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.

867 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 "libavutil/internal.h"
  26. #include <float.h>
  27. #define ALIGN 32
  28. #include "libavutil/ffversion.h"
  29. const char swr_ffversion[] = "FFmpeg version " FFMPEG_VERSION;
  30. unsigned swresample_version(void)
  31. {
  32. av_assert0(LIBSWRESAMPLE_VERSION_MICRO >= 100);
  33. return LIBSWRESAMPLE_VERSION_INT;
  34. }
  35. const char *swresample_configuration(void)
  36. {
  37. return FFMPEG_CONFIGURATION;
  38. }
  39. const char *swresample_license(void)
  40. {
  41. #define LICENSE_PREFIX "libswresample license: "
  42. return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
  43. }
  44. int swr_set_channel_mapping(struct SwrContext *s, const int *channel_map){
  45. if(!s || s->in_convert) // s needs to be allocated but not initialized
  46. return AVERROR(EINVAL);
  47. s->channel_map = channel_map;
  48. return 0;
  49. }
  50. struct SwrContext *swr_alloc_set_opts(struct SwrContext *s,
  51. int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
  52. int64_t in_ch_layout, enum AVSampleFormat in_sample_fmt, int in_sample_rate,
  53. int log_offset, void *log_ctx){
  54. if(!s) s= swr_alloc();
  55. if(!s) return NULL;
  56. s->log_level_offset= log_offset;
  57. s->log_ctx= log_ctx;
  58. if (av_opt_set_int(s, "ocl", out_ch_layout, 0) < 0)
  59. goto fail;
  60. if (av_opt_set_int(s, "osf", out_sample_fmt, 0) < 0)
  61. goto fail;
  62. if (av_opt_set_int(s, "osr", out_sample_rate, 0) < 0)
  63. goto fail;
  64. if (av_opt_set_int(s, "icl", in_ch_layout, 0) < 0)
  65. goto fail;
  66. if (av_opt_set_int(s, "isf", in_sample_fmt, 0) < 0)
  67. goto fail;
  68. if (av_opt_set_int(s, "isr", in_sample_rate, 0) < 0)
  69. goto fail;
  70. if (av_opt_set_int(s, "tsf", AV_SAMPLE_FMT_NONE, 0) < 0)
  71. goto fail;
  72. if (av_opt_set_int(s, "ich", av_get_channel_layout_nb_channels(s-> user_in_ch_layout), 0) < 0)
  73. goto fail;
  74. if (av_opt_set_int(s, "och", av_get_channel_layout_nb_channels(s->user_out_ch_layout), 0) < 0)
  75. goto fail;
  76. av_opt_set_int(s, "uch", 0, 0);
  77. return s;
  78. fail:
  79. av_log(s, AV_LOG_ERROR, "Failed to set option\n");
  80. swr_free(&s);
  81. return NULL;
  82. }
  83. static void set_audiodata_fmt(AudioData *a, enum AVSampleFormat fmt){
  84. a->fmt = fmt;
  85. a->bps = av_get_bytes_per_sample(fmt);
  86. a->planar= av_sample_fmt_is_planar(fmt);
  87. if (a->ch_count == 1)
  88. a->planar = 1;
  89. }
  90. static void free_temp(AudioData *a){
  91. av_free(a->data);
  92. memset(a, 0, sizeof(*a));
  93. }
  94. static void clear_context(SwrContext *s){
  95. s->in_buffer_index= 0;
  96. s->in_buffer_count= 0;
  97. s->resample_in_constraint= 0;
  98. memset(s->in.ch, 0, sizeof(s->in.ch));
  99. memset(s->out.ch, 0, sizeof(s->out.ch));
  100. free_temp(&s->postin);
  101. free_temp(&s->midbuf);
  102. free_temp(&s->preout);
  103. free_temp(&s->in_buffer);
  104. free_temp(&s->silence);
  105. free_temp(&s->drop_temp);
  106. free_temp(&s->dither.noise);
  107. free_temp(&s->dither.temp);
  108. swri_audio_convert_free(&s-> in_convert);
  109. swri_audio_convert_free(&s->out_convert);
  110. swri_audio_convert_free(&s->full_convert);
  111. swri_rematrix_free(s);
  112. s->flushed = 0;
  113. }
  114. av_cold void swr_free(SwrContext **ss){
  115. SwrContext *s= *ss;
  116. if(s){
  117. clear_context(s);
  118. if (s->resampler)
  119. s->resampler->free(&s->resample);
  120. }
  121. av_freep(ss);
  122. }
  123. av_cold void swr_close(SwrContext *s){
  124. clear_context(s);
  125. }
  126. av_cold int swr_init(struct SwrContext *s){
  127. int ret;
  128. char l1[1024], l2[1024];
  129. clear_context(s);
  130. if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
  131. av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
  132. return AVERROR(EINVAL);
  133. }
  134. if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
  135. av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
  136. return AVERROR(EINVAL);
  137. }
  138. s->out.ch_count = s-> user_out_ch_count;
  139. s-> in.ch_count = s-> user_in_ch_count;
  140. s->used_ch_count = s->user_used_ch_count;
  141. s-> in_ch_layout = s-> user_in_ch_layout;
  142. s->out_ch_layout = s->user_out_ch_layout;
  143. if(av_get_channel_layout_nb_channels(s-> in_ch_layout) > SWR_CH_MAX) {
  144. av_log(s, AV_LOG_WARNING, "Input channel layout 0x%"PRIx64" is invalid or unsupported.\n", s-> in_ch_layout);
  145. s->in_ch_layout = 0;
  146. }
  147. if(av_get_channel_layout_nb_channels(s->out_ch_layout) > SWR_CH_MAX) {
  148. av_log(s, AV_LOG_WARNING, "Output channel layout 0x%"PRIx64" is invalid or unsupported.\n", s->out_ch_layout);
  149. s->out_ch_layout = 0;
  150. }
  151. switch(s->engine){
  152. #if CONFIG_LIBSOXR
  153. extern struct Resampler const soxr_resampler;
  154. case SWR_ENGINE_SOXR: s->resampler = &soxr_resampler; break;
  155. #endif
  156. case SWR_ENGINE_SWR : s->resampler = &swri_resampler; break;
  157. default:
  158. av_log(s, AV_LOG_ERROR, "Requested resampling engine is unavailable\n");
  159. return AVERROR(EINVAL);
  160. }
  161. if(!s->used_ch_count)
  162. s->used_ch_count= s->in.ch_count;
  163. if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
  164. av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
  165. s-> in_ch_layout= 0;
  166. }
  167. if(!s-> in_ch_layout)
  168. s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
  169. if(!s->out_ch_layout)
  170. s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
  171. s->rematrix= s->out_ch_layout !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
  172. s->rematrix_custom;
  173. if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
  174. if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_S16P){
  175. s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
  176. }else if( av_get_planar_sample_fmt(s-> in_sample_fmt) == AV_SAMPLE_FMT_S32P
  177. && av_get_planar_sample_fmt(s->out_sample_fmt) == AV_SAMPLE_FMT_S32P
  178. && !s->rematrix
  179. && s->engine != SWR_ENGINE_SOXR){
  180. s->int_sample_fmt= AV_SAMPLE_FMT_S32P;
  181. }else if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_FLTP){
  182. s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
  183. }else{
  184. av_log(s, AV_LOG_DEBUG, "Using double precision mode\n");
  185. s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
  186. }
  187. }
  188. if( s->int_sample_fmt != AV_SAMPLE_FMT_S16P
  189. &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
  190. &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
  191. &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
  192. 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));
  193. return AVERROR(EINVAL);
  194. }
  195. set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
  196. set_audiodata_fmt(&s->out, s->out_sample_fmt);
  197. if (s->firstpts_in_samples != AV_NOPTS_VALUE) {
  198. if (!s->async && s->min_compensation >= FLT_MAX/2)
  199. s->async = 1;
  200. s->firstpts =
  201. s->outpts = s->firstpts_in_samples * s->out_sample_rate;
  202. } else
  203. s->firstpts = AV_NOPTS_VALUE;
  204. if (s->async) {
  205. if (s->min_compensation >= FLT_MAX/2)
  206. s->min_compensation = 0.001;
  207. if (s->async > 1.0001) {
  208. s->max_soft_compensation = s->async / (double) s->in_sample_rate;
  209. }
  210. }
  211. if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
  212. 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);
  213. }else
  214. s->resampler->free(&s->resample);
  215. if( s->int_sample_fmt != AV_SAMPLE_FMT_S16P
  216. && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
  217. && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
  218. && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
  219. && s->resample){
  220. av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16/s32/flt/dbl\n");
  221. return -1;
  222. }
  223. #define RSC 1 //FIXME finetune
  224. if(!s-> in.ch_count)
  225. s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
  226. if(!s->used_ch_count)
  227. s->used_ch_count= s->in.ch_count;
  228. if(!s->out.ch_count)
  229. s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
  230. if(!s-> in.ch_count){
  231. av_assert0(!s->in_ch_layout);
  232. av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
  233. return -1;
  234. }
  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. if (s->out_ch_layout && s->out.ch_count != av_get_channel_layout_nb_channels(s->out_ch_layout)) {
  238. av_log(s, AV_LOG_ERROR, "Output channel layout %s mismatches specified channel count %d\n", l2, s->out.ch_count);
  239. return AVERROR(EINVAL);
  240. }
  241. if (s->in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s->in_ch_layout)) {
  242. av_log(s, AV_LOG_ERROR, "Input channel layout %s mismatches specified channel count %d\n", l1, s->used_ch_count);
  243. return AVERROR(EINVAL);
  244. }
  245. if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
  246. av_log(s, AV_LOG_ERROR, "Rematrix is needed between %s and %s "
  247. "but there is not enough information to do it\n", l1, l2);
  248. return -1;
  249. }
  250. av_assert0(s->used_ch_count);
  251. av_assert0(s->out.ch_count);
  252. s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
  253. s->in_buffer= s->in;
  254. s->silence = s->in;
  255. s->drop_temp= s->out;
  256. if(!s->resample && !s->rematrix && !s->channel_map && !s->dither.method){
  257. s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
  258. s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
  259. return 0;
  260. }
  261. s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
  262. s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
  263. s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
  264. s->int_sample_fmt, s->out.ch_count, NULL, 0);
  265. if (!s->in_convert || !s->out_convert)
  266. return AVERROR(ENOMEM);
  267. s->postin= s->in;
  268. s->preout= s->out;
  269. s->midbuf= s->in;
  270. if(s->channel_map){
  271. s->postin.ch_count=
  272. s->midbuf.ch_count= s->used_ch_count;
  273. if(s->resample)
  274. s->in_buffer.ch_count= s->used_ch_count;
  275. }
  276. if(!s->resample_first){
  277. s->midbuf.ch_count= s->out.ch_count;
  278. if(s->resample)
  279. s->in_buffer.ch_count = s->out.ch_count;
  280. }
  281. set_audiodata_fmt(&s->postin, s->int_sample_fmt);
  282. set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
  283. set_audiodata_fmt(&s->preout, s->int_sample_fmt);
  284. if(s->resample){
  285. set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
  286. }
  287. if ((ret = swri_dither_init(s, s->out_sample_fmt, s->int_sample_fmt)) < 0)
  288. return ret;
  289. if(s->rematrix || s->dither.method)
  290. return swri_rematrix_init(s);
  291. return 0;
  292. }
  293. int swri_realloc_audio(AudioData *a, int count){
  294. int i, countb;
  295. AudioData old;
  296. if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
  297. return AVERROR(EINVAL);
  298. if(a->count >= count)
  299. return 0;
  300. count*=2;
  301. countb= FFALIGN(count*a->bps, ALIGN);
  302. old= *a;
  303. av_assert0(a->bps);
  304. av_assert0(a->ch_count);
  305. a->data= av_mallocz(countb*a->ch_count);
  306. if(!a->data)
  307. return AVERROR(ENOMEM);
  308. for(i=0; i<a->ch_count; i++){
  309. a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
  310. if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
  311. }
  312. if(!a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
  313. av_freep(&old.data);
  314. a->count= count;
  315. return 1;
  316. }
  317. static void copy(AudioData *out, AudioData *in,
  318. int count){
  319. av_assert0(out->planar == in->planar);
  320. av_assert0(out->bps == in->bps);
  321. av_assert0(out->ch_count == in->ch_count);
  322. if(out->planar){
  323. int ch;
  324. for(ch=0; ch<out->ch_count; ch++)
  325. memcpy(out->ch[ch], in->ch[ch], count*out->bps);
  326. }else
  327. memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
  328. }
  329. static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
  330. int i;
  331. if(!in_arg){
  332. memset(out->ch, 0, sizeof(out->ch));
  333. }else if(out->planar){
  334. for(i=0; i<out->ch_count; i++)
  335. out->ch[i]= in_arg[i];
  336. }else{
  337. for(i=0; i<out->ch_count; i++)
  338. out->ch[i]= in_arg[0] + i*out->bps;
  339. }
  340. }
  341. static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
  342. int i;
  343. if(out->planar){
  344. for(i=0; i<out->ch_count; i++)
  345. in_arg[i]= out->ch[i];
  346. }else{
  347. in_arg[0]= out->ch[0];
  348. }
  349. }
  350. /**
  351. *
  352. * out may be equal in.
  353. */
  354. static void buf_set(AudioData *out, AudioData *in, int count){
  355. int ch;
  356. if(in->planar){
  357. for(ch=0; ch<out->ch_count; ch++)
  358. out->ch[ch]= in->ch[ch] + count*out->bps;
  359. }else{
  360. for(ch=out->ch_count-1; ch>=0; ch--)
  361. out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
  362. }
  363. }
  364. /**
  365. *
  366. * @return number of samples output per channel
  367. */
  368. static int resample(SwrContext *s, AudioData *out_param, int out_count,
  369. const AudioData * in_param, int in_count){
  370. AudioData in, out, tmp;
  371. int ret_sum=0;
  372. int border=0;
  373. int padless = ARCH_X86 && s->engine == SWR_ENGINE_SWR ? 7 : 0;
  374. av_assert1(s->in_buffer.ch_count == in_param->ch_count);
  375. av_assert1(s->in_buffer.planar == in_param->planar);
  376. av_assert1(s->in_buffer.fmt == in_param->fmt);
  377. tmp=out=*out_param;
  378. in = *in_param;
  379. border = s->resampler->invert_initial_buffer(s->resample, &s->in_buffer,
  380. &in, in_count, &s->in_buffer_index, &s->in_buffer_count);
  381. if (border == INT_MAX) return 0;
  382. else if (border < 0) return border;
  383. else if (border) { buf_set(&in, &in, border); in_count -= border; s->resample_in_constraint = 0; }
  384. do{
  385. int ret, size, consumed;
  386. if(!s->resample_in_constraint && s->in_buffer_count){
  387. buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
  388. ret= s->resampler->multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
  389. out_count -= ret;
  390. ret_sum += ret;
  391. buf_set(&out, &out, ret);
  392. s->in_buffer_count -= consumed;
  393. s->in_buffer_index += consumed;
  394. if(!in_count)
  395. break;
  396. if(s->in_buffer_count <= border){
  397. buf_set(&in, &in, -s->in_buffer_count);
  398. in_count += s->in_buffer_count;
  399. s->in_buffer_count=0;
  400. s->in_buffer_index=0;
  401. border = 0;
  402. }
  403. }
  404. if((s->flushed || in_count > padless) && !s->in_buffer_count){
  405. s->in_buffer_index=0;
  406. ret= s->resampler->multiple_resample(s->resample, &out, out_count, &in, FFMAX(in_count-padless, 0), &consumed);
  407. out_count -= ret;
  408. ret_sum += ret;
  409. buf_set(&out, &out, ret);
  410. in_count -= consumed;
  411. buf_set(&in, &in, consumed);
  412. }
  413. //TODO is this check sane considering the advanced copy avoidance below
  414. size= s->in_buffer_index + s->in_buffer_count + in_count;
  415. if( size > s->in_buffer.count
  416. && s->in_buffer_count + in_count <= s->in_buffer_index){
  417. buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
  418. copy(&s->in_buffer, &tmp, s->in_buffer_count);
  419. s->in_buffer_index=0;
  420. }else
  421. if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
  422. return ret;
  423. if(in_count){
  424. int count= in_count;
  425. if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
  426. buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
  427. copy(&tmp, &in, /*in_*/count);
  428. s->in_buffer_count += count;
  429. in_count -= count;
  430. border += count;
  431. buf_set(&in, &in, count);
  432. s->resample_in_constraint= 0;
  433. if(s->in_buffer_count != count || in_count)
  434. continue;
  435. if (padless) {
  436. padless = 0;
  437. continue;
  438. }
  439. }
  440. break;
  441. }while(1);
  442. s->resample_in_constraint= !!out_count;
  443. return ret_sum;
  444. }
  445. static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
  446. AudioData *in , int in_count){
  447. AudioData *postin, *midbuf, *preout;
  448. int ret/*, in_max*/;
  449. AudioData preout_tmp, midbuf_tmp;
  450. if(s->full_convert){
  451. av_assert0(!s->resample);
  452. swri_audio_convert(s->full_convert, out, in, in_count);
  453. return out_count;
  454. }
  455. // in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
  456. // in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
  457. if((ret=swri_realloc_audio(&s->postin, in_count))<0)
  458. return ret;
  459. if(s->resample_first){
  460. av_assert0(s->midbuf.ch_count == s->used_ch_count);
  461. if((ret=swri_realloc_audio(&s->midbuf, out_count))<0)
  462. return ret;
  463. }else{
  464. av_assert0(s->midbuf.ch_count == s->out.ch_count);
  465. if((ret=swri_realloc_audio(&s->midbuf, in_count))<0)
  466. return ret;
  467. }
  468. if((ret=swri_realloc_audio(&s->preout, out_count))<0)
  469. return ret;
  470. postin= &s->postin;
  471. midbuf_tmp= s->midbuf;
  472. midbuf= &midbuf_tmp;
  473. preout_tmp= s->preout;
  474. preout= &preout_tmp;
  475. if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
  476. postin= in;
  477. if(s->resample_first ? !s->resample : !s->rematrix)
  478. midbuf= postin;
  479. if(s->resample_first ? !s->rematrix : !s->resample)
  480. preout= midbuf;
  481. if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar
  482. && !(s->out_sample_fmt==AV_SAMPLE_FMT_S32P && (s->dither.output_sample_bits&31))){
  483. if(preout==in){
  484. out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
  485. av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
  486. copy(out, in, out_count);
  487. return out_count;
  488. }
  489. else if(preout==postin) preout= midbuf= postin= out;
  490. else if(preout==midbuf) preout= midbuf= out;
  491. else preout= out;
  492. }
  493. if(in != postin){
  494. swri_audio_convert(s->in_convert, postin, in, in_count);
  495. }
  496. if(s->resample_first){
  497. if(postin != midbuf)
  498. out_count= resample(s, midbuf, out_count, postin, in_count);
  499. if(midbuf != preout)
  500. swri_rematrix(s, preout, midbuf, out_count, preout==out);
  501. }else{
  502. if(postin != midbuf)
  503. swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
  504. if(midbuf != preout)
  505. out_count= resample(s, preout, out_count, midbuf, in_count);
  506. }
  507. if(preout != out && out_count){
  508. AudioData *conv_src = preout;
  509. if(s->dither.method){
  510. int ch;
  511. int dither_count= FFMAX(out_count, 1<<16);
  512. if (preout == in) {
  513. conv_src = &s->dither.temp;
  514. if((ret=swri_realloc_audio(&s->dither.temp, dither_count))<0)
  515. return ret;
  516. }
  517. if((ret=swri_realloc_audio(&s->dither.noise, dither_count))<0)
  518. return ret;
  519. if(ret)
  520. for(ch=0; ch<s->dither.noise.ch_count; ch++)
  521. swri_get_dither(s, s->dither.noise.ch[ch], s->dither.noise.count, 12345678913579<<ch, s->dither.noise.fmt);
  522. av_assert0(s->dither.noise.ch_count == preout->ch_count);
  523. if(s->dither.noise_pos + out_count > s->dither.noise.count)
  524. s->dither.noise_pos = 0;
  525. if (s->dither.method < SWR_DITHER_NS){
  526. if (s->mix_2_1_simd) {
  527. int len1= out_count&~15;
  528. int off = len1 * preout->bps;
  529. if(len1)
  530. for(ch=0; ch<preout->ch_count; ch++)
  531. 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);
  532. if(out_count != len1)
  533. for(ch=0; ch<preout->ch_count; ch++)
  534. 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);
  535. } else {
  536. for(ch=0; ch<preout->ch_count; ch++)
  537. 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);
  538. }
  539. } else {
  540. switch(s->int_sample_fmt) {
  541. case AV_SAMPLE_FMT_S16P :swri_noise_shaping_int16(s, conv_src, preout, &s->dither.noise, out_count); break;
  542. case AV_SAMPLE_FMT_S32P :swri_noise_shaping_int32(s, conv_src, preout, &s->dither.noise, out_count); break;
  543. case AV_SAMPLE_FMT_FLTP :swri_noise_shaping_float(s, conv_src, preout, &s->dither.noise, out_count); break;
  544. case AV_SAMPLE_FMT_DBLP :swri_noise_shaping_double(s,conv_src, preout, &s->dither.noise, out_count); break;
  545. }
  546. }
  547. s->dither.noise_pos += out_count;
  548. }
  549. //FIXME packed doesn't need more than 1 chan here!
  550. swri_audio_convert(s->out_convert, out, conv_src, out_count);
  551. }
  552. return out_count;
  553. }
  554. int swr_is_initialized(struct SwrContext *s) {
  555. return !!s->in_buffer.ch_count;
  556. }
  557. int attribute_align_arg swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
  558. const uint8_t *in_arg [SWR_CH_MAX], int in_count){
  559. AudioData * in= &s->in;
  560. AudioData *out= &s->out;
  561. if (!swr_is_initialized(s)) {
  562. av_log(s, AV_LOG_ERROR, "Context has not been initialized\n");
  563. return AVERROR(EINVAL);
  564. }
  565. while(s->drop_output > 0){
  566. int ret;
  567. uint8_t *tmp_arg[SWR_CH_MAX];
  568. #define MAX_DROP_STEP 16384
  569. if((ret=swri_realloc_audio(&s->drop_temp, FFMIN(s->drop_output, MAX_DROP_STEP)))<0)
  570. return ret;
  571. reversefill_audiodata(&s->drop_temp, tmp_arg);
  572. s->drop_output *= -1; //FIXME find a less hackish solution
  573. 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
  574. s->drop_output *= -1;
  575. in_count = 0;
  576. if(ret>0) {
  577. s->drop_output -= ret;
  578. if (!s->drop_output && !out_arg)
  579. return 0;
  580. continue;
  581. }
  582. if(s->drop_output || !out_arg)
  583. return 0;
  584. }
  585. if(!in_arg){
  586. if(s->resample){
  587. if (!s->flushed)
  588. s->resampler->flush(s);
  589. s->resample_in_constraint = 0;
  590. s->flushed = 1;
  591. }else if(!s->in_buffer_count){
  592. return 0;
  593. }
  594. }else
  595. fill_audiodata(in , (void*)in_arg);
  596. fill_audiodata(out, out_arg);
  597. if(s->resample){
  598. int ret = swr_convert_internal(s, out, out_count, in, in_count);
  599. if(ret>0 && !s->drop_output)
  600. s->outpts += ret * (int64_t)s->in_sample_rate;
  601. return ret;
  602. }else{
  603. AudioData tmp= *in;
  604. int ret2=0;
  605. int ret, size;
  606. size = FFMIN(out_count, s->in_buffer_count);
  607. if(size){
  608. buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
  609. ret= swr_convert_internal(s, out, size, &tmp, size);
  610. if(ret<0)
  611. return ret;
  612. ret2= ret;
  613. s->in_buffer_count -= ret;
  614. s->in_buffer_index += ret;
  615. buf_set(out, out, ret);
  616. out_count -= ret;
  617. if(!s->in_buffer_count)
  618. s->in_buffer_index = 0;
  619. }
  620. if(in_count){
  621. size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
  622. if(in_count > out_count) { //FIXME move after swr_convert_internal
  623. if( size > s->in_buffer.count
  624. && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
  625. buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
  626. copy(&s->in_buffer, &tmp, s->in_buffer_count);
  627. s->in_buffer_index=0;
  628. }else
  629. if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
  630. return ret;
  631. }
  632. if(out_count){
  633. size = FFMIN(in_count, out_count);
  634. ret= swr_convert_internal(s, out, size, in, size);
  635. if(ret<0)
  636. return ret;
  637. buf_set(in, in, ret);
  638. in_count -= ret;
  639. ret2 += ret;
  640. }
  641. if(in_count){
  642. buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
  643. copy(&tmp, in, in_count);
  644. s->in_buffer_count += in_count;
  645. }
  646. }
  647. if(ret2>0 && !s->drop_output)
  648. s->outpts += ret2 * (int64_t)s->in_sample_rate;
  649. return ret2;
  650. }
  651. }
  652. int swr_drop_output(struct SwrContext *s, int count){
  653. s->drop_output += count;
  654. if(s->drop_output <= 0)
  655. return 0;
  656. av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
  657. return swr_convert(s, NULL, s->drop_output, NULL, 0);
  658. }
  659. int swr_inject_silence(struct SwrContext *s, int count){
  660. int ret, i;
  661. uint8_t *tmp_arg[SWR_CH_MAX];
  662. if(count <= 0)
  663. return 0;
  664. #define MAX_SILENCE_STEP 16384
  665. while (count > MAX_SILENCE_STEP) {
  666. if ((ret = swr_inject_silence(s, MAX_SILENCE_STEP)) < 0)
  667. return ret;
  668. count -= MAX_SILENCE_STEP;
  669. }
  670. if((ret=swri_realloc_audio(&s->silence, count))<0)
  671. return ret;
  672. if(s->silence.planar) for(i=0; i<s->silence.ch_count; i++) {
  673. memset(s->silence.ch[i], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps);
  674. } else
  675. memset(s->silence.ch[0], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps*s->silence.ch_count);
  676. reversefill_audiodata(&s->silence, tmp_arg);
  677. av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
  678. ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
  679. return ret;
  680. }
  681. int64_t swr_get_delay(struct SwrContext *s, int64_t base){
  682. if (s->resampler && s->resample){
  683. return s->resampler->get_delay(s, base);
  684. }else{
  685. return (s->in_buffer_count*base + (s->in_sample_rate>>1))/ s->in_sample_rate;
  686. }
  687. }
  688. int swr_set_compensation(struct SwrContext *s, int sample_delta, int compensation_distance){
  689. int ret;
  690. if (!s || compensation_distance < 0)
  691. return AVERROR(EINVAL);
  692. if (!compensation_distance && sample_delta)
  693. return AVERROR(EINVAL);
  694. if (!s->resample) {
  695. s->flags |= SWR_FLAG_RESAMPLE;
  696. ret = swr_init(s);
  697. if (ret < 0)
  698. return ret;
  699. }
  700. if (!s->resampler->set_compensation){
  701. return AVERROR(EINVAL);
  702. }else{
  703. return s->resampler->set_compensation(s->resample, sample_delta, compensation_distance);
  704. }
  705. }
  706. int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
  707. if(pts == INT64_MIN)
  708. return s->outpts;
  709. if (s->firstpts == AV_NOPTS_VALUE)
  710. s->outpts = s->firstpts = pts;
  711. if(s->min_compensation >= FLT_MAX) {
  712. return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
  713. } else {
  714. 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;
  715. double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
  716. if(fabs(fdelta) > s->min_compensation) {
  717. if(s->outpts == s->firstpts || fabs(fdelta) > s->min_hard_compensation){
  718. int ret;
  719. if(delta > 0) ret = swr_inject_silence(s, delta / s->out_sample_rate);
  720. else ret = swr_drop_output (s, -delta / s-> in_sample_rate);
  721. if(ret<0){
  722. av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
  723. }
  724. } else if(s->soft_compensation_duration && s->max_soft_compensation) {
  725. int duration = s->out_sample_rate * s->soft_compensation_duration;
  726. double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
  727. int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
  728. av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
  729. swr_set_compensation(s, comp, duration);
  730. }
  731. }
  732. return s->outpts;
  733. }
  734. }