2d063cd2cea4f88fd5859837d1629e9a06ae881d
[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   , INT_MAX, PARAM},
55 {"in_sample_fmt"        , "set input sample format"     , OFFSET( in_sample_fmt ), AV_OPT_TYPE_SAMPLE_FMT , {.i64=AV_SAMPLE_FMT_NONE}, -1   , INT_MAX, PARAM},
56 {"osf"                  , "set output sample format"    , OFFSET(out_sample_fmt ), AV_OPT_TYPE_SAMPLE_FMT , {.i64=AV_SAMPLE_FMT_NONE}, -1   , INT_MAX, PARAM},
57 {"out_sample_fmt"       , "set output sample format"    , OFFSET(out_sample_fmt ), AV_OPT_TYPE_SAMPLE_FMT , {.i64=AV_SAMPLE_FMT_NONE}, -1   , INT_MAX, PARAM},
58 {"tsf"                  , "set internal sample format"  , OFFSET(int_sample_fmt ), AV_OPT_TYPE_SAMPLE_FMT , {.i64=AV_SAMPLE_FMT_NONE}, -1   , INT_MAX, PARAM},
59 {"internal_sample_fmt"  , "set internal sample format"  , OFFSET(int_sample_fmt ), AV_OPT_TYPE_SAMPLE_FMT , {.i64=AV_SAMPLE_FMT_NONE}, -1   , INT_MAX, PARAM},
60 {"icl"                  , "set input channel layout"    , OFFSET( in_ch_layout  ), AV_OPT_TYPE_CHANNEL_LAYOUT, {.i64=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
61 {"in_channel_layout"    , "set input channel layout"    , OFFSET( in_ch_layout  ), AV_OPT_TYPE_CHANNEL_LAYOUT, {.i64=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
62 {"ocl"                  , "set output channel layout"   , OFFSET(out_ch_layout  ), AV_OPT_TYPE_CHANNEL_LAYOUT, {.i64=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
63 {"out_channel_layout"   , "set output channel layout"   , OFFSET(out_ch_layout  ), AV_OPT_TYPE_CHANNEL_LAYOUT, {.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 {"rematrix_maxval"      , "set rematrix maxval"         , OFFSET(rematrix_maxval), AV_OPT_TYPE_FLOAT, {.dbl=0.0                   }, 0      , 1000      , PARAM},
72
73 {"flags"                , "set flags"                   , OFFSET(flags          ), AV_OPT_TYPE_FLAGS, {.i64=0                     }, 0      , UINT_MAX  , PARAM, "flags"},
74 {"swr_flags"            , "set flags"                   , OFFSET(flags          ), AV_OPT_TYPE_FLAGS, {.i64=0                     }, 0      , UINT_MAX  , PARAM, "flags"},
75 {"res"                  , "force resampling"            , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_FLAG_RESAMPLE     }, INT_MIN, INT_MAX   , PARAM, "flags"},
76
77 {"dither_scale"         , "set dither scale"            , OFFSET(dither.scale   ), AV_OPT_TYPE_FLOAT, {.dbl=1                     }, 0      , INT_MAX   , PARAM},
78
79 {"dither_method"        , "set dither method"           , OFFSET(dither.method  ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_DITHER_NB-1, PARAM, "dither_method"},
80 {"rectangular"          , "select rectangular dither"   , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_RECTANGULAR}, INT_MIN, INT_MAX   , PARAM, "dither_method"},
81 {"triangular"           , "select triangular dither"    , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_TRIANGULAR }, INT_MIN, INT_MAX   , PARAM, "dither_method"},
82 {"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"},
83 {"lipshitz"             , "select lipshitz noise shaping dither" , 0             , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_NS_LIPSHITZ}, INT_MIN, INT_MAX, PARAM, "dither_method"},
84 {"shibata"              , "select shibata noise shaping dither" , 0              , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_NS_SHIBATA }, INT_MIN, INT_MAX, PARAM, "dither_method"},
85 {"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"},
86 {"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"},
87 {"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"},
88 {"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"},
89 {"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"},
90
91 {"filter_size"          , "set swr resampling filter size", OFFSET(filter_size)  , AV_OPT_TYPE_INT  , {.i64=32                    }, 0      , INT_MAX   , PARAM },
92 {"phase_shift"          , "set swr resampling phase shift", OFFSET(phase_shift)  , AV_OPT_TYPE_INT  , {.i64=10                    }, 0      , 24        , PARAM },
93 {"linear_interp"        , "enable linear interpolation" , OFFSET(linear_interp)  , AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , 1         , PARAM },
94 {"cutoff"               , "set cutoff frequency ratio"  , OFFSET(cutoff)         , AV_OPT_TYPE_DOUBLE,{.dbl=0.                    }, 0      , 1         , PARAM },
95
96 /* duplicate option in order to work with avconv */
97 {"resample_cutoff"      , "set cutoff frequency ratio"  , OFFSET(cutoff)         , AV_OPT_TYPE_DOUBLE,{.dbl=0.                    }, 0      , 1         , PARAM },
98
99 {"resampler"            , "set resampling Engine"       , OFFSET(engine)         , AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_ENGINE_NB-1, PARAM, "resampler"},
100 {"swr"                  , "select SW Resampler"         , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_ENGINE_SWR        }, INT_MIN, INT_MAX   , PARAM, "resampler"},
101 {"soxr"                 , "select SoX Resampler"        , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_ENGINE_SOXR       }, INT_MIN, INT_MAX   , PARAM, "resampler"},
102 {"precision"            , "set soxr resampling precision (in bits)"
103                                                         , OFFSET(precision)      , AV_OPT_TYPE_DOUBLE,{.dbl=20.0                  }, 15.0   , 33.0      , PARAM },
104 {"cheby"                , "enable soxr Chebyshev passband & higher-precision irrational ratio approximation"
105                                                         , OFFSET(cheby)          , AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , 1         , PARAM },
106 {"min_comp"             , "set minimum difference between timestamps and audio data (in seconds) below which no timestamp compensation of either kind is applied"
107                                                         , OFFSET(min_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=FLT_MAX               }, 0      , FLT_MAX   , PARAM },
108 {"min_hard_comp"        , "set minimum difference between timestamps and audio data (in seconds) to trigger padding/trimming the data."
109                                                         , OFFSET(min_hard_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=0.1                   }, 0      , INT_MAX   , PARAM },
110 {"comp_duration"        , "set duration (in seconds) over which data is stretched/squeezed to make it match the timestamps."
111                                                         , OFFSET(soft_compensation_duration),AV_OPT_TYPE_FLOAT ,{.dbl=1                     }, 0      , INT_MAX   , PARAM },
112 {"max_soft_comp"        , "set maximum factor by which data is stretched/squeezed to make it match the timestamps."
113                                                         , OFFSET(max_soft_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=0                     }, INT_MIN, INT_MAX   , PARAM },
114 {"async"                , "simplified 1 parameter audio timestamp matching, 0(disabled), 1(filling and trimming), >1(maximum stretch/squeeze in samples per second)"
115                                                         , OFFSET(async)          , AV_OPT_TYPE_FLOAT ,{.dbl=0                     }, INT_MIN, INT_MAX   , PARAM },
116 {"first_pts"            , "Assume the first pts should be this value (in samples)."
117                                                         , OFFSET(firstpts_in_samples), AV_OPT_TYPE_INT64 ,{.i64=AV_NOPTS_VALUE    }, INT64_MIN,INT64_MAX, PARAM },
118
119 { "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" },
120     { "none",  "select none",               0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_NONE  }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
121     { "dolby", "select Dolby",              0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_DOLBY }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
122     { "dplii", "select Dolby Pro Logic II", 0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_DPLII }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
123
124 { "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" },
125     { "cubic"           , "select cubic"                , 0                      , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_CUBIC            }, INT_MIN, INT_MAX, PARAM, "filter_type" },
126     { "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" },
127     { "kaiser"          , "select Kaiser Windowed Sinc" , 0                      , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_KAISER           }, INT_MIN, INT_MAX, PARAM, "filter_type" },
128
129 { "kaiser_beta"         , "set swr Kaiser Window Beta"  , OFFSET(kaiser_beta)    , AV_OPT_TYPE_INT  , {.i64=9                     }, 2      , 16        , PARAM },
130
131 { "output_sample_bits"  , "set swr number of output sample bits", OFFSET(dither.output_sample_bits), AV_OPT_TYPE_INT  , {.i64=0   }, 0      , 64        , PARAM },
132 {0}
133 };
134
135 static const char* context_to_name(void* ptr) {
136     return "SWR";
137 }
138
139 static const AVClass av_class = {
140     .class_name                = "SWResampler",
141     .item_name                 = context_to_name,
142     .option                    = options,
143     .version                   = LIBAVUTIL_VERSION_INT,
144     .log_level_offset_offset   = OFFSET(log_level_offset),
145     .parent_log_context_offset = OFFSET(log_ctx),
146     .category                  = AV_CLASS_CATEGORY_SWRESAMPLER,
147 };
148
149 unsigned swresample_version(void)
150 {
151     av_assert0(LIBSWRESAMPLE_VERSION_MICRO >= 100);
152     return LIBSWRESAMPLE_VERSION_INT;
153 }
154
155 const char *swresample_configuration(void)
156 {
157     return FFMPEG_CONFIGURATION;
158 }
159
160 const char *swresample_license(void)
161 {
162 #define LICENSE_PREFIX "libswresample license: "
163     return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
164 }
165
166 int swr_set_channel_mapping(struct SwrContext *s, const int *channel_map){
167     if(!s || s->in_convert) // s needs to be allocated but not initialized
168         return AVERROR(EINVAL);
169     s->channel_map = channel_map;
170     return 0;
171 }
172
173 const AVClass *swr_get_class(void)
174 {
175     return &av_class;
176 }
177
178 av_cold struct SwrContext *swr_alloc(void){
179     SwrContext *s= av_mallocz(sizeof(SwrContext));
180     if(s){
181         s->av_class= &av_class;
182         av_opt_set_defaults(s);
183     }
184     return s;
185 }
186
187 struct SwrContext *swr_alloc_set_opts(struct SwrContext *s,
188                                       int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
189                                       int64_t  in_ch_layout, enum AVSampleFormat  in_sample_fmt, int  in_sample_rate,
190                                       int log_offset, void *log_ctx){
191     if(!s) s= swr_alloc();
192     if(!s) return NULL;
193
194     s->log_level_offset= log_offset;
195     s->log_ctx= log_ctx;
196
197     av_opt_set_int(s, "ocl", out_ch_layout,   0);
198     av_opt_set_int(s, "osf", out_sample_fmt,  0);
199     av_opt_set_int(s, "osr", out_sample_rate, 0);
200     av_opt_set_int(s, "icl", in_ch_layout,    0);
201     av_opt_set_int(s, "isf", in_sample_fmt,   0);
202     av_opt_set_int(s, "isr", in_sample_rate,  0);
203     av_opt_set_int(s, "tsf", AV_SAMPLE_FMT_NONE,   0);
204     av_opt_set_int(s, "ich", av_get_channel_layout_nb_channels(s-> in_ch_layout), 0);
205     av_opt_set_int(s, "och", av_get_channel_layout_nb_channels(s->out_ch_layout), 0);
206     av_opt_set_int(s, "uch", 0, 0);
207     return s;
208 }
209
210 static void set_audiodata_fmt(AudioData *a, enum AVSampleFormat fmt){
211     a->fmt   = fmt;
212     a->bps   = av_get_bytes_per_sample(fmt);
213     a->planar= av_sample_fmt_is_planar(fmt);
214 }
215
216 static void free_temp(AudioData *a){
217     av_free(a->data);
218     memset(a, 0, sizeof(*a));
219 }
220
221 static void clear_context(SwrContext *s){
222     s->in_buffer_index= 0;
223     s->in_buffer_count= 0;
224     s->resample_in_constraint= 0;
225     memset(s->in.ch, 0, sizeof(s->in.ch));
226     memset(s->out.ch, 0, sizeof(s->out.ch));
227     free_temp(&s->postin);
228     free_temp(&s->midbuf);
229     free_temp(&s->preout);
230     free_temp(&s->in_buffer);
231     free_temp(&s->silence);
232     free_temp(&s->drop_temp);
233     free_temp(&s->dither.noise);
234     free_temp(&s->dither.temp);
235     swri_audio_convert_free(&s-> in_convert);
236     swri_audio_convert_free(&s->out_convert);
237     swri_audio_convert_free(&s->full_convert);
238     swri_rematrix_free(s);
239
240     s->flushed = 0;
241 }
242
243 av_cold void swr_free(SwrContext **ss){
244     SwrContext *s= *ss;
245     if(s){
246         clear_context(s);
247         if (s->resampler)
248             s->resampler->free(&s->resample);
249     }
250
251     av_freep(ss);
252 }
253
254 av_cold void swr_close(SwrContext *s){
255     clear_context(s);
256 }
257
258 av_cold int swr_init(struct SwrContext *s){
259     int ret;
260
261     clear_context(s);
262
263     if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
264         av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
265         return AVERROR(EINVAL);
266     }
267     if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
268         av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
269         return AVERROR(EINVAL);
270     }
271
272     if(av_get_channel_layout_nb_channels(s-> in_ch_layout) > SWR_CH_MAX) {
273         av_log(s, AV_LOG_WARNING, "Input channel layout 0x%"PRIx64" is invalid or unsupported.\n", s-> in_ch_layout);
274         s->in_ch_layout = 0;
275     }
276
277     if(av_get_channel_layout_nb_channels(s->out_ch_layout) > SWR_CH_MAX) {
278         av_log(s, AV_LOG_WARNING, "Output channel layout 0x%"PRIx64" is invalid or unsupported.\n", s->out_ch_layout);
279         s->out_ch_layout = 0;
280     }
281
282     switch(s->engine){
283 #if CONFIG_LIBSOXR
284         extern struct Resampler const soxr_resampler;
285         case SWR_ENGINE_SOXR: s->resampler = &soxr_resampler; break;
286 #endif
287         case SWR_ENGINE_SWR : s->resampler = &swri_resampler; break;
288         default:
289             av_log(s, AV_LOG_ERROR, "Requested resampling engine is unavailable\n");
290             return AVERROR(EINVAL);
291     }
292
293     if(!s->used_ch_count)
294         s->used_ch_count= s->in.ch_count;
295
296     if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
297         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
298         s-> in_ch_layout= 0;
299     }
300
301     if(!s-> in_ch_layout)
302         s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
303     if(!s->out_ch_layout)
304         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
305
306     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
307                  s->rematrix_custom;
308
309     if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
310         if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_S16P){
311             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
312         }else if(   av_get_planar_sample_fmt(s-> in_sample_fmt) == AV_SAMPLE_FMT_S32P
313                  && av_get_planar_sample_fmt(s->out_sample_fmt) == AV_SAMPLE_FMT_S32P
314                  && !s->rematrix
315                  && s->engine != SWR_ENGINE_SOXR){
316             s->int_sample_fmt= AV_SAMPLE_FMT_S32P;
317         }else if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_FLTP){
318             s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
319         }else{
320             av_log(s, AV_LOG_DEBUG, "Using double precision mode\n");
321             s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
322         }
323     }
324
325     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16P
326         &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
327         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
328         &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
329         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));
330         return AVERROR(EINVAL);
331     }
332
333     set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
334     set_audiodata_fmt(&s->out, s->out_sample_fmt);
335
336     if (s->firstpts_in_samples != AV_NOPTS_VALUE) {
337         if (!s->async && s->min_compensation >= FLT_MAX/2)
338             s->async = 1;
339         s->firstpts =
340         s->outpts   = s->firstpts_in_samples * s->out_sample_rate;
341     } else
342         s->firstpts = AV_NOPTS_VALUE;
343
344     if (s->async) {
345         if (s->min_compensation >= FLT_MAX/2)
346             s->min_compensation = 0.001;
347         if (s->async > 1.0001) {
348             s->max_soft_compensation = s->async / (double) s->in_sample_rate;
349         }
350     }
351
352     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
353         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);
354     }else
355         s->resampler->free(&s->resample);
356     if(    s->int_sample_fmt != AV_SAMPLE_FMT_S16P
357         && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
358         && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
359         && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
360         && s->resample){
361         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16/s32/flt/dbl\n");
362         return -1;
363     }
364
365 #define RSC 1 //FIXME finetune
366     if(!s-> in.ch_count)
367         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
368     if(!s->used_ch_count)
369         s->used_ch_count= s->in.ch_count;
370     if(!s->out.ch_count)
371         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
372
373     if(!s-> in.ch_count){
374         av_assert0(!s->in_ch_layout);
375         av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
376         return -1;
377     }
378
379     if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
380         char l1[1024], l2[1024];
381         av_get_channel_layout_string(l1, sizeof(l1), s-> in.ch_count, s-> in_ch_layout);
382         av_get_channel_layout_string(l2, sizeof(l2), s->out.ch_count, s->out_ch_layout);
383         av_log(s, AV_LOG_ERROR, "Rematrix is needed between %s and %s "
384                "but there is not enough information to do it\n", l1, l2);
385         return -1;
386     }
387
388 av_assert0(s->used_ch_count);
389 av_assert0(s->out.ch_count);
390     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
391
392     s->in_buffer= s->in;
393     s->silence  = s->in;
394     s->drop_temp= s->out;
395
396     if(!s->resample && !s->rematrix && !s->channel_map && !s->dither.method){
397         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
398                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
399         return 0;
400     }
401
402     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
403                                              s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
404     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
405                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
406
407     if (!s->in_convert || !s->out_convert)
408         return AVERROR(ENOMEM);
409
410     s->postin= s->in;
411     s->preout= s->out;
412     s->midbuf= s->in;
413
414     if(s->channel_map){
415         s->postin.ch_count=
416         s->midbuf.ch_count= s->used_ch_count;
417         if(s->resample)
418             s->in_buffer.ch_count= s->used_ch_count;
419     }
420     if(!s->resample_first){
421         s->midbuf.ch_count= s->out.ch_count;
422         if(s->resample)
423             s->in_buffer.ch_count = s->out.ch_count;
424     }
425
426     set_audiodata_fmt(&s->postin, s->int_sample_fmt);
427     set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
428     set_audiodata_fmt(&s->preout, s->int_sample_fmt);
429
430     if(s->resample){
431         set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
432     }
433
434     if ((ret = swri_dither_init(s, s->out_sample_fmt, s->int_sample_fmt)) < 0)
435         return ret;
436
437     if(s->rematrix || s->dither.method)
438         return swri_rematrix_init(s);
439
440     return 0;
441 }
442
443 int swri_realloc_audio(AudioData *a, int count){
444     int i, countb;
445     AudioData old;
446
447     if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
448         return AVERROR(EINVAL);
449
450     if(a->count >= count)
451         return 0;
452
453     count*=2;
454
455     countb= FFALIGN(count*a->bps, ALIGN);
456     old= *a;
457
458     av_assert0(a->bps);
459     av_assert0(a->ch_count);
460
461     a->data= av_mallocz(countb*a->ch_count);
462     if(!a->data)
463         return AVERROR(ENOMEM);
464     for(i=0; i<a->ch_count; i++){
465         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
466         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
467     }
468     if(!a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
469     av_freep(&old.data);
470     a->count= count;
471
472     return 1;
473 }
474
475 static void copy(AudioData *out, AudioData *in,
476                  int count){
477     av_assert0(out->planar == in->planar);
478     av_assert0(out->bps == in->bps);
479     av_assert0(out->ch_count == in->ch_count);
480     if(out->planar){
481         int ch;
482         for(ch=0; ch<out->ch_count; ch++)
483             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
484     }else
485         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
486 }
487
488 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
489     int i;
490     if(!in_arg){
491         memset(out->ch, 0, sizeof(out->ch));
492     }else if(out->planar){
493         for(i=0; i<out->ch_count; i++)
494             out->ch[i]= in_arg[i];
495     }else{
496         for(i=0; i<out->ch_count; i++)
497             out->ch[i]= in_arg[0] + i*out->bps;
498     }
499 }
500
501 static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
502     int i;
503     if(out->planar){
504         for(i=0; i<out->ch_count; i++)
505             in_arg[i]= out->ch[i];
506     }else{
507         in_arg[0]= out->ch[0];
508     }
509 }
510
511 /**
512  *
513  * out may be equal in.
514  */
515 static void buf_set(AudioData *out, AudioData *in, int count){
516     int ch;
517     if(in->planar){
518         for(ch=0; ch<out->ch_count; ch++)
519             out->ch[ch]= in->ch[ch] + count*out->bps;
520     }else{
521         for(ch=out->ch_count-1; ch>=0; ch--)
522             out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
523     }
524 }
525
526 /**
527  *
528  * @return number of samples output per channel
529  */
530 static int resample(SwrContext *s, AudioData *out_param, int out_count,
531                              const AudioData * in_param, int in_count){
532     AudioData in, out, tmp;
533     int ret_sum=0;
534     int border=0;
535     int padless = ARCH_X86 && s->engine == SWR_ENGINE_SWR ? 7 : 0;
536
537     av_assert1(s->in_buffer.ch_count == in_param->ch_count);
538     av_assert1(s->in_buffer.planar   == in_param->planar);
539     av_assert1(s->in_buffer.fmt      == in_param->fmt);
540
541     tmp=out=*out_param;
542     in =  *in_param;
543
544     border = s->resampler->invert_initial_buffer(s->resample, &s->in_buffer,
545                  &in, in_count, &s->in_buffer_index, &s->in_buffer_count);
546     if (border == INT_MAX) return 0;
547     else if (border < 0) return border;
548     else if (border) { buf_set(&in, &in, border); in_count -= border; s->resample_in_constraint = 0; }
549
550     do{
551         int ret, size, consumed;
552         if(!s->resample_in_constraint && s->in_buffer_count){
553             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
554             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
555             out_count -= ret;
556             ret_sum += ret;
557             buf_set(&out, &out, ret);
558             s->in_buffer_count -= consumed;
559             s->in_buffer_index += consumed;
560
561             if(!in_count)
562                 break;
563             if(s->in_buffer_count <= border){
564                 buf_set(&in, &in, -s->in_buffer_count);
565                 in_count += s->in_buffer_count;
566                 s->in_buffer_count=0;
567                 s->in_buffer_index=0;
568                 border = 0;
569             }
570         }
571
572         if((s->flushed || in_count > padless) && !s->in_buffer_count){
573             s->in_buffer_index=0;
574             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &in, FFMAX(in_count-padless, 0), &consumed);
575             out_count -= ret;
576             ret_sum += ret;
577             buf_set(&out, &out, ret);
578             in_count -= consumed;
579             buf_set(&in, &in, consumed);
580         }
581
582         //TODO is this check sane considering the advanced copy avoidance below
583         size= s->in_buffer_index + s->in_buffer_count + in_count;
584         if(   size > s->in_buffer.count
585            && s->in_buffer_count + in_count <= s->in_buffer_index){
586             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
587             copy(&s->in_buffer, &tmp, s->in_buffer_count);
588             s->in_buffer_index=0;
589         }else
590             if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
591                 return ret;
592
593         if(in_count){
594             int count= in_count;
595             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
596
597             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
598             copy(&tmp, &in, /*in_*/count);
599             s->in_buffer_count += count;
600             in_count -= count;
601             border += count;
602             buf_set(&in, &in, count);
603             s->resample_in_constraint= 0;
604             if(s->in_buffer_count != count || in_count)
605                 continue;
606             if (padless) {
607                 padless = 0;
608                 continue;
609             }
610         }
611         break;
612     }while(1);
613
614     s->resample_in_constraint= !!out_count;
615
616     return ret_sum;
617 }
618
619 static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
620                                                       AudioData *in , int  in_count){
621     AudioData *postin, *midbuf, *preout;
622     int ret/*, in_max*/;
623     AudioData preout_tmp, midbuf_tmp;
624
625     if(s->full_convert){
626         av_assert0(!s->resample);
627         swri_audio_convert(s->full_convert, out, in, in_count);
628         return out_count;
629     }
630
631 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
632 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
633
634     if((ret=swri_realloc_audio(&s->postin, in_count))<0)
635         return ret;
636     if(s->resample_first){
637         av_assert0(s->midbuf.ch_count == s->used_ch_count);
638         if((ret=swri_realloc_audio(&s->midbuf, out_count))<0)
639             return ret;
640     }else{
641         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
642         if((ret=swri_realloc_audio(&s->midbuf,  in_count))<0)
643             return ret;
644     }
645     if((ret=swri_realloc_audio(&s->preout, out_count))<0)
646         return ret;
647
648     postin= &s->postin;
649
650     midbuf_tmp= s->midbuf;
651     midbuf= &midbuf_tmp;
652     preout_tmp= s->preout;
653     preout= &preout_tmp;
654
655     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
656         postin= in;
657
658     if(s->resample_first ? !s->resample : !s->rematrix)
659         midbuf= postin;
660
661     if(s->resample_first ? !s->rematrix : !s->resample)
662         preout= midbuf;
663
664     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar
665        && !(s->out_sample_fmt==AV_SAMPLE_FMT_S32P && (s->dither.output_sample_bits&31))){
666         if(preout==in){
667             out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
668             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
669             copy(out, in, out_count);
670             return out_count;
671         }
672         else if(preout==postin) preout= midbuf= postin= out;
673         else if(preout==midbuf) preout= midbuf= out;
674         else                    preout= out;
675     }
676
677     if(in != postin){
678         swri_audio_convert(s->in_convert, postin, in, in_count);
679     }
680
681     if(s->resample_first){
682         if(postin != midbuf)
683             out_count= resample(s, midbuf, out_count, postin, in_count);
684         if(midbuf != preout)
685             swri_rematrix(s, preout, midbuf, out_count, preout==out);
686     }else{
687         if(postin != midbuf)
688             swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
689         if(midbuf != preout)
690             out_count= resample(s, preout, out_count, midbuf, in_count);
691     }
692
693     if(preout != out && out_count){
694         AudioData *conv_src = preout;
695         if(s->dither.method){
696             int ch;
697             int dither_count= FFMAX(out_count, 1<<16);
698
699             if (preout == in) {
700                 conv_src = &s->dither.temp;
701                 if((ret=swri_realloc_audio(&s->dither.temp, dither_count))<0)
702                     return ret;
703             }
704
705             if((ret=swri_realloc_audio(&s->dither.noise, dither_count))<0)
706                 return ret;
707             if(ret)
708                 for(ch=0; ch<s->dither.noise.ch_count; ch++)
709                     swri_get_dither(s, s->dither.noise.ch[ch], s->dither.noise.count, 12345678913579<<ch, s->dither.noise.fmt);
710             av_assert0(s->dither.noise.ch_count == preout->ch_count);
711
712             if(s->dither.noise_pos + out_count > s->dither.noise.count)
713                 s->dither.noise_pos = 0;
714
715             if (s->dither.method < SWR_DITHER_NS){
716                 if (s->mix_2_1_simd) {
717                     int len1= out_count&~15;
718                     int off = len1 * preout->bps;
719
720                     if(len1)
721                         for(ch=0; ch<preout->ch_count; ch++)
722                             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_simd_one, 0, 0, len1);
723                     if(out_count != len1)
724                         for(ch=0; ch<preout->ch_count; ch++)
725                             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);
726                 } else {
727                     for(ch=0; ch<preout->ch_count; ch++)
728                         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);
729                 }
730             } else {
731                 switch(s->int_sample_fmt) {
732                 case AV_SAMPLE_FMT_S16P :swri_noise_shaping_int16(s, conv_src, preout, &s->dither.noise, out_count); break;
733                 case AV_SAMPLE_FMT_S32P :swri_noise_shaping_int32(s, conv_src, preout, &s->dither.noise, out_count); break;
734                 case AV_SAMPLE_FMT_FLTP :swri_noise_shaping_float(s, conv_src, preout, &s->dither.noise, out_count); break;
735                 case AV_SAMPLE_FMT_DBLP :swri_noise_shaping_double(s,conv_src, preout, &s->dither.noise, out_count); break;
736                 }
737             }
738             s->dither.noise_pos += out_count;
739         }
740 //FIXME packed doesn't need more than 1 chan here!
741         swri_audio_convert(s->out_convert, out, conv_src, out_count);
742     }
743     return out_count;
744 }
745
746 int swr_is_initialized(struct SwrContext *s) {
747     return !!s->in_buffer.ch_count;
748 }
749
750 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
751                                 const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
752     AudioData * in= &s->in;
753     AudioData *out= &s->out;
754
755     if (!swr_is_initialized(s)) {
756         av_log(s, AV_LOG_ERROR, "Context has not been initialized\n");
757         return AVERROR(EINVAL);
758     }
759
760     while(s->drop_output > 0){
761         int ret;
762         uint8_t *tmp_arg[SWR_CH_MAX];
763 #define MAX_DROP_STEP 16384
764         if((ret=swri_realloc_audio(&s->drop_temp, FFMIN(s->drop_output, MAX_DROP_STEP)))<0)
765             return ret;
766
767         reversefill_audiodata(&s->drop_temp, tmp_arg);
768         s->drop_output *= -1; //FIXME find a less hackish solution
769         ret = swr_convert(s, tmp_arg, FFMIN(-s->drop_output, MAX_DROP_STEP), in_arg, in_count); //FIXME optimize but this is as good as never called so maybe it doesn't matter
770         s->drop_output *= -1;
771         in_count = 0;
772         if(ret>0) {
773             s->drop_output -= ret;
774             continue;
775         }
776
777         if(s->drop_output || !out_arg)
778             return 0;
779     }
780
781     if(!in_arg){
782         if(s->resample){
783             if (!s->flushed)
784                 s->resampler->flush(s);
785             s->resample_in_constraint = 0;
786             s->flushed = 1;
787         }else if(!s->in_buffer_count){
788             return 0;
789         }
790     }else
791         fill_audiodata(in ,  (void*)in_arg);
792
793     fill_audiodata(out, out_arg);
794
795     if(s->resample){
796         int ret = swr_convert_internal(s, out, out_count, in, in_count);
797         if(ret>0 && !s->drop_output)
798             s->outpts += ret * (int64_t)s->in_sample_rate;
799         return ret;
800     }else{
801         AudioData tmp= *in;
802         int ret2=0;
803         int ret, size;
804         size = FFMIN(out_count, s->in_buffer_count);
805         if(size){
806             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
807             ret= swr_convert_internal(s, out, size, &tmp, size);
808             if(ret<0)
809                 return ret;
810             ret2= ret;
811             s->in_buffer_count -= ret;
812             s->in_buffer_index += ret;
813             buf_set(out, out, ret);
814             out_count -= ret;
815             if(!s->in_buffer_count)
816                 s->in_buffer_index = 0;
817         }
818
819         if(in_count){
820             size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
821
822             if(in_count > out_count) { //FIXME move after swr_convert_internal
823                 if(   size > s->in_buffer.count
824                 && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
825                     buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
826                     copy(&s->in_buffer, &tmp, s->in_buffer_count);
827                     s->in_buffer_index=0;
828                 }else
829                     if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
830                         return ret;
831             }
832
833             if(out_count){
834                 size = FFMIN(in_count, out_count);
835                 ret= swr_convert_internal(s, out, size, in, size);
836                 if(ret<0)
837                     return ret;
838                 buf_set(in, in, ret);
839                 in_count -= ret;
840                 ret2 += ret;
841             }
842             if(in_count){
843                 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
844                 copy(&tmp, in, in_count);
845                 s->in_buffer_count += in_count;
846             }
847         }
848         if(ret2>0 && !s->drop_output)
849             s->outpts += ret2 * (int64_t)s->in_sample_rate;
850         return ret2;
851     }
852 }
853
854 int swr_drop_output(struct SwrContext *s, int count){
855     s->drop_output += count;
856
857     if(s->drop_output <= 0)
858         return 0;
859
860     av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
861     return swr_convert(s, NULL, s->drop_output, NULL, 0);
862 }
863
864 int swr_inject_silence(struct SwrContext *s, int count){
865     int ret, i;
866     uint8_t *tmp_arg[SWR_CH_MAX];
867
868     if(count <= 0)
869         return 0;
870
871 #define MAX_SILENCE_STEP 16384
872     while (count > MAX_SILENCE_STEP) {
873         if ((ret = swr_inject_silence(s, MAX_SILENCE_STEP)) < 0)
874             return ret;
875         count -= MAX_SILENCE_STEP;
876     }
877
878     if((ret=swri_realloc_audio(&s->silence, count))<0)
879         return ret;
880
881     if(s->silence.planar) for(i=0; i<s->silence.ch_count; i++) {
882         memset(s->silence.ch[i], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps);
883     } else
884         memset(s->silence.ch[0], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps*s->silence.ch_count);
885
886     reversefill_audiodata(&s->silence, tmp_arg);
887     av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
888     ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
889     return ret;
890 }
891
892 int64_t swr_get_delay(struct SwrContext *s, int64_t base){
893     if (s->resampler && s->resample){
894         return s->resampler->get_delay(s, base);
895     }else{
896         return (s->in_buffer_count*base + (s->in_sample_rate>>1))/ s->in_sample_rate;
897     }
898 }
899
900 int swr_set_compensation(struct SwrContext *s, int sample_delta, int compensation_distance){
901     int ret;
902
903     if (!s || compensation_distance < 0)
904         return AVERROR(EINVAL);
905     if (!compensation_distance && sample_delta)
906         return AVERROR(EINVAL);
907     if (!s->resample) {
908         s->flags |= SWR_FLAG_RESAMPLE;
909         ret = swr_init(s);
910         if (ret < 0)
911             return ret;
912     }
913     if (!s->resampler->set_compensation){
914         return AVERROR(EINVAL);
915     }else{
916         return s->resampler->set_compensation(s->resample, sample_delta, compensation_distance);
917     }
918 }
919
920 int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
921     if(pts == INT64_MIN)
922         return s->outpts;
923
924     if (s->firstpts == AV_NOPTS_VALUE)
925         s->outpts = s->firstpts = pts;
926
927     if(s->min_compensation >= FLT_MAX) {
928         return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
929     } else {
930         int64_t delta = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate) - s->outpts + s->drop_output*(int64_t)s->in_sample_rate;
931         double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
932
933         if(fabs(fdelta) > s->min_compensation) {
934             if(s->outpts == s->firstpts || fabs(fdelta) > s->min_hard_compensation){
935                 int ret;
936                 if(delta > 0) ret = swr_inject_silence(s,  delta / s->out_sample_rate);
937                 else          ret = swr_drop_output   (s, -delta / s-> in_sample_rate);
938                 if(ret<0){
939                     av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
940                 }
941             } else if(s->soft_compensation_duration && s->max_soft_compensation) {
942                 int duration = s->out_sample_rate * s->soft_compensation_duration;
943                 double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
944                 int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
945                 av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
946                 swr_set_compensation(s, comp, duration);
947             }
948         }
949
950         return s->outpts;
951     }
952 }