1/* 2 * Copyright (c) 2013 The WebM project authors. All Rights Reserved. 3 * 4 * Use of this source code is governed by a BSD-style license 5 * that can be found in the LICENSE file in the root of the source 6 * tree. An additional intellectual property rights grant can be found 7 * in the file PATENTS. All contributing project authors may 8 * be found in the AUTHORS file in the root of the source tree. 9 */ 10 11/** 12 * @file 13 * VP9 SVC encoding support via libvpx 14 */ 15 16#include <assert.h> 17#include <math.h> 18#include <limits.h> 19#include <stdarg.h> 20#include <stdio.h> 21#include <stdlib.h> 22#include <string.h> 23#define VPX_DISABLE_CTRL_TYPECHECKS 1 24#include "./vpx_config.h" 25#include "vpx/svc_context.h" 26#include "vpx/vp8cx.h" 27#include "vpx/vpx_encoder.h" 28#include "vpx_mem/vpx_mem.h" 29#include "vp9/common/vp9_onyxc_int.h" 30 31#ifdef __MINGW32__ 32#define strtok_r strtok_s 33#ifndef MINGW_HAS_SECURE_API 34// proto from /usr/x86_64-w64-mingw32/include/sec_api/string_s.h 35_CRTIMP char *__cdecl strtok_s(char *str, const char *delim, char **context); 36#endif /* MINGW_HAS_SECURE_API */ 37#endif /* __MINGW32__ */ 38 39#ifdef _MSC_VER 40#define strdup _strdup 41#define strtok_r strtok_s 42#endif 43 44#define SVC_REFERENCE_FRAMES 8 45#define SUPERFRAME_SLOTS (8) 46#define SUPERFRAME_BUFFER_SIZE (SUPERFRAME_SLOTS * sizeof(uint32_t) + 2) 47 48#define MAX_QUANTIZER 63 49 50static const int DEFAULT_SCALE_FACTORS_NUM[VPX_SS_MAX_LAYERS] = { 51 4, 5, 7, 11, 16 52}; 53 54static const int DEFAULT_SCALE_FACTORS_DEN[VPX_SS_MAX_LAYERS] = { 55 16, 16, 16, 16, 16 56}; 57 58typedef enum { 59 QUANTIZER = 0, 60 BITRATE, 61 SCALE_FACTOR, 62 AUTO_ALT_REF, 63 ALL_OPTION_TYPES 64} LAYER_OPTION_TYPE; 65 66static const int option_max_values[ALL_OPTION_TYPES] = { 67 63, INT_MAX, INT_MAX, 1 68}; 69 70static const int option_min_values[ALL_OPTION_TYPES] = { 71 0, 0, 1, 0 72}; 73 74// One encoded frame 75typedef struct FrameData { 76 void *buf; // compressed data buffer 77 size_t size; // length of compressed data 78 vpx_codec_frame_flags_t flags; /**< flags for this frame */ 79 struct FrameData *next; 80} FrameData; 81 82static SvcInternal_t *get_svc_internal(SvcContext *svc_ctx) { 83 if (svc_ctx == NULL) return NULL; 84 if (svc_ctx->internal == NULL) { 85 SvcInternal_t *const si = (SvcInternal_t *)malloc(sizeof(*si)); 86 if (si != NULL) { 87 memset(si, 0, sizeof(*si)); 88 } 89 svc_ctx->internal = si; 90 } 91 return (SvcInternal_t *)svc_ctx->internal; 92} 93 94static const SvcInternal_t *get_const_svc_internal( 95 const SvcContext *svc_ctx) { 96 if (svc_ctx == NULL) return NULL; 97 return (const SvcInternal_t *)svc_ctx->internal; 98} 99 100static void svc_log_reset(SvcContext *svc_ctx) { 101 SvcInternal_t *const si = (SvcInternal_t *)svc_ctx->internal; 102 si->message_buffer[0] = '\0'; 103} 104 105static int svc_log(SvcContext *svc_ctx, SVC_LOG_LEVEL level, 106 const char *fmt, ...) { 107 char buf[512]; 108 int retval = 0; 109 va_list ap; 110 SvcInternal_t *const si = get_svc_internal(svc_ctx); 111 112 if (level > svc_ctx->log_level) { 113 return retval; 114 } 115 116 va_start(ap, fmt); 117 retval = vsnprintf(buf, sizeof(buf), fmt, ap); 118 va_end(ap); 119 120 if (svc_ctx->log_print) { 121 printf("%s", buf); 122 } else { 123 strncat(si->message_buffer, buf, 124 sizeof(si->message_buffer) - strlen(si->message_buffer) - 1); 125 } 126 127 if (level == SVC_LOG_ERROR) { 128 si->codec_ctx->err_detail = si->message_buffer; 129 } 130 return retval; 131} 132 133static vpx_codec_err_t extract_option(LAYER_OPTION_TYPE type, 134 char *input, 135 int *value0, 136 int *value1) { 137 if (type == SCALE_FACTOR) { 138 *value0 = strtol(input, &input, 10); 139 if (*input++ != '/') 140 return VPX_CODEC_INVALID_PARAM; 141 *value1 = strtol(input, &input, 10); 142 143 if (*value0 < option_min_values[SCALE_FACTOR] || 144 *value1 < option_min_values[SCALE_FACTOR] || 145 *value0 > option_max_values[SCALE_FACTOR] || 146 *value1 > option_max_values[SCALE_FACTOR] || 147 *value0 > *value1) // num shouldn't be greater than den 148 return VPX_CODEC_INVALID_PARAM; 149 } else { 150 *value0 = atoi(input); 151 if (*value0 < option_min_values[type] || 152 *value0 > option_max_values[type]) 153 return VPX_CODEC_INVALID_PARAM; 154 } 155 return VPX_CODEC_OK; 156} 157 158static vpx_codec_err_t parse_layer_options_from_string(SvcContext *svc_ctx, 159 LAYER_OPTION_TYPE type, 160 const char *input, 161 int *option0, 162 int *option1) { 163 int i; 164 vpx_codec_err_t res = VPX_CODEC_OK; 165 char *input_string; 166 char *token; 167 const char *delim = ","; 168 char *save_ptr; 169 170 if (input == NULL || option0 == NULL || 171 (option1 == NULL && type == SCALE_FACTOR)) 172 return VPX_CODEC_INVALID_PARAM; 173 174 input_string = strdup(input); 175 token = strtok_r(input_string, delim, &save_ptr); 176 for (i = 0; i < svc_ctx->spatial_layers; ++i) { 177 if (token != NULL) { 178 res = extract_option(type, token, option0 + i, option1 + i); 179 if (res != VPX_CODEC_OK) 180 break; 181 token = strtok_r(NULL, delim, &save_ptr); 182 } else { 183 break; 184 } 185 } 186 if (res == VPX_CODEC_OK && i != svc_ctx->spatial_layers) { 187 svc_log(svc_ctx, SVC_LOG_ERROR, 188 "svc: layer params type: %d %d values required, " 189 "but only %d specified\n", type, svc_ctx->spatial_layers, i); 190 res = VPX_CODEC_INVALID_PARAM; 191 } 192 free(input_string); 193 return res; 194} 195 196/** 197 * Parse SVC encoding options 198 * Format: encoding-mode=<svc_mode>,layers=<layer_count> 199 * scale-factors=<n1>/<d1>,<n2>/<d2>,... 200 * quantizers=<q1>,<q2>,... 201 * svc_mode = [i|ip|alt_ip|gf] 202 */ 203static vpx_codec_err_t parse_options(SvcContext *svc_ctx, const char *options) { 204 char *input_string; 205 char *option_name; 206 char *option_value; 207 char *input_ptr; 208 SvcInternal_t *const si = get_svc_internal(svc_ctx); 209 vpx_codec_err_t res = VPX_CODEC_OK; 210 int i, alt_ref_enabled = 0; 211 212 if (options == NULL) return VPX_CODEC_OK; 213 input_string = strdup(options); 214 215 // parse option name 216 option_name = strtok_r(input_string, "=", &input_ptr); 217 while (option_name != NULL) { 218 // parse option value 219 option_value = strtok_r(NULL, " ", &input_ptr); 220 if (option_value == NULL) { 221 svc_log(svc_ctx, SVC_LOG_ERROR, "option missing value: %s\n", 222 option_name); 223 res = VPX_CODEC_INVALID_PARAM; 224 break; 225 } 226 if (strcmp("spatial-layers", option_name) == 0) { 227 svc_ctx->spatial_layers = atoi(option_value); 228 } else if (strcmp("temporal-layers", option_name) == 0) { 229 svc_ctx->temporal_layers = atoi(option_value); 230 } else if (strcmp("scale-factors", option_name) == 0) { 231 res = parse_layer_options_from_string(svc_ctx, SCALE_FACTOR, option_value, 232 si->svc_params.scaling_factor_num, 233 si->svc_params.scaling_factor_den); 234 if (res != VPX_CODEC_OK) break; 235 } else if (strcmp("max-quantizers", option_name) == 0) { 236 res = parse_layer_options_from_string(svc_ctx, QUANTIZER, option_value, 237 si->svc_params.max_quantizers, 238 NULL); 239 if (res != VPX_CODEC_OK) break; 240 } else if (strcmp("min-quantizers", option_name) == 0) { 241 res = parse_layer_options_from_string(svc_ctx, QUANTIZER, option_value, 242 si->svc_params.min_quantizers, 243 NULL); 244 if (res != VPX_CODEC_OK) break; 245 } else if (strcmp("auto-alt-refs", option_name) == 0) { 246 res = parse_layer_options_from_string(svc_ctx, AUTO_ALT_REF, option_value, 247 si->enable_auto_alt_ref, NULL); 248 if (res != VPX_CODEC_OK) break; 249 } else if (strcmp("bitrates", option_name) == 0) { 250 res = parse_layer_options_from_string(svc_ctx, BITRATE, option_value, 251 si->bitrates, NULL); 252 if (res != VPX_CODEC_OK) break; 253 } else if (strcmp("multi-frame-contexts", option_name) == 0) { 254 si->use_multiple_frame_contexts = atoi(option_value); 255 } else { 256 svc_log(svc_ctx, SVC_LOG_ERROR, "invalid option: %s\n", option_name); 257 res = VPX_CODEC_INVALID_PARAM; 258 break; 259 } 260 option_name = strtok_r(NULL, "=", &input_ptr); 261 } 262 free(input_string); 263 264 for (i = 0; i < svc_ctx->spatial_layers; ++i) { 265 if (si->svc_params.max_quantizers[i] > MAX_QUANTIZER || 266 si->svc_params.max_quantizers[i] < 0 || 267 si->svc_params.min_quantizers[i] > si->svc_params.max_quantizers[i] || 268 si->svc_params.min_quantizers[i] < 0) 269 res = VPX_CODEC_INVALID_PARAM; 270 } 271 272 if (si->use_multiple_frame_contexts && 273 (svc_ctx->spatial_layers > 3 || 274 svc_ctx->spatial_layers * svc_ctx->temporal_layers > 4)) 275 res = VPX_CODEC_INVALID_PARAM; 276 277 for (i = 0; i < svc_ctx->spatial_layers; ++i) 278 alt_ref_enabled += si->enable_auto_alt_ref[i]; 279 if (alt_ref_enabled > REF_FRAMES - svc_ctx->spatial_layers) { 280 svc_log(svc_ctx, SVC_LOG_ERROR, 281 "svc: auto alt ref: Maxinum %d(REF_FRAMES - layers) layers could" 282 "enabled auto alt reference frame, but % layers are enabled\n", 283 REF_FRAMES - svc_ctx->spatial_layers, alt_ref_enabled); 284 res = VPX_CODEC_INVALID_PARAM; 285 } 286 287 return res; 288} 289 290vpx_codec_err_t vpx_svc_set_options(SvcContext *svc_ctx, 291 const char *options) { 292 SvcInternal_t *const si = get_svc_internal(svc_ctx); 293 if (svc_ctx == NULL || options == NULL || si == NULL) { 294 return VPX_CODEC_INVALID_PARAM; 295 } 296 strncpy(si->options, options, sizeof(si->options)); 297 si->options[sizeof(si->options) - 1] = '\0'; 298 return VPX_CODEC_OK; 299} 300 301void assign_layer_bitrates(const SvcContext *svc_ctx, 302 vpx_codec_enc_cfg_t *const enc_cfg) { 303 int i; 304 const SvcInternal_t *const si = get_const_svc_internal(svc_ctx); 305 int sl, tl, spatial_layer_target; 306 307 if (svc_ctx->temporal_layering_mode != 0) { 308 if (si->bitrates[0] != 0) { 309 enc_cfg->rc_target_bitrate = 0; 310 for (sl = 0; sl < svc_ctx->spatial_layers; ++sl) { 311 enc_cfg->ss_target_bitrate[sl*svc_ctx->temporal_layers] = 0; 312 for (tl = 0; tl < svc_ctx->temporal_layers; ++tl) { 313 enc_cfg->ss_target_bitrate[sl*svc_ctx->temporal_layers] 314 += (unsigned int)si->bitrates[sl * svc_ctx->temporal_layers + tl]; 315 enc_cfg->layer_target_bitrate[sl*svc_ctx->temporal_layers + tl] 316 = si->bitrates[sl * svc_ctx->temporal_layers + tl]; 317 } 318 } 319 } else { 320 float total = 0; 321 float alloc_ratio[VPX_MAX_LAYERS] = {0}; 322 323 for (sl = 0; sl < svc_ctx->spatial_layers; ++sl) { 324 if (si->svc_params.scaling_factor_den[sl] > 0) { 325 alloc_ratio[sl] = (float)(si->svc_params.scaling_factor_num[sl] * 326 1.0 / si->svc_params.scaling_factor_den[sl]); 327 total += alloc_ratio[sl]; 328 } 329 } 330 331 for (sl = 0; sl < svc_ctx->spatial_layers; ++sl) { 332 enc_cfg->ss_target_bitrate[sl] = spatial_layer_target = 333 (unsigned int)(enc_cfg->rc_target_bitrate * 334 alloc_ratio[sl] / total); 335 if (svc_ctx->temporal_layering_mode == 3) { 336 enc_cfg->layer_target_bitrate[sl * svc_ctx->temporal_layers] = 337 spatial_layer_target >> 1; 338 enc_cfg->layer_target_bitrate[sl * svc_ctx->temporal_layers + 1] = 339 (spatial_layer_target >> 1) + (spatial_layer_target >> 2); 340 enc_cfg->layer_target_bitrate[sl * svc_ctx->temporal_layers + 2] = 341 spatial_layer_target; 342 } else if (svc_ctx->temporal_layering_mode == 2 || 343 svc_ctx->temporal_layering_mode == 1) { 344 enc_cfg->layer_target_bitrate[sl * svc_ctx->temporal_layers] = 345 spatial_layer_target * 2 / 3; 346 enc_cfg->layer_target_bitrate[sl * svc_ctx->temporal_layers + 1] = 347 spatial_layer_target; 348 } else { 349 // User should explicitly assign bitrates in this case. 350 assert(0); 351 } 352 } 353 } 354 } else { 355 if (si->bitrates[0] != 0) { 356 enc_cfg->rc_target_bitrate = 0; 357 for (i = 0; i < svc_ctx->spatial_layers; ++i) { 358 enc_cfg->ss_target_bitrate[i] = (unsigned int)si->bitrates[i]; 359 enc_cfg->rc_target_bitrate += si->bitrates[i]; 360 } 361 } else { 362 float total = 0; 363 float alloc_ratio[VPX_MAX_LAYERS] = {0}; 364 365 for (i = 0; i < svc_ctx->spatial_layers; ++i) { 366 if (si->svc_params.scaling_factor_den[i] > 0) { 367 alloc_ratio[i] = (float)(si->svc_params.scaling_factor_num[i] * 1.0 / 368 si->svc_params.scaling_factor_den[i]); 369 370 alloc_ratio[i] *= alloc_ratio[i]; 371 total += alloc_ratio[i]; 372 } 373 } 374 for (i = 0; i < VPX_SS_MAX_LAYERS; ++i) { 375 if (total > 0) { 376 enc_cfg->layer_target_bitrate[i] = (unsigned int) 377 (enc_cfg->rc_target_bitrate * alloc_ratio[i] / total); 378 } 379 } 380 } 381 } 382} 383 384vpx_codec_err_t vpx_svc_init(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx, 385 vpx_codec_iface_t *iface, 386 vpx_codec_enc_cfg_t *enc_cfg) { 387 vpx_codec_err_t res; 388 int i; 389 SvcInternal_t *const si = get_svc_internal(svc_ctx); 390 if (svc_ctx == NULL || codec_ctx == NULL || iface == NULL || 391 enc_cfg == NULL) { 392 return VPX_CODEC_INVALID_PARAM; 393 } 394 if (si == NULL) return VPX_CODEC_MEM_ERROR; 395 396 si->codec_ctx = codec_ctx; 397 398 si->width = enc_cfg->g_w; 399 si->height = enc_cfg->g_h; 400 401 if (enc_cfg->kf_max_dist < 2) { 402 svc_log(svc_ctx, SVC_LOG_ERROR, "key frame distance too small: %d\n", 403 enc_cfg->kf_max_dist); 404 return VPX_CODEC_INVALID_PARAM; 405 } 406 si->kf_dist = enc_cfg->kf_max_dist; 407 408 if (svc_ctx->spatial_layers == 0) 409 svc_ctx->spatial_layers = VPX_SS_DEFAULT_LAYERS; 410 if (svc_ctx->spatial_layers < 1 || 411 svc_ctx->spatial_layers > VPX_SS_MAX_LAYERS) { 412 svc_log(svc_ctx, SVC_LOG_ERROR, "spatial layers: invalid value: %d\n", 413 svc_ctx->spatial_layers); 414 return VPX_CODEC_INVALID_PARAM; 415 } 416 417 // Note: temporal_layering_mode only applies to one-pass CBR 418 // si->svc_params.temporal_layering_mode = svc_ctx->temporal_layering_mode; 419 if (svc_ctx->temporal_layering_mode == 3) { 420 svc_ctx->temporal_layers = 3; 421 } else if (svc_ctx->temporal_layering_mode == 2 || 422 svc_ctx->temporal_layering_mode == 1) { 423 svc_ctx->temporal_layers = 2; 424 } 425 426 for (i = 0; i < VPX_SS_MAX_LAYERS; ++i) { 427 si->svc_params.max_quantizers[i] = MAX_QUANTIZER; 428 si->svc_params.min_quantizers[i] = 0; 429 si->svc_params.scaling_factor_num[i] = DEFAULT_SCALE_FACTORS_NUM[i]; 430 si->svc_params.scaling_factor_den[i] = DEFAULT_SCALE_FACTORS_DEN[i]; 431 } 432 433 // Parse aggregate command line options. Options must start with 434 // "layers=xx" then followed by other options 435 res = parse_options(svc_ctx, si->options); 436 if (res != VPX_CODEC_OK) return res; 437 438 if (svc_ctx->spatial_layers < 1) 439 svc_ctx->spatial_layers = 1; 440 if (svc_ctx->spatial_layers > VPX_SS_MAX_LAYERS) 441 svc_ctx->spatial_layers = VPX_SS_MAX_LAYERS; 442 443 if (svc_ctx->temporal_layers < 1) 444 svc_ctx->temporal_layers = 1; 445 if (svc_ctx->temporal_layers > VPX_TS_MAX_LAYERS) 446 svc_ctx->temporal_layers = VPX_TS_MAX_LAYERS; 447 448 if (svc_ctx->temporal_layers * svc_ctx->spatial_layers > VPX_MAX_LAYERS) { 449 svc_log(svc_ctx, SVC_LOG_ERROR, 450 "spatial layers * temporal layers exceeds the maximum number of " 451 "allowed layers of %d\n", 452 svc_ctx->spatial_layers * svc_ctx->temporal_layers, 453 (int) VPX_MAX_LAYERS); 454 return VPX_CODEC_INVALID_PARAM; 455 } 456 assign_layer_bitrates(svc_ctx, enc_cfg); 457 458#if CONFIG_SPATIAL_SVC 459 for (i = 0; i < svc_ctx->spatial_layers; ++i) 460 enc_cfg->ss_enable_auto_alt_ref[i] = si->enable_auto_alt_ref[i]; 461#endif 462 463 if (svc_ctx->temporal_layers > 1) { 464 int i; 465 for (i = 0; i < svc_ctx->temporal_layers; ++i) { 466 enc_cfg->ts_target_bitrate[i] = enc_cfg->rc_target_bitrate / 467 svc_ctx->temporal_layers; 468 enc_cfg->ts_rate_decimator[i] = 1 << (svc_ctx->temporal_layers - 1 - i); 469 } 470 } 471 472 if (svc_ctx->threads) 473 enc_cfg->g_threads = svc_ctx->threads; 474 475 // Modify encoder configuration 476 enc_cfg->ss_number_layers = svc_ctx->spatial_layers; 477 enc_cfg->ts_number_layers = svc_ctx->temporal_layers; 478 479 if (enc_cfg->rc_end_usage == VPX_CBR) { 480 enc_cfg->rc_resize_allowed = 0; 481 enc_cfg->rc_min_quantizer = 2; 482 enc_cfg->rc_max_quantizer = 56; 483 enc_cfg->rc_undershoot_pct = 50; 484 enc_cfg->rc_overshoot_pct = 50; 485 enc_cfg->rc_buf_initial_sz = 500; 486 enc_cfg->rc_buf_optimal_sz = 600; 487 enc_cfg->rc_buf_sz = 1000; 488 } 489 490 if (enc_cfg->g_error_resilient == 0 && si->use_multiple_frame_contexts == 0) 491 enc_cfg->g_error_resilient = 1; 492 493 // Initialize codec 494 res = vpx_codec_enc_init(codec_ctx, iface, enc_cfg, VPX_CODEC_USE_PSNR); 495 if (res != VPX_CODEC_OK) { 496 svc_log(svc_ctx, SVC_LOG_ERROR, "svc_enc_init error\n"); 497 return res; 498 } 499 if (svc_ctx->spatial_layers > 1 || svc_ctx->temporal_layers > 1) { 500 vpx_codec_control(codec_ctx, VP9E_SET_SVC, 1); 501 vpx_codec_control(codec_ctx, VP9E_SET_SVC_PARAMETERS, &si->svc_params); 502 } 503 return VPX_CODEC_OK; 504} 505 506/** 507 * Encode a frame into multiple layers 508 * Create a superframe containing the individual layers 509 */ 510vpx_codec_err_t vpx_svc_encode(SvcContext *svc_ctx, 511 vpx_codec_ctx_t *codec_ctx, 512 struct vpx_image *rawimg, 513 vpx_codec_pts_t pts, 514 int64_t duration, int deadline) { 515 vpx_codec_err_t res; 516 vpx_codec_iter_t iter; 517 const vpx_codec_cx_pkt_t *cx_pkt; 518 SvcInternal_t *const si = get_svc_internal(svc_ctx); 519 if (svc_ctx == NULL || codec_ctx == NULL || si == NULL) { 520 return VPX_CODEC_INVALID_PARAM; 521 } 522 523 svc_log_reset(svc_ctx); 524 525 res = vpx_codec_encode(codec_ctx, rawimg, pts, (uint32_t)duration, 0, 526 deadline); 527 if (res != VPX_CODEC_OK) { 528 return res; 529 } 530 // save compressed data 531 iter = NULL; 532 while ((cx_pkt = vpx_codec_get_cx_data(codec_ctx, &iter))) { 533 switch (cx_pkt->kind) { 534#if VPX_ENCODER_ABI_VERSION > (5 + VPX_CODEC_ABI_VERSION) 535#if CONFIG_SPATIAL_SVC 536 case VPX_CODEC_SPATIAL_SVC_LAYER_PSNR: { 537 int i; 538 for (i = 0; i < svc_ctx->spatial_layers; ++i) { 539 int j; 540 svc_log(svc_ctx, SVC_LOG_DEBUG, 541 "SVC frame: %d, layer: %d, PSNR(Total/Y/U/V): " 542 "%2.3f %2.3f %2.3f %2.3f \n", 543 si->psnr_pkt_received, i, 544 cx_pkt->data.layer_psnr[i].psnr[0], 545 cx_pkt->data.layer_psnr[i].psnr[1], 546 cx_pkt->data.layer_psnr[i].psnr[2], 547 cx_pkt->data.layer_psnr[i].psnr[3]); 548 svc_log(svc_ctx, SVC_LOG_DEBUG, 549 "SVC frame: %d, layer: %d, SSE(Total/Y/U/V): " 550 "%2.3f %2.3f %2.3f %2.3f \n", 551 si->psnr_pkt_received, i, 552 cx_pkt->data.layer_psnr[i].sse[0], 553 cx_pkt->data.layer_psnr[i].sse[1], 554 cx_pkt->data.layer_psnr[i].sse[2], 555 cx_pkt->data.layer_psnr[i].sse[3]); 556 557 for (j = 0; j < COMPONENTS; ++j) { 558 si->psnr_sum[i][j] += 559 cx_pkt->data.layer_psnr[i].psnr[j]; 560 si->sse_sum[i][j] += cx_pkt->data.layer_psnr[i].sse[j]; 561 } 562 } 563 ++si->psnr_pkt_received; 564 break; 565 } 566 case VPX_CODEC_SPATIAL_SVC_LAYER_SIZES: { 567 int i; 568 for (i = 0; i < svc_ctx->spatial_layers; ++i) 569 si->bytes_sum[i] += cx_pkt->data.layer_sizes[i]; 570 break; 571 } 572#endif 573#endif 574 default: { 575 break; 576 } 577 } 578 } 579 580 return VPX_CODEC_OK; 581} 582 583const char *vpx_svc_get_message(const SvcContext *svc_ctx) { 584 const SvcInternal_t *const si = get_const_svc_internal(svc_ctx); 585 if (svc_ctx == NULL || si == NULL) return NULL; 586 return si->message_buffer; 587} 588 589static double calc_psnr(double d) { 590 if (d == 0) return 100; 591 return -10.0 * log(d) / log(10.0); 592} 593 594// dump accumulated statistics and reset accumulated values 595const char *vpx_svc_dump_statistics(SvcContext *svc_ctx) { 596 int number_of_frames; 597 int i, j; 598 uint32_t bytes_total = 0; 599 double scale[COMPONENTS]; 600 double psnr[COMPONENTS]; 601 double mse[COMPONENTS]; 602 double y_scale; 603 604 SvcInternal_t *const si = get_svc_internal(svc_ctx); 605 if (svc_ctx == NULL || si == NULL) return NULL; 606 607 svc_log_reset(svc_ctx); 608 609 number_of_frames = si->psnr_pkt_received; 610 if (number_of_frames <= 0) return vpx_svc_get_message(svc_ctx); 611 612 svc_log(svc_ctx, SVC_LOG_INFO, "\n"); 613 for (i = 0; i < svc_ctx->spatial_layers; ++i) { 614 615 svc_log(svc_ctx, SVC_LOG_INFO, 616 "Layer %d Average PSNR=[%2.3f, %2.3f, %2.3f, %2.3f], Bytes=[%u]\n", 617 i, (double)si->psnr_sum[i][0] / number_of_frames, 618 (double)si->psnr_sum[i][1] / number_of_frames, 619 (double)si->psnr_sum[i][2] / number_of_frames, 620 (double)si->psnr_sum[i][3] / number_of_frames, si->bytes_sum[i]); 621 // the following psnr calculation is deduced from ffmpeg.c#print_report 622 y_scale = si->width * si->height * 255.0 * 255.0 * number_of_frames; 623 scale[1] = y_scale; 624 scale[2] = scale[3] = y_scale / 4; // U or V 625 scale[0] = y_scale * 1.5; // total 626 627 for (j = 0; j < COMPONENTS; j++) { 628 psnr[j] = calc_psnr(si->sse_sum[i][j] / scale[j]); 629 mse[j] = si->sse_sum[i][j] * 255.0 * 255.0 / scale[j]; 630 } 631 svc_log(svc_ctx, SVC_LOG_INFO, 632 "Layer %d Overall PSNR=[%2.3f, %2.3f, %2.3f, %2.3f]\n", i, psnr[0], 633 psnr[1], psnr[2], psnr[3]); 634 svc_log(svc_ctx, SVC_LOG_INFO, 635 "Layer %d Overall MSE=[%2.3f, %2.3f, %2.3f, %2.3f]\n", i, mse[0], 636 mse[1], mse[2], mse[3]); 637 638 bytes_total += si->bytes_sum[i]; 639 // Clear sums for next time. 640 si->bytes_sum[i] = 0; 641 for (j = 0; j < COMPONENTS; ++j) { 642 si->psnr_sum[i][j] = 0; 643 si->sse_sum[i][j] = 0; 644 } 645 } 646 647 // only display statistics once 648 si->psnr_pkt_received = 0; 649 650 svc_log(svc_ctx, SVC_LOG_INFO, "Total Bytes=[%u]\n", bytes_total); 651 return vpx_svc_get_message(svc_ctx); 652} 653 654void vpx_svc_release(SvcContext *svc_ctx) { 655 SvcInternal_t *si; 656 if (svc_ctx == NULL) return; 657 // do not use get_svc_internal as it will unnecessarily allocate an 658 // SvcInternal_t if it was not already allocated 659 si = (SvcInternal_t *)svc_ctx->internal; 660 if (si != NULL) { 661 free(si); 662 svc_ctx->internal = NULL; 663 } 664} 665 666