| /* |
| * Copyright (c) 2013 The WebRTC project authors. All Rights Reserved. |
| * |
| * Use of this source code is governed by a BSD-style license |
| * that can be found in the LICENSE file in the root of the source |
| * tree. An additional intellectual property rights grant can be found |
| * in the file PATENTS. All contributing project authors may |
| * be found in the AUTHORS file in the root of the source tree. |
| */ |
| |
| #include <memory> |
| |
| #include "api/test/simulated_network.h" |
| #include "api/test/video/function_video_encoder_factory.h" |
| #include "call/fake_network_pipe.h" |
| #include "call/simulated_network.h" |
| #include "media/engine/internal_encoder_factory.h" |
| #include "media/engine/simulcast_encoder_adapter.h" |
| #include "modules/rtp_rtcp/source/create_video_rtp_depacketizer.h" |
| #include "modules/rtp_rtcp/source/rtp_packet.h" |
| #include "modules/video_coding/codecs/vp8/include/vp8.h" |
| #include "modules/video_coding/codecs/vp9/include/vp9.h" |
| #include "rtc_base/numerics/safe_conversions.h" |
| #include "rtc_base/numerics/sequence_number_util.h" |
| #include "rtc_base/synchronization/mutex.h" |
| #include "rtc_base/task_queue_for_test.h" |
| #include "test/call_test.h" |
| #include "test/video_test_constants.h" |
| |
| namespace webrtc { |
| namespace { |
| const int kFrameMaxWidth = 1280; |
| const int kFrameMaxHeight = 720; |
| const int kFrameRate = 30; |
| const int kMaxSecondsLost = 5; |
| const int kMaxFramesLost = kFrameRate * kMaxSecondsLost; |
| const int kMinPacketsToObserve = 10; |
| const int kEncoderBitrateBps = 300000; |
| const uint32_t kPictureIdWraparound = (1 << 15); |
| const size_t kNumTemporalLayers[] = {1, 2, 3}; |
| |
| } // namespace |
| |
| class PictureIdObserver : public test::RtpRtcpObserver { |
| public: |
| explicit PictureIdObserver(VideoCodecType codec_type) |
| : test::RtpRtcpObserver(test::VideoTestConstants::kDefaultTimeout), |
| depacketizer_(CreateVideoRtpDepacketizer(codec_type)), |
| max_expected_picture_id_gap_(0), |
| max_expected_tl0_idx_gap_(0), |
| num_ssrcs_to_observe_(1) {} |
| |
| void SetExpectedSsrcs(size_t num_expected_ssrcs) { |
| MutexLock lock(&mutex_); |
| num_ssrcs_to_observe_ = num_expected_ssrcs; |
| } |
| |
| void ResetObservedSsrcs() { |
| MutexLock lock(&mutex_); |
| // Do not clear the timestamp and picture_id, to ensure that we check |
| // consistency between reinits and recreations. |
| num_packets_sent_.clear(); |
| observed_ssrcs_.clear(); |
| } |
| |
| void SetMaxExpectedPictureIdGap(int max_expected_picture_id_gap) { |
| MutexLock lock(&mutex_); |
| max_expected_picture_id_gap_ = max_expected_picture_id_gap; |
| // Expect smaller gap for `tl0_pic_idx` (running index for temporal_idx 0). |
| max_expected_tl0_idx_gap_ = max_expected_picture_id_gap_ / 2; |
| } |
| |
| private: |
| struct ParsedPacket { |
| uint32_t timestamp; |
| uint32_t ssrc; |
| int16_t picture_id; |
| int16_t tl0_pic_idx; |
| uint8_t temporal_idx; |
| VideoFrameType frame_type; |
| }; |
| |
| bool ParsePayload(const uint8_t* packet, |
| size_t length, |
| ParsedPacket* parsed) const { |
| RtpPacket rtp_packet; |
| EXPECT_TRUE(rtp_packet.Parse(packet, length)); |
| EXPECT_TRUE( |
| rtp_packet.Ssrc() == test::VideoTestConstants::kVideoSendSsrcs[0] || |
| rtp_packet.Ssrc() == test::VideoTestConstants::kVideoSendSsrcs[1] || |
| rtp_packet.Ssrc() == test::VideoTestConstants::kVideoSendSsrcs[2]) |
| << "Unknown SSRC sent."; |
| |
| if (rtp_packet.payload_size() == 0) { |
| return false; // Padding packet. |
| } |
| |
| parsed->timestamp = rtp_packet.Timestamp(); |
| parsed->ssrc = rtp_packet.Ssrc(); |
| |
| absl::optional<VideoRtpDepacketizer::ParsedRtpPayload> parsed_payload = |
| depacketizer_->Parse(rtp_packet.PayloadBuffer()); |
| EXPECT_TRUE(parsed_payload); |
| |
| if (const auto* vp8_header = absl::get_if<RTPVideoHeaderVP8>( |
| &parsed_payload->video_header.video_type_header)) { |
| parsed->picture_id = vp8_header->pictureId; |
| parsed->tl0_pic_idx = vp8_header->tl0PicIdx; |
| parsed->temporal_idx = vp8_header->temporalIdx; |
| } else if (const auto* vp9_header = absl::get_if<RTPVideoHeaderVP9>( |
| &parsed_payload->video_header.video_type_header)) { |
| parsed->picture_id = vp9_header->picture_id; |
| parsed->tl0_pic_idx = vp9_header->tl0_pic_idx; |
| parsed->temporal_idx = vp9_header->temporal_idx; |
| } else { |
| RTC_DCHECK_NOTREACHED(); |
| } |
| |
| parsed->frame_type = parsed_payload->video_header.frame_type; |
| return true; |
| } |
| |
| // Verify continuity and monotonicity of picture_id sequence. |
| void VerifyPictureId(const ParsedPacket& current, |
| const ParsedPacket& last) const |
| RTC_EXCLUSIVE_LOCKS_REQUIRED(&mutex_) { |
| if (current.timestamp == last.timestamp) { |
| EXPECT_EQ(last.picture_id, current.picture_id); |
| return; // Same frame. |
| } |
| |
| // Packet belongs to a new frame. |
| // Picture id should be increasing. |
| EXPECT_TRUE((AheadOf<uint16_t, kPictureIdWraparound>(current.picture_id, |
| last.picture_id))); |
| |
| // Expect continuously increasing picture id. |
| int diff = ForwardDiff<uint16_t, kPictureIdWraparound>(last.picture_id, |
| current.picture_id); |
| EXPECT_LE(diff - 1, max_expected_picture_id_gap_); |
| if (diff > 2) { |
| // If the VideoSendStream is destroyed, any frames still in queue is lost. |
| // This can result in a two-frame gap, which will result in logs like |
| // "packet transmission failed, no matching RTP module found, or |
| // transmission error". |
| // A larger gap is only possible for first frame after a recreation, i.e. |
| // key frames. |
| EXPECT_EQ(VideoFrameType::kVideoFrameKey, current.frame_type); |
| } |
| } |
| |
| void VerifyTl0Idx(const ParsedPacket& current, const ParsedPacket& last) const |
| RTC_EXCLUSIVE_LOCKS_REQUIRED(&mutex_) { |
| if (current.tl0_pic_idx == kNoTl0PicIdx || |
| current.temporal_idx == kNoTemporalIdx) { |
| return; // No temporal layers. |
| } |
| |
| if (current.timestamp == last.timestamp || current.temporal_idx != 0) { |
| EXPECT_EQ(last.tl0_pic_idx, current.tl0_pic_idx); |
| return; |
| } |
| |
| // New frame with `temporal_idx` 0. |
| // `tl0_pic_idx` should be increasing. |
| EXPECT_TRUE(AheadOf<uint8_t>(current.tl0_pic_idx, last.tl0_pic_idx)); |
| |
| // Expect continuously increasing idx. |
| int diff = ForwardDiff<uint8_t>(last.tl0_pic_idx, current.tl0_pic_idx); |
| if (diff > 1) { |
| // If the VideoSendStream is destroyed, any frames still in queue is lost. |
| // Gaps only possible for first frame after a recreation, i.e. key frames. |
| EXPECT_EQ(VideoFrameType::kVideoFrameKey, current.frame_type); |
| EXPECT_LE(diff - 1, max_expected_tl0_idx_gap_); |
| } |
| } |
| |
| Action OnSendRtp(const uint8_t* packet, size_t length) override { |
| MutexLock lock(&mutex_); |
| |
| ParsedPacket parsed; |
| if (!ParsePayload(packet, length, &parsed)) |
| return SEND_PACKET; |
| |
| uint32_t ssrc = parsed.ssrc; |
| if (last_observed_packet_.find(ssrc) != last_observed_packet_.end()) { |
| // Compare to last packet. |
| VerifyPictureId(parsed, last_observed_packet_[ssrc]); |
| VerifyTl0Idx(parsed, last_observed_packet_[ssrc]); |
| } |
| |
| last_observed_packet_[ssrc] = parsed; |
| |
| // Pass the test when enough media packets have been received on all |
| // streams. |
| if (++num_packets_sent_[ssrc] >= kMinPacketsToObserve && |
| observed_ssrcs_.find(ssrc) == observed_ssrcs_.end()) { |
| observed_ssrcs_.insert(ssrc); |
| if (observed_ssrcs_.size() == num_ssrcs_to_observe_) { |
| observation_complete_.Set(); |
| } |
| } |
| return SEND_PACKET; |
| } |
| |
| Mutex mutex_; |
| const std::unique_ptr<VideoRtpDepacketizer> depacketizer_; |
| std::map<uint32_t, ParsedPacket> last_observed_packet_ RTC_GUARDED_BY(mutex_); |
| std::map<uint32_t, size_t> num_packets_sent_ RTC_GUARDED_BY(mutex_); |
| int max_expected_picture_id_gap_ RTC_GUARDED_BY(mutex_); |
| int max_expected_tl0_idx_gap_ RTC_GUARDED_BY(mutex_); |
| size_t num_ssrcs_to_observe_ RTC_GUARDED_BY(mutex_); |
| std::set<uint32_t> observed_ssrcs_ RTC_GUARDED_BY(mutex_); |
| }; |
| |
| class PictureIdTest : public test::CallTest, |
| public ::testing::WithParamInterface<size_t> { |
| public: |
| PictureIdTest() : num_temporal_layers_(GetParam()) {} |
| |
| virtual ~PictureIdTest() { |
| SendTask(task_queue(), [this]() { |
| send_transport_.reset(); |
| receive_transport_.reset(); |
| DestroyCalls(); |
| }); |
| } |
| |
| void SetupEncoder(VideoEncoderFactory* encoder_factory, |
| const std::string& payload_name); |
| void SetVideoEncoderConfig(int num_streams); |
| void TestPictureIdContinuousAfterReconfigure( |
| const std::vector<int>& ssrc_counts); |
| void TestPictureIdIncreaseAfterRecreateStreams( |
| const std::vector<int>& ssrc_counts); |
| |
| private: |
| const size_t num_temporal_layers_; |
| std::unique_ptr<PictureIdObserver> observer_; |
| }; |
| |
| // TODO(bugs.webrtc.org/13725): Enable on android when flakiness fixed. |
| #if defined(WEBRTC_ANDROID) |
| #define MAYBE_TemporalLayers DISABLED_TemporalLayers |
| #else |
| #define MAYBE_TemporalLayers TemporalLayers |
| #endif |
| |
| INSTANTIATE_TEST_SUITE_P(MAYBE_TemporalLayers, |
| PictureIdTest, |
| ::testing::ValuesIn(kNumTemporalLayers)); |
| |
| void PictureIdTest::SetupEncoder(VideoEncoderFactory* encoder_factory, |
| const std::string& payload_name) { |
| observer_.reset( |
| new PictureIdObserver(PayloadStringToCodecType(payload_name))); |
| |
| SendTask(task_queue(), [this, encoder_factory, payload_name]() { |
| CreateCalls(); |
| CreateSendTransport(BuiltInNetworkBehaviorConfig(), observer_.get()); |
| CreateSendConfig(test::VideoTestConstants::kNumSimulcastStreams, 0, 0, |
| send_transport_.get()); |
| GetVideoSendConfig()->encoder_settings.encoder_factory = encoder_factory; |
| GetVideoSendConfig()->rtp.payload_name = payload_name; |
| GetVideoEncoderConfig()->codec_type = |
| PayloadStringToCodecType(payload_name); |
| SetVideoEncoderConfig(/* number_of_streams */ 1); |
| }); |
| } |
| |
| void PictureIdTest::SetVideoEncoderConfig(int num_streams) { |
| GetVideoEncoderConfig()->number_of_streams = num_streams; |
| GetVideoEncoderConfig()->max_bitrate_bps = kEncoderBitrateBps; |
| |
| // Always divide the same total bitrate across all streams so that sending a |
| // single stream avoids lowering the bitrate estimate and requiring a |
| // subsequent rampup. |
| const int encoder_stream_bps = kEncoderBitrateBps / num_streams; |
| double scale_factor = 1.0; |
| for (int i = num_streams - 1; i >= 0; --i) { |
| VideoStream& stream = GetVideoEncoderConfig()->simulcast_layers[i]; |
| // Reduce the min bitrate by 10% to account for overhead that might |
| // otherwise cause streams to not be enabled. |
| stream.min_bitrate_bps = static_cast<int>(encoder_stream_bps * 0.9); |
| stream.target_bitrate_bps = encoder_stream_bps; |
| stream.max_bitrate_bps = encoder_stream_bps; |
| stream.num_temporal_layers = num_temporal_layers_; |
| stream.scale_resolution_down_by = scale_factor; |
| scale_factor *= 2.0; |
| } |
| } |
| |
| void PictureIdTest::TestPictureIdContinuousAfterReconfigure( |
| const std::vector<int>& ssrc_counts) { |
| SendTask(task_queue(), [this]() { |
| CreateVideoStreams(); |
| CreateFrameGeneratorCapturer(kFrameRate, kFrameMaxWidth, kFrameMaxHeight); |
| |
| // Initial test with a single stream. |
| Start(); |
| }); |
| |
| EXPECT_TRUE(observer_->Wait()) << "Timed out waiting for packets."; |
| |
| // Reconfigure VideoEncoder and test picture id increase. |
| // Expect continuously increasing picture id, equivalent to no gaps. |
| observer_->SetMaxExpectedPictureIdGap(0); |
| for (int ssrc_count : ssrc_counts) { |
| SetVideoEncoderConfig(ssrc_count); |
| observer_->SetExpectedSsrcs(ssrc_count); |
| observer_->ResetObservedSsrcs(); |
| // Make sure the picture_id sequence is continuous on reinit and recreate. |
| SendTask(task_queue(), [this]() { |
| GetVideoSendStream()->ReconfigureVideoEncoder( |
| GetVideoEncoderConfig()->Copy()); |
| }); |
| EXPECT_TRUE(observer_->Wait()) << "Timed out waiting for packets."; |
| } |
| |
| SendTask(task_queue(), [this]() { |
| Stop(); |
| DestroyStreams(); |
| }); |
| } |
| |
| void PictureIdTest::TestPictureIdIncreaseAfterRecreateStreams( |
| const std::vector<int>& ssrc_counts) { |
| SendTask(task_queue(), [this]() { |
| CreateVideoStreams(); |
| CreateFrameGeneratorCapturer(kFrameRate, kFrameMaxWidth, kFrameMaxHeight); |
| |
| // Initial test with a single stream. |
| Start(); |
| }); |
| |
| EXPECT_TRUE(observer_->Wait()) << "Timed out waiting for packets."; |
| |
| // Recreate VideoSendStream and test picture id increase. |
| // When the VideoSendStream is destroyed, any frames still in queue is lost |
| // with it, therefore it is expected that some frames might be lost. |
| observer_->SetMaxExpectedPictureIdGap(kMaxFramesLost); |
| for (int ssrc_count : ssrc_counts) { |
| SendTask(task_queue(), [this, &ssrc_count]() { |
| DestroyVideoSendStreams(); |
| |
| SetVideoEncoderConfig(ssrc_count); |
| observer_->SetExpectedSsrcs(ssrc_count); |
| observer_->ResetObservedSsrcs(); |
| |
| CreateVideoSendStreams(); |
| GetVideoSendStream()->Start(); |
| CreateFrameGeneratorCapturer(kFrameRate, kFrameMaxWidth, kFrameMaxHeight); |
| }); |
| |
| EXPECT_TRUE(observer_->Wait()) << "Timed out waiting for packets."; |
| } |
| |
| SendTask(task_queue(), [this]() { |
| Stop(); |
| DestroyStreams(); |
| }); |
| } |
| |
| TEST_P(PictureIdTest, ContinuousAfterReconfigureVp8) { |
| test::FunctionVideoEncoderFactory encoder_factory( |
| []() { return VP8Encoder::Create(); }); |
| SetupEncoder(&encoder_factory, "VP8"); |
| TestPictureIdContinuousAfterReconfigure({1, 3, 3, 1, 1}); |
| } |
| |
| // TODO(bugs.webrtc.org/14985): Investigate and reenable. |
| TEST_P(PictureIdTest, DISABLED_IncreasingAfterRecreateStreamVp8) { |
| test::FunctionVideoEncoderFactory encoder_factory( |
| []() { return VP8Encoder::Create(); }); |
| SetupEncoder(&encoder_factory, "VP8"); |
| TestPictureIdIncreaseAfterRecreateStreams({1, 3, 3, 1, 1}); |
| } |
| |
| TEST_P(PictureIdTest, ContinuousAfterStreamCountChangeVp8) { |
| test::FunctionVideoEncoderFactory encoder_factory( |
| []() { return VP8Encoder::Create(); }); |
| // Make sure that the picture id is not reset if the stream count goes |
| // down and then up. |
| SetupEncoder(&encoder_factory, "VP8"); |
| TestPictureIdContinuousAfterReconfigure({3, 1, 3}); |
| } |
| |
| TEST_P(PictureIdTest, ContinuousAfterReconfigureSimulcastEncoderAdapter) { |
| InternalEncoderFactory internal_encoder_factory; |
| test::FunctionVideoEncoderFactory encoder_factory( |
| [&internal_encoder_factory]() { |
| return std::make_unique<SimulcastEncoderAdapter>( |
| &internal_encoder_factory, SdpVideoFormat("VP8")); |
| }); |
| SetupEncoder(&encoder_factory, "VP8"); |
| TestPictureIdContinuousAfterReconfigure({1, 3, 3, 1, 1}); |
| } |
| |
| // TODO(bugs.webrtc.org/14985): Investigate and reenable. |
| TEST_P(PictureIdTest, |
| DISABLED_IncreasingAfterRecreateStreamSimulcastEncoderAdapter) { |
| InternalEncoderFactory internal_encoder_factory; |
| test::FunctionVideoEncoderFactory encoder_factory( |
| [&internal_encoder_factory]() { |
| return std::make_unique<SimulcastEncoderAdapter>( |
| &internal_encoder_factory, SdpVideoFormat("VP8")); |
| }); |
| SetupEncoder(&encoder_factory, "VP8"); |
| TestPictureIdIncreaseAfterRecreateStreams({1, 3, 3, 1, 1}); |
| } |
| |
| TEST_P(PictureIdTest, ContinuousAfterStreamCountChangeSimulcastEncoderAdapter) { |
| InternalEncoderFactory internal_encoder_factory; |
| test::FunctionVideoEncoderFactory encoder_factory( |
| [&internal_encoder_factory]() { |
| return std::make_unique<SimulcastEncoderAdapter>( |
| &internal_encoder_factory, SdpVideoFormat("VP8")); |
| }); |
| // Make sure that the picture id is not reset if the stream count goes |
| // down and then up. |
| SetupEncoder(&encoder_factory, "VP8"); |
| TestPictureIdContinuousAfterReconfigure({3, 1, 3}); |
| } |
| |
| // TODO(bugs.webrtc.org/14985): Investigate and reenable. |
| TEST_P(PictureIdTest, DISABLED_IncreasingAfterRecreateStreamVp9) { |
| test::FunctionVideoEncoderFactory encoder_factory( |
| []() { return VP9Encoder::Create(); }); |
| SetupEncoder(&encoder_factory, "VP9"); |
| TestPictureIdIncreaseAfterRecreateStreams({1, 1}); |
| } |
| |
| } // namespace webrtc |