f11bbd715c9a71785402af3c33aeed62cb188981
[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);
221         swri_audio_convert_free(&s-> in_convert);
222         swri_audio_convert_free(&s->out_convert);
223         swri_audio_convert_free(&s->full_convert);
224         if (s->resampler)
225             s->resampler->free(&s->resample);
226         swri_rematrix_free(s);
227     }
228
229     av_freep(ss);
230 }
231
232 av_cold int swr_init(struct SwrContext *s){
233     s->in_buffer_index= 0;
234     s->in_buffer_count= 0;
235     s->resample_in_constraint= 0;
236     free_temp(&s->postin);
237     free_temp(&s->midbuf);
238     free_temp(&s->preout);
239     free_temp(&s->in_buffer);
240     free_temp(&s->dither);
241     memset(s->in.ch, 0, sizeof(s->in.ch));
242     memset(s->out.ch, 0, sizeof(s->out.ch));
243     swri_audio_convert_free(&s-> in_convert);
244     swri_audio_convert_free(&s->out_convert);
245     swri_audio_convert_free(&s->full_convert);
246     swri_rematrix_free(s);
247
248     s->flushed = 0;
249
250     if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
251         av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
252         return AVERROR(EINVAL);
253     }
254     if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
255         av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
256         return AVERROR(EINVAL);
257     }
258
259     if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
260         if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_S16P){
261             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
262         }else if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_FLTP){
263             s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
264         }else{
265             av_log(s, AV_LOG_DEBUG, "Using double precision mode\n");
266             s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
267         }
268     }
269
270     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16P
271         &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
272         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
273         &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
274         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));
275         return AVERROR(EINVAL);
276     }
277
278     switch(s->engine){
279 #if CONFIG_LIBSOXR
280         extern struct Resampler const soxr_resampler;
281         case SWR_ENGINE_SOXR: s->resampler = &soxr_resampler; break;
282 #endif
283         case SWR_ENGINE_SWR : s->resampler = &swri_resampler; break;
284         default:
285             av_log(s, AV_LOG_ERROR, "Requested resampling engine is unavailable\n");
286             return AVERROR(EINVAL);
287     }
288
289     set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
290     set_audiodata_fmt(&s->out, s->out_sample_fmt);
291
292     if (s->async) {
293         if (s->min_compensation >= FLT_MAX/2)
294             s->min_compensation = 0.001;
295         if (s->async > 1.0001) {
296             s->max_soft_compensation = s->async / (double) s->in_sample_rate;
297         }
298     }
299
300     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
301         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);
302     }else
303         s->resampler->free(&s->resample);
304     if(    s->int_sample_fmt != AV_SAMPLE_FMT_S16P
305         && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
306         && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
307         && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
308         && s->resample){
309         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16/s32/flt/dbl\n");
310         return -1;
311     }
312
313     if(!s->used_ch_count)
314         s->used_ch_count= s->in.ch_count;
315
316     if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
317         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
318         s-> in_ch_layout= 0;
319     }
320
321     if(!s-> in_ch_layout)
322         s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
323     if(!s->out_ch_layout)
324         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
325
326     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
327                  s->rematrix_custom;
328
329 #define RSC 1 //FIXME finetune
330     if(!s-> in.ch_count)
331         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
332     if(!s->used_ch_count)
333         s->used_ch_count= s->in.ch_count;
334     if(!s->out.ch_count)
335         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
336
337     if(!s-> in.ch_count){
338         av_assert0(!s->in_ch_layout);
339         av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
340         return -1;
341     }
342
343     if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
344         char l1[1024], l2[1024];
345         av_get_channel_layout_string(l1, sizeof(l1), s-> in.ch_count, s-> in_ch_layout);
346         av_get_channel_layout_string(l2, sizeof(l2), s->out.ch_count, s->out_ch_layout);
347         av_log(s, AV_LOG_ERROR, "Rematrix is needed between %s and %s "
348                "but there is not enough information to do it\n", l1, l2);
349         return -1;
350     }
351
352 av_assert0(s->used_ch_count);
353 av_assert0(s->out.ch_count);
354     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
355
356     s->in_buffer= s->in;
357
358     if(!s->resample && !s->rematrix && !s->channel_map && !s->dither_method){
359         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
360                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
361         return 0;
362     }
363
364     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
365                                              s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
366     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
367                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
368
369
370     s->postin= s->in;
371     s->preout= s->out;
372     s->midbuf= s->in;
373
374     if(s->channel_map){
375         s->postin.ch_count=
376         s->midbuf.ch_count= s->used_ch_count;
377         if(s->resample)
378             s->in_buffer.ch_count= s->used_ch_count;
379     }
380     if(!s->resample_first){
381         s->midbuf.ch_count= s->out.ch_count;
382         if(s->resample)
383             s->in_buffer.ch_count = s->out.ch_count;
384     }
385
386     set_audiodata_fmt(&s->postin, s->int_sample_fmt);
387     set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
388     set_audiodata_fmt(&s->preout, s->int_sample_fmt);
389
390     if(s->resample){
391         set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
392     }
393
394     s->dither = s->preout;
395
396     if(s->rematrix || s->dither_method)
397         return swri_rematrix_init(s);
398
399     return 0;
400 }
401
402 int swri_realloc_audio(AudioData *a, int count){
403     int i, countb;
404     AudioData old;
405
406     if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
407         return AVERROR(EINVAL);
408
409     if(a->count >= count)
410         return 0;
411
412     count*=2;
413
414     countb= FFALIGN(count*a->bps, ALIGN);
415     old= *a;
416
417     av_assert0(a->bps);
418     av_assert0(a->ch_count);
419
420     a->data= av_mallocz(countb*a->ch_count);
421     if(!a->data)
422         return AVERROR(ENOMEM);
423     for(i=0; i<a->ch_count; i++){
424         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
425         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
426     }
427     if(!a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
428     av_free(old.data);
429     a->count= count;
430
431     return 1;
432 }
433
434 static void copy(AudioData *out, AudioData *in,
435                  int count){
436     av_assert0(out->planar == in->planar);
437     av_assert0(out->bps == in->bps);
438     av_assert0(out->ch_count == in->ch_count);
439     if(out->planar){
440         int ch;
441         for(ch=0; ch<out->ch_count; ch++)
442             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
443     }else
444         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
445 }
446
447 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
448     int i;
449     if(!in_arg){
450         memset(out->ch, 0, sizeof(out->ch));
451     }else if(out->planar){
452         for(i=0; i<out->ch_count; i++)
453             out->ch[i]= in_arg[i];
454     }else{
455         for(i=0; i<out->ch_count; i++)
456             out->ch[i]= in_arg[0] + i*out->bps;
457     }
458 }
459
460 static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
461     int i;
462     if(out->planar){
463         for(i=0; i<out->ch_count; i++)
464             in_arg[i]= out->ch[i];
465     }else{
466         in_arg[0]= out->ch[0];
467     }
468 }
469
470 /**
471  *
472  * out may be equal in.
473  */
474 static void buf_set(AudioData *out, AudioData *in, int count){
475     int ch;
476     if(in->planar){
477         for(ch=0; ch<out->ch_count; ch++)
478             out->ch[ch]= in->ch[ch] + count*out->bps;
479     }else{
480         for(ch=out->ch_count-1; ch>=0; ch--)
481             out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
482     }
483 }
484
485 /**
486  *
487  * @return number of samples output per channel
488  */
489 static int resample(SwrContext *s, AudioData *out_param, int out_count,
490                              const AudioData * in_param, int in_count){
491     AudioData in, out, tmp;
492     int ret_sum=0;
493     int border=0;
494
495     av_assert1(s->in_buffer.ch_count == in_param->ch_count);
496     av_assert1(s->in_buffer.planar   == in_param->planar);
497     av_assert1(s->in_buffer.fmt      == in_param->fmt);
498
499     tmp=out=*out_param;
500     in =  *in_param;
501
502     do{
503         int ret, size, consumed;
504         if(!s->resample_in_constraint && s->in_buffer_count){
505             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
506             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
507             out_count -= ret;
508             ret_sum += ret;
509             buf_set(&out, &out, ret);
510             s->in_buffer_count -= consumed;
511             s->in_buffer_index += consumed;
512
513             if(!in_count)
514                 break;
515             if(s->in_buffer_count <= border){
516                 buf_set(&in, &in, -s->in_buffer_count);
517                 in_count += s->in_buffer_count;
518                 s->in_buffer_count=0;
519                 s->in_buffer_index=0;
520                 border = 0;
521             }
522         }
523
524         if((s->flushed || in_count) && !s->in_buffer_count){
525             s->in_buffer_index=0;
526             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &in, in_count, &consumed);
527             out_count -= ret;
528             ret_sum += ret;
529             buf_set(&out, &out, ret);
530             in_count -= consumed;
531             buf_set(&in, &in, consumed);
532         }
533
534         //TODO is this check sane considering the advanced copy avoidance below
535         size= s->in_buffer_index + s->in_buffer_count + in_count;
536         if(   size > s->in_buffer.count
537            && s->in_buffer_count + in_count <= s->in_buffer_index){
538             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
539             copy(&s->in_buffer, &tmp, s->in_buffer_count);
540             s->in_buffer_index=0;
541         }else
542             if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
543                 return ret;
544
545         if(in_count){
546             int count= in_count;
547             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
548
549             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
550             copy(&tmp, &in, /*in_*/count);
551             s->in_buffer_count += count;
552             in_count -= count;
553             border += count;
554             buf_set(&in, &in, count);
555             s->resample_in_constraint= 0;
556             if(s->in_buffer_count != count || in_count)
557                 continue;
558         }
559         break;
560     }while(1);
561
562     s->resample_in_constraint= !!out_count;
563
564     return ret_sum;
565 }
566
567 static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
568                                                       AudioData *in , int  in_count){
569     AudioData *postin, *midbuf, *preout;
570     int ret/*, in_max*/;
571     AudioData preout_tmp, midbuf_tmp;
572
573     if(s->full_convert){
574         av_assert0(!s->resample);
575         swri_audio_convert(s->full_convert, out, in, in_count);
576         return out_count;
577     }
578
579 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
580 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
581
582     if((ret=swri_realloc_audio(&s->postin, in_count))<0)
583         return ret;
584     if(s->resample_first){
585         av_assert0(s->midbuf.ch_count == s->used_ch_count);
586         if((ret=swri_realloc_audio(&s->midbuf, out_count))<0)
587             return ret;
588     }else{
589         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
590         if((ret=swri_realloc_audio(&s->midbuf,  in_count))<0)
591             return ret;
592     }
593     if((ret=swri_realloc_audio(&s->preout, out_count))<0)
594         return ret;
595
596     postin= &s->postin;
597
598     midbuf_tmp= s->midbuf;
599     midbuf= &midbuf_tmp;
600     preout_tmp= s->preout;
601     preout= &preout_tmp;
602
603     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
604         postin= in;
605
606     if(s->resample_first ? !s->resample : !s->rematrix)
607         midbuf= postin;
608
609     if(s->resample_first ? !s->rematrix : !s->resample)
610         preout= midbuf;
611
612     if (preout == in && s->dither_method) {
613         av_assert1(postin == midbuf && midbuf == preout);
614         postin = midbuf = preout = &preout_tmp;
615     }
616
617     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar){
618         if(preout==in){
619             out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
620             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
621             copy(out, in, out_count);
622             return out_count;
623         }
624         else if(preout==postin) preout= midbuf= postin= out;
625         else if(preout==midbuf) preout= midbuf= out;
626         else                    preout= out;
627     }
628
629     if(in != postin){
630         swri_audio_convert(s->in_convert, postin, in, in_count);
631     }
632
633     if(s->resample_first){
634         if(postin != midbuf)
635             out_count= resample(s, midbuf, out_count, postin, in_count);
636         if(midbuf != preout)
637             swri_rematrix(s, preout, midbuf, out_count, preout==out);
638     }else{
639         if(postin != midbuf)
640             swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
641         if(midbuf != preout)
642             out_count= resample(s, preout, out_count, midbuf, in_count);
643     }
644
645     if(preout != out && out_count){
646         if(s->dither_method){
647             int ch, len1;
648             int dither_count= FFMAX(out_count, 1<<16);
649             av_assert0(preout != in);
650
651             if((ret=swri_realloc_audio(&s->dither, dither_count))<0)
652                 return ret;
653             if(ret)
654                 for(ch=0; ch<s->dither.ch_count; ch++)
655                     swri_get_dither(s, s->dither.ch[ch], s->dither.count, 12345678913579<<ch, s->out_sample_fmt, s->int_sample_fmt);
656             av_assert0(s->dither.ch_count == preout->ch_count);
657
658             if(s->dither_pos + out_count > s->dither.count)
659                 s->dither_pos = 0;
660
661             if (s->dither_method < SWR_DITHER_NS){
662                 if (s->mix_2_1_simd) {
663                     int len1= out_count&~15;
664                     int off = len1 * preout->bps;
665
666                     if(len1)
667                         for(ch=0; ch<preout->ch_count; ch++)
668                             s->mix_2_1_simd(preout->ch[ch], preout->ch[ch], s->dither.ch[ch] + s->dither.bps * s->dither_pos, s->native_one, 0, 0, len1);
669                     if(out_count != len1)
670                         for(ch=0; ch<preout->ch_count; ch++)
671                             s->mix_2_1_f(preout->ch[ch] + off, preout->ch[ch] + off, s->dither.ch[ch] + s->dither.bps * s->dither_pos + off + len1, s->native_one, 0, 0, out_count - len1);
672                 } else {
673                     for(ch=0; ch<preout->ch_count; ch++)
674                         s->mix_2_1_f(preout->ch[ch], preout->ch[ch], s->dither.ch[ch] + s->dither.bps * s->dither_pos, s->native_one, 0, 0, out_count);
675                 }
676             } else {
677                 switch(s->int_sample_fmt) {
678                 case AV_SAMPLE_FMT_S16P :swri_noise_shaping_int16(s, preout, &s->dither, out_count); break;
679                 case AV_SAMPLE_FMT_S32P :swri_noise_shaping_int32(s, preout, &s->dither, out_count); break;
680                 case AV_SAMPLE_FMT_FLTP :swri_noise_shaping_float(s, preout, &s->dither, out_count); break;
681                 case AV_SAMPLE_FMT_DBLP :swri_noise_shaping_double(s,preout, &s->dither, out_count); break;
682                 }
683             }
684             s->dither_pos += out_count;
685         }
686 //FIXME packed doesnt need more than 1 chan here!
687         swri_audio_convert(s->out_convert, out, preout, out_count);
688     }
689     return out_count;
690 }
691
692 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
693                                 const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
694     AudioData * in= &s->in;
695     AudioData *out= &s->out;
696
697     if(s->drop_output > 0){
698         int ret;
699         AudioData tmp = s->out;
700         uint8_t *tmp_arg[SWR_CH_MAX];
701         tmp.count = 0;
702         tmp.data  = NULL;
703         if((ret=swri_realloc_audio(&tmp, s->drop_output))<0)
704             return ret;
705
706         reversefill_audiodata(&tmp, tmp_arg);
707         s->drop_output *= -1; //FIXME find a less hackish solution
708         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
709         s->drop_output *= -1;
710         if(ret>0)
711             s->drop_output -= ret;
712
713         av_freep(&tmp.data);
714         if(s->drop_output || !out_arg)
715             return 0;
716         in_count = 0;
717     }
718
719     if(!in_arg){
720         if(s->resample){
721             if (!s->flushed)
722                 s->resampler->flush(s);
723             s->resample_in_constraint = 0;
724             s->flushed = 1;
725         }else if(!s->in_buffer_count){
726             return 0;
727         }
728     }else
729         fill_audiodata(in ,  (void*)in_arg);
730
731     fill_audiodata(out, out_arg);
732
733     if(s->resample){
734         int ret = swr_convert_internal(s, out, out_count, in, in_count);
735         if(ret>0 && !s->drop_output)
736             s->outpts += ret * (int64_t)s->in_sample_rate;
737         return ret;
738     }else{
739         AudioData tmp= *in;
740         int ret2=0;
741         int ret, size;
742         size = FFMIN(out_count, s->in_buffer_count);
743         if(size){
744             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
745             ret= swr_convert_internal(s, out, size, &tmp, size);
746             if(ret<0)
747                 return ret;
748             ret2= ret;
749             s->in_buffer_count -= ret;
750             s->in_buffer_index += ret;
751             buf_set(out, out, ret);
752             out_count -= ret;
753             if(!s->in_buffer_count)
754                 s->in_buffer_index = 0;
755         }
756
757         if(in_count){
758             size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
759
760             if(in_count > out_count) { //FIXME move after swr_convert_internal
761                 if(   size > s->in_buffer.count
762                 && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
763                     buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
764                     copy(&s->in_buffer, &tmp, s->in_buffer_count);
765                     s->in_buffer_index=0;
766                 }else
767                     if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
768                         return ret;
769             }
770
771             if(out_count){
772                 size = FFMIN(in_count, out_count);
773                 ret= swr_convert_internal(s, out, size, in, size);
774                 if(ret<0)
775                     return ret;
776                 buf_set(in, in, ret);
777                 in_count -= ret;
778                 ret2 += ret;
779             }
780             if(in_count){
781                 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
782                 copy(&tmp, in, in_count);
783                 s->in_buffer_count += in_count;
784             }
785         }
786         if(ret2>0 && !s->drop_output)
787             s->outpts += ret2 * (int64_t)s->in_sample_rate;
788         return ret2;
789     }
790 }
791
792 int swr_drop_output(struct SwrContext *s, int count){
793     s->drop_output += count;
794
795     if(s->drop_output <= 0)
796         return 0;
797
798     av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
799     return swr_convert(s, NULL, s->drop_output, NULL, 0);
800 }
801
802 int swr_inject_silence(struct SwrContext *s, int count){
803     int ret, i;
804     AudioData silence = s->in;
805     uint8_t *tmp_arg[SWR_CH_MAX];
806
807     if(count <= 0)
808         return 0;
809
810     silence.count = 0;
811     silence.data  = NULL;
812     if((ret=swri_realloc_audio(&silence, count))<0)
813         return ret;
814
815     if(silence.planar) for(i=0; i<silence.ch_count; i++) {
816         memset(silence.ch[i], silence.bps==1 ? 0x80 : 0, count*silence.bps);
817     } else
818         memset(silence.ch[0], silence.bps==1 ? 0x80 : 0, count*silence.bps*silence.ch_count);
819
820     reversefill_audiodata(&silence, tmp_arg);
821     av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
822     ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
823     av_freep(&silence.data);
824     return ret;
825 }
826
827 int64_t swr_get_delay(struct SwrContext *s, int64_t base){
828     if (s->resampler && s->resample){
829         return s->resampler->get_delay(s, base);
830     }else{
831         return (s->in_buffer_count*base + (s->in_sample_rate>>1))/ s->in_sample_rate;
832     }
833 }
834
835 int swr_set_compensation(struct SwrContext *s, int sample_delta, int compensation_distance){
836     int ret;
837
838     if (!s || compensation_distance < 0)
839         return AVERROR(EINVAL);
840     if (!compensation_distance && sample_delta)
841         return AVERROR(EINVAL);
842     if (!s->resample) {
843         s->flags |= SWR_FLAG_RESAMPLE;
844         ret = swr_init(s);
845         if (ret < 0)
846             return ret;
847     }
848     if (!s->resampler->set_compensation){
849         return AVERROR(EINVAL);
850     }else{
851         return s->resampler->set_compensation(s->resample, sample_delta, compensation_distance);
852     }
853 }
854
855 int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
856     if(pts == INT64_MIN)
857         return s->outpts;
858     if(s->min_compensation >= FLT_MAX) {
859         return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
860     } else {
861         int64_t delta = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate) - s->outpts;
862         double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
863
864         if(fabs(fdelta) > s->min_compensation) {
865             if(!s->outpts || fabs(fdelta) > s->min_hard_compensation){
866                 int ret;
867                 if(delta > 0) ret = swr_inject_silence(s,  delta / s->out_sample_rate);
868                 else          ret = swr_drop_output   (s, -delta / s-> in_sample_rate);
869                 if(ret<0){
870                     av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
871                 }
872             } else if(s->soft_compensation_duration && s->max_soft_compensation) {
873                 int duration = s->out_sample_rate * s->soft_compensation_duration;
874                 double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
875                 int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
876                 av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
877                 swr_set_compensation(s, comp, duration);
878             }
879         }
880
881         return s->outpts;
882     }
883 }