You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

823 lines
29KB

  1. /*
  2. * Copyright (C) 2011-2013 Michael Niedermayer (michaelni@gmx.at)
  3. *
  4. * This file is part of libswresample
  5. *
  6. * libswresample is free software; you can redistribute it and/or
  7. * modify it under the terms of the GNU Lesser General Public
  8. * License as published by the Free Software Foundation; either
  9. * version 2.1 of the License, or (at your option) any later version.
  10. *
  11. * libswresample is distributed in the hope that it will be useful,
  12. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  14. * Lesser General Public License for more details.
  15. *
  16. * You should have received a copy of the GNU Lesser General Public
  17. * License along with libswresample; if not, write to the Free Software
  18. * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  19. */
  20. #include "libavutil/opt.h"
  21. #include "swresample_internal.h"
  22. #include "audioconvert.h"
  23. #include "libavutil/avassert.h"
  24. #include "libavutil/channel_layout.h"
  25. #include <float.h>
  26. #define ALIGN 32
  27. unsigned swresample_version(void)
  28. {
  29. av_assert0(LIBSWRESAMPLE_VERSION_MICRO >= 100);
  30. return LIBSWRESAMPLE_VERSION_INT;
  31. }
  32. const char *swresample_configuration(void)
  33. {
  34. return FFMPEG_CONFIGURATION;
  35. }
  36. const char *swresample_license(void)
  37. {
  38. #define LICENSE_PREFIX "libswresample license: "
  39. return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
  40. }
  41. int swr_set_channel_mapping(struct SwrContext *s, const int *channel_map){
  42. if(!s || s->in_convert) // s needs to be allocated but not initialized
  43. return AVERROR(EINVAL);
  44. s->channel_map = channel_map;
  45. return 0;
  46. }
  47. struct SwrContext *swr_alloc_set_opts(struct SwrContext *s,
  48. int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
  49. int64_t in_ch_layout, enum AVSampleFormat in_sample_fmt, int in_sample_rate,
  50. int log_offset, void *log_ctx){
  51. if(!s) s= swr_alloc();
  52. if(!s) return NULL;
  53. s->log_level_offset= log_offset;
  54. s->log_ctx= log_ctx;
  55. av_opt_set_int(s, "ocl", out_ch_layout, 0);
  56. av_opt_set_int(s, "osf", out_sample_fmt, 0);
  57. av_opt_set_int(s, "osr", out_sample_rate, 0);
  58. av_opt_set_int(s, "icl", in_ch_layout, 0);
  59. av_opt_set_int(s, "isf", in_sample_fmt, 0);
  60. av_opt_set_int(s, "isr", in_sample_rate, 0);
  61. av_opt_set_int(s, "tsf", AV_SAMPLE_FMT_NONE, 0);
  62. av_opt_set_int(s, "ich", av_get_channel_layout_nb_channels(s-> in_ch_layout), 0);
  63. av_opt_set_int(s, "och", av_get_channel_layout_nb_channels(s->out_ch_layout), 0);
  64. av_opt_set_int(s, "uch", 0, 0);
  65. return s;
  66. }
  67. static void set_audiodata_fmt(AudioData *a, enum AVSampleFormat fmt){
  68. a->fmt = fmt;
  69. a->bps = av_get_bytes_per_sample(fmt);
  70. a->planar= av_sample_fmt_is_planar(fmt);
  71. if (a->ch_count == 1)
  72. a->planar = 1;
  73. }
  74. static void free_temp(AudioData *a){
  75. av_free(a->data);
  76. memset(a, 0, sizeof(*a));
  77. }
  78. static void clear_context(SwrContext *s){
  79. s->in_buffer_index= 0;
  80. s->in_buffer_count= 0;
  81. s->resample_in_constraint= 0;
  82. memset(s->in.ch, 0, sizeof(s->in.ch));
  83. memset(s->out.ch, 0, sizeof(s->out.ch));
  84. free_temp(&s->postin);
  85. free_temp(&s->midbuf);
  86. free_temp(&s->preout);
  87. free_temp(&s->in_buffer);
  88. free_temp(&s->silence);
  89. free_temp(&s->drop_temp);
  90. free_temp(&s->dither.noise);
  91. free_temp(&s->dither.temp);
  92. swri_audio_convert_free(&s-> in_convert);
  93. swri_audio_convert_free(&s->out_convert);
  94. swri_audio_convert_free(&s->full_convert);
  95. swri_rematrix_free(s);
  96. s->flushed = 0;
  97. }
  98. av_cold void swr_free(SwrContext **ss){
  99. SwrContext *s= *ss;
  100. if(s){
  101. clear_context(s);
  102. if (s->resampler)
  103. s->resampler->free(&s->resample);
  104. }
  105. av_freep(ss);
  106. }
  107. av_cold void swr_close(SwrContext *s){
  108. clear_context(s);
  109. }
  110. av_cold int swr_init(struct SwrContext *s){
  111. int ret;
  112. clear_context(s);
  113. if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
  114. av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
  115. return AVERROR(EINVAL);
  116. }
  117. if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
  118. av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
  119. return AVERROR(EINVAL);
  120. }
  121. if(av_get_channel_layout_nb_channels(s-> in_ch_layout) > SWR_CH_MAX) {
  122. av_log(s, AV_LOG_WARNING, "Input channel layout 0x%"PRIx64" is invalid or unsupported.\n", s-> in_ch_layout);
  123. s->in_ch_layout = 0;
  124. }
  125. if(av_get_channel_layout_nb_channels(s->out_ch_layout) > SWR_CH_MAX) {
  126. av_log(s, AV_LOG_WARNING, "Output channel layout 0x%"PRIx64" is invalid or unsupported.\n", s->out_ch_layout);
  127. s->out_ch_layout = 0;
  128. }
  129. switch(s->engine){
  130. #if CONFIG_LIBSOXR
  131. extern struct Resampler const soxr_resampler;
  132. case SWR_ENGINE_SOXR: s->resampler = &soxr_resampler; break;
  133. #endif
  134. case SWR_ENGINE_SWR : s->resampler = &swri_resampler; break;
  135. default:
  136. av_log(s, AV_LOG_ERROR, "Requested resampling engine is unavailable\n");
  137. return AVERROR(EINVAL);
  138. }
  139. if(!s->used_ch_count)
  140. s->used_ch_count= s->in.ch_count;
  141. if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
  142. av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
  143. s-> in_ch_layout= 0;
  144. }
  145. if(!s-> in_ch_layout)
  146. s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
  147. if(!s->out_ch_layout)
  148. s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
  149. s->rematrix= s->out_ch_layout !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
  150. s->rematrix_custom;
  151. if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
  152. if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_S16P){
  153. s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
  154. }else if( av_get_planar_sample_fmt(s-> in_sample_fmt) == AV_SAMPLE_FMT_S32P
  155. && av_get_planar_sample_fmt(s->out_sample_fmt) == AV_SAMPLE_FMT_S32P
  156. && !s->rematrix
  157. && s->engine != SWR_ENGINE_SOXR){
  158. s->int_sample_fmt= AV_SAMPLE_FMT_S32P;
  159. }else if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_FLTP){
  160. s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
  161. }else{
  162. av_log(s, AV_LOG_DEBUG, "Using double precision mode\n");
  163. s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
  164. }
  165. }
  166. if( s->int_sample_fmt != AV_SAMPLE_FMT_S16P
  167. &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
  168. &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
  169. &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
  170. av_log(s, AV_LOG_ERROR, "Requested sample format %s is not supported internally, S16/S32/FLT/DBL is supported\n", av_get_sample_fmt_name(s->int_sample_fmt));
  171. return AVERROR(EINVAL);
  172. }
  173. set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
  174. set_audiodata_fmt(&s->out, s->out_sample_fmt);
  175. if (s->firstpts_in_samples != AV_NOPTS_VALUE) {
  176. if (!s->async && s->min_compensation >= FLT_MAX/2)
  177. s->async = 1;
  178. s->firstpts =
  179. s->outpts = s->firstpts_in_samples * s->out_sample_rate;
  180. } else
  181. s->firstpts = AV_NOPTS_VALUE;
  182. if (s->async) {
  183. if (s->min_compensation >= FLT_MAX/2)
  184. s->min_compensation = 0.001;
  185. if (s->async > 1.0001) {
  186. s->max_soft_compensation = s->async / (double) s->in_sample_rate;
  187. }
  188. }
  189. if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
  190. s->resample = s->resampler->init(s->resample, s->out_sample_rate, s->in_sample_rate, s->filter_size, s->phase_shift, s->linear_interp, s->cutoff, s->int_sample_fmt, s->filter_type, s->kaiser_beta, s->precision, s->cheby);
  191. }else
  192. s->resampler->free(&s->resample);
  193. if( s->int_sample_fmt != AV_SAMPLE_FMT_S16P
  194. && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
  195. && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
  196. && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
  197. && s->resample){
  198. av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16/s32/flt/dbl\n");
  199. return -1;
  200. }
  201. #define RSC 1 //FIXME finetune
  202. if(!s-> in.ch_count)
  203. s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
  204. if(!s->used_ch_count)
  205. s->used_ch_count= s->in.ch_count;
  206. if(!s->out.ch_count)
  207. s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
  208. if(!s-> in.ch_count){
  209. av_assert0(!s->in_ch_layout);
  210. av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
  211. return -1;
  212. }
  213. if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
  214. char l1[1024], l2[1024];
  215. av_get_channel_layout_string(l1, sizeof(l1), s-> in.ch_count, s-> in_ch_layout);
  216. av_get_channel_layout_string(l2, sizeof(l2), s->out.ch_count, s->out_ch_layout);
  217. av_log(s, AV_LOG_ERROR, "Rematrix is needed between %s and %s "
  218. "but there is not enough information to do it\n", l1, l2);
  219. return -1;
  220. }
  221. av_assert0(s->used_ch_count);
  222. av_assert0(s->out.ch_count);
  223. s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
  224. s->in_buffer= s->in;
  225. s->silence = s->in;
  226. s->drop_temp= s->out;
  227. if(!s->resample && !s->rematrix && !s->channel_map && !s->dither.method){
  228. s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
  229. s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
  230. return 0;
  231. }
  232. s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
  233. s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
  234. s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
  235. s->int_sample_fmt, s->out.ch_count, NULL, 0);
  236. if (!s->in_convert || !s->out_convert)
  237. return AVERROR(ENOMEM);
  238. s->postin= s->in;
  239. s->preout= s->out;
  240. s->midbuf= s->in;
  241. if(s->channel_map){
  242. s->postin.ch_count=
  243. s->midbuf.ch_count= s->used_ch_count;
  244. if(s->resample)
  245. s->in_buffer.ch_count= s->used_ch_count;
  246. }
  247. if(!s->resample_first){
  248. s->midbuf.ch_count= s->out.ch_count;
  249. if(s->resample)
  250. s->in_buffer.ch_count = s->out.ch_count;
  251. }
  252. set_audiodata_fmt(&s->postin, s->int_sample_fmt);
  253. set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
  254. set_audiodata_fmt(&s->preout, s->int_sample_fmt);
  255. if(s->resample){
  256. set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
  257. }
  258. if ((ret = swri_dither_init(s, s->out_sample_fmt, s->int_sample_fmt)) < 0)
  259. return ret;
  260. if(s->rematrix || s->dither.method)
  261. return swri_rematrix_init(s);
  262. return 0;
  263. }
  264. int swri_realloc_audio(AudioData *a, int count){
  265. int i, countb;
  266. AudioData old;
  267. if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
  268. return AVERROR(EINVAL);
  269. if(a->count >= count)
  270. return 0;
  271. count*=2;
  272. countb= FFALIGN(count*a->bps, ALIGN);
  273. old= *a;
  274. av_assert0(a->bps);
  275. av_assert0(a->ch_count);
  276. a->data= av_mallocz(countb*a->ch_count);
  277. if(!a->data)
  278. return AVERROR(ENOMEM);
  279. for(i=0; i<a->ch_count; i++){
  280. a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
  281. if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
  282. }
  283. if(!a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
  284. av_freep(&old.data);
  285. a->count= count;
  286. return 1;
  287. }
  288. static void copy(AudioData *out, AudioData *in,
  289. int count){
  290. av_assert0(out->planar == in->planar);
  291. av_assert0(out->bps == in->bps);
  292. av_assert0(out->ch_count == in->ch_count);
  293. if(out->planar){
  294. int ch;
  295. for(ch=0; ch<out->ch_count; ch++)
  296. memcpy(out->ch[ch], in->ch[ch], count*out->bps);
  297. }else
  298. memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
  299. }
  300. static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
  301. int i;
  302. if(!in_arg){
  303. memset(out->ch, 0, sizeof(out->ch));
  304. }else if(out->planar){
  305. for(i=0; i<out->ch_count; i++)
  306. out->ch[i]= in_arg[i];
  307. }else{
  308. for(i=0; i<out->ch_count; i++)
  309. out->ch[i]= in_arg[0] + i*out->bps;
  310. }
  311. }
  312. static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
  313. int i;
  314. if(out->planar){
  315. for(i=0; i<out->ch_count; i++)
  316. in_arg[i]= out->ch[i];
  317. }else{
  318. in_arg[0]= out->ch[0];
  319. }
  320. }
  321. /**
  322. *
  323. * out may be equal in.
  324. */
  325. static void buf_set(AudioData *out, AudioData *in, int count){
  326. int ch;
  327. if(in->planar){
  328. for(ch=0; ch<out->ch_count; ch++)
  329. out->ch[ch]= in->ch[ch] + count*out->bps;
  330. }else{
  331. for(ch=out->ch_count-1; ch>=0; ch--)
  332. out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
  333. }
  334. }
  335. /**
  336. *
  337. * @return number of samples output per channel
  338. */
  339. static int resample(SwrContext *s, AudioData *out_param, int out_count,
  340. const AudioData * in_param, int in_count){
  341. AudioData in, out, tmp;
  342. int ret_sum=0;
  343. int border=0;
  344. int padless = ARCH_X86 && s->engine == SWR_ENGINE_SWR ? 7 : 0;
  345. av_assert1(s->in_buffer.ch_count == in_param->ch_count);
  346. av_assert1(s->in_buffer.planar == in_param->planar);
  347. av_assert1(s->in_buffer.fmt == in_param->fmt);
  348. tmp=out=*out_param;
  349. in = *in_param;
  350. border = s->resampler->invert_initial_buffer(s->resample, &s->in_buffer,
  351. &in, in_count, &s->in_buffer_index, &s->in_buffer_count);
  352. if (border == INT_MAX) return 0;
  353. else if (border < 0) return border;
  354. else if (border) { buf_set(&in, &in, border); in_count -= border; s->resample_in_constraint = 0; }
  355. do{
  356. int ret, size, consumed;
  357. if(!s->resample_in_constraint && s->in_buffer_count){
  358. buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
  359. ret= s->resampler->multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
  360. out_count -= ret;
  361. ret_sum += ret;
  362. buf_set(&out, &out, ret);
  363. s->in_buffer_count -= consumed;
  364. s->in_buffer_index += consumed;
  365. if(!in_count)
  366. break;
  367. if(s->in_buffer_count <= border){
  368. buf_set(&in, &in, -s->in_buffer_count);
  369. in_count += s->in_buffer_count;
  370. s->in_buffer_count=0;
  371. s->in_buffer_index=0;
  372. border = 0;
  373. }
  374. }
  375. if((s->flushed || in_count > padless) && !s->in_buffer_count){
  376. s->in_buffer_index=0;
  377. ret= s->resampler->multiple_resample(s->resample, &out, out_count, &in, FFMAX(in_count-padless, 0), &consumed);
  378. out_count -= ret;
  379. ret_sum += ret;
  380. buf_set(&out, &out, ret);
  381. in_count -= consumed;
  382. buf_set(&in, &in, consumed);
  383. }
  384. //TODO is this check sane considering the advanced copy avoidance below
  385. size= s->in_buffer_index + s->in_buffer_count + in_count;
  386. if( size > s->in_buffer.count
  387. && s->in_buffer_count + in_count <= s->in_buffer_index){
  388. buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
  389. copy(&s->in_buffer, &tmp, s->in_buffer_count);
  390. s->in_buffer_index=0;
  391. }else
  392. if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
  393. return ret;
  394. if(in_count){
  395. int count= in_count;
  396. if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
  397. buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
  398. copy(&tmp, &in, /*in_*/count);
  399. s->in_buffer_count += count;
  400. in_count -= count;
  401. border += count;
  402. buf_set(&in, &in, count);
  403. s->resample_in_constraint= 0;
  404. if(s->in_buffer_count != count || in_count)
  405. continue;
  406. if (padless) {
  407. padless = 0;
  408. continue;
  409. }
  410. }
  411. break;
  412. }while(1);
  413. s->resample_in_constraint= !!out_count;
  414. return ret_sum;
  415. }
  416. static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
  417. AudioData *in , int in_count){
  418. AudioData *postin, *midbuf, *preout;
  419. int ret/*, in_max*/;
  420. AudioData preout_tmp, midbuf_tmp;
  421. if(s->full_convert){
  422. av_assert0(!s->resample);
  423. swri_audio_convert(s->full_convert, out, in, in_count);
  424. return out_count;
  425. }
  426. // in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
  427. // in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
  428. if((ret=swri_realloc_audio(&s->postin, in_count))<0)
  429. return ret;
  430. if(s->resample_first){
  431. av_assert0(s->midbuf.ch_count == s->used_ch_count);
  432. if((ret=swri_realloc_audio(&s->midbuf, out_count))<0)
  433. return ret;
  434. }else{
  435. av_assert0(s->midbuf.ch_count == s->out.ch_count);
  436. if((ret=swri_realloc_audio(&s->midbuf, in_count))<0)
  437. return ret;
  438. }
  439. if((ret=swri_realloc_audio(&s->preout, out_count))<0)
  440. return ret;
  441. postin= &s->postin;
  442. midbuf_tmp= s->midbuf;
  443. midbuf= &midbuf_tmp;
  444. preout_tmp= s->preout;
  445. preout= &preout_tmp;
  446. if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
  447. postin= in;
  448. if(s->resample_first ? !s->resample : !s->rematrix)
  449. midbuf= postin;
  450. if(s->resample_first ? !s->rematrix : !s->resample)
  451. preout= midbuf;
  452. if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar
  453. && !(s->out_sample_fmt==AV_SAMPLE_FMT_S32P && (s->dither.output_sample_bits&31))){
  454. if(preout==in){
  455. out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
  456. av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
  457. copy(out, in, out_count);
  458. return out_count;
  459. }
  460. else if(preout==postin) preout= midbuf= postin= out;
  461. else if(preout==midbuf) preout= midbuf= out;
  462. else preout= out;
  463. }
  464. if(in != postin){
  465. swri_audio_convert(s->in_convert, postin, in, in_count);
  466. }
  467. if(s->resample_first){
  468. if(postin != midbuf)
  469. out_count= resample(s, midbuf, out_count, postin, in_count);
  470. if(midbuf != preout)
  471. swri_rematrix(s, preout, midbuf, out_count, preout==out);
  472. }else{
  473. if(postin != midbuf)
  474. swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
  475. if(midbuf != preout)
  476. out_count= resample(s, preout, out_count, midbuf, in_count);
  477. }
  478. if(preout != out && out_count){
  479. AudioData *conv_src = preout;
  480. if(s->dither.method){
  481. int ch;
  482. int dither_count= FFMAX(out_count, 1<<16);
  483. if (preout == in) {
  484. conv_src = &s->dither.temp;
  485. if((ret=swri_realloc_audio(&s->dither.temp, dither_count))<0)
  486. return ret;
  487. }
  488. if((ret=swri_realloc_audio(&s->dither.noise, dither_count))<0)
  489. return ret;
  490. if(ret)
  491. for(ch=0; ch<s->dither.noise.ch_count; ch++)
  492. swri_get_dither(s, s->dither.noise.ch[ch], s->dither.noise.count, 12345678913579<<ch, s->dither.noise.fmt);
  493. av_assert0(s->dither.noise.ch_count == preout->ch_count);
  494. if(s->dither.noise_pos + out_count > s->dither.noise.count)
  495. s->dither.noise_pos = 0;
  496. if (s->dither.method < SWR_DITHER_NS){
  497. if (s->mix_2_1_simd) {
  498. int len1= out_count&~15;
  499. int off = len1 * preout->bps;
  500. if(len1)
  501. for(ch=0; ch<preout->ch_count; ch++)
  502. s->mix_2_1_simd(conv_src->ch[ch], preout->ch[ch], s->dither.noise.ch[ch] + s->dither.noise.bps * s->dither.noise_pos, s->native_simd_one, 0, 0, len1);
  503. if(out_count != len1)
  504. for(ch=0; ch<preout->ch_count; ch++)
  505. s->mix_2_1_f(conv_src->ch[ch] + off, preout->ch[ch] + off, s->dither.noise.ch[ch] + s->dither.noise.bps * s->dither.noise_pos + off + len1, s->native_one, 0, 0, out_count - len1);
  506. } else {
  507. for(ch=0; ch<preout->ch_count; ch++)
  508. s->mix_2_1_f(conv_src->ch[ch], preout->ch[ch], s->dither.noise.ch[ch] + s->dither.noise.bps * s->dither.noise_pos, s->native_one, 0, 0, out_count);
  509. }
  510. } else {
  511. switch(s->int_sample_fmt) {
  512. case AV_SAMPLE_FMT_S16P :swri_noise_shaping_int16(s, conv_src, preout, &s->dither.noise, out_count); break;
  513. case AV_SAMPLE_FMT_S32P :swri_noise_shaping_int32(s, conv_src, preout, &s->dither.noise, out_count); break;
  514. case AV_SAMPLE_FMT_FLTP :swri_noise_shaping_float(s, conv_src, preout, &s->dither.noise, out_count); break;
  515. case AV_SAMPLE_FMT_DBLP :swri_noise_shaping_double(s,conv_src, preout, &s->dither.noise, out_count); break;
  516. }
  517. }
  518. s->dither.noise_pos += out_count;
  519. }
  520. //FIXME packed doesn't need more than 1 chan here!
  521. swri_audio_convert(s->out_convert, out, conv_src, out_count);
  522. }
  523. return out_count;
  524. }
  525. int swr_is_initialized(struct SwrContext *s) {
  526. return !!s->in_buffer.ch_count;
  527. }
  528. int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
  529. const uint8_t *in_arg [SWR_CH_MAX], int in_count){
  530. AudioData * in= &s->in;
  531. AudioData *out= &s->out;
  532. if (!swr_is_initialized(s)) {
  533. av_log(s, AV_LOG_ERROR, "Context has not been initialized\n");
  534. return AVERROR(EINVAL);
  535. }
  536. while(s->drop_output > 0){
  537. int ret;
  538. uint8_t *tmp_arg[SWR_CH_MAX];
  539. #define MAX_DROP_STEP 16384
  540. if((ret=swri_realloc_audio(&s->drop_temp, FFMIN(s->drop_output, MAX_DROP_STEP)))<0)
  541. return ret;
  542. reversefill_audiodata(&s->drop_temp, tmp_arg);
  543. s->drop_output *= -1; //FIXME find a less hackish solution
  544. ret = swr_convert(s, tmp_arg, FFMIN(-s->drop_output, MAX_DROP_STEP), in_arg, in_count); //FIXME optimize but this is as good as never called so maybe it doesn't matter
  545. s->drop_output *= -1;
  546. in_count = 0;
  547. if(ret>0) {
  548. s->drop_output -= ret;
  549. continue;
  550. }
  551. if(s->drop_output || !out_arg)
  552. return 0;
  553. }
  554. if(!in_arg){
  555. if(s->resample){
  556. if (!s->flushed)
  557. s->resampler->flush(s);
  558. s->resample_in_constraint = 0;
  559. s->flushed = 1;
  560. }else if(!s->in_buffer_count){
  561. return 0;
  562. }
  563. }else
  564. fill_audiodata(in , (void*)in_arg);
  565. fill_audiodata(out, out_arg);
  566. if(s->resample){
  567. int ret = swr_convert_internal(s, out, out_count, in, in_count);
  568. if(ret>0 && !s->drop_output)
  569. s->outpts += ret * (int64_t)s->in_sample_rate;
  570. return ret;
  571. }else{
  572. AudioData tmp= *in;
  573. int ret2=0;
  574. int ret, size;
  575. size = FFMIN(out_count, s->in_buffer_count);
  576. if(size){
  577. buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
  578. ret= swr_convert_internal(s, out, size, &tmp, size);
  579. if(ret<0)
  580. return ret;
  581. ret2= ret;
  582. s->in_buffer_count -= ret;
  583. s->in_buffer_index += ret;
  584. buf_set(out, out, ret);
  585. out_count -= ret;
  586. if(!s->in_buffer_count)
  587. s->in_buffer_index = 0;
  588. }
  589. if(in_count){
  590. size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
  591. if(in_count > out_count) { //FIXME move after swr_convert_internal
  592. if( size > s->in_buffer.count
  593. && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
  594. buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
  595. copy(&s->in_buffer, &tmp, s->in_buffer_count);
  596. s->in_buffer_index=0;
  597. }else
  598. if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
  599. return ret;
  600. }
  601. if(out_count){
  602. size = FFMIN(in_count, out_count);
  603. ret= swr_convert_internal(s, out, size, in, size);
  604. if(ret<0)
  605. return ret;
  606. buf_set(in, in, ret);
  607. in_count -= ret;
  608. ret2 += ret;
  609. }
  610. if(in_count){
  611. buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
  612. copy(&tmp, in, in_count);
  613. s->in_buffer_count += in_count;
  614. }
  615. }
  616. if(ret2>0 && !s->drop_output)
  617. s->outpts += ret2 * (int64_t)s->in_sample_rate;
  618. return ret2;
  619. }
  620. }
  621. int swr_drop_output(struct SwrContext *s, int count){
  622. s->drop_output += count;
  623. if(s->drop_output <= 0)
  624. return 0;
  625. av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
  626. return swr_convert(s, NULL, s->drop_output, NULL, 0);
  627. }
  628. int swr_inject_silence(struct SwrContext *s, int count){
  629. int ret, i;
  630. uint8_t *tmp_arg[SWR_CH_MAX];
  631. if(count <= 0)
  632. return 0;
  633. #define MAX_SILENCE_STEP 16384
  634. while (count > MAX_SILENCE_STEP) {
  635. if ((ret = swr_inject_silence(s, MAX_SILENCE_STEP)) < 0)
  636. return ret;
  637. count -= MAX_SILENCE_STEP;
  638. }
  639. if((ret=swri_realloc_audio(&s->silence, count))<0)
  640. return ret;
  641. if(s->silence.planar) for(i=0; i<s->silence.ch_count; i++) {
  642. memset(s->silence.ch[i], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps);
  643. } else
  644. memset(s->silence.ch[0], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps*s->silence.ch_count);
  645. reversefill_audiodata(&s->silence, tmp_arg);
  646. av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
  647. ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
  648. return ret;
  649. }
  650. int64_t swr_get_delay(struct SwrContext *s, int64_t base){
  651. if (s->resampler && s->resample){
  652. return s->resampler->get_delay(s, base);
  653. }else{
  654. return (s->in_buffer_count*base + (s->in_sample_rate>>1))/ s->in_sample_rate;
  655. }
  656. }
  657. int swr_set_compensation(struct SwrContext *s, int sample_delta, int compensation_distance){
  658. int ret;
  659. if (!s || compensation_distance < 0)
  660. return AVERROR(EINVAL);
  661. if (!compensation_distance && sample_delta)
  662. return AVERROR(EINVAL);
  663. if (!s->resample) {
  664. s->flags |= SWR_FLAG_RESAMPLE;
  665. ret = swr_init(s);
  666. if (ret < 0)
  667. return ret;
  668. }
  669. if (!s->resampler->set_compensation){
  670. return AVERROR(EINVAL);
  671. }else{
  672. return s->resampler->set_compensation(s->resample, sample_delta, compensation_distance);
  673. }
  674. }
  675. int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
  676. if(pts == INT64_MIN)
  677. return s->outpts;
  678. if (s->firstpts == AV_NOPTS_VALUE)
  679. s->outpts = s->firstpts = pts;
  680. if(s->min_compensation >= FLT_MAX) {
  681. return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
  682. } else {
  683. int64_t delta = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate) - s->outpts + s->drop_output*(int64_t)s->in_sample_rate;
  684. double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
  685. if(fabs(fdelta) > s->min_compensation) {
  686. if(s->outpts == s->firstpts || fabs(fdelta) > s->min_hard_compensation){
  687. int ret;
  688. if(delta > 0) ret = swr_inject_silence(s, delta / s->out_sample_rate);
  689. else ret = swr_drop_output (s, -delta / s-> in_sample_rate);
  690. if(ret<0){
  691. av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
  692. }
  693. } else if(s->soft_compensation_duration && s->max_soft_compensation) {
  694. int duration = s->out_sample_rate * s->soft_compensation_duration;
  695. double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
  696. int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
  697. av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
  698. swr_set_compensation(s, comp, duration);
  699. }
  700. }
  701. return s->outpts;
  702. }
  703. }