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"
30 #define RENAME(x) x ## _float
31 #include "rematrix_template.c"
44 #define RENAME(x) x ## _double
45 #include "rematrix_template.c"
54 #define R(x) (((x) + 16384)>>15)
55 #define SAMPLE int16_t
58 #define RENAME(x) x ## _s16
59 #include "rematrix_template.c"
64 #define FRONT_CENTER 2
65 #define LOW_FREQUENCY 3
68 #define FRONT_LEFT_OF_CENTER 6
69 #define FRONT_RIGHT_OF_CENTER 7
74 #define TOP_FRONT_LEFT 12
75 #define TOP_FRONT_CENTER 13
76 #define TOP_FRONT_RIGHT 14
77 #define TOP_BACK_LEFT 15
78 #define TOP_BACK_CENTER 16
79 #define TOP_BACK_RIGHT 17
81 int swr_set_matrix(struct SwrContext *s, const double *matrix, int stride)
83 int nb_in, nb_out, in, out;
85 if (!s || s->in_convert) // s needs to be allocated but not initialized
86 return AVERROR(EINVAL);
87 memset(s->matrix, 0, sizeof(s->matrix));
88 nb_in = av_get_channel_layout_nb_channels(s->in_ch_layout);
89 nb_out = av_get_channel_layout_nb_channels(s->out_ch_layout);
90 for (out = 0; out < nb_out; out++) {
91 for (in = 0; in < nb_in; in++)
92 s->matrix[out][in] = matrix[in];
95 s->rematrix_custom = 1;
99 static int even(int64_t layout){
100 if(!layout) return 1;
101 if(layout&(layout-1)) return 1;
105 static int clean_layout(SwrContext *s, int64_t layout){
106 if((layout & AV_CH_LAYOUT_STEREO_DOWNMIX) == AV_CH_LAYOUT_STEREO_DOWNMIX)
107 return AV_CH_LAYOUT_STEREO;
109 if(layout && layout != AV_CH_FRONT_CENTER && !(layout&(layout-1))) {
111 av_get_channel_layout_string(buf, sizeof(buf), -1, layout);
112 av_log(s, AV_LOG_VERBOSE, "Treating %s as mono\n", buf);
113 return AV_CH_FRONT_CENTER;
119 static int sane_layout(int64_t layout){
120 if(!(layout & AV_CH_LAYOUT_SURROUND)) // at least 1 front speaker
122 if(!even(layout & (AV_CH_FRONT_LEFT | AV_CH_FRONT_RIGHT))) // no asymetric front
124 if(!even(layout & (AV_CH_SIDE_LEFT | AV_CH_SIDE_RIGHT))) // no asymetric side
126 if(!even(layout & (AV_CH_BACK_LEFT | AV_CH_BACK_RIGHT)))
128 if(!even(layout & (AV_CH_FRONT_LEFT_OF_CENTER | AV_CH_FRONT_RIGHT_OF_CENTER)))
130 if(av_get_channel_layout_nb_channels(layout) >= SWR_CH_MAX)
136 av_cold static int auto_matrix(SwrContext *s)
139 double matrix[64][64]={{0}};
140 int64_t unaccounted, in_ch_layout, out_ch_layout;
143 const int matrix_encoding = s->matrix_encoding;
145 in_ch_layout = clean_layout(s, s->in_ch_layout);
146 if(!sane_layout(in_ch_layout)){
147 av_get_channel_layout_string(buf, sizeof(buf), -1, s->in_ch_layout);
148 av_log(s, AV_LOG_ERROR, "Input channel layout '%s' is not supported\n", buf);
149 return AVERROR(EINVAL);
152 out_ch_layout = clean_layout(s, s->out_ch_layout);
153 if(!sane_layout(out_ch_layout)){
154 av_get_channel_layout_string(buf, sizeof(buf), -1, s->out_ch_layout);
155 av_log(s, AV_LOG_ERROR, "Output channel layout '%s' is not supported\n", buf);
156 return AVERROR(EINVAL);
159 memset(s->matrix, 0, sizeof(s->matrix));
161 if(in_ch_layout & out_ch_layout & (1ULL<<i))
165 unaccounted= in_ch_layout & ~out_ch_layout;
167 //FIXME implement dolby surround
168 //FIXME implement full ac3
171 if(unaccounted & AV_CH_FRONT_CENTER){
172 if((out_ch_layout & AV_CH_LAYOUT_STEREO) == AV_CH_LAYOUT_STEREO){
173 if(in_ch_layout & AV_CH_LAYOUT_STEREO) {
174 matrix[ FRONT_LEFT][FRONT_CENTER]+= s->clev;
175 matrix[FRONT_RIGHT][FRONT_CENTER]+= s->clev;
177 matrix[ FRONT_LEFT][FRONT_CENTER]+= M_SQRT1_2;
178 matrix[FRONT_RIGHT][FRONT_CENTER]+= M_SQRT1_2;
183 if(unaccounted & AV_CH_LAYOUT_STEREO){
184 if(out_ch_layout & AV_CH_FRONT_CENTER){
185 matrix[FRONT_CENTER][ FRONT_LEFT]+= M_SQRT1_2;
186 matrix[FRONT_CENTER][FRONT_RIGHT]+= M_SQRT1_2;
187 if(in_ch_layout & AV_CH_FRONT_CENTER)
188 matrix[FRONT_CENTER][ FRONT_CENTER] = s->clev*sqrt(2);
193 if(unaccounted & AV_CH_BACK_CENTER){
194 if(out_ch_layout & AV_CH_BACK_LEFT){
195 matrix[ BACK_LEFT][BACK_CENTER]+= M_SQRT1_2;
196 matrix[BACK_RIGHT][BACK_CENTER]+= M_SQRT1_2;
197 }else if(out_ch_layout & AV_CH_SIDE_LEFT){
198 matrix[ SIDE_LEFT][BACK_CENTER]+= M_SQRT1_2;
199 matrix[SIDE_RIGHT][BACK_CENTER]+= M_SQRT1_2;
200 }else if(out_ch_layout & AV_CH_FRONT_LEFT){
201 if (matrix_encoding == AV_MATRIX_ENCODING_DOLBY ||
202 matrix_encoding == AV_MATRIX_ENCODING_DPLII) {
203 if (unaccounted & (AV_CH_BACK_LEFT | AV_CH_SIDE_LEFT)) {
204 matrix[FRONT_LEFT ][BACK_CENTER] -= s->slev * M_SQRT1_2;
205 matrix[FRONT_RIGHT][BACK_CENTER] += s->slev * M_SQRT1_2;
207 matrix[FRONT_LEFT ][BACK_CENTER] -= s->slev;
208 matrix[FRONT_RIGHT][BACK_CENTER] += s->slev;
211 matrix[ FRONT_LEFT][BACK_CENTER]+= s->slev*M_SQRT1_2;
212 matrix[FRONT_RIGHT][BACK_CENTER]+= s->slev*M_SQRT1_2;
214 }else if(out_ch_layout & AV_CH_FRONT_CENTER){
215 matrix[ FRONT_CENTER][BACK_CENTER]+= s->slev*M_SQRT1_2;
219 if(unaccounted & AV_CH_BACK_LEFT){
220 if(out_ch_layout & AV_CH_BACK_CENTER){
221 matrix[BACK_CENTER][ BACK_LEFT]+= M_SQRT1_2;
222 matrix[BACK_CENTER][BACK_RIGHT]+= M_SQRT1_2;
223 }else if(out_ch_layout & AV_CH_SIDE_LEFT){
224 if(in_ch_layout & AV_CH_SIDE_LEFT){
225 matrix[ SIDE_LEFT][ BACK_LEFT]+= M_SQRT1_2;
226 matrix[SIDE_RIGHT][BACK_RIGHT]+= M_SQRT1_2;
228 matrix[ SIDE_LEFT][ BACK_LEFT]+= 1.0;
229 matrix[SIDE_RIGHT][BACK_RIGHT]+= 1.0;
231 }else if(out_ch_layout & AV_CH_FRONT_LEFT){
232 if (matrix_encoding == AV_MATRIX_ENCODING_DOLBY) {
233 matrix[FRONT_LEFT ][BACK_LEFT ] -= s->slev * M_SQRT1_2;
234 matrix[FRONT_LEFT ][BACK_RIGHT] -= s->slev * M_SQRT1_2;
235 matrix[FRONT_RIGHT][BACK_LEFT ] += s->slev * M_SQRT1_2;
236 matrix[FRONT_RIGHT][BACK_RIGHT] += s->slev * M_SQRT1_2;
237 } else if (matrix_encoding == AV_MATRIX_ENCODING_DPLII) {
238 matrix[FRONT_LEFT ][BACK_LEFT ] -= s->slev * SQRT3_2;
239 matrix[FRONT_LEFT ][BACK_RIGHT] -= s->slev * M_SQRT1_2;
240 matrix[FRONT_RIGHT][BACK_LEFT ] += s->slev * M_SQRT1_2;
241 matrix[FRONT_RIGHT][BACK_RIGHT] += s->slev * SQRT3_2;
243 matrix[ FRONT_LEFT][ BACK_LEFT] += s->slev;
244 matrix[FRONT_RIGHT][BACK_RIGHT] += s->slev;
246 }else if(out_ch_layout & AV_CH_FRONT_CENTER){
247 matrix[ FRONT_CENTER][BACK_LEFT ]+= s->slev*M_SQRT1_2;
248 matrix[ FRONT_CENTER][BACK_RIGHT]+= s->slev*M_SQRT1_2;
253 if(unaccounted & AV_CH_SIDE_LEFT){
254 if(out_ch_layout & AV_CH_BACK_LEFT){
255 /* if back channels do not exist in the input, just copy side
256 channels to back channels, otherwise mix side into back */
257 if (in_ch_layout & AV_CH_BACK_LEFT) {
258 matrix[BACK_LEFT ][SIDE_LEFT ] += M_SQRT1_2;
259 matrix[BACK_RIGHT][SIDE_RIGHT] += M_SQRT1_2;
261 matrix[BACK_LEFT ][SIDE_LEFT ] += 1.0;
262 matrix[BACK_RIGHT][SIDE_RIGHT] += 1.0;
264 }else if(out_ch_layout & AV_CH_BACK_CENTER){
265 matrix[BACK_CENTER][ SIDE_LEFT]+= M_SQRT1_2;
266 matrix[BACK_CENTER][SIDE_RIGHT]+= M_SQRT1_2;
267 }else if(out_ch_layout & AV_CH_FRONT_LEFT){
268 if (matrix_encoding == AV_MATRIX_ENCODING_DOLBY) {
269 matrix[FRONT_LEFT ][SIDE_LEFT ] -= s->slev * M_SQRT1_2;
270 matrix[FRONT_LEFT ][SIDE_RIGHT] -= s->slev * M_SQRT1_2;
271 matrix[FRONT_RIGHT][SIDE_LEFT ] += s->slev * M_SQRT1_2;
272 matrix[FRONT_RIGHT][SIDE_RIGHT] += s->slev * M_SQRT1_2;
273 } else if (matrix_encoding == AV_MATRIX_ENCODING_DPLII) {
274 matrix[FRONT_LEFT ][SIDE_LEFT ] -= s->slev * SQRT3_2;
275 matrix[FRONT_LEFT ][SIDE_RIGHT] -= s->slev * M_SQRT1_2;
276 matrix[FRONT_RIGHT][SIDE_LEFT ] += s->slev * M_SQRT1_2;
277 matrix[FRONT_RIGHT][SIDE_RIGHT] += s->slev * SQRT3_2;
279 matrix[ FRONT_LEFT][ SIDE_LEFT] += s->slev;
280 matrix[FRONT_RIGHT][SIDE_RIGHT] += s->slev;
282 }else if(out_ch_layout & AV_CH_FRONT_CENTER){
283 matrix[ FRONT_CENTER][SIDE_LEFT ]+= s->slev*M_SQRT1_2;
284 matrix[ FRONT_CENTER][SIDE_RIGHT]+= s->slev*M_SQRT1_2;
289 if(unaccounted & AV_CH_FRONT_LEFT_OF_CENTER){
290 if(out_ch_layout & AV_CH_FRONT_LEFT){
291 matrix[ FRONT_LEFT][ FRONT_LEFT_OF_CENTER]+= 1.0;
292 matrix[FRONT_RIGHT][FRONT_RIGHT_OF_CENTER]+= 1.0;
293 }else if(out_ch_layout & AV_CH_FRONT_CENTER){
294 matrix[ FRONT_CENTER][ FRONT_LEFT_OF_CENTER]+= M_SQRT1_2;
295 matrix[ FRONT_CENTER][FRONT_RIGHT_OF_CENTER]+= M_SQRT1_2;
299 /* mix LFE into front left/right or center */
300 if (unaccounted & AV_CH_LOW_FREQUENCY) {
301 if (out_ch_layout & AV_CH_FRONT_CENTER) {
302 matrix[FRONT_CENTER][LOW_FREQUENCY] += s->lfe_mix_level;
303 } else if (out_ch_layout & AV_CH_FRONT_LEFT) {
304 matrix[FRONT_LEFT ][LOW_FREQUENCY] += s->lfe_mix_level * M_SQRT1_2;
305 matrix[FRONT_RIGHT][LOW_FREQUENCY] += s->lfe_mix_level * M_SQRT1_2;
310 for(out_i=i=0; i<64; i++){
314 s->matrix[out_i][in_i]= matrix[i][j];
316 sum += fabs(matrix[i][j]);
318 if(in_ch_layout & (1ULL<<j))
321 maxcoef= FFMAX(maxcoef, sum);
322 if(out_ch_layout & (1ULL<<i))
325 if(s->rematrix_volume < 0)
326 maxcoef = -s->rematrix_volume;
328 if(( av_get_packed_sample_fmt(s->out_sample_fmt) < AV_SAMPLE_FMT_FLT
329 || av_get_packed_sample_fmt(s->int_sample_fmt) < AV_SAMPLE_FMT_FLT) && maxcoef > 1.0){
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 for(i=0; i<av_get_channel_layout_nb_channels(out_ch_layout); i++){
344 for(j=0; j<av_get_channel_layout_nb_channels(in_ch_layout); j++){
345 av_log(NULL, AV_LOG_DEBUG, "%f ", s->matrix[i][j]);
347 av_log(NULL, AV_LOG_DEBUG, "\n");
352 av_cold int swri_rematrix_init(SwrContext *s){
354 int nb_in = av_get_channel_layout_nb_channels(s->in_ch_layout);
355 int nb_out = av_get_channel_layout_nb_channels(s->out_ch_layout);
359 if (!s->rematrix_custom) {
360 int r = auto_matrix(s);
364 if (s->midbuf.fmt == AV_SAMPLE_FMT_S16P){
365 s->native_matrix = av_mallocz(nb_in * nb_out * sizeof(int));
366 s->native_one = av_mallocz(sizeof(int));
367 for (i = 0; i < nb_out; i++)
368 for (j = 0; j < nb_in; j++)
369 ((int*)s->native_matrix)[i * nb_in + j] = lrintf(s->matrix[i][j] * 32768);
370 *((int*)s->native_one) = 32768;
371 s->mix_1_1_f = (mix_1_1_func_type*)copy_s16;
372 s->mix_2_1_f = (mix_2_1_func_type*)sum2_s16;
373 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_s16(s);
374 }else if(s->midbuf.fmt == AV_SAMPLE_FMT_FLTP){
375 s->native_matrix = av_mallocz(nb_in * nb_out * sizeof(float));
376 s->native_one = av_mallocz(sizeof(float));
377 for (i = 0; i < nb_out; i++)
378 for (j = 0; j < nb_in; j++)
379 ((float*)s->native_matrix)[i * nb_in + j] = s->matrix[i][j];
380 *((float*)s->native_one) = 1.0;
381 s->mix_1_1_f = (mix_1_1_func_type*)copy_float;
382 s->mix_2_1_f = (mix_2_1_func_type*)sum2_float;
383 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_float(s);
384 }else if(s->midbuf.fmt == AV_SAMPLE_FMT_DBLP){
385 s->native_matrix = av_mallocz(nb_in * nb_out * sizeof(double));
386 s->native_one = av_mallocz(sizeof(double));
387 for (i = 0; i < nb_out; i++)
388 for (j = 0; j < nb_in; j++)
389 ((double*)s->native_matrix)[i * nb_in + j] = s->matrix[i][j];
390 *((double*)s->native_one) = 1.0;
391 s->mix_1_1_f = (mix_1_1_func_type*)copy_double;
392 s->mix_2_1_f = (mix_2_1_func_type*)sum2_double;
393 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_double(s);
396 //FIXME quantize for integeres
397 for (i = 0; i < SWR_CH_MAX; i++) {
399 for (j = 0; j < SWR_CH_MAX; j++) {
400 s->matrix32[i][j]= lrintf(s->matrix[i][j] * 32768);
402 s->matrix_ch[i][++ch_in]= j;
404 s->matrix_ch[i][0]= ch_in;
407 if(HAVE_YASM && HAVE_MMX) swri_rematrix_init_x86(s);
412 av_cold void swri_rematrix_free(SwrContext *s){
413 av_freep(&s->native_matrix);
414 av_freep(&s->native_one);
415 av_freep(&s->native_simd_matrix);
418 int swri_rematrix(SwrContext *s, AudioData *out, AudioData *in, int len, int mustcopy){
419 int out_i, in_i, i, j;
424 s->mix_any_f(out->ch, (const uint8_t **)in->ch, s->native_matrix, len);
428 if(s->mix_2_1_simd || s->mix_1_1_simd){
430 off = len1 * out->bps;
433 av_assert0(out->ch_count == av_get_channel_layout_nb_channels(s->out_ch_layout));
434 av_assert0(in ->ch_count == av_get_channel_layout_nb_channels(s-> in_ch_layout));
436 for(out_i=0; out_i<out->ch_count; out_i++){
437 switch(s->matrix_ch[out_i][0]){
440 memset(out->ch[out_i], 0, len * av_get_bytes_per_sample(s->int_sample_fmt));
443 in_i= s->matrix_ch[out_i][1];
444 if(s->matrix[out_i][in_i]!=1.0){
445 if(s->mix_1_1_simd && len1)
446 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);
448 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);
450 memcpy(out->ch[out_i], in->ch[in_i], len*out->bps);
452 out->ch[out_i]= in->ch[in_i];
456 int in_i1 = s->matrix_ch[out_i][1];
457 int in_i2 = s->matrix_ch[out_i][2];
458 if(s->mix_2_1_simd && len1)
459 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);
461 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);
463 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);
466 if(s->int_sample_fmt == AV_SAMPLE_FMT_FLTP){
467 for(i=0; i<len; i++){
469 for(j=0; j<s->matrix_ch[out_i][0]; j++){
470 in_i= s->matrix_ch[out_i][1+j];
471 v+= ((float*)in->ch[in_i])[i] * s->matrix[out_i][in_i];
473 ((float*)out->ch[out_i])[i]= v;
475 }else if(s->int_sample_fmt == AV_SAMPLE_FMT_DBLP){
476 for(i=0; i<len; i++){
478 for(j=0; j<s->matrix_ch[out_i][0]; j++){
479 in_i= s->matrix_ch[out_i][1+j];
480 v+= ((double*)in->ch[in_i])[i] * s->matrix[out_i][in_i];
482 ((double*)out->ch[out_i])[i]= v;
485 for(i=0; i<len; i++){
487 for(j=0; j<s->matrix_ch[out_i][0]; j++){
488 in_i= s->matrix_ch[out_i][1+j];
489 v+= ((int16_t*)in->ch[in_i])[i] * s->matrix32[out_i][in_i];
491 ((int16_t*)out->ch[out_i])[i]= (v + 16384)>>15;