1 /*
2 * Copyright (C) 2011-2012 Michael Niedermayer (michaelni@gmx.at)
3 *
4 * This file is part of libswresample
5 *
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.
10 *
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.
15 *
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
19 */
20
21 #include "swresample_internal.h"
22 #include "libavutil/avassert.h"
23 #include "libavutil/channel_layout.h"
24
25 #define TEMPLATE_REMATRIX_FLT
26 #include "rematrix_template.c"
27 #undef TEMPLATE_REMATRIX_FLT
28
29 #define TEMPLATE_REMATRIX_DBL
30 #include "rematrix_template.c"
31 #undef TEMPLATE_REMATRIX_DBL
32
33 #define TEMPLATE_REMATRIX_S16
34 #include "rematrix_template.c"
35 #define TEMPLATE_CLIP
36 #include "rematrix_template.c"
37 #undef TEMPLATE_CLIP
38 #undef TEMPLATE_REMATRIX_S16
39
40 #define TEMPLATE_REMATRIX_S32
41 #include "rematrix_template.c"
42 #undef TEMPLATE_REMATRIX_S32
43
44 #define FRONT_LEFT 0
45 #define FRONT_RIGHT 1
46 #define FRONT_CENTER 2
47 #define LOW_FREQUENCY 3
48 #define BACK_LEFT 4
49 #define BACK_RIGHT 5
50 #define FRONT_LEFT_OF_CENTER 6
51 #define FRONT_RIGHT_OF_CENTER 7
52 #define BACK_CENTER 8
53 #define SIDE_LEFT 9
54 #define SIDE_RIGHT 10
55 #define TOP_CENTER 11
56 #define TOP_FRONT_LEFT 12
57 #define TOP_FRONT_CENTER 13
58 #define TOP_FRONT_RIGHT 14
59 #define TOP_BACK_LEFT 15
60 #define TOP_BACK_CENTER 16
61 #define TOP_BACK_RIGHT 17
62 #define NUM_NAMED_CHANNELS 18
63
swr_set_matrix(struct SwrContext * s,const double * matrix,int stride)64 int swr_set_matrix(struct SwrContext *s, const double *matrix, int stride)
65 {
66 int nb_in, nb_out, in, out;
67
68 if (!s || s->in_convert) // s needs to be allocated but not initialized
69 return AVERROR(EINVAL);
70 memset(s->matrix, 0, sizeof(s->matrix));
71 memset(s->matrix_flt, 0, sizeof(s->matrix_flt));
72 nb_in = (s->user_in_ch_count > 0) ? s->user_in_ch_count :
73 av_get_channel_layout_nb_channels(s->user_in_ch_layout);
74 nb_out = (s->user_out_ch_count > 0) ? s->user_out_ch_count :
75 av_get_channel_layout_nb_channels(s->user_out_ch_layout);
76 for (out = 0; out < nb_out; out++) {
77 for (in = 0; in < nb_in; in++)
78 s->matrix_flt[out][in] = s->matrix[out][in] = matrix[in];
79 matrix += stride;
80 }
81 s->rematrix_custom = 1;
82 return 0;
83 }
84
even(int64_t layout)85 static int even(int64_t layout){
86 if(!layout) return 1;
87 if(layout&(layout-1)) return 1;
88 return 0;
89 }
90
clean_layout(void * s,int64_t layout)91 static int clean_layout(void *s, int64_t layout){
92 if(layout && layout != AV_CH_FRONT_CENTER && !(layout&(layout-1))) {
93 char buf[128];
94 av_get_channel_layout_string(buf, sizeof(buf), -1, layout);
95 av_log(s, AV_LOG_VERBOSE, "Treating %s as mono\n", buf);
96 return AV_CH_FRONT_CENTER;
97 }
98
99 return layout;
100 }
101
sane_layout(int64_t layout)102 static int sane_layout(int64_t layout){
103 if(!(layout & AV_CH_LAYOUT_SURROUND)) // at least 1 front speaker
104 return 0;
105 if(!even(layout & (AV_CH_FRONT_LEFT | AV_CH_FRONT_RIGHT))) // no asymetric front
106 return 0;
107 if(!even(layout & (AV_CH_SIDE_LEFT | AV_CH_SIDE_RIGHT))) // no asymetric side
108 return 0;
109 if(!even(layout & (AV_CH_BACK_LEFT | AV_CH_BACK_RIGHT)))
110 return 0;
111 if(!even(layout & (AV_CH_FRONT_LEFT_OF_CENTER | AV_CH_FRONT_RIGHT_OF_CENTER)))
112 return 0;
113 if(av_get_channel_layout_nb_channels(layout) >= SWR_CH_MAX)
114 return 0;
115
116 return 1;
117 }
118
swr_build_matrix(uint64_t in_ch_layout_param,uint64_t out_ch_layout_param,double center_mix_level,double surround_mix_level,double lfe_mix_level,double maxval,double rematrix_volume,double * matrix_param,int stride,enum AVMatrixEncoding matrix_encoding,void * log_context)119 av_cold int swr_build_matrix(uint64_t in_ch_layout_param, uint64_t out_ch_layout_param,
120 double center_mix_level, double surround_mix_level,
121 double lfe_mix_level, double maxval,
122 double rematrix_volume, double *matrix_param,
123 int stride, enum AVMatrixEncoding matrix_encoding, void *log_context)
124 {
125 int i, j, out_i;
126 double matrix[NUM_NAMED_CHANNELS][NUM_NAMED_CHANNELS]={{0}};
127 int64_t unaccounted, in_ch_layout, out_ch_layout;
128 double maxcoef=0;
129 char buf[128];
130
131 in_ch_layout = clean_layout(log_context, in_ch_layout_param);
132 out_ch_layout = clean_layout(log_context, out_ch_layout_param);
133
134 if( out_ch_layout == AV_CH_LAYOUT_STEREO_DOWNMIX
135 && (in_ch_layout & AV_CH_LAYOUT_STEREO_DOWNMIX) == 0
136 )
137 out_ch_layout = AV_CH_LAYOUT_STEREO;
138
139 if( in_ch_layout == AV_CH_LAYOUT_STEREO_DOWNMIX
140 && (out_ch_layout & AV_CH_LAYOUT_STEREO_DOWNMIX) == 0
141 )
142 in_ch_layout = AV_CH_LAYOUT_STEREO;
143
144 if(!sane_layout(in_ch_layout)){
145 av_get_channel_layout_string(buf, sizeof(buf), -1, in_ch_layout_param);
146 av_log(log_context, AV_LOG_ERROR, "Input channel layout '%s' is not supported\n", buf);
147 return AVERROR(EINVAL);
148 }
149
150 if(!sane_layout(out_ch_layout)){
151 av_get_channel_layout_string(buf, sizeof(buf), -1, out_ch_layout_param);
152 av_log(log_context, AV_LOG_ERROR, "Output channel layout '%s' is not supported\n", buf);
153 return AVERROR(EINVAL);
154 }
155
156 for(i=0; i<FF_ARRAY_ELEMS(matrix); i++){
157 if(in_ch_layout & out_ch_layout & (1ULL<<i))
158 matrix[i][i]= 1.0;
159 }
160
161 unaccounted= in_ch_layout & ~out_ch_layout;
162
163 //FIXME implement dolby surround
164 //FIXME implement full ac3
165
166
167 if(unaccounted & AV_CH_FRONT_CENTER){
168 if((out_ch_layout & AV_CH_LAYOUT_STEREO) == AV_CH_LAYOUT_STEREO){
169 if(in_ch_layout & AV_CH_LAYOUT_STEREO) {
170 matrix[ FRONT_LEFT][FRONT_CENTER]+= center_mix_level;
171 matrix[FRONT_RIGHT][FRONT_CENTER]+= center_mix_level;
172 } else {
173 matrix[ FRONT_LEFT][FRONT_CENTER]+= M_SQRT1_2;
174 matrix[FRONT_RIGHT][FRONT_CENTER]+= M_SQRT1_2;
175 }
176 }else
177 av_assert0(0);
178 }
179 if(unaccounted & AV_CH_LAYOUT_STEREO){
180 if(out_ch_layout & AV_CH_FRONT_CENTER){
181 matrix[FRONT_CENTER][ FRONT_LEFT]+= M_SQRT1_2;
182 matrix[FRONT_CENTER][FRONT_RIGHT]+= M_SQRT1_2;
183 if(in_ch_layout & AV_CH_FRONT_CENTER)
184 matrix[FRONT_CENTER][ FRONT_CENTER] = center_mix_level*sqrt(2);
185 }else
186 av_assert0(0);
187 }
188
189 if(unaccounted & AV_CH_BACK_CENTER){
190 if(out_ch_layout & AV_CH_BACK_LEFT){
191 matrix[ BACK_LEFT][BACK_CENTER]+= M_SQRT1_2;
192 matrix[BACK_RIGHT][BACK_CENTER]+= M_SQRT1_2;
193 }else if(out_ch_layout & AV_CH_SIDE_LEFT){
194 matrix[ SIDE_LEFT][BACK_CENTER]+= M_SQRT1_2;
195 matrix[SIDE_RIGHT][BACK_CENTER]+= M_SQRT1_2;
196 }else if(out_ch_layout & AV_CH_FRONT_LEFT){
197 if (matrix_encoding == AV_MATRIX_ENCODING_DOLBY ||
198 matrix_encoding == AV_MATRIX_ENCODING_DPLII) {
199 if (unaccounted & (AV_CH_BACK_LEFT | AV_CH_SIDE_LEFT)) {
200 matrix[FRONT_LEFT ][BACK_CENTER] -= surround_mix_level * M_SQRT1_2;
201 matrix[FRONT_RIGHT][BACK_CENTER] += surround_mix_level * M_SQRT1_2;
202 } else {
203 matrix[FRONT_LEFT ][BACK_CENTER] -= surround_mix_level;
204 matrix[FRONT_RIGHT][BACK_CENTER] += surround_mix_level;
205 }
206 } else {
207 matrix[ FRONT_LEFT][BACK_CENTER]+= surround_mix_level * M_SQRT1_2;
208 matrix[FRONT_RIGHT][BACK_CENTER]+= surround_mix_level * M_SQRT1_2;
209 }
210 }else if(out_ch_layout & AV_CH_FRONT_CENTER){
211 matrix[ FRONT_CENTER][BACK_CENTER]+= surround_mix_level * M_SQRT1_2;
212 }else
213 av_assert0(0);
214 }
215 if(unaccounted & AV_CH_BACK_LEFT){
216 if(out_ch_layout & AV_CH_BACK_CENTER){
217 matrix[BACK_CENTER][ BACK_LEFT]+= M_SQRT1_2;
218 matrix[BACK_CENTER][BACK_RIGHT]+= M_SQRT1_2;
219 }else if(out_ch_layout & AV_CH_SIDE_LEFT){
220 if(in_ch_layout & AV_CH_SIDE_LEFT){
221 matrix[ SIDE_LEFT][ BACK_LEFT]+= M_SQRT1_2;
222 matrix[SIDE_RIGHT][BACK_RIGHT]+= M_SQRT1_2;
223 }else{
224 matrix[ SIDE_LEFT][ BACK_LEFT]+= 1.0;
225 matrix[SIDE_RIGHT][BACK_RIGHT]+= 1.0;
226 }
227 }else if(out_ch_layout & AV_CH_FRONT_LEFT){
228 if (matrix_encoding == AV_MATRIX_ENCODING_DOLBY) {
229 matrix[FRONT_LEFT ][BACK_LEFT ] -= surround_mix_level * M_SQRT1_2;
230 matrix[FRONT_LEFT ][BACK_RIGHT] -= surround_mix_level * M_SQRT1_2;
231 matrix[FRONT_RIGHT][BACK_LEFT ] += surround_mix_level * M_SQRT1_2;
232 matrix[FRONT_RIGHT][BACK_RIGHT] += surround_mix_level * M_SQRT1_2;
233 } else if (matrix_encoding == AV_MATRIX_ENCODING_DPLII) {
234 matrix[FRONT_LEFT ][BACK_LEFT ] -= surround_mix_level * SQRT3_2;
235 matrix[FRONT_LEFT ][BACK_RIGHT] -= surround_mix_level * M_SQRT1_2;
236 matrix[FRONT_RIGHT][BACK_LEFT ] += surround_mix_level * M_SQRT1_2;
237 matrix[FRONT_RIGHT][BACK_RIGHT] += surround_mix_level * SQRT3_2;
238 } else {
239 matrix[ FRONT_LEFT][ BACK_LEFT] += surround_mix_level;
240 matrix[FRONT_RIGHT][BACK_RIGHT] += surround_mix_level;
241 }
242 }else if(out_ch_layout & AV_CH_FRONT_CENTER){
243 matrix[ FRONT_CENTER][BACK_LEFT ]+= surround_mix_level*M_SQRT1_2;
244 matrix[ FRONT_CENTER][BACK_RIGHT]+= surround_mix_level*M_SQRT1_2;
245 }else
246 av_assert0(0);
247 }
248
249 if(unaccounted & AV_CH_SIDE_LEFT){
250 if(out_ch_layout & AV_CH_BACK_LEFT){
251 /* if back channels do not exist in the input, just copy side
252 channels to back channels, otherwise mix side into back */
253 if (in_ch_layout & AV_CH_BACK_LEFT) {
254 matrix[BACK_LEFT ][SIDE_LEFT ] += M_SQRT1_2;
255 matrix[BACK_RIGHT][SIDE_RIGHT] += M_SQRT1_2;
256 } else {
257 matrix[BACK_LEFT ][SIDE_LEFT ] += 1.0;
258 matrix[BACK_RIGHT][SIDE_RIGHT] += 1.0;
259 }
260 }else if(out_ch_layout & AV_CH_BACK_CENTER){
261 matrix[BACK_CENTER][ SIDE_LEFT]+= M_SQRT1_2;
262 matrix[BACK_CENTER][SIDE_RIGHT]+= M_SQRT1_2;
263 }else if(out_ch_layout & AV_CH_FRONT_LEFT){
264 if (matrix_encoding == AV_MATRIX_ENCODING_DOLBY) {
265 matrix[FRONT_LEFT ][SIDE_LEFT ] -= surround_mix_level * M_SQRT1_2;
266 matrix[FRONT_LEFT ][SIDE_RIGHT] -= surround_mix_level * M_SQRT1_2;
267 matrix[FRONT_RIGHT][SIDE_LEFT ] += surround_mix_level * M_SQRT1_2;
268 matrix[FRONT_RIGHT][SIDE_RIGHT] += surround_mix_level * M_SQRT1_2;
269 } else if (matrix_encoding == AV_MATRIX_ENCODING_DPLII) {
270 matrix[FRONT_LEFT ][SIDE_LEFT ] -= surround_mix_level * SQRT3_2;
271 matrix[FRONT_LEFT ][SIDE_RIGHT] -= surround_mix_level * M_SQRT1_2;
272 matrix[FRONT_RIGHT][SIDE_LEFT ] += surround_mix_level * M_SQRT1_2;
273 matrix[FRONT_RIGHT][SIDE_RIGHT] += surround_mix_level * SQRT3_2;
274 } else {
275 matrix[ FRONT_LEFT][ SIDE_LEFT] += surround_mix_level;
276 matrix[FRONT_RIGHT][SIDE_RIGHT] += surround_mix_level;
277 }
278 }else if(out_ch_layout & AV_CH_FRONT_CENTER){
279 matrix[ FRONT_CENTER][SIDE_LEFT ]+= surround_mix_level * M_SQRT1_2;
280 matrix[ FRONT_CENTER][SIDE_RIGHT]+= surround_mix_level * M_SQRT1_2;
281 }else
282 av_assert0(0);
283 }
284
285 if(unaccounted & AV_CH_FRONT_LEFT_OF_CENTER){
286 if(out_ch_layout & AV_CH_FRONT_LEFT){
287 matrix[ FRONT_LEFT][ FRONT_LEFT_OF_CENTER]+= 1.0;
288 matrix[FRONT_RIGHT][FRONT_RIGHT_OF_CENTER]+= 1.0;
289 }else if(out_ch_layout & AV_CH_FRONT_CENTER){
290 matrix[ FRONT_CENTER][ FRONT_LEFT_OF_CENTER]+= M_SQRT1_2;
291 matrix[ FRONT_CENTER][FRONT_RIGHT_OF_CENTER]+= M_SQRT1_2;
292 }else
293 av_assert0(0);
294 }
295 /* mix LFE into front left/right or center */
296 if (unaccounted & AV_CH_LOW_FREQUENCY) {
297 if (out_ch_layout & AV_CH_FRONT_CENTER) {
298 matrix[FRONT_CENTER][LOW_FREQUENCY] += lfe_mix_level;
299 } else if (out_ch_layout & AV_CH_FRONT_LEFT) {
300 matrix[FRONT_LEFT ][LOW_FREQUENCY] += lfe_mix_level * M_SQRT1_2;
301 matrix[FRONT_RIGHT][LOW_FREQUENCY] += lfe_mix_level * M_SQRT1_2;
302 } else
303 av_assert0(0);
304 }
305
306 for(out_i=i=0; i<64; i++){
307 double sum=0;
308 int in_i=0;
309 if((out_ch_layout & (1ULL<<i)) == 0)
310 continue;
311 for(j=0; j<64; j++){
312 if((in_ch_layout & (1ULL<<j)) == 0)
313 continue;
314 if (i < FF_ARRAY_ELEMS(matrix) && j < FF_ARRAY_ELEMS(matrix[0]))
315 matrix_param[stride*out_i + in_i] = matrix[i][j];
316 else
317 matrix_param[stride*out_i + in_i] = i == j && (in_ch_layout & out_ch_layout & (1ULL<<i));
318 sum += fabs(matrix_param[stride*out_i + in_i]);
319 in_i++;
320 }
321 maxcoef= FFMAX(maxcoef, sum);
322 out_i++;
323 }
324 if(rematrix_volume < 0)
325 maxcoef = -rematrix_volume;
326
327 if(maxcoef > maxval || rematrix_volume < 0){
328 maxcoef /= maxval;
329 for(i=0; i<SWR_CH_MAX; i++)
330 for(j=0; j<SWR_CH_MAX; j++){
331 matrix_param[stride*i + j] /= maxcoef;
332 }
333 }
334
335 if(rematrix_volume > 0){
336 for(i=0; i<SWR_CH_MAX; i++)
337 for(j=0; j<SWR_CH_MAX; j++){
338 matrix_param[stride*i + j] *= rematrix_volume;
339 }
340 }
341
342 av_log(log_context, AV_LOG_DEBUG, "Matrix coefficients:\n");
343 for(i=0; i<av_get_channel_layout_nb_channels(out_ch_layout); i++){
344 const char *c =
345 av_get_channel_name(av_channel_layout_extract_channel(out_ch_layout, i));
346 av_log(log_context, AV_LOG_DEBUG, "%s: ", c ? c : "?");
347 for(j=0; j<av_get_channel_layout_nb_channels(in_ch_layout); j++){
348 c = av_get_channel_name(av_channel_layout_extract_channel(in_ch_layout, j));
349 av_log(log_context, AV_LOG_DEBUG, "%s:%f ", c ? c : "?", matrix_param[stride*i + j]);
350 }
351 av_log(log_context, AV_LOG_DEBUG, "\n");
352 }
353 return 0;
354 }
355
auto_matrix(SwrContext * s)356 av_cold static int auto_matrix(SwrContext *s)
357 {
358 double maxval;
359 int ret;
360
361 if (s->rematrix_maxval > 0) {
362 maxval = s->rematrix_maxval;
363 } else if ( av_get_packed_sample_fmt(s->out_sample_fmt) < AV_SAMPLE_FMT_FLT
364 || av_get_packed_sample_fmt(s->int_sample_fmt) < AV_SAMPLE_FMT_FLT) {
365 maxval = 1.0;
366 } else
367 maxval = INT_MAX;
368
369 memset(s->matrix, 0, sizeof(s->matrix));
370 ret = swr_build_matrix(s->in_ch_layout, s->out_ch_layout,
371 s->clev, s->slev, s->lfe_mix_level,
372 maxval, s->rematrix_volume, (double*)s->matrix,
373 s->matrix[1] - s->matrix[0], s->matrix_encoding, s);
374
375 if (ret >= 0 && s->int_sample_fmt == AV_SAMPLE_FMT_FLTP) {
376 int i, j;
377 for (i = 0; i < FF_ARRAY_ELEMS(s->matrix[0]); i++)
378 for (j = 0; j < FF_ARRAY_ELEMS(s->matrix[0]); j++)
379 s->matrix_flt[i][j] = s->matrix[i][j];
380 }
381
382 return ret;
383 }
384
swri_rematrix_init(SwrContext * s)385 av_cold int swri_rematrix_init(SwrContext *s){
386 int i, j;
387 int nb_in = s->used_ch_count;
388 int nb_out = s->out.ch_count;
389
390 s->mix_any_f = NULL;
391
392 if (!s->rematrix_custom) {
393 int r = auto_matrix(s);
394 if (r)
395 return r;
396 }
397 if (s->midbuf.fmt == AV_SAMPLE_FMT_S16P){
398 int maxsum = 0;
399 s->native_matrix = av_calloc(nb_in * nb_out, sizeof(int));
400 s->native_one = av_mallocz(sizeof(int));
401 if (!s->native_matrix || !s->native_one)
402 return AVERROR(ENOMEM);
403 for (i = 0; i < nb_out; i++) {
404 double rem = 0;
405 int sum = 0;
406
407 for (j = 0; j < nb_in; j++) {
408 double target = s->matrix[i][j] * 32768 + rem;
409 ((int*)s->native_matrix)[i * nb_in + j] = lrintf(target);
410 rem += target - ((int*)s->native_matrix)[i * nb_in + j];
411 sum += FFABS(((int*)s->native_matrix)[i * nb_in + j]);
412 }
413 maxsum = FFMAX(maxsum, sum);
414 }
415 *((int*)s->native_one) = 32768;
416 if (maxsum <= 32768) {
417 s->mix_1_1_f = (mix_1_1_func_type*)copy_s16;
418 s->mix_2_1_f = (mix_2_1_func_type*)sum2_s16;
419 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_s16(s);
420 } else {
421 s->mix_1_1_f = (mix_1_1_func_type*)copy_clip_s16;
422 s->mix_2_1_f = (mix_2_1_func_type*)sum2_clip_s16;
423 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_clip_s16(s);
424 }
425 }else if(s->midbuf.fmt == AV_SAMPLE_FMT_FLTP){
426 s->native_matrix = av_calloc(nb_in * nb_out, sizeof(float));
427 s->native_one = av_mallocz(sizeof(float));
428 if (!s->native_matrix || !s->native_one)
429 return AVERROR(ENOMEM);
430 for (i = 0; i < nb_out; i++)
431 for (j = 0; j < nb_in; j++)
432 ((float*)s->native_matrix)[i * nb_in + j] = s->matrix[i][j];
433 *((float*)s->native_one) = 1.0;
434 s->mix_1_1_f = (mix_1_1_func_type*)copy_float;
435 s->mix_2_1_f = (mix_2_1_func_type*)sum2_float;
436 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_float(s);
437 }else if(s->midbuf.fmt == AV_SAMPLE_FMT_DBLP){
438 s->native_matrix = av_calloc(nb_in * nb_out, sizeof(double));
439 s->native_one = av_mallocz(sizeof(double));
440 if (!s->native_matrix || !s->native_one)
441 return AVERROR(ENOMEM);
442 for (i = 0; i < nb_out; i++)
443 for (j = 0; j < nb_in; j++)
444 ((double*)s->native_matrix)[i * nb_in + j] = s->matrix[i][j];
445 *((double*)s->native_one) = 1.0;
446 s->mix_1_1_f = (mix_1_1_func_type*)copy_double;
447 s->mix_2_1_f = (mix_2_1_func_type*)sum2_double;
448 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_double(s);
449 }else if(s->midbuf.fmt == AV_SAMPLE_FMT_S32P){
450 s->native_one = av_mallocz(sizeof(int));
451 if (!s->native_one)
452 return AVERROR(ENOMEM);
453 s->native_matrix = av_calloc(nb_in * nb_out, sizeof(int));
454 if (!s->native_matrix) {
455 av_freep(&s->native_one);
456 return AVERROR(ENOMEM);
457 }
458 for (i = 0; i < nb_out; i++) {
459 double rem = 0;
460
461 for (j = 0; j < nb_in; j++) {
462 double target = s->matrix[i][j] * 32768 + rem;
463 ((int*)s->native_matrix)[i * nb_in + j] = lrintf(target);
464 rem += target - ((int*)s->native_matrix)[i * nb_in + j];
465 }
466 }
467 *((int*)s->native_one) = 32768;
468 s->mix_1_1_f = (mix_1_1_func_type*)copy_s32;
469 s->mix_2_1_f = (mix_2_1_func_type*)sum2_s32;
470 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_s32(s);
471 }else
472 av_assert0(0);
473 //FIXME quantize for integeres
474 for (i = 0; i < SWR_CH_MAX; i++) {
475 int ch_in=0;
476 for (j = 0; j < SWR_CH_MAX; j++) {
477 s->matrix32[i][j]= lrintf(s->matrix[i][j] * 32768);
478 if(s->matrix[i][j])
479 s->matrix_ch[i][++ch_in]= j;
480 }
481 s->matrix_ch[i][0]= ch_in;
482 }
483
484 if(HAVE_X86ASM && HAVE_MMX)
485 return swri_rematrix_init_x86(s);
486
487 return 0;
488 }
489
swri_rematrix_free(SwrContext * s)490 av_cold void swri_rematrix_free(SwrContext *s){
491 av_freep(&s->native_matrix);
492 av_freep(&s->native_one);
493 av_freep(&s->native_simd_matrix);
494 av_freep(&s->native_simd_one);
495 }
496
swri_rematrix(SwrContext * s,AudioData * out,AudioData * in,int len,int mustcopy)497 int swri_rematrix(SwrContext *s, AudioData *out, AudioData *in, int len, int mustcopy){
498 int out_i, in_i, i, j;
499 int len1 = 0;
500 int off = 0;
501
502 if(s->mix_any_f) {
503 s->mix_any_f(out->ch, (const uint8_t **)in->ch, s->native_matrix, len);
504 return 0;
505 }
506
507 if(s->mix_2_1_simd || s->mix_1_1_simd){
508 len1= len&~15;
509 off = len1 * out->bps;
510 }
511
512 av_assert0(!s->out_ch_layout || out->ch_count == av_get_channel_layout_nb_channels(s->out_ch_layout));
513 av_assert0(!s-> in_ch_layout || in ->ch_count == av_get_channel_layout_nb_channels(s-> in_ch_layout));
514
515 for(out_i=0; out_i<out->ch_count; out_i++){
516 switch(s->matrix_ch[out_i][0]){
517 case 0:
518 if(mustcopy)
519 memset(out->ch[out_i], 0, len * av_get_bytes_per_sample(s->int_sample_fmt));
520 break;
521 case 1:
522 in_i= s->matrix_ch[out_i][1];
523 if(s->matrix[out_i][in_i]!=1.0){
524 if(s->mix_1_1_simd && len1)
525 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);
526 if(len != len1)
527 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);
528 }else if(mustcopy){
529 memcpy(out->ch[out_i], in->ch[in_i], len*out->bps);
530 }else{
531 out->ch[out_i]= in->ch[in_i];
532 }
533 break;
534 case 2: {
535 int in_i1 = s->matrix_ch[out_i][1];
536 int in_i2 = s->matrix_ch[out_i][2];
537 if(s->mix_2_1_simd && len1)
538 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);
539 else
540 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);
541 if(len != len1)
542 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);
543 break;}
544 default:
545 if(s->int_sample_fmt == AV_SAMPLE_FMT_FLTP){
546 for(i=0; i<len; i++){
547 float v=0;
548 for(j=0; j<s->matrix_ch[out_i][0]; j++){
549 in_i= s->matrix_ch[out_i][1+j];
550 v+= ((float*)in->ch[in_i])[i] * s->matrix_flt[out_i][in_i];
551 }
552 ((float*)out->ch[out_i])[i]= v;
553 }
554 }else if(s->int_sample_fmt == AV_SAMPLE_FMT_DBLP){
555 for(i=0; i<len; i++){
556 double v=0;
557 for(j=0; j<s->matrix_ch[out_i][0]; j++){
558 in_i= s->matrix_ch[out_i][1+j];
559 v+= ((double*)in->ch[in_i])[i] * s->matrix[out_i][in_i];
560 }
561 ((double*)out->ch[out_i])[i]= v;
562 }
563 }else{
564 for(i=0; i<len; i++){
565 int v=0;
566 for(j=0; j<s->matrix_ch[out_i][0]; j++){
567 in_i= s->matrix_ch[out_i][1+j];
568 v+= ((int16_t*)in->ch[in_i])[i] * s->matrix32[out_i][in_i];
569 }
570 ((int16_t*)out->ch[out_i])[i]= (v + 16384)>>15;
571 }
572 }
573 }
574 }
575 return 0;
576 }
577