Skip to content

GitLab

A
aom-rav1e
Commit be3b08da authored by Minghai Shang's avatar Minghai Shang
Browse files
Options

[svc] Temporal svc with two pass rate control 

It's built based on current spatial svc code.
We only support one spatial two temporal layers at this time.
Change-Id: I1fdc8584354b910331e626bfae60473b3b701ba1
parent 6b649a0d
Hide whitespace changes
Inline Side-by-side
Showing with 485 additions and 206 deletions
examples/vp9_spatial_svc_encoder.c
View file @ be3b08da
......@@ -38,8 +38,10 @@ static const arg_def_t timebase_arg =
ARG_DEF("t", "timebase", 1, "timebase (num/den)");
static const arg_def_t bitrate_arg = ARG_DEF(
"b", "target-bitrate", 1, "encoding bitrate, in kilobits per second");
static const arg_def_t layers_arg =
ARG_DEF("l", "layers", 1, "number of SVC layers");
static const arg_def_t spatial_layers_arg =
ARG_DEF("sl", "spatial-layers", 1, "number of spatial SVC layers");
static const arg_def_t temporal_layers_arg =
ARG_DEF("tl", "temporal-layers", 1, "number of temporal SVC layers");
static const arg_def_t kf_dist_arg =
ARG_DEF("k", "kf-dist", 1, "number of frames between keyframes");
static const arg_def_t scale_factors_arg =
......@@ -65,10 +67,11 @@ static const arg_def_t max_bitrate_arg =
static const arg_def_t *svc_args[] = {
&frames_arg, &width_arg, &height_arg,
&timebase_arg, &bitrate_arg, &skip_frames_arg, &layers_arg,
&timebase_arg, &bitrate_arg, &skip_frames_arg, &spatial_layers_arg,
&kf_dist_arg, &scale_factors_arg, &quantizers_arg, &passes_arg,
&pass_arg, &fpf_name_arg, &min_q_arg, &max_q_arg,
&min_bitrate_arg, &max_bitrate_arg, NULL
&min_bitrate_arg, &max_bitrate_arg, &temporal_layers_arg,
NULL
};
static const uint32_t default_frames_to_skip = 0;
......@@ -79,6 +82,7 @@ static const uint32_t default_timebase_num = 1;
static const uint32_t default_timebase_den = 60;
static const uint32_t default_bitrate = 1000;
static const uint32_t default_spatial_layers = 5;
static const uint32_t default_temporal_layers = 1;
static const uint32_t default_kf_dist = 100;
typedef struct {
......@@ -119,6 +123,7 @@ static void parse_command_line(int argc, const char **argv_,
// initialize SvcContext with parameters that will be passed to vpx_svc_init
svc_ctx->log_level = SVC_LOG_DEBUG;
svc_ctx->spatial_layers = default_spatial_layers;
svc_ctx->temporal_layers = default_temporal_layers;
// start with default encoder configuration
res = vpx_codec_enc_config_default(vpx_codec_vp9_cx(), enc_cfg, 0);
......@@ -156,8 +161,10 @@ static void parse_command_line(int argc, const char **argv_,
enc_cfg->rc_target_bitrate = arg_parse_uint(&arg);
} else if (arg_match(&arg, &skip_frames_arg, argi)) {
app_input->frames_to_skip = arg_parse_uint(&arg);
} else if (arg_match(&arg, &layers_arg, argi)) {
} else if (arg_match(&arg, &spatial_layers_arg, argi)) {
svc_ctx->spatial_layers = arg_parse_uint(&arg);
} else if (arg_match(&arg, &temporal_layers_arg, argi)) {
svc_ctx->temporal_layers = arg_parse_uint(&arg);
} else if (arg_match(&arg, &kf_dist_arg, argi)) {
enc_cfg->kf_min_dist = arg_parse_uint(&arg);
enc_cfg->kf_max_dist = enc_cfg->kf_min_dist;
......
test/svc_test.cc
View file @ be3b08da
......@@ -212,13 +212,13 @@ class SvcTest : public ::testing::Test {
EXPECT_EQ(received_frames, n);
}
void DropEnhancementLayers(struct vpx_fixed_buf *const inputs,
const int num_super_frames,
const int remained_layers,
const bool is_multiple_frame_context) {
void DropLayersAndMakeItVP9Comaptible(struct vpx_fixed_buf *const inputs,
const int num_super_frames,
const int remained_spatial_layers,
const bool is_multiple_frame_contexts) {
ASSERT_TRUE(inputs != NULL);
ASSERT_GT(num_super_frames, 0);
ASSERT_GT(remained_layers, 0);
ASSERT_GT(remained_spatial_layers, 0);
for (int i = 0; i < num_super_frames; ++i) {
uint32_t frame_sizes[8] = {0};
......@@ -234,64 +234,110 @@ class SvcTest : public ::testing::Test {
NULL, NULL);
ASSERT_EQ(VPX_CODEC_OK, res);
uint8_t *frame_data = static_cast<uint8_t *>(inputs[i].buf);
uint8_t *frame_start = frame_data;
for (frame = 0; frame < frame_count; ++frame) {
// Looking for a visible frame.
if (frame_data[0] & 0x02) {
++frames_found;
if (frames_found == remained_layers)
break;
if (frame_count == 0) {
// There's no super frame but only a single frame.
ASSERT_EQ(1, remained_spatial_layers);
if (is_multiple_frame_contexts) {
// Make a new super frame.
uint8_t marker = 0xc1;
unsigned int mask;
int mag;
// Choose the magnitude.
for (mag = 0, mask = 0xff; mag < 4; ++mag) {
if (inputs[i].sz < mask)
break;
mask <<= 8;
mask |= 0xff;
}
marker |= mag << 3;
int index_sz = 2 + (mag + 1) * 2;
inputs[i].buf = realloc(inputs[i].buf, inputs[i].sz + index_sz + 16);
ASSERT_TRUE(inputs[i].buf != NULL);
uint8_t *frame_data = static_cast<uint8_t*>(inputs[i].buf);
frame_data[0] &= ~2; // Set the show_frame flag to 0.
frame_data += inputs[i].sz;
// Add an one byte frame with show_existing_frame.
*frame_data++ = 0x88;
// Write the super frame index.
*frame_data++ = marker;
frame_sizes[0] = inputs[i].sz;
frame_sizes[1] = 1;
for (int j = 0; j < 2; ++j) {
unsigned int this_sz = frame_sizes[j];
for (int k = 0; k <= mag; k++) {
*frame_data++ = this_sz & 0xff;
this_sz >>= 8;
}
}
*frame_data++ = marker;
inputs[i].sz += index_sz + 1;
}
} else {
// Found a super frame.
uint8_t *frame_data = static_cast<uint8_t*>(inputs[i].buf);
uint8_t *frame_start = frame_data;
for (frame = 0; frame < frame_count; ++frame) {
// Looking for a visible frame.
if (frame_data[0] & 0x02) {
++frames_found;
if (frames_found == remained_spatial_layers)
break;
}
frame_data += frame_sizes[frame];
}
ASSERT_LT(frame, frame_count) << "Couldn't find a visible frame. "
<< "remained_spatial_layers: " << remained_spatial_layers
<< " super_frame: " << i
<< " is_multiple_frame_context: " << is_multiple_frame_contexts;
if (frame == frame_count - 1 && !is_multiple_frame_contexts)
continue;
frame_data += frame_sizes[frame];
}
ASSERT_LT(frame, frame_count) << "Couldn't find a visible frame. "
<< "remaining_layers: " << remained_layers
<< " super_frame: " << i
<< " is_multiple_frame_context: " << is_multiple_frame_context;
if (frame == frame_count - 1 && !is_multiple_frame_context)
continue;
frame_data += frame_sizes[frame];
// We need to add one more frame for multiple frame context.
if (is_multiple_frame_context)
++frame;
uint8_t marker =
static_cast<const uint8_t *>(inputs[i].buf)[inputs[i].sz - 1];
const uint32_t mag = ((marker >> 3) & 0x3) + 1;
const size_t index_sz = 2 + mag * frame_count;
const size_t new_index_sz = 2 + mag * (frame + 1);
marker &= 0x0f8;
marker |= frame;
// Copy existing frame sizes.
memmove(frame_data + (is_multiple_frame_context ? 2 : 1),
frame_start + inputs[i].sz - index_sz + 1, new_index_sz - 2);
if (is_multiple_frame_context) {
// Add a one byte frame with flag show_existing frame.
*frame_data++ = 0x88 | (remained_layers - 1);
}
// New marker.
frame_data[0] = marker;
frame_data += (mag * (frame + 1) + 1);
if (is_multiple_frame_context) {
// Write the frame size for the one byte frame.
frame_data -= mag;
*frame_data++ = 1;
for (uint32_t j = 1; j < mag; ++j) {
*frame_data++ = 0;
// We need to add one more frame for multiple frame contexts.
if (is_multiple_frame_contexts)
++frame;
uint8_t marker =
static_cast<const uint8_t*>(inputs[i].buf)[inputs[i].sz - 1];
const uint32_t mag = ((marker >> 3) & 0x3) + 1;
const size_t index_sz = 2 + mag * frame_count;
const size_t new_index_sz = 2 + mag * (frame + 1);
marker &= 0x0f8;
marker |= frame;
// Copy existing frame sizes.
memmove(frame_data + (is_multiple_frame_contexts ? 2 : 1),
frame_start + inputs[i].sz - index_sz + 1, new_index_sz - 2);
if (is_multiple_frame_contexts) {
// Add a one byte frame with flag show_existing_frame.
*frame_data++ = 0x88 | (remained_spatial_layers - 1);
}
// New marker.
frame_data[0] = marker;
frame_data += (mag * (frame + 1) + 1);
if (is_multiple_frame_contexts) {
// Write the frame size for the one byte frame.
frame_data -= mag;
*frame_data++ = 1;
for (uint32_t j = 1; j < mag; ++j) {
*frame_data++ = 0;
}
}
}
*frame_data++ = marker;
inputs[i].sz = frame_data - frame_start;
*frame_data++ = marker;
inputs[i].sz = frame_data - frame_start;
if (is_multiple_frame_context) {
// Change the show frame flag to 0 for all frames.
for (int j = 0; j < frame; ++j) {
frame_start[0] &= ~2;
frame_start += frame_sizes[j];
if (is_multiple_frame_contexts) {
// Change the show frame flag to 0 for all frames.
for (int j = 0; j < frame; ++j) {
frame_start[0] &= ~2;
frame_start += frame_sizes[j];
}
}
}
}
......@@ -359,7 +405,7 @@ TEST_F(SvcTest, InvalidOptions) {
}
TEST_F(SvcTest, SetLayersOption) {
vpx_codec_err_t res = vpx_svc_set_options(&svc_, "layers=3");
vpx_codec_err_t res = vpx_svc_set_options(&svc_, "spatial-layers=3");
EXPECT_EQ(VPX_CODEC_OK, res);
InitializeEncoder();
EXPECT_EQ(3, svc_.spatial_layers);
......@@ -367,7 +413,7 @@ TEST_F(SvcTest, SetLayersOption) {
TEST_F(SvcTest, SetMultipleOptions) {
vpx_codec_err_t res =
vpx_svc_set_options(&svc_, "layers=2 scale-factors=1/3,2/3");
vpx_svc_set_options(&svc_, "spatial-layers=2 scale-factors=1/3,2/3");
EXPECT_EQ(VPX_CODEC_OK, res);
InitializeEncoder();
EXPECT_EQ(2, svc_.spatial_layers);
......@@ -529,7 +575,7 @@ TEST_F(SvcTest, TwoPassEncode20FramesWithAltRef) {
FreeBitstreamBuffers(&outputs[0], 20);
}
TEST_F(SvcTest, TwoPassEncode2LayersDecodeBaseLayerOnly) {
TEST_F(SvcTest, TwoPassEncode2SpatialLayersDecodeBaseLayerOnly) {
// First pass encode
std::string stats_buf;
Pass1EncodeNFrames(10, 2, &stats_buf);
......@@ -540,12 +586,12 @@ TEST_F(SvcTest, TwoPassEncode2LayersDecodeBaseLayerOnly) {
vpx_fixed_buf outputs[10];
memset(&outputs[0], 0, sizeof(outputs));
Pass2EncodeNFrames(&stats_buf, 10, 2, &outputs[0]);
DropEnhancementLayers(&outputs[0], 10, 1, false);
DropLayersAndMakeItVP9Comaptible(&outputs[0], 10, 1, false);
DecodeNFrames(&outputs[0], 10);
FreeBitstreamBuffers(&outputs[0], 10);
}
TEST_F(SvcTest, TwoPassEncode5LayersDecode54321Layers) {
TEST_F(SvcTest, TwoPassEncode5SpatialLayersDecode54321Layers) {
// First pass encode
std::string stats_buf;
Pass1EncodeNFrames(10, 5, &stats_buf);
......@@ -558,13 +604,13 @@ TEST_F(SvcTest, TwoPassEncode5LayersDecode54321Layers) {
Pass2EncodeNFrames(&stats_buf, 10, 5, &outputs[0]);
DecodeNFrames(&outputs[0], 10);
DropEnhancementLayers(&outputs[0], 10, 4, false);
DropLayersAndMakeItVP9Comaptible(&outputs[0], 10, 4, false);
DecodeNFrames(&outputs[0], 10);
DropEnhancementLayers(&outputs[0], 10, 3, false);
DropLayersAndMakeItVP9Comaptible(&outputs[0], 10, 3, false);
DecodeNFrames(&outputs[0], 10);
DropEnhancementLayers(&outputs[0], 10, 2, false);
DropLayersAndMakeItVP9Comaptible(&outputs[0], 10, 2, false);
DecodeNFrames(&outputs[0], 10);
DropEnhancementLayers(&outputs[0], 10, 1, false);
DropLayersAndMakeItVP9Comaptible(&outputs[0], 10, 1, false);
DecodeNFrames(&outputs[0], 10);
FreeBitstreamBuffers(&outputs[0], 10);
......@@ -601,15 +647,15 @@ TEST_F(SvcTest, TwoPassEncode3SNRLayersDecode321Layers) {
memset(&outputs[0], 0, sizeof(outputs));
Pass2EncodeNFrames(&stats_buf, 20, 3, &outputs[0]);
DecodeNFrames(&outputs[0], 20);
DropEnhancementLayers(&outputs[0], 20, 2, false);
DropLayersAndMakeItVP9Comaptible(&outputs[0], 20, 2, false);
DecodeNFrames(&outputs[0], 20);
DropEnhancementLayers(&outputs[0], 20, 1, false);
DropLayersAndMakeItVP9Comaptible(&outputs[0], 20, 1, false);
DecodeNFrames(&outputs[0], 20);
FreeBitstreamBuffers(&outputs[0], 20);
}
TEST_F(SvcTest, SetMultipleFrameContextOption) {
TEST_F(SvcTest, SetMultipleFrameContextsOption) {
svc_.spatial_layers = 5;
vpx_codec_err_t res =
vpx_svc_set_options(&svc_, "multi-frame-contexts=1");
......@@ -622,7 +668,7 @@ TEST_F(SvcTest, SetMultipleFrameContextOption) {
InitializeEncoder();
}
TEST_F(SvcTest, TwoPassEncode2LayersWithMultipleFrameContext) {
TEST_F(SvcTest, TwoPassEncode2SpatialLayersWithMultipleFrameContexts) {
// First pass encode
std::string stats_buf;
Pass1EncodeNFrames(10, 2, &stats_buf);
......@@ -634,12 +680,13 @@ TEST_F(SvcTest, TwoPassEncode2LayersWithMultipleFrameContext) {
vpx_fixed_buf outputs[10];
memset(&outputs[0], 0, sizeof(outputs));
Pass2EncodeNFrames(&stats_buf, 10, 2, &outputs[0]);
DropEnhancementLayers(&outputs[0], 10, 2, true);
DropLayersAndMakeItVP9Comaptible(&outputs[0], 10, 2, true);
DecodeNFrames(&outputs[0], 10);
FreeBitstreamBuffers(&outputs[0], 10);
}
TEST_F(SvcTest, TwoPassEncode2LayersWithMultipleFrameContextDecodeBaselayer) {
TEST_F(SvcTest,
TwoPassEncode2SpatialLayersWithMultipleFrameContextsDecodeBaselayer) {
// First pass encode
std::string stats_buf;
Pass1EncodeNFrames(10, 2, &stats_buf);
......@@ -651,12 +698,12 @@ TEST_F(SvcTest, TwoPassEncode2LayersWithMultipleFrameContextDecodeBaselayer) {
vpx_fixed_buf outputs[10];
memset(&outputs[0], 0, sizeof(outputs));
Pass2EncodeNFrames(&stats_buf, 10, 2, &outputs[0]);
DropEnhancementLayers(&outputs[0], 10, 1, true);
DropLayersAndMakeItVP9Comaptible(&outputs[0], 10, 1, true);
DecodeNFrames(&outputs[0], 10);
FreeBitstreamBuffers(&outputs[0], 10);
}
TEST_F(SvcTest, TwoPassEncode2SNRLayersWithMultipleFrameContext) {
TEST_F(SvcTest, TwoPassEncode2SNRLayersWithMultipleFrameContexts) {
// First pass encode
std::string stats_buf;
vpx_svc_set_options(&svc_, "scale-factors=1/1,1/1");
......@@ -670,12 +717,13 @@ TEST_F(SvcTest, TwoPassEncode2SNRLayersWithMultipleFrameContext) {
vpx_fixed_buf outputs[10];
memset(&outputs[0], 0, sizeof(outputs));
Pass2EncodeNFrames(&stats_buf, 10, 2, &outputs[0]);
DropEnhancementLayers(&outputs[0], 10, 2, true);
DropLayersAndMakeItVP9Comaptible(&outputs[0], 10, 2, true);
DecodeNFrames(&outputs[0], 10);
FreeBitstreamBuffers(&outputs[0], 10);
}
TEST_F(SvcTest, TwoPassEncode3SNRLayersWithMultipleFrameContextDecode321Layer) {
TEST_F(SvcTest,
TwoPassEncode3SNRLayersWithMultipleFrameContextsDecode321Layer) {
// First pass encode
std::string stats_buf;
vpx_svc_set_options(&svc_, "scale-factors=1/1,1/1,1/1");
......@@ -697,25 +745,114 @@ TEST_F(SvcTest, TwoPassEncode3SNRLayersWithMultipleFrameContextDecode321Layer) {
memcpy(outputs_new[i].buf, outputs[i].buf, outputs[i].sz);
outputs_new[i].sz = outputs[i].sz;
}
DropEnhancementLayers(&outputs_new[0], 10, 3, true);
DropLayersAndMakeItVP9Comaptible(&outputs_new[0], 10, 3, true);
DecodeNFrames(&outputs_new[0], 10);
for (int i = 0; i < 10; ++i) {
memcpy(outputs_new[i].buf, outputs[i].buf, outputs[i].sz);
outputs_new[i].sz = outputs[i].sz;
}
DropEnhancementLayers(&outputs_new[0], 10, 2, true);
DropLayersAndMakeItVP9Comaptible(&outputs_new[0], 10, 2, true);
DecodeNFrames(&outputs_new[0], 10);
for (int i = 0; i < 10; ++i) {
memcpy(outputs_new[i].buf, outputs[i].buf, outputs[i].sz);
outputs_new[i].sz = outputs[i].sz;
}
DropEnhancementLayers(&outputs_new[0], 10, 1, true);
DropLayersAndMakeItVP9Comaptible(&outputs_new[0], 10, 1, true);
DecodeNFrames(&outputs_new[0], 10);
FreeBitstreamBuffers(&outputs[0], 10);
FreeBitstreamBuffers(&outputs_new[0], 10);
}
TEST_F(SvcTest, TwoPassEncode2TemporalLayers) {
// First pass encode
std::string stats_buf;
vpx_svc_set_options(&svc_, "scale-factors=1/1");
svc_.temporal_layers = 2;
Pass1EncodeNFrames(10, 1, &stats_buf);
// Second pass encode
codec_enc_.g_pass = VPX_RC_LAST_PASS;
svc_.temporal_layers = 2;
vpx_svc_set_options(&svc_, "auto-alt-refs=1 scale-factors=1/1");
vpx_fixed_buf outputs[10];
memset(&outputs[0], 0, sizeof(outputs));
Pass2EncodeNFrames(&stats_buf, 10, 1, &outputs[0]);
DecodeNFrames(&outputs[0], 10);
FreeBitstreamBuffers(&outputs[0], 10);
}
TEST_F(SvcTest, TwoPassEncode2TemporalLayersWithMultipleFrameContexts) {
// First pass encode
std::string stats_buf;
vpx_svc_set_options(&svc_, "scale-factors=1/1");
svc_.temporal_layers = 2;
Pass1EncodeNFrames(10, 1, &stats_buf);
// Second pass encode
codec_enc_.g_pass = VPX_RC_LAST_PASS;
svc_.temporal_layers = 2;
codec_enc_.g_error_resilient = 0;
vpx_svc_set_options(&svc_, "auto-alt-refs=1 scale-factors=1/1 "
"multi-frame-contexts=1");
vpx_fixed_buf outputs[10];
memset(&outputs[0], 0, sizeof(outputs));
Pass2EncodeNFrames(&stats_buf, 10, 1, &outputs[0]);
DropLayersAndMakeItVP9Comaptible(&outputs[0], 10, 1, true);
DecodeNFrames(&outputs[0], 10);
FreeBitstreamBuffers(&outputs[0], 10);
}
TEST_F(SvcTest, TwoPassEncode2TemporalLayersDecodeBaseLayer) {
// First pass encode
std::string stats_buf;
vpx_svc_set_options(&svc_, "scale-factors=1/1");
svc_.temporal_layers = 2;
Pass1EncodeNFrames(10, 1, &stats_buf);
// Second pass encode
codec_enc_.g_pass = VPX_RC_LAST_PASS;
svc_.temporal_layers = 2;
vpx_svc_set_options(&svc_, "auto-alt-refs=1 scale-factors=1/1");
vpx_fixed_buf outputs[10];
memset(&outputs[0], 0, sizeof(outputs));
Pass2EncodeNFrames(&stats_buf, 10, 1, &outputs[0]);
vpx_fixed_buf base_layer[5];
for (int i = 0; i < 5; ++i)
base_layer[i] = outputs[i * 2];
DecodeNFrames(&base_layer[0], 5);
FreeBitstreamBuffers(&outputs[0], 10);
}
TEST_F(SvcTest,
TwoPassEncode2TemporalLayersWithMultipleFrameContextsDecodeBaseLayer) {
// First pass encode
std::string stats_buf;
vpx_svc_set_options(&svc_, "scale-factors=1/1");
svc_.temporal_layers = 2;
Pass1EncodeNFrames(10, 1, &stats_buf);
// Second pass encode
codec_enc_.g_pass = VPX_RC_LAST_PASS;
svc_.temporal_layers = 2;
codec_enc_.g_error_resilient = 0;
vpx_svc_set_options(&svc_, "auto-alt-refs=1 scale-factors=1/1 "
"multi-frame-contexts=1");
vpx_fixed_buf outputs[10];
memset(&outputs[0], 0, sizeof(outputs));
Pass2EncodeNFrames(&stats_buf, 10, 1, &outputs[0]);
DropLayersAndMakeItVP9Comaptible(&outputs[0], 10, 1, true);
vpx_fixed_buf base_layer[5];
for (int i = 0; i < 5; ++i)
base_layer[i] = outputs[i * 2];
DecodeNFrames(&base_layer[0], 5);
FreeBitstreamBuffers(&outputs[0], 10);
}
} // namespace
vp9/encoder/vp9_bitstream.c
View file @ be3b08da
......@@ -999,8 +999,10 @@ static void write_frame_size_with_refs(VP9_COMP *cpi,
// Set "found" to 0 for temporal svc and for spatial svc key frame
if (cpi->use_svc &&
(cpi->svc.number_spatial_layers == 1 ||
cpi->svc.layer_context[cpi->svc.spatial_layer_id].is_key_frame)) {
((cpi->svc.number_temporal_layers > 1 &&
cpi->oxcf.rc_mode == VPX_CBR) ||
(cpi->svc.number_spatial_layers > 1 &&
cpi->svc.layer_context[cpi->svc.spatial_layer_id].is_key_frame))) {
found = 0;
}
vp9_wb_write_bit(wb, found);
......@@ -1093,7 +1095,7 @@ static void write_uncompressed_header(VP9_COMP *cpi,
// show_existing_frame flag which tells the decoder which frame we want to
// show.
if (!cm->show_frame ||
(is_spatial_svc(cpi) && cm->error_resilient_mode == 0))
(is_two_pass_svc(cpi) && cm->error_resilient_mode == 0))
vp9_wb_write_bit(wb, cm->intra_only);
if (!cm->error_resilient_mode)
......
vp9/encoder/vp9_bitstream.h
View file @ be3b08da
......@@ -26,7 +26,7 @@ static INLINE int vp9_preserve_existing_gf(VP9_COMP *cpi) {
return !cpi->multi_arf_allowed && cpi->refresh_golden_frame &&
cpi->rc.is_src_frame_alt_ref &&
(!cpi->use_svc || // Add spatial svc base layer case here
(is_spatial_svc(cpi) &&
(is_two_pass_svc(cpi) &&
cpi->svc.spatial_layer_id == 0 &&
cpi->svc.layer_context[0].gold_ref_idx >=0 &&
cpi->oxcf.ss_play_alternate[0]));
......
vp9/encoder/vp9_encoder.c
View file @ be3b08da
......@@ -128,7 +128,7 @@ static void setup_frame(VP9_COMP *cpi) {
}
if (cm->frame_type == KEY_FRAME) {
if (!is_spatial_svc(cpi))
if (!is_two_pass_svc(cpi))
cpi->refresh_golden_frame = 1;
cpi->refresh_alt_ref_frame = 1;
vp9_zero(cpi->interp_filter_selected);
......@@ -510,7 +510,7 @@ static void update_frame_size(VP9_COMP *cpi) {
vp9_init_context_buffers(cm);
init_macroblockd(cm, xd);
if (is_spatial_svc(cpi)) {
if (is_two_pass_svc(cpi)) {
if (vp9_realloc_frame_buffer(&cpi->alt_ref_buffer,
cm->width, cm->height,
cm->subsampling_x, cm->subsampling_y,
......@@ -562,7 +562,9 @@ static void init_config(struct VP9_COMP *cpi, VP9EncoderConfig *oxcf) {
cpi->svc.number_temporal_layers = oxcf->ts_number_layers;
if ((cpi->svc.number_temporal_layers > 1 && cpi->oxcf.rc_mode == VPX_CBR) ||
(cpi->svc.number_spatial_layers > 1 && cpi->oxcf.pass == 2)) {
((cpi->svc.number_temporal_layers > 1 ||
cpi->svc.number_spatial_layers > 1) &&
cpi->oxcf.pass == 2)) {
vp9_init_layer_context(cpi);
}
......@@ -654,7 +656,9 @@ void vp9_change_config(struct VP9_COMP *cpi, const VP9EncoderConfig *oxcf) {
if ((cpi->svc.number_temporal_layers > 1 &&
cpi->oxcf.rc_mode == VPX_CBR) ||
(cpi->svc.number_spatial_layers > 1 && cpi->oxcf.pass == 2)) {
((cpi->svc.number_temporal_layers > 1 ||
cpi->svc.number_spatial_layers > 1) &&
cpi->oxcf.pass == 2)) {
vp9_update_layer_context_change_config(cpi,
(int)cpi->oxcf.target_bandwidth);
}
......@@ -902,7 +906,7 @@ VP9_COMP *vp9_create_compressor(VP9EncoderConfig *oxcf) {
const int packets = (int)(oxcf->two_pass_stats_in.sz / packet_sz);
if (cpi->svc.number_spatial_layers > 1
&& cpi->svc.number_temporal_layers == 1) {
|| cpi->svc.number_temporal_layers > 1) {
FIRSTPASS_STATS *const stats = oxcf->two_pass_stats_in.buf;
FIRSTPASS_STATS *stats_copy[VPX_SS_MAX_LAYERS] = {0};
int i;
......@@ -1510,7 +1514,7 @@ void vp9_update_reference_frames(VP9_COMP *cpi) {
cpi->alt_fb_idx = cpi->gld_fb_idx;
cpi->gld_fb_idx = tmp;
if (is_spatial_svc(cpi)) {
if (is_two_pass_svc(cpi)) {
cpi->svc.layer_context[0].gold_ref_idx = cpi->gld_fb_idx;
cpi->svc.layer_context[0].alt_ref_idx = cpi->alt_fb_idx;
}
......@@ -1936,8 +1940,7 @@ static int get_ref_frame_flags(const VP9_COMP *cpi) {
if (gold_is_last)
flags &= ~VP9_GOLD_FLAG;
if (cpi->rc.frames_till_gf_update_due == INT_MAX &&
!is_spatial_svc(cpi))
if (cpi->rc.frames_till_gf_update_due == INT_MAX && !is_two_pass_svc(cpi))
flags &= ~VP9_GOLD_FLAG;
if (alt_is_last)
......@@ -1984,7 +1987,7 @@ static int is_skippable_frame(const VP9_COMP *cpi) {
// can be skipped for partition check, and the partition size is assigned
// according to the variance
const SVC *const svc = &cpi->svc;
const TWO_PASS *const twopass = is_spatial_svc(cpi) ?
const TWO_PASS *const twopass = is_two_pass_svc(cpi) ?
&svc->layer_context[svc->spatial_layer_id].twopass : &cpi->twopass;
return (!frame_is_intra_only(&cpi->common) &&
......@@ -2136,18 +2139,34 @@ static void encode_frame_to_data_rate(VP9_COMP *cpi,
cm->reset_frame_context = 2;
}
}
if (is_spatial_svc(cpi) && cm->error_resilient_mode == 0) {
cm->frame_context_idx = cpi->svc.spatial_layer_id;
if (is_two_pass_svc(cpi) && cm->error_resilient_mode == 0) {
cm->frame_context_idx =
cpi->svc.spatial_layer_id * cpi->svc.number_temporal_layers +
cpi->svc.temporal_layer_id;
// The probs will be updated based on the frame type of its previous
// frame if frame_parallel_decoding_mode is 0. The type may vary for
// the frame after a key frame in base layer since we may drop enhancement
// layers. So set frame_parallel_decoding_mode to 1 in this case.
if (cpi->svc.spatial_layer_id == 0 &&
cpi->svc.layer_context[0].last_frame_type == KEY_FRAME)
cm->frame_parallel_decoding_mode = 1;
else
cm->frame_parallel_decoding_mode = 0;
if (cpi->svc.number_temporal_layers == 1) {
if (cpi->svc.spatial_layer_id == 0 &&
cpi->svc.layer_context[0].last_frame_type == KEY_FRAME)
cm->frame_parallel_decoding_mode = 1;
else
cm->frame_parallel_decoding_mode = 0;
} else if (cpi->svc.spatial_layer_id == 0) {
// Find the 2nd frame in temporal base layer and 1st frame in temporal
// enhancement layers from the key frame.
int i;
for (i = 0; i < cpi->svc.number_temporal_layers; ++i) {
if (cpi->svc.layer_context[0].frames_from_key_frame == 1 << i) {