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 int64_t 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 (in_ch_layout == AV_CH_LAYOUT_22POINT2 &&
145 out_ch_layout != AV_CH_LAYOUT_22POINT2) {
146 in_ch_layout = (AV_CH_LAYOUT_7POINT1_WIDE_BACK|AV_CH_BACK_CENTER);
147 av_get_channel_layout_string(buf, sizeof(buf), -1, in_ch_layout);
148 av_log(log_context, AV_LOG_WARNING,
149 "Full-on remixing from 22.2 has not yet been implemented! "
150 "Processing the input as '%s'\n",
151 buf);
152 }
153
154 if(!sane_layout(in_ch_layout)){
155 av_get_channel_layout_string(buf, sizeof(buf), -1, in_ch_layout_param);
156 av_log(log_context, AV_LOG_ERROR, "Input channel layout '%s' is not supported\n", buf);
157 return AVERROR(EINVAL);
158 }
159
160 if(!sane_layout(out_ch_layout)){
161 av_get_channel_layout_string(buf, sizeof(buf), -1, out_ch_layout_param);
162 av_log(log_context, AV_LOG_ERROR, "Output channel layout '%s' is not supported\n", buf);
163 return AVERROR(EINVAL);
164 }
165
166 for(i=0; i<FF_ARRAY_ELEMS(matrix); i++){
167 if(in_ch_layout & out_ch_layout & (1ULL<<i))
168 matrix[i][i]= 1.0;
169 }
170
171 unaccounted= in_ch_layout & ~out_ch_layout;
172
173 //FIXME implement dolby surround
174 //FIXME implement full ac3
175
176
177 if(unaccounted & AV_CH_FRONT_CENTER){
178 if((out_ch_layout & AV_CH_LAYOUT_STEREO) == AV_CH_LAYOUT_STEREO){
179 if(in_ch_layout & AV_CH_LAYOUT_STEREO) {
180 matrix[ FRONT_LEFT][FRONT_CENTER]+= center_mix_level;
181 matrix[FRONT_RIGHT][FRONT_CENTER]+= center_mix_level;
182 } else {
183 matrix[ FRONT_LEFT][FRONT_CENTER]+= M_SQRT1_2;
184 matrix[FRONT_RIGHT][FRONT_CENTER]+= M_SQRT1_2;
185 }
186 }else
187 av_assert0(0);
188 }
189 if(unaccounted & AV_CH_LAYOUT_STEREO){
190 if(out_ch_layout & AV_CH_FRONT_CENTER){
191 matrix[FRONT_CENTER][ FRONT_LEFT]+= M_SQRT1_2;
192 matrix[FRONT_CENTER][FRONT_RIGHT]+= M_SQRT1_2;
193 if(in_ch_layout & AV_CH_FRONT_CENTER)
194 matrix[FRONT_CENTER][ FRONT_CENTER] = center_mix_level*sqrt(2);
195 }else
196 av_assert0(0);
197 }
198
199 if(unaccounted & AV_CH_BACK_CENTER){
200 if(out_ch_layout & AV_CH_BACK_LEFT){
201 matrix[ BACK_LEFT][BACK_CENTER]+= M_SQRT1_2;
202 matrix[BACK_RIGHT][BACK_CENTER]+= M_SQRT1_2;
203 }else if(out_ch_layout & AV_CH_SIDE_LEFT){
204 matrix[ SIDE_LEFT][BACK_CENTER]+= M_SQRT1_2;
205 matrix[SIDE_RIGHT][BACK_CENTER]+= M_SQRT1_2;
206 }else if(out_ch_layout & AV_CH_FRONT_LEFT){
207 if (matrix_encoding == AV_MATRIX_ENCODING_DOLBY ||
208 matrix_encoding == AV_MATRIX_ENCODING_DPLII) {
209 if (unaccounted & (AV_CH_BACK_LEFT | AV_CH_SIDE_LEFT)) {
210 matrix[FRONT_LEFT ][BACK_CENTER] -= surround_mix_level * M_SQRT1_2;
211 matrix[FRONT_RIGHT][BACK_CENTER] += surround_mix_level * M_SQRT1_2;
212 } else {
213 matrix[FRONT_LEFT ][BACK_CENTER] -= surround_mix_level;
214 matrix[FRONT_RIGHT][BACK_CENTER] += surround_mix_level;
215 }
216 } else {
217 matrix[ FRONT_LEFT][BACK_CENTER]+= surround_mix_level * M_SQRT1_2;
218 matrix[FRONT_RIGHT][BACK_CENTER]+= surround_mix_level * M_SQRT1_2;
219 }
220 }else if(out_ch_layout & AV_CH_FRONT_CENTER){
221 matrix[ FRONT_CENTER][BACK_CENTER]+= surround_mix_level * M_SQRT1_2;
222 }else
223 av_assert0(0);
224 }
225 if(unaccounted & AV_CH_BACK_LEFT){
226 if(out_ch_layout & AV_CH_BACK_CENTER){
227 matrix[BACK_CENTER][ BACK_LEFT]+= M_SQRT1_2;
228 matrix[BACK_CENTER][BACK_RIGHT]+= M_SQRT1_2;
229 }else if(out_ch_layout & AV_CH_SIDE_LEFT){
230 if(in_ch_layout & AV_CH_SIDE_LEFT){
231 matrix[ SIDE_LEFT][ BACK_LEFT]+= M_SQRT1_2;
232 matrix[SIDE_RIGHT][BACK_RIGHT]+= M_SQRT1_2;
233 }else{
234 matrix[ SIDE_LEFT][ BACK_LEFT]+= 1.0;
235 matrix[SIDE_RIGHT][BACK_RIGHT]+= 1.0;
236 }
237 }else if(out_ch_layout & AV_CH_FRONT_LEFT){
238 if (matrix_encoding == AV_MATRIX_ENCODING_DOLBY) {
239 matrix[FRONT_LEFT ][BACK_LEFT ] -= surround_mix_level * M_SQRT1_2;
240 matrix[FRONT_LEFT ][BACK_RIGHT] -= surround_mix_level * M_SQRT1_2;
241 matrix[FRONT_RIGHT][BACK_LEFT ] += surround_mix_level * M_SQRT1_2;
242 matrix[FRONT_RIGHT][BACK_RIGHT] += surround_mix_level * M_SQRT1_2;
243 } else if (matrix_encoding == AV_MATRIX_ENCODING_DPLII) {
244 matrix[FRONT_LEFT ][BACK_LEFT ] -= surround_mix_level * SQRT3_2;
245 matrix[FRONT_LEFT ][BACK_RIGHT] -= surround_mix_level * M_SQRT1_2;
246 matrix[FRONT_RIGHT][BACK_LEFT ] += surround_mix_level * M_SQRT1_2;
247 matrix[FRONT_RIGHT][BACK_RIGHT] += surround_mix_level * SQRT3_2;
248 } else {
249 matrix[ FRONT_LEFT][ BACK_LEFT] += surround_mix_level;
250 matrix[FRONT_RIGHT][BACK_RIGHT] += surround_mix_level;
251 }
252 }else if(out_ch_layout & AV_CH_FRONT_CENTER){
253 matrix[ FRONT_CENTER][BACK_LEFT ]+= surround_mix_level*M_SQRT1_2;
254 matrix[ FRONT_CENTER][BACK_RIGHT]+= surround_mix_level*M_SQRT1_2;
255 }else
256 av_assert0(0);
257 }
258
259 if(unaccounted & AV_CH_SIDE_LEFT){
260 if(out_ch_layout & AV_CH_BACK_LEFT){
261 /* if back channels do not exist in the input, just copy side
262 channels to back channels, otherwise mix side into back */
263 if (in_ch_layout & AV_CH_BACK_LEFT) {
264 matrix[BACK_LEFT ][SIDE_LEFT ] += M_SQRT1_2;
265 matrix[BACK_RIGHT][SIDE_RIGHT] += M_SQRT1_2;
266 } else {
267 matrix[BACK_LEFT ][SIDE_LEFT ] += 1.0;
268 matrix[BACK_RIGHT][SIDE_RIGHT] += 1.0;
269 }
270 }else if(out_ch_layout & AV_CH_BACK_CENTER){
271 matrix[BACK_CENTER][ SIDE_LEFT]+= M_SQRT1_2;
272 matrix[BACK_CENTER][SIDE_RIGHT]+= M_SQRT1_2;
273 }else if(out_ch_layout & AV_CH_FRONT_LEFT){
274 if (matrix_encoding == AV_MATRIX_ENCODING_DOLBY) {
275 matrix[FRONT_LEFT ][SIDE_LEFT ] -= surround_mix_level * M_SQRT1_2;
276 matrix[FRONT_LEFT ][SIDE_RIGHT] -= surround_mix_level * M_SQRT1_2;
277 matrix[FRONT_RIGHT][SIDE_LEFT ] += surround_mix_level * M_SQRT1_2;
278 matrix[FRONT_RIGHT][SIDE_RIGHT] += surround_mix_level * M_SQRT1_2;
279 } else if (matrix_encoding == AV_MATRIX_ENCODING_DPLII) {
280 matrix[FRONT_LEFT ][SIDE_LEFT ] -= surround_mix_level * SQRT3_2;
281 matrix[FRONT_LEFT ][SIDE_RIGHT] -= surround_mix_level * M_SQRT1_2;
282 matrix[FRONT_RIGHT][SIDE_LEFT ] += surround_mix_level * M_SQRT1_2;
283 matrix[FRONT_RIGHT][SIDE_RIGHT] += surround_mix_level * SQRT3_2;
284 } else {
285 matrix[ FRONT_LEFT][ SIDE_LEFT] += surround_mix_level;
286 matrix[FRONT_RIGHT][SIDE_RIGHT] += surround_mix_level;
287 }
288 }else if(out_ch_layout & AV_CH_FRONT_CENTER){
289 matrix[ FRONT_CENTER][SIDE_LEFT ]+= surround_mix_level * M_SQRT1_2;
290 matrix[ FRONT_CENTER][SIDE_RIGHT]+= surround_mix_level * M_SQRT1_2;
291 }else
292 av_assert0(0);
293 }
294
295 if(unaccounted & AV_CH_FRONT_LEFT_OF_CENTER){
296 if(out_ch_layout & AV_CH_FRONT_LEFT){
297 matrix[ FRONT_LEFT][ FRONT_LEFT_OF_CENTER]+= 1.0;
298 matrix[FRONT_RIGHT][FRONT_RIGHT_OF_CENTER]+= 1.0;
299 }else if(out_ch_layout & AV_CH_FRONT_CENTER){
300 matrix[ FRONT_CENTER][ FRONT_LEFT_OF_CENTER]+= M_SQRT1_2;
301 matrix[ FRONT_CENTER][FRONT_RIGHT_OF_CENTER]+= M_SQRT1_2;
302 }else
303 av_assert0(0);
304 }
305 /* mix LFE into front left/right or center */
306 if (unaccounted & AV_CH_LOW_FREQUENCY) {
307 if (out_ch_layout & AV_CH_FRONT_CENTER) {
308 matrix[FRONT_CENTER][LOW_FREQUENCY] += lfe_mix_level;
309 } else if (out_ch_layout & AV_CH_FRONT_LEFT) {
310 matrix[FRONT_LEFT ][LOW_FREQUENCY] += lfe_mix_level * M_SQRT1_2;
311 matrix[FRONT_RIGHT][LOW_FREQUENCY] += lfe_mix_level * M_SQRT1_2;
312 } else
313 av_assert0(0);
314 }
315
316 for(out_i=i=0; i<64; i++){
317 double sum=0;
318 int in_i=0;
319 if((out_ch_layout & (1ULL<<i)) == 0)
320 continue;
321 for(j=0; j<64; j++){
322 if((in_ch_layout & (1ULL<<j)) == 0)
323 continue;
324 if (i < FF_ARRAY_ELEMS(matrix) && j < FF_ARRAY_ELEMS(matrix[0]))
325 matrix_param[stride*out_i + in_i] = matrix[i][j];
326 else
327 matrix_param[stride*out_i + in_i] = i == j && (in_ch_layout & out_ch_layout & (1ULL<<i));
328 sum += fabs(matrix_param[stride*out_i + in_i]);
329 in_i++;
330 }
331 maxcoef= FFMAX(maxcoef, sum);
332 out_i++;
333 }
334 if(rematrix_volume < 0)
335 maxcoef = -rematrix_volume;
336
337 if(maxcoef > maxval || rematrix_volume < 0){
338 maxcoef /= maxval;
339 for(i=0; i<SWR_CH_MAX; i++)
340 for(j=0; j<SWR_CH_MAX; j++){
341 matrix_param[stride*i + j] /= maxcoef;
342 }
343 }
344
345 if(rematrix_volume > 0){
346 for(i=0; i<SWR_CH_MAX; i++)
347 for(j=0; j<SWR_CH_MAX; j++){
348 matrix_param[stride*i + j] *= rematrix_volume;
349 }
350 }
351
352 av_log(log_context, AV_LOG_DEBUG, "Matrix coefficients:\n");
353 for(i=0; i<av_get_channel_layout_nb_channels(out_ch_layout); i++){
354 const char *c =
355 av_get_channel_name(av_channel_layout_extract_channel(out_ch_layout, i));
356 av_log(log_context, AV_LOG_DEBUG, "%s: ", c ? c : "?");
357 for(j=0; j<av_get_channel_layout_nb_channels(in_ch_layout); j++){
358 c = av_get_channel_name(av_channel_layout_extract_channel(in_ch_layout, j));
359 av_log(log_context, AV_LOG_DEBUG, "%s:%f ", c ? c : "?", matrix_param[stride*i + j]);
360 }
361 av_log(log_context, AV_LOG_DEBUG, "\n");
362 }
363 return 0;
364 }
365
auto_matrix(SwrContext * s)366 av_cold static int auto_matrix(SwrContext *s)
367 {
368 double maxval;
369 int ret;
370
371 if (s->rematrix_maxval > 0) {
372 maxval = s->rematrix_maxval;
373 } else if ( av_get_packed_sample_fmt(s->out_sample_fmt) < AV_SAMPLE_FMT_FLT
374 || av_get_packed_sample_fmt(s->int_sample_fmt) < AV_SAMPLE_FMT_FLT) {
375 maxval = 1.0;
376 } else
377 maxval = INT_MAX;
378
379 memset(s->matrix, 0, sizeof(s->matrix));
380 ret = swr_build_matrix(s->in_ch_layout, s->out_ch_layout,
381 s->clev, s->slev, s->lfe_mix_level,
382 maxval, s->rematrix_volume, (double*)s->matrix,
383 s->matrix[1] - s->matrix[0], s->matrix_encoding, s);
384
385 if (ret >= 0 && s->int_sample_fmt == AV_SAMPLE_FMT_FLTP) {
386 int i, j;
387 for (i = 0; i < FF_ARRAY_ELEMS(s->matrix[0]); i++)
388 for (j = 0; j < FF_ARRAY_ELEMS(s->matrix[0]); j++)
389 s->matrix_flt[i][j] = s->matrix[i][j];
390 }
391
392 return ret;
393 }
394
swri_rematrix_init(SwrContext * s)395 av_cold int swri_rematrix_init(SwrContext *s){
396 int i, j;
397 int nb_in = s->used_ch_count;
398 int nb_out = s->out.ch_count;
399
400 s->mix_any_f = NULL;
401
402 if (!s->rematrix_custom) {
403 int r = auto_matrix(s);
404 if (r)
405 return r;
406 }
407 if (s->midbuf.fmt == AV_SAMPLE_FMT_S16P){
408 int maxsum = 0;
409 s->native_matrix = av_calloc(nb_in * nb_out, sizeof(int));
410 s->native_one = av_mallocz(sizeof(int));
411 if (!s->native_matrix || !s->native_one)
412 return AVERROR(ENOMEM);
413 for (i = 0; i < nb_out; i++) {
414 double rem = 0;
415 int sum = 0;
416
417 for (j = 0; j < nb_in; j++) {
418 double target = s->matrix[i][j] * 32768 + rem;
419 ((int*)s->native_matrix)[i * nb_in + j] = lrintf(target);
420 rem += target - ((int*)s->native_matrix)[i * nb_in + j];
421 sum += FFABS(((int*)s->native_matrix)[i * nb_in + j]);
422 }
423 maxsum = FFMAX(maxsum, sum);
424 }
425 *((int*)s->native_one) = 32768;
426 if (maxsum <= 32768) {
427 s->mix_1_1_f = (mix_1_1_func_type*)copy_s16;
428 s->mix_2_1_f = (mix_2_1_func_type*)sum2_s16;
429 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_s16(s);
430 } else {
431 s->mix_1_1_f = (mix_1_1_func_type*)copy_clip_s16;
432 s->mix_2_1_f = (mix_2_1_func_type*)sum2_clip_s16;
433 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_clip_s16(s);
434 }
435 }else if(s->midbuf.fmt == AV_SAMPLE_FMT_FLTP){
436 s->native_matrix = av_calloc(nb_in * nb_out, sizeof(float));
437 s->native_one = av_mallocz(sizeof(float));
438 if (!s->native_matrix || !s->native_one)
439 return AVERROR(ENOMEM);
440 for (i = 0; i < nb_out; i++)
441 for (j = 0; j < nb_in; j++)
442 ((float*)s->native_matrix)[i * nb_in + j] = s->matrix[i][j];
443 *((float*)s->native_one) = 1.0;
444 s->mix_1_1_f = (mix_1_1_func_type*)copy_float;
445 s->mix_2_1_f = (mix_2_1_func_type*)sum2_float;
446 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_float(s);
447 }else if(s->midbuf.fmt == AV_SAMPLE_FMT_DBLP){
448 s->native_matrix = av_calloc(nb_in * nb_out, sizeof(double));
449 s->native_one = av_mallocz(sizeof(double));
450 if (!s->native_matrix || !s->native_one)
451 return AVERROR(ENOMEM);
452 for (i = 0; i < nb_out; i++)
453 for (j = 0; j < nb_in; j++)
454 ((double*)s->native_matrix)[i * nb_in + j] = s->matrix[i][j];
455 *((double*)s->native_one) = 1.0;
456 s->mix_1_1_f = (mix_1_1_func_type*)copy_double;
457 s->mix_2_1_f = (mix_2_1_func_type*)sum2_double;
458 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_double(s);
459 }else if(s->midbuf.fmt == AV_SAMPLE_FMT_S32P){
460 s->native_one = av_mallocz(sizeof(int));
461 if (!s->native_one)
462 return AVERROR(ENOMEM);
463 s->native_matrix = av_calloc(nb_in * nb_out, sizeof(int));
464 if (!s->native_matrix) {
465 av_freep(&s->native_one);
466 return AVERROR(ENOMEM);
467 }
468 for (i = 0; i < nb_out; i++) {
469 double rem = 0;
470
471 for (j = 0; j < nb_in; j++) {
472 double target = s->matrix[i][j] * 32768 + rem;
473 ((int*)s->native_matrix)[i * nb_in + j] = lrintf(target);
474 rem += target - ((int*)s->native_matrix)[i * nb_in + j];
475 }
476 }
477 *((int*)s->native_one) = 32768;
478 s->mix_1_1_f = (mix_1_1_func_type*)copy_s32;
479 s->mix_2_1_f = (mix_2_1_func_type*)sum2_s32;
480 s->mix_any_f = (mix_any_func_type*)get_mix_any_func_s32(s);
481 }else
482 av_assert0(0);
483 //FIXME quantize for integeres
484 for (i = 0; i < SWR_CH_MAX; i++) {
485 int ch_in=0;
486 for (j = 0; j < SWR_CH_MAX; j++) {
487 s->matrix32[i][j]= lrintf(s->matrix[i][j] * 32768);
488 if(s->matrix[i][j])
489 s->matrix_ch[i][++ch_in]= j;
490 }
491 s->matrix_ch[i][0]= ch_in;
492 }
493
494 if(HAVE_X86ASM && HAVE_MMX)
495 return swri_rematrix_init_x86(s);
496
497 return 0;
498 }
499
swri_rematrix_free(SwrContext * s)500 av_cold void swri_rematrix_free(SwrContext *s){
501 av_freep(&s->native_matrix);
502 av_freep(&s->native_one);
503 av_freep(&s->native_simd_matrix);
504 av_freep(&s->native_simd_one);
505 }
506
swri_rematrix(SwrContext * s,AudioData * out,AudioData * in,int len,int mustcopy)507 int swri_rematrix(SwrContext *s, AudioData *out, AudioData *in, int len, int mustcopy){
508 int out_i, in_i, i, j;
509 int len1 = 0;
510 int off = 0;
511
512 if(s->mix_any_f) {
513 s->mix_any_f(out->ch, (const uint8_t **)in->ch, s->native_matrix, len);
514 return 0;
515 }
516
517 if(s->mix_2_1_simd || s->mix_1_1_simd){
518 len1= len&~15;
519 off = len1 * out->bps;
520 }
521
522 av_assert0(!s->out_ch_layout || out->ch_count == av_get_channel_layout_nb_channels(s->out_ch_layout));
523 av_assert0(!s-> in_ch_layout || in ->ch_count == av_get_channel_layout_nb_channels(s-> in_ch_layout));
524
525 for(out_i=0; out_i<out->ch_count; out_i++){
526 switch(s->matrix_ch[out_i][0]){
527 case 0:
528 if(mustcopy)
529 memset(out->ch[out_i], 0, len * av_get_bytes_per_sample(s->int_sample_fmt));
530 break;
531 case 1:
532 in_i= s->matrix_ch[out_i][1];
533 if(s->matrix[out_i][in_i]!=1.0){
534 if(s->mix_1_1_simd && len1)
535 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);
536 if(len != len1)
537 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);
538 }else if(mustcopy){
539 memcpy(out->ch[out_i], in->ch[in_i], len*out->bps);
540 }else{
541 out->ch[out_i]= in->ch[in_i];
542 }
543 break;
544 case 2: {
545 int in_i1 = s->matrix_ch[out_i][1];
546 int in_i2 = s->matrix_ch[out_i][2];
547 if(s->mix_2_1_simd && len1)
548 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);
549 else
550 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);
551 if(len != len1)
552 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);
553 break;}
554 default:
555 if(s->int_sample_fmt == AV_SAMPLE_FMT_FLTP){
556 for(i=0; i<len; i++){
557 float v=0;
558 for(j=0; j<s->matrix_ch[out_i][0]; j++){
559 in_i= s->matrix_ch[out_i][1+j];
560 v+= ((float*)in->ch[in_i])[i] * s->matrix_flt[out_i][in_i];
561 }
562 ((float*)out->ch[out_i])[i]= v;
563 }
564 }else if(s->int_sample_fmt == AV_SAMPLE_FMT_DBLP){
565 for(i=0; i<len; i++){
566 double v=0;
567 for(j=0; j<s->matrix_ch[out_i][0]; j++){
568 in_i= s->matrix_ch[out_i][1+j];
569 v+= ((double*)in->ch[in_i])[i] * s->matrix[out_i][in_i];
570 }
571 ((double*)out->ch[out_i])[i]= v;
572 }
573 }else{
574 for(i=0; i<len; i++){
575 int v=0;
576 for(j=0; j<s->matrix_ch[out_i][0]; j++){
577 in_i= s->matrix_ch[out_i][1+j];
578 v+= ((int16_t*)in->ch[in_i])[i] * s->matrix32[out_i][in_i];
579 }
580 ((int16_t*)out->ch[out_i])[i]= (v + 16384)>>15;
581 }
582 }
583 }
584 }
585 return 0;
586 }
587