2 * Copyright (C) 2011 Michael Niedermayer (michaelni@gmx.at)
4 * This file is part of libswresample
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.
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.
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
21 #include "libavutil/opt.h"
22 #include "swresample_internal.h"
23 #include "audioconvert.h"
24 #include "libavutil/avassert.h"
25 #include "libavutil/audioconvert.h"
28 #define C15DB 1.189207115
30 #define C_15DB 0.840896415
31 #define C_30DB M_SQRT1_2
32 #define C_45DB 0.594603558
36 //TODO split options array out?
37 #define OFFSET(x) offsetof(SwrContext,x)
38 static const AVOption options[]={
39 {"ich", "input channel count", OFFSET( in.ch_count ), AV_OPT_TYPE_INT, {.dbl=2}, 0, SWR_CH_MAX, 0},
40 {"och", "output channel count", OFFSET(out.ch_count ), AV_OPT_TYPE_INT, {.dbl=2}, 0, SWR_CH_MAX, 0},
41 {"uch", "used channel count", OFFSET(used_ch_count ), AV_OPT_TYPE_INT, {.dbl=0}, 0, SWR_CH_MAX, 0},
42 {"isr", "input sample rate" , OFFSET( in_sample_rate), AV_OPT_TYPE_INT, {.dbl=48000}, 1, INT_MAX, 0},
43 {"osr", "output sample rate" , OFFSET(out_sample_rate), AV_OPT_TYPE_INT, {.dbl=48000}, 1, INT_MAX, 0},
44 //{"ip" , "input planar" , OFFSET( in.planar ), AV_OPT_TYPE_INT, {.dbl=0}, 0, 1, 0},
45 //{"op" , "output planar" , OFFSET(out.planar ), AV_OPT_TYPE_INT, {.dbl=0}, 0, 1, 0},
46 {"isf", "input sample format", OFFSET( in_sample_fmt ), AV_OPT_TYPE_INT, {.dbl=AV_SAMPLE_FMT_S16}, 0, AV_SAMPLE_FMT_NB-1+256, 0},
47 {"osf", "output sample format", OFFSET(out_sample_fmt ), AV_OPT_TYPE_INT, {.dbl=AV_SAMPLE_FMT_S16}, 0, AV_SAMPLE_FMT_NB-1+256, 0},
48 {"tsf", "internal sample format", OFFSET(int_sample_fmt ), AV_OPT_TYPE_INT, {.dbl=AV_SAMPLE_FMT_NONE}, -1, AV_SAMPLE_FMT_FLT, 0},
49 {"icl", "input channel layout" , OFFSET( in_ch_layout), AV_OPT_TYPE_INT64, {.dbl=0}, 0, INT64_MAX, 0, "channel_layout"},
50 {"ocl", "output channel layout", OFFSET(out_ch_layout), AV_OPT_TYPE_INT64, {.dbl=0}, 0, INT64_MAX, 0, "channel_layout"},
51 {"clev", "center mix level" , OFFSET(clev) , AV_OPT_TYPE_FLOAT, {.dbl=C_30DB}, 0, 4, 0},
52 {"slev", "sourround mix level" , OFFSET(slev) , AV_OPT_TYPE_FLOAT, {.dbl=C_30DB}, 0, 4, 0},
53 {"rmvol", "rematrix volume" , OFFSET(rematrix_volume), AV_OPT_TYPE_FLOAT, {.dbl=1.0}, -1000, 1000, 0},
54 {"flags", NULL , OFFSET(flags) , AV_OPT_TYPE_FLAGS, {.dbl=0}, 0, UINT_MAX, 0, "flags"},
55 {"res", "force resampling", 0, AV_OPT_TYPE_CONST, {.dbl=SWR_FLAG_RESAMPLE}, INT_MIN, INT_MAX, 0, "flags"},
56 {"dither", "dither method" , OFFSET(dither_method), AV_OPT_TYPE_INT, {.dbl=0}, 0, SWR_DITHER_NB-1, 0, "dither_method"},
57 {"rectangular", "rectangular dither", 0, AV_OPT_TYPE_CONST, {.dbl=SWR_DITHER_RECTANGULAR}, INT_MIN, INT_MAX, 0, "dither_method"},
62 static const char* context_to_name(void* ptr) {
66 static const AVClass av_class = {
67 .class_name = "SwrContext",
68 .item_name = context_to_name,
70 .version = LIBAVUTIL_VERSION_INT,
71 .log_level_offset_offset = OFFSET(log_level_offset),
72 .parent_log_context_offset = OFFSET(log_ctx),
75 unsigned swresample_version(void)
77 av_assert0(LIBSWRESAMPLE_VERSION_MICRO >= 100);
78 return LIBSWRESAMPLE_VERSION_INT;
81 const char *swresample_configuration(void)
83 return FFMPEG_CONFIGURATION;
86 const char *swresample_license(void)
88 #define LICENSE_PREFIX "libswresample license: "
89 return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
92 int swr_set_channel_mapping(struct SwrContext *s, const int *channel_map){
93 if(!s || s->in_convert) // s needs to be allocated but not initialized
94 return AVERROR(EINVAL);
95 s->channel_map = channel_map;
99 struct SwrContext *swr_alloc(void){
100 SwrContext *s= av_mallocz(sizeof(SwrContext));
102 s->av_class= &av_class;
103 av_opt_set_defaults(s);
108 struct SwrContext *swr_alloc_set_opts(struct SwrContext *s,
109 int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
110 int64_t in_ch_layout, enum AVSampleFormat in_sample_fmt, int in_sample_rate,
111 int log_offset, void *log_ctx){
112 if(!s) s= swr_alloc();
115 s->log_level_offset= log_offset;
118 av_opt_set_int(s, "ocl", out_ch_layout, 0);
119 av_opt_set_int(s, "osf", out_sample_fmt, 0);
120 av_opt_set_int(s, "osr", out_sample_rate, 0);
121 av_opt_set_int(s, "icl", in_ch_layout, 0);
122 av_opt_set_int(s, "isf", in_sample_fmt, 0);
123 av_opt_set_int(s, "isr", in_sample_rate, 0);
124 av_opt_set_int(s, "tsf", AV_SAMPLE_FMT_NONE, 0);
125 av_opt_set_int(s, "ich", av_get_channel_layout_nb_channels(s-> in_ch_layout), 0);
126 av_opt_set_int(s, "och", av_get_channel_layout_nb_channels(s->out_ch_layout), 0);
127 av_opt_set_int(s, "uch", 0, 0);
132 static void free_temp(AudioData *a){
134 memset(a, 0, sizeof(*a));
137 void swr_free(SwrContext **ss){
140 free_temp(&s->postin);
141 free_temp(&s->midbuf);
142 free_temp(&s->preout);
143 free_temp(&s->in_buffer);
144 free_temp(&s->dither);
145 swri_audio_convert_free(&s-> in_convert);
146 swri_audio_convert_free(&s->out_convert);
147 swri_audio_convert_free(&s->full_convert);
148 swri_resample_free(&s->resample);
154 int swr_init(struct SwrContext *s){
155 s->in_buffer_index= 0;
156 s->in_buffer_count= 0;
157 s->resample_in_constraint= 0;
158 free_temp(&s->postin);
159 free_temp(&s->midbuf);
160 free_temp(&s->preout);
161 free_temp(&s->in_buffer);
162 free_temp(&s->dither);
163 swri_audio_convert_free(&s-> in_convert);
164 swri_audio_convert_free(&s->out_convert);
165 swri_audio_convert_free(&s->full_convert);
169 s-> in.planar= av_sample_fmt_is_planar(s-> in_sample_fmt);
170 s->out.planar= av_sample_fmt_is_planar(s->out_sample_fmt);
171 s-> in_sample_fmt= av_get_alt_sample_fmt(s-> in_sample_fmt, 0);
172 s->out_sample_fmt= av_get_alt_sample_fmt(s->out_sample_fmt, 0);
174 if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
175 av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
176 return AVERROR(EINVAL);
178 if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
179 av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
180 return AVERROR(EINVAL);
183 //FIXME should we allow/support using FLT on material that doesnt need it ?
184 if(s->in_sample_fmt <= AV_SAMPLE_FMT_S16 || s->int_sample_fmt==AV_SAMPLE_FMT_S16){
185 s->int_sample_fmt= AV_SAMPLE_FMT_S16;
187 s->int_sample_fmt= AV_SAMPLE_FMT_FLT;
189 if( s->int_sample_fmt != AV_SAMPLE_FMT_S16
190 &&s->int_sample_fmt != AV_SAMPLE_FMT_S32
191 &&s->int_sample_fmt != AV_SAMPLE_FMT_FLT){
192 av_log(s, AV_LOG_ERROR, "Requested sample format %s is not supported internally, S16/S32/FLT is supported\n", av_get_sample_fmt_name(s->int_sample_fmt));
193 return AVERROR(EINVAL);
196 if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
197 s->resample = swri_resample_init(s->resample, s->out_sample_rate, s->in_sample_rate, 16, 10, 0, 0.8, s->int_sample_fmt);
199 swri_resample_free(&s->resample);
200 if( s->int_sample_fmt != AV_SAMPLE_FMT_S16
201 && s->int_sample_fmt != AV_SAMPLE_FMT_S32
202 && s->int_sample_fmt != AV_SAMPLE_FMT_FLT
204 av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16/s32/flt\n");
208 if(!s->used_ch_count)
209 s->used_ch_count= s->in.ch_count;
211 if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
212 av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
216 if(!s-> in_ch_layout)
217 s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
218 if(!s->out_ch_layout)
219 s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
221 s->rematrix= s->out_ch_layout !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
224 #define RSC 1 //FIXME finetune
226 s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
227 if(!s->used_ch_count)
228 s->used_ch_count= s->in.ch_count;
230 s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
232 if(!s-> in.ch_count){
233 av_assert0(!s->in_ch_layout);
234 av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
238 if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
239 av_log(s, AV_LOG_ERROR, "Rematrix is needed but there is not enough information to do it\n");
243 av_assert0(s->used_ch_count);
244 av_assert0(s->out.ch_count);
245 s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
247 s-> in.bps= av_get_bytes_per_sample(s-> in_sample_fmt);
248 s->int_bps= av_get_bytes_per_sample(s->int_sample_fmt);
249 s->out.bps= av_get_bytes_per_sample(s->out_sample_fmt);
252 if(!s->resample && !s->rematrix && !s->channel_map){
253 s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
254 s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
258 s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
259 s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
260 s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
261 s->int_sample_fmt, s->out.ch_count, NULL, 0);
270 s->midbuf.ch_count= s->used_ch_count;
272 s->in_buffer.ch_count= s->used_ch_count;
274 if(!s->resample_first){
275 s->midbuf.ch_count= s->out.ch_count;
277 s->in_buffer.ch_count = s->out.ch_count;
280 s->postin.bps = s->midbuf.bps = s->preout.bps = s->int_bps;
281 s->postin.planar = s->midbuf.planar = s->preout.planar = 1;
284 s->in_buffer.bps = s->int_bps;
285 s->in_buffer.planar = 1;
288 s->dither = s->preout;
291 return swri_rematrix_init(s);
296 static int realloc_audio(AudioData *a, int count){
300 if(a->count >= count)
305 countb= FFALIGN(count*a->bps, 32);
309 av_assert0(a->ch_count);
311 a->data= av_malloc(countb*a->ch_count);
313 return AVERROR(ENOMEM);
314 for(i=0; i<a->ch_count; i++){
315 a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
316 if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
318 if(!a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
325 static void copy(AudioData *out, AudioData *in,
327 av_assert0(out->planar == in->planar);
328 av_assert0(out->bps == in->bps);
329 av_assert0(out->ch_count == in->ch_count);
332 for(ch=0; ch<out->ch_count; ch++)
333 memcpy(out->ch[ch], in->ch[ch], count*out->bps);
335 memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
338 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
341 for(i=0; i<out->ch_count; i++)
342 out->ch[i]= in_arg[i];
344 for(i=0; i<out->ch_count; i++)
345 out->ch[i]= in_arg[0] + i*out->bps;
351 * out may be equal in.
353 static void buf_set(AudioData *out, AudioData *in, int count){
356 for(ch=0; ch<out->ch_count; ch++)
357 out->ch[ch]= in->ch[ch] + count*out->bps;
359 out->ch[0]= in->ch[0] + count*out->ch_count*out->bps;
364 * @return number of samples output per channel
366 static int resample(SwrContext *s, AudioData *out_param, int out_count,
367 const AudioData * in_param, int in_count){
368 AudioData in, out, tmp;
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= swri_multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
382 buf_set(&out, &out, ret);
383 s->in_buffer_count -= consumed;
384 s->in_buffer_index += consumed;
388 if(s->in_buffer_count <= border){
389 buf_set(&in, &in, -s->in_buffer_count);
390 in_count += s->in_buffer_count;
391 s->in_buffer_count=0;
392 s->in_buffer_index=0;
397 if(in_count && !s->in_buffer_count){
398 s->in_buffer_index=0;
399 ret= swri_multiple_resample(s->resample, &out, out_count, &in, in_count, &consumed);
402 buf_set(&out, &out, ret);
403 in_count -= consumed;
404 buf_set(&in, &in, consumed);
407 //TODO is this check sane considering the advanced copy avoidance below
408 size= s->in_buffer_index + s->in_buffer_count + in_count;
409 if( size > s->in_buffer.count
410 && s->in_buffer_count + in_count <= s->in_buffer_index){
411 buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
412 copy(&s->in_buffer, &tmp, s->in_buffer_count);
413 s->in_buffer_index=0;
415 if((ret=realloc_audio(&s->in_buffer, size)) < 0)
420 if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
422 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
423 copy(&tmp, &in, /*in_*/count);
424 s->in_buffer_count += count;
427 buf_set(&in, &in, count);
428 s->resample_in_constraint= 0;
429 if(s->in_buffer_count != count || in_count)
435 s->resample_in_constraint= !!out_count;
440 static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
441 AudioData *in , int in_count){
442 AudioData *postin, *midbuf, *preout;
444 AudioData preout_tmp, midbuf_tmp;
447 av_assert0(!s->resample);
448 swri_audio_convert(s->full_convert, out, in, in_count);
452 // in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
453 // in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
455 if((ret=realloc_audio(&s->postin, in_count))<0)
457 if(s->resample_first){
458 av_assert0(s->midbuf.ch_count == s->used_ch_count);
459 if((ret=realloc_audio(&s->midbuf, out_count))<0)
462 av_assert0(s->midbuf.ch_count == s->out.ch_count);
463 if((ret=realloc_audio(&s->midbuf, in_count))<0)
466 if((ret=realloc_audio(&s->preout, out_count))<0)
471 midbuf_tmp= s->midbuf;
473 preout_tmp= s->preout;
476 if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar)
479 if(s->resample_first ? !s->resample : !s->rematrix)
482 if(s->resample_first ? !s->rematrix : !s->resample)
485 if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar){
487 out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
488 av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
489 copy(out, in, out_count);
492 else if(preout==postin) preout= midbuf= postin= out;
493 else if(preout==midbuf) preout= midbuf= out;
498 swri_audio_convert(s->in_convert, postin, in, in_count);
501 if(s->resample_first){
503 out_count= resample(s, midbuf, out_count, postin, in_count);
505 swri_rematrix(s, preout, midbuf, out_count, preout==out);
508 swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
510 out_count= resample(s, preout, out_count, midbuf, in_count);
513 if(preout != out && out_count){
514 if(s->dither_method){
516 av_assert0(preout != in);
518 if((ret=realloc_audio(&s->dither, out_count))<0)
521 for(ch=0; ch<s->dither.ch_count; ch++)
522 swri_get_dither(s->dither.ch[ch], s->dither.count, 12345678913579<<ch, s->out_sample_fmt, s->int_sample_fmt, s->dither_method);
523 av_assert0(s->dither.ch_count == preout->ch_count);
525 for(ch=0; ch<preout->ch_count; ch++){
526 swri_sum2(s->int_sample_fmt, preout->ch[ch], preout->ch[ch], s->dither.ch[ch], 1, 1, out_count);
529 //FIXME packed doesnt need more than 1 chan here!
530 swri_audio_convert(s->out_convert, out, preout, out_count);
535 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
536 const uint8_t *in_arg [SWR_CH_MAX], int in_count){
537 AudioData * in= &s->in;
538 AudioData *out= &s->out;
541 if(s->in_buffer_count){
542 if (s->resample && !s->flushed) {
543 AudioData *a= &s->in_buffer;
545 if((ret=realloc_audio(a, s->in_buffer_index + 2*s->in_buffer_count)) < 0)
547 av_assert0(a->planar);
548 for(i=0; i<a->ch_count; i++){
549 for(j=0; j<s->in_buffer_count; j++){
550 memcpy(a->ch[i] + (s->in_buffer_index+s->in_buffer_count+j )*a->bps,
551 a->ch[i] + (s->in_buffer_index+s->in_buffer_count-j-1)*a->bps, a->bps);
554 s->in_buffer_count += (s->in_buffer_count+1)/2;
555 s->resample_in_constraint = 0;
562 fill_audiodata(in , (void*)in_arg);
564 fill_audiodata(out, out_arg);
567 return swr_convert_internal(s, out, out_count, in, in_count);
572 size = FFMIN(out_count, s->in_buffer_count);
574 buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
575 ret= swr_convert_internal(s, out, size, &tmp, size);
579 s->in_buffer_count -= ret;
580 s->in_buffer_index += ret;
581 buf_set(out, out, ret);
583 if(!s->in_buffer_count)
584 s->in_buffer_index = 0;
588 size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
590 if(in_count > out_count) { //FIXME move after swr_convert_internal
591 if( size > s->in_buffer.count
592 && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
593 buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
594 copy(&s->in_buffer, &tmp, s->in_buffer_count);
595 s->in_buffer_index=0;
597 if((ret=realloc_audio(&s->in_buffer, size)) < 0)
602 size = FFMIN(in_count, out_count);
603 ret= swr_convert_internal(s, out, size, in, size);
606 buf_set(in, in, ret);
611 buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
612 copy(&tmp, in, in_count);
613 s->in_buffer_count += in_count;