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
60 int swr_set_matrix(struct SwrContext *s, const double *matrix, int stride)
62 int nb_in, nb_out, in, out;
64 if (!s || s->in_convert) // s needs to be allocated but not initialized
65 return AVERROR(EINVAL);
66 memset(s->matrix, 0, sizeof(s->matrix));
67 nb_in = av_get_channel_layout_nb_channels(s->in_ch_layout);
68 nb_out = av_get_channel_layout_nb_channels(s->out_ch_layout);
69 for (out = 0; out < nb_out; out++) {
70 for (in = 0; in < nb_in; in++)
71 s->matrix[out][in] = matrix[in];
74 s->rematrix_custom = 1;
78 static int even(int64_t layout){
80 if(layout&(layout-1)) return 1;
84 static int clean_layout(SwrContext *s, int64_t layout){
85 if((layout & AV_CH_LAYOUT_STEREO_DOWNMIX) == AV_CH_LAYOUT_STEREO_DOWNMIX)
86 return AV_CH_LAYOUT_STEREO;
88 if(layout && layout != AV_CH_FRONT_CENTER && !(layout&(layout-1))) {
90 av_get_channel_layout_string(buf, sizeof(buf), -1, layout);
91 av_log(s, AV_LOG_VERBOSE, "Treating %s as mono\n", buf);
92 return AV_CH_FRONT_CENTER;
98 static int sane_layout(int64_t layout){
99 if(!(layout & AV_CH_LAYOUT_SURROUND)) // at least 1 front speaker
101 if(!even(layout & (AV_CH_FRONT_LEFT | AV_CH_FRONT_RIGHT))) // no asymetric front
103 if(!even(layout & (AV_CH_SIDE_LEFT | AV_CH_SIDE_RIGHT))) // no asymetric side
105 if(!even(layout & (AV_CH_BACK_LEFT | AV_CH_BACK_RIGHT)))
107 if(!even(layout & (AV_CH_FRONT_LEFT_OF_CENTER | AV_CH_FRONT_RIGHT_OF_CENTER)))
109 if(av_get_channel_layout_nb_channels(layout) >= SWR_CH_MAX)
115 av_cold static int auto_matrix(SwrContext *s)
118 double matrix[64][64]={{0}};
119 int64_t unaccounted, in_ch_layout, out_ch_layout;
122 const int matrix_encoding = s->matrix_encoding;
124 in_ch_layout = clean_layout(s, s->in_ch_layout);
125 if(!sane_layout(in_ch_layout)){
126 av_get_channel_layout_string(buf, sizeof(buf), -1, s->in_ch_layout);
127 av_log(s, AV_LOG_ERROR, "Input channel layout '%s' is not supported\n", buf);
128 return AVERROR(EINVAL);
131 out_ch_layout = clean_layout(s, s->out_ch_layout);
132 if(!sane_layout(out_ch_layout)){
133 av_get_channel_layout_string(buf, sizeof(buf), -1, s->out_ch_layout);
134 av_log(s, AV_LOG_ERROR, "Output channel layout '%s' is not supported\n", buf);
135 return AVERROR(EINVAL);
138 memset(s->matrix, 0, sizeof(s->matrix));
140 if(in_ch_layout & out_ch_layout & (1ULL<<i))
144 unaccounted= in_ch_layout & ~out_ch_layout;
146 //FIXME implement dolby surround
147 //FIXME implement full ac3
150 if(unaccounted & AV_CH_FRONT_CENTER){
151 if((out_ch_layout & AV_CH_LAYOUT_STEREO) == AV_CH_LAYOUT_STEREO){
152 if(in_ch_layout & AV_CH_LAYOUT_STEREO) {
153 matrix[ FRONT_LEFT][FRONT_CENTER]+= s->clev;
154 matrix[FRONT_RIGHT][FRONT_CENTER]+= s->clev;
156 matrix[ FRONT_LEFT][FRONT_CENTER]+= M_SQRT1_2;
157 matrix[FRONT_RIGHT][FRONT_CENTER]+= M_SQRT1_2;
162 if(unaccounted & AV_CH_LAYOUT_STEREO){
163 if(out_ch_layout & AV_CH_FRONT_CENTER){
164 matrix[FRONT_CENTER][ FRONT_LEFT]+= M_SQRT1_2;
165 matrix[FRONT_CENTER][FRONT_RIGHT]+= M_SQRT1_2;
166 if(in_ch_layout & AV_CH_FRONT_CENTER)
167 matrix[FRONT_CENTER][ FRONT_CENTER] = s->clev*sqrt(2);
172 if(unaccounted & AV_CH_BACK_CENTER){
173 if(out_ch_layout & AV_CH_BACK_LEFT){
174 matrix[ BACK_LEFT][BACK_CENTER]+= M_SQRT1_2;
175 matrix[BACK_RIGHT][BACK_CENTER]+= M_SQRT1_2;
176 }else if(out_ch_layout & AV_CH_SIDE_LEFT){
177 matrix[ SIDE_LEFT][BACK_CENTER]+= M_SQRT1_2;
178 matrix[SIDE_RIGHT][BACK_CENTER]+= M_SQRT1_2;
179 }else if(out_ch_layout & AV_CH_FRONT_LEFT){
180 if (matrix_encoding == AV_MATRIX_ENCODING_DOLBY ||
181 matrix_encoding == AV_MATRIX_ENCODING_DPLII) {
182 if (unaccounted & (AV_CH_BACK_LEFT | AV_CH_SIDE_LEFT)) {
183 matrix[FRONT_LEFT ][BACK_CENTER] -= s->slev * M_SQRT1_2;
184 matrix[FRONT_RIGHT][BACK_CENTER] += s->slev * M_SQRT1_2;
186 matrix[FRONT_LEFT ][BACK_CENTER] -= s->slev;
187 matrix[FRONT_RIGHT][BACK_CENTER] += s->slev;
190 matrix[ FRONT_LEFT][BACK_CENTER]+= s->slev*M_SQRT1_2;
191 matrix[FRONT_RIGHT][BACK_CENTER]+= s->slev*M_SQRT1_2;
193 }else if(out_ch_layout & AV_CH_FRONT_CENTER){
194 matrix[ FRONT_CENTER][BACK_CENTER]+= s->slev*M_SQRT1_2;
198 if(unaccounted & AV_CH_BACK_LEFT){
199 if(out_ch_layout & AV_CH_BACK_CENTER){
200 matrix[BACK_CENTER][ BACK_LEFT]+= M_SQRT1_2;
201 matrix[BACK_CENTER][BACK_RIGHT]+= M_SQRT1_2;
202 }else if(out_ch_layout & AV_CH_SIDE_LEFT){
203 if(in_ch_layout & AV_CH_SIDE_LEFT){
204 matrix[ SIDE_LEFT][ BACK_LEFT]+= M_SQRT1_2;
205 matrix[SIDE_RIGHT][BACK_RIGHT]+= M_SQRT1_2;
207 matrix[ SIDE_LEFT][ BACK_LEFT]+= 1.0;
208 matrix[SIDE_RIGHT][BACK_RIGHT]+= 1.0;
210 }else if(out_ch_layout & AV_CH_FRONT_LEFT){
211 if (matrix_encoding == AV_MATRIX_ENCODING_DOLBY) {
212 matrix[FRONT_LEFT ][BACK_LEFT ] -= s->slev * M_SQRT1_2;
213 matrix[FRONT_LEFT ][BACK_RIGHT] -= s->slev * M_SQRT1_2;
214 matrix[FRONT_RIGHT][BACK_LEFT ] += s->slev * M_SQRT1_2;
215 matrix[FRONT_RIGHT][BACK_RIGHT] += s->slev * M_SQRT1_2;
216 } else if (matrix_encoding == AV_MATRIX_ENCODING_DPLII) {
217 matrix[FRONT_LEFT ][BACK_LEFT ] -= s->slev * SQRT3_2;
218 matrix[FRONT_LEFT ][BACK_RIGHT] -= s->slev * M_SQRT1_2;
219 matrix[FRONT_RIGHT][BACK_LEFT ] += s->slev * M_SQRT1_2;
220 matrix[FRONT_RIGHT][BACK_RIGHT] += s->slev * SQRT3_2;
222 matrix[ FRONT_LEFT][ BACK_LEFT] += s->slev;
223 matrix[FRONT_RIGHT][BACK_RIGHT] += s->slev;
225 }else if(out_ch_layout & AV_CH_FRONT_CENTER){
226 matrix[ FRONT_CENTER][BACK_LEFT ]+= s->slev*M_SQRT1_2;
227 matrix[ FRONT_CENTER][BACK_RIGHT]+= s->slev*M_SQRT1_2;
232 if(unaccounted & AV_CH_SIDE_LEFT){
233 if(out_ch_layout & AV_CH_BACK_LEFT){
234 /* if back channels do not exist in the input, just copy side
235 channels to back channels, otherwise mix side into back */
236 if (in_ch_layout & AV_CH_BACK_LEFT) {
237 matrix[BACK_LEFT ][SIDE_LEFT ] += M_SQRT1_2;
238 matrix[BACK_RIGHT][SIDE_RIGHT] += M_SQRT1_2;
240 matrix[BACK_LEFT ][SIDE_LEFT ] += 1.0;
241 matrix[BACK_RIGHT][SIDE_RIGHT] += 1.0;
243 }else if(out_ch_layout & AV_CH_BACK_CENTER){
244 matrix[BACK_CENTER][ SIDE_LEFT]+= M_SQRT1_2;
245 matrix[BACK_CENTER][SIDE_RIGHT]+= M_SQRT1_2;
246 }else if(out_ch_layout & AV_CH_FRONT_LEFT){
247 if (matrix_encoding == AV_MATRIX_ENCODING_DOLBY) {
248 matrix[FRONT_LEFT ][SIDE_LEFT ] -= s->slev * M_SQRT1_2;
249 matrix[FRONT_LEFT ][SIDE_RIGHT] -= s->slev * M_SQRT1_2;
250 matrix[FRONT_RIGHT][SIDE_LEFT ] += s->slev * M_SQRT1_2;
251 matrix[FRONT_RIGHT][SIDE_RIGHT] += s->slev * M_SQRT1_2;
252 } else if (matrix_encoding == AV_MATRIX_ENCODING_DPLII) {
253 matrix[FRONT_LEFT ][SIDE_LEFT ] -= s->slev * SQRT3_2;
254 matrix[FRONT_LEFT ][SIDE_RIGHT] -= s->slev * M_SQRT1_2;
255 matrix[FRONT_RIGHT][SIDE_LEFT ] += s->slev * M_SQRT1_2;
256 matrix[FRONT_RIGHT][SIDE_RIGHT] += s->slev * SQRT3_2;
258 matrix[ FRONT_LEFT][ SIDE_LEFT] += s->slev;
259 matrix[FRONT_RIGHT][SIDE_RIGHT] += s->slev;
261 }else if(out_ch_layout & AV_CH_FRONT_CENTER){
262 matrix[ FRONT_CENTER][SIDE_LEFT ]+= s->slev*M_SQRT1_2;
263 matrix[ FRONT_CENTER][SIDE_RIGHT]+= s->slev*M_SQRT1_2;
268 if(unaccounted & AV_CH_FRONT_LEFT_OF_CENTER){
269 if(out_ch_layout & AV_CH_FRONT_LEFT){
270 matrix[ FRONT_LEFT][ FRONT_LEFT_OF_CENTER]+= 1.0;
271 matrix[FRONT_RIGHT][FRONT_RIGHT_OF_CENTER]+= 1.0;
272 }else if(out_ch_layout & AV_CH_FRONT_CENTER){
273 matrix[ FRONT_CENTER][ FRONT_LEFT_OF_CENTER]+= M_SQRT1_2;
274 matrix[ FRONT_CENTER][FRONT_RIGHT_OF_CENTER]+= M_SQRT1_2;
278 /* mix LFE into front left/right or center */
279 if (unaccounted & AV_CH_LOW_FREQUENCY) {
280 if (out_ch_layout & AV_CH_FRONT_CENTER) {
281 matrix[FRONT_CENTER][LOW_FREQUENCY] += s->lfe_mix_level;
282 } else if (out_ch_layout & AV_CH_FRONT_LEFT) {
283 matrix[FRONT_LEFT ][LOW_FREQUENCY] += s->lfe_mix_level * M_SQRT1_2;
284 matrix[FRONT_RIGHT][LOW_FREQUENCY] += s->lfe_mix_level * M_SQRT1_2;
289 for(out_i=i=0; i<64; i++){
293 s->matrix[out_i][in_i]= matrix[i][j];
295 sum += fabs(matrix[i][j]);
297 if(in_ch_layout & (1ULL<<j))
300 maxcoef= FFMAX(maxcoef, sum);
301 if(out_ch_layout & (1ULL<<i))
304 if(s->rematrix_volume < 0)
305 maxcoef = -s->rematrix_volume;
307 if(( av_get_packed_sample_fmt(s->out_sample_fmt) < AV_SAMPLE_FMT_FLT
308 || av_get_packed_sample_fmt(s->int_sample_fmt) < AV_SAMPLE_FMT_FLT) && maxcoef > 1.0){
309 for(i=0; i<SWR_CH_MAX; i++)
310 for(j=0; j<SWR_CH_MAX; j++){
311 s->matrix[i][j] /= maxcoef;
315 if(s->rematrix_volume > 0){
316 for(i=0; i<SWR_CH_MAX; i++)
317 for(j=0; j<SWR_CH_MAX; j++){
318 s->matrix[i][j] *= s->rematrix_volume;
322 for(i=0; i<av_get_channel_layout_nb_channels(out_ch_layout); i++){
323 for(j=0; j<av_get_channel_layout_nb_channels(in_ch_layout); j++){
324 av_log(NULL, AV_LOG_DEBUG, "%f ", s->matrix[i][j]);
326 av_log(NULL, AV_LOG_DEBUG, "\n");
331 av_cold int swri_rematrix_init(SwrContext *s){
333 int nb_in = av_get_channel_layout_nb_channels(s->in_ch_layout);
334 int nb_out = av_get_channel_layout_nb_channels(s->out_ch_layout);
338 if (!s->rematrix_custom) {
339 int r = auto_matrix(s);
343 if (s->midbuf.fmt == AV_SAMPLE_FMT_S16P){
344 s->native_matrix = av_calloc(nb_in * nb_out, sizeof(int));
345 s->native_one = av_mallocz(sizeof(int));
346 for (i = 0; i < nb_out; i++)
347 for (j = 0; j < nb_in; j++)
348 ((int*)s->native_matrix)[i * nb_in + j] = lrintf(s->matrix[i][j] * 32768);
349 *((int*)s->native_one) = 32768;
350 s->mix_1_1_f = (mix_1_1_func_type*)copy_s16;
351 s->mix_2_1_f = (mix_2_1_func_type*)sum2_s16;
352 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_s16(s);
353 }else if(s->midbuf.fmt == AV_SAMPLE_FMT_FLTP){
354 s->native_matrix = av_calloc(nb_in * nb_out, sizeof(float));
355 s->native_one = av_mallocz(sizeof(float));
356 for (i = 0; i < nb_out; i++)
357 for (j = 0; j < nb_in; j++)
358 ((float*)s->native_matrix)[i * nb_in + j] = s->matrix[i][j];
359 *((float*)s->native_one) = 1.0;
360 s->mix_1_1_f = (mix_1_1_func_type*)copy_float;
361 s->mix_2_1_f = (mix_2_1_func_type*)sum2_float;
362 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_float(s);
363 }else if(s->midbuf.fmt == AV_SAMPLE_FMT_DBLP){
364 s->native_matrix = av_calloc(nb_in * nb_out, sizeof(double));
365 s->native_one = av_mallocz(sizeof(double));
366 for (i = 0; i < nb_out; i++)
367 for (j = 0; j < nb_in; j++)
368 ((double*)s->native_matrix)[i * nb_in + j] = s->matrix[i][j];
369 *((double*)s->native_one) = 1.0;
370 s->mix_1_1_f = (mix_1_1_func_type*)copy_double;
371 s->mix_2_1_f = (mix_2_1_func_type*)sum2_double;
372 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_double(s);
373 }else if(s->midbuf.fmt == AV_SAMPLE_FMT_S32P){
374 // Only for dithering currently
375 // s->native_matrix = av_calloc(nb_in * nb_out, sizeof(double));
376 s->native_one = av_mallocz(sizeof(int));
377 // for (i = 0; i < nb_out; i++)
378 // for (j = 0; j < nb_in; j++)
379 // ((double*)s->native_matrix)[i * nb_in + j] = s->matrix[i][j];
380 *((int*)s->native_one) = 32768;
381 s->mix_1_1_f = (mix_1_1_func_type*)copy_s32;
382 s->mix_2_1_f = (mix_2_1_func_type*)sum2_s32;
383 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_s32(s);
386 //FIXME quantize for integeres
387 for (i = 0; i < SWR_CH_MAX; i++) {
389 for (j = 0; j < SWR_CH_MAX; j++) {
390 s->matrix32[i][j]= lrintf(s->matrix[i][j] * 32768);
392 s->matrix_ch[i][++ch_in]= j;
394 s->matrix_ch[i][0]= ch_in;
397 if(HAVE_YASM && HAVE_MMX) swri_rematrix_init_x86(s);
402 av_cold void swri_rematrix_free(SwrContext *s){
403 av_freep(&s->native_matrix);
404 av_freep(&s->native_one);
405 av_freep(&s->native_simd_matrix);
406 av_freep(&s->native_simd_one);
409 int swri_rematrix(SwrContext *s, AudioData *out, AudioData *in, int len, int mustcopy){
410 int out_i, in_i, i, j;
415 s->mix_any_f(out->ch, (const uint8_t **)in->ch, s->native_matrix, len);
419 if(s->mix_2_1_simd || s->mix_1_1_simd){
421 off = len1 * out->bps;
424 av_assert0(out->ch_count == av_get_channel_layout_nb_channels(s->out_ch_layout));
425 av_assert0(in ->ch_count == av_get_channel_layout_nb_channels(s-> in_ch_layout));
427 for(out_i=0; out_i<out->ch_count; out_i++){
428 switch(s->matrix_ch[out_i][0]){
431 memset(out->ch[out_i], 0, len * av_get_bytes_per_sample(s->int_sample_fmt));
434 in_i= s->matrix_ch[out_i][1];
435 if(s->matrix[out_i][in_i]!=1.0){
436 if(s->mix_1_1_simd && len1)
437 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);
439 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);
441 memcpy(out->ch[out_i], in->ch[in_i], len*out->bps);
443 out->ch[out_i]= in->ch[in_i];
447 int in_i1 = s->matrix_ch[out_i][1];
448 int in_i2 = s->matrix_ch[out_i][2];
449 if(s->mix_2_1_simd && len1)
450 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);
452 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);
454 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);
457 if(s->int_sample_fmt == AV_SAMPLE_FMT_FLTP){
458 for(i=0; i<len; i++){
460 for(j=0; j<s->matrix_ch[out_i][0]; j++){
461 in_i= s->matrix_ch[out_i][1+j];
462 v+= ((float*)in->ch[in_i])[i] * s->matrix[out_i][in_i];
464 ((float*)out->ch[out_i])[i]= v;
466 }else if(s->int_sample_fmt == AV_SAMPLE_FMT_DBLP){
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+= ((double*)in->ch[in_i])[i] * s->matrix[out_i][in_i];
473 ((double*)out->ch[out_i])[i]= v;
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+= ((int16_t*)in->ch[in_i])[i] * s->matrix32[out_i][in_i];
482 ((int16_t*)out->ch[out_i])[i]= (v + 16384)>>15;