4c971dc8e0d81940d08e3d28c1f07a16efcf888a
[ffmpeg.git] / libswresample / swresample.c
1 /*
2  * Copyright (C) 2011-2013 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 "libavutil/opt.h"
22 #include "swresample_internal.h"
23 #include "audioconvert.h"
24 #include "libavutil/avassert.h"
25 #include "libavutil/channel_layout.h"
26
27 #include <float.h>
28
29 #define  C30DB  M_SQRT2
30 #define  C15DB  1.189207115
31 #define C__0DB  1.0
32 #define C_15DB  0.840896415
33 #define C_30DB  M_SQRT1_2
34 #define C_45DB  0.594603558
35 #define C_60DB  0.5
36
37 #define ALIGN 32
38
39 //TODO split options array out?
40 #define OFFSET(x) offsetof(SwrContext,x)
41 #define PARAM AV_OPT_FLAG_AUDIO_PARAM
42
43 static const AVOption options[]={
44 {"ich"                  , "set input channel count"     , OFFSET( in.ch_count   ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_CH_MAX, PARAM},
45 {"in_channel_count"     , "set input channel count"     , OFFSET( in.ch_count   ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_CH_MAX, PARAM},
46 {"och"                  , "set output channel count"    , OFFSET(out.ch_count   ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_CH_MAX, PARAM},
47 {"out_channel_count"    , "set output channel count"    , OFFSET(out.ch_count   ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_CH_MAX, PARAM},
48 {"uch"                  , "set used channel count"      , OFFSET(used_ch_count  ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_CH_MAX, PARAM},
49 {"used_channel_count"   , "set used channel count"      , OFFSET(used_ch_count  ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_CH_MAX, PARAM},
50 {"isr"                  , "set input sample rate"       , OFFSET( in_sample_rate), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , INT_MAX   , PARAM},
51 {"in_sample_rate"       , "set input sample rate"       , OFFSET( in_sample_rate), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , INT_MAX   , PARAM},
52 {"osr"                  , "set output sample rate"      , OFFSET(out_sample_rate), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , INT_MAX   , PARAM},
53 {"out_sample_rate"      , "set output sample rate"      , OFFSET(out_sample_rate), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , INT_MAX   , PARAM},
54 {"isf"                  , "set input sample format"     , OFFSET( in_sample_fmt ), AV_OPT_TYPE_SAMPLE_FMT , {.i64=AV_SAMPLE_FMT_NONE}, -1   , AV_SAMPLE_FMT_NB-1, PARAM},
55 {"in_sample_fmt"        , "set input sample format"     , OFFSET( in_sample_fmt ), AV_OPT_TYPE_SAMPLE_FMT , {.i64=AV_SAMPLE_FMT_NONE}, -1   , AV_SAMPLE_FMT_NB-1, PARAM},
56 {"osf"                  , "set output sample format"    , OFFSET(out_sample_fmt ), AV_OPT_TYPE_SAMPLE_FMT , {.i64=AV_SAMPLE_FMT_NONE}, -1   , AV_SAMPLE_FMT_NB-1, PARAM},
57 {"out_sample_fmt"       , "set output sample format"    , OFFSET(out_sample_fmt ), AV_OPT_TYPE_SAMPLE_FMT , {.i64=AV_SAMPLE_FMT_NONE}, -1   , AV_SAMPLE_FMT_NB-1, PARAM},
58 {"tsf"                  , "set internal sample format"  , OFFSET(int_sample_fmt ), AV_OPT_TYPE_SAMPLE_FMT , {.i64=AV_SAMPLE_FMT_NONE}, -1   , AV_SAMPLE_FMT_NB-1, PARAM},
59 {"internal_sample_fmt"  , "set internal sample format"  , OFFSET(int_sample_fmt ), AV_OPT_TYPE_SAMPLE_FMT , {.i64=AV_SAMPLE_FMT_NONE}, -1   , AV_SAMPLE_FMT_NB-1, PARAM},
60 {"icl"                  , "set input channel layout"    , OFFSET( in_ch_layout  ), AV_OPT_TYPE_INT64, {.i64=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
61 {"in_channel_layout"    , "set input channel layout"    , OFFSET( in_ch_layout  ), AV_OPT_TYPE_INT64, {.i64=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
62 {"ocl"                  , "set output channel layout"   , OFFSET(out_ch_layout  ), AV_OPT_TYPE_INT64, {.i64=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
63 {"out_channel_layout"   , "set output channel layout"   , OFFSET(out_ch_layout  ), AV_OPT_TYPE_INT64, {.i64=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
64 {"clev"                 , "set center mix level"        , OFFSET(clev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
65 {"center_mix_level"     , "set center mix level"        , OFFSET(clev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
66 {"slev"                 , "set surround mix level"      , OFFSET(slev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
67 {"surround_mix_level"   , "set surround mix Level"      , OFFSET(slev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
68 {"lfe_mix_level"        , "set LFE mix level"           , OFFSET(lfe_mix_level  ), AV_OPT_TYPE_FLOAT, {.dbl=0                     }, -32    , 32        , PARAM},
69 {"rmvol"                , "set rematrix volume"         , OFFSET(rematrix_volume), AV_OPT_TYPE_FLOAT, {.dbl=1.0                   }, -1000  , 1000      , PARAM},
70 {"rematrix_volume"      , "set rematrix volume"         , OFFSET(rematrix_volume), AV_OPT_TYPE_FLOAT, {.dbl=1.0                   }, -1000  , 1000      , PARAM},
71
72 {"flags"                , "set flags"                   , OFFSET(flags          ), AV_OPT_TYPE_FLAGS, {.i64=0                     }, 0      , UINT_MAX  , PARAM, "flags"},
73 {"swr_flags"            , "set flags"                   , OFFSET(flags          ), AV_OPT_TYPE_FLAGS, {.i64=0                     }, 0      , UINT_MAX  , PARAM, "flags"},
74 {"res"                  , "force resampling"            , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_FLAG_RESAMPLE     }, INT_MIN, INT_MAX   , PARAM, "flags"},
75
76 {"dither_scale"         , "set dither scale"            , OFFSET(dither.scale   ), AV_OPT_TYPE_FLOAT, {.dbl=1                     }, 0      , INT_MAX   , PARAM},
77
78 {"dither_method"        , "set dither method"           , OFFSET(dither.method  ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_DITHER_NB-1, PARAM, "dither_method"},
79 {"rectangular"          , "select rectangular dither"   , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_RECTANGULAR}, INT_MIN, INT_MAX   , PARAM, "dither_method"},
80 {"triangular"           , "select triangular dither"    , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_TRIANGULAR }, INT_MIN, INT_MAX   , PARAM, "dither_method"},
81 {"triangular_hp"        , "select triangular dither with high pass" , 0          , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_TRIANGULAR_HIGHPASS }, INT_MIN, INT_MAX, PARAM, "dither_method"},
82 {"lipshitz"             , "select lipshitz noise shaping dither" , 0             , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_NS_LIPSHITZ}, INT_MIN, INT_MAX, PARAM, "dither_method"},
83 {"shibata"              , "select shibata noise shaping dither" , 0              , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_NS_SHIBATA }, INT_MIN, INT_MAX, PARAM, "dither_method"},
84 {"low_shibata"          , "select low shibata noise shaping dither" , 0          , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_NS_LOW_SHIBATA }, INT_MIN, INT_MAX, PARAM, "dither_method"},
85 {"high_shibata"         , "select high shibata noise shaping dither" , 0         , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_NS_HIGH_SHIBATA }, INT_MIN, INT_MAX, PARAM, "dither_method"},
86 {"f_weighted"           , "select f-weighted noise shaping dither" , 0           , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_NS_F_WEIGHTED }, INT_MIN, INT_MAX, PARAM, "dither_method"},
87 {"modified_e_weighted"  , "select modified-e-weighted noise shaping dither" , 0  , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_NS_MODIFIED_E_WEIGHTED }, INT_MIN, INT_MAX, PARAM, "dither_method"},
88 {"improved_e_weighted"  , "select improved-e-weighted noise shaping dither" , 0  , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_NS_IMPROVED_E_WEIGHTED }, INT_MIN, INT_MAX, PARAM, "dither_method"},
89
90 {"filter_size"          , "set swr resampling filter size", OFFSET(filter_size)  , AV_OPT_TYPE_INT  , {.i64=32                    }, 0      , INT_MAX   , PARAM },
91 {"phase_shift"          , "set swr resampling phase shift", OFFSET(phase_shift)  , AV_OPT_TYPE_INT  , {.i64=10                    }, 0      , 30        , PARAM },
92 {"linear_interp"        , "enable linear interpolation" , OFFSET(linear_interp)  , AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , 1         , PARAM },
93 {"cutoff"               , "set cutoff frequency ratio"  , OFFSET(cutoff)         , AV_OPT_TYPE_DOUBLE,{.dbl=0.                    }, 0      , 1         , PARAM },
94 {"resampler"            , "set resampling Engine"       , OFFSET(engine)         , AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_ENGINE_NB-1, PARAM, "resampler"},
95 {"swr"                  , "select SW Resampler"         , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_ENGINE_SWR        }, INT_MIN, INT_MAX   , PARAM, "resampler"},
96 {"soxr"                 , "select SoX Resampler"        , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_ENGINE_SOXR       }, INT_MIN, INT_MAX   , PARAM, "resampler"},
97 {"precision"            , "set soxr resampling precision (in bits)"
98                                                         , OFFSET(precision)      , AV_OPT_TYPE_DOUBLE,{.dbl=20.0                  }, 15.0   , 33.0      , PARAM },
99 {"cheby"                , "enable soxr Chebyshev passband & higher-precision irrational ratio approximation"
100                                                         , OFFSET(cheby)          , AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , 1         , PARAM },
101 {"min_comp"             , "set minimum difference between timestamps and audio data (in seconds) below which no timestamp compensation of either kind is applied"
102                                                         , OFFSET(min_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=FLT_MAX               }, 0      , FLT_MAX   , PARAM },
103 {"min_hard_comp"        , "set minimum difference between timestamps and audio data (in seconds) to trigger padding/trimming the data."
104                                                         , OFFSET(min_hard_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=0.1                   }, 0      , INT_MAX   , PARAM },
105 {"comp_duration"        , "set duration (in seconds) over which data is stretched/squeezed to make it match the timestamps."
106                                                         , OFFSET(soft_compensation_duration),AV_OPT_TYPE_FLOAT ,{.dbl=1                     }, 0      , INT_MAX   , PARAM },
107 {"max_soft_comp"        , "set maximum factor by which data is stretched/squeezed to make it match the timestamps."
108                                                         , OFFSET(max_soft_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=0                     }, INT_MIN, INT_MAX   , PARAM },
109 {"async"                , "simplified 1 parameter audio timestamp matching, 0(disabled), 1(filling and trimming), >1(maximum stretch/squeeze in samples per second)"
110                                                         , OFFSET(async)          , AV_OPT_TYPE_FLOAT ,{.dbl=0                     }, INT_MIN, INT_MAX   , PARAM },
111
112 { "matrix_encoding"     , "set matrixed stereo encoding" , OFFSET(matrix_encoding), AV_OPT_TYPE_INT   ,{.i64 = AV_MATRIX_ENCODING_NONE}, AV_MATRIX_ENCODING_NONE,     AV_MATRIX_ENCODING_NB-1, PARAM, "matrix_encoding" },
113     { "none",  "select none",               0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_NONE  }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
114     { "dolby", "select Dolby",              0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_DOLBY }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
115     { "dplii", "select Dolby Pro Logic II", 0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_DPLII }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
116
117 { "filter_type"         , "select swr filter type"      , OFFSET(filter_type)    , AV_OPT_TYPE_INT  , { .i64 = SWR_FILTER_TYPE_KAISER }, SWR_FILTER_TYPE_CUBIC, SWR_FILTER_TYPE_KAISER, PARAM, "filter_type" },
118     { "cubic"           , "select cubic"                , 0                      , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_CUBIC            }, INT_MIN, INT_MAX, PARAM, "filter_type" },
119     { "blackman_nuttall", "select Blackman Nuttall Windowed Sinc", 0             , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_BLACKMAN_NUTTALL }, INT_MIN, INT_MAX, PARAM, "filter_type" },
120     { "kaiser"          , "select Kaiser Windowed Sinc" , 0                      , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_KAISER           }, INT_MIN, INT_MAX, PARAM, "filter_type" },
121
122 { "kaiser_beta"         , "set swr Kaiser Window Beta"  , OFFSET(kaiser_beta)    , AV_OPT_TYPE_INT  , {.i64=9                     }, 2      , 16        , PARAM },
123
124 {0}
125 };
126
127 static const char* context_to_name(void* ptr) {
128     return "SWR";
129 }
130
131 static const AVClass av_class = {
132     .class_name                = "SWResampler",
133     .item_name                 = context_to_name,
134     .option                    = options,
135     .version                   = LIBAVUTIL_VERSION_INT,
136     .log_level_offset_offset   = OFFSET(log_level_offset),
137     .parent_log_context_offset = OFFSET(log_ctx),
138     .category                  = AV_CLASS_CATEGORY_SWRESAMPLER,
139 };
140
141 unsigned swresample_version(void)
142 {
143     av_assert0(LIBSWRESAMPLE_VERSION_MICRO >= 100);
144     return LIBSWRESAMPLE_VERSION_INT;
145 }
146
147 const char *swresample_configuration(void)
148 {
149     return FFMPEG_CONFIGURATION;
150 }
151
152 const char *swresample_license(void)
153 {
154 #define LICENSE_PREFIX "libswresample license: "
155     return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
156 }
157
158 int swr_set_channel_mapping(struct SwrContext *s, const int *channel_map){
159     if(!s || s->in_convert) // s needs to be allocated but not initialized
160         return AVERROR(EINVAL);
161     s->channel_map = channel_map;
162     return 0;
163 }
164
165 const AVClass *swr_get_class(void)
166 {
167     return &av_class;
168 }
169
170 av_cold struct SwrContext *swr_alloc(void){
171     SwrContext *s= av_mallocz(sizeof(SwrContext));
172     if(s){
173         s->av_class= &av_class;
174         av_opt_set_defaults(s);
175     }
176     return s;
177 }
178
179 struct SwrContext *swr_alloc_set_opts(struct SwrContext *s,
180                                       int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
181                                       int64_t  in_ch_layout, enum AVSampleFormat  in_sample_fmt, int  in_sample_rate,
182                                       int log_offset, void *log_ctx){
183     if(!s) s= swr_alloc();
184     if(!s) return NULL;
185
186     s->log_level_offset= log_offset;
187     s->log_ctx= log_ctx;
188
189     av_opt_set_int(s, "ocl", out_ch_layout,   0);
190     av_opt_set_int(s, "osf", out_sample_fmt,  0);
191     av_opt_set_int(s, "osr", out_sample_rate, 0);
192     av_opt_set_int(s, "icl", in_ch_layout,    0);
193     av_opt_set_int(s, "isf", in_sample_fmt,   0);
194     av_opt_set_int(s, "isr", in_sample_rate,  0);
195     av_opt_set_int(s, "tsf", AV_SAMPLE_FMT_NONE,   0);
196     av_opt_set_int(s, "ich", av_get_channel_layout_nb_channels(s-> in_ch_layout), 0);
197     av_opt_set_int(s, "och", av_get_channel_layout_nb_channels(s->out_ch_layout), 0);
198     av_opt_set_int(s, "uch", 0, 0);
199     return s;
200 }
201
202 static void set_audiodata_fmt(AudioData *a, enum AVSampleFormat fmt){
203     a->fmt   = fmt;
204     a->bps   = av_get_bytes_per_sample(fmt);
205     a->planar= av_sample_fmt_is_planar(fmt);
206 }
207
208 static void free_temp(AudioData *a){
209     av_free(a->data);
210     memset(a, 0, sizeof(*a));
211 }
212
213 av_cold void swr_free(SwrContext **ss){
214     SwrContext *s= *ss;
215     if(s){
216         free_temp(&s->postin);
217         free_temp(&s->midbuf);
218         free_temp(&s->preout);
219         free_temp(&s->in_buffer);
220         free_temp(&s->dither.noise);
221         free_temp(&s->dither.temp);
222         swri_audio_convert_free(&s-> in_convert);
223         swri_audio_convert_free(&s->out_convert);
224         swri_audio_convert_free(&s->full_convert);
225         if (s->resampler)
226             s->resampler->free(&s->resample);
227         swri_rematrix_free(s);
228     }
229
230     av_freep(ss);
231 }
232
233 av_cold int swr_init(struct SwrContext *s){
234     int ret;
235     s->in_buffer_index= 0;
236     s->in_buffer_count= 0;
237     s->resample_in_constraint= 0;
238     free_temp(&s->postin);
239     free_temp(&s->midbuf);
240     free_temp(&s->preout);
241     free_temp(&s->in_buffer);
242     free_temp(&s->dither.noise);
243     free_temp(&s->dither.temp);
244     memset(s->in.ch, 0, sizeof(s->in.ch));
245     memset(s->out.ch, 0, sizeof(s->out.ch));
246     swri_audio_convert_free(&s-> in_convert);
247     swri_audio_convert_free(&s->out_convert);
248     swri_audio_convert_free(&s->full_convert);
249     swri_rematrix_free(s);
250
251     s->flushed = 0;
252
253     if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
254         av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
255         return AVERROR(EINVAL);
256     }
257     if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
258         av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
259         return AVERROR(EINVAL);
260     }
261
262     if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
263         if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_S16P){
264             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
265         }else if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_FLTP){
266             s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
267         }else{
268             av_log(s, AV_LOG_DEBUG, "Using double precision mode\n");
269             s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
270         }
271     }
272
273     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16P
274         &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
275         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
276         &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
277         av_log(s, AV_LOG_ERROR, "Requested sample format %s is not supported internally, S16/S32/FLT/DBL is supported\n", av_get_sample_fmt_name(s->int_sample_fmt));
278         return AVERROR(EINVAL);
279     }
280
281     switch(s->engine){
282 #if CONFIG_LIBSOXR
283         extern struct Resampler const soxr_resampler;
284         case SWR_ENGINE_SOXR: s->resampler = &soxr_resampler; break;
285 #endif
286         case SWR_ENGINE_SWR : s->resampler = &swri_resampler; break;
287         default:
288             av_log(s, AV_LOG_ERROR, "Requested resampling engine is unavailable\n");
289             return AVERROR(EINVAL);
290     }
291
292     set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
293     set_audiodata_fmt(&s->out, s->out_sample_fmt);
294
295     if (s->async) {
296         if (s->min_compensation >= FLT_MAX/2)
297             s->min_compensation = 0.001;
298         if (s->async > 1.0001) {
299             s->max_soft_compensation = s->async / (double) s->in_sample_rate;
300         }
301     }
302
303     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
304         s->resample = s->resampler->init(s->resample, s->out_sample_rate, s->in_sample_rate, s->filter_size, s->phase_shift, s->linear_interp, s->cutoff, s->int_sample_fmt, s->filter_type, s->kaiser_beta, s->precision, s->cheby);
305     }else
306         s->resampler->free(&s->resample);
307     if(    s->int_sample_fmt != AV_SAMPLE_FMT_S16P
308         && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
309         && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
310         && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
311         && s->resample){
312         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16/s32/flt/dbl\n");
313         return -1;
314     }
315
316     if(!s->used_ch_count)
317         s->used_ch_count= s->in.ch_count;
318
319     if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
320         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
321         s-> in_ch_layout= 0;
322     }
323
324     if(!s-> in_ch_layout)
325         s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
326     if(!s->out_ch_layout)
327         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
328
329     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
330                  s->rematrix_custom;
331
332 #define RSC 1 //FIXME finetune
333     if(!s-> in.ch_count)
334         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
335     if(!s->used_ch_count)
336         s->used_ch_count= s->in.ch_count;
337     if(!s->out.ch_count)
338         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
339
340     if(!s-> in.ch_count){
341         av_assert0(!s->in_ch_layout);
342         av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
343         return -1;
344     }
345
346     if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
347         char l1[1024], l2[1024];
348         av_get_channel_layout_string(l1, sizeof(l1), s-> in.ch_count, s-> in_ch_layout);
349         av_get_channel_layout_string(l2, sizeof(l2), s->out.ch_count, s->out_ch_layout);
350         av_log(s, AV_LOG_ERROR, "Rematrix is needed between %s and %s "
351                "but there is not enough information to do it\n", l1, l2);
352         return -1;
353     }
354
355 av_assert0(s->used_ch_count);
356 av_assert0(s->out.ch_count);
357     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
358
359     s->in_buffer= s->in;
360
361     if(!s->resample && !s->rematrix && !s->channel_map && !s->dither.method){
362         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
363                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
364         return 0;
365     }
366
367     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
368                                              s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
369     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
370                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
371
372
373     s->postin= s->in;
374     s->preout= s->out;
375     s->midbuf= s->in;
376
377     if(s->channel_map){
378         s->postin.ch_count=
379         s->midbuf.ch_count= s->used_ch_count;
380         if(s->resample)
381             s->in_buffer.ch_count= s->used_ch_count;
382     }
383     if(!s->resample_first){
384         s->midbuf.ch_count= s->out.ch_count;
385         if(s->resample)
386             s->in_buffer.ch_count = s->out.ch_count;
387     }
388
389     set_audiodata_fmt(&s->postin, s->int_sample_fmt);
390     set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
391     set_audiodata_fmt(&s->preout, s->int_sample_fmt);
392
393     if(s->resample){
394         set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
395     }
396
397     if ((ret = swri_dither_init(s, s->out_sample_fmt, s->int_sample_fmt)) < 0)
398         return ret;
399
400     if(s->rematrix || s->dither.method)
401         return swri_rematrix_init(s);
402
403     return 0;
404 }
405
406 int swri_realloc_audio(AudioData *a, int count){
407     int i, countb;
408     AudioData old;
409
410     if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
411         return AVERROR(EINVAL);
412
413     if(a->count >= count)
414         return 0;
415
416     count*=2;
417
418     countb= FFALIGN(count*a->bps, ALIGN);
419     old= *a;
420
421     av_assert0(a->bps);
422     av_assert0(a->ch_count);
423
424     a->data= av_mallocz(countb*a->ch_count);
425     if(!a->data)
426         return AVERROR(ENOMEM);
427     for(i=0; i<a->ch_count; i++){
428         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
429         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
430     }
431     if(!a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
432     av_free(old.data);
433     a->count= count;
434
435     return 1;
436 }
437
438 static void copy(AudioData *out, AudioData *in,
439                  int count){
440     av_assert0(out->planar == in->planar);
441     av_assert0(out->bps == in->bps);
442     av_assert0(out->ch_count == in->ch_count);
443     if(out->planar){
444         int ch;
445         for(ch=0; ch<out->ch_count; ch++)
446             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
447     }else
448         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
449 }
450
451 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
452     int i;
453     if(!in_arg){
454         memset(out->ch, 0, sizeof(out->ch));
455     }else if(out->planar){
456         for(i=0; i<out->ch_count; i++)
457             out->ch[i]= in_arg[i];
458     }else{
459         for(i=0; i<out->ch_count; i++)
460             out->ch[i]= in_arg[0] + i*out->bps;
461     }
462 }
463
464 static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
465     int i;
466     if(out->planar){
467         for(i=0; i<out->ch_count; i++)
468             in_arg[i]= out->ch[i];
469     }else{
470         in_arg[0]= out->ch[0];
471     }
472 }
473
474 /**
475  *
476  * out may be equal in.
477  */
478 static void buf_set(AudioData *out, AudioData *in, int count){
479     int ch;
480     if(in->planar){
481         for(ch=0; ch<out->ch_count; ch++)
482             out->ch[ch]= in->ch[ch] + count*out->bps;
483     }else{
484         for(ch=out->ch_count-1; ch>=0; ch--)
485             out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
486     }
487 }
488
489 /**
490  *
491  * @return number of samples output per channel
492  */
493 static int resample(SwrContext *s, AudioData *out_param, int out_count,
494                              const AudioData * in_param, int in_count){
495     AudioData in, out, tmp;
496     int ret_sum=0;
497     int border=0;
498
499     av_assert1(s->in_buffer.ch_count == in_param->ch_count);
500     av_assert1(s->in_buffer.planar   == in_param->planar);
501     av_assert1(s->in_buffer.fmt      == in_param->fmt);
502
503     tmp=out=*out_param;
504     in =  *in_param;
505
506     do{
507         int ret, size, consumed;
508         if(!s->resample_in_constraint && s->in_buffer_count){
509             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
510             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
511             out_count -= ret;
512             ret_sum += ret;
513             buf_set(&out, &out, ret);
514             s->in_buffer_count -= consumed;
515             s->in_buffer_index += consumed;
516
517             if(!in_count)
518                 break;
519             if(s->in_buffer_count <= border){
520                 buf_set(&in, &in, -s->in_buffer_count);
521                 in_count += s->in_buffer_count;
522                 s->in_buffer_count=0;
523                 s->in_buffer_index=0;
524                 border = 0;
525             }
526         }
527
528         if((s->flushed || in_count) && !s->in_buffer_count){
529             s->in_buffer_index=0;
530             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &in, in_count, &consumed);
531             out_count -= ret;
532             ret_sum += ret;
533             buf_set(&out, &out, ret);
534             in_count -= consumed;
535             buf_set(&in, &in, consumed);
536         }
537
538         //TODO is this check sane considering the advanced copy avoidance below
539         size= s->in_buffer_index + s->in_buffer_count + in_count;
540         if(   size > s->in_buffer.count
541            && s->in_buffer_count + in_count <= s->in_buffer_index){
542             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
543             copy(&s->in_buffer, &tmp, s->in_buffer_count);
544             s->in_buffer_index=0;
545         }else
546             if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
547                 return ret;
548
549         if(in_count){
550             int count= in_count;
551             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
552
553             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
554             copy(&tmp, &in, /*in_*/count);
555             s->in_buffer_count += count;
556             in_count -= count;
557             border += count;
558             buf_set(&in, &in, count);
559             s->resample_in_constraint= 0;
560             if(s->in_buffer_count != count || in_count)
561                 continue;
562         }
563         break;
564     }while(1);
565
566     s->resample_in_constraint= !!out_count;
567
568     return ret_sum;
569 }
570
571 static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
572                                                       AudioData *in , int  in_count){
573     AudioData *postin, *midbuf, *preout;
574     int ret/*, in_max*/;
575     AudioData preout_tmp, midbuf_tmp;
576
577     if(s->full_convert){
578         av_assert0(!s->resample);
579         swri_audio_convert(s->full_convert, out, in, in_count);
580         return out_count;
581     }
582
583 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
584 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
585
586     if((ret=swri_realloc_audio(&s->postin, in_count))<0)
587         return ret;
588     if(s->resample_first){
589         av_assert0(s->midbuf.ch_count == s->used_ch_count);
590         if((ret=swri_realloc_audio(&s->midbuf, out_count))<0)
591             return ret;
592     }else{
593         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
594         if((ret=swri_realloc_audio(&s->midbuf,  in_count))<0)
595             return ret;
596     }
597     if((ret=swri_realloc_audio(&s->preout, out_count))<0)
598         return ret;
599
600     postin= &s->postin;
601
602     midbuf_tmp= s->midbuf;
603     midbuf= &midbuf_tmp;
604     preout_tmp= s->preout;
605     preout= &preout_tmp;
606
607     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
608         postin= in;
609
610     if(s->resample_first ? !s->resample : !s->rematrix)
611         midbuf= postin;
612
613     if(s->resample_first ? !s->rematrix : !s->resample)
614         preout= midbuf;
615
616     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar){
617         if(preout==in){
618             out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
619             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
620             copy(out, in, out_count);
621             return out_count;
622         }
623         else if(preout==postin) preout= midbuf= postin= out;
624         else if(preout==midbuf) preout= midbuf= out;
625         else                    preout= out;
626     }
627
628     if(in != postin){
629         swri_audio_convert(s->in_convert, postin, in, in_count);
630     }
631
632     if(s->resample_first){
633         if(postin != midbuf)
634             out_count= resample(s, midbuf, out_count, postin, in_count);
635         if(midbuf != preout)
636             swri_rematrix(s, preout, midbuf, out_count, preout==out);
637     }else{
638         if(postin != midbuf)
639             swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
640         if(midbuf != preout)
641             out_count= resample(s, preout, out_count, midbuf, in_count);
642     }
643
644     if(preout != out && out_count){
645         AudioData *conv_src = preout;
646         if(s->dither.method){
647             int ch;
648             int dither_count= FFMAX(out_count, 1<<16);
649
650             if (preout == in) {
651                 conv_src = &s->dither.temp;
652                 if((ret=swri_realloc_audio(&s->dither.temp, dither_count))<0)
653                     return ret;
654             }
655
656             if((ret=swri_realloc_audio(&s->dither.noise, dither_count))<0)
657                 return ret;
658             if(ret)
659                 for(ch=0; ch<s->dither.noise.ch_count; ch++)
660                     swri_get_dither(s, s->dither.noise.ch[ch], s->dither.noise.count, 12345678913579<<ch, s->dither.noise.fmt);
661             av_assert0(s->dither.noise.ch_count == preout->ch_count);
662
663             if(s->dither.noise_pos + out_count > s->dither.noise.count)
664                 s->dither.noise_pos = 0;
665
666             if (s->dither.method < SWR_DITHER_NS){
667                 if (s->mix_2_1_simd) {
668                     int len1= out_count&~15;
669                     int off = len1 * preout->bps;
670
671                     if(len1)
672                         for(ch=0; ch<preout->ch_count; ch++)
673                             s->mix_2_1_simd(conv_src->ch[ch], preout->ch[ch], s->dither.noise.ch[ch] + s->dither.noise.bps * s->dither.noise_pos, s->native_one, 0, 0, len1);
674                     if(out_count != len1)
675                         for(ch=0; ch<preout->ch_count; ch++)
676                             s->mix_2_1_f(conv_src->ch[ch] + off, preout->ch[ch] + off, s->dither.noise.ch[ch] + s->dither.noise.bps * s->dither.noise_pos + off + len1, s->native_one, 0, 0, out_count - len1);
677                 } else {
678                     for(ch=0; ch<preout->ch_count; ch++)
679                         s->mix_2_1_f(conv_src->ch[ch], preout->ch[ch], s->dither.noise.ch[ch] + s->dither.noise.bps * s->dither.noise_pos, s->native_one, 0, 0, out_count);
680                 }
681             } else {
682                 switch(s->int_sample_fmt) {
683                 case AV_SAMPLE_FMT_S16P :swri_noise_shaping_int16(s, conv_src, preout, &s->dither.noise, out_count); break;
684                 case AV_SAMPLE_FMT_S32P :swri_noise_shaping_int32(s, conv_src, preout, &s->dither.noise, out_count); break;
685                 case AV_SAMPLE_FMT_FLTP :swri_noise_shaping_float(s, conv_src, preout, &s->dither.noise, out_count); break;
686                 case AV_SAMPLE_FMT_DBLP :swri_noise_shaping_double(s,conv_src, preout, &s->dither.noise, out_count); break;
687                 }
688             }
689             s->dither.noise_pos += out_count;
690         }
691 //FIXME packed doesnt need more than 1 chan here!
692         swri_audio_convert(s->out_convert, out, conv_src, out_count);
693     }
694     return out_count;
695 }
696
697 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
698                                 const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
699     AudioData * in= &s->in;
700     AudioData *out= &s->out;
701
702     if(s->drop_output > 0){
703         int ret;
704         AudioData tmp = s->out;
705         uint8_t *tmp_arg[SWR_CH_MAX];
706         tmp.count = 0;
707         tmp.data  = NULL;
708         if((ret=swri_realloc_audio(&tmp, s->drop_output))<0)
709             return ret;
710
711         reversefill_audiodata(&tmp, tmp_arg);
712         s->drop_output *= -1; //FIXME find a less hackish solution
713         ret = swr_convert(s, tmp_arg, -s->drop_output, in_arg, in_count); //FIXME optimize but this is as good as never called so maybe it doesnt matter
714         s->drop_output *= -1;
715         if(ret>0)
716             s->drop_output -= ret;
717
718         av_freep(&tmp.data);
719         if(s->drop_output || !out_arg)
720             return 0;
721         in_count = 0;
722     }
723
724     if(!in_arg){
725         if(s->resample){
726             if (!s->flushed)
727                 s->resampler->flush(s);
728             s->resample_in_constraint = 0;
729             s->flushed = 1;
730         }else if(!s->in_buffer_count){
731             return 0;
732         }
733     }else
734         fill_audiodata(in ,  (void*)in_arg);
735
736     fill_audiodata(out, out_arg);
737
738     if(s->resample){
739         int ret = swr_convert_internal(s, out, out_count, in, in_count);
740         if(ret>0 && !s->drop_output)
741             s->outpts += ret * (int64_t)s->in_sample_rate;
742         return ret;
743     }else{
744         AudioData tmp= *in;
745         int ret2=0;
746         int ret, size;
747         size = FFMIN(out_count, s->in_buffer_count);
748         if(size){
749             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
750             ret= swr_convert_internal(s, out, size, &tmp, size);
751             if(ret<0)
752                 return ret;
753             ret2= ret;
754             s->in_buffer_count -= ret;
755             s->in_buffer_index += ret;
756             buf_set(out, out, ret);
757             out_count -= ret;
758             if(!s->in_buffer_count)
759                 s->in_buffer_index = 0;
760         }
761
762         if(in_count){
763             size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
764
765             if(in_count > out_count) { //FIXME move after swr_convert_internal
766                 if(   size > s->in_buffer.count
767                 && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
768                     buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
769                     copy(&s->in_buffer, &tmp, s->in_buffer_count);
770                     s->in_buffer_index=0;
771                 }else
772                     if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
773                         return ret;
774             }
775
776             if(out_count){
777                 size = FFMIN(in_count, out_count);
778                 ret= swr_convert_internal(s, out, size, in, size);
779                 if(ret<0)
780                     return ret;
781                 buf_set(in, in, ret);
782                 in_count -= ret;
783                 ret2 += ret;
784             }
785             if(in_count){
786                 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
787                 copy(&tmp, in, in_count);
788                 s->in_buffer_count += in_count;
789             }
790         }
791         if(ret2>0 && !s->drop_output)
792             s->outpts += ret2 * (int64_t)s->in_sample_rate;
793         return ret2;
794     }
795 }
796
797 int swr_drop_output(struct SwrContext *s, int count){
798     s->drop_output += count;
799
800     if(s->drop_output <= 0)
801         return 0;
802
803     av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
804     return swr_convert(s, NULL, s->drop_output, NULL, 0);
805 }
806
807 int swr_inject_silence(struct SwrContext *s, int count){
808     int ret, i;
809     AudioData silence = s->in;
810     uint8_t *tmp_arg[SWR_CH_MAX];
811
812     if(count <= 0)
813         return 0;
814
815     silence.count = 0;
816     silence.data  = NULL;
817     if((ret=swri_realloc_audio(&silence, count))<0)
818         return ret;
819
820     if(silence.planar) for(i=0; i<silence.ch_count; i++) {
821         memset(silence.ch[i], silence.bps==1 ? 0x80 : 0, count*silence.bps);
822     } else
823         memset(silence.ch[0], silence.bps==1 ? 0x80 : 0, count*silence.bps*silence.ch_count);
824
825     reversefill_audiodata(&silence, tmp_arg);
826     av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
827     ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
828     av_freep(&silence.data);
829     return ret;
830 }
831
832 int64_t swr_get_delay(struct SwrContext *s, int64_t base){
833     if (s->resampler && s->resample){
834         return s->resampler->get_delay(s, base);
835     }else{
836         return (s->in_buffer_count*base + (s->in_sample_rate>>1))/ s->in_sample_rate;
837     }
838 }
839
840 int swr_set_compensation(struct SwrContext *s, int sample_delta, int compensation_distance){
841     int ret;
842
843     if (!s || compensation_distance < 0)
844         return AVERROR(EINVAL);
845     if (!compensation_distance && sample_delta)
846         return AVERROR(EINVAL);
847     if (!s->resample) {
848         s->flags |= SWR_FLAG_RESAMPLE;
849         ret = swr_init(s);
850         if (ret < 0)
851             return ret;
852     }
853     if (!s->resampler->set_compensation){
854         return AVERROR(EINVAL);
855     }else{
856         return s->resampler->set_compensation(s->resample, sample_delta, compensation_distance);
857     }
858 }
859
860 int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
861     if(pts == INT64_MIN)
862         return s->outpts;
863     if(s->min_compensation >= FLT_MAX) {
864         return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
865     } else {
866         int64_t delta = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate) - s->outpts;
867         double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
868
869         if(fabs(fdelta) > s->min_compensation) {
870             if(!s->outpts || fabs(fdelta) > s->min_hard_compensation){
871                 int ret;
872                 if(delta > 0) ret = swr_inject_silence(s,  delta / s->out_sample_rate);
873                 else          ret = swr_drop_output   (s, -delta / s-> in_sample_rate);
874                 if(ret<0){
875                     av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
876                 }
877             } else if(s->soft_compensation_duration && s->max_soft_compensation) {
878                 int duration = s->out_sample_rate * s->soft_compensation_duration;
879                 double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
880                 int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
881                 av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
882                 swr_set_compensation(s, comp, duration);
883             }
884         }
885
886         return s->outpts;
887     }
888 }