generateKeyframe: pass frame_types in bypass mode

Passes frame_types to the underlying encoder in bypass mode.
For libvpx this has no effect, for H264 this changes the behavior
to allow generating keyframes on a per-layer basis.

BUG=chromium:1354101

Change-Id: I26fc22d9e2ec4681a57ce591e9eafd0b1ec962b0
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/285083
Reviewed-by: Ilya Nikolaevskiy <ilnik@webrtc.org>
Reviewed-by: Harald Alvestrand <hta@webrtc.org>
Commit-Queue: Philipp Hancke <phancke@microsoft.com>
Cr-Commit-Position: refs/heads/main@{#38821}

api/test/simulcast_test_fixture.h

@@ -19,6 +19,7 @@ class SimulcastTestFixture {
   virtual ~SimulcastTestFixture() = default;
 
   virtual void TestKeyFrameRequestsOnAllStreams() = 0;
+  virtual void TestKeyFrameRequestsOnSpecificStreams() = 0;
   virtual void TestPaddingAllStreams() = 0;
   virtual void TestPaddingTwoStreams() = 0;
   virtual void TestPaddingTwoStreamsOneMaxedOut() = 0;

media/engine/simulcast_encoder_adapter.cc

@@ -466,7 +466,7 @@ int SimulcastEncoderAdapter::Encode(
     if (layer.is_keyframe_needed()) {
       // This is legacy behavior, generating a keyframe on all layers
       // when generating one for a layer that became active for the first time
-      // or after being disabled
+      // or after being disabled.
       is_keyframe_needed = true;
       break;
     }
@@ -491,8 +491,11 @@ int SimulcastEncoderAdapter::Encode(
     // frame types for all streams should be passed to the encoder unchanged.
     // Otherwise a single per-encoder frame type is passed.
     std::vector<VideoFrameType> stream_frame_types(
-        bypass_mode_ ? total_streams_count_ : 1,
+        bypass_mode_
+            ? std::max<unsigned char>(codec_.numberOfSimulcastStreams, 1)
+            : 1,
         VideoFrameType::kVideoFrameDelta);
+
     bool keyframe_requested = false;
     if (is_keyframe_needed) {
       std::fill(stream_frame_types.begin(), stream_frame_types.end(),
@@ -500,16 +503,13 @@ int SimulcastEncoderAdapter::Encode(
       keyframe_requested = true;
     } else if (frame_types) {
       if (bypass_mode_) {
-        // In bypass mode, requesting a key frame on any layer triggers a
-        // key frame request on all layers.
-        for (const auto& frame_type : *frame_types) {
-          if (frame_type == VideoFrameType::kVideoFrameKey) {
-            std::fill(stream_frame_types.begin(), stream_frame_types.end(),
-                      VideoFrameType::kVideoFrameKey);
-            keyframe_requested = true;
-            break;
-          }
-        }
+        // In bypass mode, we effectively pass on frame_types.
+        RTC_DCHECK_EQ(frame_types->size(), stream_frame_types.size());
+        stream_frame_types = *frame_types;
+        keyframe_requested =
+            absl::c_any_of(*frame_types, [](const VideoFrameType frame_type) {
+              return frame_type == VideoFrameType::kVideoFrameKey;
+            });
       } else {
         size_t stream_idx = static_cast<size_t>(layer.stream_idx());
         if (frame_types->size() >= stream_idx &&

modules/video_coding/codecs/h264/h264_encoder_impl.cc

@@ -416,26 +416,17 @@ int32_t H264EncoderImpl::Encode(
   RTC_CHECK(frame_buffer->type() == VideoFrameBuffer::Type::kI420 ||
             frame_buffer->type() == VideoFrameBuffer::Type::kI420A);
 
-  bool send_key_frame = false;
+  bool is_keyframe_needed = false;
   for (size_t i = 0; i < configurations_.size(); ++i) {
     if (configurations_[i].key_frame_request && configurations_[i].sending) {
-      send_key_frame = true;
+      // This is legacy behavior, generating a keyframe on all layers
+      // when generating one for a layer that became active for the first time
+      // or after being disabled.
+      is_keyframe_needed = true;
       break;
     }
   }
 
-  if (!send_key_frame && frame_types) {
-    for (size_t i = 0; i < configurations_.size(); ++i) {
-      const size_t simulcast_idx =
-          static_cast<size_t>(configurations_[i].simulcast_idx);
-      if (configurations_[i].sending && simulcast_idx < frame_types->size() &&
-          (*frame_types)[simulcast_idx] == VideoFrameType::kVideoFrameKey) {
-        send_key_frame = true;
-        break;
-      }
-    }
-  }
-
   RTC_DCHECK_EQ(configurations_[0].width, frame_buffer->width());
   RTC_DCHECK_EQ(configurations_[0].height, frame_buffer->height());
 
@@ -480,12 +471,20 @@ int32_t H264EncoderImpl::Encode(
     if (!configurations_[i].sending) {
       continue;
     }
-    if (frame_types != nullptr) {
+    if (frame_types != nullptr && i < frame_types->size()) {
       // Skip frame?
       if ((*frame_types)[i] == VideoFrameType::kEmptyFrame) {
         continue;
       }
     }
+    // Send a key frame either when this layer is configured to require one
+    // or we have explicitly been asked to.
+    const size_t simulcast_idx =
+        static_cast<size_t>(configurations_[i].simulcast_idx);
+    bool send_key_frame =
+        is_keyframe_needed ||
+        (frame_types && simulcast_idx < frame_types->size() &&
+         (*frame_types)[simulcast_idx] == VideoFrameType::kVideoFrameKey);
     if (send_key_frame) {
       // API doc says ForceIntraFrame(false) does nothing, but calling this
       // function forces a key frame regardless of the `bIDR` argument's value.

modules/video_coding/codecs/h264/h264_simulcast_unittest.cc

@@ -35,8 +35,12 @@ std::unique_ptr<SimulcastTestFixture> CreateSpecificSimulcastTestFixture() {
 }  // namespace
 
 TEST(TestH264Simulcast, TestKeyFrameRequestsOnAllStreams) {
+  GTEST_SKIP() << "Not applicable to H264.";
+}
+
+TEST(TestH264Simulcast, TestKeyFrameRequestsOnSpecificStreams) {
   auto fixture = CreateSpecificSimulcastTestFixture();
-  fixture->TestKeyFrameRequestsOnAllStreams();
+  fixture->TestKeyFrameRequestsOnSpecificStreams();
 }
 
 TEST(TestH264Simulcast, TestPaddingAllStreams) {

modules/video_coding/codecs/test/videoprocessor.cc

@@ -307,8 +307,10 @@ void VideoProcessor::ProcessFrame() {
   // Encode.
   const std::vector<VideoFrameType> frame_types =
       (frame_number == 0)
-          ? std::vector<VideoFrameType>{VideoFrameType::kVideoFrameKey}
-          : std::vector<VideoFrameType>{VideoFrameType::kVideoFrameDelta};
+          ? std::vector<VideoFrameType>(num_simulcast_or_spatial_layers_,
+                                        VideoFrameType::kVideoFrameKey)
+          : std::vector<VideoFrameType>(num_simulcast_or_spatial_layers_,
+                                        VideoFrameType::kVideoFrameDelta);
   const int encode_return_code = encoder_->Encode(input_frame, &frame_types);
   for (size_t i = 0; i < num_simulcast_or_spatial_layers_; ++i) {
     FrameStatistics* frame_stat = stats_->GetFrame(frame_number, i);

modules/video_coding/codecs/vp8/libvpx_vp8_simulcast_test.cc

@@ -39,6 +39,10 @@ TEST(LibvpxVp8SimulcastTest, TestKeyFrameRequestsOnAllStreams) {
   fixture->TestKeyFrameRequestsOnAllStreams();
 }
 
+TEST(LibvpxVp8SimulcastTest, TestKeyFrameRequestsOnSpecificStreams) {
+  GTEST_SKIP() << "Not applicable to VP8.";
+}
+
 TEST(LibvpxVp8SimulcastTest, TestPaddingAllStreams) {
   auto fixture = CreateSpecificSimulcastTestFixture();
   fixture->TestPaddingAllStreams();

modules/video_coding/utility/simulcast_test_fixture_impl.cc

@@ -44,6 +44,7 @@ const int kMaxBitrates[kNumberOfSimulcastStreams] = {150, 600, 1200};
 const int kMinBitrates[kNumberOfSimulcastStreams] = {50, 150, 600};
 const int kTargetBitrates[kNumberOfSimulcastStreams] = {100, 450, 1000};
 const float kMaxFramerates[kNumberOfSimulcastStreams] = {30, 30, 30};
+const int kScaleResolutionDownBy[kNumberOfSimulcastStreams] = {4, 2, 1};
 const int kDefaultTemporalLayerProfile[3] = {3, 3, 3};
 const int kNoTemporalLayerProfile[3] = {0, 0, 0};
 
@@ -353,14 +354,10 @@ void SimulcastTestFixtureImpl::ExpectStreams(
     const std::vector<bool> expected_streams_active) {
   ASSERT_EQ(static_cast<int>(expected_streams_active.size()),
             kNumberOfSimulcastStreams);
-  if (expected_streams_active[0]) {
-    ExpectStream(frame_type, 4);
+  for (size_t i = 0; i < kNumberOfSimulcastStreams; i++) {
+    if (expected_streams_active[i]) {
+      ExpectStream(frame_type, kScaleResolutionDownBy[i]);
     }
-  if (expected_streams_active[1]) {
-    ExpectStream(frame_type, 2);
-  }
-  if (expected_streams_active[2]) {
-    ExpectStream(frame_type, 1);
   }
 }
 
@@ -389,8 +386,8 @@ void SimulcastTestFixtureImpl::VerifyTemporalIdxAndSyncForAllSpatialLayers(
   }
 }
 
-// We currently expect all active streams to generate a key frame even though
-// a key frame was only requested for some of them.
+// For some codecs (VP8) expect all active streams to generate a key frame even
+// though a key frame was only requested for some of them.
 void SimulcastTestFixtureImpl::TestKeyFrameRequestsOnAllStreams() {
   SetRates(kMaxBitrates[2], 30);  // To get all three streams.
   std::vector<VideoFrameType> frame_types(kNumberOfSimulcastStreams,
@@ -428,6 +425,69 @@ void SimulcastTestFixtureImpl::TestKeyFrameRequestsOnAllStreams() {
   EXPECT_EQ(0, encoder_->Encode(*input_frame_, &frame_types));
 }
 
+// For some codecs (H264) expect only particular active streams to generate a
+// key frame when a key frame was only requested for some of them.
+void SimulcastTestFixtureImpl::TestKeyFrameRequestsOnSpecificStreams() {
+  SetRates(kMaxBitrates[2], 30);  // To get all three streams.
+  std::vector<VideoFrameType> frame_types(kNumberOfSimulcastStreams,
+                                          VideoFrameType::kVideoFrameDelta);
+  ExpectStreams(VideoFrameType::kVideoFrameKey, kNumberOfSimulcastStreams);
+  EXPECT_EQ(0, encoder_->Encode(*input_frame_, &frame_types));
+
+  ExpectStreams(VideoFrameType::kVideoFrameDelta, kNumberOfSimulcastStreams);
+  input_frame_->set_timestamp(input_frame_->timestamp() + 3000);
+  EXPECT_EQ(0, encoder_->Encode(*input_frame_, &frame_types));
+
+  frame_types[0] = VideoFrameType::kVideoFrameKey;
+  ExpectStream(VideoFrameType::kVideoFrameKey, kScaleResolutionDownBy[0]);
+  ExpectStream(VideoFrameType::kVideoFrameDelta, kScaleResolutionDownBy[1]);
+  ExpectStream(VideoFrameType::kVideoFrameDelta, kScaleResolutionDownBy[2]);
+  input_frame_->set_timestamp(input_frame_->timestamp() + 3000);
+  EXPECT_EQ(0, encoder_->Encode(*input_frame_, &frame_types));
+
+  std::fill(frame_types.begin(), frame_types.end(),
+            VideoFrameType::kVideoFrameDelta);
+  frame_types[1] = VideoFrameType::kVideoFrameKey;
+  ExpectStream(VideoFrameType::kVideoFrameDelta, kScaleResolutionDownBy[0]);
+  ExpectStream(VideoFrameType::kVideoFrameKey, kScaleResolutionDownBy[1]);
+  ExpectStream(VideoFrameType::kVideoFrameDelta, kScaleResolutionDownBy[2]);
+  input_frame_->set_timestamp(input_frame_->timestamp() + 3000);
+  EXPECT_EQ(0, encoder_->Encode(*input_frame_, &frame_types));
+
+  std::fill(frame_types.begin(), frame_types.end(),
+            VideoFrameType::kVideoFrameDelta);
+  frame_types[2] = VideoFrameType::kVideoFrameKey;
+  ExpectStream(VideoFrameType::kVideoFrameDelta, kScaleResolutionDownBy[0]);
+  ExpectStream(VideoFrameType::kVideoFrameDelta, kScaleResolutionDownBy[1]);
+  ExpectStream(VideoFrameType::kVideoFrameKey, kScaleResolutionDownBy[2]);
+  input_frame_->set_timestamp(input_frame_->timestamp() + 3000);
+  EXPECT_EQ(0, encoder_->Encode(*input_frame_, &frame_types));
+
+  std::fill(frame_types.begin(), frame_types.end(),
+            VideoFrameType::kVideoFrameDelta);
+  frame_types[0] = VideoFrameType::kVideoFrameKey;
+  frame_types[2] = VideoFrameType::kVideoFrameKey;
+  ExpectStream(VideoFrameType::kVideoFrameKey, kScaleResolutionDownBy[0]);
+  ExpectStream(VideoFrameType::kVideoFrameDelta, kScaleResolutionDownBy[1]);
+  ExpectStream(VideoFrameType::kVideoFrameKey, kScaleResolutionDownBy[2]);
+  input_frame_->set_timestamp(input_frame_->timestamp() + 3000);
+  EXPECT_EQ(0, encoder_->Encode(*input_frame_, &frame_types));
+
+  std::fill(frame_types.begin(), frame_types.end(),
+            VideoFrameType::kVideoFrameKey);
+  ExpectStream(VideoFrameType::kVideoFrameKey, kScaleResolutionDownBy[0]);
+  ExpectStream(VideoFrameType::kVideoFrameKey, kScaleResolutionDownBy[1]);
+  ExpectStream(VideoFrameType::kVideoFrameKey, kScaleResolutionDownBy[2]);
+  input_frame_->set_timestamp(input_frame_->timestamp() + 3000);
+  EXPECT_EQ(0, encoder_->Encode(*input_frame_, &frame_types));
+
+  std::fill(frame_types.begin(), frame_types.end(),
+            VideoFrameType::kVideoFrameDelta);
+  ExpectStreams(VideoFrameType::kVideoFrameDelta, kNumberOfSimulcastStreams);
+  input_frame_->set_timestamp(input_frame_->timestamp() + 3000);
+  EXPECT_EQ(0, encoder_->Encode(*input_frame_, &frame_types));
+}
+
 void SimulcastTestFixtureImpl::TestPaddingAllStreams() {
   // We should always encode the base layer.
   SetRates(kMinBitrates[0] - 1, 30);

modules/video_coding/utility/simulcast_test_fixture_impl.h

@@ -35,6 +35,7 @@ class SimulcastTestFixtureImpl final : public SimulcastTestFixture {
 
   // Implements SimulcastTestFixture.
   void TestKeyFrameRequestsOnAllStreams() override;
+  void TestKeyFrameRequestsOnSpecificStreams() override;
   void TestPaddingAllStreams() override;
   void TestPaddingTwoStreams() override;
   void TestPaddingTwoStreamsOneMaxedOut() override;