2 * Copyright (C) 2011-2012 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 "swresample_internal.h"
22 #include "libavutil/avassert.h"
23 #include "libavutil/channel_layout.h"
25 #define TEMPLATE_REMATRIX_FLT
26 #include "rematrix_template.c"
27 #undef TEMPLATE_REMATRIX_FLT
29 #define TEMPLATE_REMATRIX_DBL
30 #include "rematrix_template.c"
31 #undef TEMPLATE_REMATRIX_DBL
33 #define TEMPLATE_REMATRIX_S16
34 #include "rematrix_template.c"
35 #undef TEMPLATE_REMATRIX_S16
37 #define TEMPLATE_REMATRIX_S32
38 #include "rematrix_template.c"
39 #undef TEMPLATE_REMATRIX_S32
43 #define FRONT_CENTER 2
44 #define LOW_FREQUENCY 3
47 #define FRONT_LEFT_OF_CENTER 6
48 #define FRONT_RIGHT_OF_CENTER 7
53 #define TOP_FRONT_LEFT 12
54 #define TOP_FRONT_CENTER 13
55 #define TOP_FRONT_RIGHT 14
56 #define TOP_BACK_LEFT 15
57 #define TOP_BACK_CENTER 16
58 #define TOP_BACK_RIGHT 17
59 #define NUM_NAMED_CHANNELS 18
61 int swr_set_matrix(struct SwrContext *s, const double *matrix, int stride)
63 int nb_in, nb_out, in, out;
65 if (!s || s->in_convert) // s needs to be allocated but not initialized
66 return AVERROR(EINVAL);
67 memset(s->matrix, 0, sizeof(s->matrix));
68 nb_in = av_get_channel_layout_nb_channels(s->user_in_ch_layout);
69 nb_out = av_get_channel_layout_nb_channels(s->user_out_ch_layout);
70 for (out = 0; out < nb_out; out++) {
71 for (in = 0; in < nb_in; in++)
72 s->matrix[out][in] = matrix[in];
75 s->rematrix_custom = 1;
79 static int even(int64_t layout){
81 if(layout&(layout-1)) return 1;
85 static int clean_layout(SwrContext *s, int64_t layout){
86 if(layout && layout != AV_CH_FRONT_CENTER && !(layout&(layout-1))) {
88 av_get_channel_layout_string(buf, sizeof(buf), -1, layout);
89 av_log(s, AV_LOG_VERBOSE, "Treating %s as mono\n", buf);
90 return AV_CH_FRONT_CENTER;
96 static int sane_layout(int64_t layout){
97 if(!(layout & AV_CH_LAYOUT_SURROUND)) // at least 1 front speaker
99 if(!even(layout & (AV_CH_FRONT_LEFT | AV_CH_FRONT_RIGHT))) // no asymetric front
101 if(!even(layout & (AV_CH_SIDE_LEFT | AV_CH_SIDE_RIGHT))) // no asymetric side
103 if(!even(layout & (AV_CH_BACK_LEFT | AV_CH_BACK_RIGHT)))
105 if(!even(layout & (AV_CH_FRONT_LEFT_OF_CENTER | AV_CH_FRONT_RIGHT_OF_CENTER)))
107 if(av_get_channel_layout_nb_channels(layout) >= SWR_CH_MAX)
113 av_cold static int auto_matrix(SwrContext *s)
116 double matrix[NUM_NAMED_CHANNELS][NUM_NAMED_CHANNELS]={{0}};
117 int64_t unaccounted, in_ch_layout, out_ch_layout;
120 const int matrix_encoding = s->matrix_encoding;
123 in_ch_layout = clean_layout(s, s->in_ch_layout);
124 out_ch_layout = clean_layout(s, s->out_ch_layout);
126 if( out_ch_layout == AV_CH_LAYOUT_STEREO_DOWNMIX
127 && (in_ch_layout & AV_CH_LAYOUT_STEREO_DOWNMIX) == 0
129 out_ch_layout = AV_CH_LAYOUT_STEREO;
131 if( in_ch_layout == AV_CH_LAYOUT_STEREO_DOWNMIX
132 && (out_ch_layout & AV_CH_LAYOUT_STEREO_DOWNMIX) == 0
134 in_ch_layout = AV_CH_LAYOUT_STEREO;
136 if(!sane_layout(in_ch_layout)){
137 av_get_channel_layout_string(buf, sizeof(buf), -1, s->in_ch_layout);
138 av_log(s, AV_LOG_ERROR, "Input channel layout '%s' is not supported\n", buf);
139 return AVERROR(EINVAL);
142 if(!sane_layout(out_ch_layout)){
143 av_get_channel_layout_string(buf, sizeof(buf), -1, s->out_ch_layout);
144 av_log(s, AV_LOG_ERROR, "Output channel layout '%s' is not supported\n", buf);
145 return AVERROR(EINVAL);
148 memset(s->matrix, 0, sizeof(s->matrix));
149 for(i=0; i<FF_ARRAY_ELEMS(matrix); i++){
150 if(in_ch_layout & out_ch_layout & (1ULL<<i))
154 unaccounted= in_ch_layout & ~out_ch_layout;
156 //FIXME implement dolby surround
157 //FIXME implement full ac3
160 if(unaccounted & AV_CH_FRONT_CENTER){
161 if((out_ch_layout & AV_CH_LAYOUT_STEREO) == AV_CH_LAYOUT_STEREO){
162 if(in_ch_layout & AV_CH_LAYOUT_STEREO) {
163 matrix[ FRONT_LEFT][FRONT_CENTER]+= s->clev;
164 matrix[FRONT_RIGHT][FRONT_CENTER]+= s->clev;
166 matrix[ FRONT_LEFT][FRONT_CENTER]+= M_SQRT1_2;
167 matrix[FRONT_RIGHT][FRONT_CENTER]+= M_SQRT1_2;
172 if(unaccounted & AV_CH_LAYOUT_STEREO){
173 if(out_ch_layout & AV_CH_FRONT_CENTER){
174 matrix[FRONT_CENTER][ FRONT_LEFT]+= M_SQRT1_2;
175 matrix[FRONT_CENTER][FRONT_RIGHT]+= M_SQRT1_2;
176 if(in_ch_layout & AV_CH_FRONT_CENTER)
177 matrix[FRONT_CENTER][ FRONT_CENTER] = s->clev*sqrt(2);
182 if(unaccounted & AV_CH_BACK_CENTER){
183 if(out_ch_layout & AV_CH_BACK_LEFT){
184 matrix[ BACK_LEFT][BACK_CENTER]+= M_SQRT1_2;
185 matrix[BACK_RIGHT][BACK_CENTER]+= M_SQRT1_2;
186 }else if(out_ch_layout & AV_CH_SIDE_LEFT){
187 matrix[ SIDE_LEFT][BACK_CENTER]+= M_SQRT1_2;
188 matrix[SIDE_RIGHT][BACK_CENTER]+= M_SQRT1_2;
189 }else if(out_ch_layout & AV_CH_FRONT_LEFT){
190 if (matrix_encoding == AV_MATRIX_ENCODING_DOLBY ||
191 matrix_encoding == AV_MATRIX_ENCODING_DPLII) {
192 if (unaccounted & (AV_CH_BACK_LEFT | AV_CH_SIDE_LEFT)) {
193 matrix[FRONT_LEFT ][BACK_CENTER] -= s->slev * M_SQRT1_2;
194 matrix[FRONT_RIGHT][BACK_CENTER] += s->slev * M_SQRT1_2;
196 matrix[FRONT_LEFT ][BACK_CENTER] -= s->slev;
197 matrix[FRONT_RIGHT][BACK_CENTER] += s->slev;
200 matrix[ FRONT_LEFT][BACK_CENTER]+= s->slev*M_SQRT1_2;
201 matrix[FRONT_RIGHT][BACK_CENTER]+= s->slev*M_SQRT1_2;
203 }else if(out_ch_layout & AV_CH_FRONT_CENTER){
204 matrix[ FRONT_CENTER][BACK_CENTER]+= s->slev*M_SQRT1_2;
208 if(unaccounted & AV_CH_BACK_LEFT){
209 if(out_ch_layout & AV_CH_BACK_CENTER){
210 matrix[BACK_CENTER][ BACK_LEFT]+= M_SQRT1_2;
211 matrix[BACK_CENTER][BACK_RIGHT]+= M_SQRT1_2;
212 }else if(out_ch_layout & AV_CH_SIDE_LEFT){
213 if(in_ch_layout & AV_CH_SIDE_LEFT){
214 matrix[ SIDE_LEFT][ BACK_LEFT]+= M_SQRT1_2;
215 matrix[SIDE_RIGHT][BACK_RIGHT]+= M_SQRT1_2;
217 matrix[ SIDE_LEFT][ BACK_LEFT]+= 1.0;
218 matrix[SIDE_RIGHT][BACK_RIGHT]+= 1.0;
220 }else if(out_ch_layout & AV_CH_FRONT_LEFT){
221 if (matrix_encoding == AV_MATRIX_ENCODING_DOLBY) {
222 matrix[FRONT_LEFT ][BACK_LEFT ] -= s->slev * M_SQRT1_2;
223 matrix[FRONT_LEFT ][BACK_RIGHT] -= s->slev * M_SQRT1_2;
224 matrix[FRONT_RIGHT][BACK_LEFT ] += s->slev * M_SQRT1_2;
225 matrix[FRONT_RIGHT][BACK_RIGHT] += s->slev * M_SQRT1_2;
226 } else if (matrix_encoding == AV_MATRIX_ENCODING_DPLII) {
227 matrix[FRONT_LEFT ][BACK_LEFT ] -= s->slev * SQRT3_2;
228 matrix[FRONT_LEFT ][BACK_RIGHT] -= s->slev * M_SQRT1_2;
229 matrix[FRONT_RIGHT][BACK_LEFT ] += s->slev * M_SQRT1_2;
230 matrix[FRONT_RIGHT][BACK_RIGHT] += s->slev * SQRT3_2;
232 matrix[ FRONT_LEFT][ BACK_LEFT] += s->slev;
233 matrix[FRONT_RIGHT][BACK_RIGHT] += s->slev;
235 }else if(out_ch_layout & AV_CH_FRONT_CENTER){
236 matrix[ FRONT_CENTER][BACK_LEFT ]+= s->slev*M_SQRT1_2;
237 matrix[ FRONT_CENTER][BACK_RIGHT]+= s->slev*M_SQRT1_2;
242 if(unaccounted & AV_CH_SIDE_LEFT){
243 if(out_ch_layout & AV_CH_BACK_LEFT){
244 /* if back channels do not exist in the input, just copy side
245 channels to back channels, otherwise mix side into back */
246 if (in_ch_layout & AV_CH_BACK_LEFT) {
247 matrix[BACK_LEFT ][SIDE_LEFT ] += M_SQRT1_2;
248 matrix[BACK_RIGHT][SIDE_RIGHT] += M_SQRT1_2;
250 matrix[BACK_LEFT ][SIDE_LEFT ] += 1.0;
251 matrix[BACK_RIGHT][SIDE_RIGHT] += 1.0;
253 }else if(out_ch_layout & AV_CH_BACK_CENTER){
254 matrix[BACK_CENTER][ SIDE_LEFT]+= M_SQRT1_2;
255 matrix[BACK_CENTER][SIDE_RIGHT]+= M_SQRT1_2;
256 }else if(out_ch_layout & AV_CH_FRONT_LEFT){
257 if (matrix_encoding == AV_MATRIX_ENCODING_DOLBY) {
258 matrix[FRONT_LEFT ][SIDE_LEFT ] -= s->slev * M_SQRT1_2;
259 matrix[FRONT_LEFT ][SIDE_RIGHT] -= s->slev * M_SQRT1_2;
260 matrix[FRONT_RIGHT][SIDE_LEFT ] += s->slev * M_SQRT1_2;
261 matrix[FRONT_RIGHT][SIDE_RIGHT] += s->slev * M_SQRT1_2;
262 } else if (matrix_encoding == AV_MATRIX_ENCODING_DPLII) {
263 matrix[FRONT_LEFT ][SIDE_LEFT ] -= s->slev * SQRT3_2;
264 matrix[FRONT_LEFT ][SIDE_RIGHT] -= s->slev * M_SQRT1_2;
265 matrix[FRONT_RIGHT][SIDE_LEFT ] += s->slev * M_SQRT1_2;
266 matrix[FRONT_RIGHT][SIDE_RIGHT] += s->slev * SQRT3_2;
268 matrix[ FRONT_LEFT][ SIDE_LEFT] += s->slev;
269 matrix[FRONT_RIGHT][SIDE_RIGHT] += s->slev;
271 }else if(out_ch_layout & AV_CH_FRONT_CENTER){
272 matrix[ FRONT_CENTER][SIDE_LEFT ]+= s->slev*M_SQRT1_2;
273 matrix[ FRONT_CENTER][SIDE_RIGHT]+= s->slev*M_SQRT1_2;
278 if(unaccounted & AV_CH_FRONT_LEFT_OF_CENTER){
279 if(out_ch_layout & AV_CH_FRONT_LEFT){
280 matrix[ FRONT_LEFT][ FRONT_LEFT_OF_CENTER]+= 1.0;
281 matrix[FRONT_RIGHT][FRONT_RIGHT_OF_CENTER]+= 1.0;
282 }else if(out_ch_layout & AV_CH_FRONT_CENTER){
283 matrix[ FRONT_CENTER][ FRONT_LEFT_OF_CENTER]+= M_SQRT1_2;
284 matrix[ FRONT_CENTER][FRONT_RIGHT_OF_CENTER]+= M_SQRT1_2;
288 /* mix LFE into front left/right or center */
289 if (unaccounted & AV_CH_LOW_FREQUENCY) {
290 if (out_ch_layout & AV_CH_FRONT_CENTER) {
291 matrix[FRONT_CENTER][LOW_FREQUENCY] += s->lfe_mix_level;
292 } else if (out_ch_layout & AV_CH_FRONT_LEFT) {
293 matrix[FRONT_LEFT ][LOW_FREQUENCY] += s->lfe_mix_level * M_SQRT1_2;
294 matrix[FRONT_RIGHT][LOW_FREQUENCY] += s->lfe_mix_level * M_SQRT1_2;
299 for(out_i=i=0; i<64; i++){
302 if((out_ch_layout & (1ULL<<i)) == 0)
305 if((in_ch_layout & (1ULL<<j)) == 0)
307 if (i < FF_ARRAY_ELEMS(matrix) && j < FF_ARRAY_ELEMS(matrix[0]))
308 s->matrix[out_i][in_i]= matrix[i][j];
310 s->matrix[out_i][in_i]= i == j && (in_ch_layout & out_ch_layout & (1ULL<<i));
311 sum += fabs(s->matrix[out_i][in_i]);
314 maxcoef= FFMAX(maxcoef, sum);
317 if(s->rematrix_volume < 0)
318 maxcoef = -s->rematrix_volume;
320 if (s->rematrix_maxval > 0) {
321 maxval = s->rematrix_maxval;
322 } else if ( av_get_packed_sample_fmt(s->out_sample_fmt) < AV_SAMPLE_FMT_FLT
323 || av_get_packed_sample_fmt(s->int_sample_fmt) < AV_SAMPLE_FMT_FLT) {
328 if(maxcoef > maxval || s->rematrix_volume < 0){
330 for(i=0; i<SWR_CH_MAX; i++)
331 for(j=0; j<SWR_CH_MAX; j++){
332 s->matrix[i][j] /= maxcoef;
336 if(s->rematrix_volume > 0){
337 for(i=0; i<SWR_CH_MAX; i++)
338 for(j=0; j<SWR_CH_MAX; j++){
339 s->matrix[i][j] *= s->rematrix_volume;
343 av_log(s, AV_LOG_DEBUG, "Matrix coefficients:\n");
344 for(i=0; i<av_get_channel_layout_nb_channels(out_ch_layout); i++){
346 av_get_channel_name(av_channel_layout_extract_channel(out_ch_layout, i));
347 av_log(s, AV_LOG_DEBUG, "%s: ", c ? c : "?");
348 for(j=0; j<av_get_channel_layout_nb_channels(in_ch_layout); j++){
349 c = av_get_channel_name(av_channel_layout_extract_channel(in_ch_layout, j));
350 av_log(s, AV_LOG_DEBUG, "%s:%f ", c ? c : "?", s->matrix[i][j]);
352 av_log(s, AV_LOG_DEBUG, "\n");
357 av_cold int swri_rematrix_init(SwrContext *s){
359 int nb_in = av_get_channel_layout_nb_channels(s->in_ch_layout);
360 int nb_out = av_get_channel_layout_nb_channels(s->out_ch_layout);
364 if (!s->rematrix_custom) {
365 int r = auto_matrix(s);
369 if (s->midbuf.fmt == AV_SAMPLE_FMT_S16P){
370 s->native_matrix = av_calloc(nb_in * nb_out, sizeof(int));
371 s->native_one = av_mallocz(sizeof(int));
372 if (!s->native_matrix || !s->native_one)
373 return AVERROR(ENOMEM);
374 for (i = 0; i < nb_out; i++)
375 for (j = 0; j < nb_in; j++)
376 ((int*)s->native_matrix)[i * nb_in + j] = lrintf(s->matrix[i][j] * 32768);
377 *((int*)s->native_one) = 32768;
378 s->mix_1_1_f = (mix_1_1_func_type*)copy_s16;
379 s->mix_2_1_f = (mix_2_1_func_type*)sum2_s16;
380 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_s16(s);
381 }else if(s->midbuf.fmt == AV_SAMPLE_FMT_FLTP){
382 s->native_matrix = av_calloc(nb_in * nb_out, sizeof(float));
383 s->native_one = av_mallocz(sizeof(float));
384 if (!s->native_matrix || !s->native_one)
385 return AVERROR(ENOMEM);
386 for (i = 0; i < nb_out; i++)
387 for (j = 0; j < nb_in; j++)
388 ((float*)s->native_matrix)[i * nb_in + j] = s->matrix[i][j];
389 *((float*)s->native_one) = 1.0;
390 s->mix_1_1_f = (mix_1_1_func_type*)copy_float;
391 s->mix_2_1_f = (mix_2_1_func_type*)sum2_float;
392 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_float(s);
393 }else if(s->midbuf.fmt == AV_SAMPLE_FMT_DBLP){
394 s->native_matrix = av_calloc(nb_in * nb_out, sizeof(double));
395 s->native_one = av_mallocz(sizeof(double));
396 if (!s->native_matrix || !s->native_one)
397 return AVERROR(ENOMEM);
398 for (i = 0; i < nb_out; i++)
399 for (j = 0; j < nb_in; j++)
400 ((double*)s->native_matrix)[i * nb_in + j] = s->matrix[i][j];
401 *((double*)s->native_one) = 1.0;
402 s->mix_1_1_f = (mix_1_1_func_type*)copy_double;
403 s->mix_2_1_f = (mix_2_1_func_type*)sum2_double;
404 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_double(s);
405 }else if(s->midbuf.fmt == AV_SAMPLE_FMT_S32P){
406 // Only for dithering currently
407 // s->native_matrix = av_calloc(nb_in * nb_out, sizeof(double));
408 s->native_one = av_mallocz(sizeof(int));
410 return AVERROR(ENOMEM);
411 // for (i = 0; i < nb_out; i++)
412 // for (j = 0; j < nb_in; j++)
413 // ((double*)s->native_matrix)[i * nb_in + j] = s->matrix[i][j];
414 *((int*)s->native_one) = 32768;
415 s->mix_1_1_f = (mix_1_1_func_type*)copy_s32;
416 s->mix_2_1_f = (mix_2_1_func_type*)sum2_s32;
417 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_s32(s);
420 //FIXME quantize for integeres
421 for (i = 0; i < SWR_CH_MAX; i++) {
423 for (j = 0; j < SWR_CH_MAX; j++) {
424 s->matrix32[i][j]= lrintf(s->matrix[i][j] * 32768);
426 s->matrix_ch[i][++ch_in]= j;
428 s->matrix_ch[i][0]= ch_in;
431 if(HAVE_YASM && HAVE_MMX)
432 return swri_rematrix_init_x86(s);
437 av_cold void swri_rematrix_free(SwrContext *s){
438 av_freep(&s->native_matrix);
439 av_freep(&s->native_one);
440 av_freep(&s->native_simd_matrix);
441 av_freep(&s->native_simd_one);
444 int swri_rematrix(SwrContext *s, AudioData *out, AudioData *in, int len, int mustcopy){
445 int out_i, in_i, i, j;
450 s->mix_any_f(out->ch, (const uint8_t **)in->ch, s->native_matrix, len);
454 if(s->mix_2_1_simd || s->mix_1_1_simd){
456 off = len1 * out->bps;
459 av_assert0(!s->out_ch_layout || out->ch_count == av_get_channel_layout_nb_channels(s->out_ch_layout));
460 av_assert0(!s-> in_ch_layout || in ->ch_count == av_get_channel_layout_nb_channels(s-> in_ch_layout));
462 for(out_i=0; out_i<out->ch_count; out_i++){
463 switch(s->matrix_ch[out_i][0]){
466 memset(out->ch[out_i], 0, len * av_get_bytes_per_sample(s->int_sample_fmt));
469 in_i= s->matrix_ch[out_i][1];
470 if(s->matrix[out_i][in_i]!=1.0){
471 if(s->mix_1_1_simd && len1)
472 s->mix_1_1_simd(out->ch[out_i] , in->ch[in_i] , s->native_simd_matrix, in->ch_count*out_i + in_i, len1);
474 s->mix_1_1_f (out->ch[out_i]+off, in->ch[in_i]+off, s->native_matrix, in->ch_count*out_i + in_i, len-len1);
476 memcpy(out->ch[out_i], in->ch[in_i], len*out->bps);
478 out->ch[out_i]= in->ch[in_i];
482 int in_i1 = s->matrix_ch[out_i][1];
483 int in_i2 = s->matrix_ch[out_i][2];
484 if(s->mix_2_1_simd && len1)
485 s->mix_2_1_simd(out->ch[out_i] , in->ch[in_i1] , in->ch[in_i2] , s->native_simd_matrix, in->ch_count*out_i + in_i1, in->ch_count*out_i + in_i2, len1);
487 s->mix_2_1_f (out->ch[out_i] , in->ch[in_i1] , in->ch[in_i2] , s->native_matrix, in->ch_count*out_i + in_i1, in->ch_count*out_i + in_i2, len1);
489 s->mix_2_1_f (out->ch[out_i]+off, in->ch[in_i1]+off, in->ch[in_i2]+off, s->native_matrix, in->ch_count*out_i + in_i1, in->ch_count*out_i + in_i2, len-len1);
492 if(s->int_sample_fmt == AV_SAMPLE_FMT_FLTP){
493 for(i=0; i<len; i++){
495 for(j=0; j<s->matrix_ch[out_i][0]; j++){
496 in_i= s->matrix_ch[out_i][1+j];
497 v+= ((float*)in->ch[in_i])[i] * s->matrix[out_i][in_i];
499 ((float*)out->ch[out_i])[i]= v;
501 }else if(s->int_sample_fmt == AV_SAMPLE_FMT_DBLP){
502 for(i=0; i<len; i++){
504 for(j=0; j<s->matrix_ch[out_i][0]; j++){
505 in_i= s->matrix_ch[out_i][1+j];
506 v+= ((double*)in->ch[in_i])[i] * s->matrix[out_i][in_i];
508 ((double*)out->ch[out_i])[i]= v;
511 for(i=0; i<len; i++){
513 for(j=0; j<s->matrix_ch[out_i][0]; j++){
514 in_i= s->matrix_ch[out_i][1+j];
515 v+= ((int16_t*)in->ch[in_i])[i] * s->matrix32[out_i][in_i];
517 ((int16_t*)out->ch[out_i])[i]= (v + 16384)>>15;