swr: add a lowpass default so that each Resampler can have its own default.
[ffmpeg.git] / libswresample / swresample.c
1 /*
2  * Copyright (C) 2011-2012 Michael Niedermayer (michaelni@gmx.at)
3  *
4  * This file is part of libswresample
5  *
6  * libswresample is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU Lesser General Public
8  * License as published by the Free Software Foundation; either
9  * version 2.1 of the License, or (at your option) any later version.
10  *
11  * libswresample is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14  * Lesser General Public License for more details.
15  *
16  * You should have received a copy of the GNU Lesser General Public
17  * License along with libswresample; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19  */
20
21 #include "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
83 {"filter_size"          , "set resampling filter size"  , OFFSET(filter_size)    , AV_OPT_TYPE_INT  , {.i64=16                    }, 0      , INT_MAX   , PARAM },
84 {"phase_shift"          , "set resampling phase shift"  , OFFSET(phase_shift)    , AV_OPT_TYPE_INT  , {.i64=10                    }, 0      , 30        , PARAM },
85 {"linear_interp"        , "enable linear interpolation" , OFFSET(linear_interp)  , AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , 1         , PARAM },
86 {"cutoff"               , "set cutoff frequency ratio"  , OFFSET(cutoff)         , AV_OPT_TYPE_DOUBLE,{.dbl=0.                    }, 0      , 1         , PARAM },
87 {"resampler"            , "set resampling Engine"       , OFFSET(engine)         , AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_ENGINE_NB-1, PARAM, "resampler"},
88 {"swr"                  , "select SW Resampler"         , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_ENGINE_SWR        }, INT_MIN, INT_MAX   , PARAM, "resampler"},
89 {"min_comp"             , "set minimum difference between timestamps and audio data (in seconds) below which no timestamp compensation of either kind is applied"
90                                                         , OFFSET(min_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=FLT_MAX               }, 0      , FLT_MAX   , PARAM },
91 {"min_hard_comp"        , "set minimum difference between timestamps and audio data (in seconds) to trigger padding/trimming the data."
92                                                         , OFFSET(min_hard_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=0.1                   }, 0      , INT_MAX   , PARAM },
93 {"comp_duration"        , "set duration (in seconds) over which data is stretched/squeezed to make it match the timestamps."
94                                                         , OFFSET(soft_compensation_duration),AV_OPT_TYPE_FLOAT ,{.dbl=1                     }, 0      , INT_MAX   , PARAM },
95 {"max_soft_comp"        , "set maximum factor by which data is stretched/squeezed to make it match the timestamps."
96                                                         , OFFSET(max_soft_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=0                     }, INT_MIN, INT_MAX   , PARAM },
97
98 { "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" },
99     { "none",  "select none",               0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_NONE  }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
100     { "dolby", "select Dolby",              0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_DOLBY }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
101     { "dplii", "select Dolby Pro Logic II", 0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_DPLII }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
102
103 { "filter_type"         , "select 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" },
104     { "cubic"           , "select cubic"                , 0                      , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_CUBIC            }, INT_MIN, INT_MAX, PARAM, "filter_type" },
105     { "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" },
106     { "kaiser"          , "select Kaiser Windowed Sinc" , 0                      , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_KAISER           }, INT_MIN, INT_MAX, PARAM, "filter_type" },
107
108 { "kaiser_beta"         , "set Kaiser Window Beta"      , OFFSET(kaiser_beta)    , AV_OPT_TYPE_INT  , {.i64=9                     }, 2      , 16        , PARAM },
109
110 {0}
111 };
112
113 static const char* context_to_name(void* ptr) {
114     return "SWR";
115 }
116
117 static const AVClass av_class = {
118     .class_name                = "SWResampler",
119     .item_name                 = context_to_name,
120     .option                    = options,
121     .version                   = LIBAVUTIL_VERSION_INT,
122     .log_level_offset_offset   = OFFSET(log_level_offset),
123     .parent_log_context_offset = OFFSET(log_ctx),
124     .category                  = AV_CLASS_CATEGORY_SWRESAMPLER,
125 };
126
127 unsigned swresample_version(void)
128 {
129     av_assert0(LIBSWRESAMPLE_VERSION_MICRO >= 100);
130     return LIBSWRESAMPLE_VERSION_INT;
131 }
132
133 const char *swresample_configuration(void)
134 {
135     return FFMPEG_CONFIGURATION;
136 }
137
138 const char *swresample_license(void)
139 {
140 #define LICENSE_PREFIX "libswresample license: "
141     return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
142 }
143
144 int swr_set_channel_mapping(struct SwrContext *s, const int *channel_map){
145     if(!s || s->in_convert) // s needs to be allocated but not initialized
146         return AVERROR(EINVAL);
147     s->channel_map = channel_map;
148     return 0;
149 }
150
151 const AVClass *swr_get_class(void)
152 {
153     return &av_class;
154 }
155
156 av_cold struct SwrContext *swr_alloc(void){
157     SwrContext *s= av_mallocz(sizeof(SwrContext));
158     if(s){
159         s->av_class= &av_class;
160         av_opt_set_defaults(s);
161     }
162     return s;
163 }
164
165 struct SwrContext *swr_alloc_set_opts(struct SwrContext *s,
166                                       int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
167                                       int64_t  in_ch_layout, enum AVSampleFormat  in_sample_fmt, int  in_sample_rate,
168                                       int log_offset, void *log_ctx){
169     if(!s) s= swr_alloc();
170     if(!s) return NULL;
171
172     s->log_level_offset= log_offset;
173     s->log_ctx= log_ctx;
174
175     av_opt_set_int(s, "ocl", out_ch_layout,   0);
176     av_opt_set_int(s, "osf", out_sample_fmt,  0);
177     av_opt_set_int(s, "osr", out_sample_rate, 0);
178     av_opt_set_int(s, "icl", in_ch_layout,    0);
179     av_opt_set_int(s, "isf", in_sample_fmt,   0);
180     av_opt_set_int(s, "isr", in_sample_rate,  0);
181     av_opt_set_int(s, "tsf", AV_SAMPLE_FMT_NONE,   0);
182     av_opt_set_int(s, "ich", av_get_channel_layout_nb_channels(s-> in_ch_layout), 0);
183     av_opt_set_int(s, "och", av_get_channel_layout_nb_channels(s->out_ch_layout), 0);
184     av_opt_set_int(s, "uch", 0, 0);
185     return s;
186 }
187
188 static void set_audiodata_fmt(AudioData *a, enum AVSampleFormat fmt){
189     a->fmt   = fmt;
190     a->bps   = av_get_bytes_per_sample(fmt);
191     a->planar= av_sample_fmt_is_planar(fmt);
192 }
193
194 static void free_temp(AudioData *a){
195     av_free(a->data);
196     memset(a, 0, sizeof(*a));
197 }
198
199 av_cold void swr_free(SwrContext **ss){
200     SwrContext *s= *ss;
201     if(s){
202         free_temp(&s->postin);
203         free_temp(&s->midbuf);
204         free_temp(&s->preout);
205         free_temp(&s->in_buffer);
206         free_temp(&s->dither);
207         swri_audio_convert_free(&s-> in_convert);
208         swri_audio_convert_free(&s->out_convert);
209         swri_audio_convert_free(&s->full_convert);
210         if (s->resampler)
211             s->resampler->free(&s->resample);
212         swri_rematrix_free(s);
213     }
214
215     av_freep(ss);
216 }
217
218 av_cold int swr_init(struct SwrContext *s){
219     s->in_buffer_index= 0;
220     s->in_buffer_count= 0;
221     s->resample_in_constraint= 0;
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->dither);
227     memset(s->in.ch, 0, sizeof(s->in.ch));
228     memset(s->out.ch, 0, sizeof(s->out.ch));
229     swri_audio_convert_free(&s-> in_convert);
230     swri_audio_convert_free(&s->out_convert);
231     swri_audio_convert_free(&s->full_convert);
232     swri_rematrix_free(s);
233
234     s->flushed = 0;
235
236     if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
237         av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
238         return AVERROR(EINVAL);
239     }
240     if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
241         av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
242         return AVERROR(EINVAL);
243     }
244
245     if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
246         if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_S16P){
247             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
248         }else if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_FLTP){
249             s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
250         }else{
251             av_log(s, AV_LOG_DEBUG, "Using double precision mode\n");
252             s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
253         }
254     }
255
256     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16P
257         &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
258         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
259         &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
260         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));
261         return AVERROR(EINVAL);
262     }
263
264     switch(s->engine){
265         case SWR_ENGINE_SWR : s->resampler = &swri_resampler; break;
266         default:
267             av_log(s, AV_LOG_ERROR, "Requested resampling engine is unavailable\n");
268             return AVERROR(EINVAL);
269     }
270
271     set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
272     set_audiodata_fmt(&s->out, s->out_sample_fmt);
273
274     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
275         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);
276     }else
277         s->resampler->free(&s->resample);
278     if(    s->int_sample_fmt != AV_SAMPLE_FMT_S16P
279         && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
280         && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
281         && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
282         && s->resample){
283         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16/s32/flt/dbl\n");
284         return -1;
285     }
286
287     if(!s->used_ch_count)
288         s->used_ch_count= s->in.ch_count;
289
290     if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
291         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
292         s-> in_ch_layout= 0;
293     }
294
295     if(!s-> in_ch_layout)
296         s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
297     if(!s->out_ch_layout)
298         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
299
300     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
301                  s->rematrix_custom;
302
303 #define RSC 1 //FIXME finetune
304     if(!s-> in.ch_count)
305         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
306     if(!s->used_ch_count)
307         s->used_ch_count= s->in.ch_count;
308     if(!s->out.ch_count)
309         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
310
311     if(!s-> in.ch_count){
312         av_assert0(!s->in_ch_layout);
313         av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
314         return -1;
315     }
316
317     if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
318         av_log(s, AV_LOG_ERROR, "Rematrix is needed but there is not enough information to do it\n");
319         return -1;
320     }
321
322 av_assert0(s->used_ch_count);
323 av_assert0(s->out.ch_count);
324     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
325
326     s->in_buffer= s->in;
327
328     if(!s->resample && !s->rematrix && !s->channel_map && !s->dither_method){
329         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
330                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
331         return 0;
332     }
333
334     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
335                                              s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
336     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
337                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
338
339
340     s->postin= s->in;
341     s->preout= s->out;
342     s->midbuf= s->in;
343
344     if(s->channel_map){
345         s->postin.ch_count=
346         s->midbuf.ch_count= s->used_ch_count;
347         if(s->resample)
348             s->in_buffer.ch_count= s->used_ch_count;
349     }
350     if(!s->resample_first){
351         s->midbuf.ch_count= s->out.ch_count;
352         if(s->resample)
353             s->in_buffer.ch_count = s->out.ch_count;
354     }
355
356     set_audiodata_fmt(&s->postin, s->int_sample_fmt);
357     set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
358     set_audiodata_fmt(&s->preout, s->int_sample_fmt);
359
360     if(s->resample){
361         set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
362     }
363
364     s->dither = s->preout;
365
366     if(s->rematrix || s->dither_method)
367         return swri_rematrix_init(s);
368
369     return 0;
370 }
371
372 int swri_realloc_audio(AudioData *a, int count){
373     int i, countb;
374     AudioData old;
375
376     if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
377         return AVERROR(EINVAL);
378
379     if(a->count >= count)
380         return 0;
381
382     count*=2;
383
384     countb= FFALIGN(count*a->bps, ALIGN);
385     old= *a;
386
387     av_assert0(a->bps);
388     av_assert0(a->ch_count);
389
390     a->data= av_mallocz(countb*a->ch_count);
391     if(!a->data)
392         return AVERROR(ENOMEM);
393     for(i=0; i<a->ch_count; i++){
394         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
395         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
396     }
397     if(!a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
398     av_free(old.data);
399     a->count= count;
400
401     return 1;
402 }
403
404 static void copy(AudioData *out, AudioData *in,
405                  int count){
406     av_assert0(out->planar == in->planar);
407     av_assert0(out->bps == in->bps);
408     av_assert0(out->ch_count == in->ch_count);
409     if(out->planar){
410         int ch;
411         for(ch=0; ch<out->ch_count; ch++)
412             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
413     }else
414         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
415 }
416
417 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
418     int i;
419     if(!in_arg){
420         memset(out->ch, 0, sizeof(out->ch));
421     }else if(out->planar){
422         for(i=0; i<out->ch_count; i++)
423             out->ch[i]= in_arg[i];
424     }else{
425         for(i=0; i<out->ch_count; i++)
426             out->ch[i]= in_arg[0] + i*out->bps;
427     }
428 }
429
430 static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
431     int i;
432     if(out->planar){
433         for(i=0; i<out->ch_count; i++)
434             in_arg[i]= out->ch[i];
435     }else{
436         in_arg[0]= out->ch[0];
437     }
438 }
439
440 /**
441  *
442  * out may be equal in.
443  */
444 static void buf_set(AudioData *out, AudioData *in, int count){
445     int ch;
446     if(in->planar){
447         for(ch=0; ch<out->ch_count; ch++)
448             out->ch[ch]= in->ch[ch] + count*out->bps;
449     }else{
450         for(ch=out->ch_count-1; ch>=0; ch--)
451             out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
452     }
453 }
454
455 /**
456  *
457  * @return number of samples output per channel
458  */
459 static int resample(SwrContext *s, AudioData *out_param, int out_count,
460                              const AudioData * in_param, int in_count){
461     AudioData in, out, tmp;
462     int ret_sum=0;
463     int border=0;
464
465     av_assert1(s->in_buffer.ch_count == in_param->ch_count);
466     av_assert1(s->in_buffer.planar   == in_param->planar);
467     av_assert1(s->in_buffer.fmt      == in_param->fmt);
468
469     tmp=out=*out_param;
470     in =  *in_param;
471
472     do{
473         int ret, size, consumed;
474         if(!s->resample_in_constraint && s->in_buffer_count){
475             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
476             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
477             out_count -= ret;
478             ret_sum += ret;
479             buf_set(&out, &out, ret);
480             s->in_buffer_count -= consumed;
481             s->in_buffer_index += consumed;
482
483             if(!in_count)
484                 break;
485             if(s->in_buffer_count <= border){
486                 buf_set(&in, &in, -s->in_buffer_count);
487                 in_count += s->in_buffer_count;
488                 s->in_buffer_count=0;
489                 s->in_buffer_index=0;
490                 border = 0;
491             }
492         }
493
494         if(in_count && !s->in_buffer_count){
495             s->in_buffer_index=0;
496             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &in, in_count, &consumed);
497             out_count -= ret;
498             ret_sum += ret;
499             buf_set(&out, &out, ret);
500             in_count -= consumed;
501             buf_set(&in, &in, consumed);
502         }
503
504         //TODO is this check sane considering the advanced copy avoidance below
505         size= s->in_buffer_index + s->in_buffer_count + in_count;
506         if(   size > s->in_buffer.count
507            && s->in_buffer_count + in_count <= s->in_buffer_index){
508             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
509             copy(&s->in_buffer, &tmp, s->in_buffer_count);
510             s->in_buffer_index=0;
511         }else
512             if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
513                 return ret;
514
515         if(in_count){
516             int count= in_count;
517             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
518
519             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
520             copy(&tmp, &in, /*in_*/count);
521             s->in_buffer_count += count;
522             in_count -= count;
523             border += count;
524             buf_set(&in, &in, count);
525             s->resample_in_constraint= 0;
526             if(s->in_buffer_count != count || in_count)
527                 continue;
528         }
529         break;
530     }while(1);
531
532     s->resample_in_constraint= !!out_count;
533
534     return ret_sum;
535 }
536
537 static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
538                                                       AudioData *in , int  in_count){
539     AudioData *postin, *midbuf, *preout;
540     int ret/*, in_max*/;
541     AudioData preout_tmp, midbuf_tmp;
542
543     if(s->full_convert){
544         av_assert0(!s->resample);
545         swri_audio_convert(s->full_convert, out, in, in_count);
546         return out_count;
547     }
548
549 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
550 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
551
552     if((ret=swri_realloc_audio(&s->postin, in_count))<0)
553         return ret;
554     if(s->resample_first){
555         av_assert0(s->midbuf.ch_count == s->used_ch_count);
556         if((ret=swri_realloc_audio(&s->midbuf, out_count))<0)
557             return ret;
558     }else{
559         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
560         if((ret=swri_realloc_audio(&s->midbuf,  in_count))<0)
561             return ret;
562     }
563     if((ret=swri_realloc_audio(&s->preout, out_count))<0)
564         return ret;
565
566     postin= &s->postin;
567
568     midbuf_tmp= s->midbuf;
569     midbuf= &midbuf_tmp;
570     preout_tmp= s->preout;
571     preout= &preout_tmp;
572
573     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
574         postin= in;
575
576     if(s->resample_first ? !s->resample : !s->rematrix)
577         midbuf= postin;
578
579     if(s->resample_first ? !s->rematrix : !s->resample)
580         preout= midbuf;
581
582     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar){
583         if(preout==in){
584             out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
585             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
586             copy(out, in, out_count);
587             return out_count;
588         }
589         else if(preout==postin) preout= midbuf= postin= out;
590         else if(preout==midbuf) preout= midbuf= out;
591         else                    preout= out;
592     }
593
594     if(in != postin){
595         swri_audio_convert(s->in_convert, postin, in, in_count);
596     }
597
598     if(s->resample_first){
599         if(postin != midbuf)
600             out_count= resample(s, midbuf, out_count, postin, in_count);
601         if(midbuf != preout)
602             swri_rematrix(s, preout, midbuf, out_count, preout==out);
603     }else{
604         if(postin != midbuf)
605             swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
606         if(midbuf != preout)
607             out_count= resample(s, preout, out_count, midbuf, in_count);
608     }
609
610     if(preout != out && out_count){
611         if(s->dither_method){
612             int ch;
613             int dither_count= FFMAX(out_count, 1<<16);
614             av_assert0(preout != in);
615
616             if((ret=swri_realloc_audio(&s->dither, dither_count))<0)
617                 return ret;
618             if(ret)
619                 for(ch=0; ch<s->dither.ch_count; ch++)
620                     swri_get_dither(s, s->dither.ch[ch], s->dither.count, 12345678913579<<ch, s->out_sample_fmt, s->int_sample_fmt);
621             av_assert0(s->dither.ch_count == preout->ch_count);
622
623             if(s->dither_pos + out_count > s->dither.count)
624                 s->dither_pos = 0;
625
626             for(ch=0; ch<preout->ch_count; ch++)
627                 s->mix_2_1_f(preout->ch[ch], preout->ch[ch], s->dither.ch[ch] + s->dither.bps * s->dither_pos, s->native_one, 0, 0, out_count);
628
629             s->dither_pos += out_count;
630         }
631 //FIXME packed doesnt need more than 1 chan here!
632         swri_audio_convert(s->out_convert, out, preout, out_count);
633     }
634     return out_count;
635 }
636
637 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
638                                 const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
639     AudioData * in= &s->in;
640     AudioData *out= &s->out;
641
642     if(s->drop_output > 0){
643         int ret;
644         AudioData tmp = s->out;
645         uint8_t *tmp_arg[SWR_CH_MAX];
646         tmp.count = 0;
647         tmp.data  = NULL;
648         if((ret=swri_realloc_audio(&tmp, s->drop_output))<0)
649             return ret;
650
651         reversefill_audiodata(&tmp, tmp_arg);
652         s->drop_output *= -1; //FIXME find a less hackish solution
653         ret = swr_convert(s, tmp_arg, -s->drop_output, in_arg, in_count); //FIXME optimize but this is as good as never called so maybe it doesnt matter
654         s->drop_output *= -1;
655         if(ret>0)
656             s->drop_output -= ret;
657
658         av_freep(&tmp.data);
659         if(s->drop_output || !out_arg)
660             return 0;
661         in_count = 0;
662     }
663
664     if(!in_arg){
665         if(s->resample){
666             if (!s->flushed)
667                 s->resampler->flush(s);
668             s->resample_in_constraint = 0;
669             s->flushed = 1;
670         }else if(!s->in_buffer_count){
671             return 0;
672         }
673     }else
674         fill_audiodata(in ,  (void*)in_arg);
675
676     fill_audiodata(out, out_arg);
677
678     if(s->resample){
679         int ret = swr_convert_internal(s, out, out_count, in, in_count);
680         if(ret>0 && !s->drop_output)
681             s->outpts += ret * (int64_t)s->in_sample_rate;
682         return ret;
683     }else{
684         AudioData tmp= *in;
685         int ret2=0;
686         int ret, size;
687         size = FFMIN(out_count, s->in_buffer_count);
688         if(size){
689             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
690             ret= swr_convert_internal(s, out, size, &tmp, size);
691             if(ret<0)
692                 return ret;
693             ret2= ret;
694             s->in_buffer_count -= ret;
695             s->in_buffer_index += ret;
696             buf_set(out, out, ret);
697             out_count -= ret;
698             if(!s->in_buffer_count)
699                 s->in_buffer_index = 0;
700         }
701
702         if(in_count){
703             size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
704
705             if(in_count > out_count) { //FIXME move after swr_convert_internal
706                 if(   size > s->in_buffer.count
707                 && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
708                     buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
709                     copy(&s->in_buffer, &tmp, s->in_buffer_count);
710                     s->in_buffer_index=0;
711                 }else
712                     if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
713                         return ret;
714             }
715
716             if(out_count){
717                 size = FFMIN(in_count, out_count);
718                 ret= swr_convert_internal(s, out, size, in, size);
719                 if(ret<0)
720                     return ret;
721                 buf_set(in, in, ret);
722                 in_count -= ret;
723                 ret2 += ret;
724             }
725             if(in_count){
726                 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
727                 copy(&tmp, in, in_count);
728                 s->in_buffer_count += in_count;
729             }
730         }
731         if(ret2>0 && !s->drop_output)
732             s->outpts += ret2 * (int64_t)s->in_sample_rate;
733         return ret2;
734     }
735 }
736
737 int swr_drop_output(struct SwrContext *s, int count){
738     s->drop_output += count;
739
740     if(s->drop_output <= 0)
741         return 0;
742
743     av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
744     return swr_convert(s, NULL, s->drop_output, NULL, 0);
745 }
746
747 int swr_inject_silence(struct SwrContext *s, int count){
748     int ret, i;
749     AudioData silence = s->in;
750     uint8_t *tmp_arg[SWR_CH_MAX];
751
752     if(count <= 0)
753         return 0;
754
755     silence.count = 0;
756     silence.data  = NULL;
757     if((ret=swri_realloc_audio(&silence, count))<0)
758         return ret;
759
760     if(silence.planar) for(i=0; i<silence.ch_count; i++) {
761         memset(silence.ch[i], silence.bps==1 ? 0x80 : 0, count*silence.bps);
762     } else
763         memset(silence.ch[0], silence.bps==1 ? 0x80 : 0, count*silence.bps*silence.ch_count);
764
765     reversefill_audiodata(&silence, tmp_arg);
766     av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
767     ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
768     av_freep(&silence.data);
769     return ret;
770 }
771
772 int64_t swr_get_delay(struct SwrContext *s, int64_t base){
773     if (s->resampler && s->resample){
774         return s->resampler->get_delay(s, base);
775     }else{
776         return (s->in_buffer_count*base + (s->in_sample_rate>>1))/ s->in_sample_rate;
777     }
778 }
779
780 int swr_set_compensation(struct SwrContext *s, int sample_delta, int compensation_distance){
781     int ret;
782
783     if (!s || compensation_distance < 0)
784         return AVERROR(EINVAL);
785     if (!compensation_distance && sample_delta)
786         return AVERROR(EINVAL);
787     if (!s->resample) {
788         s->flags |= SWR_FLAG_RESAMPLE;
789         ret = swr_init(s);
790         if (ret < 0)
791             return ret;
792     }
793     if (!s->resampler->set_compensation){
794         return AVERROR(EINVAL);
795     }else{
796         return s->resampler->set_compensation(s->resample, sample_delta, compensation_distance);
797     }
798 }
799
800 int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
801     if(pts == INT64_MIN)
802         return s->outpts;
803     if(s->min_compensation >= FLT_MAX) {
804         return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
805     } else {
806         int64_t delta = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate) - s->outpts;
807         double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
808
809         if(fabs(fdelta) > s->min_compensation) {
810             if(!s->outpts || fabs(fdelta) > s->min_hard_compensation){
811                 int ret;
812                 if(delta > 0) ret = swr_inject_silence(s,  delta / s->out_sample_rate);
813                 else          ret = swr_drop_output   (s, -delta / s-> in_sample_rate);
814                 if(ret<0){
815                     av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
816                 }
817             } else if(s->soft_compensation_duration && s->max_soft_compensation) {
818                 int duration = s->out_sample_rate * s->soft_compensation_duration;
819                 double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
820                 int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
821                 av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
822                 swr_set_compensation(s, comp, duration);
823             }
824         }
825
826         return s->outpts;
827     }
828 }