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.

857 lines
30KB

  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. #include "libavutil/ffversion.h"
  28. const char swr_ffversion[] = "FFmpeg version " FFMPEG_VERSION;
  29. unsigned swresample_version(void)
  30. {
  31. av_assert0(LIBSWRESAMPLE_VERSION_MICRO >= 100);
  32. return LIBSWRESAMPLE_VERSION_INT;
  33. }
  34. const char *swresample_configuration(void)
  35. {
  36. return FFMPEG_CONFIGURATION;
  37. }
  38. const char *swresample_license(void)
  39. {
  40. #define LICENSE_PREFIX "libswresample license: "
  41. return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
  42. }
  43. int swr_set_channel_mapping(struct SwrContext *s, const int *channel_map){
  44. if(!s || s->in_convert) // s needs to be allocated but not initialized
  45. return AVERROR(EINVAL);
  46. s->channel_map = channel_map;
  47. return 0;
  48. }
  49. struct SwrContext *swr_alloc_set_opts(struct SwrContext *s,
  50. int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
  51. int64_t in_ch_layout, enum AVSampleFormat in_sample_fmt, int in_sample_rate,
  52. int log_offset, void *log_ctx){
  53. if(!s) s= swr_alloc();
  54. if(!s) return NULL;
  55. s->log_level_offset= log_offset;
  56. s->log_ctx= log_ctx;
  57. if (av_opt_set_int(s, "ocl", out_ch_layout, 0) < 0)
  58. goto fail;
  59. if (av_opt_set_int(s, "osf", out_sample_fmt, 0) < 0)
  60. goto fail;
  61. if (av_opt_set_int(s, "osr", out_sample_rate, 0) < 0)
  62. goto fail;
  63. if (av_opt_set_int(s, "icl", in_ch_layout, 0) < 0)
  64. goto fail;
  65. if (av_opt_set_int(s, "isf", in_sample_fmt, 0) < 0)
  66. goto fail;
  67. if (av_opt_set_int(s, "isr", in_sample_rate, 0) < 0)
  68. goto fail;
  69. if (av_opt_set_int(s, "tsf", AV_SAMPLE_FMT_NONE, 0) < 0)
  70. goto fail;
  71. if (av_opt_set_int(s, "ich", av_get_channel_layout_nb_channels(s-> user_in_ch_layout), 0) < 0)
  72. goto fail;
  73. if (av_opt_set_int(s, "och", av_get_channel_layout_nb_channels(s->user_out_ch_layout), 0) < 0)
  74. goto fail;
  75. av_opt_set_int(s, "uch", 0, 0);
  76. return s;
  77. fail:
  78. av_log(s, AV_LOG_ERROR, "Failed to set option\n");
  79. swr_free(&s);
  80. return NULL;
  81. }
  82. static void set_audiodata_fmt(AudioData *a, enum AVSampleFormat fmt){
  83. a->fmt = fmt;
  84. a->bps = av_get_bytes_per_sample(fmt);
  85. a->planar= av_sample_fmt_is_planar(fmt);
  86. if (a->ch_count == 1)
  87. a->planar = 1;
  88. }
  89. static void free_temp(AudioData *a){
  90. av_free(a->data);
  91. memset(a, 0, sizeof(*a));
  92. }
  93. static void clear_context(SwrContext *s){
  94. s->in_buffer_index= 0;
  95. s->in_buffer_count= 0;
  96. s->resample_in_constraint= 0;
  97. memset(s->in.ch, 0, sizeof(s->in.ch));
  98. memset(s->out.ch, 0, sizeof(s->out.ch));
  99. free_temp(&s->postin);
  100. free_temp(&s->midbuf);
  101. free_temp(&s->preout);
  102. free_temp(&s->in_buffer);
  103. free_temp(&s->silence);
  104. free_temp(&s->drop_temp);
  105. free_temp(&s->dither.noise);
  106. free_temp(&s->dither.temp);
  107. swri_audio_convert_free(&s-> in_convert);
  108. swri_audio_convert_free(&s->out_convert);
  109. swri_audio_convert_free(&s->full_convert);
  110. swri_rematrix_free(s);
  111. s->flushed = 0;
  112. }
  113. av_cold void swr_free(SwrContext **ss){
  114. SwrContext *s= *ss;
  115. if(s){
  116. clear_context(s);
  117. if (s->resampler)
  118. s->resampler->free(&s->resample);
  119. }
  120. av_freep(ss);
  121. }
  122. av_cold void swr_close(SwrContext *s){
  123. clear_context(s);
  124. }
  125. av_cold int swr_init(struct SwrContext *s){
  126. int ret;
  127. clear_context(s);
  128. if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
  129. av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
  130. return AVERROR(EINVAL);
  131. }
  132. if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
  133. av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
  134. return AVERROR(EINVAL);
  135. }
  136. s->out.ch_count = s-> user_out_ch_count;
  137. s-> in.ch_count = s-> user_in_ch_count;
  138. s->used_ch_count = s->user_used_ch_count;
  139. s-> in_ch_layout = s-> user_in_ch_layout;
  140. s->out_ch_layout = s->user_out_ch_layout;
  141. if(av_get_channel_layout_nb_channels(s-> in_ch_layout) > SWR_CH_MAX) {
  142. av_log(s, AV_LOG_WARNING, "Input channel layout 0x%"PRIx64" is invalid or unsupported.\n", s-> in_ch_layout);
  143. s->in_ch_layout = 0;
  144. }
  145. if(av_get_channel_layout_nb_channels(s->out_ch_layout) > SWR_CH_MAX) {
  146. av_log(s, AV_LOG_WARNING, "Output channel layout 0x%"PRIx64" is invalid or unsupported.\n", s->out_ch_layout);
  147. s->out_ch_layout = 0;
  148. }
  149. switch(s->engine){
  150. #if CONFIG_LIBSOXR
  151. extern struct Resampler const soxr_resampler;
  152. case SWR_ENGINE_SOXR: s->resampler = &soxr_resampler; break;
  153. #endif
  154. case SWR_ENGINE_SWR : s->resampler = &swri_resampler; break;
  155. default:
  156. av_log(s, AV_LOG_ERROR, "Requested resampling engine is unavailable\n");
  157. return AVERROR(EINVAL);
  158. }
  159. if(!s->used_ch_count)
  160. s->used_ch_count= s->in.ch_count;
  161. if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
  162. av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
  163. s-> in_ch_layout= 0;
  164. }
  165. if(!s-> in_ch_layout)
  166. s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
  167. if(!s->out_ch_layout)
  168. s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
  169. s->rematrix= s->out_ch_layout !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
  170. s->rematrix_custom;
  171. if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
  172. if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_S16P){
  173. s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
  174. }else if( av_get_planar_sample_fmt(s-> in_sample_fmt) == AV_SAMPLE_FMT_S32P
  175. && av_get_planar_sample_fmt(s->out_sample_fmt) == AV_SAMPLE_FMT_S32P
  176. && !s->rematrix
  177. && s->engine != SWR_ENGINE_SOXR){
  178. s->int_sample_fmt= AV_SAMPLE_FMT_S32P;
  179. }else if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_FLTP){
  180. s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
  181. }else{
  182. av_log(s, AV_LOG_DEBUG, "Using double precision mode\n");
  183. s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
  184. }
  185. }
  186. if( s->int_sample_fmt != AV_SAMPLE_FMT_S16P
  187. &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
  188. &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
  189. &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
  190. 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));
  191. return AVERROR(EINVAL);
  192. }
  193. set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
  194. set_audiodata_fmt(&s->out, s->out_sample_fmt);
  195. if (s->firstpts_in_samples != AV_NOPTS_VALUE) {
  196. if (!s->async && s->min_compensation >= FLT_MAX/2)
  197. s->async = 1;
  198. s->firstpts =
  199. s->outpts = s->firstpts_in_samples * s->out_sample_rate;
  200. } else
  201. s->firstpts = AV_NOPTS_VALUE;
  202. if (s->async) {
  203. if (s->min_compensation >= FLT_MAX/2)
  204. s->min_compensation = 0.001;
  205. if (s->async > 1.0001) {
  206. s->max_soft_compensation = s->async / (double) s->in_sample_rate;
  207. }
  208. }
  209. if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
  210. 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);
  211. }else
  212. s->resampler->free(&s->resample);
  213. if( s->int_sample_fmt != AV_SAMPLE_FMT_S16P
  214. && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
  215. && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
  216. && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
  217. && s->resample){
  218. av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16/s32/flt/dbl\n");
  219. return -1;
  220. }
  221. #define RSC 1 //FIXME finetune
  222. if(!s-> in.ch_count)
  223. s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
  224. if(!s->used_ch_count)
  225. s->used_ch_count= s->in.ch_count;
  226. if(!s->out.ch_count)
  227. s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
  228. if(!s-> in.ch_count){
  229. av_assert0(!s->in_ch_layout);
  230. av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
  231. return -1;
  232. }
  233. if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
  234. char l1[1024], l2[1024];
  235. av_get_channel_layout_string(l1, sizeof(l1), s-> in.ch_count, s-> in_ch_layout);
  236. av_get_channel_layout_string(l2, sizeof(l2), s->out.ch_count, s->out_ch_layout);
  237. av_log(s, AV_LOG_ERROR, "Rematrix is needed between %s and %s "
  238. "but there is not enough information to do it\n", l1, l2);
  239. return -1;
  240. }
  241. av_assert0(s->used_ch_count);
  242. av_assert0(s->out.ch_count);
  243. s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
  244. s->in_buffer= s->in;
  245. s->silence = s->in;
  246. s->drop_temp= s->out;
  247. if(!s->resample && !s->rematrix && !s->channel_map && !s->dither.method){
  248. s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
  249. s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
  250. return 0;
  251. }
  252. s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
  253. s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
  254. s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
  255. s->int_sample_fmt, s->out.ch_count, NULL, 0);
  256. if (!s->in_convert || !s->out_convert)
  257. return AVERROR(ENOMEM);
  258. s->postin= s->in;
  259. s->preout= s->out;
  260. s->midbuf= s->in;
  261. if(s->channel_map){
  262. s->postin.ch_count=
  263. s->midbuf.ch_count= s->used_ch_count;
  264. if(s->resample)
  265. s->in_buffer.ch_count= s->used_ch_count;
  266. }
  267. if(!s->resample_first){
  268. s->midbuf.ch_count= s->out.ch_count;
  269. if(s->resample)
  270. s->in_buffer.ch_count = s->out.ch_count;
  271. }
  272. set_audiodata_fmt(&s->postin, s->int_sample_fmt);
  273. set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
  274. set_audiodata_fmt(&s->preout, s->int_sample_fmt);
  275. if(s->resample){
  276. set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
  277. }
  278. if ((ret = swri_dither_init(s, s->out_sample_fmt, s->int_sample_fmt)) < 0)
  279. return ret;
  280. if(s->rematrix || s->dither.method)
  281. return swri_rematrix_init(s);
  282. return 0;
  283. }
  284. int swri_realloc_audio(AudioData *a, int count){
  285. int i, countb;
  286. AudioData old;
  287. if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
  288. return AVERROR(EINVAL);
  289. if(a->count >= count)
  290. return 0;
  291. count*=2;
  292. countb= FFALIGN(count*a->bps, ALIGN);
  293. old= *a;
  294. av_assert0(a->bps);
  295. av_assert0(a->ch_count);
  296. a->data= av_mallocz(countb*a->ch_count);
  297. if(!a->data)
  298. return AVERROR(ENOMEM);
  299. for(i=0; i<a->ch_count; i++){
  300. a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
  301. if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
  302. }
  303. if(!a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
  304. av_freep(&old.data);
  305. a->count= count;
  306. return 1;
  307. }
  308. static void copy(AudioData *out, AudioData *in,
  309. int count){
  310. av_assert0(out->planar == in->planar);
  311. av_assert0(out->bps == in->bps);
  312. av_assert0(out->ch_count == in->ch_count);
  313. if(out->planar){
  314. int ch;
  315. for(ch=0; ch<out->ch_count; ch++)
  316. memcpy(out->ch[ch], in->ch[ch], count*out->bps);
  317. }else
  318. memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
  319. }
  320. static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
  321. int i;
  322. if(!in_arg){
  323. memset(out->ch, 0, sizeof(out->ch));
  324. }else if(out->planar){
  325. for(i=0; i<out->ch_count; i++)
  326. out->ch[i]= in_arg[i];
  327. }else{
  328. for(i=0; i<out->ch_count; i++)
  329. out->ch[i]= in_arg[0] + i*out->bps;
  330. }
  331. }
  332. static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
  333. int i;
  334. if(out->planar){
  335. for(i=0; i<out->ch_count; i++)
  336. in_arg[i]= out->ch[i];
  337. }else{
  338. in_arg[0]= out->ch[0];
  339. }
  340. }
  341. /**
  342. *
  343. * out may be equal in.
  344. */
  345. static void buf_set(AudioData *out, AudioData *in, int count){
  346. int ch;
  347. if(in->planar){
  348. for(ch=0; ch<out->ch_count; ch++)
  349. out->ch[ch]= in->ch[ch] + count*out->bps;
  350. }else{
  351. for(ch=out->ch_count-1; ch>=0; ch--)
  352. out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
  353. }
  354. }
  355. /**
  356. *
  357. * @return number of samples output per channel
  358. */
  359. static int resample(SwrContext *s, AudioData *out_param, int out_count,
  360. const AudioData * in_param, int in_count){
  361. AudioData in, out, tmp;
  362. int ret_sum=0;
  363. int border=0;
  364. int padless = ARCH_X86 && s->engine == SWR_ENGINE_SWR ? 7 : 0;
  365. av_assert1(s->in_buffer.ch_count == in_param->ch_count);
  366. av_assert1(s->in_buffer.planar == in_param->planar);
  367. av_assert1(s->in_buffer.fmt == in_param->fmt);
  368. tmp=out=*out_param;
  369. in = *in_param;
  370. border = s->resampler->invert_initial_buffer(s->resample, &s->in_buffer,
  371. &in, in_count, &s->in_buffer_index, &s->in_buffer_count);
  372. if (border == INT_MAX) return 0;
  373. else if (border < 0) return border;
  374. else if (border) { buf_set(&in, &in, border); in_count -= border; s->resample_in_constraint = 0; }
  375. do{
  376. int ret, size, consumed;
  377. if(!s->resample_in_constraint && s->in_buffer_count){
  378. buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
  379. ret= s->resampler->multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
  380. out_count -= ret;
  381. ret_sum += ret;
  382. buf_set(&out, &out, ret);
  383. s->in_buffer_count -= consumed;
  384. s->in_buffer_index += consumed;
  385. if(!in_count)
  386. break;
  387. if(s->in_buffer_count <= border){
  388. buf_set(&in, &in, -s->in_buffer_count);
  389. in_count += s->in_buffer_count;
  390. s->in_buffer_count=0;
  391. s->in_buffer_index=0;
  392. border = 0;
  393. }
  394. }
  395. if((s->flushed || in_count > padless) && !s->in_buffer_count){
  396. s->in_buffer_index=0;
  397. ret= s->resampler->multiple_resample(s->resample, &out, out_count, &in, FFMAX(in_count-padless, 0), &consumed);
  398. out_count -= ret;
  399. ret_sum += ret;
  400. buf_set(&out, &out, ret);
  401. in_count -= consumed;
  402. buf_set(&in, &in, consumed);
  403. }
  404. //TODO is this check sane considering the advanced copy avoidance below
  405. size= s->in_buffer_index + s->in_buffer_count + in_count;
  406. if( size > s->in_buffer.count
  407. && s->in_buffer_count + in_count <= s->in_buffer_index){
  408. buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
  409. copy(&s->in_buffer, &tmp, s->in_buffer_count);
  410. s->in_buffer_index=0;
  411. }else
  412. if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
  413. return ret;
  414. if(in_count){
  415. int count= in_count;
  416. if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
  417. buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
  418. copy(&tmp, &in, /*in_*/count);
  419. s->in_buffer_count += count;
  420. in_count -= count;
  421. border += count;
  422. buf_set(&in, &in, count);
  423. s->resample_in_constraint= 0;
  424. if(s->in_buffer_count != count || in_count)
  425. continue;
  426. if (padless) {
  427. padless = 0;
  428. continue;
  429. }
  430. }
  431. break;
  432. }while(1);
  433. s->resample_in_constraint= !!out_count;
  434. return ret_sum;
  435. }
  436. static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
  437. AudioData *in , int in_count){
  438. AudioData *postin, *midbuf, *preout;
  439. int ret/*, in_max*/;
  440. AudioData preout_tmp, midbuf_tmp;
  441. if(s->full_convert){
  442. av_assert0(!s->resample);
  443. swri_audio_convert(s->full_convert, out, in, in_count);
  444. return out_count;
  445. }
  446. // in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
  447. // in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
  448. if((ret=swri_realloc_audio(&s->postin, in_count))<0)
  449. return ret;
  450. if(s->resample_first){
  451. av_assert0(s->midbuf.ch_count == s->used_ch_count);
  452. if((ret=swri_realloc_audio(&s->midbuf, out_count))<0)
  453. return ret;
  454. }else{
  455. av_assert0(s->midbuf.ch_count == s->out.ch_count);
  456. if((ret=swri_realloc_audio(&s->midbuf, in_count))<0)
  457. return ret;
  458. }
  459. if((ret=swri_realloc_audio(&s->preout, out_count))<0)
  460. return ret;
  461. postin= &s->postin;
  462. midbuf_tmp= s->midbuf;
  463. midbuf= &midbuf_tmp;
  464. preout_tmp= s->preout;
  465. preout= &preout_tmp;
  466. if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
  467. postin= in;
  468. if(s->resample_first ? !s->resample : !s->rematrix)
  469. midbuf= postin;
  470. if(s->resample_first ? !s->rematrix : !s->resample)
  471. preout= midbuf;
  472. if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar
  473. && !(s->out_sample_fmt==AV_SAMPLE_FMT_S32P && (s->dither.output_sample_bits&31))){
  474. if(preout==in){
  475. out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
  476. av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
  477. copy(out, in, out_count);
  478. return out_count;
  479. }
  480. else if(preout==postin) preout= midbuf= postin= out;
  481. else if(preout==midbuf) preout= midbuf= out;
  482. else preout= out;
  483. }
  484. if(in != postin){
  485. swri_audio_convert(s->in_convert, postin, in, in_count);
  486. }
  487. if(s->resample_first){
  488. if(postin != midbuf)
  489. out_count= resample(s, midbuf, out_count, postin, in_count);
  490. if(midbuf != preout)
  491. swri_rematrix(s, preout, midbuf, out_count, preout==out);
  492. }else{
  493. if(postin != midbuf)
  494. swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
  495. if(midbuf != preout)
  496. out_count= resample(s, preout, out_count, midbuf, in_count);
  497. }
  498. if(preout != out && out_count){
  499. AudioData *conv_src = preout;
  500. if(s->dither.method){
  501. int ch;
  502. int dither_count= FFMAX(out_count, 1<<16);
  503. if (preout == in) {
  504. conv_src = &s->dither.temp;
  505. if((ret=swri_realloc_audio(&s->dither.temp, dither_count))<0)
  506. return ret;
  507. }
  508. if((ret=swri_realloc_audio(&s->dither.noise, dither_count))<0)
  509. return ret;
  510. if(ret)
  511. for(ch=0; ch<s->dither.noise.ch_count; ch++)
  512. swri_get_dither(s, s->dither.noise.ch[ch], s->dither.noise.count, 12345678913579<<ch, s->dither.noise.fmt);
  513. av_assert0(s->dither.noise.ch_count == preout->ch_count);
  514. if(s->dither.noise_pos + out_count > s->dither.noise.count)
  515. s->dither.noise_pos = 0;
  516. if (s->dither.method < SWR_DITHER_NS){
  517. if (s->mix_2_1_simd) {
  518. int len1= out_count&~15;
  519. int off = len1 * preout->bps;
  520. if(len1)
  521. for(ch=0; ch<preout->ch_count; ch++)
  522. 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);
  523. if(out_count != len1)
  524. for(ch=0; ch<preout->ch_count; ch++)
  525. 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);
  526. } else {
  527. for(ch=0; ch<preout->ch_count; ch++)
  528. 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);
  529. }
  530. } else {
  531. switch(s->int_sample_fmt) {
  532. case AV_SAMPLE_FMT_S16P :swri_noise_shaping_int16(s, conv_src, preout, &s->dither.noise, out_count); break;
  533. case AV_SAMPLE_FMT_S32P :swri_noise_shaping_int32(s, conv_src, preout, &s->dither.noise, out_count); break;
  534. case AV_SAMPLE_FMT_FLTP :swri_noise_shaping_float(s, conv_src, preout, &s->dither.noise, out_count); break;
  535. case AV_SAMPLE_FMT_DBLP :swri_noise_shaping_double(s,conv_src, preout, &s->dither.noise, out_count); break;
  536. }
  537. }
  538. s->dither.noise_pos += out_count;
  539. }
  540. //FIXME packed doesn't need more than 1 chan here!
  541. swri_audio_convert(s->out_convert, out, conv_src, out_count);
  542. }
  543. return out_count;
  544. }
  545. int swr_is_initialized(struct SwrContext *s) {
  546. return !!s->in_buffer.ch_count;
  547. }
  548. int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
  549. const uint8_t *in_arg [SWR_CH_MAX], int in_count){
  550. AudioData * in= &s->in;
  551. AudioData *out= &s->out;
  552. if (!swr_is_initialized(s)) {
  553. av_log(s, AV_LOG_ERROR, "Context has not been initialized\n");
  554. return AVERROR(EINVAL);
  555. }
  556. while(s->drop_output > 0){
  557. int ret;
  558. uint8_t *tmp_arg[SWR_CH_MAX];
  559. #define MAX_DROP_STEP 16384
  560. if((ret=swri_realloc_audio(&s->drop_temp, FFMIN(s->drop_output, MAX_DROP_STEP)))<0)
  561. return ret;
  562. reversefill_audiodata(&s->drop_temp, tmp_arg);
  563. s->drop_output *= -1; //FIXME find a less hackish solution
  564. 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
  565. s->drop_output *= -1;
  566. in_count = 0;
  567. if(ret>0) {
  568. s->drop_output -= ret;
  569. if (!s->drop_output && !out_arg)
  570. return 0;
  571. continue;
  572. }
  573. if(s->drop_output || !out_arg)
  574. return 0;
  575. }
  576. if(!in_arg){
  577. if(s->resample){
  578. if (!s->flushed)
  579. s->resampler->flush(s);
  580. s->resample_in_constraint = 0;
  581. s->flushed = 1;
  582. }else if(!s->in_buffer_count){
  583. return 0;
  584. }
  585. }else
  586. fill_audiodata(in , (void*)in_arg);
  587. fill_audiodata(out, out_arg);
  588. if(s->resample){
  589. int ret = swr_convert_internal(s, out, out_count, in, in_count);
  590. if(ret>0 && !s->drop_output)
  591. s->outpts += ret * (int64_t)s->in_sample_rate;
  592. return ret;
  593. }else{
  594. AudioData tmp= *in;
  595. int ret2=0;
  596. int ret, size;
  597. size = FFMIN(out_count, s->in_buffer_count);
  598. if(size){
  599. buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
  600. ret= swr_convert_internal(s, out, size, &tmp, size);
  601. if(ret<0)
  602. return ret;
  603. ret2= ret;
  604. s->in_buffer_count -= ret;
  605. s->in_buffer_index += ret;
  606. buf_set(out, out, ret);
  607. out_count -= ret;
  608. if(!s->in_buffer_count)
  609. s->in_buffer_index = 0;
  610. }
  611. if(in_count){
  612. size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
  613. if(in_count > out_count) { //FIXME move after swr_convert_internal
  614. if( size > s->in_buffer.count
  615. && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
  616. buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
  617. copy(&s->in_buffer, &tmp, s->in_buffer_count);
  618. s->in_buffer_index=0;
  619. }else
  620. if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
  621. return ret;
  622. }
  623. if(out_count){
  624. size = FFMIN(in_count, out_count);
  625. ret= swr_convert_internal(s, out, size, in, size);
  626. if(ret<0)
  627. return ret;
  628. buf_set(in, in, ret);
  629. in_count -= ret;
  630. ret2 += ret;
  631. }
  632. if(in_count){
  633. buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
  634. copy(&tmp, in, in_count);
  635. s->in_buffer_count += in_count;
  636. }
  637. }
  638. if(ret2>0 && !s->drop_output)
  639. s->outpts += ret2 * (int64_t)s->in_sample_rate;
  640. return ret2;
  641. }
  642. }
  643. int swr_drop_output(struct SwrContext *s, int count){
  644. s->drop_output += count;
  645. if(s->drop_output <= 0)
  646. return 0;
  647. av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
  648. return swr_convert(s, NULL, s->drop_output, NULL, 0);
  649. }
  650. int swr_inject_silence(struct SwrContext *s, int count){
  651. int ret, i;
  652. uint8_t *tmp_arg[SWR_CH_MAX];
  653. if(count <= 0)
  654. return 0;
  655. #define MAX_SILENCE_STEP 16384
  656. while (count > MAX_SILENCE_STEP) {
  657. if ((ret = swr_inject_silence(s, MAX_SILENCE_STEP)) < 0)
  658. return ret;
  659. count -= MAX_SILENCE_STEP;
  660. }
  661. if((ret=swri_realloc_audio(&s->silence, count))<0)
  662. return ret;
  663. if(s->silence.planar) for(i=0; i<s->silence.ch_count; i++) {
  664. memset(s->silence.ch[i], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps);
  665. } else
  666. memset(s->silence.ch[0], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps*s->silence.ch_count);
  667. reversefill_audiodata(&s->silence, tmp_arg);
  668. av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
  669. ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
  670. return ret;
  671. }
  672. int64_t swr_get_delay(struct SwrContext *s, int64_t base){
  673. if (s->resampler && s->resample){
  674. return s->resampler->get_delay(s, base);
  675. }else{
  676. return (s->in_buffer_count*base + (s->in_sample_rate>>1))/ s->in_sample_rate;
  677. }
  678. }
  679. int swr_set_compensation(struct SwrContext *s, int sample_delta, int compensation_distance){
  680. int ret;
  681. if (!s || compensation_distance < 0)
  682. return AVERROR(EINVAL);
  683. if (!compensation_distance && sample_delta)
  684. return AVERROR(EINVAL);
  685. if (!s->resample) {
  686. s->flags |= SWR_FLAG_RESAMPLE;
  687. ret = swr_init(s);
  688. if (ret < 0)
  689. return ret;
  690. }
  691. if (!s->resampler->set_compensation){
  692. return AVERROR(EINVAL);
  693. }else{
  694. return s->resampler->set_compensation(s->resample, sample_delta, compensation_distance);
  695. }
  696. }
  697. int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
  698. if(pts == INT64_MIN)
  699. return s->outpts;
  700. if (s->firstpts == AV_NOPTS_VALUE)
  701. s->outpts = s->firstpts = pts;
  702. if(s->min_compensation >= FLT_MAX) {
  703. return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
  704. } else {
  705. 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;
  706. double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
  707. if(fabs(fdelta) > s->min_compensation) {
  708. if(s->outpts == s->firstpts || fabs(fdelta) > s->min_hard_compensation){
  709. int ret;
  710. if(delta > 0) ret = swr_inject_silence(s, delta / s->out_sample_rate);
  711. else ret = swr_drop_output (s, -delta / s-> in_sample_rate);
  712. if(ret<0){
  713. av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
  714. }
  715. } else if(s->soft_compensation_duration && s->max_soft_compensation) {
  716. int duration = s->out_sample_rate * s->soft_compensation_duration;
  717. double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
  718. int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
  719. av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
  720. swr_set_compensation(s, comp, duration);
  721. }
  722. }
  723. return s->outpts;
  724. }
  725. }