1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vpx/src/svc_encodeframe.c Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,983 @@
1.4 +/*
1.5 + * Copyright (c) 2013 The WebM project authors. All Rights Reserved.
1.6 + *
1.7 + * Use of this source code is governed by a BSD-style license
1.8 + * that can be found in the LICENSE file in the root of the source
1.9 + * tree. An additional intellectual property rights grant can be found
1.10 + * in the file PATENTS. All contributing project authors may
1.11 + * be found in the AUTHORS file in the root of the source tree.
1.12 + */
1.13 +
1.14 +/**
1.15 + * @file
1.16 + * VP9 SVC encoding support via libvpx
1.17 + */
1.18 +
1.19 +#include <stdarg.h>
1.20 +#include <stdio.h>
1.21 +#include <stdlib.h>
1.22 +#include <string.h>
1.23 +#define VPX_DISABLE_CTRL_TYPECHECKS 1
1.24 +#define VPX_CODEC_DISABLE_COMPAT 1
1.25 +#include "vpx/svc_context.h"
1.26 +#include "vpx/vp8cx.h"
1.27 +#include "vpx/vpx_encoder.h"
1.28 +
1.29 +#ifdef __MINGW32__
1.30 +#define strtok_r strtok_s
1.31 +#ifndef MINGW_HAS_SECURE_API
1.32 +// proto from /usr/x86_64-w64-mingw32/include/sec_api/string_s.h
1.33 +_CRTIMP char *__cdecl strtok_s(char *str, const char *delim, char **context);
1.34 +#endif /* MINGW_HAS_SECURE_API */
1.35 +#endif /* __MINGW32__ */
1.36 +
1.37 +#ifdef _MSC_VER
1.38 +#define strdup _strdup
1.39 +#define strtok_r strtok_s
1.40 +#endif
1.41 +
1.42 +#define SVC_REFERENCE_FRAMES 8
1.43 +#define SUPERFRAME_SLOTS (8)
1.44 +#define SUPERFRAME_BUFFER_SIZE (SUPERFRAME_SLOTS * sizeof(uint32_t) + 2)
1.45 +#define OPTION_BUFFER_SIZE 256
1.46 +
1.47 +static const char *DEFAULT_QUANTIZER_VALUES = "60,53,39,33,27";
1.48 +static const char *DEFAULT_SCALE_FACTORS = "4/16,5/16,7/16,11/16,16/16";
1.49 +
1.50 +typedef struct SvcInternal {
1.51 + char options[OPTION_BUFFER_SIZE]; // set by vpx_svc_set_options
1.52 + char quantizers[OPTION_BUFFER_SIZE]; // set by vpx_svc_set_quantizers
1.53 + char scale_factors[OPTION_BUFFER_SIZE]; // set by vpx_svc_set_scale_factors
1.54 +
1.55 + // values extracted from option, quantizers
1.56 + int scaling_factor_num[VPX_SS_MAX_LAYERS];
1.57 + int scaling_factor_den[VPX_SS_MAX_LAYERS];
1.58 + int quantizer[VPX_SS_MAX_LAYERS];
1.59 +
1.60 + // accumulated statistics
1.61 + double psnr_in_layer[VPX_SS_MAX_LAYERS];
1.62 + uint32_t bytes_in_layer[VPX_SS_MAX_LAYERS];
1.63 +
1.64 + // codec encoding values
1.65 + int width; // width of highest layer
1.66 + int height; // height of highest layer
1.67 + int kf_dist; // distance between keyframes
1.68 +
1.69 + // state variables
1.70 + int encode_frame_count;
1.71 + int frame_within_gop;
1.72 + vpx_enc_frame_flags_t enc_frame_flags;
1.73 + int layers;
1.74 + int layer;
1.75 + int is_keyframe;
1.76 +
1.77 + size_t frame_size;
1.78 + size_t buffer_size;
1.79 + void *buffer;
1.80 +
1.81 + char message_buffer[2048];
1.82 + vpx_codec_ctx_t *codec_ctx;
1.83 +} SvcInternal;
1.84 +
1.85 +// Superframe is used to generate an index of individual frames (i.e., layers)
1.86 +struct Superframe {
1.87 + int count;
1.88 + uint32_t sizes[SUPERFRAME_SLOTS];
1.89 + uint32_t magnitude;
1.90 + uint8_t buffer[SUPERFRAME_BUFFER_SIZE];
1.91 + size_t index_size;
1.92 +};
1.93 +
1.94 +// One encoded frame layer
1.95 +struct LayerData {
1.96 + void *buf; // compressed data buffer
1.97 + size_t size; // length of compressed data
1.98 + struct LayerData *next;
1.99 +};
1.100 +
1.101 +// create LayerData from encoder output
1.102 +static struct LayerData *ld_create(void *buf, size_t size) {
1.103 + struct LayerData *const layer_data = malloc(sizeof(*layer_data));
1.104 + if (layer_data == NULL) {
1.105 + return NULL;
1.106 + }
1.107 + layer_data->buf = malloc(size);
1.108 + if (layer_data->buf == NULL) {
1.109 + free(layer_data);
1.110 + return NULL;
1.111 + }
1.112 + memcpy(layer_data->buf, buf, size);
1.113 + layer_data->size = size;
1.114 + return layer_data;
1.115 +}
1.116 +
1.117 +// free LayerData
1.118 +static void ld_free(struct LayerData *layer_data) {
1.119 + if (layer_data) {
1.120 + if (layer_data->buf) {
1.121 + free(layer_data->buf);
1.122 + layer_data->buf = NULL;
1.123 + }
1.124 + free(layer_data);
1.125 + }
1.126 +}
1.127 +
1.128 +// add layer data to list
1.129 +static void ld_list_add(struct LayerData **list, struct LayerData *layer_data) {
1.130 + struct LayerData **p = list;
1.131 +
1.132 + while (*p != NULL) p = &(*p)->next;
1.133 + *p = layer_data;
1.134 + layer_data->next = NULL;
1.135 +}
1.136 +
1.137 +// get accumulated size of layer data
1.138 +static size_t ld_list_get_buffer_size(struct LayerData *list) {
1.139 + struct LayerData *p;
1.140 + size_t size = 0;
1.141 +
1.142 + for (p = list; p != NULL; p = p->next) {
1.143 + size += p->size;
1.144 + }
1.145 + return size;
1.146 +}
1.147 +
1.148 +// copy layer data to buffer
1.149 +static void ld_list_copy_to_buffer(struct LayerData *list, uint8_t *buffer) {
1.150 + struct LayerData *p;
1.151 +
1.152 + for (p = list; p != NULL; p = p->next) {
1.153 + buffer[0] = 1;
1.154 + memcpy(buffer, p->buf, p->size);
1.155 + buffer += p->size;
1.156 + }
1.157 +}
1.158 +
1.159 +// free layer data list
1.160 +static void ld_list_free(struct LayerData *list) {
1.161 + struct LayerData *p = list;
1.162 +
1.163 + while (p) {
1.164 + list = list->next;
1.165 + ld_free(p);
1.166 + p = list;
1.167 + }
1.168 +}
1.169 +
1.170 +static void sf_create_index(struct Superframe *sf) {
1.171 + uint8_t marker = 0xc0;
1.172 + int i;
1.173 + uint32_t mag, mask;
1.174 + uint8_t *bufp;
1.175 +
1.176 + if (sf->count == 0 || sf->count >= 8) return;
1.177 +
1.178 + // Add the number of frames to the marker byte
1.179 + marker |= sf->count - 1;
1.180 +
1.181 + // Choose the magnitude
1.182 + for (mag = 0, mask = 0xff; mag < 4; ++mag) {
1.183 + if (sf->magnitude < mask) break;
1.184 + mask <<= 8;
1.185 + mask |= 0xff;
1.186 + }
1.187 + marker |= mag << 3;
1.188 +
1.189 + // Write the index
1.190 + sf->index_size = 2 + (mag + 1) * sf->count;
1.191 + bufp = sf->buffer;
1.192 +
1.193 + *bufp++ = marker;
1.194 + for (i = 0; i < sf->count; ++i) {
1.195 + int this_sz = sf->sizes[i];
1.196 + uint32_t j;
1.197 +
1.198 + for (j = 0; j <= mag; ++j) {
1.199 + *bufp++ = this_sz & 0xff;
1.200 + this_sz >>= 8;
1.201 + }
1.202 + }
1.203 + *bufp++ = marker;
1.204 +}
1.205 +
1.206 +static SvcInternal *get_svc_internal(SvcContext *svc_ctx) {
1.207 + if (svc_ctx == NULL) return NULL;
1.208 + if (svc_ctx->internal == NULL) {
1.209 + SvcInternal *const si = malloc(sizeof(*si));
1.210 + if (si != NULL) {
1.211 + memset(si, 0, sizeof(*si));
1.212 + }
1.213 + svc_ctx->internal = si;
1.214 + }
1.215 + return svc_ctx->internal;
1.216 +}
1.217 +
1.218 +static const SvcInternal *get_const_svc_internal(const SvcContext *svc_ctx) {
1.219 + if (svc_ctx == NULL) return NULL;
1.220 + return svc_ctx->internal;
1.221 +}
1.222 +
1.223 +static void svc_log_reset(SvcContext *svc_ctx) {
1.224 + SvcInternal *const si = (SvcInternal *)svc_ctx->internal;
1.225 + si->message_buffer[0] = '\0';
1.226 +}
1.227 +
1.228 +static int svc_log(SvcContext *svc_ctx, int level, const char *fmt, ...) {
1.229 + char buf[512];
1.230 + int retval = 0;
1.231 + va_list ap;
1.232 + SvcInternal *const si = get_svc_internal(svc_ctx);
1.233 +
1.234 + if (level > svc_ctx->log_level) {
1.235 + return retval;
1.236 + }
1.237 +
1.238 + va_start(ap, fmt);
1.239 + retval = vsnprintf(buf, sizeof(buf), fmt, ap);
1.240 + va_end(ap);
1.241 +
1.242 + if (svc_ctx->log_print) {
1.243 + printf("%s", buf);
1.244 + } else {
1.245 + strncat(si->message_buffer, buf,
1.246 + sizeof(si->message_buffer) - strlen(si->message_buffer) - 1);
1.247 + }
1.248 +
1.249 + if (level == SVC_LOG_ERROR) {
1.250 + si->codec_ctx->err_detail = si->message_buffer;
1.251 + }
1.252 + return retval;
1.253 +}
1.254 +
1.255 +static vpx_codec_err_t set_option_encoding_mode(SvcContext *svc_ctx,
1.256 + const char *value_str) {
1.257 + if (strcmp(value_str, "i") == 0) {
1.258 + svc_ctx->encoding_mode = INTER_LAYER_PREDICTION_I;
1.259 + } else if (strcmp(value_str, "alt-ip") == 0) {
1.260 + svc_ctx->encoding_mode = ALT_INTER_LAYER_PREDICTION_IP;
1.261 + } else if (strcmp(value_str, "ip") == 0) {
1.262 + svc_ctx->encoding_mode = INTER_LAYER_PREDICTION_IP;
1.263 + } else if (strcmp(value_str, "gf") == 0) {
1.264 + svc_ctx->encoding_mode = USE_GOLDEN_FRAME;
1.265 + } else {
1.266 + svc_log(svc_ctx, SVC_LOG_ERROR, "invalid encoding mode: %s", value_str);
1.267 + return VPX_CODEC_INVALID_PARAM;
1.268 + }
1.269 + return VPX_CODEC_OK;
1.270 +}
1.271 +
1.272 +static vpx_codec_err_t parse_quantizer_values(SvcContext *svc_ctx,
1.273 + const char *quantizer_values) {
1.274 + char *input_string;
1.275 + char *token;
1.276 + const char *delim = ",";
1.277 + char *save_ptr;
1.278 + int found = 0;
1.279 + int i, q;
1.280 + int res = VPX_CODEC_OK;
1.281 + SvcInternal *const si = get_svc_internal(svc_ctx);
1.282 +
1.283 + if (quantizer_values == NULL || strlen(quantizer_values) == 0) {
1.284 + input_string = strdup(DEFAULT_QUANTIZER_VALUES);
1.285 + } else {
1.286 + input_string = strdup(quantizer_values);
1.287 + }
1.288 +
1.289 + token = strtok_r(input_string, delim, &save_ptr);
1.290 + for (i = 0; i < svc_ctx->spatial_layers; ++i) {
1.291 + if (token != NULL) {
1.292 + q = atoi(token);
1.293 + if (q <= 0 || q > 100) {
1.294 + svc_log(svc_ctx, SVC_LOG_ERROR,
1.295 + "svc-quantizer-values: invalid value %s\n", token);
1.296 + res = VPX_CODEC_INVALID_PARAM;
1.297 + break;
1.298 + }
1.299 + token = strtok_r(NULL, delim, &save_ptr);
1.300 + found = i + 1;
1.301 + } else {
1.302 + q = 0;
1.303 + }
1.304 + si->quantizer[i + VPX_SS_MAX_LAYERS - svc_ctx->spatial_layers] = q;
1.305 + }
1.306 + if (res == VPX_CODEC_OK && found != svc_ctx->spatial_layers) {
1.307 + svc_log(svc_ctx, SVC_LOG_ERROR,
1.308 + "svc: quantizers: %d values required, but only %d specified\n",
1.309 + svc_ctx->spatial_layers, found);
1.310 + res = VPX_CODEC_INVALID_PARAM;
1.311 + }
1.312 + free(input_string);
1.313 + return res;
1.314 +}
1.315 +
1.316 +static void log_invalid_scale_factor(SvcContext *svc_ctx, const char *value) {
1.317 + svc_log(svc_ctx, SVC_LOG_ERROR, "svc scale-factors: invalid value %s\n",
1.318 + value);
1.319 +}
1.320 +
1.321 +static vpx_codec_err_t parse_scale_factors(SvcContext *svc_ctx,
1.322 + const char *scale_factors) {
1.323 + char *input_string;
1.324 + char *token;
1.325 + const char *delim = ",";
1.326 + char *save_ptr;
1.327 + int found = 0;
1.328 + int i;
1.329 + int64_t num, den;
1.330 + int res = VPX_CODEC_OK;
1.331 + SvcInternal *const si = get_svc_internal(svc_ctx);
1.332 +
1.333 + if (scale_factors == NULL || strlen(scale_factors) == 0) {
1.334 + input_string = strdup(DEFAULT_SCALE_FACTORS);
1.335 + } else {
1.336 + input_string = strdup(scale_factors);
1.337 + }
1.338 + token = strtok_r(input_string, delim, &save_ptr);
1.339 + for (i = 0; i < svc_ctx->spatial_layers; ++i) {
1.340 + num = den = 0;
1.341 + if (token != NULL) {
1.342 + num = strtol(token, &token, 10);
1.343 + if (num <= 0) {
1.344 + log_invalid_scale_factor(svc_ctx, token);
1.345 + res = VPX_CODEC_INVALID_PARAM;
1.346 + break;
1.347 + }
1.348 + if (*token++ != '/') {
1.349 + log_invalid_scale_factor(svc_ctx, token);
1.350 + res = VPX_CODEC_INVALID_PARAM;
1.351 + break;
1.352 + }
1.353 + den = strtol(token, &token, 10);
1.354 + if (den <= 0) {
1.355 + log_invalid_scale_factor(svc_ctx, token);
1.356 + res = VPX_CODEC_INVALID_PARAM;
1.357 + break;
1.358 + }
1.359 + token = strtok_r(NULL, delim, &save_ptr);
1.360 + found = i + 1;
1.361 + }
1.362 + si->scaling_factor_num[i + VPX_SS_MAX_LAYERS - svc_ctx->spatial_layers] =
1.363 + (int)num;
1.364 + si->scaling_factor_den[i + VPX_SS_MAX_LAYERS - svc_ctx->spatial_layers] =
1.365 + (int)den;
1.366 + }
1.367 + if (res == VPX_CODEC_OK && found != svc_ctx->spatial_layers) {
1.368 + svc_log(svc_ctx, SVC_LOG_ERROR,
1.369 + "svc: scale-factors: %d values required, but only %d specified\n",
1.370 + svc_ctx->spatial_layers, found);
1.371 + res = VPX_CODEC_INVALID_PARAM;
1.372 + }
1.373 + free(input_string);
1.374 + return res;
1.375 +}
1.376 +
1.377 +/**
1.378 + * Parse SVC encoding options
1.379 + * Format: encoding-mode=<svc_mode>,layers=<layer_count>
1.380 + * scale-factors=<n1>/<d1>,<n2>/<d2>,...
1.381 + * quantizers=<q1>,<q2>,...
1.382 + * svc_mode = [i|ip|alt_ip|gf]
1.383 + */
1.384 +static vpx_codec_err_t parse_options(SvcContext *svc_ctx, const char *options) {
1.385 + char *input_string;
1.386 + char *option_name;
1.387 + char *option_value;
1.388 + char *input_ptr;
1.389 + int res = VPX_CODEC_OK;
1.390 +
1.391 + if (options == NULL) return VPX_CODEC_OK;
1.392 + input_string = strdup(options);
1.393 +
1.394 + // parse option name
1.395 + option_name = strtok_r(input_string, "=", &input_ptr);
1.396 + while (option_name != NULL) {
1.397 + // parse option value
1.398 + option_value = strtok_r(NULL, " ", &input_ptr);
1.399 + if (option_value == NULL) {
1.400 + svc_log(svc_ctx, SVC_LOG_ERROR, "option missing value: %s\n",
1.401 + option_name);
1.402 + res = VPX_CODEC_INVALID_PARAM;
1.403 + break;
1.404 + }
1.405 + if (strcmp("encoding-mode", option_name) == 0) {
1.406 + res = set_option_encoding_mode(svc_ctx, option_value);
1.407 + if (res != VPX_CODEC_OK) break;
1.408 + } else if (strcmp("layers", option_name) == 0) {
1.409 + svc_ctx->spatial_layers = atoi(option_value);
1.410 + } else if (strcmp("scale-factors", option_name) == 0) {
1.411 + res = parse_scale_factors(svc_ctx, option_value);
1.412 + if (res != VPX_CODEC_OK) break;
1.413 + } else if (strcmp("quantizers", option_name) == 0) {
1.414 + res = parse_quantizer_values(svc_ctx, option_value);
1.415 + if (res != VPX_CODEC_OK) break;
1.416 + } else {
1.417 + svc_log(svc_ctx, SVC_LOG_ERROR, "invalid option: %s\n", option_name);
1.418 + res = VPX_CODEC_INVALID_PARAM;
1.419 + break;
1.420 + }
1.421 + option_name = strtok_r(NULL, "=", &input_ptr);
1.422 + }
1.423 + free(input_string);
1.424 + return res;
1.425 +}
1.426 +
1.427 +vpx_codec_err_t vpx_svc_set_options(SvcContext *svc_ctx, const char *options) {
1.428 + SvcInternal *const si = get_svc_internal(svc_ctx);
1.429 + if (svc_ctx == NULL || options == NULL || si == NULL) {
1.430 + return VPX_CODEC_INVALID_PARAM;
1.431 + }
1.432 + strncpy(si->options, options, sizeof(si->options));
1.433 + si->options[sizeof(si->options) - 1] = '\0';
1.434 + return VPX_CODEC_OK;
1.435 +}
1.436 +
1.437 +vpx_codec_err_t vpx_svc_set_quantizers(SvcContext *svc_ctx,
1.438 + const char *quantizers) {
1.439 + SvcInternal *const si = get_svc_internal(svc_ctx);
1.440 + if (svc_ctx == NULL || quantizers == NULL || si == NULL) {
1.441 + return VPX_CODEC_INVALID_PARAM;
1.442 + }
1.443 + strncpy(si->quantizers, quantizers, sizeof(si->quantizers));
1.444 + si->quantizers[sizeof(si->quantizers) - 1] = '\0';
1.445 + return VPX_CODEC_OK;
1.446 +}
1.447 +
1.448 +vpx_codec_err_t vpx_svc_set_scale_factors(SvcContext *svc_ctx,
1.449 + const char *scale_factors) {
1.450 + SvcInternal *const si = get_svc_internal(svc_ctx);
1.451 + if (svc_ctx == NULL || scale_factors == NULL || si == NULL) {
1.452 + return VPX_CODEC_INVALID_PARAM;
1.453 + }
1.454 + strncpy(si->scale_factors, scale_factors, sizeof(si->scale_factors));
1.455 + si->scale_factors[sizeof(si->scale_factors) - 1] = '\0';
1.456 + return VPX_CODEC_OK;
1.457 +}
1.458 +
1.459 +vpx_codec_err_t vpx_svc_init(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
1.460 + vpx_codec_iface_t *iface,
1.461 + vpx_codec_enc_cfg_t *enc_cfg) {
1.462 + int max_intra_size_pct;
1.463 + vpx_codec_err_t res;
1.464 + SvcInternal *const si = get_svc_internal(svc_ctx);
1.465 + if (svc_ctx == NULL || codec_ctx == NULL || iface == NULL ||
1.466 + enc_cfg == NULL) {
1.467 + return VPX_CODEC_INVALID_PARAM;
1.468 + }
1.469 + if (si == NULL) return VPX_CODEC_MEM_ERROR;
1.470 +
1.471 + si->codec_ctx = codec_ctx;
1.472 +
1.473 + si->width = enc_cfg->g_w;
1.474 + si->height = enc_cfg->g_h;
1.475 +
1.476 + if (enc_cfg->kf_max_dist < 2) {
1.477 + svc_log(svc_ctx, SVC_LOG_ERROR, "key frame distance too small: %d\n",
1.478 + enc_cfg->kf_max_dist);
1.479 + return VPX_CODEC_INVALID_PARAM;
1.480 + }
1.481 + si->kf_dist = enc_cfg->kf_max_dist;
1.482 +
1.483 + if (svc_ctx->spatial_layers == 0)
1.484 + svc_ctx->spatial_layers = VPX_SS_DEFAULT_LAYERS;
1.485 + if (svc_ctx->spatial_layers < 1 ||
1.486 + svc_ctx->spatial_layers > VPX_SS_MAX_LAYERS) {
1.487 + svc_log(svc_ctx, SVC_LOG_ERROR, "spatial layers: invalid value: %d\n",
1.488 + svc_ctx->spatial_layers);
1.489 + return VPX_CODEC_INVALID_PARAM;
1.490 + }
1.491 + // use SvcInternal value for number of layers to enable forcing single layer
1.492 + // for first frame
1.493 + si->layers = svc_ctx->spatial_layers;
1.494 +
1.495 + res = parse_quantizer_values(svc_ctx, si->quantizers);
1.496 + if (res != VPX_CODEC_OK) return res;
1.497 +
1.498 + res = parse_scale_factors(svc_ctx, si->scale_factors);
1.499 + if (res != VPX_CODEC_OK) return res;
1.500 +
1.501 + // parse aggregate command line options
1.502 + res = parse_options(svc_ctx, si->options);
1.503 + if (res != VPX_CODEC_OK) return res;
1.504 +
1.505 + // modify encoder configuration
1.506 + enc_cfg->ss_number_layers = si->layers;
1.507 + enc_cfg->kf_mode = VPX_KF_DISABLED;
1.508 + enc_cfg->g_pass = VPX_RC_ONE_PASS;
1.509 + // Lag in frames not currently supported
1.510 + enc_cfg->g_lag_in_frames = 0;
1.511 +
1.512 + // TODO(ivanmaltz): determine if these values need to be set explicitly for
1.513 + // svc, or if the normal default/override mechanism can be used
1.514 + enc_cfg->rc_dropframe_thresh = 0;
1.515 + enc_cfg->rc_end_usage = VPX_CBR;
1.516 + enc_cfg->rc_resize_allowed = 0;
1.517 + enc_cfg->rc_min_quantizer = 33;
1.518 + enc_cfg->rc_max_quantizer = 33;
1.519 + enc_cfg->rc_undershoot_pct = 100;
1.520 + enc_cfg->rc_overshoot_pct = 15;
1.521 + enc_cfg->rc_buf_initial_sz = 500;
1.522 + enc_cfg->rc_buf_optimal_sz = 600;
1.523 + enc_cfg->rc_buf_sz = 1000;
1.524 + enc_cfg->g_error_resilient = 1;
1.525 +
1.526 + // Initialize codec
1.527 + res = vpx_codec_enc_init(codec_ctx, iface, enc_cfg, VPX_CODEC_USE_PSNR);
1.528 + if (res != VPX_CODEC_OK) {
1.529 + svc_log(svc_ctx, SVC_LOG_ERROR, "svc_enc_init error\n");
1.530 + return res;
1.531 + }
1.532 +
1.533 + vpx_codec_control(codec_ctx, VP9E_SET_SVC, 1);
1.534 + vpx_codec_control(codec_ctx, VP8E_SET_CPUUSED, 1);
1.535 + vpx_codec_control(codec_ctx, VP8E_SET_STATIC_THRESHOLD, 1);
1.536 + vpx_codec_control(codec_ctx, VP8E_SET_NOISE_SENSITIVITY, 1);
1.537 + vpx_codec_control(codec_ctx, VP8E_SET_TOKEN_PARTITIONS, 1);
1.538 +
1.539 + max_intra_size_pct =
1.540 + (int)(((double)enc_cfg->rc_buf_optimal_sz * 0.5) *
1.541 + ((double)enc_cfg->g_timebase.den / enc_cfg->g_timebase.num) / 10.0);
1.542 + vpx_codec_control(codec_ctx, VP8E_SET_MAX_INTRA_BITRATE_PCT,
1.543 + max_intra_size_pct);
1.544 + return VPX_CODEC_OK;
1.545 +}
1.546 +
1.547 +// SVC Algorithm flags - these get mapped to VP8_EFLAG_* defined in vp8cx.h
1.548 +
1.549 +// encoder should reference the last frame
1.550 +#define USE_LAST (1 << 0)
1.551 +
1.552 +// encoder should reference the alt ref frame
1.553 +#define USE_ARF (1 << 1)
1.554 +
1.555 +// encoder should reference the golden frame
1.556 +#define USE_GF (1 << 2)
1.557 +
1.558 +// encoder should copy current frame to the last frame buffer
1.559 +#define UPDATE_LAST (1 << 3)
1.560 +
1.561 +// encoder should copy current frame to the alt ref frame buffer
1.562 +#define UPDATE_ARF (1 << 4)
1.563 +
1.564 +// encoder should copy current frame to the golden frame
1.565 +#define UPDATE_GF (1 << 5)
1.566 +
1.567 +static int map_vp8_flags(int svc_flags) {
1.568 + int flags = 0;
1.569 +
1.570 + if (!(svc_flags & USE_LAST)) flags |= VP8_EFLAG_NO_REF_LAST;
1.571 + if (!(svc_flags & USE_ARF)) flags |= VP8_EFLAG_NO_REF_ARF;
1.572 + if (!(svc_flags & USE_GF)) flags |= VP8_EFLAG_NO_REF_GF;
1.573 +
1.574 + if (svc_flags & UPDATE_LAST) {
1.575 + // last is updated automatically
1.576 + } else {
1.577 + flags |= VP8_EFLAG_NO_UPD_LAST;
1.578 + }
1.579 + if (svc_flags & UPDATE_ARF) {
1.580 + flags |= VP8_EFLAG_FORCE_ARF;
1.581 + } else {
1.582 + flags |= VP8_EFLAG_NO_UPD_ARF;
1.583 + }
1.584 + if (svc_flags & UPDATE_GF) {
1.585 + flags |= VP8_EFLAG_FORCE_GF;
1.586 + } else {
1.587 + flags |= VP8_EFLAG_NO_UPD_GF;
1.588 + }
1.589 + return flags;
1.590 +}
1.591 +
1.592 +/**
1.593 + * Helper to check if the current frame is the first, full resolution dummy.
1.594 + */
1.595 +static int vpx_svc_dummy_frame(SvcContext *svc_ctx) {
1.596 + SvcInternal *const si = get_svc_internal(svc_ctx);
1.597 + return svc_ctx->first_frame_full_size == 1 && si->encode_frame_count == 0;
1.598 +}
1.599 +
1.600 +static void calculate_enc_frame_flags(SvcContext *svc_ctx) {
1.601 + vpx_enc_frame_flags_t flags = VPX_EFLAG_FORCE_KF;
1.602 + SvcInternal *const si = get_svc_internal(svc_ctx);
1.603 + const int is_keyframe = (si->frame_within_gop == 0);
1.604 +
1.605 + // keyframe layer zero is identical for all modes
1.606 + if ((is_keyframe && si->layer == 0) || vpx_svc_dummy_frame(svc_ctx)) {
1.607 + si->enc_frame_flags = VPX_EFLAG_FORCE_KF;
1.608 + return;
1.609 + }
1.610 +
1.611 + switch (svc_ctx->encoding_mode) {
1.612 + case ALT_INTER_LAYER_PREDICTION_IP:
1.613 + if (si->layer == 0) {
1.614 + flags = map_vp8_flags(USE_LAST | UPDATE_LAST);
1.615 + } else if (is_keyframe) {
1.616 + if (si->layer == si->layers - 1) {
1.617 + flags = map_vp8_flags(USE_ARF | UPDATE_LAST);
1.618 + } else {
1.619 + flags = map_vp8_flags(USE_ARF | UPDATE_LAST | UPDATE_GF);
1.620 + }
1.621 + } else {
1.622 + flags = map_vp8_flags(USE_LAST | USE_ARF | UPDATE_LAST);
1.623 + }
1.624 + break;
1.625 + case INTER_LAYER_PREDICTION_I:
1.626 + if (si->layer == 0) {
1.627 + flags = map_vp8_flags(USE_LAST | UPDATE_LAST);
1.628 + } else if (is_keyframe) {
1.629 + flags = map_vp8_flags(USE_ARF | UPDATE_LAST);
1.630 + } else {
1.631 + flags = map_vp8_flags(USE_LAST | UPDATE_LAST);
1.632 + }
1.633 + break;
1.634 + case INTER_LAYER_PREDICTION_IP:
1.635 + if (si->layer == 0) {
1.636 + flags = map_vp8_flags(USE_LAST | UPDATE_LAST);
1.637 + } else if (is_keyframe) {
1.638 + flags = map_vp8_flags(USE_ARF | UPDATE_LAST);
1.639 + } else {
1.640 + flags = map_vp8_flags(USE_LAST | USE_ARF | UPDATE_LAST);
1.641 + }
1.642 + break;
1.643 + case USE_GOLDEN_FRAME:
1.644 + if (2 * si->layers - SVC_REFERENCE_FRAMES <= si->layer) {
1.645 + if (si->layer == 0) {
1.646 + flags = map_vp8_flags(USE_LAST | USE_GF | UPDATE_LAST);
1.647 + } else if (is_keyframe) {
1.648 + flags = map_vp8_flags(USE_ARF | UPDATE_LAST | UPDATE_GF);
1.649 + } else {
1.650 + flags = map_vp8_flags(USE_LAST | USE_ARF | USE_GF | UPDATE_LAST);
1.651 + }
1.652 + } else {
1.653 + if (si->layer == 0) {
1.654 + flags = map_vp8_flags(USE_LAST | UPDATE_LAST);
1.655 + } else if (is_keyframe) {
1.656 + flags = map_vp8_flags(USE_ARF | UPDATE_LAST);
1.657 + } else {
1.658 + flags = map_vp8_flags(USE_LAST | UPDATE_LAST);
1.659 + }
1.660 + }
1.661 + break;
1.662 + default:
1.663 + svc_log(svc_ctx, SVC_LOG_ERROR, "unexpected encoding mode: %d\n",
1.664 + svc_ctx->encoding_mode);
1.665 + break;
1.666 + }
1.667 + si->enc_frame_flags = flags;
1.668 +}
1.669 +
1.670 +vpx_codec_err_t vpx_svc_get_layer_resolution(const SvcContext *svc_ctx,
1.671 + int layer,
1.672 + unsigned int *width,
1.673 + unsigned int *height) {
1.674 + int w, h, index, num, den;
1.675 + const SvcInternal *const si = get_const_svc_internal(svc_ctx);
1.676 +
1.677 + if (svc_ctx == NULL || si == NULL || width == NULL || height == NULL) {
1.678 + return VPX_CODEC_INVALID_PARAM;
1.679 + }
1.680 + if (layer < 0 || layer >= si->layers) return VPX_CODEC_INVALID_PARAM;
1.681 +
1.682 + index = layer + VPX_SS_MAX_LAYERS - si->layers;
1.683 + num = si->scaling_factor_num[index];
1.684 + den = si->scaling_factor_den[index];
1.685 + if (num == 0 || den == 0) return VPX_CODEC_INVALID_PARAM;
1.686 +
1.687 + w = si->width * num / den;
1.688 + h = si->height * num / den;
1.689 +
1.690 + // make height and width even to make chrome player happy
1.691 + w += w % 2;
1.692 + h += h % 2;
1.693 +
1.694 + *width = w;
1.695 + *height = h;
1.696 +
1.697 + return VPX_CODEC_OK;
1.698 +}
1.699 +
1.700 +static void set_svc_parameters(SvcContext *svc_ctx,
1.701 + vpx_codec_ctx_t *codec_ctx) {
1.702 + int layer, layer_index;
1.703 + vpx_svc_parameters_t svc_params;
1.704 + SvcInternal *const si = get_svc_internal(svc_ctx);
1.705 +
1.706 + memset(&svc_params, 0, sizeof(svc_params));
1.707 + svc_params.layer = si->layer;
1.708 + svc_params.flags = si->enc_frame_flags;
1.709 +
1.710 + layer = si->layer;
1.711 + if (svc_ctx->encoding_mode == ALT_INTER_LAYER_PREDICTION_IP &&
1.712 + si->frame_within_gop == 0) {
1.713 + // layers 1 & 3 don't exist in this mode, use the higher one
1.714 + if (layer == 0 || layer == 2) {
1.715 + layer += 1;
1.716 + }
1.717 + }
1.718 + if (VPX_CODEC_OK != vpx_svc_get_layer_resolution(svc_ctx, layer,
1.719 + &svc_params.width,
1.720 + &svc_params.height)) {
1.721 + svc_log(svc_ctx, SVC_LOG_ERROR, "vpx_svc_get_layer_resolution failed\n");
1.722 + }
1.723 + layer_index = layer + VPX_SS_MAX_LAYERS - si->layers;
1.724 + svc_params.min_quantizer = si->quantizer[layer_index];
1.725 + svc_params.max_quantizer = si->quantizer[layer_index];
1.726 + svc_params.distance_from_i_frame = si->frame_within_gop;
1.727 +
1.728 + // Use buffer i for layer i LST
1.729 + svc_params.lst_fb_idx = si->layer;
1.730 +
1.731 + // Use buffer i-1 for layer i Alt (Inter-layer prediction)
1.732 + if (si->layer != 0) {
1.733 + const int use_higher_layer =
1.734 + svc_ctx->encoding_mode == ALT_INTER_LAYER_PREDICTION_IP &&
1.735 + si->frame_within_gop == 0;
1.736 + svc_params.alt_fb_idx = use_higher_layer ? si->layer - 2 : si->layer - 1;
1.737 + }
1.738 +
1.739 + if (svc_ctx->encoding_mode == ALT_INTER_LAYER_PREDICTION_IP) {
1.740 + svc_params.gld_fb_idx = si->layer + 1;
1.741 + } else {
1.742 + if (si->layer < 2 * si->layers - SVC_REFERENCE_FRAMES)
1.743 + svc_params.gld_fb_idx = svc_params.lst_fb_idx;
1.744 + else
1.745 + svc_params.gld_fb_idx = 2 * si->layers - 1 - si->layer;
1.746 + }
1.747 +
1.748 + svc_log(svc_ctx, SVC_LOG_DEBUG, "SVC frame: %d, layer: %d, %dx%d, q: %d\n",
1.749 + si->encode_frame_count, si->layer, svc_params.width,
1.750 + svc_params.height, svc_params.min_quantizer);
1.751 +
1.752 + if (svc_params.flags == VPX_EFLAG_FORCE_KF) {
1.753 + svc_log(svc_ctx, SVC_LOG_DEBUG, "flags == VPX_EFLAG_FORCE_KF\n");
1.754 + } else {
1.755 + svc_log(
1.756 + svc_ctx, SVC_LOG_DEBUG, "Using: LST/GLD/ALT [%2d|%2d|%2d]\n",
1.757 + svc_params.flags & VP8_EFLAG_NO_REF_LAST ? -1 : svc_params.lst_fb_idx,
1.758 + svc_params.flags & VP8_EFLAG_NO_REF_GF ? -1 : svc_params.gld_fb_idx,
1.759 + svc_params.flags & VP8_EFLAG_NO_REF_ARF ? -1 : svc_params.alt_fb_idx);
1.760 + svc_log(
1.761 + svc_ctx, SVC_LOG_DEBUG, "Updating: LST/GLD/ALT [%2d|%2d|%2d]\n",
1.762 + svc_params.flags & VP8_EFLAG_NO_UPD_LAST ? -1 : svc_params.lst_fb_idx,
1.763 + svc_params.flags & VP8_EFLAG_NO_UPD_GF ? -1 : svc_params.gld_fb_idx,
1.764 + svc_params.flags & VP8_EFLAG_NO_UPD_ARF ? -1 : svc_params.alt_fb_idx);
1.765 + }
1.766 +
1.767 + vpx_codec_control(codec_ctx, VP9E_SET_SVC_PARAMETERS, &svc_params);
1.768 +}
1.769 +
1.770 +/**
1.771 + * Encode a frame into multiple layers
1.772 + * Create a superframe containing the individual layers
1.773 + */
1.774 +vpx_codec_err_t vpx_svc_encode(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
1.775 + struct vpx_image *rawimg, vpx_codec_pts_t pts,
1.776 + int64_t duration, int deadline) {
1.777 + vpx_codec_err_t res;
1.778 + vpx_codec_iter_t iter;
1.779 + const vpx_codec_cx_pkt_t *cx_pkt;
1.780 + struct LayerData *cx_layer_list = NULL;
1.781 + struct LayerData *layer_data;
1.782 + struct Superframe superframe;
1.783 + SvcInternal *const si = get_svc_internal(svc_ctx);
1.784 + if (svc_ctx == NULL || codec_ctx == NULL || rawimg == NULL || si == NULL) {
1.785 + return VPX_CODEC_INVALID_PARAM;
1.786 + }
1.787 +
1.788 + memset(&superframe, 0, sizeof(superframe));
1.789 + svc_log_reset(svc_ctx);
1.790 +
1.791 + si->layers = vpx_svc_dummy_frame(svc_ctx) ? 1 : svc_ctx->spatial_layers;
1.792 + if (si->frame_within_gop >= si->kf_dist ||
1.793 + si->encode_frame_count == 0 ||
1.794 + (si->encode_frame_count == 1 && svc_ctx->first_frame_full_size == 1)) {
1.795 + si->frame_within_gop = 0;
1.796 + }
1.797 + si->is_keyframe = (si->frame_within_gop == 0);
1.798 + si->frame_size = 0;
1.799 +
1.800 + svc_log(svc_ctx, SVC_LOG_DEBUG,
1.801 + "vpx_svc_encode layers: %d, frame_count: %d, frame_within_gop: %d\n",
1.802 + si->layers, si->encode_frame_count, si->frame_within_gop);
1.803 +
1.804 + // encode each layer
1.805 + for (si->layer = 0; si->layer < si->layers; ++si->layer) {
1.806 + if (svc_ctx->encoding_mode == ALT_INTER_LAYER_PREDICTION_IP &&
1.807 + si->is_keyframe && (si->layer == 1 || si->layer == 3)) {
1.808 + svc_log(svc_ctx, SVC_LOG_DEBUG, "Skip encoding layer %d\n", si->layer);
1.809 + continue;
1.810 + }
1.811 + calculate_enc_frame_flags(svc_ctx);
1.812 +
1.813 + if (vpx_svc_dummy_frame(svc_ctx)) {
1.814 + // do not set svc parameters, use normal encode
1.815 + svc_log(svc_ctx, SVC_LOG_DEBUG, "encoding full size first frame\n");
1.816 + } else {
1.817 + set_svc_parameters(svc_ctx, codec_ctx);
1.818 + }
1.819 + res = vpx_codec_encode(codec_ctx, rawimg, pts, (uint32_t)duration,
1.820 + si->enc_frame_flags, deadline);
1.821 + if (res != VPX_CODEC_OK) {
1.822 + return res;
1.823 + }
1.824 + // save compressed data
1.825 + iter = NULL;
1.826 + while ((cx_pkt = vpx_codec_get_cx_data(codec_ctx, &iter))) {
1.827 + switch (cx_pkt->kind) {
1.828 + case VPX_CODEC_CX_FRAME_PKT: {
1.829 + const uint32_t frame_pkt_size = (uint32_t)(cx_pkt->data.frame.sz);
1.830 + if (!vpx_svc_dummy_frame(svc_ctx)) {
1.831 + si->bytes_in_layer[si->layer] += frame_pkt_size;
1.832 + svc_log(svc_ctx, SVC_LOG_DEBUG,
1.833 + "SVC frame: %d, layer: %d, size: %u\n",
1.834 + si->encode_frame_count, si->layer, frame_pkt_size);
1.835 + }
1.836 + layer_data =
1.837 + ld_create(cx_pkt->data.frame.buf, (size_t)frame_pkt_size);
1.838 + if (layer_data == NULL) {
1.839 + svc_log(svc_ctx, SVC_LOG_ERROR, "Error allocating LayerData\n");
1.840 + return 0;
1.841 + }
1.842 + ld_list_add(&cx_layer_list, layer_data);
1.843 +
1.844 + // save layer size in superframe index
1.845 + superframe.sizes[superframe.count++] = frame_pkt_size;
1.846 + superframe.magnitude |= frame_pkt_size;
1.847 + break;
1.848 + }
1.849 + case VPX_CODEC_PSNR_PKT: {
1.850 + if (!vpx_svc_dummy_frame(svc_ctx)) {
1.851 + svc_log(svc_ctx, SVC_LOG_DEBUG,
1.852 + "SVC frame: %d, layer: %d, PSNR(Total/Y/U/V): "
1.853 + "%2.3f %2.3f %2.3f %2.3f \n",
1.854 + si->encode_frame_count, si->layer,
1.855 + cx_pkt->data.psnr.psnr[0], cx_pkt->data.psnr.psnr[1],
1.856 + cx_pkt->data.psnr.psnr[2], cx_pkt->data.psnr.psnr[3]);
1.857 + si->psnr_in_layer[si->layer] += cx_pkt->data.psnr.psnr[0];
1.858 + }
1.859 + break;
1.860 + }
1.861 + default: {
1.862 + break;
1.863 + }
1.864 + }
1.865 + }
1.866 + }
1.867 + // add superframe index to layer data list
1.868 + if (!vpx_svc_dummy_frame(svc_ctx)) {
1.869 + sf_create_index(&superframe);
1.870 + layer_data = ld_create(superframe.buffer, superframe.index_size);
1.871 + ld_list_add(&cx_layer_list, layer_data);
1.872 + }
1.873 + // get accumulated size of layer data
1.874 + si->frame_size = ld_list_get_buffer_size(cx_layer_list);
1.875 + if (si->frame_size == 0) return VPX_CODEC_ERROR;
1.876 +
1.877 + // all layers encoded, create single buffer with concatenated layers
1.878 + if (si->frame_size > si->buffer_size) {
1.879 + free(si->buffer);
1.880 + si->buffer = malloc(si->frame_size);
1.881 + if (si->buffer == NULL) {
1.882 + ld_list_free(cx_layer_list);
1.883 + return VPX_CODEC_MEM_ERROR;
1.884 + }
1.885 + si->buffer_size = si->frame_size;
1.886 + }
1.887 + // copy layer data into packet
1.888 + ld_list_copy_to_buffer(cx_layer_list, si->buffer);
1.889 +
1.890 + ld_list_free(cx_layer_list);
1.891 +
1.892 + svc_log(svc_ctx, SVC_LOG_DEBUG, "SVC frame: %d, kf: %d, size: %d, pts: %d\n",
1.893 + si->encode_frame_count, si->is_keyframe, (int)si->frame_size,
1.894 + (int)pts);
1.895 + ++si->frame_within_gop;
1.896 + ++si->encode_frame_count;
1.897 +
1.898 + return VPX_CODEC_OK;
1.899 +}
1.900 +
1.901 +const char *vpx_svc_get_message(const SvcContext *svc_ctx) {
1.902 + const SvcInternal *const si = get_const_svc_internal(svc_ctx);
1.903 + if (svc_ctx == NULL || si == NULL) return NULL;
1.904 + return si->message_buffer;
1.905 +}
1.906 +
1.907 +void *vpx_svc_get_buffer(const SvcContext *svc_ctx) {
1.908 + const SvcInternal *const si = get_const_svc_internal(svc_ctx);
1.909 + if (svc_ctx == NULL || si == NULL) return NULL;
1.910 + return si->buffer;
1.911 +}
1.912 +
1.913 +size_t vpx_svc_get_frame_size(const SvcContext *svc_ctx) {
1.914 + const SvcInternal *const si = get_const_svc_internal(svc_ctx);
1.915 + if (svc_ctx == NULL || si == NULL) return 0;
1.916 + return si->frame_size;
1.917 +}
1.918 +
1.919 +int vpx_svc_get_encode_frame_count(const SvcContext *svc_ctx) {
1.920 + const SvcInternal *const si = get_const_svc_internal(svc_ctx);
1.921 + if (svc_ctx == NULL || si == NULL) return 0;
1.922 + return si->encode_frame_count;
1.923 +}
1.924 +
1.925 +int vpx_svc_is_keyframe(const SvcContext *svc_ctx) {
1.926 + const SvcInternal *const si = get_const_svc_internal(svc_ctx);
1.927 + if (svc_ctx == NULL || si == NULL) return 0;
1.928 + return si->is_keyframe;
1.929 +}
1.930 +
1.931 +void vpx_svc_set_keyframe(SvcContext *svc_ctx) {
1.932 + SvcInternal *const si = get_svc_internal(svc_ctx);
1.933 + if (svc_ctx == NULL || si == NULL) return;
1.934 + si->frame_within_gop = 0;
1.935 +}
1.936 +
1.937 +// dump accumulated statistics and reset accumulated values
1.938 +const char *vpx_svc_dump_statistics(SvcContext *svc_ctx) {
1.939 + int number_of_frames, number_of_keyframes, encode_frame_count;
1.940 + int i;
1.941 + uint32_t bytes_total = 0;
1.942 + SvcInternal *const si = get_svc_internal(svc_ctx);
1.943 + if (svc_ctx == NULL || si == NULL) return NULL;
1.944 +
1.945 + svc_log_reset(svc_ctx);
1.946 +
1.947 + encode_frame_count = si->encode_frame_count;
1.948 + if (svc_ctx->first_frame_full_size) encode_frame_count--;
1.949 + if (si->encode_frame_count <= 0) return vpx_svc_get_message(svc_ctx);
1.950 +
1.951 + svc_log(svc_ctx, SVC_LOG_INFO, "\n");
1.952 + number_of_keyframes = encode_frame_count / si->kf_dist + 1;
1.953 + for (i = 0; i < si->layers; ++i) {
1.954 + number_of_frames = encode_frame_count;
1.955 +
1.956 + if (svc_ctx->encoding_mode == ALT_INTER_LAYER_PREDICTION_IP &&
1.957 + (i == 1 || i == 3)) {
1.958 + number_of_frames -= number_of_keyframes;
1.959 + }
1.960 + svc_log(svc_ctx, SVC_LOG_INFO, "Layer %d PSNR=[%2.3f], Bytes=[%u]\n", i,
1.961 + (double)si->psnr_in_layer[i] / number_of_frames,
1.962 + si->bytes_in_layer[i]);
1.963 + bytes_total += si->bytes_in_layer[i];
1.964 + si->psnr_in_layer[i] = 0;
1.965 + si->bytes_in_layer[i] = 0;
1.966 + }
1.967 +
1.968 + // only display statistics once
1.969 + si->encode_frame_count = 0;
1.970 +
1.971 + svc_log(svc_ctx, SVC_LOG_INFO, "Total Bytes=[%u]\n", bytes_total);
1.972 + return vpx_svc_get_message(svc_ctx);
1.973 +}
1.974 +
1.975 +void vpx_svc_release(SvcContext *svc_ctx) {
1.976 + SvcInternal *si;
1.977 + if (svc_ctx == NULL) return;
1.978 + // do not use get_svc_internal as it will unnecessarily allocate an
1.979 + // SvcInternal if it was not already allocated
1.980 + si = (SvcInternal *)svc_ctx->internal;
1.981 + if (si != NULL) {
1.982 + free(si->buffer);
1.983 + free(si);
1.984 + svc_ctx->internal = NULL;
1.985 + }
1.986 +}
