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.

873 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. case SWR_ENGINE_SOXR: s->resampler = &swri_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_array(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) {
  381. return 0;
  382. } else if (border < 0) {
  383. return border;
  384. } else if (border) {
  385. buf_set(&in, &in, border);
  386. in_count -= border;
  387. s->resample_in_constraint = 0;
  388. }
  389. do{
  390. int ret, size, consumed;
  391. if(!s->resample_in_constraint && s->in_buffer_count){
  392. buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
  393. ret= s->resampler->multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
  394. out_count -= ret;
  395. ret_sum += ret;
  396. buf_set(&out, &out, ret);
  397. s->in_buffer_count -= consumed;
  398. s->in_buffer_index += consumed;
  399. if(!in_count)
  400. break;
  401. if(s->in_buffer_count <= border){
  402. buf_set(&in, &in, -s->in_buffer_count);
  403. in_count += s->in_buffer_count;
  404. s->in_buffer_count=0;
  405. s->in_buffer_index=0;
  406. border = 0;
  407. }
  408. }
  409. if((s->flushed || in_count > padless) && !s->in_buffer_count){
  410. s->in_buffer_index=0;
  411. ret= s->resampler->multiple_resample(s->resample, &out, out_count, &in, FFMAX(in_count-padless, 0), &consumed);
  412. out_count -= ret;
  413. ret_sum += ret;
  414. buf_set(&out, &out, ret);
  415. in_count -= consumed;
  416. buf_set(&in, &in, consumed);
  417. }
  418. //TODO is this check sane considering the advanced copy avoidance below
  419. size= s->in_buffer_index + s->in_buffer_count + in_count;
  420. if( size > s->in_buffer.count
  421. && s->in_buffer_count + in_count <= s->in_buffer_index){
  422. buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
  423. copy(&s->in_buffer, &tmp, s->in_buffer_count);
  424. s->in_buffer_index=0;
  425. }else
  426. if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
  427. return ret;
  428. if(in_count){
  429. int count= in_count;
  430. if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
  431. buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
  432. copy(&tmp, &in, /*in_*/count);
  433. s->in_buffer_count += count;
  434. in_count -= count;
  435. border += count;
  436. buf_set(&in, &in, count);
  437. s->resample_in_constraint= 0;
  438. if(s->in_buffer_count != count || in_count)
  439. continue;
  440. if (padless) {
  441. padless = 0;
  442. continue;
  443. }
  444. }
  445. break;
  446. }while(1);
  447. s->resample_in_constraint= !!out_count;
  448. return ret_sum;
  449. }
  450. static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
  451. AudioData *in , int in_count){
  452. AudioData *postin, *midbuf, *preout;
  453. int ret/*, in_max*/;
  454. AudioData preout_tmp, midbuf_tmp;
  455. if(s->full_convert){
  456. av_assert0(!s->resample);
  457. swri_audio_convert(s->full_convert, out, in, in_count);
  458. return out_count;
  459. }
  460. // in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
  461. // in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
  462. if((ret=swri_realloc_audio(&s->postin, in_count))<0)
  463. return ret;
  464. if(s->resample_first){
  465. av_assert0(s->midbuf.ch_count == s->used_ch_count);
  466. if((ret=swri_realloc_audio(&s->midbuf, out_count))<0)
  467. return ret;
  468. }else{
  469. av_assert0(s->midbuf.ch_count == s->out.ch_count);
  470. if((ret=swri_realloc_audio(&s->midbuf, in_count))<0)
  471. return ret;
  472. }
  473. if((ret=swri_realloc_audio(&s->preout, out_count))<0)
  474. return ret;
  475. postin= &s->postin;
  476. midbuf_tmp= s->midbuf;
  477. midbuf= &midbuf_tmp;
  478. preout_tmp= s->preout;
  479. preout= &preout_tmp;
  480. if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
  481. postin= in;
  482. if(s->resample_first ? !s->resample : !s->rematrix)
  483. midbuf= postin;
  484. if(s->resample_first ? !s->rematrix : !s->resample)
  485. preout= midbuf;
  486. if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar
  487. && !(s->out_sample_fmt==AV_SAMPLE_FMT_S32P && (s->dither.output_sample_bits&31))){
  488. if(preout==in){
  489. out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
  490. av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
  491. copy(out, in, out_count);
  492. return out_count;
  493. }
  494. else if(preout==postin) preout= midbuf= postin= out;
  495. else if(preout==midbuf) preout= midbuf= out;
  496. else preout= out;
  497. }
  498. if(in != postin){
  499. swri_audio_convert(s->in_convert, postin, in, in_count);
  500. }
  501. if(s->resample_first){
  502. if(postin != midbuf)
  503. out_count= resample(s, midbuf, out_count, postin, in_count);
  504. if(midbuf != preout)
  505. swri_rematrix(s, preout, midbuf, out_count, preout==out);
  506. }else{
  507. if(postin != midbuf)
  508. swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
  509. if(midbuf != preout)
  510. out_count= resample(s, preout, out_count, midbuf, in_count);
  511. }
  512. if(preout != out && out_count){
  513. AudioData *conv_src = preout;
  514. if(s->dither.method){
  515. int ch;
  516. int dither_count= FFMAX(out_count, 1<<16);
  517. if (preout == in) {
  518. conv_src = &s->dither.temp;
  519. if((ret=swri_realloc_audio(&s->dither.temp, dither_count))<0)
  520. return ret;
  521. }
  522. if((ret=swri_realloc_audio(&s->dither.noise, dither_count))<0)
  523. return ret;
  524. if(ret)
  525. for(ch=0; ch<s->dither.noise.ch_count; ch++)
  526. swri_get_dither(s, s->dither.noise.ch[ch], s->dither.noise.count, 12345678913579<<ch, s->dither.noise.fmt);
  527. av_assert0(s->dither.noise.ch_count == preout->ch_count);
  528. if(s->dither.noise_pos + out_count > s->dither.noise.count)
  529. s->dither.noise_pos = 0;
  530. if (s->dither.method < SWR_DITHER_NS){
  531. if (s->mix_2_1_simd) {
  532. int len1= out_count&~15;
  533. int off = len1 * preout->bps;
  534. if(len1)
  535. for(ch=0; ch<preout->ch_count; ch++)
  536. 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);
  537. if(out_count != len1)
  538. for(ch=0; ch<preout->ch_count; ch++)
  539. 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);
  540. } else {
  541. for(ch=0; ch<preout->ch_count; ch++)
  542. 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);
  543. }
  544. } else {
  545. switch(s->int_sample_fmt) {
  546. case AV_SAMPLE_FMT_S16P :swri_noise_shaping_int16(s, conv_src, preout, &s->dither.noise, out_count); break;
  547. case AV_SAMPLE_FMT_S32P :swri_noise_shaping_int32(s, conv_src, preout, &s->dither.noise, out_count); break;
  548. case AV_SAMPLE_FMT_FLTP :swri_noise_shaping_float(s, conv_src, preout, &s->dither.noise, out_count); break;
  549. case AV_SAMPLE_FMT_DBLP :swri_noise_shaping_double(s,conv_src, preout, &s->dither.noise, out_count); break;
  550. }
  551. }
  552. s->dither.noise_pos += out_count;
  553. }
  554. //FIXME packed doesn't need more than 1 chan here!
  555. swri_audio_convert(s->out_convert, out, conv_src, out_count);
  556. }
  557. return out_count;
  558. }
  559. int swr_is_initialized(struct SwrContext *s) {
  560. return !!s->in_buffer.ch_count;
  561. }
  562. int attribute_align_arg swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
  563. const uint8_t *in_arg [SWR_CH_MAX], int in_count){
  564. AudioData * in= &s->in;
  565. AudioData *out= &s->out;
  566. if (!swr_is_initialized(s)) {
  567. av_log(s, AV_LOG_ERROR, "Context has not been initialized\n");
  568. return AVERROR(EINVAL);
  569. }
  570. while(s->drop_output > 0){
  571. int ret;
  572. uint8_t *tmp_arg[SWR_CH_MAX];
  573. #define MAX_DROP_STEP 16384
  574. if((ret=swri_realloc_audio(&s->drop_temp, FFMIN(s->drop_output, MAX_DROP_STEP)))<0)
  575. return ret;
  576. reversefill_audiodata(&s->drop_temp, tmp_arg);
  577. s->drop_output *= -1; //FIXME find a less hackish solution
  578. 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
  579. s->drop_output *= -1;
  580. in_count = 0;
  581. if(ret>0) {
  582. s->drop_output -= ret;
  583. if (!s->drop_output && !out_arg)
  584. return 0;
  585. continue;
  586. }
  587. av_assert0(s->drop_output);
  588. return 0;
  589. }
  590. if(!in_arg){
  591. if(s->resample){
  592. if (!s->flushed)
  593. s->resampler->flush(s);
  594. s->resample_in_constraint = 0;
  595. s->flushed = 1;
  596. }else if(!s->in_buffer_count){
  597. return 0;
  598. }
  599. }else
  600. fill_audiodata(in , (void*)in_arg);
  601. fill_audiodata(out, out_arg);
  602. if(s->resample){
  603. int ret = swr_convert_internal(s, out, out_count, in, in_count);
  604. if(ret>0 && !s->drop_output)
  605. s->outpts += ret * (int64_t)s->in_sample_rate;
  606. return ret;
  607. }else{
  608. AudioData tmp= *in;
  609. int ret2=0;
  610. int ret, size;
  611. size = FFMIN(out_count, s->in_buffer_count);
  612. if(size){
  613. buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
  614. ret= swr_convert_internal(s, out, size, &tmp, size);
  615. if(ret<0)
  616. return ret;
  617. ret2= ret;
  618. s->in_buffer_count -= ret;
  619. s->in_buffer_index += ret;
  620. buf_set(out, out, ret);
  621. out_count -= ret;
  622. if(!s->in_buffer_count)
  623. s->in_buffer_index = 0;
  624. }
  625. if(in_count){
  626. size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
  627. if(in_count > out_count) { //FIXME move after swr_convert_internal
  628. if( size > s->in_buffer.count
  629. && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
  630. buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
  631. copy(&s->in_buffer, &tmp, s->in_buffer_count);
  632. s->in_buffer_index=0;
  633. }else
  634. if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
  635. return ret;
  636. }
  637. if(out_count){
  638. size = FFMIN(in_count, out_count);
  639. ret= swr_convert_internal(s, out, size, in, size);
  640. if(ret<0)
  641. return ret;
  642. buf_set(in, in, ret);
  643. in_count -= ret;
  644. ret2 += ret;
  645. }
  646. if(in_count){
  647. buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
  648. copy(&tmp, in, in_count);
  649. s->in_buffer_count += in_count;
  650. }
  651. }
  652. if(ret2>0 && !s->drop_output)
  653. s->outpts += ret2 * (int64_t)s->in_sample_rate;
  654. return ret2;
  655. }
  656. }
  657. int swr_drop_output(struct SwrContext *s, int count){
  658. const uint8_t *tmp_arg[SWR_CH_MAX];
  659. s->drop_output += count;
  660. if(s->drop_output <= 0)
  661. return 0;
  662. av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
  663. return swr_convert(s, NULL, s->drop_output, tmp_arg, 0);
  664. }
  665. int swr_inject_silence(struct SwrContext *s, int count){
  666. int ret, i;
  667. uint8_t *tmp_arg[SWR_CH_MAX];
  668. if(count <= 0)
  669. return 0;
  670. #define MAX_SILENCE_STEP 16384
  671. while (count > MAX_SILENCE_STEP) {
  672. if ((ret = swr_inject_silence(s, MAX_SILENCE_STEP)) < 0)
  673. return ret;
  674. count -= MAX_SILENCE_STEP;
  675. }
  676. if((ret=swri_realloc_audio(&s->silence, count))<0)
  677. return ret;
  678. if(s->silence.planar) for(i=0; i<s->silence.ch_count; i++) {
  679. memset(s->silence.ch[i], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps);
  680. } else
  681. memset(s->silence.ch[0], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps*s->silence.ch_count);
  682. reversefill_audiodata(&s->silence, tmp_arg);
  683. av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
  684. ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
  685. return ret;
  686. }
  687. int64_t swr_get_delay(struct SwrContext *s, int64_t base){
  688. if (s->resampler && s->resample){
  689. return s->resampler->get_delay(s, base);
  690. }else{
  691. return (s->in_buffer_count*base + (s->in_sample_rate>>1))/ s->in_sample_rate;
  692. }
  693. }
  694. int swr_set_compensation(struct SwrContext *s, int sample_delta, int compensation_distance){
  695. int ret;
  696. if (!s || compensation_distance < 0)
  697. return AVERROR(EINVAL);
  698. if (!compensation_distance && sample_delta)
  699. return AVERROR(EINVAL);
  700. if (!s->resample) {
  701. s->flags |= SWR_FLAG_RESAMPLE;
  702. ret = swr_init(s);
  703. if (ret < 0)
  704. return ret;
  705. }
  706. if (!s->resampler->set_compensation){
  707. return AVERROR(EINVAL);
  708. }else{
  709. return s->resampler->set_compensation(s->resample, sample_delta, compensation_distance);
  710. }
  711. }
  712. int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
  713. if(pts == INT64_MIN)
  714. return s->outpts;
  715. if (s->firstpts == AV_NOPTS_VALUE)
  716. s->outpts = s->firstpts = pts;
  717. if(s->min_compensation >= FLT_MAX) {
  718. return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
  719. } else {
  720. 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;
  721. double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
  722. if(fabs(fdelta) > s->min_compensation) {
  723. if(s->outpts == s->firstpts || fabs(fdelta) > s->min_hard_compensation){
  724. int ret;
  725. if(delta > 0) ret = swr_inject_silence(s, delta / s->out_sample_rate);
  726. else ret = swr_drop_output (s, -delta / s-> in_sample_rate);
  727. if(ret<0){
  728. av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
  729. }
  730. } else if(s->soft_compensation_duration && s->max_soft_compensation) {
  731. int duration = s->out_sample_rate * s->soft_compensation_duration;
  732. double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
  733. int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
  734. av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
  735. swr_set_compensation(s, comp, duration);
  736. }
  737. }
  738. return s->outpts;
  739. }
  740. }