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
39 #define FRONT_CENTER 2
40 #define LOW_FREQUENCY 3
43 #define FRONT_LEFT_OF_CENTER 6
44 #define FRONT_RIGHT_OF_CENTER 7
49 #define TOP_FRONT_LEFT 12
50 #define TOP_FRONT_CENTER 13
51 #define TOP_FRONT_RIGHT 14
52 #define TOP_BACK_LEFT 15
53 #define TOP_BACK_CENTER 16
54 #define TOP_BACK_RIGHT 17
56 int swr_set_matrix(struct SwrContext *s, const double *matrix, int stride)
58 int nb_in, nb_out, in, out;
60 if (!s || s->in_convert) // s needs to be allocated but not initialized
61 return AVERROR(EINVAL);
62 memset(s->matrix, 0, sizeof(s->matrix));
63 nb_in = av_get_channel_layout_nb_channels(s->in_ch_layout);
64 nb_out = av_get_channel_layout_nb_channels(s->out_ch_layout);
65 for (out = 0; out < nb_out; out++) {
66 for (in = 0; in < nb_in; in++)
67 s->matrix[out][in] = matrix[in];
70 s->rematrix_custom = 1;
74 static int even(int64_t layout){
76 if(layout&(layout-1)) return 1;
80 static int clean_layout(SwrContext *s, int64_t layout){
81 if((layout & AV_CH_LAYOUT_STEREO_DOWNMIX) == AV_CH_LAYOUT_STEREO_DOWNMIX)
82 return AV_CH_LAYOUT_STEREO;
84 if(layout && layout != AV_CH_FRONT_CENTER && !(layout&(layout-1))) {
86 av_get_channel_layout_string(buf, sizeof(buf), -1, layout);
87 av_log(s, AV_LOG_VERBOSE, "Treating %s as mono\n", buf);
88 return AV_CH_FRONT_CENTER;
94 static int sane_layout(int64_t layout){
95 if(!(layout & AV_CH_LAYOUT_SURROUND)) // at least 1 front speaker
97 if(!even(layout & (AV_CH_FRONT_LEFT | AV_CH_FRONT_RIGHT))) // no asymetric front
99 if(!even(layout & (AV_CH_SIDE_LEFT | AV_CH_SIDE_RIGHT))) // no asymetric side
101 if(!even(layout & (AV_CH_BACK_LEFT | AV_CH_BACK_RIGHT)))
103 if(!even(layout & (AV_CH_FRONT_LEFT_OF_CENTER | AV_CH_FRONT_RIGHT_OF_CENTER)))
105 if(av_get_channel_layout_nb_channels(layout) >= SWR_CH_MAX)
111 av_cold static int auto_matrix(SwrContext *s)
114 double matrix[64][64]={{0}};
115 int64_t unaccounted, in_ch_layout, out_ch_layout;
118 const int matrix_encoding = s->matrix_encoding;
120 in_ch_layout = clean_layout(s, s->in_ch_layout);
121 if(!sane_layout(in_ch_layout)){
122 av_get_channel_layout_string(buf, sizeof(buf), -1, s->in_ch_layout);
123 av_log(s, AV_LOG_ERROR, "Input channel layout '%s' is not supported\n", buf);
124 return AVERROR(EINVAL);
127 out_ch_layout = clean_layout(s, s->out_ch_layout);
128 if(!sane_layout(out_ch_layout)){
129 av_get_channel_layout_string(buf, sizeof(buf), -1, s->out_ch_layout);
130 av_log(s, AV_LOG_ERROR, "Output channel layout '%s' is not supported\n", buf);
131 return AVERROR(EINVAL);
134 memset(s->matrix, 0, sizeof(s->matrix));
136 if(in_ch_layout & out_ch_layout & (1ULL<<i))
140 unaccounted= in_ch_layout & ~out_ch_layout;
142 //FIXME implement dolby surround
143 //FIXME implement full ac3
146 if(unaccounted & AV_CH_FRONT_CENTER){
147 if((out_ch_layout & AV_CH_LAYOUT_STEREO) == AV_CH_LAYOUT_STEREO){
148 if(in_ch_layout & AV_CH_LAYOUT_STEREO) {
149 matrix[ FRONT_LEFT][FRONT_CENTER]+= s->clev;
150 matrix[FRONT_RIGHT][FRONT_CENTER]+= s->clev;
152 matrix[ FRONT_LEFT][FRONT_CENTER]+= M_SQRT1_2;
153 matrix[FRONT_RIGHT][FRONT_CENTER]+= M_SQRT1_2;
158 if(unaccounted & AV_CH_LAYOUT_STEREO){
159 if(out_ch_layout & AV_CH_FRONT_CENTER){
160 matrix[FRONT_CENTER][ FRONT_LEFT]+= M_SQRT1_2;
161 matrix[FRONT_CENTER][FRONT_RIGHT]+= M_SQRT1_2;
162 if(in_ch_layout & AV_CH_FRONT_CENTER)
163 matrix[FRONT_CENTER][ FRONT_CENTER] = s->clev*sqrt(2);
168 if(unaccounted & AV_CH_BACK_CENTER){
169 if(out_ch_layout & AV_CH_BACK_LEFT){
170 matrix[ BACK_LEFT][BACK_CENTER]+= M_SQRT1_2;
171 matrix[BACK_RIGHT][BACK_CENTER]+= M_SQRT1_2;
172 }else if(out_ch_layout & AV_CH_SIDE_LEFT){
173 matrix[ SIDE_LEFT][BACK_CENTER]+= M_SQRT1_2;
174 matrix[SIDE_RIGHT][BACK_CENTER]+= M_SQRT1_2;
175 }else if(out_ch_layout & AV_CH_FRONT_LEFT){
176 if (matrix_encoding == AV_MATRIX_ENCODING_DOLBY ||
177 matrix_encoding == AV_MATRIX_ENCODING_DPLII) {
178 if (unaccounted & (AV_CH_BACK_LEFT | AV_CH_SIDE_LEFT)) {
179 matrix[FRONT_LEFT ][BACK_CENTER] -= s->slev * M_SQRT1_2;
180 matrix[FRONT_RIGHT][BACK_CENTER] += s->slev * M_SQRT1_2;
182 matrix[FRONT_LEFT ][BACK_CENTER] -= s->slev;
183 matrix[FRONT_RIGHT][BACK_CENTER] += s->slev;
186 matrix[ FRONT_LEFT][BACK_CENTER]+= s->slev*M_SQRT1_2;
187 matrix[FRONT_RIGHT][BACK_CENTER]+= s->slev*M_SQRT1_2;
189 }else if(out_ch_layout & AV_CH_FRONT_CENTER){
190 matrix[ FRONT_CENTER][BACK_CENTER]+= s->slev*M_SQRT1_2;
194 if(unaccounted & AV_CH_BACK_LEFT){
195 if(out_ch_layout & AV_CH_BACK_CENTER){
196 matrix[BACK_CENTER][ BACK_LEFT]+= M_SQRT1_2;
197 matrix[BACK_CENTER][BACK_RIGHT]+= M_SQRT1_2;
198 }else if(out_ch_layout & AV_CH_SIDE_LEFT){
199 if(in_ch_layout & AV_CH_SIDE_LEFT){
200 matrix[ SIDE_LEFT][ BACK_LEFT]+= M_SQRT1_2;
201 matrix[SIDE_RIGHT][BACK_RIGHT]+= M_SQRT1_2;
203 matrix[ SIDE_LEFT][ BACK_LEFT]+= 1.0;
204 matrix[SIDE_RIGHT][BACK_RIGHT]+= 1.0;
206 }else if(out_ch_layout & AV_CH_FRONT_LEFT){
207 if (matrix_encoding == AV_MATRIX_ENCODING_DOLBY) {
208 matrix[FRONT_LEFT ][BACK_LEFT ] -= s->slev * M_SQRT1_2;
209 matrix[FRONT_LEFT ][BACK_RIGHT] -= s->slev * M_SQRT1_2;
210 matrix[FRONT_RIGHT][BACK_LEFT ] += s->slev * M_SQRT1_2;
211 matrix[FRONT_RIGHT][BACK_RIGHT] += s->slev * M_SQRT1_2;
212 } else if (matrix_encoding == AV_MATRIX_ENCODING_DPLII) {
213 matrix[FRONT_LEFT ][BACK_LEFT ] -= s->slev * SQRT3_2;
214 matrix[FRONT_LEFT ][BACK_RIGHT] -= s->slev * M_SQRT1_2;
215 matrix[FRONT_RIGHT][BACK_LEFT ] += s->slev * M_SQRT1_2;
216 matrix[FRONT_RIGHT][BACK_RIGHT] += s->slev * SQRT3_2;
218 matrix[ FRONT_LEFT][ BACK_LEFT] += s->slev;
219 matrix[FRONT_RIGHT][BACK_RIGHT] += s->slev;
221 }else if(out_ch_layout & AV_CH_FRONT_CENTER){
222 matrix[ FRONT_CENTER][BACK_LEFT ]+= s->slev*M_SQRT1_2;
223 matrix[ FRONT_CENTER][BACK_RIGHT]+= s->slev*M_SQRT1_2;
228 if(unaccounted & AV_CH_SIDE_LEFT){
229 if(out_ch_layout & AV_CH_BACK_LEFT){
230 /* if back channels do not exist in the input, just copy side
231 channels to back channels, otherwise mix side into back */
232 if (in_ch_layout & AV_CH_BACK_LEFT) {
233 matrix[BACK_LEFT ][SIDE_LEFT ] += M_SQRT1_2;
234 matrix[BACK_RIGHT][SIDE_RIGHT] += M_SQRT1_2;
236 matrix[BACK_LEFT ][SIDE_LEFT ] += 1.0;
237 matrix[BACK_RIGHT][SIDE_RIGHT] += 1.0;
239 }else if(out_ch_layout & AV_CH_BACK_CENTER){
240 matrix[BACK_CENTER][ SIDE_LEFT]+= M_SQRT1_2;
241 matrix[BACK_CENTER][SIDE_RIGHT]+= M_SQRT1_2;
242 }else if(out_ch_layout & AV_CH_FRONT_LEFT){
243 if (matrix_encoding == AV_MATRIX_ENCODING_DOLBY) {
244 matrix[FRONT_LEFT ][SIDE_LEFT ] -= s->slev * M_SQRT1_2;
245 matrix[FRONT_LEFT ][SIDE_RIGHT] -= s->slev * M_SQRT1_2;
246 matrix[FRONT_RIGHT][SIDE_LEFT ] += s->slev * M_SQRT1_2;
247 matrix[FRONT_RIGHT][SIDE_RIGHT] += s->slev * M_SQRT1_2;
248 } else if (matrix_encoding == AV_MATRIX_ENCODING_DPLII) {
249 matrix[FRONT_LEFT ][SIDE_LEFT ] -= s->slev * SQRT3_2;
250 matrix[FRONT_LEFT ][SIDE_RIGHT] -= s->slev * M_SQRT1_2;
251 matrix[FRONT_RIGHT][SIDE_LEFT ] += s->slev * M_SQRT1_2;
252 matrix[FRONT_RIGHT][SIDE_RIGHT] += s->slev * SQRT3_2;
254 matrix[ FRONT_LEFT][ SIDE_LEFT] += s->slev;
255 matrix[FRONT_RIGHT][SIDE_RIGHT] += s->slev;
257 }else if(out_ch_layout & AV_CH_FRONT_CENTER){
258 matrix[ FRONT_CENTER][SIDE_LEFT ]+= s->slev*M_SQRT1_2;
259 matrix[ FRONT_CENTER][SIDE_RIGHT]+= s->slev*M_SQRT1_2;
264 if(unaccounted & AV_CH_FRONT_LEFT_OF_CENTER){
265 if(out_ch_layout & AV_CH_FRONT_LEFT){
266 matrix[ FRONT_LEFT][ FRONT_LEFT_OF_CENTER]+= 1.0;
267 matrix[FRONT_RIGHT][FRONT_RIGHT_OF_CENTER]+= 1.0;
268 }else if(out_ch_layout & AV_CH_FRONT_CENTER){
269 matrix[ FRONT_CENTER][ FRONT_LEFT_OF_CENTER]+= M_SQRT1_2;
270 matrix[ FRONT_CENTER][FRONT_RIGHT_OF_CENTER]+= M_SQRT1_2;
274 /* mix LFE into front left/right or center */
275 if (unaccounted & AV_CH_LOW_FREQUENCY) {
276 if (out_ch_layout & AV_CH_FRONT_CENTER) {
277 matrix[FRONT_CENTER][LOW_FREQUENCY] += s->lfe_mix_level;
278 } else if (out_ch_layout & AV_CH_FRONT_LEFT) {
279 matrix[FRONT_LEFT ][LOW_FREQUENCY] += s->lfe_mix_level * M_SQRT1_2;
280 matrix[FRONT_RIGHT][LOW_FREQUENCY] += s->lfe_mix_level * M_SQRT1_2;
285 for(out_i=i=0; i<64; i++){
289 s->matrix[out_i][in_i]= matrix[i][j];
291 sum += fabs(matrix[i][j]);
293 if(in_ch_layout & (1ULL<<j))
296 maxcoef= FFMAX(maxcoef, sum);
297 if(out_ch_layout & (1ULL<<i))
300 if(s->rematrix_volume < 0)
301 maxcoef = -s->rematrix_volume;
303 if(( av_get_packed_sample_fmt(s->out_sample_fmt) < AV_SAMPLE_FMT_FLT
304 || av_get_packed_sample_fmt(s->int_sample_fmt) < AV_SAMPLE_FMT_FLT) && maxcoef > 1.0){
305 for(i=0; i<SWR_CH_MAX; i++)
306 for(j=0; j<SWR_CH_MAX; j++){
307 s->matrix[i][j] /= maxcoef;
311 if(s->rematrix_volume > 0){
312 for(i=0; i<SWR_CH_MAX; i++)
313 for(j=0; j<SWR_CH_MAX; j++){
314 s->matrix[i][j] *= s->rematrix_volume;
318 for(i=0; i<av_get_channel_layout_nb_channels(out_ch_layout); i++){
319 for(j=0; j<av_get_channel_layout_nb_channels(in_ch_layout); j++){
320 av_log(NULL, AV_LOG_DEBUG, "%f ", s->matrix[i][j]);
322 av_log(NULL, AV_LOG_DEBUG, "\n");
327 av_cold int swri_rematrix_init(SwrContext *s){
329 int nb_in = av_get_channel_layout_nb_channels(s->in_ch_layout);
330 int nb_out = av_get_channel_layout_nb_channels(s->out_ch_layout);
334 if (!s->rematrix_custom) {
335 int r = auto_matrix(s);
339 if (s->midbuf.fmt == AV_SAMPLE_FMT_S16P){
340 s->native_matrix = av_mallocz(nb_in * nb_out * sizeof(int));
341 s->native_one = av_mallocz(sizeof(int));
342 for (i = 0; i < nb_out; i++)
343 for (j = 0; j < nb_in; j++)
344 ((int*)s->native_matrix)[i * nb_in + j] = lrintf(s->matrix[i][j] * 32768);
345 *((int*)s->native_one) = 32768;
346 s->mix_1_1_f = (mix_1_1_func_type*)copy_s16;
347 s->mix_2_1_f = (mix_2_1_func_type*)sum2_s16;
348 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_s16(s);
349 }else if(s->midbuf.fmt == AV_SAMPLE_FMT_FLTP){
350 s->native_matrix = av_mallocz(nb_in * nb_out * sizeof(float));
351 s->native_one = av_mallocz(sizeof(float));
352 for (i = 0; i < nb_out; i++)
353 for (j = 0; j < nb_in; j++)
354 ((float*)s->native_matrix)[i * nb_in + j] = s->matrix[i][j];
355 *((float*)s->native_one) = 1.0;
356 s->mix_1_1_f = (mix_1_1_func_type*)copy_float;
357 s->mix_2_1_f = (mix_2_1_func_type*)sum2_float;
358 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_float(s);
359 }else if(s->midbuf.fmt == AV_SAMPLE_FMT_DBLP){
360 s->native_matrix = av_mallocz(nb_in * nb_out * sizeof(double));
361 s->native_one = av_mallocz(sizeof(double));
362 for (i = 0; i < nb_out; i++)
363 for (j = 0; j < nb_in; j++)
364 ((double*)s->native_matrix)[i * nb_in + j] = s->matrix[i][j];
365 *((double*)s->native_one) = 1.0;
366 s->mix_1_1_f = (mix_1_1_func_type*)copy_double;
367 s->mix_2_1_f = (mix_2_1_func_type*)sum2_double;
368 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_double(s);
371 //FIXME quantize for integeres
372 for (i = 0; i < SWR_CH_MAX; i++) {
374 for (j = 0; j < SWR_CH_MAX; j++) {
375 s->matrix32[i][j]= lrintf(s->matrix[i][j] * 32768);
377 s->matrix_ch[i][++ch_in]= j;
379 s->matrix_ch[i][0]= ch_in;
382 if(HAVE_YASM && HAVE_MMX) swri_rematrix_init_x86(s);
387 av_cold void swri_rematrix_free(SwrContext *s){
388 av_freep(&s->native_matrix);
389 av_freep(&s->native_one);
390 av_freep(&s->native_simd_matrix);
393 int swri_rematrix(SwrContext *s, AudioData *out, AudioData *in, int len, int mustcopy){
394 int out_i, in_i, i, j;
399 s->mix_any_f(out->ch, (const uint8_t **)in->ch, s->native_matrix, len);
403 if(s->mix_2_1_simd || s->mix_1_1_simd){
405 off = len1 * out->bps;
408 av_assert0(out->ch_count == av_get_channel_layout_nb_channels(s->out_ch_layout));
409 av_assert0(in ->ch_count == av_get_channel_layout_nb_channels(s-> in_ch_layout));
411 for(out_i=0; out_i<out->ch_count; out_i++){
412 switch(s->matrix_ch[out_i][0]){
415 memset(out->ch[out_i], 0, len * av_get_bytes_per_sample(s->int_sample_fmt));
418 in_i= s->matrix_ch[out_i][1];
419 if(s->matrix[out_i][in_i]!=1.0){
420 if(s->mix_1_1_simd && len1)
421 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);
423 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);
425 memcpy(out->ch[out_i], in->ch[in_i], len*out->bps);
427 out->ch[out_i]= in->ch[in_i];
431 int in_i1 = s->matrix_ch[out_i][1];
432 int in_i2 = s->matrix_ch[out_i][2];
433 if(s->mix_2_1_simd && len1)
434 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);
436 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);
438 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);
441 if(s->int_sample_fmt == AV_SAMPLE_FMT_FLTP){
442 for(i=0; i<len; i++){
444 for(j=0; j<s->matrix_ch[out_i][0]; j++){
445 in_i= s->matrix_ch[out_i][1+j];
446 v+= ((float*)in->ch[in_i])[i] * s->matrix[out_i][in_i];
448 ((float*)out->ch[out_i])[i]= v;
450 }else if(s->int_sample_fmt == AV_SAMPLE_FMT_DBLP){
451 for(i=0; i<len; i++){
453 for(j=0; j<s->matrix_ch[out_i][0]; j++){
454 in_i= s->matrix_ch[out_i][1+j];
455 v+= ((double*)in->ch[in_i])[i] * s->matrix[out_i][in_i];
457 ((double*)out->ch[out_i])[i]= v;
460 for(i=0; i<len; i++){
462 for(j=0; j<s->matrix_ch[out_i][0]; j++){
463 in_i= s->matrix_ch[out_i][1+j];
464 v+= ((int16_t*)in->ch[in_i])[i] * s->matrix32[out_i][in_i];
466 ((int16_t*)out->ch[out_i])[i]= (v + 16384)>>15;