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"},
60 static const char* context_to_name(void* ptr) {
64 static const AVClass av_class = {
65 .class_name = "SwrContext",
66 .item_name = context_to_name,
68 .version = LIBAVUTIL_VERSION_INT,
69 .log_level_offset_offset = OFFSET(log_level_offset),
70 .parent_log_context_offset = OFFSET(log_ctx),
73 unsigned swresample_version(void)
75 av_assert0(LIBSWRESAMPLE_VERSION_MICRO >= 100);
76 return LIBSWRESAMPLE_VERSION_INT;
79 const char *swresample_configuration(void)
81 return FFMPEG_CONFIGURATION;
84 const char *swresample_license(void)
86 #define LICENSE_PREFIX "libswresample license: "
87 return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
90 int swr_set_channel_mapping(struct SwrContext *s, const int *channel_map){
91 if(!s || s->in_convert) // s needs to be allocated but not initialized
92 return AVERROR(EINVAL);
93 s->channel_map = channel_map;
97 struct SwrContext *swr_alloc(void){
98 SwrContext *s= av_mallocz(sizeof(SwrContext));
100 s->av_class= &av_class;
101 av_opt_set_defaults(s);
106 struct SwrContext *swr_alloc_set_opts(struct SwrContext *s,
107 int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
108 int64_t in_ch_layout, enum AVSampleFormat in_sample_fmt, int in_sample_rate,
109 int log_offset, void *log_ctx){
110 if(!s) s= swr_alloc();
113 s->log_level_offset= log_offset;
116 av_opt_set_int(s, "ocl", out_ch_layout, 0);
117 av_opt_set_int(s, "osf", out_sample_fmt, 0);
118 av_opt_set_int(s, "osr", out_sample_rate, 0);
119 av_opt_set_int(s, "icl", in_ch_layout, 0);
120 av_opt_set_int(s, "isf", in_sample_fmt, 0);
121 av_opt_set_int(s, "isr", in_sample_rate, 0);
122 av_opt_set_int(s, "tsf", AV_SAMPLE_FMT_S16, 0);
123 av_opt_set_int(s, "ich", av_get_channel_layout_nb_channels(s-> in_ch_layout), 0);
124 av_opt_set_int(s, "och", av_get_channel_layout_nb_channels(s->out_ch_layout), 0);
125 av_opt_set_int(s, "uch", 0, 0);
130 static void free_temp(AudioData *a){
132 memset(a, 0, sizeof(*a));
135 void swr_free(SwrContext **ss){
138 free_temp(&s->postin);
139 free_temp(&s->midbuf);
140 free_temp(&s->preout);
141 free_temp(&s->in_buffer);
142 swri_audio_convert_free(&s-> in_convert);
143 swri_audio_convert_free(&s->out_convert);
144 swri_audio_convert_free(&s->full_convert);
145 swri_resample_free(&s->resample);
151 int swr_init(struct SwrContext *s){
152 s->in_buffer_index= 0;
153 s->in_buffer_count= 0;
154 s->resample_in_constraint= 0;
155 free_temp(&s->postin);
156 free_temp(&s->midbuf);
157 free_temp(&s->preout);
158 free_temp(&s->in_buffer);
159 swri_audio_convert_free(&s-> in_convert);
160 swri_audio_convert_free(&s->out_convert);
161 swri_audio_convert_free(&s->full_convert);
163 s-> in.planar= av_sample_fmt_is_planar(s-> in_sample_fmt);
164 s->out.planar= av_sample_fmt_is_planar(s->out_sample_fmt);
165 s-> in_sample_fmt= av_get_alt_sample_fmt(s-> in_sample_fmt, 0);
166 s->out_sample_fmt= av_get_alt_sample_fmt(s->out_sample_fmt, 0);
168 if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
169 av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
170 return AVERROR(EINVAL);
172 if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
173 av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
174 return AVERROR(EINVAL);
177 if( s->int_sample_fmt != AV_SAMPLE_FMT_S16
178 &&s->int_sample_fmt != AV_SAMPLE_FMT_FLT){
179 av_log(s, AV_LOG_ERROR, "Requested sample format %s is not supported internally, only float & S16 is supported\n", av_get_sample_fmt_name(s->int_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;
190 if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
191 s->resample = swri_resample_init(s->resample, s->out_sample_rate, s->in_sample_rate, 16, 10, 0, 0.8);
193 swri_resample_free(&s->resample);
194 if(s->int_sample_fmt != AV_SAMPLE_FMT_S16 && s->resample){
195 av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16 currently\n"); //FIXME
199 if(!s->used_ch_count)
200 s->used_ch_count= s->in.ch_count;
202 if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
203 av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
207 if(!s-> in_ch_layout)
208 s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
209 if(!s->out_ch_layout)
210 s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
212 s->rematrix= s->out_ch_layout !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
215 #define RSC 1 //FIXME finetune
217 s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
218 if(!s->used_ch_count)
219 s->used_ch_count= s->in.ch_count;
221 s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
223 if(!s-> in.ch_count){
224 av_assert0(!s->in_ch_layout);
225 av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
229 av_assert0(s->used_ch_count);
230 av_assert0(s->out.ch_count);
231 s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
233 s-> in.bps= av_get_bytes_per_sample(s-> in_sample_fmt);
234 s->int_bps= av_get_bytes_per_sample(s->int_sample_fmt);
235 s->out.bps= av_get_bytes_per_sample(s->out_sample_fmt);
237 if(!s->resample && !s->rematrix && !s->channel_map){
238 s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
239 s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
243 s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
244 s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
245 s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
246 s->int_sample_fmt, s->out.ch_count, NULL, 0);
256 s->in_buffer.ch_count= s->used_ch_count;
258 if(!s->resample_first){
259 s->midbuf.ch_count= s->out.ch_count;
260 s->in_buffer.ch_count = s->out.ch_count;
263 s->in_buffer.bps = s->postin.bps = s->midbuf.bps = s->preout.bps = s->int_bps;
264 s->in_buffer.planar = s->postin.planar = s->midbuf.planar = s->preout.planar = 1;
268 return swri_rematrix_init(s);
273 static int realloc_audio(AudioData *a, int count){
277 if(a->count >= count)
282 countb= FFALIGN(count*a->bps, 32);
285 av_assert0(a->planar);
287 av_assert0(a->ch_count);
289 a->data= av_malloc(countb*a->ch_count);
291 return AVERROR(ENOMEM);
292 for(i=0; i<a->ch_count; i++){
293 a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
294 if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
302 static void copy(AudioData *out, AudioData *in,
304 av_assert0(out->planar == in->planar);
305 av_assert0(out->bps == in->bps);
306 av_assert0(out->ch_count == in->ch_count);
309 for(ch=0; ch<out->ch_count; ch++)
310 memcpy(out->ch[ch], in->ch[ch], count*out->bps);
312 memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
315 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
318 for(i=0; i<out->ch_count; i++)
319 out->ch[i]= in_arg[i];
321 for(i=0; i<out->ch_count; i++)
322 out->ch[i]= in_arg[0] + i*out->bps;
328 * out may be equal in.
330 static void buf_set(AudioData *out, AudioData *in, int count){
333 for(ch=0; ch<out->ch_count; ch++)
334 out->ch[ch]= in->ch[ch] + count*out->bps;
336 out->ch[0]= in->ch[0] + count*out->ch_count*out->bps;
341 * @return number of samples output per channel
343 static int resample(SwrContext *s, AudioData *out_param, int out_count,
344 const AudioData * in_param, int in_count){
345 AudioData in, out, tmp;
353 int ret, size, consumed;
354 if(!s->resample_in_constraint && s->in_buffer_count){
355 buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
356 ret= swri_multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
359 buf_set(&out, &out, ret);
360 s->in_buffer_count -= consumed;
361 s->in_buffer_index += consumed;
365 if(s->in_buffer_count <= border){
366 buf_set(&in, &in, -s->in_buffer_count);
367 in_count += s->in_buffer_count;
368 s->in_buffer_count=0;
369 s->in_buffer_index=0;
374 if(in_count && !s->in_buffer_count){
375 s->in_buffer_index=0;
376 ret= swri_multiple_resample(s->resample, &out, out_count, &in, in_count, &consumed);
379 buf_set(&out, &out, ret);
380 in_count -= consumed;
381 buf_set(&in, &in, consumed);
384 //TODO is this check sane considering the advanced copy avoidance below
385 size= s->in_buffer_index + s->in_buffer_count + in_count;
386 if( size > s->in_buffer.count
387 && s->in_buffer_count + in_count <= s->in_buffer_index){
388 buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
389 copy(&s->in_buffer, &tmp, s->in_buffer_count);
390 s->in_buffer_index=0;
392 if((ret=realloc_audio(&s->in_buffer, size)) < 0)
397 if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
399 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
400 copy(&tmp, &in, /*in_*/count);
401 s->in_buffer_count += count;
404 buf_set(&in, &in, count);
405 s->resample_in_constraint= 0;
406 if(s->in_buffer_count != count || in_count)
412 s->resample_in_constraint= !!out_count;
417 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
418 const uint8_t *in_arg [SWR_CH_MAX], int in_count){
419 AudioData *postin, *midbuf, *preout;
421 AudioData * in= &s->in;
422 AudioData *out= &s->out;
423 AudioData preout_tmp, midbuf_tmp;
426 if(in_count > out_count)
428 out_count = in_count;
432 if(s->in_buffer_count){
433 AudioData *a= &s->in_buffer;
435 if((ret=realloc_audio(a, s->in_buffer_index + 2*s->in_buffer_count)) < 0)
437 av_assert0(a->planar);
438 for(i=0; i<a->ch_count; i++){
439 for(j=0; j<s->in_buffer_count; j++){
440 memcpy(a->ch[i] + (s->in_buffer_index+s->in_buffer_count+j )*a->bps,
441 a->ch[i] + (s->in_buffer_index+s->in_buffer_count-j-1)*a->bps, a->bps);
444 s->in_buffer_count += (s->in_buffer_count+1)/2;
445 s->resample_in_constraint = 0;
450 fill_audiodata(in , (void*)in_arg);
451 fill_audiodata(out, out_arg);
454 av_assert0(!s->resample);
455 swri_audio_convert(s->full_convert, out, in, in_count);
459 // in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
460 // in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
462 if((ret=realloc_audio(&s->postin, in_count))<0)
464 if(s->resample_first){
465 av_assert0(s->midbuf.ch_count == s->used_ch_count);
466 if((ret=realloc_audio(&s->midbuf, out_count))<0)
469 av_assert0(s->midbuf.ch_count == s->out.ch_count);
470 if((ret=realloc_audio(&s->midbuf, in_count))<0)
473 if((ret=realloc_audio(&s->preout, out_count))<0)
478 midbuf_tmp= s->midbuf;
480 preout_tmp= s->preout;
483 if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar)
486 if(s->resample_first ? !s->resample : !s->rematrix)
489 if(s->resample_first ? !s->rematrix : !s->resample)
492 if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar){
494 out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
495 av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
496 copy(out, in, out_count);
499 else if(preout==postin) preout= midbuf= postin= out;
500 else if(preout==midbuf) preout= midbuf= out;
505 swri_audio_convert(s->in_convert, postin, in, in_count);
508 if(s->resample_first){
510 out_count= resample(s, midbuf, out_count, postin, in_count);
512 swri_rematrix(s, preout, midbuf, out_count, preout==out);
515 swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
517 out_count= resample(s, preout, out_count, midbuf, in_count);
521 //FIXME packed doesnt need more than 1 chan here!
522 swri_audio_convert(s->out_convert, out, preout, out_count);
525 s->in_buffer_count = 0;