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 enc_cfg->layer_target_bitrate[sl * svc_ctx->temporal_layers] = 344 spatial_layer_target * 2 / 3; 345 enc_cfg->layer_target_bitrate[sl * svc_ctx->temporal_layers + 1] = 346 spatial_layer_target; 347 } else { 348 // User should explicitly assign bitrates in this case. 349 assert(0); 350 } 351 } 352 } 353 } else { 354 if (si->bitrates[0] != 0) { 355 enc_cfg->rc_target_bitrate = 0; 356 for (i = 0; i < svc_ctx->spatial_layers; ++i) { 357 enc_cfg->ss_target_bitrate[i] = (unsigned int)si->bitrates[i]; 358 enc_cfg->rc_target_bitrate += si->bitrates[i]; 359 } 360 } else { 361 float total = 0; 362 float alloc_ratio[VPX_MAX_LAYERS] = {0}; 363 364 for (i = 0; i < svc_ctx->spatial_layers; ++i) { 365 if (si->svc_params.scaling_factor_den[i] > 0) { 366 alloc_ratio[i] = (float)(si->svc_params.scaling_factor_num[i] * 1.0 / 367 si->svc_params.scaling_factor_den[i]); 368 369 alloc_ratio[i] *= alloc_ratio[i]; 370 total += alloc_ratio[i]; 371 } 372 } 373 for (i = 0; i < VPX_SS_MAX_LAYERS; ++i) { 374 if (total > 0) { 375 enc_cfg->layer_target_bitrate[i] = (unsigned int) 376 (enc_cfg->rc_target_bitrate * alloc_ratio[i] / total); 377 } 378 } 379 } 380 } 381} 382 383vpx_codec_err_t vpx_svc_init(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx, 384 vpx_codec_iface_t *iface, 385 vpx_codec_enc_cfg_t *enc_cfg) { 386 vpx_codec_err_t res; 387 int i; 388 SvcInternal_t *const si = get_svc_internal(svc_ctx); 389 if (svc_ctx == NULL || codec_ctx == NULL || iface == NULL || 390 enc_cfg == NULL) { 391 return VPX_CODEC_INVALID_PARAM; 392 } 393 if (si == NULL) return VPX_CODEC_MEM_ERROR; 394 395 si->codec_ctx = codec_ctx; 396 397 si->width = enc_cfg->g_w; 398 si->height = enc_cfg->g_h; 399 400 if (enc_cfg->kf_max_dist < 2) { 401 svc_log(svc_ctx, SVC_LOG_ERROR, "key frame distance too small: %d\n", 402 enc_cfg->kf_max_dist); 403 return VPX_CODEC_INVALID_PARAM; 404 } 405 si->kf_dist = enc_cfg->kf_max_dist; 406 407 if (svc_ctx->spatial_layers == 0) 408 svc_ctx->spatial_layers = VPX_SS_DEFAULT_LAYERS; 409 if (svc_ctx->spatial_layers < 1 || 410 svc_ctx->spatial_layers > VPX_SS_MAX_LAYERS) { 411 svc_log(svc_ctx, SVC_LOG_ERROR, "spatial layers: invalid value: %d\n", 412 svc_ctx->spatial_layers); 413 return VPX_CODEC_INVALID_PARAM; 414 } 415 416 // Note: temporal_layering_mode only applies to one-pass CBR 417 // si->svc_params.temporal_layering_mode = svc_ctx->temporal_layering_mode; 418 if (svc_ctx->temporal_layering_mode == 3) { 419 svc_ctx->temporal_layers = 3; 420 } else if (svc_ctx->temporal_layering_mode == 2) { 421 svc_ctx->temporal_layers = 2; 422 } 423 424 for (i = 0; i < VPX_SS_MAX_LAYERS; ++i) { 425 si->svc_params.max_quantizers[i] = MAX_QUANTIZER; 426 si->svc_params.min_quantizers[i] = 0; 427 si->svc_params.scaling_factor_num[i] = DEFAULT_SCALE_FACTORS_NUM[i]; 428 si->svc_params.scaling_factor_den[i] = DEFAULT_SCALE_FACTORS_DEN[i]; 429 } 430 431 // Parse aggregate command line options. Options must start with 432 // "layers=xx" then followed by other options 433 res = parse_options(svc_ctx, si->options); 434 if (res != VPX_CODEC_OK) return res; 435 436 if (svc_ctx->spatial_layers < 1) 437 svc_ctx->spatial_layers = 1; 438 if (svc_ctx->spatial_layers > VPX_SS_MAX_LAYERS) 439 svc_ctx->spatial_layers = VPX_SS_MAX_LAYERS; 440 441 if (svc_ctx->temporal_layers < 1) 442 svc_ctx->temporal_layers = 1; 443 if (svc_ctx->temporal_layers > VPX_TS_MAX_LAYERS) 444 svc_ctx->temporal_layers = VPX_TS_MAX_LAYERS; 445 446 if (svc_ctx->temporal_layers * svc_ctx->spatial_layers > VPX_MAX_LAYERS) { 447 svc_log(svc_ctx, SVC_LOG_ERROR, 448 "spatial layers * temporal layers exceeds the maximum number of " 449 "allowed layers of %d\n", 450 svc_ctx->spatial_layers * svc_ctx->temporal_layers, 451 (int) VPX_MAX_LAYERS); 452 return VPX_CODEC_INVALID_PARAM; 453 } 454 assign_layer_bitrates(svc_ctx, enc_cfg); 455 456#if CONFIG_SPATIAL_SVC 457 for (i = 0; i < svc_ctx->spatial_layers; ++i) 458 enc_cfg->ss_enable_auto_alt_ref[i] = si->enable_auto_alt_ref[i]; 459#endif 460 461 if (svc_ctx->temporal_layers > 1) { 462 int i; 463 for (i = 0; i < svc_ctx->temporal_layers; ++i) { 464 enc_cfg->ts_target_bitrate[i] = enc_cfg->rc_target_bitrate / 465 svc_ctx->temporal_layers; 466 enc_cfg->ts_rate_decimator[i] = 1 << (svc_ctx->temporal_layers - 1 - i); 467 } 468 } 469 470 if (svc_ctx->threads) 471 enc_cfg->g_threads = svc_ctx->threads; 472 473 // Modify encoder configuration 474 enc_cfg->ss_number_layers = svc_ctx->spatial_layers; 475 enc_cfg->ts_number_layers = svc_ctx->temporal_layers; 476 477 if (enc_cfg->rc_end_usage == VPX_CBR) { 478 enc_cfg->rc_resize_allowed = 0; 479 enc_cfg->rc_min_quantizer = 2; 480 enc_cfg->rc_max_quantizer = 63; 481 enc_cfg->rc_undershoot_pct = 50; 482 enc_cfg->rc_overshoot_pct = 50; 483 enc_cfg->rc_buf_initial_sz = 20; 484 enc_cfg->rc_buf_optimal_sz = 600; 485 enc_cfg->rc_buf_sz = 1000; 486 } 487 488 if (enc_cfg->g_error_resilient == 0 && si->use_multiple_frame_contexts == 0) 489 enc_cfg->g_error_resilient = 1; 490 491 // Initialize codec 492 res = vpx_codec_enc_init(codec_ctx, iface, enc_cfg, VPX_CODEC_USE_PSNR); 493 if (res != VPX_CODEC_OK) { 494 svc_log(svc_ctx, SVC_LOG_ERROR, "svc_enc_init error\n"); 495 return res; 496 } 497 498 vpx_codec_control(codec_ctx, VP9E_SET_SVC, 1); 499 vpx_codec_control(codec_ctx, VP9E_SET_SVC_PARAMETERS, &si->svc_params); 500 501 return VPX_CODEC_OK; 502} 503 504/** 505 * Encode a frame into multiple layers 506 * Create a superframe containing the individual layers 507 */ 508vpx_codec_err_t vpx_svc_encode(SvcContext *svc_ctx, 509 vpx_codec_ctx_t *codec_ctx, 510 struct vpx_image *rawimg, 511 vpx_codec_pts_t pts, 512 int64_t duration, int deadline) { 513 vpx_codec_err_t res; 514 vpx_codec_iter_t iter; 515 const vpx_codec_cx_pkt_t *cx_pkt; 516 SvcInternal_t *const si = get_svc_internal(svc_ctx); 517 if (svc_ctx == NULL || codec_ctx == NULL || si == NULL) { 518 return VPX_CODEC_INVALID_PARAM; 519 } 520 521 svc_log_reset(svc_ctx); 522 523 res = vpx_codec_encode(codec_ctx, rawimg, pts, (uint32_t)duration, 0, 524 deadline); 525 if (res != VPX_CODEC_OK) { 526 return res; 527 } 528 // save compressed data 529 iter = NULL; 530 while ((cx_pkt = vpx_codec_get_cx_data(codec_ctx, &iter))) { 531 switch (cx_pkt->kind) { 532#if VPX_ENCODER_ABI_VERSION > (5 + VPX_CODEC_ABI_VERSION) 533#if CONFIG_SPATIAL_SVC 534 case VPX_CODEC_SPATIAL_SVC_LAYER_PSNR: { 535 int i; 536 for (i = 0; i < svc_ctx->spatial_layers; ++i) { 537 int j; 538 svc_log(svc_ctx, SVC_LOG_DEBUG, 539 "SVC frame: %d, layer: %d, PSNR(Total/Y/U/V): " 540 "%2.3f %2.3f %2.3f %2.3f \n", 541 si->psnr_pkt_received, i, 542 cx_pkt->data.layer_psnr[i].psnr[0], 543 cx_pkt->data.layer_psnr[i].psnr[1], 544 cx_pkt->data.layer_psnr[i].psnr[2], 545 cx_pkt->data.layer_psnr[i].psnr[3]); 546 svc_log(svc_ctx, SVC_LOG_DEBUG, 547 "SVC frame: %d, layer: %d, SSE(Total/Y/U/V): " 548 "%2.3f %2.3f %2.3f %2.3f \n", 549 si->psnr_pkt_received, i, 550 cx_pkt->data.layer_psnr[i].sse[0], 551 cx_pkt->data.layer_psnr[i].sse[1], 552 cx_pkt->data.layer_psnr[i].sse[2], 553 cx_pkt->data.layer_psnr[i].sse[3]); 554 555 for (j = 0; j < COMPONENTS; ++j) { 556 si->psnr_sum[i][j] += 557 cx_pkt->data.layer_psnr[i].psnr[j]; 558 si->sse_sum[i][j] += cx_pkt->data.layer_psnr[i].sse[j]; 559 } 560 } 561 ++si->psnr_pkt_received; 562 break; 563 } 564 case VPX_CODEC_SPATIAL_SVC_LAYER_SIZES: { 565 int i; 566 for (i = 0; i < svc_ctx->spatial_layers; ++i) 567 si->bytes_sum[i] += cx_pkt->data.layer_sizes[i]; 568 break; 569 } 570#endif 571#endif 572 default: { 573 break; 574 } 575 } 576 } 577 578 return VPX_CODEC_OK; 579} 580 581const char *vpx_svc_get_message(const SvcContext *svc_ctx) { 582 const SvcInternal_t *const si = get_const_svc_internal(svc_ctx); 583 if (svc_ctx == NULL || si == NULL) return NULL; 584 return si->message_buffer; 585} 586 587static double calc_psnr(double d) { 588 if (d == 0) return 100; 589 return -10.0 * log(d) / log(10.0); 590} 591 592// dump accumulated statistics and reset accumulated values 593const char *vpx_svc_dump_statistics(SvcContext *svc_ctx) { 594 int number_of_frames; 595 int i, j; 596 uint32_t bytes_total = 0; 597 double scale[COMPONENTS]; 598 double psnr[COMPONENTS]; 599 double mse[COMPONENTS]; 600 double y_scale; 601 602 SvcInternal_t *const si = get_svc_internal(svc_ctx); 603 if (svc_ctx == NULL || si == NULL) return NULL; 604 605 svc_log_reset(svc_ctx); 606 607 number_of_frames = si->psnr_pkt_received; 608 if (number_of_frames <= 0) return vpx_svc_get_message(svc_ctx); 609 610 svc_log(svc_ctx, SVC_LOG_INFO, "\n"); 611 for (i = 0; i < svc_ctx->spatial_layers; ++i) { 612 613 svc_log(svc_ctx, SVC_LOG_INFO, 614 "Layer %d Average PSNR=[%2.3f, %2.3f, %2.3f, %2.3f], Bytes=[%u]\n", 615 i, (double)si->psnr_sum[i][0] / number_of_frames, 616 (double)si->psnr_sum[i][1] / number_of_frames, 617 (double)si->psnr_sum[i][2] / number_of_frames, 618 (double)si->psnr_sum[i][3] / number_of_frames, si->bytes_sum[i]); 619 // the following psnr calculation is deduced from ffmpeg.c#print_report 620 y_scale = si->width * si->height * 255.0 * 255.0 * number_of_frames; 621 scale[1] = y_scale; 622 scale[2] = scale[3] = y_scale / 4; // U or V 623 scale[0] = y_scale * 1.5; // total 624 625 for (j = 0; j < COMPONENTS; j++) { 626 psnr[j] = calc_psnr(si->sse_sum[i][j] / scale[j]); 627 mse[j] = si->sse_sum[i][j] * 255.0 * 255.0 / scale[j]; 628 } 629 svc_log(svc_ctx, SVC_LOG_INFO, 630 "Layer %d Overall PSNR=[%2.3f, %2.3f, %2.3f, %2.3f]\n", i, psnr[0], 631 psnr[1], psnr[2], psnr[3]); 632 svc_log(svc_ctx, SVC_LOG_INFO, 633 "Layer %d Overall MSE=[%2.3f, %2.3f, %2.3f, %2.3f]\n", i, mse[0], 634 mse[1], mse[2], mse[3]); 635 636 bytes_total += si->bytes_sum[i]; 637 // Clear sums for next time. 638 si->bytes_sum[i] = 0; 639 for (j = 0; j < COMPONENTS; ++j) { 640 si->psnr_sum[i][j] = 0; 641 si->sse_sum[i][j] = 0; 642 } 643 } 644 645 // only display statistics once 646 si->psnr_pkt_received = 0; 647 648 svc_log(svc_ctx, SVC_LOG_INFO, "Total Bytes=[%u]\n", bytes_total); 649 return vpx_svc_get_message(svc_ctx); 650} 651 652void vpx_svc_release(SvcContext *svc_ctx) { 653 SvcInternal_t *si; 654 if (svc_ctx == NULL) return; 655 // do not use get_svc_internal as it will unnecessarily allocate an 656 // SvcInternal_t if it was not already allocated 657 si = (SvcInternal_t *)svc_ctx->internal; 658 if (si != NULL) { 659 free(si); 660 svc_ctx->internal = NULL; 661 } 662} 663 664