frame);
148   }
149 
SendRtp(const uint8_t * packet,size_t length,const PacketOptions & options)150   bool SendRtp(const uint8_t* packet,
151                size_t length,
152                const PacketOptions& options) override {
153     RtpUtility::RtpHeaderParser parser(packet, length);
154     RTPHeader header;
155     parser.Parse(&header);
156 
157     int64_t current_time =
158         Clock::GetRealTimeClock()->CurrentNtpInMilliseconds();
159     bool result = transport_->SendRtp(packet, length, options);
160     {
161       rtc::CritScope lock(&crit_);
162       if (rtp_timestamp_delta_ == 0) {
163         rtp_timestamp_delta_ = header.timestamp - first_send_frame_.timestamp();
164         first_send_frame_.Reset();
165       }
166       uint32_t timestamp = header.timestamp - rtp_timestamp_delta_;
167       send_times_[timestamp] = current_time;
168       if (!transport_->DiscardedLastPacket() &&
169           header.ssrc == ssrc_to_analyze_) {
170         encoded_frame_sizes_[timestamp] +=
171             length - (header.headerLength + header.paddingLength);
172       }
173     }
174     return result;
175   }
176 
SendRtcp(const uint8_t * packet,size_t length)177   bool SendRtcp(const uint8_t* packet, size_t length) override {
178     return transport_->SendRtcp(packet, length);
179   }
180 
EncodedFrameCallback(const EncodedFrame & frame)181   void EncodedFrameCallback(const EncodedFrame& frame) override {
182     rtc::CritScope lock(&comparison_lock_);
183     if (frames_recorded_ < frames_to_process_)
184       encoded_frame_size_.AddSample(frame.length_);
185   }
186 
RenderFrame(const VideoFrame & video_frame,int time_to_render_ms)187   void RenderFrame(const VideoFrame& video_frame,
188                    int time_to_render_ms) override {
189     int64_t render_time_ms =
190         Clock::GetRealTimeClock()->CurrentNtpInMilliseconds();
191     uint32_t send_timestamp = video_frame.timestamp() - rtp_timestamp_delta_;
192 
193     rtc::CritScope lock(&crit_);
194 
195     while (frames_.front().timestamp() < send_timestamp) {
196       AddFrameComparison(frames_.front(), last_rendered_frame_, true,
197                          render_time_ms);
198       frames_.pop_front();
199     }
200 
201     VideoFrame reference_frame = frames_.front();
202     frames_.pop_front();
203     assert(!reference_frame.IsZeroSize());
204     if (send_timestamp == reference_frame.timestamp() - 1) {
205       // TODO(ivica): Make this work for > 2 streams.
206       // Look at rtp_sender.c:RTPSender::BuildRTPHeader.
207       ++send_timestamp;
208     }
209     EXPECT_EQ(reference_frame.timestamp(), send_timestamp);
210     assert(reference_frame.timestamp() == send_timestamp);
211 
212     AddFrameComparison(reference_frame, video_frame, false, render_time_ms);
213 
214     last_rendered_frame_ = video_frame;
215   }
216 
IsTextureSupported() const217   bool IsTextureSupported() const override { return false; }
218 
Wait()219   void Wait() {
220     // Frame comparisons can be very expensive. Wait for test to be done, but
221     // at time-out check if frames_processed is going up. If so, give it more
222     // time, otherwise fail. Hopefully this will reduce test flakiness.
223 
224     stats_polling_thread_.Start();
225 
226     int last_frames_processed = -1;
227     int iteration = 0;
228     while (!done_.Wait(VideoQualityTest::kDefaultTimeoutMs)) {
229       int frames_processed;
230       {
231         rtc::CritScope crit(&comparison_lock_);
232         frames_processed = frames_processed_;
233       }
234 
235       // Print some output so test infrastructure won't think we've crashed.
236       const char* kKeepAliveMessages[3] = {
237           "Uh, I'm-I'm not quite dead, sir.",
238           "Uh, I-I think uh, I could pull through, sir.",
239           "Actually, I think I'm all right to come with you--"};
240       printf("- %s\n", kKeepAliveMessages[iteration++ % 3]);
241 
242       if (last_frames_processed == -1) {
243         last_frames_processed = frames_processed;
244         continue;
245       }
246       ASSERT_GT(frames_processed, last_frames_processed)
247           << "Analyzer stalled while waiting for test to finish.";
248       last_frames_processed = frames_processed;
249     }
250 
251     if (iteration > 0)
252       printf("- Farewell, sweet Concorde!\n");
253 
254     // Signal stats polling thread if that is still waiting and stop it now,
255     // since it uses the send_stream_ reference that might be reclaimed after
256     // returning from this method.
257     done_.Set();
258     stats_polling_thread_.Stop();
259   }
260 
261   VideoCaptureInput* input_;
262   test::LayerFilteringTransport* const transport_;
263   PacketReceiver* receiver_;
264   VideoSendStream* send_stream_;
265 
266  private:
267   struct FrameComparison {
FrameComparisonwebrtc::VideoAnalyzer::FrameComparison268     FrameComparison()
269         : dropped(false),
270           send_time_ms(0),
271           recv_time_ms(0),
272           render_time_ms(0),
273           encoded_frame_size(0) {}
274 
FrameComparisonwebrtc::VideoAnalyzer::FrameComparison275     FrameComparison(const VideoFrame& reference,
276                     const VideoFrame& render,
277                     bool dropped,
278                     int64_t send_time_ms,
279                     int64_t recv_time_ms,
280                     int64_t render_time_ms,
281                     size_t encoded_frame_size)
282         : reference(reference),
283           render(render),
284           dropped(dropped),
285           send_time_ms(send_time_ms),
286           recv_time_ms(recv_time_ms),
287           render_time_ms(render_time_ms),
288           encoded_frame_size(encoded_frame_size) {}
289 
290     VideoFrame reference;
291     VideoFrame render;
292     bool dropped;
293     int64_t send_time_ms;
294     int64_t recv_time_ms;
295     int64_t render_time_ms;
296     size_t encoded_frame_size;
297   };
298 
299   struct Sample {
Samplewebrtc::VideoAnalyzer::Sample300     Sample(int dropped,
301            int64_t input_time_ms,
302            int64_t send_time_ms,
303            int64_t recv_time_ms,
304            int64_t render_time_ms,
305            size_t encoded_frame_size,
306            double psnr,
307            double ssim)
308         : dropped(dropped),
309           input_time_ms(input_time_ms),
310           send_time_ms(send_time_ms),
311           recv_time_ms(recv_time_ms),
312           render_time_ms(render_time_ms),
313           encoded_frame_size(encoded_frame_size),
314           psnr(psnr),
315           ssim(ssim) {}
316 
317     int dropped;
318     int64_t input_time_ms;
319     int64_t send_time_ms;
320     int64_t recv_time_ms;
321     int64_t render_time_ms;
322     size_t encoded_frame_size;
323     double psnr;
324     double ssim;
325   };
326 
AddFrameComparison(const VideoFrame & reference,const VideoFrame & render,bool dropped,int64_t render_time_ms)327   void AddFrameComparison(const VideoFrame& reference,
328                           const VideoFrame& render,
329                           bool dropped,
330                           int64_t render_time_ms)
331       EXCLUSIVE_LOCKS_REQUIRED(crit_) {
332     int64_t send_time_ms = send_times_[reference.timestamp()];
333     send_times_.erase(reference.timestamp());
334     int64_t recv_time_ms = recv_times_[reference.timestamp()];
335     recv_times_.erase(reference.timestamp());
336 
337     // TODO(ivica): Make this work for > 2 streams.
338     auto it = encoded_frame_sizes_.find(reference.timestamp());
339     if (it == encoded_frame_sizes_.end())
340       it = encoded_frame_sizes_.find(reference.timestamp() - 1);
341     size_t encoded_size = it == encoded_frame_sizes_.end() ? 0 : it->second;
342     if (it != encoded_frame_sizes_.end())
343       encoded_frame_sizes_.erase(it);
344 
345     VideoFrame reference_copy;
346     VideoFrame render_copy;
347     reference_copy.CopyFrame(reference);
348     render_copy.CopyFrame(render);
349 
350     rtc::CritScope crit(&comparison_lock_);
351     comparisons_.push_back(FrameComparison(reference_copy, render_copy, dropped,
352                                            send_time_ms, recv_time_ms,
353                                            render_time_ms, encoded_size));
354     comparison_available_event_.Set();
355   }
356 
PollStatsThread(void * obj)357   static bool PollStatsThread(void* obj) {
358     return static_cast<VideoAnalyzer*>(obj)->PollStats();
359   }
360 
PollStats()361   bool PollStats() {
362     if (done_.Wait(kSendStatsPollingIntervalMs)) {
363       // Set event again to make sure main thread is also signaled, then we're
364       // done.
365       done_.Set();
366       return false;
367     }
368 
369     VideoSendStream::Stats stats = send_stream_->GetStats();
370 
371     rtc::CritScope crit(&comparison_lock_);
372     encode_frame_rate_.AddSample(stats.encode_frame_rate);
373     encode_time_ms.AddSample(stats.avg_encode_time_ms);
374     encode_usage_percent.AddSample(stats.encode_usage_percent);
375     media_bitrate_bps.AddSample(stats.media_bitrate_bps);
376 
377     return true;
378   }
379 
FrameComparisonThread(void * obj)380   static bool FrameComparisonThread(void* obj) {
381     return static_cast<VideoAnalyzer*>(obj)->CompareFrames();
382   }
383 
CompareFrames()384   bool CompareFrames() {
385     if (AllFramesRecorded())
386       return false;
387 
388     VideoFrame reference;
389     VideoFrame render;
390     FrameComparison comparison;
391 
392     if (!PopComparison(&comparison)) {
393       // Wait until new comparison task is available, or test is done.
394       // If done, wake up remaining threads waiting.
395       comparison_available_event_.Wait(1000);
396       if (AllFramesRecorded()) {
397         comparison_available_event_.Set();
398         return false;
399       }
400       return true;  // Try again.
401     }
402 
403     PerformFrameComparison(comparison);
404 
405     if (FrameProcessed()) {
406       PrintResults();
407       if (graph_data_output_file_)
408         PrintSamplesToFile();
409       done_.Set();
410       comparison_available_event_.Set();
411       return false;
412     }
413 
414     return true;
415   }
416 
PopComparison(FrameComparison * comparison)417   bool PopComparison(FrameComparison* comparison) {
418     rtc::CritScope crit(&comparison_lock_);
419     // If AllFramesRecorded() is true, it means we have already popped
420     // frames_to_process_ frames from comparisons_, so there is no more work
421     // for this thread to be done. frames_processed_ might still be lower if
422     // all comparisons are not done, but those frames are currently being
423     // worked on by other threads.
424     if (comparisons_.empty() || AllFramesRecorded())
425       return false;
426 
427     *comparison = comparisons_.front();
428     comparisons_.pop_front();
429 
430     FrameRecorded();
431     return true;
432   }
433 
434   // Increment counter for number of frames received for comparison.
FrameRecorded()435   void FrameRecorded() {
436     rtc::CritScope crit(&comparison_lock_);
437     ++frames_recorded_;
438   }
439 
440   // Returns true if all frames to be compared have been taken from the queue.
AllFramesRecorded()441   bool AllFramesRecorded() {
442     rtc::CritScope crit(&comparison_lock_);
443     assert(frames_recorded_ <= frames_to_process_);
444     return frames_recorded_ == frames_to_process_;
445   }
446 
447   // Increase count of number of frames processed. Returns true if this was the
448   // last frame to be processed.
FrameProcessed()449   bool FrameProcessed() {
450     rtc::CritScope crit(&comparison_lock_);
451     ++frames_processed_;
452     assert(frames_processed_ <= frames_to_process_);
453     return frames_processed_ == frames_to_process_;
454   }
455 
PrintResults()456   void PrintResults() {
457     rtc::CritScope crit(&comparison_lock_);
458     PrintResult("psnr", psnr_, " dB");
459     PrintResult("ssim", ssim_, "");
460     PrintResult("sender_time", sender_time_, " ms");
461     printf("RESULT dropped_frames: %s = %d frames\n", test_label_.c_str(),
462            dropped_frames_);
463     PrintResult("receiver_time", receiver_time_, " ms");
464     PrintResult("total_delay_incl_network", end_to_end_, " ms");
465     PrintResult("time_between_rendered_frames", rendered_delta_, " ms");
466     PrintResult("encoded_frame_size", encoded_frame_size_, " bytes");
467     PrintResult("encode_frame_rate", encode_frame_rate_, " fps");
468     PrintResult("encode_time", encode_time_ms, " ms");
469     PrintResult("encode_usage_percent", encode_usage_percent, " percent");
470     PrintResult("media_bitrate", media_bitrate_bps, " bps");
471 
472     EXPECT_GT(psnr_.Mean(), avg_psnr_threshold_);
473     EXPECT_GT(ssim_.Mean(), avg_ssim_threshold_);
474   }
475 
PerformFrameComparison(const FrameComparison & comparison)476   void PerformFrameComparison(const FrameComparison& comparison) {
477     // Perform expensive psnr and ssim calculations while not holding lock.
478     double psnr = I420PSNR(&comparison.reference, &comparison.render);
479     double ssim = I420SSIM(&comparison.reference, &comparison.render);
480 
481     int64_t input_time_ms = comparison.reference.ntp_time_ms();
482 
483     rtc::CritScope crit(&comparison_lock_);
484     if (graph_data_output_file_) {
485       samples_.push_back(
486           Sample(comparison.dropped, input_time_ms, comparison.send_time_ms,
487                  comparison.recv_time_ms, comparison.render_time_ms,
488                  comparison.encoded_frame_size, psnr, ssim));
489     }
490     psnr_.AddSample(psnr);
491     ssim_.AddSample(ssim);
492 
493     if (comparison.dropped) {
494       ++dropped_frames_;
495       return;
496     }
497     if (last_render_time_ != 0)
498       rendered_delta_.AddSample(comparison.render_time_ms - last_render_time_);
499     last_render_time_ = comparison.render_time_ms;
500 
501     sender_time_.AddSample(comparison.send_time_ms - input_time_ms);
502     receiver_time_.AddSample(comparison.render_time_ms -
503                              comparison.recv_time_ms);
504     end_to_end_.AddSample(comparison.render_time_ms - input_time_ms);
505     encoded_frame_size_.AddSample(comparison.encoded_frame_size);
506   }
507 
PrintResult(const char * result_type,test::Statistics stats,const char * unit)508   void PrintResult(const char* result_type,
509                    test::Statistics stats,
510                    const char* unit) {
511     printf("RESULT %s: %s = {%f, %f}%s\n",
512            result_type,
513            test_label_.c_str(),
514            stats.Mean(),
515            stats.StandardDeviation(),
516            unit);
517   }
518 
PrintSamplesToFile(void)519   void PrintSamplesToFile(void) {
520     FILE* out = graph_data_output_file_;
521     rtc::CritScope crit(&comparison_lock_);
522     std::sort(samples_.begin(), samples_.end(),
523               [](const Sample& A, const Sample& B) -> bool {
524                 return A.input_time_ms < B.input_time_ms;
525               });
526 
527     fprintf(out, "%s\n", graph_title_.c_str());
528     fprintf(out, "%" PRIuS "\n", samples_.size());
529     fprintf(out,
530             "dropped "
531             "input_time_ms "
532             "send_time_ms "
533             "recv_time_ms "
534             "render_time_ms "
535             "encoded_frame_size "
536             "psnr "
537             "ssim "
538             "encode_time_ms\n");
539     int missing_encode_time_samples = 0;
540     for (const Sample& sample : samples_) {
541       auto it = samples_encode_time_ms_.find(sample.input_time_ms);
542       int encode_time_ms;
543       if (it != samples_encode_time_ms_.end()) {
544         encode_time_ms = it->second;
545       } else {
546         ++missing_encode_time_samples;
547         encode_time_ms = -1;
548       }
549       fprintf(out, "%d %" PRId64 " %" PRId64 " %" PRId64 " %" PRId64 " %" PRIuS
550                    " %lf %lf %d\n",
551               sample.dropped, sample.input_time_ms, sample.send_time_ms,
552               sample.recv_time_ms, sample.render_time_ms,
553               sample.encoded_frame_size, sample.psnr, sample.ssim,
554               encode_time_ms);
555     }
556     if (missing_encode_time_samples) {
557       fprintf(stderr,
558               "Warning: Missing encode_time_ms samples for %d frame(s).\n",
559               missing_encode_time_samples);
560     }
561   }
562 
563   const std::string test_label_;
564   FILE* const graph_data_output_file_;
565   const std::string graph_title_;
566   const uint32_t ssrc_to_analyze_;
567   std::vector<Sample> samples_ GUARDED_BY(comparison_lock_);
568   std::map<int64_t, int> samples_encode_time_ms_ GUARDED_BY(comparison_lock_);
569   test::Statistics sender_time_ GUARDED_BY(comparison_lock_);
570   test::Statistics receiver_time_ GUARDED_BY(comparison_lock_);
571   test::Statistics psnr_ GUARDED_BY(comparison_lock_);
572   test::Statistics ssim_ GUARDED_BY(comparison_lock_);
573   test::Statistics end_to_end_ GUARDED_BY(comparison_lock_);
574   test::Statistics rendered_delta_ GUARDED_BY(comparison_lock_);
575   test::Statistics encoded_frame_size_ GUARDED_BY(comparison_lock_);
576   test::Statistics encode_frame_rate_ GUARDED_BY(comparison_lock_);
577   test::Statistics encode_time_ms GUARDED_BY(comparison_lock_);
578   test::Statistics encode_usage_percent GUARDED_BY(comparison_lock_);
579   test::Statistics media_bitrate_bps GUARDED_BY(comparison_lock_);
580 
581   const int frames_to_process_;
582   int frames_recorded_;
583   int frames_processed_;
584   int dropped_frames_;
585   int64_t last_render_time_;
586   uint32_t rtp_timestamp_delta_;
587 
588   rtc::CriticalSection crit_;
589   std::deque<VideoFrame> frames_ GUARDED_BY(crit_);
590   VideoFrame last_rendered_frame_ GUARDED_BY(crit_);
591   std::map<uint32_t, int64_t> send_times_ GUARDED_BY(crit_);
592   std::map<uint32_t, int64_t> recv_times_ GUARDED_BY(crit_);
593   std::map<uint32_t, size_t> encoded_frame_sizes_ GUARDED_BY(crit_);
594   VideoFrame first_send_frame_ GUARDED_BY(crit_);
595   const double avg_psnr_threshold_;
596   const double avg_ssim_threshold_;
597 
598   rtc::CriticalSection comparison_lock_;
599   std::vector<rtc::PlatformThread*> comparison_thread_pool_;
600   rtc::PlatformThread stats_polling_thread_;
601   rtc::Event comparison_available_event_;
602   std::deque<FrameComparison> comparisons_ GUARDED_BY(comparison_lock_);
603   rtc::Event done_;
604 };
605 
VideoQualityTest()606 VideoQualityTest::VideoQualityTest() : clock_(Clock::GetRealTimeClock()) {}
607 
TestBody()608 void VideoQualityTest::TestBody() {}
609 
GenerateGraphTitle() const610 std::string VideoQualityTest::GenerateGraphTitle() const {
611   std::stringstream ss;
612   ss << params_.common.codec;
613   ss << " (" << params_.common.target_bitrate_bps / 1000 << "kbps";
614   ss << ", " << params_.common.fps << " FPS";
615   if (params_.screenshare.scroll_duration)
616     ss << ", " << params_.screenshare.scroll_duration << "s scroll";
617   if (params_.ss.streams.size() > 1)
618     ss << ", Stream #" << params_.ss.selected_stream;
619   if (params_.ss.num_spatial_layers > 1)
620     ss << ", Layer #" << params_.ss.selected_sl;
621   ss << ")";
622   return ss.str();
623 }
624 
CheckParams()625 void VideoQualityTest::CheckParams() {
626   // Add a default stream in none specified.
627   if (params_.ss.streams.empty())
628     params_.ss.streams.push_back(VideoQualityTest::DefaultVideoStream(params_));
629   if (params_.ss.num_spatial_layers == 0)
630     params_.ss.num_spatial_layers = 1;
631 
632   if (params_.pipe.loss_percent != 0 ||
633       params_.pipe.queue_length_packets != 0) {
634     // Since LayerFilteringTransport changes the sequence numbers, we can't
635     // use that feature with pack loss, since the NACK request would end up
636     // retransmitting the wrong packets.
637     RTC_CHECK(params_.ss.selected_sl == -1 ||
638               params_.ss.selected_sl == params_.ss.num_spatial_layers - 1);
639     RTC_CHECK(params_.common.selected_tl == -1 ||
640               params_.common.selected_tl ==
641                   params_.common.num_temporal_layers - 1);
642   }
643 
644   // TODO(ivica): Should max_bitrate_bps == -1 represent inf max bitrate, as it
645   // does in some parts of the code?
646   RTC_CHECK_GE(params_.common.max_bitrate_bps,
647                params_.common.target_bitrate_bps);
648   RTC_CHECK_GE(params_.common.target_bitrate_bps,
649                params_.common.min_bitrate_bps);
650   RTC_CHECK_LT(params_.common.selected_tl, params_.common.num_temporal_layers);
651   RTC_CHECK_LT(params_.ss.selected_stream, params_.ss.streams.size());
652   for (const VideoStream& stream : params_.ss.streams) {
653     RTC_CHECK_GE(stream.min_bitrate_bps, 0);
654     RTC_CHECK_GE(stream.target_bitrate_bps, stream.min_bitrate_bps);
655     RTC_CHECK_GE(stream.max_bitrate_bps, stream.target_bitrate_bps);
656     RTC_CHECK_EQ(static_cast<int>(stream.temporal_layer_thresholds_bps.size()),
657                  params_.common.num_temporal_layers - 1);
658   }
659   // TODO(ivica): Should we check if the sum of all streams/layers is equal to
660   // the total bitrate? We anyway have to update them in the case bitrate
661   // estimator changes the total bitrates.
662   RTC_CHECK_GE(params_.ss.num_spatial_layers, 1);
663   RTC_CHECK_LE(params_.ss.selected_sl, params_.ss.num_spatial_layers);
664   RTC_CHECK(params_.ss.spatial_layers.empty() ||
665             params_.ss.spatial_layers.size() ==
666                 static_cast<size_t>(params_.ss.num_spatial_layers));
667   if (params_.common.codec == "VP8") {
668     RTC_CHECK_EQ(params_.ss.num_spatial_layers, 1);
669   } else if (params_.common.codec == "VP9") {
670     RTC_CHECK_EQ(params_.ss.streams.size(), 1u);
671   }
672 }
673 
674 // Static.
ParseCSV(const std::string & str)675 std::vector<int> VideoQualityTest::ParseCSV(const std::string& str) {
676   // Parse comma separated nonnegative integers, where some elements may be
677   // empty. The empty values are replaced with -1.
678   // E.g. "10,-20,,30,40" --> {10, 20, -1, 30,40}
679   // E.g. ",,10,,20," --> {-1, -1, 10, -1, 20, -1}
680   std::vector<int> result;
681   if (str.empty())
682     return result;
683 
684   const char* p = str.c_str();
685   int value = -1;
686   int pos;
687   while (*p) {
688     if (*p == ',') {
689       result.push_back(value);
690       value = -1;
691       ++p;
692       continue;
693     }
694     RTC_CHECK_EQ(sscanf(p, "%d%n", &value, &pos), 1)
695         << "Unexpected non-number value.";
696     p += pos;
697   }
698   result.push_back(value);
699   return result;
700 }
701 
702 // Static.
DefaultVideoStream(const Params & params)703 VideoStream VideoQualityTest::DefaultVideoStream(const Params& params) {
704   VideoStream stream;
705   stream.width = params.common.width;
706   stream.height = params.common.height;
707   stream.max_framerate = params.common.fps;
708   stream.min_bitrate_bps = params.common.min_bitrate_bps;
709   stream.target_bitrate_bps = params.common.target_bitrate_bps;
710   stream.max_bitrate_bps = params.common.max_bitrate_bps;
711   stream.max_qp = 52;
712   if (params.common.num_temporal_layers == 2)
713     stream.temporal_layer_thresholds_bps.push_back(stream.target_bitrate_bps);
714   return stream;
715 }
716 
717 // Static.
FillScalabilitySettings(Params * params,const std::vector<std::string> & stream_descriptors,size_t selected_stream,int num_spatial_layers,int selected_sl,const std::vector<std::string> & sl_descriptors)718 void VideoQualityTest::FillScalabilitySettings(
719     Params* params,
720     const std::vector<std::string>& stream_descriptors,
721     size_t selected_stream,
722     int num_spatial_layers,
723     int selected_sl,
724     const std::vector<std::string>& sl_descriptors) {
725   // Read VideoStream and SpatialLayer elements from a list of comma separated
726   // lists. To use a default value for an element, use -1 or leave empty.
727   // Validity checks performed in CheckParams.
728 
729   RTC_CHECK(params->ss.streams.empty());
730   for (auto descriptor : stream_descriptors) {
731     if (descriptor.empty())
732       continue;
733     VideoStream stream = VideoQualityTest::DefaultVideoStream(*params);
734     std::vector<int> v = VideoQualityTest::ParseCSV(descriptor);
735     if (v[0] != -1)
736       stream.width = static_cast<size_t>(v[0]);
737     if (v[1] != -1)
738       stream.height = static_cast<size_t>(v[1]);
739     if (v[2] != -1)
740       stream.max_framerate = v[2];
741     if (v[3] != -1)
742       stream.min_bitrate_bps = v[3];
743     if (v[4] != -1)
744       stream.target_bitrate_bps = v[4];
745     if (v[5] != -1)
746       stream.max_bitrate_bps = v[5];
747     if (v.size() > 6 && v[6] != -1)
748       stream.max_qp = v[6];
749     if (v.size() > 7) {
750       stream.temporal_layer_thresholds_bps.clear();
751       stream.temporal_layer_thresholds_bps.insert(
752           stream.temporal_layer_thresholds_bps.end(), v.begin() + 7, v.end());
753     } else {
754       // Automatic TL thresholds for more than two layers not supported.
755       RTC_CHECK_LE(params->common.num_temporal_layers, 2);
756     }
757     params->ss.streams.push_back(stream);
758   }
759   params->ss.selected_stream = selected_stream;
760 
761   params->ss.num_spatial_layers = num_spatial_layers ? num_spatial_layers : 1;
762   params->ss.selected_sl = selected_sl;
763   RTC_CHECK(params->ss.spatial_layers.empty());
764   for (auto descriptor : sl_descriptors) {
765     if (descriptor.empty())
766       continue;
767     std::vector<int> v = VideoQualityTest::ParseCSV(descriptor);
768     RTC_CHECK_GT(v[2], 0);
769 
770     SpatialLayer layer;
771     layer.scaling_factor_num = v[0] == -1 ? 1 : v[0];
772     layer.scaling_factor_den = v[1] == -1 ? 1 : v[1];
773     layer.target_bitrate_bps = v[2];
774     params->ss.spatial_layers.push_back(layer);
775   }
776 }
777 
SetupCommon(Transport * send_transport,Transport * recv_transport)778 void VideoQualityTest::SetupCommon(Transport* send_transport,
779                                    Transport* recv_transport) {
780   if (params_.logs)
781     trace_to_stderr_.reset(new test::TraceToStderr);
782 
783   size_t num_streams = params_.ss.streams.size();
784   CreateSendConfig(num_streams, 0, send_transport);
785 
786   int payload_type;
787   if (params_.common.codec == "VP8") {
788     encoder_.reset(VideoEncoder::Create(VideoEncoder::kVp8));
789     payload_type = kPayloadTypeVP8;
790   } else if (params_.common.codec == "VP9") {
791     encoder_.reset(VideoEncoder::Create(VideoEncoder::kVp9));
792     payload_type = kPayloadTypeVP9;
793   } else {
794     RTC_NOTREACHED() << "Codec not supported!";
795     return;
796   }
797   video_send_config_.encoder_settings.encoder = encoder_.get();
798   video_send_config_.encoder_settings.payload_name = params_.common.codec;
799   video_send_config_.encoder_settings.payload_type = payload_type;
800   video_send_config_.rtp.nack.rtp_history_ms = kNackRtpHistoryMs;
801   video_send_config_.rtp.rtx.payload_type = kSendRtxPayloadType;
802   for (size_t i = 0; i < num_streams; ++i)
803     video_send_config_.rtp.rtx.ssrcs.push_back(kSendRtxSsrcs[i]);
804 
805   video_send_config_.rtp.extensions.clear();
806   if (params_.common.send_side_bwe) {
807     video_send_config_.rtp.extensions.push_back(
808         RtpExtension(RtpExtension::kTransportSequenceNumber,
809                      test::kTransportSequenceNumberExtensionId));
810   } else {
811     video_send_config_.rtp.extensions.push_back(RtpExtension(
812         RtpExtension::kAbsSendTime, test::kAbsSendTimeExtensionId));
813   }
814 
815   video_encoder_config_.min_transmit_bitrate_bps =
816       params_.common.min_transmit_bps;
817   video_encoder_config_.streams = params_.ss.streams;
818   video_encoder_config_.spatial_layers = params_.ss.spatial_layers;
819 
820   CreateMatchingReceiveConfigs(recv_transport);
821 
822   for (size_t i = 0; i < num_streams; ++i) {
823     video_receive_configs_[i].rtp.nack.rtp_history_ms = kNackRtpHistoryMs;
824     video_receive_configs_[i].rtp.rtx[kSendRtxPayloadType].ssrc =
825         kSendRtxSsrcs[i];
826     video_receive_configs_[i].rtp.rtx[kSendRtxPayloadType].payload_type =
827         kSendRtxPayloadType;
828     video_receive_configs_[i].rtp.transport_cc = params_.common.send_side_bwe;
829   }
830 }
831 
SetupScreenshare()832 void VideoQualityTest::SetupScreenshare() {
833   RTC_CHECK(params_.screenshare.enabled);
834 
835   // Fill out codec settings.
836   video_encoder_config_.content_type = VideoEncoderConfig::ContentType::kScreen;
837   if (params_.common.codec == "VP8") {
838     codec_settings_.VP8 = VideoEncoder::GetDefaultVp8Settings();
839     codec_settings_.VP8.denoisingOn = false;
840     codec_settings_.VP8.frameDroppingOn = false;
841     codec_settings_.VP8.numberOfTemporalLayers =
842         static_cast<unsigned char>(params_.common.num_temporal_layers);
843     video_encoder_config_.encoder_specific_settings = &codec_settings_.VP8;
844   } else if (params_.common.codec == "VP9") {
845     codec_settings_.VP9 = VideoEncoder::GetDefaultVp9Settings();
846     codec_settings_.VP9.denoisingOn = false;
847     codec_settings_.VP9.frameDroppingOn = false;
848     codec_settings_.VP9.numberOfTemporalLayers =
849         static_cast<unsigned char>(params_.common.num_temporal_layers);
850     video_encoder_config_.encoder_specific_settings = &codec_settings_.VP9;
851     codec_settings_.VP9.numberOfSpatialLayers =
852         static_cast<unsigned char>(params_.ss.num_spatial_layers);
853   }
854 
855   // Setup frame generator.
856   const size_t kWidth = 1850;
857   const size_t kHeight = 1110;
858   std::vector<std::string> slides;
859   slides.push_back(test::ResourcePath("web_screenshot_1850_1110", "yuv"));
860   slides.push_back(test::ResourcePath("presentation_1850_1110", "yuv"));
861   slides.push_back(test::ResourcePath("photo_1850_1110", "yuv"));
862   slides.push_back(test::ResourcePath("difficult_photo_1850_1110", "yuv"));
863 
864   if (params_.screenshare.scroll_duration == 0) {
865     // Cycle image every slide_change_interval seconds.
866     frame_generator_.reset(test::FrameGenerator::CreateFromYuvFile(
867         slides, kWidth, kHeight,
868         params_.screenshare.slide_change_interval * params_.common.fps));
869   } else {
870     RTC_CHECK_LE(params_.common.width, kWidth);
871     RTC_CHECK_LE(params_.common.height, kHeight);
872     RTC_CHECK_GT(params_.screenshare.slide_change_interval, 0);
873     const int kPauseDurationMs = (params_.screenshare.slide_change_interval -
874                                   params_.screenshare.scroll_duration) *
875                                  1000;
876     RTC_CHECK_LE(params_.screenshare.scroll_duration,
877                  params_.screenshare.slide_change_interval);
878 
879     frame_generator_.reset(
880         test::FrameGenerator::CreateScrollingInputFromYuvFiles(
881             clock_, slides, kWidth, kHeight, params_.common.width,
882             params_.common.height, params_.screenshare.scroll_duration * 1000,
883             kPauseDurationMs));
884   }
885 }
886 
CreateCapturer(VideoCaptureInput * input)887 void VideoQualityTest::CreateCapturer(VideoCaptureInput* input) {
888   if (params_.screenshare.enabled) {
889     test::FrameGeneratorCapturer* frame_generator_capturer =
890         new test::FrameGeneratorCapturer(
891             clock_, input, frame_generator_.release(), params_.common.fps);
892     EXPECT_TRUE(frame_generator_capturer->Init());
893     capturer_.reset(frame_generator_capturer);
894   } else {
895     if (params_.video.clip_name.empty()) {
896       capturer_.reset(test::VideoCapturer::Create(input, params_.common.width,
897                                                   params_.common.height,
898                                                   params_.common.fps, clock_));
899     } else {
900       capturer_.reset(test::FrameGeneratorCapturer::CreateFromYuvFile(
901           input, test::ResourcePath(params_.video.clip_name, "yuv"),
902           params_.common.width, params_.common.height, params_.common.fps,
903           clock_));
904       ASSERT_TRUE(capturer_.get() != nullptr)
905           << "Could not create capturer for " << params_.video.clip_name
906           << ".yuv. Is this resource file present?";
907     }
908   }
909 }
910 
RunWithAnalyzer(const Params & params)911 void VideoQualityTest::RunWithAnalyzer(const Params& params) {
912   params_ = params;
913 
914   // TODO(ivica): Merge with RunWithRenderer and use a flag / argument to
915   // differentiate between the analyzer and the renderer case.
916   CheckParams();
917 
918   FILE* graph_data_output_file = nullptr;
919   if (!params_.analyzer.graph_data_output_filename.empty()) {
920     graph_data_output_file =
921         fopen(params_.analyzer.graph_data_output_filename.c_str(), "w");
922     RTC_CHECK(graph_data_output_file != nullptr)
923         << "Can't open the file " << params_.analyzer.graph_data_output_filename
924         << "!";
925   }
926 
927   Call::Config call_config;
928   call_config.bitrate_config = params.common.call_bitrate_config;
929   CreateCalls(call_config, call_config);
930 
931   test::LayerFilteringTransport send_transport(
932       params.pipe, sender_call_.get(), kPayloadTypeVP8, kPayloadTypeVP9,
933       params.common.selected_tl, params_.ss.selected_sl);
934   test::DirectTransport recv_transport(params.pipe, receiver_call_.get());
935 
936   std::string graph_title = params_.analyzer.graph_title;
937   if (graph_title.empty())
938     graph_title = VideoQualityTest::GenerateGraphTitle();
939 
940   // In the case of different resolutions, the functions calculating PSNR and
941   // SSIM return -1.0, instead of a positive value as usual. VideoAnalyzer
942   // aborts if the average psnr/ssim are below the given threshold, which is
943   // 0.0 by default. Setting the thresholds to -1.1 prevents the unnecessary
944   // abort.
945   VideoStream& selected_stream = params_.ss.streams[params_.ss.selected_stream];
946   int selected_sl = params_.ss.selected_sl != -1
947                         ? params_.ss.selected_sl
948                         : params_.ss.num_spatial_layers - 1;
949   bool disable_quality_check =
950       selected_stream.width != params_.common.width ||
951       selected_stream.height != params_.common.height ||
952       (!params_.ss.spatial_layers.empty() &&
953        params_.ss.spatial_layers[selected_sl].scaling_factor_num !=
954            params_.ss.spatial_layers[selected_sl].scaling_factor_den);
955   if (disable_quality_check) {
956     fprintf(stderr,
957             "Warning: Calculating PSNR and SSIM for downsized resolution "
958             "not implemented yet! Skipping PSNR and SSIM calculations!");
959   }
960 
961   VideoAnalyzer analyzer(
962       &send_transport, params_.analyzer.test_label,
963       disable_quality_check ? -1.1 : params_.analyzer.avg_psnr_threshold,
964       disable_quality_check ? -1.1 : params_.analyzer.avg_ssim_threshold,
965       params_.analyzer.test_durations_secs * params_.common.fps,
966       graph_data_output_file, graph_title,
967       kVideoSendSsrcs[params_.ss.selected_stream]);
968 
969   analyzer.SetReceiver(receiver_call_->Receiver());
970   send_transport.SetReceiver(&analyzer);
971   recv_transport.SetReceiver(sender_call_->Receiver());
972 
973   SetupCommon(&analyzer, &recv_transport);
974   video_send_config_.encoding_time_observer = &analyzer;
975   video_receive_configs_[params_.ss.selected_stream].renderer = &analyzer;
976   for (auto& config : video_receive_configs_)
977     config.pre_decode_callback = &analyzer;
978 
979   if (params_.screenshare.enabled)
980     SetupScreenshare();
981 
982   CreateVideoStreams();
983   analyzer.input_ = video_send_stream_->Input();
984   analyzer.send_stream_ = video_send_stream_;
985 
986   CreateCapturer(&analyzer);
987 
988   video_send_stream_->Start();
989   for (VideoReceiveStream* receive_stream : video_receive_streams_)
990     receive_stream->Start();
991   capturer_->Start();
992 
993   analyzer.Wait();
994 
995   send_transport.StopSending();
996   recv_transport.StopSending();
997 
998   capturer_->Stop();
999   for (VideoReceiveStream* receive_stream : video_receive_streams_)
1000     receive_stream->Stop();
1001   video_send_stream_->Stop();
1002 
1003   DestroyStreams();
1004 
1005   if (graph_data_output_file)
1006     fclose(graph_data_output_file);
1007 }
1008 
RunWithVideoRenderer(const Params & params)1009 void VideoQualityTest::RunWithVideoRenderer(const Params& params) {
1010   params_ = params;
1011   CheckParams();
1012 
1013   rtc::scoped_ptr<test::VideoRenderer> local_preview(
1014       test::VideoRenderer::Create("Local Preview", params_.common.width,
1015                                   params_.common.height));
1016   size_t stream_id = params_.ss.selected_stream;
1017   std::string title = "Loopback Video";
1018   if (params_.ss.streams.size() > 1) {
1019     std::ostringstream s;
1020     s << stream_id;
1021     title += " - Stream #" + s.str();
1022   }
1023 
1024   rtc::scoped_ptr<test::VideoRenderer> loopback_video(
1025       test::VideoRenderer::Create(title.c_str(),
1026                                   params_.ss.streams[stream_id].width,
1027                                   params_.ss.streams[stream_id].height));
1028 
1029   // TODO(ivica): Remove bitrate_config and use the default Call::Config(), to
1030   // match the full stack tests.
1031   Call::Config call_config;
1032   call_config.bitrate_config = params_.common.call_bitrate_config;
1033   rtc::scoped_ptr<Call> call(Call::Create(call_config));
1034 
1035   test::LayerFilteringTransport transport(
1036       params.pipe, call.get(), kPayloadTypeVP8, kPayloadTypeVP9,
1037       params.common.selected_tl, params_.ss.selected_sl);
1038   // TODO(ivica): Use two calls to be able to merge with RunWithAnalyzer or at
1039   // least share as much code as possible. That way this test would also match
1040   // the full stack tests better.
1041   transport.SetReceiver(call->Receiver());
1042 
1043   SetupCommon(&transport, &transport);
1044 
1045   video_send_config_.local_renderer = local_preview.get();
1046   video_receive_configs_[stream_id].renderer = loopback_video.get();
1047 
1048   if (params_.screenshare.enabled)
1049     SetupScreenshare();
1050 
1051   video_send_stream_ =
1052       call->CreateVideoSendStream(video_send_config_, video_encoder_config_);
1053   VideoReceiveStream* receive_stream =
1054       call->CreateVideoReceiveStream(video_receive_configs_[stream_id]);
1055   CreateCapturer(video_send_stream_->Input());
1056 
1057   receive_stream->Start();
1058   video_send_stream_->Start();
1059   capturer_->Start();
1060 
1061   test::PressEnterToContinue();
1062 
1063   capturer_->Stop();
1064   video_send_stream_->Stop();
1065   receive_stream->Stop();
1066 
1067   call->DestroyVideoReceiveStream(receive_stream);
1068   call->DestroyVideoSendStream(video_send_stream_);
1069 
1070   transport.StopSending();
1071 }
1072 
1073 }  // namespace webrtc
1074