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/audioconvert.h"
23 #include "libavutil/avassert.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 sane_layout(int64_t layout){
106 if(!(layout & AV_CH_LAYOUT_SURROUND)) // at least 1 front speaker
108 if(!even(layout & (AV_CH_FRONT_LEFT | AV_CH_FRONT_RIGHT))) // no asymetric front
110 if(!even(layout & (AV_CH_SIDE_LEFT | AV_CH_SIDE_RIGHT))) // no asymetric side
112 if(!even(layout & (AV_CH_BACK_LEFT | AV_CH_BACK_RIGHT)))
114 if(!even(layout & (AV_CH_FRONT_LEFT_OF_CENTER | AV_CH_FRONT_RIGHT_OF_CENTER)))
116 if(av_get_channel_layout_nb_channels(layout) >= SWR_CH_MAX)
122 static int auto_matrix(SwrContext *s)
125 double matrix[64][64]={{0}};
126 int64_t unaccounted= s->in_ch_layout & ~s->out_ch_layout;
130 memset(s->matrix, 0, sizeof(s->matrix));
132 if(s->in_ch_layout & s->out_ch_layout & (1LL<<i))
136 if(!sane_layout(s->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);
141 if(!sane_layout(s->out_ch_layout)){
142 av_get_channel_layout_string(buf, sizeof(buf), -1, s->out_ch_layout);
143 av_log(s, AV_LOG_ERROR, "Output channel layout '%s' is not supported\n", buf);
144 return AVERROR(EINVAL);
147 //FIXME implement dolby surround
148 //FIXME implement full ac3
151 if(unaccounted & AV_CH_FRONT_CENTER){
152 if((s->out_ch_layout & AV_CH_LAYOUT_STEREO) == AV_CH_LAYOUT_STEREO){
153 matrix[ FRONT_LEFT][FRONT_CENTER]+= M_SQRT1_2;
154 matrix[FRONT_RIGHT][FRONT_CENTER]+= M_SQRT1_2;
158 if(unaccounted & AV_CH_LAYOUT_STEREO){
159 if(s->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(s->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(s->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(s->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(s->out_ch_layout & AV_CH_FRONT_LEFT){
176 matrix[ FRONT_LEFT][BACK_CENTER]+= s->slev*M_SQRT1_2;
177 matrix[FRONT_RIGHT][BACK_CENTER]+= s->slev*M_SQRT1_2;
178 }else if(s->out_ch_layout & AV_CH_FRONT_CENTER){
179 matrix[ FRONT_CENTER][BACK_CENTER]+= s->slev*M_SQRT1_2;
183 if(unaccounted & AV_CH_BACK_LEFT){
184 if(s->out_ch_layout & AV_CH_BACK_CENTER){
185 matrix[BACK_CENTER][ BACK_LEFT]+= M_SQRT1_2;
186 matrix[BACK_CENTER][BACK_RIGHT]+= M_SQRT1_2;
187 }else if(s->out_ch_layout & AV_CH_SIDE_LEFT){
188 if(s->in_ch_layout & AV_CH_SIDE_LEFT){
189 matrix[ SIDE_LEFT][ BACK_LEFT]+= M_SQRT1_2;
190 matrix[SIDE_RIGHT][BACK_RIGHT]+= M_SQRT1_2;
192 matrix[ SIDE_LEFT][ BACK_LEFT]+= 1.0;
193 matrix[SIDE_RIGHT][BACK_RIGHT]+= 1.0;
195 }else if(s->out_ch_layout & AV_CH_FRONT_LEFT){
196 matrix[ FRONT_LEFT][ BACK_LEFT]+= s->slev;
197 matrix[FRONT_RIGHT][BACK_RIGHT]+= s->slev;
198 }else if(s->out_ch_layout & AV_CH_FRONT_CENTER){
199 matrix[ FRONT_CENTER][BACK_LEFT ]+= s->slev*M_SQRT1_2;
200 matrix[ FRONT_CENTER][BACK_RIGHT]+= s->slev*M_SQRT1_2;
205 if(unaccounted & AV_CH_SIDE_LEFT){
206 if(s->out_ch_layout & AV_CH_BACK_LEFT){
207 /* if back channels do not exist in the input, just copy side
208 channels to back channels, otherwise mix side into back */
209 if (s->in_ch_layout & AV_CH_BACK_LEFT) {
210 matrix[BACK_LEFT ][SIDE_LEFT ] += M_SQRT1_2;
211 matrix[BACK_RIGHT][SIDE_RIGHT] += M_SQRT1_2;
213 matrix[BACK_LEFT ][SIDE_LEFT ] += 1.0;
214 matrix[BACK_RIGHT][SIDE_RIGHT] += 1.0;
216 }else if(s->out_ch_layout & AV_CH_BACK_CENTER){
217 matrix[BACK_CENTER][ SIDE_LEFT]+= M_SQRT1_2;
218 matrix[BACK_CENTER][SIDE_RIGHT]+= M_SQRT1_2;
219 }else if(s->out_ch_layout & AV_CH_FRONT_LEFT){
220 matrix[ FRONT_LEFT][ SIDE_LEFT]+= s->slev;
221 matrix[FRONT_RIGHT][SIDE_RIGHT]+= s->slev;
222 }else if(s->out_ch_layout & AV_CH_FRONT_CENTER){
223 matrix[ FRONT_CENTER][SIDE_LEFT ]+= s->slev*M_SQRT1_2;
224 matrix[ FRONT_CENTER][SIDE_RIGHT]+= s->slev*M_SQRT1_2;
229 if(unaccounted & AV_CH_FRONT_LEFT_OF_CENTER){
230 if(s->out_ch_layout & AV_CH_FRONT_LEFT){
231 matrix[ FRONT_LEFT][ FRONT_LEFT_OF_CENTER]+= 1.0;
232 matrix[FRONT_RIGHT][FRONT_RIGHT_OF_CENTER]+= 1.0;
233 }else if(s->out_ch_layout & AV_CH_FRONT_CENTER){
234 matrix[ FRONT_CENTER][ FRONT_LEFT_OF_CENTER]+= M_SQRT1_2;
235 matrix[ FRONT_CENTER][FRONT_RIGHT_OF_CENTER]+= M_SQRT1_2;
239 /* mix LFE into front left/right or center */
240 if (unaccounted & AV_CH_LOW_FREQUENCY) {
241 if (s->out_ch_layout & AV_CH_FRONT_CENTER) {
242 matrix[FRONT_CENTER][LOW_FREQUENCY] += s->lfe_mix_level;
243 } else if (s->out_ch_layout & AV_CH_FRONT_LEFT) {
244 matrix[FRONT_LEFT ][LOW_FREQUENCY] += s->lfe_mix_level * M_SQRT1_2;
245 matrix[FRONT_RIGHT][LOW_FREQUENCY] += s->lfe_mix_level * M_SQRT1_2;
250 for(out_i=i=0; i<64; i++){
254 s->matrix[out_i][in_i]= matrix[i][j];
256 sum += fabs(matrix[i][j]);
258 if(s->in_ch_layout & (1ULL<<j))
261 maxcoef= FFMAX(maxcoef, sum);
262 if(s->out_ch_layout & (1ULL<<i))
265 if(s->rematrix_volume < 0)
266 maxcoef = -s->rematrix_volume;
268 if(( av_get_packed_sample_fmt(s->out_sample_fmt) < AV_SAMPLE_FMT_FLT
269 || av_get_packed_sample_fmt(s->int_sample_fmt) < AV_SAMPLE_FMT_FLT) && maxcoef > 1.0){
270 for(i=0; i<SWR_CH_MAX; i++)
271 for(j=0; j<SWR_CH_MAX; j++){
272 s->matrix[i][j] /= maxcoef;
276 if(s->rematrix_volume > 0){
277 for(i=0; i<SWR_CH_MAX; i++)
278 for(j=0; j<SWR_CH_MAX; j++){
279 s->matrix[i][j] *= s->rematrix_volume;
283 for(i=0; i<av_get_channel_layout_nb_channels(s->out_ch_layout); i++){
284 for(j=0; j<av_get_channel_layout_nb_channels(s->in_ch_layout); j++){
285 av_log(NULL, AV_LOG_DEBUG, "%f ", s->matrix[i][j]);
287 av_log(NULL, AV_LOG_DEBUG, "\n");
292 int swri_rematrix_init(SwrContext *s){
294 int nb_in = av_get_channel_layout_nb_channels(s->in_ch_layout);
295 int nb_out = av_get_channel_layout_nb_channels(s->out_ch_layout);
299 if (!s->rematrix_custom) {
300 int r = auto_matrix(s);
304 if (s->midbuf.fmt == AV_SAMPLE_FMT_S16P){
305 s->native_matrix = av_mallocz(nb_in * nb_out * sizeof(int));
306 s->native_one = av_mallocz(sizeof(int));
307 for (i = 0; i < nb_out; i++)
308 for (j = 0; j < nb_in; j++)
309 ((int*)s->native_matrix)[i * nb_in + j] = lrintf(s->matrix[i][j] * 32768);
310 *((int*)s->native_one) = 32768;
311 s->mix_1_1_f = (mix_1_1_func_type*)copy_s16;
312 s->mix_2_1_f = (mix_2_1_func_type*)sum2_s16;
313 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_s16(s);
314 }else if(s->midbuf.fmt == AV_SAMPLE_FMT_FLTP){
315 s->native_matrix = av_mallocz(nb_in * nb_out * sizeof(float));
316 s->native_one = av_mallocz(sizeof(float));
317 for (i = 0; i < nb_out; i++)
318 for (j = 0; j < nb_in; j++)
319 ((float*)s->native_matrix)[i * nb_in + j] = s->matrix[i][j];
320 *((float*)s->native_one) = 1.0;
321 s->mix_1_1_f = (mix_1_1_func_type*)copy_float;
322 s->mix_2_1_f = (mix_2_1_func_type*)sum2_float;
323 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_float(s);
324 }else if(s->midbuf.fmt == AV_SAMPLE_FMT_DBLP){
325 s->native_matrix = av_mallocz(nb_in * nb_out * sizeof(double));
326 s->native_one = av_mallocz(sizeof(double));
327 for (i = 0; i < nb_out; i++)
328 for (j = 0; j < nb_in; j++)
329 ((double*)s->native_matrix)[i * nb_in + j] = s->matrix[i][j];
330 *((double*)s->native_one) = 1.0;
331 s->mix_1_1_f = (mix_1_1_func_type*)copy_double;
332 s->mix_2_1_f = (mix_2_1_func_type*)sum2_double;
333 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_double(s);
336 //FIXME quantize for integeres
337 for (i = 0; i < SWR_CH_MAX; i++) {
339 for (j = 0; j < SWR_CH_MAX; j++) {
340 s->matrix32[i][j]= lrintf(s->matrix[i][j] * 32768);
342 s->matrix_ch[i][++ch_in]= j;
344 s->matrix_ch[i][0]= ch_in;
347 if(HAVE_YASM && HAVE_MMX) swri_rematrix_init_x86(s);
352 void swri_rematrix_free(SwrContext *s){
353 av_freep(&s->native_matrix);
354 av_freep(&s->native_one);
355 av_freep(&s->native_simd_matrix);
358 int swri_rematrix(SwrContext *s, AudioData *out, AudioData *in, int len, int mustcopy){
359 int out_i, in_i, i, j;
364 s->mix_any_f(out->ch, (const uint8_t **)in->ch, s->native_matrix, len);
368 if(s->mix_2_1_simd || s->mix_1_1_simd){
370 off = len1 * out->bps;
373 av_assert0(out->ch_count == av_get_channel_layout_nb_channels(s->out_ch_layout));
374 av_assert0(in ->ch_count == av_get_channel_layout_nb_channels(s-> in_ch_layout));
376 for(out_i=0; out_i<out->ch_count; out_i++){
377 switch(s->matrix_ch[out_i][0]){
380 memset(out->ch[out_i], 0, len * av_get_bytes_per_sample(s->int_sample_fmt));
383 in_i= s->matrix_ch[out_i][1];
384 if(s->matrix[out_i][in_i]!=1.0){
385 if(s->mix_1_1_simd && len1)
386 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);
388 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);
390 memcpy(out->ch[out_i], in->ch[in_i], len*out->bps);
392 out->ch[out_i]= in->ch[in_i];
396 int in_i1 = s->matrix_ch[out_i][1];
397 int in_i2 = s->matrix_ch[out_i][2];
398 if(s->mix_2_1_simd && len1)
399 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);
401 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);
403 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);
406 if(s->int_sample_fmt == AV_SAMPLE_FMT_FLTP){
407 for(i=0; i<len; i++){
409 for(j=0; j<s->matrix_ch[out_i][0]; j++){
410 in_i= s->matrix_ch[out_i][1+j];
411 v+= ((float*)in->ch[in_i])[i] * s->matrix[out_i][in_i];
413 ((float*)out->ch[out_i])[i]= v;
415 }else if(s->int_sample_fmt == AV_SAMPLE_FMT_DBLP){
416 for(i=0; i<len; i++){
418 for(j=0; j<s->matrix_ch[out_i][0]; j++){
419 in_i= s->matrix_ch[out_i][1+j];
420 v+= ((double*)in->ch[in_i])[i] * s->matrix[out_i][in_i];
422 ((double*)out->ch[out_i])[i]= v;
425 for(i=0; i<len; i++){
427 for(j=0; j<s->matrix_ch[out_i][0]; j++){
428 in_i= s->matrix_ch[out_i][1+j];
429 v+= ((int16_t*)in->ch[in_i])[i] * s->matrix32[out_i][in_i];
431 ((int16_t*)out->ch[out_i])[i]= (v + 16384)>>15;