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.

866 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. char l1[1024], l2[1024];
  128. clear_context(s);
  129. if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
  130. av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
  131. return AVERROR(EINVAL);
  132. }
  133. if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
  134. av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
  135. return AVERROR(EINVAL);
  136. }
  137. s->out.ch_count = s-> user_out_ch_count;
  138. s-> in.ch_count = s-> user_in_ch_count;
  139. s->used_ch_count = s->user_used_ch_count;
  140. s-> in_ch_layout = s-> user_in_ch_layout;
  141. s->out_ch_layout = s->user_out_ch_layout;
  142. if(av_get_channel_layout_nb_channels(s-> in_ch_layout) > SWR_CH_MAX) {
  143. av_log(s, AV_LOG_WARNING, "Input channel layout 0x%"PRIx64" is invalid or unsupported.\n", s-> in_ch_layout);
  144. s->in_ch_layout = 0;
  145. }
  146. if(av_get_channel_layout_nb_channels(s->out_ch_layout) > SWR_CH_MAX) {
  147. av_log(s, AV_LOG_WARNING, "Output channel layout 0x%"PRIx64" is invalid or unsupported.\n", s->out_ch_layout);
  148. s->out_ch_layout = 0;
  149. }
  150. switch(s->engine){
  151. #if CONFIG_LIBSOXR
  152. extern struct Resampler const soxr_resampler;
  153. case SWR_ENGINE_SOXR: s->resampler = &soxr_resampler; break;
  154. #endif
  155. case SWR_ENGINE_SWR : s->resampler = &swri_resampler; break;
  156. default:
  157. av_log(s, AV_LOG_ERROR, "Requested resampling engine is unavailable\n");
  158. return AVERROR(EINVAL);
  159. }
  160. if(!s->used_ch_count)
  161. s->used_ch_count= s->in.ch_count;
  162. if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
  163. av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
  164. s-> in_ch_layout= 0;
  165. }
  166. if(!s-> in_ch_layout)
  167. s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
  168. if(!s->out_ch_layout)
  169. s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
  170. s->rematrix= s->out_ch_layout !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
  171. s->rematrix_custom;
  172. if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
  173. if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_S16P){
  174. s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
  175. }else if( av_get_planar_sample_fmt(s-> in_sample_fmt) == AV_SAMPLE_FMT_S32P
  176. && av_get_planar_sample_fmt(s->out_sample_fmt) == AV_SAMPLE_FMT_S32P
  177. && !s->rematrix
  178. && s->engine != SWR_ENGINE_SOXR){
  179. s->int_sample_fmt= AV_SAMPLE_FMT_S32P;
  180. }else if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_FLTP){
  181. s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
  182. }else{
  183. av_log(s, AV_LOG_DEBUG, "Using double precision mode\n");
  184. s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
  185. }
  186. }
  187. if( s->int_sample_fmt != AV_SAMPLE_FMT_S16P
  188. &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
  189. &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
  190. &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
  191. 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));
  192. return AVERROR(EINVAL);
  193. }
  194. set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
  195. set_audiodata_fmt(&s->out, s->out_sample_fmt);
  196. if (s->firstpts_in_samples != AV_NOPTS_VALUE) {
  197. if (!s->async && s->min_compensation >= FLT_MAX/2)
  198. s->async = 1;
  199. s->firstpts =
  200. s->outpts = s->firstpts_in_samples * s->out_sample_rate;
  201. } else
  202. s->firstpts = AV_NOPTS_VALUE;
  203. if (s->async) {
  204. if (s->min_compensation >= FLT_MAX/2)
  205. s->min_compensation = 0.001;
  206. if (s->async > 1.0001) {
  207. s->max_soft_compensation = s->async / (double) s->in_sample_rate;
  208. }
  209. }
  210. if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
  211. 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);
  212. }else
  213. s->resampler->free(&s->resample);
  214. if( s->int_sample_fmt != AV_SAMPLE_FMT_S16P
  215. && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
  216. && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
  217. && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
  218. && s->resample){
  219. av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16/s32/flt/dbl\n");
  220. return -1;
  221. }
  222. #define RSC 1 //FIXME finetune
  223. if(!s-> in.ch_count)
  224. s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
  225. if(!s->used_ch_count)
  226. s->used_ch_count= s->in.ch_count;
  227. if(!s->out.ch_count)
  228. s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
  229. if(!s-> in.ch_count){
  230. av_assert0(!s->in_ch_layout);
  231. av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
  232. return -1;
  233. }
  234. av_get_channel_layout_string(l1, sizeof(l1), s-> in.ch_count, s-> in_ch_layout);
  235. av_get_channel_layout_string(l2, sizeof(l2), s->out.ch_count, s->out_ch_layout);
  236. if (s->out_ch_layout && s->out.ch_count != av_get_channel_layout_nb_channels(s->out_ch_layout)) {
  237. av_log(s, AV_LOG_ERROR, "Output channel layout %s mismatches specified channel count %d\n", l2, s->out.ch_count);
  238. return AVERROR(EINVAL);
  239. }
  240. if (s->in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s->in_ch_layout)) {
  241. av_log(s, AV_LOG_ERROR, "Input channel layout %s mismatches specified channel count %d\n", l1, s->used_ch_count);
  242. return AVERROR(EINVAL);
  243. }
  244. if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
  245. av_log(s, AV_LOG_ERROR, "Rematrix is needed between %s and %s "
  246. "but there is not enough information to do it\n", l1, l2);
  247. return -1;
  248. }
  249. av_assert0(s->used_ch_count);
  250. av_assert0(s->out.ch_count);
  251. s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
  252. s->in_buffer= s->in;
  253. s->silence = s->in;
  254. s->drop_temp= s->out;
  255. if(!s->resample && !s->rematrix && !s->channel_map && !s->dither.method){
  256. s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
  257. s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
  258. return 0;
  259. }
  260. s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
  261. s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
  262. s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
  263. s->int_sample_fmt, s->out.ch_count, NULL, 0);
  264. if (!s->in_convert || !s->out_convert)
  265. return AVERROR(ENOMEM);
  266. s->postin= s->in;
  267. s->preout= s->out;
  268. s->midbuf= s->in;
  269. if(s->channel_map){
  270. s->postin.ch_count=
  271. s->midbuf.ch_count= s->used_ch_count;
  272. if(s->resample)
  273. s->in_buffer.ch_count= s->used_ch_count;
  274. }
  275. if(!s->resample_first){
  276. s->midbuf.ch_count= s->out.ch_count;
  277. if(s->resample)
  278. s->in_buffer.ch_count = s->out.ch_count;
  279. }
  280. set_audiodata_fmt(&s->postin, s->int_sample_fmt);
  281. set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
  282. set_audiodata_fmt(&s->preout, s->int_sample_fmt);
  283. if(s->resample){
  284. set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
  285. }
  286. if ((ret = swri_dither_init(s, s->out_sample_fmt, s->int_sample_fmt)) < 0)
  287. return ret;
  288. if(s->rematrix || s->dither.method)
  289. return swri_rematrix_init(s);
  290. return 0;
  291. }
  292. int swri_realloc_audio(AudioData *a, int count){
  293. int i, countb;
  294. AudioData old;
  295. if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
  296. return AVERROR(EINVAL);
  297. if(a->count >= count)
  298. return 0;
  299. count*=2;
  300. countb= FFALIGN(count*a->bps, ALIGN);
  301. old= *a;
  302. av_assert0(a->bps);
  303. av_assert0(a->ch_count);
  304. a->data= av_mallocz(countb*a->ch_count);
  305. if(!a->data)
  306. return AVERROR(ENOMEM);
  307. for(i=0; i<a->ch_count; i++){
  308. a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
  309. if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
  310. }
  311. if(!a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
  312. av_freep(&old.data);
  313. a->count= count;
  314. return 1;
  315. }
  316. static void copy(AudioData *out, AudioData *in,
  317. int count){
  318. av_assert0(out->planar == in->planar);
  319. av_assert0(out->bps == in->bps);
  320. av_assert0(out->ch_count == in->ch_count);
  321. if(out->planar){
  322. int ch;
  323. for(ch=0; ch<out->ch_count; ch++)
  324. memcpy(out->ch[ch], in->ch[ch], count*out->bps);
  325. }else
  326. memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
  327. }
  328. static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
  329. int i;
  330. if(!in_arg){
  331. memset(out->ch, 0, sizeof(out->ch));
  332. }else if(out->planar){
  333. for(i=0; i<out->ch_count; i++)
  334. out->ch[i]= in_arg[i];
  335. }else{
  336. for(i=0; i<out->ch_count; i++)
  337. out->ch[i]= in_arg[0] + i*out->bps;
  338. }
  339. }
  340. static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
  341. int i;
  342. if(out->planar){
  343. for(i=0; i<out->ch_count; i++)
  344. in_arg[i]= out->ch[i];
  345. }else{
  346. in_arg[0]= out->ch[0];
  347. }
  348. }
  349. /**
  350. *
  351. * out may be equal in.
  352. */
  353. static void buf_set(AudioData *out, AudioData *in, int count){
  354. int ch;
  355. if(in->planar){
  356. for(ch=0; ch<out->ch_count; ch++)
  357. out->ch[ch]= in->ch[ch] + count*out->bps;
  358. }else{
  359. for(ch=out->ch_count-1; ch>=0; ch--)
  360. out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
  361. }
  362. }
  363. /**
  364. *
  365. * @return number of samples output per channel
  366. */
  367. static int resample(SwrContext *s, AudioData *out_param, int out_count,
  368. const AudioData * in_param, int in_count){
  369. AudioData in, out, tmp;
  370. int ret_sum=0;
  371. int border=0;
  372. int padless = ARCH_X86 && s->engine == SWR_ENGINE_SWR ? 7 : 0;
  373. av_assert1(s->in_buffer.ch_count == in_param->ch_count);
  374. av_assert1(s->in_buffer.planar == in_param->planar);
  375. av_assert1(s->in_buffer.fmt == in_param->fmt);
  376. tmp=out=*out_param;
  377. in = *in_param;
  378. border = s->resampler->invert_initial_buffer(s->resample, &s->in_buffer,
  379. &in, in_count, &s->in_buffer_index, &s->in_buffer_count);
  380. if (border == INT_MAX) return 0;
  381. else if (border < 0) return border;
  382. else if (border) { buf_set(&in, &in, border); in_count -= border; s->resample_in_constraint = 0; }
  383. do{
  384. int ret, size, consumed;
  385. if(!s->resample_in_constraint && s->in_buffer_count){
  386. buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
  387. ret= s->resampler->multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
  388. out_count -= ret;
  389. ret_sum += ret;
  390. buf_set(&out, &out, ret);
  391. s->in_buffer_count -= consumed;
  392. s->in_buffer_index += consumed;
  393. if(!in_count)
  394. break;
  395. if(s->in_buffer_count <= border){
  396. buf_set(&in, &in, -s->in_buffer_count);
  397. in_count += s->in_buffer_count;
  398. s->in_buffer_count=0;
  399. s->in_buffer_index=0;
  400. border = 0;
  401. }
  402. }
  403. if((s->flushed || in_count > padless) && !s->in_buffer_count){
  404. s->in_buffer_index=0;
  405. ret= s->resampler->multiple_resample(s->resample, &out, out_count, &in, FFMAX(in_count-padless, 0), &consumed);
  406. out_count -= ret;
  407. ret_sum += ret;
  408. buf_set(&out, &out, ret);
  409. in_count -= consumed;
  410. buf_set(&in, &in, consumed);
  411. }
  412. //TODO is this check sane considering the advanced copy avoidance below
  413. size= s->in_buffer_index + s->in_buffer_count + in_count;
  414. if( size > s->in_buffer.count
  415. && s->in_buffer_count + in_count <= s->in_buffer_index){
  416. buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
  417. copy(&s->in_buffer, &tmp, s->in_buffer_count);
  418. s->in_buffer_index=0;
  419. }else
  420. if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
  421. return ret;
  422. if(in_count){
  423. int count= in_count;
  424. if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
  425. buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
  426. copy(&tmp, &in, /*in_*/count);
  427. s->in_buffer_count += count;
  428. in_count -= count;
  429. border += count;
  430. buf_set(&in, &in, count);
  431. s->resample_in_constraint= 0;
  432. if(s->in_buffer_count != count || in_count)
  433. continue;
  434. if (padless) {
  435. padless = 0;
  436. continue;
  437. }
  438. }
  439. break;
  440. }while(1);
  441. s->resample_in_constraint= !!out_count;
  442. return ret_sum;
  443. }
  444. static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
  445. AudioData *in , int in_count){
  446. AudioData *postin, *midbuf, *preout;
  447. int ret/*, in_max*/;
  448. AudioData preout_tmp, midbuf_tmp;
  449. if(s->full_convert){
  450. av_assert0(!s->resample);
  451. swri_audio_convert(s->full_convert, out, in, in_count);
  452. return out_count;
  453. }
  454. // in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
  455. // in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
  456. if((ret=swri_realloc_audio(&s->postin, in_count))<0)
  457. return ret;
  458. if(s->resample_first){
  459. av_assert0(s->midbuf.ch_count == s->used_ch_count);
  460. if((ret=swri_realloc_audio(&s->midbuf, out_count))<0)
  461. return ret;
  462. }else{
  463. av_assert0(s->midbuf.ch_count == s->out.ch_count);
  464. if((ret=swri_realloc_audio(&s->midbuf, in_count))<0)
  465. return ret;
  466. }
  467. if((ret=swri_realloc_audio(&s->preout, out_count))<0)
  468. return ret;
  469. postin= &s->postin;
  470. midbuf_tmp= s->midbuf;
  471. midbuf= &midbuf_tmp;
  472. preout_tmp= s->preout;
  473. preout= &preout_tmp;
  474. if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
  475. postin= in;
  476. if(s->resample_first ? !s->resample : !s->rematrix)
  477. midbuf= postin;
  478. if(s->resample_first ? !s->rematrix : !s->resample)
  479. preout= midbuf;
  480. if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar
  481. && !(s->out_sample_fmt==AV_SAMPLE_FMT_S32P && (s->dither.output_sample_bits&31))){
  482. if(preout==in){
  483. out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
  484. av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
  485. copy(out, in, out_count);
  486. return out_count;
  487. }
  488. else if(preout==postin) preout= midbuf= postin= out;
  489. else if(preout==midbuf) preout= midbuf= out;
  490. else preout= out;
  491. }
  492. if(in != postin){
  493. swri_audio_convert(s->in_convert, postin, in, in_count);
  494. }
  495. if(s->resample_first){
  496. if(postin != midbuf)
  497. out_count= resample(s, midbuf, out_count, postin, in_count);
  498. if(midbuf != preout)
  499. swri_rematrix(s, preout, midbuf, out_count, preout==out);
  500. }else{
  501. if(postin != midbuf)
  502. swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
  503. if(midbuf != preout)
  504. out_count= resample(s, preout, out_count, midbuf, in_count);
  505. }
  506. if(preout != out && out_count){
  507. AudioData *conv_src = preout;
  508. if(s->dither.method){
  509. int ch;
  510. int dither_count= FFMAX(out_count, 1<<16);
  511. if (preout == in) {
  512. conv_src = &s->dither.temp;
  513. if((ret=swri_realloc_audio(&s->dither.temp, dither_count))<0)
  514. return ret;
  515. }
  516. if((ret=swri_realloc_audio(&s->dither.noise, dither_count))<0)
  517. return ret;
  518. if(ret)
  519. for(ch=0; ch<s->dither.noise.ch_count; ch++)
  520. swri_get_dither(s, s->dither.noise.ch[ch], s->dither.noise.count, 12345678913579<<ch, s->dither.noise.fmt);
  521. av_assert0(s->dither.noise.ch_count == preout->ch_count);
  522. if(s->dither.noise_pos + out_count > s->dither.noise.count)
  523. s->dither.noise_pos = 0;
  524. if (s->dither.method < SWR_DITHER_NS){
  525. if (s->mix_2_1_simd) {
  526. int len1= out_count&~15;
  527. int off = len1 * preout->bps;
  528. if(len1)
  529. for(ch=0; ch<preout->ch_count; ch++)
  530. 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);
  531. if(out_count != len1)
  532. for(ch=0; ch<preout->ch_count; ch++)
  533. 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);
  534. } else {
  535. for(ch=0; ch<preout->ch_count; ch++)
  536. 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);
  537. }
  538. } else {
  539. switch(s->int_sample_fmt) {
  540. case AV_SAMPLE_FMT_S16P :swri_noise_shaping_int16(s, conv_src, preout, &s->dither.noise, out_count); break;
  541. case AV_SAMPLE_FMT_S32P :swri_noise_shaping_int32(s, conv_src, preout, &s->dither.noise, out_count); break;
  542. case AV_SAMPLE_FMT_FLTP :swri_noise_shaping_float(s, conv_src, preout, &s->dither.noise, out_count); break;
  543. case AV_SAMPLE_FMT_DBLP :swri_noise_shaping_double(s,conv_src, preout, &s->dither.noise, out_count); break;
  544. }
  545. }
  546. s->dither.noise_pos += out_count;
  547. }
  548. //FIXME packed doesn't need more than 1 chan here!
  549. swri_audio_convert(s->out_convert, out, conv_src, out_count);
  550. }
  551. return out_count;
  552. }
  553. int swr_is_initialized(struct SwrContext *s) {
  554. return !!s->in_buffer.ch_count;
  555. }
  556. int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
  557. const uint8_t *in_arg [SWR_CH_MAX], int in_count){
  558. AudioData * in= &s->in;
  559. AudioData *out= &s->out;
  560. if (!swr_is_initialized(s)) {
  561. av_log(s, AV_LOG_ERROR, "Context has not been initialized\n");
  562. return AVERROR(EINVAL);
  563. }
  564. while(s->drop_output > 0){
  565. int ret;
  566. uint8_t *tmp_arg[SWR_CH_MAX];
  567. #define MAX_DROP_STEP 16384
  568. if((ret=swri_realloc_audio(&s->drop_temp, FFMIN(s->drop_output, MAX_DROP_STEP)))<0)
  569. return ret;
  570. reversefill_audiodata(&s->drop_temp, tmp_arg);
  571. s->drop_output *= -1; //FIXME find a less hackish solution
  572. 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
  573. s->drop_output *= -1;
  574. in_count = 0;
  575. if(ret>0) {
  576. s->drop_output -= ret;
  577. if (!s->drop_output && !out_arg)
  578. return 0;
  579. continue;
  580. }
  581. if(s->drop_output || !out_arg)
  582. return 0;
  583. }
  584. if(!in_arg){
  585. if(s->resample){
  586. if (!s->flushed)
  587. s->resampler->flush(s);
  588. s->resample_in_constraint = 0;
  589. s->flushed = 1;
  590. }else if(!s->in_buffer_count){
  591. return 0;
  592. }
  593. }else
  594. fill_audiodata(in , (void*)in_arg);
  595. fill_audiodata(out, out_arg);
  596. if(s->resample){
  597. int ret = swr_convert_internal(s, out, out_count, in, in_count);
  598. if(ret>0 && !s->drop_output)
  599. s->outpts += ret * (int64_t)s->in_sample_rate;
  600. return ret;
  601. }else{
  602. AudioData tmp= *in;
  603. int ret2=0;
  604. int ret, size;
  605. size = FFMIN(out_count, s->in_buffer_count);
  606. if(size){
  607. buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
  608. ret= swr_convert_internal(s, out, size, &tmp, size);
  609. if(ret<0)
  610. return ret;
  611. ret2= ret;
  612. s->in_buffer_count -= ret;
  613. s->in_buffer_index += ret;
  614. buf_set(out, out, ret);
  615. out_count -= ret;
  616. if(!s->in_buffer_count)
  617. s->in_buffer_index = 0;
  618. }
  619. if(in_count){
  620. size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
  621. if(in_count > out_count) { //FIXME move after swr_convert_internal
  622. if( size > s->in_buffer.count
  623. && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
  624. buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
  625. copy(&s->in_buffer, &tmp, s->in_buffer_count);
  626. s->in_buffer_index=0;
  627. }else
  628. if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
  629. return ret;
  630. }
  631. if(out_count){
  632. size = FFMIN(in_count, out_count);
  633. ret= swr_convert_internal(s, out, size, in, size);
  634. if(ret<0)
  635. return ret;
  636. buf_set(in, in, ret);
  637. in_count -= ret;
  638. ret2 += ret;
  639. }
  640. if(in_count){
  641. buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
  642. copy(&tmp, in, in_count);
  643. s->in_buffer_count += in_count;
  644. }
  645. }
  646. if(ret2>0 && !s->drop_output)
  647. s->outpts += ret2 * (int64_t)s->in_sample_rate;
  648. return ret2;
  649. }
  650. }
  651. int swr_drop_output(struct SwrContext *s, int count){
  652. s->drop_output += count;
  653. if(s->drop_output <= 0)
  654. return 0;
  655. av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
  656. return swr_convert(s, NULL, s->drop_output, NULL, 0);
  657. }
  658. int swr_inject_silence(struct SwrContext *s, int count){
  659. int ret, i;
  660. uint8_t *tmp_arg[SWR_CH_MAX];
  661. if(count <= 0)
  662. return 0;
  663. #define MAX_SILENCE_STEP 16384
  664. while (count > MAX_SILENCE_STEP) {
  665. if ((ret = swr_inject_silence(s, MAX_SILENCE_STEP)) < 0)
  666. return ret;
  667. count -= MAX_SILENCE_STEP;
  668. }
  669. if((ret=swri_realloc_audio(&s->silence, count))<0)
  670. return ret;
  671. if(s->silence.planar) for(i=0; i<s->silence.ch_count; i++) {
  672. memset(s->silence.ch[i], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps);
  673. } else
  674. memset(s->silence.ch[0], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps*s->silence.ch_count);
  675. reversefill_audiodata(&s->silence, tmp_arg);
  676. av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
  677. ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
  678. return ret;
  679. }
  680. int64_t swr_get_delay(struct SwrContext *s, int64_t base){
  681. if (s->resampler && s->resample){
  682. return s->resampler->get_delay(s, base);
  683. }else{
  684. return (s->in_buffer_count*base + (s->in_sample_rate>>1))/ s->in_sample_rate;
  685. }
  686. }
  687. int swr_set_compensation(struct SwrContext *s, int sample_delta, int compensation_distance){
  688. int ret;
  689. if (!s || compensation_distance < 0)
  690. return AVERROR(EINVAL);
  691. if (!compensation_distance && sample_delta)
  692. return AVERROR(EINVAL);
  693. if (!s->resample) {
  694. s->flags |= SWR_FLAG_RESAMPLE;
  695. ret = swr_init(s);
  696. if (ret < 0)
  697. return ret;
  698. }
  699. if (!s->resampler->set_compensation){
  700. return AVERROR(EINVAL);
  701. }else{
  702. return s->resampler->set_compensation(s->resample, sample_delta, compensation_distance);
  703. }
  704. }
  705. int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
  706. if(pts == INT64_MIN)
  707. return s->outpts;
  708. if (s->firstpts == AV_NOPTS_VALUE)
  709. s->outpts = s->firstpts = pts;
  710. if(s->min_compensation >= FLT_MAX) {
  711. return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
  712. } else {
  713. 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;
  714. double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
  715. if(fabs(fdelta) > s->min_compensation) {
  716. if(s->outpts == s->firstpts || fabs(fdelta) > s->min_hard_compensation){
  717. int ret;
  718. if(delta > 0) ret = swr_inject_silence(s, delta / s->out_sample_rate);
  719. else ret = swr_drop_output (s, -delta / s-> in_sample_rate);
  720. if(ret<0){
  721. av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
  722. }
  723. } else if(s->soft_compensation_duration && s->max_soft_compensation) {
  724. int duration = s->out_sample_rate * s->soft_compensation_duration;
  725. double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
  726. int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
  727. av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
  728. swr_set_compensation(s, comp, duration);
  729. }
  730. }
  731. return s->outpts;
  732. }
  733. }