// dump accumulated statistics and reset accumulated values
const char *vpx_svc_dump_statistics(SvcContext *svc_ctx) {
int number_of_frames, number_of_keyframes, encode_frame_count;
int i;
uint32_t bytes_total = 0;
SvcInternal *const si = get_svc_internal(svc_ctx);
if (svc_ctx == NULL || si == NULL) return NULL;
svc_log_reset(svc_ctx);
encode_frame_count = si->encode_frame_count;
if (si->encode_frame_count <= 0) return vpx_svc_get_message(svc_ctx);
svc_log(svc_ctx, SVC_LOG_INFO, "\n");
number_of_keyframes = encode_frame_count / si->kf_dist + 1;
for (i = 0; i < si->layers; ++i) {
number_of_frames = encode_frame_count;
if (svc_ctx->encoding_mode == ALT_INTER_LAYER_PREDICTION_IP &&
(i == 1 || i == 3)) {
number_of_frames -= number_of_keyframes;
}
svc_log(svc_ctx, SVC_LOG_INFO, "Layer %d PSNR=[%2.3f], Bytes=[%u]\n", i,
(double)si->psnr_in_layer[i] / number_of_frames,
si->bytes_in_layer[i]);
bytes_total += si->bytes_in_layer[i];
si->psnr_in_layer[i] = 0;
si->bytes_in_layer[i] = 0;
}
// only display statistics once
si->encode_frame_count = 0;
svc_log(svc_ctx, SVC_LOG_INFO, "Total Bytes=[%u]\n", bytes_total);
return vpx_svc_get_message(svc_ctx);
}
// dump accumulated statistics and reset accumulated values
const char *vpx_svc_dump_statistics(SvcContext *svc_ctx) {
int number_of_frames;
int i, j;
uint32_t bytes_total = 0;
double scale[COMPONENTS];
double psnr[COMPONENTS];
double mse[COMPONENTS];
double y_scale;
SvcInternal *const si = get_svc_internal(svc_ctx);
if (svc_ctx == NULL || si == NULL) return NULL;
svc_log_reset(svc_ctx);
number_of_frames = si->psnr_pkt_received;
if (number_of_frames <= 0) return vpx_svc_get_message(svc_ctx);
svc_log(svc_ctx, SVC_LOG_INFO, "\n");
for (i = 0; i < svc_ctx->spatial_layers; ++i) {
svc_log(svc_ctx, SVC_LOG_INFO,
"Layer %d Average PSNR=[%2.3f, %2.3f, %2.3f, %2.3f], Bytes=[%u]\n",
i, (double)si->psnr_sum[i][0] / number_of_frames,
(double)si->psnr_sum[i][1] / number_of_frames,
(double)si->psnr_sum[i][2] / number_of_frames,
(double)si->psnr_sum[i][3] / number_of_frames, si->bytes_sum[i]);
// the following psnr calculation is deduced from ffmpeg.c#print_report
y_scale = si->width * si->height * 255.0 * 255.0 * number_of_frames;
scale[1] = y_scale;
scale[2] = scale[3] = y_scale / 4; // U or V
scale[0] = y_scale * 1.5; // total
for (j = 0; j < COMPONENTS; j++) {
psnr[j] = calc_psnr(si->sse_sum[i][j] / scale[j]);
mse[j] = si->sse_sum[i][j] * 255.0 * 255.0 / scale[j];
}
svc_log(svc_ctx, SVC_LOG_INFO,
"Layer %d Overall PSNR=[%2.3f, %2.3f, %2.3f, %2.3f]\n", i, psnr[0],
psnr[1], psnr[2], psnr[3]);
svc_log(svc_ctx, SVC_LOG_INFO,
"Layer %d Overall MSE=[%2.3f, %2.3f, %2.3f, %2.3f]\n", i, mse[0],
mse[1], mse[2], mse[3]);
bytes_total += si->bytes_sum[i];
// clear sums for next time
si->bytes_sum[i] = 0;
for (j = 0; j < COMPONENTS; ++j) {
si->psnr_sum[i][j] = 0;
si->sse_sum[i][j] = 0;
}
}
// only display statistics once
si->psnr_pkt_received = 0;
svc_log(svc_ctx, SVC_LOG_INFO, "Total Bytes=[%u]\n", bytes_total);
return vpx_svc_get_message(svc_ctx);
}
static void set_svc_parameters(SvcContext *svc_ctx,
vpx_codec_ctx_t *codec_ctx) {
int layer;
vpx_svc_parameters_t svc_params;
SvcInternal *const si = get_svc_internal(svc_ctx);
memset(&svc_params, 0, sizeof(svc_params));
svc_params.temporal_layer = 0;
svc_params.spatial_layer = si->layer;
layer = si->layer;
if (VPX_CODEC_OK != vpx_svc_get_layer_resolution(svc_ctx, layer,
&svc_params.width,
&svc_params.height)) {
svc_log(svc_ctx, SVC_LOG_ERROR, "vpx_svc_get_layer_resolution failed\n");
}
if (codec_ctx->config.enc->g_pass == VPX_RC_ONE_PASS) {
svc_params.min_quantizer = si->quantizer[layer];
svc_params.max_quantizer = si->quantizer[layer];
} else {
svc_params.min_quantizer = codec_ctx->config.enc->rc_min_quantizer;
svc_params.max_quantizer = codec_ctx->config.enc->rc_max_quantizer;
}
vpx_codec_control(codec_ctx, VP9E_SET_SVC_PARAMETERS, &svc_params);
}
static vpx_codec_err_t parse_quantizer_values(SvcContext *svc_ctx,
const char *quantizer_values) {
char *input_string;
char *token;
const char *delim = ",";
char *save_ptr;
int found = 0;
int i, q;
int res = VPX_CODEC_OK;
SvcInternal *const si = get_svc_internal(svc_ctx);
if (quantizer_values == NULL || strlen(quantizer_values) == 0) {
input_string = strdup(DEFAULT_QUANTIZER_VALUES);
} else {
input_string = strdup(quantizer_values);
}
token = strtok_r(input_string, delim, &save_ptr);
for (i = 0; i < svc_ctx->spatial_layers; ++i) {
if (token != NULL) {
q = atoi(token);
if (q <= 0 || q > 100) {
svc_log(svc_ctx, SVC_LOG_ERROR,
"svc-quantizer-values: invalid value %s\n", token);
res = VPX_CODEC_INVALID_PARAM;
break;
}
token = strtok_r(NULL, delim, &save_ptr);
found = i + 1;
} else {
q = 0;
}
si->quantizer[i + VPX_SS_MAX_LAYERS - svc_ctx->spatial_layers] = q;
}
if (res == VPX_CODEC_OK && found != svc_ctx->spatial_layers) {
svc_log(svc_ctx, SVC_LOG_ERROR,
"svc: quantizers: %d values required, but only %d specified\n",
svc_ctx->spatial_layers, found);
res = VPX_CODEC_INVALID_PARAM;
}
free(input_string);
return res;
}
static vpx_codec_err_t parse_scale_factors(SvcContext *svc_ctx,
const char *scale_factors) {
char *input_string;
char *token;
const char *delim = ",";
char *save_ptr;
int found = 0;
int i;
int64_t num, den;
int res = VPX_CODEC_OK;
SvcInternal *const si = get_svc_internal(svc_ctx);
if (scale_factors == NULL || strlen(scale_factors) == 0) {
input_string = strdup(DEFAULT_SCALE_FACTORS);
} else {
input_string = strdup(scale_factors);
}
token = strtok_r(input_string, delim, &save_ptr);
for (i = 0; i < svc_ctx->spatial_layers; ++i) {
num = den = 0;
if (token != NULL) {
num = strtol(token, &token, 10);
if (num <= 0) {
log_invalid_scale_factor(svc_ctx, token);
res = VPX_CODEC_INVALID_PARAM;
break;
}
if (*token++ != '/') {
log_invalid_scale_factor(svc_ctx, token);
res = VPX_CODEC_INVALID_PARAM;
break;
}
den = strtol(token, &token, 10);
if (den <= 0) {
log_invalid_scale_factor(svc_ctx, token);
res = VPX_CODEC_INVALID_PARAM;
break;
}
token = strtok_r(NULL, delim, &save_ptr);
found = i + 1;
}
si->scaling_factor_num[i + VPX_SS_MAX_LAYERS - svc_ctx->spatial_layers] =
(int)num;
si->scaling_factor_den[i + VPX_SS_MAX_LAYERS - svc_ctx->spatial_layers] =
(int)den;
}
if (res == VPX_CODEC_OK && found != svc_ctx->spatial_layers) {
svc_log(svc_ctx, SVC_LOG_ERROR,
"svc: scale-factors: %d values required, but only %d specified\n",
svc_ctx->spatial_layers, found);
res = VPX_CODEC_INVALID_PARAM;
}
free(input_string);
return res;
}
/**
* Parse SVC encoding options
* Format: encoding-mode=<svc_mode>,layers=<layer_count>
* scale-factors=<n1>/<d1>,<n2>/<d2>,...
* quantizers=<q1>,<q2>,...
* svc_mode = [i|ip|alt_ip|gf]
*/
static vpx_codec_err_t parse_options(SvcContext *svc_ctx, const char *options) {
char *input_string;
char *option_name;
char *option_value;
char *input_ptr;
int res = VPX_CODEC_OK;
if (options == NULL) return VPX_CODEC_OK;
input_string = strdup(options);
// parse option name
option_name = strtok_r(input_string, "=", &input_ptr);
while (option_name != NULL) {
// parse option value
option_value = strtok_r(NULL, " ", &input_ptr);
if (option_value == NULL) {
svc_log(svc_ctx, SVC_LOG_ERROR, "option missing value: %s\n",
option_name);
res = VPX_CODEC_INVALID_PARAM;
break;
}
if (strcmp("encoding-mode", option_name) == 0) {
res = set_option_encoding_mode(svc_ctx, option_value);
if (res != VPX_CODEC_OK) break;
} else if (strcmp("layers", option_name) == 0) {
svc_ctx->spatial_layers = atoi(option_value);
} else if (strcmp("scale-factors", option_name) == 0) {
res = parse_scale_factors(svc_ctx, option_value);
if (res != VPX_CODEC_OK) break;
} else if (strcmp("quantizers", option_name) == 0) {
res = parse_quantizer_values(svc_ctx, option_value);
if (res != VPX_CODEC_OK) break;
} else {
svc_log(svc_ctx, SVC_LOG_ERROR, "invalid option: %s\n", option_name);
res = VPX_CODEC_INVALID_PARAM;
break;
}
option_name = strtok_r(NULL, "=", &input_ptr);
}
free(input_string);
return res;
}
static vpx_codec_err_t set_option_encoding_mode(SvcContext *svc_ctx,
const char *value_str) {
if (strcmp(value_str, "i") == 0) {
svc_ctx->encoding_mode = INTER_LAYER_PREDICTION_I;
} else if (strcmp(value_str, "alt-ip") == 0) {
svc_ctx->encoding_mode = ALT_INTER_LAYER_PREDICTION_IP;
} else if (strcmp(value_str, "ip") == 0) {
svc_ctx->encoding_mode = INTER_LAYER_PREDICTION_IP;
} else if (strcmp(value_str, "gf") == 0) {
svc_ctx->encoding_mode = USE_GOLDEN_FRAME;
} else {
svc_log(svc_ctx, SVC_LOG_ERROR, "invalid encoding mode: %s", value_str);
return VPX_CODEC_INVALID_PARAM;
}
return VPX_CODEC_OK;
}
static vpx_codec_err_t parse_layer_options_from_string(SvcContext *svc_ctx,
LAYER_OPTION_TYPE type,
const char *input,
int *option0,
int *option1) {
int i;
vpx_codec_err_t res = VPX_CODEC_OK;
char *input_string;
char *token;
const char *delim = ",";
char *save_ptr;
int num_layers = svc_ctx->spatial_layers;
if (type == BITRATE)
num_layers = svc_ctx->spatial_layers * svc_ctx->temporal_layers;
if (input == NULL || option0 == NULL ||
(option1 == NULL && type == SCALE_FACTOR))
return VPX_CODEC_INVALID_PARAM;
input_string = strdup(input);
token = strtok_r(input_string, delim, &save_ptr);
for (i = 0; i < num_layers; ++i) {
if (token != NULL) {
res = extract_option(type, token, option0 + i, option1 + i);
if (res != VPX_CODEC_OK) break;
token = strtok_r(NULL, delim, &save_ptr);
} else {
break;
}
}
if (res == VPX_CODEC_OK && i != num_layers) {
svc_log(svc_ctx, SVC_LOG_ERROR,
"svc: layer params type: %d %d values required, "
"but only %d specified\n",
type, num_layers, i);
res = VPX_CODEC_INVALID_PARAM;
}
free(input_string);
return res;
}
/**
* Parse SVC encoding options
* Format: encoding-mode=<svc_mode>,layers=<layer_count>
* scale-factors=<n1>/<d1>,<n2>/<d2>,...
* quantizers=<q1>,<q2>,...
* svc_mode = [i|ip|alt_ip|gf]
*/
static vpx_codec_err_t parse_options(SvcContext *svc_ctx, const char *options) {
char *input_string;
char *option_name;
char *option_value;
char *input_ptr = NULL;
SvcInternal_t *const si = get_svc_internal(svc_ctx);
vpx_codec_err_t res = VPX_CODEC_OK;
int i, alt_ref_enabled = 0;
if (options == NULL) return VPX_CODEC_OK;
input_string = strdup(options);
// parse option name
option_name = strtok_r(input_string, "=", &input_ptr);
while (option_name != NULL) {
// parse option value
option_value = strtok_r(NULL, " ", &input_ptr);
if (option_value == NULL) {
svc_log(svc_ctx, SVC_LOG_ERROR, "option missing value: %s\n",
option_name);
res = VPX_CODEC_INVALID_PARAM;
break;
}
if (strcmp("spatial-layers", option_name) == 0) {
svc_ctx->spatial_layers = atoi(option_value);
} else if (strcmp("temporal-layers", option_name) == 0) {
svc_ctx->temporal_layers = atoi(option_value);
} else if (strcmp("scale-factors", option_name) == 0) {
res = parse_layer_options_from_string(svc_ctx, SCALE_FACTOR, option_value,
si->svc_params.scaling_factor_num,
si->svc_params.scaling_factor_den);
if (res != VPX_CODEC_OK) break;
} else if (strcmp("max-quantizers", option_name) == 0) {
res =
parse_layer_options_from_string(svc_ctx, QUANTIZER, option_value,
si->svc_params.max_quantizers, NULL);
if (res != VPX_CODEC_OK) break;
} else if (strcmp("min-quantizers", option_name) == 0) {
res =
parse_layer_options_from_string(svc_ctx, QUANTIZER, option_value,
si->svc_params.min_quantizers, NULL);
if (res != VPX_CODEC_OK) break;
} else if (strcmp("auto-alt-refs", option_name) == 0) {
res = parse_layer_options_from_string(svc_ctx, AUTO_ALT_REF, option_value,
si->enable_auto_alt_ref, NULL);
if (res != VPX_CODEC_OK) break;
} else if (strcmp("bitrates", option_name) == 0) {
res = parse_layer_options_from_string(svc_ctx, BITRATE, option_value,
si->bitrates, NULL);
if (res != VPX_CODEC_OK) break;
} else if (strcmp("multi-frame-contexts", option_name) == 0) {
si->use_multiple_frame_contexts = atoi(option_value);
} else {
svc_log(svc_ctx, SVC_LOG_ERROR, "invalid option: %s\n", option_name);
res = VPX_CODEC_INVALID_PARAM;
break;
}
option_name = strtok_r(NULL, "=", &input_ptr);
}
free(input_string);
for (i = 0; i < svc_ctx->spatial_layers; ++i) {
if (si->svc_params.max_quantizers[i] > MAX_QUANTIZER ||
si->svc_params.max_quantizers[i] < 0 ||
si->svc_params.min_quantizers[i] > si->svc_params.max_quantizers[i] ||
si->svc_params.min_quantizers[i] < 0)
res = VPX_CODEC_INVALID_PARAM;
}
if (si->use_multiple_frame_contexts &&
(svc_ctx->spatial_layers > 3 ||
svc_ctx->spatial_layers * svc_ctx->temporal_layers > 4))
res = VPX_CODEC_INVALID_PARAM;
for (i = 0; i < svc_ctx->spatial_layers; ++i)
alt_ref_enabled += si->enable_auto_alt_ref[i];
if (alt_ref_enabled > REF_FRAMES - svc_ctx->spatial_layers) {
svc_log(svc_ctx, SVC_LOG_ERROR,
"svc: auto alt ref: Maxinum %d(REF_FRAMES - layers) layers could"
"enabled auto alt reference frame, but % layers are enabled\n",
REF_FRAMES - svc_ctx->spatial_layers, alt_ref_enabled);
res = VPX_CODEC_INVALID_PARAM;
}
return res;
}
/**
* Encode a frame into multiple layers
* Create a superframe containing the individual layers
*/
vpx_codec_err_t vpx_svc_encode(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
struct vpx_image *rawimg, vpx_codec_pts_t pts,
int64_t duration, int deadline) {
vpx_codec_err_t res;
vpx_codec_iter_t iter;
const vpx_codec_cx_pkt_t *cx_pkt;
struct LayerData *cx_layer_list = NULL;
struct LayerData *layer_data;
struct Superframe superframe;
SvcInternal *const si = get_svc_internal(svc_ctx);
if (svc_ctx == NULL || codec_ctx == NULL || rawimg == NULL || si == NULL) {
return VPX_CODEC_INVALID_PARAM;
}
memset(&superframe, 0, sizeof(superframe));
svc_log_reset(svc_ctx);
si->layers = svc_ctx->spatial_layers;
if (si->frame_within_gop >= si->kf_dist ||
si->encode_frame_count == 0) {
si->frame_within_gop = 0;
}
si->is_keyframe = (si->frame_within_gop == 0);
si->frame_size = 0;
svc_log(svc_ctx, SVC_LOG_DEBUG,
"vpx_svc_encode layers: %d, frame_count: %d, frame_within_gop: %d\n",
si->layers, si->encode_frame_count, si->frame_within_gop);
// encode each layer
for (si->layer = 0; si->layer < si->layers; ++si->layer) {
if (svc_ctx->encoding_mode == ALT_INTER_LAYER_PREDICTION_IP &&
si->is_keyframe && (si->layer == 1 || si->layer == 3)) {
svc_log(svc_ctx, SVC_LOG_DEBUG, "Skip encoding layer %d\n", si->layer);
continue;
}
calculate_enc_frame_flags(svc_ctx);
set_svc_parameters(svc_ctx, codec_ctx);
res = vpx_codec_encode(codec_ctx, rawimg, pts, (uint32_t)duration,
si->enc_frame_flags, deadline);
if (res != VPX_CODEC_OK) {
return res;
}
// save compressed data
iter = NULL;
while ((cx_pkt = vpx_codec_get_cx_data(codec_ctx, &iter))) {
switch (cx_pkt->kind) {
case VPX_CODEC_CX_FRAME_PKT: {
const uint32_t frame_pkt_size = (uint32_t)(cx_pkt->data.frame.sz);
si->bytes_in_layer[si->layer] += frame_pkt_size;
svc_log(svc_ctx, SVC_LOG_DEBUG,
"SVC frame: %d, layer: %d, size: %u\n",
si->encode_frame_count, si->layer, frame_pkt_size);
layer_data =
ld_create(cx_pkt->data.frame.buf, (size_t)frame_pkt_size);
if (layer_data == NULL) {
svc_log(svc_ctx, SVC_LOG_ERROR, "Error allocating LayerData\n");
return 0;
}
ld_list_add(&cx_layer_list, layer_data);
// save layer size in superframe index
superframe.sizes[superframe.count++] = frame_pkt_size;
superframe.magnitude |= frame_pkt_size;
break;
}
case VPX_CODEC_PSNR_PKT: {
svc_log(svc_ctx, SVC_LOG_DEBUG,
"SVC frame: %d, layer: %d, PSNR(Total/Y/U/V): "
"%2.3f %2.3f %2.3f %2.3f \n",
si->encode_frame_count, si->layer,
cx_pkt->data.psnr.psnr[0], cx_pkt->data.psnr.psnr[1],
cx_pkt->data.psnr.psnr[2], cx_pkt->data.psnr.psnr[3]);
si->psnr_in_layer[si->layer] += cx_pkt->data.psnr.psnr[0];
break;
}
default: {
break;
}
}
}
}
// add superframe index to layer data list
sf_create_index(&superframe);
layer_data = ld_create(superframe.buffer, superframe.index_size);
ld_list_add(&cx_layer_list, layer_data);
// get accumulated size of layer data
si->frame_size = ld_list_get_buffer_size(cx_layer_list);
if (si->frame_size == 0) return VPX_CODEC_ERROR;
// all layers encoded, create single buffer with concatenated layers
if (si->frame_size > si->buffer_size) {
free(si->buffer);
si->buffer = malloc(si->frame_size);
if (si->buffer == NULL) {
ld_list_free(cx_layer_list);
return VPX_CODEC_MEM_ERROR;
}
si->buffer_size = si->frame_size;
}
// copy layer data into packet
ld_list_copy_to_buffer(cx_layer_list, si->buffer);
ld_list_free(cx_layer_list);
svc_log(svc_ctx, SVC_LOG_DEBUG, "SVC frame: %d, kf: %d, size: %d, pts: %d\n",
si->encode_frame_count, si->is_keyframe, (int)si->frame_size,
(int)pts);
++si->frame_within_gop;
++si->encode_frame_count;
return VPX_CODEC_OK;
}
static void set_svc_parameters(SvcContext *svc_ctx,
vpx_codec_ctx_t *codec_ctx) {
int layer, layer_index;
vpx_svc_parameters_t svc_params;
SvcInternal *const si = get_svc_internal(svc_ctx);
memset(&svc_params, 0, sizeof(svc_params));
svc_params.layer = si->layer;
svc_params.flags = si->enc_frame_flags;
layer = si->layer;
if (svc_ctx->encoding_mode == ALT_INTER_LAYER_PREDICTION_IP &&
si->frame_within_gop == 0) {
// layers 1 & 3 don't exist in this mode, use the higher one
if (layer == 0 || layer == 2) {
layer += 1;
}
}
if (VPX_CODEC_OK != vpx_svc_get_layer_resolution(svc_ctx, layer,
&svc_params.width,
&svc_params.height)) {
svc_log(svc_ctx, SVC_LOG_ERROR, "vpx_svc_get_layer_resolution failed\n");
}
layer_index = layer + VPX_SS_MAX_LAYERS - si->layers;
svc_params.min_quantizer = si->quantizer[layer_index];
svc_params.max_quantizer = si->quantizer[layer_index];
svc_params.distance_from_i_frame = si->frame_within_gop;
// Use buffer i for layer i LST
svc_params.lst_fb_idx = si->layer;
// Use buffer i-1 for layer i Alt (Inter-layer prediction)
if (si->layer != 0) {
const int use_higher_layer =
svc_ctx->encoding_mode == ALT_INTER_LAYER_PREDICTION_IP &&
si->frame_within_gop == 0;
svc_params.alt_fb_idx = use_higher_layer ? si->layer - 2 : si->layer - 1;
}
if (svc_ctx->encoding_mode == ALT_INTER_LAYER_PREDICTION_IP) {
svc_params.gld_fb_idx = si->layer + 1;
} else {
if (si->layer < 2 * si->layers - SVC_REFERENCE_FRAMES)
svc_params.gld_fb_idx = svc_params.lst_fb_idx;
else
svc_params.gld_fb_idx = 2 * si->layers - 1 - si->layer;
}
svc_log(svc_ctx, SVC_LOG_DEBUG, "SVC frame: %d, layer: %d, %dx%d, q: %d\n",
si->encode_frame_count, si->layer, svc_params.width,
svc_params.height, svc_params.min_quantizer);
if (svc_params.flags == VPX_EFLAG_FORCE_KF) {
svc_log(svc_ctx, SVC_LOG_DEBUG, "flags == VPX_EFLAG_FORCE_KF\n");
} else {
svc_log(
svc_ctx, SVC_LOG_DEBUG, "Using: LST/GLD/ALT [%2d|%2d|%2d]\n",
svc_params.flags & VP8_EFLAG_NO_REF_LAST ? -1 : svc_params.lst_fb_idx,
svc_params.flags & VP8_EFLAG_NO_REF_GF ? -1 : svc_params.gld_fb_idx,
svc_params.flags & VP8_EFLAG_NO_REF_ARF ? -1 : svc_params.alt_fb_idx);
svc_log(
svc_ctx, SVC_LOG_DEBUG, "Updating: LST/GLD/ALT [%2d|%2d|%2d]\n",
svc_params.flags & VP8_EFLAG_NO_UPD_LAST ? -1 : svc_params.lst_fb_idx,
svc_params.flags & VP8_EFLAG_NO_UPD_GF ? -1 : svc_params.gld_fb_idx,
svc_params.flags & VP8_EFLAG_NO_UPD_ARF ? -1 : svc_params.alt_fb_idx);
}
vpx_codec_control(codec_ctx, VP9E_SET_SVC_PARAMETERS, &svc_params);
}
static void calculate_enc_frame_flags(SvcContext *svc_ctx) {
vpx_enc_frame_flags_t flags = VPX_EFLAG_FORCE_KF;
SvcInternal *const si = get_svc_internal(svc_ctx);
const int is_keyframe = (si->frame_within_gop == 0);
// keyframe layer zero is identical for all modes
if (is_keyframe && si->layer == 0) {
si->enc_frame_flags = VPX_EFLAG_FORCE_KF;
return;
}
switch (svc_ctx->encoding_mode) {
case ALT_INTER_LAYER_PREDICTION_IP:
if (si->layer == 0) {
flags = map_vp8_flags(USE_LAST | UPDATE_LAST);
} else if (is_keyframe) {
if (si->layer == si->layers - 1) {
flags = map_vp8_flags(USE_ARF | UPDATE_LAST);
} else {
flags = map_vp8_flags(USE_ARF | UPDATE_LAST | UPDATE_GF);
}
} else {
flags = map_vp8_flags(USE_LAST | USE_ARF | UPDATE_LAST);
}
break;
case INTER_LAYER_PREDICTION_I:
if (si->layer == 0) {
flags = map_vp8_flags(USE_LAST | UPDATE_LAST);
} else if (is_keyframe) {
flags = map_vp8_flags(USE_ARF | UPDATE_LAST);
} else {
flags = map_vp8_flags(USE_LAST | UPDATE_LAST);
}
break;
case INTER_LAYER_PREDICTION_IP:
if (si->layer == 0) {
flags = map_vp8_flags(USE_LAST | UPDATE_LAST);
} else if (is_keyframe) {
flags = map_vp8_flags(USE_ARF | UPDATE_LAST);
} else {
flags = map_vp8_flags(USE_LAST | USE_ARF | UPDATE_LAST);
}
break;
case USE_GOLDEN_FRAME:
if (2 * si->layers - SVC_REFERENCE_FRAMES <= si->layer) {
if (si->layer == 0) {
flags = map_vp8_flags(USE_LAST | USE_GF | UPDATE_LAST);
} else if (is_keyframe) {
flags = map_vp8_flags(USE_ARF | UPDATE_LAST | UPDATE_GF);
} else {
flags = map_vp8_flags(USE_LAST | USE_ARF | USE_GF | UPDATE_LAST);
}
} else {
if (si->layer == 0) {
flags = map_vp8_flags(USE_LAST | UPDATE_LAST);
} else if (is_keyframe) {
flags = map_vp8_flags(USE_ARF | UPDATE_LAST);
} else {
flags = map_vp8_flags(USE_LAST | UPDATE_LAST);
}
}
break;
default:
svc_log(svc_ctx, SVC_LOG_ERROR, "unexpected encoding mode: %d\n",
svc_ctx->encoding_mode);
break;
}
si->enc_frame_flags = flags;
}
static void log_invalid_scale_factor(SvcContext *svc_ctx, const char *value) {
svc_log(svc_ctx, SVC_LOG_ERROR, "svc scale-factors: invalid value %s\n",
value);
}
vpx_codec_err_t vpx_svc_init(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
vpx_codec_iface_t *iface,
vpx_codec_enc_cfg_t *enc_cfg) {
vpx_codec_err_t res;
int i, sl, tl;
SvcInternal_t *const si = get_svc_internal(svc_ctx);
if (svc_ctx == NULL || codec_ctx == NULL || iface == NULL ||
enc_cfg == NULL) {
return VPX_CODEC_INVALID_PARAM;
}
if (si == NULL) return VPX_CODEC_MEM_ERROR;
si->codec_ctx = codec_ctx;
si->width = enc_cfg->g_w;
si->height = enc_cfg->g_h;
si->kf_dist = enc_cfg->kf_max_dist;
if (svc_ctx->spatial_layers == 0)
svc_ctx->spatial_layers = VPX_SS_DEFAULT_LAYERS;
if (svc_ctx->spatial_layers < 1 ||
svc_ctx->spatial_layers > VPX_SS_MAX_LAYERS) {
svc_log(svc_ctx, SVC_LOG_ERROR, "spatial layers: invalid value: %d\n",
svc_ctx->spatial_layers);
return VPX_CODEC_INVALID_PARAM;
}
// Note: temporal_layering_mode only applies to one-pass CBR
// si->svc_params.temporal_layering_mode = svc_ctx->temporal_layering_mode;
if (svc_ctx->temporal_layering_mode == 3) {
svc_ctx->temporal_layers = 3;
} else if (svc_ctx->temporal_layering_mode == 2 ||
svc_ctx->temporal_layering_mode == 1) {
svc_ctx->temporal_layers = 2;
}
for (sl = 0; sl < VPX_SS_MAX_LAYERS; ++sl) {
si->svc_params.scaling_factor_num[sl] = DEFAULT_SCALE_FACTORS_NUM[sl];
si->svc_params.scaling_factor_den[sl] = DEFAULT_SCALE_FACTORS_DEN[sl];
si->svc_params.speed_per_layer[sl] = svc_ctx->speed;
}
if (enc_cfg->rc_end_usage == VPX_CBR && enc_cfg->g_pass == VPX_RC_ONE_PASS &&
svc_ctx->spatial_layers <= 3) {
for (sl = 0; sl < svc_ctx->spatial_layers; ++sl) {
int sl2 = (svc_ctx->spatial_layers == 2) ? sl + 1 : sl;
si->svc_params.scaling_factor_num[sl] = DEFAULT_SCALE_FACTORS_NUM_2x[sl2];
si->svc_params.scaling_factor_den[sl] = DEFAULT_SCALE_FACTORS_DEN_2x[sl2];
}
if (svc_ctx->spatial_layers == 1) {
si->svc_params.scaling_factor_num[0] = 1;
si->svc_params.scaling_factor_den[0] = 1;
}
}
for (tl = 0; tl < svc_ctx->temporal_layers; ++tl) {
for (sl = 0; sl < svc_ctx->spatial_layers; ++sl) {
i = sl * svc_ctx->temporal_layers + tl;
si->svc_params.max_quantizers[i] = MAX_QUANTIZER;
si->svc_params.min_quantizers[i] = 0;
if (enc_cfg->rc_end_usage == VPX_CBR &&
enc_cfg->g_pass == VPX_RC_ONE_PASS) {
si->svc_params.max_quantizers[i] = 56;
si->svc_params.min_quantizers[i] = 2;
}
}
}
// Parse aggregate command line options. Options must start with
// "layers=xx" then followed by other options
res = parse_options(svc_ctx, si->options);
if (res != VPX_CODEC_OK) return res;
if (svc_ctx->spatial_layers < 1) svc_ctx->spatial_layers = 1;
if (svc_ctx->spatial_layers > VPX_SS_MAX_LAYERS)
svc_ctx->spatial_layers = VPX_SS_MAX_LAYERS;
if (svc_ctx->temporal_layers < 1) svc_ctx->temporal_layers = 1;
if (svc_ctx->temporal_layers > VPX_TS_MAX_LAYERS)
svc_ctx->temporal_layers = VPX_TS_MAX_LAYERS;
if (svc_ctx->temporal_layers * svc_ctx->spatial_layers > VPX_MAX_LAYERS) {
svc_log(svc_ctx, SVC_LOG_ERROR,
"spatial layers * temporal layers exceeds the maximum number of "
"allowed layers of %d\n",
svc_ctx->spatial_layers * svc_ctx->temporal_layers,
(int)VPX_MAX_LAYERS);
return VPX_CODEC_INVALID_PARAM;
}
res = assign_layer_bitrates(svc_ctx, enc_cfg);
if (res != VPX_CODEC_OK) {
svc_log(svc_ctx, SVC_LOG_ERROR,
"layer bitrates incorrect: \n"
"1) spatial layer bitrates should sum up to target \n"
"2) temporal layer bitrates should be increasing within \n"
"a spatial layer \n");
return VPX_CODEC_INVALID_PARAM;
}
#if CONFIG_SPATIAL_SVC
for (i = 0; i < svc_ctx->spatial_layers; ++i)
enc_cfg->ss_enable_auto_alt_ref[i] = si->enable_auto_alt_ref[i];
#endif
if (svc_ctx->temporal_layers > 1) {
int i;
for (i = 0; i < svc_ctx->temporal_layers; ++i) {
enc_cfg->ts_target_bitrate[i] =
enc_cfg->rc_target_bitrate / svc_ctx->temporal_layers;
enc_cfg->ts_rate_decimator[i] = 1 << (svc_ctx->temporal_layers - 1 - i);
}
}
if (svc_ctx->threads) enc_cfg->g_threads = svc_ctx->threads;
// Modify encoder configuration
enc_cfg->ss_number_layers = svc_ctx->spatial_layers;
enc_cfg->ts_number_layers = svc_ctx->temporal_layers;
if (enc_cfg->rc_end_usage == VPX_CBR) {
enc_cfg->rc_resize_allowed = 0;
enc_cfg->rc_min_quantizer = 2;
enc_cfg->rc_max_quantizer = 56;
enc_cfg->rc_undershoot_pct = 50;
enc_cfg->rc_overshoot_pct = 50;
enc_cfg->rc_buf_initial_sz = 500;
enc_cfg->rc_buf_optimal_sz = 600;
enc_cfg->rc_buf_sz = 1000;
enc_cfg->rc_dropframe_thresh = 0;
}
if (enc_cfg->g_error_resilient == 0 && si->use_multiple_frame_contexts == 0)
enc_cfg->g_error_resilient = 1;
// Initialize codec
res = vpx_codec_enc_init(codec_ctx, iface, enc_cfg, VPX_CODEC_USE_PSNR);
if (res != VPX_CODEC_OK) {
svc_log(svc_ctx, SVC_LOG_ERROR, "svc_enc_init error\n");
return res;
}
if (svc_ctx->spatial_layers > 1 || svc_ctx->temporal_layers > 1) {
vpx_codec_control(codec_ctx, VP9E_SET_SVC, 1);
vpx_codec_control(codec_ctx, VP9E_SET_SVC_PARAMETERS, &si->svc_params);
}
return VPX_CODEC_OK;
}
vpx_codec_err_t vpx_svc_init(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
vpx_codec_iface_t *iface,
vpx_codec_enc_cfg_t *enc_cfg) {
int max_intra_size_pct;
vpx_codec_err_t res;
SvcInternal *const si = get_svc_internal(svc_ctx);
if (svc_ctx == NULL || codec_ctx == NULL || iface == NULL ||
enc_cfg == NULL) {
return VPX_CODEC_INVALID_PARAM;
}
if (si == NULL) return VPX_CODEC_MEM_ERROR;
si->codec_ctx = codec_ctx;
si->width = enc_cfg->g_w;
si->height = enc_cfg->g_h;
if (enc_cfg->kf_max_dist < 2) {
svc_log(svc_ctx, SVC_LOG_ERROR, "key frame distance too small: %d\n",
enc_cfg->kf_max_dist);
return VPX_CODEC_INVALID_PARAM;
}
si->kf_dist = enc_cfg->kf_max_dist;
if (svc_ctx->spatial_layers == 0)
svc_ctx->spatial_layers = VPX_SS_DEFAULT_LAYERS;
if (svc_ctx->spatial_layers < 1 ||
svc_ctx->spatial_layers > VPX_SS_MAX_LAYERS) {
svc_log(svc_ctx, SVC_LOG_ERROR, "spatial layers: invalid value: %d\n",
svc_ctx->spatial_layers);
return VPX_CODEC_INVALID_PARAM;
}
// use SvcInternal value for number of layers to enable forcing single layer
// for first frame
si->layers = svc_ctx->spatial_layers;
res = parse_quantizer_values(svc_ctx, si->quantizers);
if (res != VPX_CODEC_OK) return res;
res = parse_scale_factors(svc_ctx, si->scale_factors);
if (res != VPX_CODEC_OK) return res;
// parse aggregate command line options
res = parse_options(svc_ctx, si->options);
if (res != VPX_CODEC_OK) return res;
// modify encoder configuration
enc_cfg->ss_number_layers = si->layers;
enc_cfg->kf_mode = VPX_KF_DISABLED;
enc_cfg->g_pass = VPX_RC_ONE_PASS;
// Lag in frames not currently supported
enc_cfg->g_lag_in_frames = 0;
// TODO(ivanmaltz): determine if these values need to be set explicitly for
// svc, or if the normal default/override mechanism can be used
enc_cfg->rc_dropframe_thresh = 0;
enc_cfg->rc_end_usage = VPX_CBR;
enc_cfg->rc_resize_allowed = 0;
enc_cfg->rc_min_quantizer = 33;
enc_cfg->rc_max_quantizer = 33;
enc_cfg->rc_undershoot_pct = 100;
enc_cfg->rc_overshoot_pct = 15;
enc_cfg->rc_buf_initial_sz = 500;
enc_cfg->rc_buf_optimal_sz = 600;
enc_cfg->rc_buf_sz = 1000;
enc_cfg->g_error_resilient = 1;
// Initialize codec
res = vpx_codec_enc_init(codec_ctx, iface, enc_cfg, VPX_CODEC_USE_PSNR);
if (res != VPX_CODEC_OK) {
svc_log(svc_ctx, SVC_LOG_ERROR, "svc_enc_init error\n");
return res;
}
vpx_codec_control(codec_ctx, VP9E_SET_SVC, 1);
vpx_codec_control(codec_ctx, VP8E_SET_CPUUSED, 1);
vpx_codec_control(codec_ctx, VP8E_SET_STATIC_THRESHOLD, 1);
vpx_codec_control(codec_ctx, VP8E_SET_NOISE_SENSITIVITY, 1);
vpx_codec_control(codec_ctx, VP8E_SET_TOKEN_PARTITIONS, 1);
max_intra_size_pct =
(int)(((double)enc_cfg->rc_buf_optimal_sz * 0.5) *
((double)enc_cfg->g_timebase.den / enc_cfg->g_timebase.num) / 10.0);
vpx_codec_control(codec_ctx, VP8E_SET_MAX_INTRA_BITRATE_PCT,
max_intra_size_pct);
return VPX_CODEC_OK;
}
/**
* Encode a frame into multiple layers
* Create a superframe containing the individual layers
*/
vpx_codec_err_t vpx_svc_encode(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
struct vpx_image *rawimg, vpx_codec_pts_t pts,
int64_t duration, int deadline) {
vpx_codec_err_t res;
vpx_codec_iter_t iter;
const vpx_codec_cx_pkt_t *cx_pkt;
SvcInternal_t *const si = get_svc_internal(svc_ctx);
if (svc_ctx == NULL || codec_ctx == NULL || si == NULL) {
return VPX_CODEC_INVALID_PARAM;
}
svc_log_reset(svc_ctx);
res =
vpx_codec_encode(codec_ctx, rawimg, pts, (uint32_t)duration, 0, deadline);
if (res != VPX_CODEC_OK) {
return res;
}
// save compressed data
iter = NULL;
while ((cx_pkt = vpx_codec_get_cx_data(codec_ctx, &iter))) {
switch (cx_pkt->kind) {
#if CONFIG_SPATIAL_SVC && defined(VPX_TEST_SPATIAL_SVC)
case VPX_CODEC_SPATIAL_SVC_LAYER_PSNR: {
int i;
for (i = 0; i < svc_ctx->spatial_layers; ++i) {
int j;
svc_log(svc_ctx, SVC_LOG_DEBUG,
"SVC frame: %d, layer: %d, PSNR(Total/Y/U/V): "
"%2.3f %2.3f %2.3f %2.3f \n",
si->psnr_pkt_received, i, cx_pkt->data.layer_psnr[i].psnr[0],
cx_pkt->data.layer_psnr[i].psnr[1],
cx_pkt->data.layer_psnr[i].psnr[2],
cx_pkt->data.layer_psnr[i].psnr[3]);
svc_log(svc_ctx, SVC_LOG_DEBUG,
"SVC frame: %d, layer: %d, SSE(Total/Y/U/V): "
"%2.3f %2.3f %2.3f %2.3f \n",
si->psnr_pkt_received, i, cx_pkt->data.layer_psnr[i].sse[0],
cx_pkt->data.layer_psnr[i].sse[1],
cx_pkt->data.layer_psnr[i].sse[2],
cx_pkt->data.layer_psnr[i].sse[3]);
for (j = 0; j < COMPONENTS; ++j) {
si->psnr_sum[i][j] += cx_pkt->data.layer_psnr[i].psnr[j];
si->sse_sum[i][j] += cx_pkt->data.layer_psnr[i].sse[j];
}
}
++si->psnr_pkt_received;
break;
}
case VPX_CODEC_SPATIAL_SVC_LAYER_SIZES: {
int i;
for (i = 0; i < svc_ctx->spatial_layers; ++i)
si->bytes_sum[i] += cx_pkt->data.layer_sizes[i];
break;
}
#endif
case VPX_CODEC_PSNR_PKT: {
#if CONFIG_SPATIAL_SVC && defined(VPX_TEST_SPATIAL_SVC)
int j;
svc_log(svc_ctx, SVC_LOG_DEBUG,
"frame: %d, layer: %d, PSNR(Total/Y/U/V): "
"%2.3f %2.3f %2.3f %2.3f \n",
si->psnr_pkt_received, 0, cx_pkt->data.layer_psnr[0].psnr[0],
cx_pkt->data.layer_psnr[0].psnr[1],
cx_pkt->data.layer_psnr[0].psnr[2],
cx_pkt->data.layer_psnr[0].psnr[3]);
for (j = 0; j < COMPONENTS; ++j) {
si->psnr_sum[0][j] += cx_pkt->data.layer_psnr[0].psnr[j];
si->sse_sum[0][j] += cx_pkt->data.layer_psnr[0].sse[j];
}
#endif
}
++si->psnr_pkt_received;
break;
default: { break; }
}
}
return VPX_CODEC_OK;
}
vpx_codec_err_t vpx_svc_init(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
vpx_codec_iface_t *iface,
vpx_codec_enc_cfg_t *enc_cfg) {
vpx_codec_err_t res;
int i;
SvcInternal *const si = get_svc_internal(svc_ctx);
if (svc_ctx == NULL || codec_ctx == NULL || iface == NULL ||
enc_cfg == NULL) {
return VPX_CODEC_INVALID_PARAM;
}
if (si == NULL) return VPX_CODEC_MEM_ERROR;
si->codec_ctx = codec_ctx;
si->width = enc_cfg->g_w;
si->height = enc_cfg->g_h;
if (enc_cfg->kf_max_dist < 2) {
svc_log(svc_ctx, SVC_LOG_ERROR, "key frame distance too small: %d\n",
enc_cfg->kf_max_dist);
return VPX_CODEC_INVALID_PARAM;
}
si->kf_dist = enc_cfg->kf_max_dist;
if (svc_ctx->spatial_layers == 0)
svc_ctx->spatial_layers = VPX_SS_DEFAULT_LAYERS;
if (svc_ctx->spatial_layers < 1 ||
svc_ctx->spatial_layers > VPX_SS_MAX_LAYERS) {
svc_log(svc_ctx, SVC_LOG_ERROR, "spatial layers: invalid value: %d\n",
svc_ctx->spatial_layers);
return VPX_CODEC_INVALID_PARAM;
}
for (i = 0; i < VPX_SS_MAX_LAYERS; ++i) {
si->quantizer[i] = DEFAULT_QUANTIZER_VALUES[i];
si->scaling_factor_num[i] = DEFAULT_SCALE_FACTORS_NUM[i];
si->scaling_factor_den[i] = DEFAULT_SCALE_FACTORS_DEN[i];
}
if (strlen(si->quantizers) > 0) {
res = parse_layer_options_from_string(svc_ctx, QUANTIZER, si->quantizers,
si->quantizer, NULL);
if (res != VPX_CODEC_OK)
return res;
}
if (strlen(si->scale_factors) > 0) {
res = parse_layer_options_from_string(svc_ctx, SCALE_FACTOR,
si->scale_factors,
si->scaling_factor_num,
si->scaling_factor_den);
if (res != VPX_CODEC_OK)
return res;
}
// Parse aggregate command line options. Options must start with
// "layers=xx" then followed by other options
res = parse_options(svc_ctx, si->options);
if (res != VPX_CODEC_OK) return res;
if (svc_ctx->spatial_layers < 1)
svc_ctx->spatial_layers = 1;
if (svc_ctx->spatial_layers > VPX_SS_MAX_LAYERS)
svc_ctx->spatial_layers = VPX_SS_MAX_LAYERS;
if (svc_ctx->temporal_layers < 1)
svc_ctx->temporal_layers = 1;
if (svc_ctx->temporal_layers > VPX_TS_MAX_LAYERS)
svc_ctx->temporal_layers = VPX_TS_MAX_LAYERS;
assign_layer_bitrates(svc_ctx, enc_cfg);
#if CONFIG_SPATIAL_SVC
for (i = 0; i < svc_ctx->spatial_layers; ++i)
enc_cfg->ss_enable_auto_alt_ref[i] = si->enable_auto_alt_ref[i];
#endif
if (svc_ctx->temporal_layers > 1) {
int i;
for (i = 0; i < svc_ctx->temporal_layers; ++i) {
enc_cfg->ts_target_bitrate[i] = enc_cfg->rc_target_bitrate /
svc_ctx->temporal_layers;
enc_cfg->ts_rate_decimator[i] = 1 << (svc_ctx->temporal_layers - 1 - i);
}
}
// modify encoder configuration
enc_cfg->ss_number_layers = svc_ctx->spatial_layers;
enc_cfg->ts_number_layers = svc_ctx->temporal_layers;
if (enc_cfg->g_error_resilient == 0 && si->use_multiple_frame_contexts == 0)
enc_cfg->g_error_resilient = 1;
// Initialize codec
res = vpx_codec_enc_init(codec_ctx, iface, enc_cfg, VPX_CODEC_USE_PSNR);
if (res != VPX_CODEC_OK) {
svc_log(svc_ctx, SVC_LOG_ERROR, "svc_enc_init error\n");
return res;
}
vpx_codec_control(codec_ctx, VP9E_SET_SVC, 1);
return VPX_CODEC_OK;
}
