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