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