From 863f860bfcf3bdc26eeecb299aa481d0f63d11ac Mon Sep 17 00:00:00 2001 From: Marco Date: Tue, 11 Apr 2017 16:17:18 -0700 Subject: [PATCH] vp9: Avoid encoder loopfilter for non-reference frames. Useful for SVC, where the top layer enhancement frames may not update any reference buffers, as is the case for the patterns in the 1 pass CBR SVC when #temporal_layers > 1. ~3% encoder speedup for SVC patterns with temporal layers in 1 pass CBR mode. Updated the SVC datarate tests for the mismatch frames. Set the frame-dropper off in some tests with #temporal_layers > 1 so we can correctly set #mismatch frames. Adjusted rate target threshold for tests where frame-dropper was turned off. Change-Id: Ia0c142f02100be0fed61cd2049691be9c59d6793 --- test/datarate_test.cc | 79 ++++++++++++++++++++++++--------------- vp9/encoder/vp9_encoder.c | 6 ++- 2 files changed, 54 insertions(+), 31 deletions(-) diff --git a/test/datarate_test.cc b/test/datarate_test.cc index d7f9148e1..ff88e9985 100644 --- a/test/datarate_test.cc +++ b/test/datarate_test.cc @@ -1221,9 +1221,10 @@ static void assign_layer_bitrates(vpx_codec_enc_cfg_t *const enc_cfg, } } -// Check basic rate targeting for 1 pass CBR SVC: 2 spatial layers and 1temporal -// layer, with screen content mode on and same speed setting for all layers. -TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2patialLayersScreenContent1) { +// Check basic rate targeting for 1 pass CBR SVC: 2 spatial layers and 1 +// temporal layer, with screen content mode on and same speed setting for all +// layers. +TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SL1TLScreenContent1) { cfg_.rc_buf_initial_sz = 500; cfg_.rc_buf_optimal_sz = 500; cfg_.rc_buf_sz = 1000; @@ -1241,7 +1242,7 @@ TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2patialLayersScreenContent1) { svc_params_.scaling_factor_den[0] = 288; svc_params_.scaling_factor_num[1] = 288; svc_params_.scaling_factor_den[1] = 288; - cfg_.rc_dropframe_thresh = 0; + cfg_.rc_dropframe_thresh = 10; cfg_.kf_max_dist = 9999; ::libvpx_test::Y4mVideoSource video("niklas_1280_720_30.y4m", 0, 300); cfg_.rc_target_bitrate = 500; @@ -1256,7 +1257,7 @@ TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2patialLayersScreenContent1) { // Check basic rate targeting for 1 pass CBR SVC: 2 spatial layers and // 3 temporal layers. Run CIF clip with 1 thread. -TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SpatialLayers) { +TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SL3TL) { cfg_.rc_buf_initial_sz = 500; cfg_.rc_buf_optimal_sz = 500; cfg_.rc_buf_sz = 1000; @@ -1276,7 +1277,7 @@ TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SpatialLayers) { svc_params_.scaling_factor_den[0] = 288; svc_params_.scaling_factor_num[1] = 288; svc_params_.scaling_factor_den[1] = 288; - cfg_.rc_dropframe_thresh = 10; + cfg_.rc_dropframe_thresh = 0; cfg_.kf_max_dist = 9999; ::libvpx_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288, 30, 1, 0, 200); @@ -1288,17 +1289,21 @@ TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SpatialLayers) { assign_layer_bitrates(&cfg_, &svc_params_, cfg_.ss_number_layers, cfg_.ts_number_layers, cfg_.temporal_layering_mode); ASSERT_NO_FATAL_FAILURE(RunLoop(&video)); - ASSERT_GE(cfg_.rc_target_bitrate, file_datarate_ * 0.85) + ASSERT_GE(cfg_.rc_target_bitrate, file_datarate_ * 0.82) << " The datarate for the file exceeds the target by too much!"; ASSERT_LE(cfg_.rc_target_bitrate, file_datarate_ * 1.15) << " The datarate for the file is lower than the target by too much!"; - EXPECT_EQ(static_cast(0), GetMismatchFrames()); + // Number of temporal layers > 1, so half of the frames in this SVC pattern + // will be non-reference frame and hence encoder will avoid loopfilter. + // Since frame dropper is off, we can expcet 100 (half of the sequence) + // mismatched frames. + EXPECT_EQ(static_cast(100), GetMismatchFrames()); } } // Check basic rate targeting for 1 pass CBR SVC with denoising. // 2 spatial layers and 3 temporal layer. Run HD clip with 2 threads. -TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SpatialLayersDenoiserOn) { +TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SL3TLDenoiserOn) { cfg_.rc_buf_initial_sz = 500; cfg_.rc_buf_optimal_sz = 500; cfg_.rc_buf_sz = 1000; @@ -1318,7 +1323,7 @@ TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SpatialLayersDenoiserOn) { svc_params_.scaling_factor_den[0] = 288; svc_params_.scaling_factor_num[1] = 288; svc_params_.scaling_factor_den[1] = 288; - cfg_.rc_dropframe_thresh = 10; + cfg_.rc_dropframe_thresh = 0; cfg_.kf_max_dist = 9999; ::libvpx_test::Y4mVideoSource video("niklas_1280_720_30.y4m", 0, 300); // TODO(marpan): Check that effective_datarate for each layer hits the @@ -1330,17 +1335,21 @@ TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SpatialLayersDenoiserOn) { assign_layer_bitrates(&cfg_, &svc_params_, cfg_.ss_number_layers, cfg_.ts_number_layers, cfg_.temporal_layering_mode); ASSERT_NO_FATAL_FAILURE(RunLoop(&video)); - ASSERT_GE(cfg_.rc_target_bitrate, file_datarate_ * 0.85) + ASSERT_GE(cfg_.rc_target_bitrate, file_datarate_ * 0.82) << " The datarate for the file exceeds the target by too much!"; ASSERT_LE(cfg_.rc_target_bitrate, file_datarate_ * 1.15) << " The datarate for the file is lower than the target by too much!"; - EXPECT_EQ(static_cast(0), GetMismatchFrames()); + // Number of temporal layers > 1, so half of the frames in this SVC pattern + // will be non-reference frame and hence encoder will avoid loopfilter. + // Since frame dropper is off, we can expcet 150 (half of the sequence) + // mismatched frames. + EXPECT_EQ(static_cast(150), GetMismatchFrames()); } } // Check basic rate targeting for 1 pass CBR SVC: 2 spatial layers and 3 // temporal layers. Run CIF clip with 1 thread, and few short key frame periods. -TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SpatialLayersSmallKf) { +TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SL3TLSmallKf) { cfg_.rc_buf_initial_sz = 500; cfg_.rc_buf_optimal_sz = 500; cfg_.rc_buf_sz = 1000; @@ -1376,13 +1385,12 @@ TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SpatialLayersSmallKf) { << " The datarate for the file exceeds the target by too much!"; ASSERT_LE(cfg_.rc_target_bitrate, file_datarate_ * 1.15) << " The datarate for the file is lower than the target by too much!"; - EXPECT_EQ(static_cast(0), GetMismatchFrames()); } } // Check basic rate targeting for 1 pass CBR SVC: 2 spatial layers and // 3 temporal layers. Run HD clip with 4 threads. -TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SpatialLayers4threads) { +TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SL3TL4threads) { cfg_.rc_buf_initial_sz = 500; cfg_.rc_buf_optimal_sz = 500; cfg_.rc_buf_sz = 1000; @@ -1402,7 +1410,7 @@ TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SpatialLayers4threads) { svc_params_.scaling_factor_den[0] = 288; svc_params_.scaling_factor_num[1] = 288; svc_params_.scaling_factor_den[1] = 288; - cfg_.rc_dropframe_thresh = 10; + cfg_.rc_dropframe_thresh = 0; cfg_.kf_max_dist = 9999; ::libvpx_test::Y4mVideoSource video("niklas_1280_720_30.y4m", 0, 300); cfg_.rc_target_bitrate = 800; @@ -1410,16 +1418,20 @@ TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SpatialLayers4threads) { assign_layer_bitrates(&cfg_, &svc_params_, cfg_.ss_number_layers, cfg_.ts_number_layers, cfg_.temporal_layering_mode); ASSERT_NO_FATAL_FAILURE(RunLoop(&video)); - ASSERT_GE(cfg_.rc_target_bitrate, file_datarate_ * 0.85) + ASSERT_GE(cfg_.rc_target_bitrate, file_datarate_ * 0.82) << " The datarate for the file exceeds the target by too much!"; ASSERT_LE(cfg_.rc_target_bitrate, file_datarate_ * 1.15) << " The datarate for the file is lower than the target by too much!"; - EXPECT_EQ(static_cast(0), GetMismatchFrames()); + // Number of temporal layers > 1, so half of the frames in this SVC pattern + // will be non-reference frame and hence encoder will avoid loopfilter. + // Since frame dropper is off, we can expcet 150 (half of the sequence) + // mismatched frames. + EXPECT_EQ(static_cast(150), GetMismatchFrames()); } // Check basic rate targeting for 1 pass CBR SVC: 3 spatial layers and // 3 temporal layers. Run CIF clip with 1 thread. -TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc3SpatialLayers) { +TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc3SL3TL) { cfg_.rc_buf_initial_sz = 500; cfg_.rc_buf_optimal_sz = 500; cfg_.rc_buf_sz = 1000; @@ -1441,7 +1453,7 @@ TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc3SpatialLayers) { svc_params_.scaling_factor_den[1] = 288; svc_params_.scaling_factor_num[2] = 288; svc_params_.scaling_factor_den[2] = 288; - cfg_.rc_dropframe_thresh = 10; + cfg_.rc_dropframe_thresh = 0; cfg_.kf_max_dist = 9999; ::libvpx_test::Y4mVideoSource video("niklas_1280_720_30.y4m", 0, 300); cfg_.rc_target_bitrate = 800; @@ -1449,16 +1461,20 @@ TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc3SpatialLayers) { assign_layer_bitrates(&cfg_, &svc_params_, cfg_.ss_number_layers, cfg_.ts_number_layers, cfg_.temporal_layering_mode); ASSERT_NO_FATAL_FAILURE(RunLoop(&video)); - ASSERT_GE(cfg_.rc_target_bitrate, file_datarate_ * 0.85) + ASSERT_GE(cfg_.rc_target_bitrate, file_datarate_ * 0.82) << " The datarate for the file exceeds the target by too much!"; ASSERT_LE(cfg_.rc_target_bitrate, file_datarate_ * 1.22) << " The datarate for the file is lower than the target by too much!"; - EXPECT_EQ(static_cast(0), GetMismatchFrames()); + // Number of temporal layers > 1, so half of the frames in this SVC pattern + // will be non-reference frame and hence encoder will avoid loopfilter. + // Since frame dropper is off, we can expcet 150 (half of the sequence) + // mismatched frames. + EXPECT_EQ(static_cast(150), GetMismatchFrames()); } // Check basic rate targeting for 1 pass CBR SVC: 3 spatial layers and 3 // temporal layers. Run CIF clip with 1 thread, and few short key frame periods. -TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc3SpatialLayersSmallKf) { +TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc3SL3TLSmallKf) { cfg_.rc_buf_initial_sz = 500; cfg_.rc_buf_optimal_sz = 500; cfg_.rc_buf_sz = 1000; @@ -1495,13 +1511,12 @@ TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc3SpatialLayersSmallKf) { << " The datarate for the file exceeds the target by too much!"; ASSERT_LE(cfg_.rc_target_bitrate, file_datarate_ * 1.30) << " The datarate for the file is lower than the target by too much!"; - EXPECT_EQ(static_cast(0), GetMismatchFrames()); } } // Check basic rate targeting for 1 pass CBR SVC: 3 spatial layers and // 3 temporal layers. Run HD clip with 4 threads. -TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc3SpatialLayers4threads) { +TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc3SL3TL4threads) { cfg_.rc_buf_initial_sz = 500; cfg_.rc_buf_optimal_sz = 500; cfg_.rc_buf_sz = 1000; @@ -1523,7 +1538,7 @@ TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc3SpatialLayers4threads) { svc_params_.scaling_factor_den[1] = 288; svc_params_.scaling_factor_num[2] = 288; svc_params_.scaling_factor_den[2] = 288; - cfg_.rc_dropframe_thresh = 10; + cfg_.rc_dropframe_thresh = 0; cfg_.kf_max_dist = 9999; ::libvpx_test::Y4mVideoSource video("niklas_1280_720_30.y4m", 0, 300); cfg_.rc_target_bitrate = 800; @@ -1531,16 +1546,20 @@ TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc3SpatialLayers4threads) { assign_layer_bitrates(&cfg_, &svc_params_, cfg_.ss_number_layers, cfg_.ts_number_layers, cfg_.temporal_layering_mode); ASSERT_NO_FATAL_FAILURE(RunLoop(&video)); - ASSERT_GE(cfg_.rc_target_bitrate, file_datarate_ * 0.85) + ASSERT_GE(cfg_.rc_target_bitrate, file_datarate_ * 0.82) << " The datarate for the file exceeds the target by too much!"; ASSERT_LE(cfg_.rc_target_bitrate, file_datarate_ * 1.22) << " The datarate for the file is lower than the target by too much!"; - EXPECT_EQ(static_cast(0), GetMismatchFrames()); + // Number of temporal layers > 1, so half of the frames in this SVC pattern + // will be non-reference frame and hence encoder will avoid loopfilter. + // Since frame dropper is off, we can expcet 150 (half of the sequence) + // mismatched frames. + EXPECT_EQ(static_cast(150), GetMismatchFrames()); } // Run SVC encoder for 1 temporal layer, 2 spatial layers, with spatial // downscale 5x5. -TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SpatialLayers5x5MultipleRuns) { +TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SL1TL5x5MultipleRuns) { cfg_.rc_buf_initial_sz = 500; cfg_.rc_buf_optimal_sz = 500; cfg_.rc_buf_sz = 1000; @@ -1558,7 +1577,7 @@ TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SpatialLayers5x5MultipleRuns) { svc_params_.scaling_factor_den[0] = 1280; svc_params_.scaling_factor_num[1] = 1280; svc_params_.scaling_factor_den[1] = 1280; - cfg_.rc_dropframe_thresh = 0; + cfg_.rc_dropframe_thresh = 10; cfg_.kf_max_dist = 999999; cfg_.kf_min_dist = 0; cfg_.ss_target_bitrate[0] = 300; diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c index 1dc70d2d3..7d56232e1 100644 --- a/vp9/encoder/vp9_encoder.c +++ b/vp9/encoder/vp9_encoder.c @@ -2618,6 +2618,10 @@ static void loopfilter_frame(VP9_COMP *cpi, VP9_COMMON *cm) { MACROBLOCKD *xd = &cpi->td.mb.e_mbd; struct loopfilter *lf = &cm->lf; + const int is_reference_frame = + (cpi->refresh_last_frame || cpi->refresh_golden_frame || + cpi->refresh_alt_ref_frame); + if (xd->lossless) { lf->filter_level = 0; lf->last_filt_level = 0; @@ -2643,7 +2647,7 @@ static void loopfilter_frame(VP9_COMP *cpi, VP9_COMMON *cm) { cpi->time_pick_lpf += vpx_usec_timer_elapsed(&timer); } - if (lf->filter_level > 0) { + if (lf->filter_level > 0 && is_reference_frame) { vp9_build_mask_frame(cm, lf->filter_level, 0); if (cpi->num_workers > 1) -- 2.40.0