Commit 4864ab21 authored by Marco Paniconi's avatar Marco Paniconi

Layer based rate control for CBR mode.

This patch adds a buffer-based rate control for temporal layers,
under CBR mode.

Added vpx_temporal_scalable_patters.c encoder for testing temporal
layers, for both vp9 and vp8 (replaces the old vp8_scalable_patterns).

Updated datarate unittest with tests for temporal layer rate-targeting.

Change-Id: I8900a854288b9354d9c697cfeb0243a9fd6790b1
parent 9602ed88
......@@ -54,9 +54,6 @@ vpxenc.SRCS += third_party/libmkv/EbmlWriter.h
vpxenc.SRCS += $(LIBYUV_SRCS)
vpxenc.GUID = 548DEC74-7A15-4B2B-AFC3-AA102E7C25C1
vpxenc.DESCRIPTION = Full featured encoder
UTILS-$(CONFIG_VP8_ENCODER) += vp8_scalable_patterns.c
vp8_scalable_patterns.GUID = 0D6A210B-F482-4D6F-8570-4A9C01ACC88C
vp8_scalable_patterns.DESCRIPTION = Temporal Scalability Encoder
UTILS-$(CONFIG_VP9_ENCODER) += vp9_spatial_scalable_encoder.c
vp9_spatial_scalable_encoder.SRCS += args.c args.h
vp9_spatial_scalable_encoder.SRCS += ivfenc.c ivfenc.h
......@@ -73,6 +70,11 @@ endif
#example_xma.GUID = A955FC4A-73F1-44F7-135E-30D84D32F022
#example_xma.DESCRIPTION = External Memory Allocation mode usage
EXAMPLES-$(CONFIG_ENCODERS) += vpx_temporal_scalable_patterns.c
vpx_temporal_scalable_patterns.SRCS += ivfenc.c ivfenc.h
vpx_temporal_scalable_patterns.SRCS += tools_common.c tools_common.h
vpx_temporal_scalable_patterns.GUID = B18C08F2-A439-4502-A78E-849BE3D60947
vpx_temporal_scalable_patterns.DESCRIPTION = Temporal Scalability Encoder
EXAMPLES-$(CONFIG_VP8_DECODER) += simple_decoder.c
simple_decoder.GUID = D3BBF1E9-2427-450D-BBFF-B2843C1D44CC
simple_decoder.SRCS += ivfdec.h ivfdec.c
......
This diff is collapsed.
This diff is collapsed.
......@@ -123,6 +123,11 @@ class Encoder {
ASSERT_EQ(VPX_CODEC_OK, res) << EncoderError();
}
void Control(int ctrl_id, struct vpx_svc_layer_id *arg) {
const vpx_codec_err_t res = vpx_codec_control_(&encoder_, ctrl_id, arg);
ASSERT_EQ(VPX_CODEC_OK, res) << EncoderError();
}
void set_deadline(unsigned long deadline) {
deadline_ = deadline;
}
......
This diff is collapsed.
......@@ -147,8 +147,12 @@ extern "C" {
// END DATARATE CONTROL OPTIONS
// ----------------------------------------------------------------
// Spatial scalability
int ss_number_layers;
// Spatial and temporal scalability.
int ss_number_layers; // Number of spatial layers.
int ts_number_layers; // Number of temporal layers.
// Bitrate allocation (CBR mode) and framerate factor, for temporal layers.
int ts_target_bitrate[VPX_TS_MAX_LAYERS];
int ts_rate_decimator[VPX_TS_MAX_LAYERS];
// these parameters aren't to be used in final build don't use!!!
int play_alternate;
......
......@@ -1144,6 +1144,109 @@ static int64_t rescale(int val, int64_t num, int denom) {
return (llval * llnum / llden);
}
// Initialize layer context data from init_config().
static void init_layer_context(VP9_COMP *const cpi) {
const VP9_CONFIG *const oxcf = &cpi->oxcf;
int temporal_layer = 0;
cpi->svc.spatial_layer_id = 0;
cpi->svc.temporal_layer_id = 0;
for (temporal_layer = 0; temporal_layer < cpi->svc.number_temporal_layers;
++temporal_layer) {
LAYER_CONTEXT *const lc = &cpi->svc.layer_context[temporal_layer];
RATE_CONTROL *const lrc = &lc->rc;
lrc->active_worst_quality = q_trans[oxcf->worst_allowed_q];
lrc->avg_frame_qindex[INTER_FRAME] = q_trans[oxcf->worst_allowed_q];
lrc->last_q[INTER_FRAME] = q_trans[oxcf->worst_allowed_q];
lrc->ni_av_qi = lrc->active_worst_quality;
lrc->total_actual_bits = 0;
lrc->total_target_vs_actual = 0;
lrc->ni_tot_qi = 0;
lrc->tot_q = 0.0;
lrc->ni_frames = 0;
lrc->rate_correction_factor = 1.0;
lrc->key_frame_rate_correction_factor = 1.0;
lc->target_bandwidth = oxcf->ts_target_bitrate[temporal_layer] *
1000;
lrc->buffer_level = rescale((int)(oxcf->starting_buffer_level),
lc->target_bandwidth, 1000);
lrc->bits_off_target = lrc->buffer_level;
}
}
// Update the layer context from a change_config() call.
static void update_layer_context_change_config(VP9_COMP *const cpi,
const int target_bandwidth) {
const VP9_CONFIG *const oxcf = &cpi->oxcf;
const RATE_CONTROL *const rc = &cpi->rc;
int temporal_layer = 0;
float bitrate_alloc = 1.0;
for (temporal_layer = 0; temporal_layer < cpi->svc.number_temporal_layers;
++temporal_layer) {
LAYER_CONTEXT *const lc = &cpi->svc.layer_context[temporal_layer];
RATE_CONTROL *const lrc = &lc->rc;
lc->target_bandwidth = oxcf->ts_target_bitrate[temporal_layer] * 1000;
bitrate_alloc = (float)lc->target_bandwidth / (float)target_bandwidth;
// Update buffer-related quantities.
lc->starting_buffer_level = oxcf->starting_buffer_level * bitrate_alloc;
lc->optimal_buffer_level = oxcf->optimal_buffer_level * bitrate_alloc;
lc->maximum_buffer_size = oxcf->maximum_buffer_size * bitrate_alloc;
lrc->bits_off_target = MIN(lrc->bits_off_target, lc->maximum_buffer_size);
lrc->buffer_level = MIN(lrc->buffer_level, lc->maximum_buffer_size);
// Update framerate-related quantities.
lc->framerate = oxcf->framerate / oxcf->ts_rate_decimator[temporal_layer];
lrc->av_per_frame_bandwidth = (int)(lc->target_bandwidth / lc->framerate);
lrc->max_frame_bandwidth = rc->max_frame_bandwidth;
// Update qp-related quantities.
lrc->worst_quality = rc->worst_quality;
lrc->best_quality = rc->best_quality;
lrc->active_worst_quality = rc->active_worst_quality;
}
}
// Prior to encoding the frame, update framerate-related quantities
// for the current layer.
static void update_layer_framerate(VP9_COMP *const cpi) {
int temporal_layer = cpi->svc.temporal_layer_id;
LAYER_CONTEXT *const lc = &cpi->svc.layer_context[temporal_layer];
RATE_CONTROL *const lrc = &lc->rc;
lc->framerate = cpi->oxcf.framerate /
cpi->oxcf.ts_rate_decimator[temporal_layer];
lrc->av_per_frame_bandwidth = (int)(lc->target_bandwidth /
lc->framerate);
lrc->max_frame_bandwidth = cpi->rc.max_frame_bandwidth;
}
// Prior to encoding the frame, set the layer context, for the current layer
// to be encoded, to the cpi struct.
static void restore_layer_context(VP9_COMP *const cpi) {
int temporal_layer = cpi->svc.temporal_layer_id;
LAYER_CONTEXT *lc = &cpi->svc.layer_context[temporal_layer];
int frame_since_key = cpi->rc.frames_since_key;
int frame_to_key = cpi->rc.frames_to_key;
cpi->rc = lc->rc;
cpi->oxcf.target_bandwidth = lc->target_bandwidth;
cpi->oxcf.starting_buffer_level = lc->starting_buffer_level;
cpi->oxcf.optimal_buffer_level = lc->optimal_buffer_level;
cpi->oxcf.maximum_buffer_size = lc->maximum_buffer_size;
cpi->output_framerate = lc->framerate;
// Reset the frames_since_key and frames_to_key counters to their values
// before the layer restore. Keep these defined for the stream (not layer).
cpi->rc.frames_since_key = frame_since_key;
cpi->rc.frames_to_key = frame_to_key;
}
// Save the layer context after encoding the frame.
static void save_layer_context(VP9_COMP *const cpi) {
int temporal_layer = cpi->svc.temporal_layer_id;
LAYER_CONTEXT *lc = &cpi->svc.layer_context[temporal_layer];
lc->rc = cpi->rc;
lc->target_bandwidth = cpi->oxcf.target_bandwidth;
lc->starting_buffer_level = cpi->oxcf.starting_buffer_level;
lc->optimal_buffer_level = cpi->oxcf.optimal_buffer_level;
lc->maximum_buffer_size = cpi->oxcf.maximum_buffer_size;
lc->framerate = cpi->output_framerate;
}
static void set_tile_limits(VP9_COMP *cpi) {
VP9_COMMON *const cm = &cpi->common;
......@@ -1170,6 +1273,16 @@ static void init_config(VP9_PTR ptr, VP9_CONFIG *oxcf) {
cm->subsampling_y = 0;
vp9_alloc_compressor_data(cpi);
// Spatial scalability.
cpi->svc.number_spatial_layers = oxcf->ss_number_layers;
// Temporal scalability.
cpi->svc.number_temporal_layers = oxcf->ts_number_layers;
if (cpi->svc.number_temporal_layers > 1 &&
cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER) {
init_layer_context(cpi);
}
// change includes all joint functionality
vp9_change_config(ptr, oxcf);
......@@ -1210,9 +1323,6 @@ static void init_config(VP9_PTR ptr, VP9_CONFIG *oxcf) {
cpi->gld_fb_idx = 1;
cpi->alt_fb_idx = 2;
cpi->current_layer = 0;
cpi->use_svc = 0;
set_tile_limits(cpi);
cpi->fixed_divide[0] = 0;
......@@ -1220,7 +1330,6 @@ static void init_config(VP9_PTR ptr, VP9_CONFIG *oxcf) {
cpi->fixed_divide[i] = 0x80000 / i;
}
void vp9_change_config(VP9_PTR ptr, VP9_CONFIG *oxcf) {
VP9_COMP *cpi = (VP9_COMP *)(ptr);
VP9_COMMON *const cm = &cpi->common;
......@@ -1312,10 +1421,10 @@ void vp9_change_config(VP9_PTR ptr, VP9_CONFIG *oxcf) {
cpi->oxcf.target_bandwidth, 1000);
// Under a configuration change, where maximum_buffer_size may change,
// keep buffer level clipped to the maximum allowed buffer size.
if (cpi->rc.bits_off_target > cpi->oxcf.maximum_buffer_size) {
cpi->rc.bits_off_target = cpi->oxcf.maximum_buffer_size;
cpi->rc.buffer_level = cpi->rc.bits_off_target;
}
cpi->rc.bits_off_target = MIN(cpi->rc.bits_off_target,
cpi->oxcf.maximum_buffer_size);
cpi->rc.buffer_level = MIN(cpi->rc.buffer_level,
cpi->oxcf.maximum_buffer_size);
// Set up frame rate and related parameters rate control values.
vp9_new_framerate(cpi, cpi->oxcf.framerate);
......@@ -1350,6 +1459,11 @@ void vp9_change_config(VP9_PTR ptr, VP9_CONFIG *oxcf) {
}
update_frame_size(cpi);
if (cpi->svc.number_temporal_layers > 1 &&
cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER) {
update_layer_context_change_config(cpi, cpi->oxcf.target_bandwidth);
}
cpi->speed = cpi->oxcf.cpu_used;
if (cpi->oxcf.lag_in_frames == 0) {
......@@ -1573,6 +1687,8 @@ VP9_PTR vp9_create_compressor(VP9_CONFIG *oxcf) {
vp9_create_common(cm);
cpi->use_svc = 0;
init_config((VP9_PTR)cpi, oxcf);
init_pick_mode_context(cpi);
......@@ -1588,9 +1704,6 @@ VP9_PTR vp9_create_compressor(VP9_CONFIG *oxcf) {
cpi->alt_is_last = 0;
cpi->gold_is_alt = 0;
// Spatial scalability
cpi->number_spatial_layers = oxcf->ss_number_layers;
// Create the encoder segmentation map and set all entries to 0
CHECK_MEM_ERROR(cm, cpi->segmentation_map,
vpx_calloc(cm->mi_rows * cm->mi_cols, 1));
......@@ -3504,6 +3617,12 @@ int vp9_get_compressed_data(VP9_PTR ptr, unsigned int *frame_flags,
adjust_frame_rate(cpi);
}
if (cpi->svc.number_temporal_layers > 1 &&
cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER) {
update_layer_framerate(cpi);
restore_layer_context(cpi);
}
// start with a 0 size frame
*size = 0;
......@@ -3579,6 +3698,12 @@ int vp9_get_compressed_data(VP9_PTR ptr, unsigned int *frame_flags,
cpi->droppable = !frame_is_reference(cpi);
}
// Save layer specific state.
if (cpi->svc.number_temporal_layers > 1 &&
cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER) {
save_layer_context(cpi);
}
vpx_usec_timer_mark(&cmptimer);
cpi->time_compress_data += vpx_usec_timer_elapsed(&cmptimer);
......
......@@ -407,6 +407,15 @@ typedef struct {
int super_fast_rtc;
} SPEED_FEATURES;
typedef struct {
RATE_CONTROL rc;
int target_bandwidth;
int64_t starting_buffer_level;
int64_t optimal_buffer_level;
int64_t maximum_buffer_size;
double framerate;
} LAYER_CONTEXT;
typedef struct VP9_COMP {
DECLARE_ALIGNED(16, int16_t, y_quant[QINDEX_RANGE][8]);
DECLARE_ALIGNED(16, int16_t, y_quant_shift[QINDEX_RANGE][8]);
......@@ -451,9 +460,6 @@ typedef struct VP9_COMP {
int gld_fb_idx;
int alt_fb_idx;
int current_layer;
int use_svc;
#if CONFIG_MULTIPLE_ARF
int alt_ref_fb_idx[REF_FRAMES - 3];
#endif
......@@ -669,7 +675,18 @@ typedef struct VP9_COMP {
int initial_width;
int initial_height;
int number_spatial_layers;
int use_svc;
struct svc {
int spatial_layer_id;
int temporal_layer_id;
int number_spatial_layers;
int number_temporal_layers;
// Layer context used for rate control in CBR mode, only defined for
// temporal layers for now.
LAYER_CONTEXT layer_context[VPX_TS_MAX_LAYERS];
} svc;
int enable_encode_breakout; // Default value is 1. From first pass stats,
// encode_breakout may be disabled.
......
......@@ -241,6 +241,26 @@ int vp9_rc_clamp_iframe_target_size(const VP9_COMP *const cpi, int target) {
return target;
}
// Update the buffer level for higher layers, given the encoded current layer.
static void update_layer_buffer_level(VP9_COMP *const cpi,
int encoded_frame_size) {
int temporal_layer = 0;
int current_temporal_layer = cpi->svc.temporal_layer_id;
for (temporal_layer = current_temporal_layer + 1;
temporal_layer < cpi->svc.number_temporal_layers; ++temporal_layer) {
LAYER_CONTEXT *lc = &cpi->svc.layer_context[temporal_layer];
RATE_CONTROL *lrc = &lc->rc;
int bits_off_for_this_layer = (int)(lc->target_bandwidth / lc->framerate -
encoded_frame_size);
lrc->bits_off_target += bits_off_for_this_layer;
// Clip buffer level to maximum buffer size for the layer.
lrc->bits_off_target = MIN(lrc->bits_off_target, lc->maximum_buffer_size);
lrc->buffer_level = lrc->bits_off_target;
}
}
// Update the buffer level: leaky bucket model.
static void update_buffer_level(VP9_COMP *cpi, int encoded_frame_size) {
const VP9_COMMON *const cm = &cpi->common;
......@@ -255,14 +275,18 @@ static void update_buffer_level(VP9_COMP *cpi, int encoded_frame_size) {
}
// Clip the buffer level to the maximum specified buffer size.
rc->buffer_level = MIN(rc->bits_off_target, oxcf->maximum_buffer_size);
rc->bits_off_target = MIN(rc->bits_off_target, oxcf->maximum_buffer_size);
rc->buffer_level = rc->bits_off_target;
if (cpi->use_svc && cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER) {
update_layer_buffer_level(cpi, encoded_frame_size);
}
}
int vp9_rc_drop_frame(VP9_COMP *cpi) {
const VP9_CONFIG *oxcf = &cpi->oxcf;
RATE_CONTROL *const rc = &cpi->rc;
if (!oxcf->drop_frames_water_mark) {
return 0;
} else {
......@@ -273,7 +297,7 @@ int vp9_rc_drop_frame(VP9_COMP *cpi) {
// If buffer is below drop_mark, for now just drop every other frame
// (starting with the next frame) until it increases back over drop_mark.
int drop_mark = (int)(oxcf->drop_frames_water_mark *
oxcf->optimal_buffer_level / 100);
oxcf->optimal_buffer_level / 100);
if ((rc->buffer_level > drop_mark) &&
(rc->decimation_factor > 0)) {
--rc->decimation_factor;
......@@ -301,7 +325,8 @@ static double get_rate_correction_factor(const VP9_COMP *cpi) {
if (cpi->common.frame_type == KEY_FRAME) {
return cpi->rc.key_frame_rate_correction_factor;
} else {
if (cpi->refresh_alt_ref_frame || cpi->refresh_golden_frame)
if ((cpi->refresh_alt_ref_frame || cpi->refresh_golden_frame) &&
!(cpi->use_svc && cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER))
return cpi->rc.gf_rate_correction_factor;
else
return cpi->rc.rate_correction_factor;
......@@ -312,7 +337,8 @@ static void set_rate_correction_factor(VP9_COMP *cpi, double factor) {
if (cpi->common.frame_type == KEY_FRAME) {
cpi->rc.key_frame_rate_correction_factor = factor;
} else {
if (cpi->refresh_alt_ref_frame || cpi->refresh_golden_frame)
if ((cpi->refresh_alt_ref_frame || cpi->refresh_golden_frame) &&
!(cpi->use_svc && cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER))
cpi->rc.gf_rate_correction_factor = factor;
else
cpi->rc.rate_correction_factor = factor;
......@@ -538,7 +564,12 @@ static int rc_pick_q_and_adjust_q_bounds_one_pass(const VP9_COMP *cpi,
if (oxcf->end_usage == USAGE_CONSTANT_QUALITY) {
active_best_quality = cpi->cq_target_quality;
} else {
active_best_quality = inter_minq[rc->avg_frame_qindex[INTER_FRAME]];
// Use the lower of active_worst_quality and recent/average Q.
if (rc->avg_frame_qindex[INTER_FRAME] < active_worst_quality)
active_best_quality = inter_minq[rc->avg_frame_qindex[INTER_FRAME]];
else
active_best_quality = inter_minq[active_worst_quality];
//
// For the constrained quality mode we don't want
// q to fall below the cq level.
if ((oxcf->end_usage == USAGE_CONSTRAINED_QUALITY) &&
......@@ -574,7 +605,6 @@ static int rc_pick_q_and_adjust_q_bounds_one_pass(const VP9_COMP *cpi,
*top_index = (active_worst_quality + active_best_quality) / 2;
}
#endif
if (oxcf->end_usage == USAGE_CONSTANT_QUALITY) {
q = active_best_quality;
// Special case code to try and match quality with forced key frames
......@@ -1002,21 +1032,6 @@ void vp9_rc_postencode_update_drop_frame(VP9_COMP *cpi) {
cpi->rc.frames_to_key--;
}
void vp9_rc_get_svc_params(VP9_COMP *cpi) {
VP9_COMMON *const cm = &cpi->common;
if ((cm->current_video_frame == 0) ||
(cm->frame_flags & FRAMEFLAGS_KEY) ||
(cpi->oxcf.auto_key && (cpi->rc.frames_since_key %
cpi->key_frame_frequency == 0))) {
cm->frame_type = KEY_FRAME;
cpi->rc.source_alt_ref_active = 0;
} else {
cm->frame_type = INTER_FRAME;
}
cpi->rc.frames_till_gf_update_due = INT_MAX;
cpi->rc.baseline_gf_interval = INT_MAX;
}
static int test_for_kf_one_pass(VP9_COMP *cpi) {
// Placeholder function for auto key frame
return 0;
......@@ -1171,6 +1186,32 @@ static int calc_iframe_target_size_one_pass_cbr(const VP9_COMP *cpi) {
return target;
}
void vp9_rc_get_svc_params(VP9_COMP *cpi) {
VP9_COMMON *const cm = &cpi->common;
int target = cpi->rc.av_per_frame_bandwidth;
if ((cm->current_video_frame == 0) ||
(cm->frame_flags & FRAMEFLAGS_KEY) ||
(cpi->oxcf.auto_key && (cpi->rc.frames_since_key %
cpi->key_frame_frequency == 0))) {
cm->frame_type = KEY_FRAME;
cpi->rc.source_alt_ref_active = 0;
if (cpi->pass == 0 && cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER) {
target = calc_iframe_target_size_one_pass_cbr(cpi);
cpi->rc.active_worst_quality = cpi->rc.worst_quality;
}
} else {
cm->frame_type = INTER_FRAME;
if (cpi->pass == 0 && cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER) {
target = calc_pframe_target_size_one_pass_cbr(cpi);
cpi->rc.active_worst_quality =
calc_active_worst_quality_one_pass_cbr(cpi);
}
}
vp9_rc_set_frame_target(cpi, target);
cpi->rc.frames_till_gf_update_due = INT_MAX;
cpi->rc.baseline_gf_interval = INT_MAX;
}
void vp9_rc_get_one_pass_cbr_params(VP9_COMP *cpi) {
VP9_COMMON *const cm = &cpi->common;
int target;
......
......@@ -175,6 +175,23 @@ static vpx_codec_err_t validate_config(vpx_codec_alg_priv_t *ctx,
RANGE_CHECK(cfg, ss_number_layers, 1,
VPX_SS_MAX_LAYERS); /*Spatial layers max */
RANGE_CHECK(cfg, ts_number_layers, 1, VPX_TS_MAX_LAYERS);
if (cfg->ts_number_layers > 1) {
int i;
for (i = 1; i < cfg->ts_number_layers; ++i) {
if (cfg->ts_target_bitrate[i] < cfg->ts_target_bitrate[i-1]) {
ERROR("ts_target_bitrate entries are not increasing");
}
}
RANGE_CHECK(cfg, ts_rate_decimator[cfg->ts_number_layers-1], 1, 1);
for (i = cfg->ts_number_layers-2; i > 0; --i) {
if (cfg->ts_rate_decimator[i-1] != 2*cfg->ts_rate_decimator[i]) {
ERROR("ts_rate_decimator factors are not powers of 2");
}
}
}
/* VP8 does not support a lower bound on the keyframe interval in
* automatic keyframe placement mode.
*/
......@@ -342,6 +359,19 @@ static vpx_codec_err_t set_vp9e_config(VP9_CONFIG *oxcf,
oxcf->aq_mode = vp8_cfg.aq_mode;
oxcf->ss_number_layers = cfg.ss_number_layers;
oxcf->ts_number_layers = cfg.ts_number_layers;
if (oxcf->ts_number_layers > 1) {
memcpy(oxcf->ts_target_bitrate, cfg.ts_target_bitrate,
sizeof(cfg.ts_target_bitrate));
memcpy(oxcf->ts_rate_decimator, cfg.ts_rate_decimator,
sizeof(cfg.ts_rate_decimator));
} else if (oxcf->ts_number_layers == 1) {
oxcf->ts_target_bitrate[0] = oxcf->target_bandwidth;
oxcf->ts_rate_decimator[0] = 1;
}
/*
printf("Current VP9 Settings: \n");
printf("target_bandwidth: %d\n", oxcf->target_bandwidth);
......@@ -1012,6 +1042,32 @@ static vpx_codec_err_t vp9e_set_svc(vpx_codec_alg_priv_t *ctx, int ctr_id,
va_list args) {
int data = va_arg(args, int);
vp9_set_svc(ctx->cpi, data);
// CBR mode for SVC with both temporal and spatial layers not yet supported.
if (data == 1 &&
ctx->cfg.rc_end_usage == VPX_CBR &&
ctx->cfg.ss_number_layers > 1 &&
ctx->cfg.ts_number_layers > 1) {
return VPX_CODEC_INVALID_PARAM;
}
return VPX_CODEC_OK;
}
static vpx_codec_err_t vp9e_set_svc_layer_id(vpx_codec_alg_priv_t *ctx,
int ctr_id,
va_list args) {
vpx_svc_layer_id_t *data = va_arg(args, vpx_svc_layer_id_t *);
VP9_COMP *cpi = (VP9_COMP *)ctx->cpi;
cpi->svc.spatial_layer_id = data->spatial_layer_id;
cpi->svc.temporal_layer_id = data->temporal_layer_id;
// Checks on valid layer_id input.
if (cpi->svc.temporal_layer_id < 0 ||
cpi->svc.temporal_layer_id >= ctx->cfg.ts_number_layers) {
return VPX_CODEC_INVALID_PARAM;
}
if (cpi->svc.spatial_layer_id < 0 ||
cpi->svc.spatial_layer_id >= ctx->cfg.ss_number_layers) {
return VPX_CODEC_INVALID_PARAM;
}
return VPX_CODEC_OK;
}
......@@ -1027,7 +1083,9 @@ static vpx_codec_err_t vp9e_set_svc_parameters(vpx_codec_alg_priv_t *ctx,
params = *(vpx_svc_parameters_t *)data;
cpi->current_layer = params.layer;
cpi->svc.spatial_layer_id = params.spatial_layer;
cpi->svc.temporal_layer_id = params.temporal_layer;
cpi->lst_fb_idx = params.lst_fb_idx;
cpi->gld_fb_idx = params.gld_fb_idx;
cpi->alt_fb_idx = params.alt_fb_idx;
......@@ -1076,6 +1134,7 @@ static vpx_codec_ctrl_fn_map_t vp9e_ctf_maps[] = {
{VP9_GET_REFERENCE, get_reference},
{VP9E_SET_SVC, vp9e_set_svc},
{VP9E_SET_SVC_PARAMETERS, vp9e_set_svc_parameters},
{VP9E_SET_SVC_LAYER_ID, vp9e_set_svc_layer_id},
{ -1, NULL},
};
......@@ -1126,7 +1185,11 @@ static vpx_codec_enc_cfg_map_t vp9e_usage_cfg_map[] = {
9999, /* kf_max_dist */
VPX_SS_DEFAULT_LAYERS, /* ss_number_layers */
1, /* ts_number_layers */
{0}, /* ts_target_bitrate */
{0}, /* ts_rate_decimator */
0, /* ts_periodicity */
{0}, /* ts_layer_id */
#if VPX_ENCODER_ABI_VERSION == (1 + VPX_CODEC_ABI_VERSION)
"vp8.fpf" /* first pass filename */
#endif
......
......@@ -499,6 +499,7 @@ vpx_codec_err_t vpx_svc_init(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
// modify encoder configuration
enc_cfg->ss_number_layers = si->layers;
enc_cfg->ts_number_layers = 1; // Temporal layers not used in this encoder.
enc_cfg->kf_mode = VPX_KF_DISABLED;
enc_cfg->g_pass = VPX_RC_ONE_PASS;
// Lag in frames not currently supported
......@@ -691,7 +692,8 @@ static void set_svc_parameters(SvcContext *svc_ctx,
SvcInternal *const si = get_svc_internal(svc_ctx);
memset(&svc_params, 0, sizeof(svc_params));
svc_params.layer = si->layer;
svc_params.temporal_layer = 0;
svc_params.spatial_layer = si->layer;
svc_params.flags = si->enc_frame_flags;
layer = si->layer;
......
......@@ -194,7 +194,8 @@ enum vp8e_enc_control_id {
VP9E_SET_AQ_MODE,
VP9E_SET_SVC,
VP9E_SET_SVC_PARAMETERS
VP9E_SET_SVC_PARAMETERS,
VP9E_SET_SVC_LAYER_ID
};
/*!\brief vpx 1-D scaling mode
......@@ -285,7 +286,8 @@ typedef enum {
typedef struct vpx_svc_parameters {
unsigned int width; /**< width of current spatial layer */
unsigned int height; /**< height of current spatial layer */
int layer; /**< current layer number - 0 = base */
int spatial_layer; /**< current spatial layer number - 0 = base */
int temporal_layer; /**< current temporal layer number - 0 = base */
int flags; /**< encode frame flags */
int max_quantizer; /**< max quantizer for current layer */
int min_quantizer; /**< min quantizer for current layer */
......@@ -295,6 +297,11 @@ typedef struct vpx_svc_parameters {
int alt_fb_idx; /**< alt reference frame frame buffer index */
} vpx_svc_parameters_t;
typedef struct vpx_svc_layer_id {
int spatial_layer_id;
int temporal_layer_id;
} vpx_svc_layer_id_t;
/*!\brief VP8 encoder control function parameter type
*
* Defines the data types that VP8E control functions take. Note that
......@@ -316,6 +323,7 @@ VPX_CTRL_USE_TYPE(VP8E_SET_SCALEMODE, vpx_scaling_mode_t *)
VPX_CTRL_USE_TYPE(VP9E_SET_SVC, int)
VPX_CTRL_USE_TYPE(VP9E_SET_SVC_PARAMETERS, vpx_svc_parameters_t *)
VPX_CTRL_USE_TYPE(VP9E_SET_SVC_LAYER_ID, vpx_svc_layer_id_t *)
VPX_CTRL_USE_TYPE(VP8E_SET_CPUUSED, int)
VPX_CTRL_USE_TYPE(VP8E_SET_ENABLEAUTOALTREF, unsigned int)
......
......@@ -604,47 +604,48 @@ extern "C" {
* Spatial scalability settings (ss)
*/
/*!\brief Number of coding layers (spatial)
/*!\brief Number of spatial coding layers.
*
* This value specifies the number of coding layers to be used.
* This value specifies the number of spatial coding layers to be used.
*/
unsigned int ss_number_layers;
/*!\brief Number of coding layers
/*!\brief Number of temporal coding layers.
*
* This value specifies the number of coding layers to be used.
* This value specifies the number of temporal layers to be used.
*/
unsigned int ts_number_layers;
/*!\brief Target bitrate for each layer
/*!\brief Target bitrate for each temporal layer.
*
* These values specify the target coding bitrate for each coding layer.
* These values specify the target coding bitrate to be used for each
* temporal layer.
*/
unsigned int ts_target_bitrate[VPX_TS_MAX_LAYERS];
/*!\brief Frame rate decimation factor for each layer
/*!\brief Frame rate decimation factor for each temporal layer.
*
* These values specify the frame rate decimation factors to apply
* to each layer.
* to each temporal layer.
*/
unsigned int ts_rate_decimator[VPX_TS_MAX_LAYERS];
/*!\brief Length of the sequence defining frame layer membership
/*!\brief Length of the sequence defining frame temporal layer membership.
*
* This value specifies the length of the sequence that defines the
* membership of frames to layers. For example, if ts_periodicity=8 then
* frames are assigned to coding layers with a repeated sequence of
* length 8.
*/
* membership of frames to temporal layers. For example, if the
* ts_periodicity = 8, then the frames are assigned to coding layers with a
* repeated sequence of length 8.
*/
unsigned int ts_periodicity;
/*!\brief Template defining the membership of frames to coding layers
/*!\brief Template defining the membership of frames to temporal layers.
*
* This array defines the membership of frames to coding layers. For a
* 2-layer encoding that assigns even numbered frames to one layer (0)
* and odd numbered frames to a second layer (1) with ts_periodicity=8,
* then ts_layer_id = (0,1,0,1,0,1,0,1).
*/
* This array defines the membership of frames to temporal coding layers.
* For a 2-layer encoding that assigns even numbered frames to one temporal
* layer (0) and odd numbered frames to a second temporal layer (1) with
* ts_periodicity=8, then ts_layer_id = (0,1,0,1,0,1,0,1).
*/
unsigned int ts_layer_id[VPX_TS_MAX_PERIODICITY];
} vpx_codec_enc_cfg_t; /**< alias for struct vpx_codec_enc_cfg */
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment