swsresample: Fix unprotected inline asm
[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/audioconvert.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"                  ,  "Input Channel Count"        , OFFSET( in.ch_count   ), AV_OPT_TYPE_INT  , {.i64=2                     }, 0      , SWR_CH_MAX, PARAM},
45 {"in_channel_count"     ,  "Input Channel Count"        , OFFSET( in.ch_count   ), AV_OPT_TYPE_INT  , {.i64=2                     }, 0      , SWR_CH_MAX, PARAM},
46 {"och"                  , "Output Channel Count"        , OFFSET(out.ch_count   ), AV_OPT_TYPE_INT  , {.i64=2                     }, 0      , SWR_CH_MAX, PARAM},
47 {"out_channel_count"    , "Output Channel Count"        , OFFSET(out.ch_count   ), AV_OPT_TYPE_INT  , {.i64=2                     }, 0      , SWR_CH_MAX, PARAM},
48 {"uch"                  ,   "Used Channel Count"        , OFFSET(used_ch_count  ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_CH_MAX, PARAM},
49 {"used_channel_count"   ,   "Used Channel Count"        , OFFSET(used_ch_count  ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_CH_MAX, PARAM},
50 {"isr"                  ,  "Input Sample Rate"          , OFFSET( in_sample_rate), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , INT_MAX   , PARAM},
51 {"in_sample_rate"       ,  "Input Sample Rate"          , OFFSET( in_sample_rate), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , INT_MAX   , PARAM},
52 {"osr"                  , "Output Sample Rate"          , OFFSET(out_sample_rate), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , INT_MAX   , PARAM},
53 {"out_sample_rate"      , "Output Sample Rate"          , OFFSET(out_sample_rate), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , INT_MAX   , PARAM},
54 {"isf"                  ,    "Input Sample Format"      , OFFSET( in_sample_fmt ), AV_OPT_TYPE_INT  , {.i64=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_NB-1+256, PARAM},
55 {"in_sample_fmt"        ,    "Input Sample Format"      , OFFSET( in_sample_fmt ), AV_OPT_TYPE_INT  , {.i64=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_NB-1+256, PARAM},
56 {"osf"                  ,   "Output Sample Format"      , OFFSET(out_sample_fmt ), AV_OPT_TYPE_INT  , {.i64=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_NB-1+256, PARAM},
57 {"out_sample_fmt"       ,   "Output Sample Format"      , OFFSET(out_sample_fmt ), AV_OPT_TYPE_INT  , {.i64=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_NB-1+256, PARAM},
58 {"tsf"                  , "Internal Sample Format"      , OFFSET(int_sample_fmt ), AV_OPT_TYPE_INT  , {.i64=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_FLTP, PARAM},
59 {"internal_sample_fmt"  , "Internal Sample Format"      , OFFSET(int_sample_fmt ), AV_OPT_TYPE_INT  , {.i64=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_FLTP, PARAM},
60 {"icl"                  ,   "Input Channel Layout"      , OFFSET( in_ch_layout  ), AV_OPT_TYPE_INT64, {.i64=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
61 {"in_channel_layout"    ,   "Input Channel Layout"      , OFFSET( in_ch_layout  ), AV_OPT_TYPE_INT64, {.i64=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
62 {"ocl"                  ,  "Output Channel Layout"      , OFFSET(out_ch_layout  ), AV_OPT_TYPE_INT64, {.i64=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
63 {"out_channel_layout"   ,  "Output Channel Layout"      , OFFSET(out_ch_layout  ), AV_OPT_TYPE_INT64, {.i64=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
64 {"clev"                 ,    "Center Mix Level"         , OFFSET(clev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
65 {"center_mix_level"     ,    "Center Mix Level"         , OFFSET(clev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
66 {"slev"                 , "Sourround Mix Level"         , OFFSET(slev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
67 {"surround_mix_level"   , "Sourround Mix Level"         , OFFSET(slev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
68 {"lfe_mix_level"        , "LFE Mix Level"               , OFFSET(lfe_mix_level  ), AV_OPT_TYPE_FLOAT, {.dbl=0                     }, -32    , 32        , PARAM},
69 {"rmvol"                , "Rematrix Volume"             , OFFSET(rematrix_volume), AV_OPT_TYPE_FLOAT, {.dbl=1.0                   }, -1000  , 1000      , PARAM},
70 {"rematrix_volume"      , "Rematrix Volume"             , OFFSET(rematrix_volume), AV_OPT_TYPE_FLOAT, {.dbl=1.0                   }, -1000  , 1000      , PARAM},
71 {"flags"                , NULL                          , OFFSET(flags          ), AV_OPT_TYPE_FLAGS, {.i64=0                     }, 0      , UINT_MAX  , PARAM, "flags"},
72 {"swr_flags"            , NULL                          , OFFSET(flags          ), AV_OPT_TYPE_FLAGS, {.i64=0                     }, 0      , UINT_MAX  , PARAM, "flags"},
73 {"res"                  , "Force Resampling"            , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_FLAG_RESAMPLE     }, INT_MIN, INT_MAX   , PARAM, "flags"},
74 {"dither_scale"         , "Dither Scale"                , OFFSET(dither_scale   ), AV_OPT_TYPE_FLOAT, {.dbl=1                     }, 0      , INT_MAX   , PARAM},
75 {"dither_method"        , "Dither Method"               , OFFSET(dither_method  ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_DITHER_NB-1, PARAM, "dither_method"},
76 {"rectangular"          , "Rectangular Dither"          , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_RECTANGULAR}, INT_MIN, INT_MAX   , PARAM, "dither_method"},
77 {"triangular"           ,  "Triangular Dither"          , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_TRIANGULAR }, INT_MIN, INT_MAX   , PARAM, "dither_method"},
78 {"triangular_hp"        , "Triangular Dither With High Pass" , 0                 , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_TRIANGULAR_HIGHPASS }, INT_MIN, INT_MAX, PARAM, "dither_method"},
79 {"filter_size"          , "Resampling Filter Size"      , OFFSET(filter_size)    , AV_OPT_TYPE_INT  , {.i64=16                    }, 0      , INT_MAX   , PARAM },
80 {"phase_shift"          , "Resampling Phase Shift"      , OFFSET(phase_shift)    , AV_OPT_TYPE_INT  , {.i64=10                    }, 0      , 30        , PARAM },
81 {"linear_interp"        , "Use Linear Interpolation"    , OFFSET(linear_interp)  , AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , 1         , PARAM },
82 {"cutoff"               , "Cutoff Frequency Ratio"      , OFFSET(cutoff)         , AV_OPT_TYPE_DOUBLE,{.dbl=0.8                   }, 0      , 1         , PARAM },
83 {"min_comp"             , "Minimum difference between timestamps and audio data (in seconds) below which no timestamp compensation of either kind is applied"
84                                                         , OFFSET(min_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=FLT_MAX               }, 0      , FLT_MAX   , PARAM },
85 {"min_hard_comp"        , "Minimum difference between timestamps and audio data (in seconds) to trigger padding/trimming the data."
86                                                    , OFFSET(min_hard_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=0.1                   }, 0      , INT_MAX   , PARAM },
87 {"comp_duration"        , "Duration (in seconds) over which data is stretched/squeezed to make it match the timestamps."
88                                               , OFFSET(soft_compensation_duration),AV_OPT_TYPE_FLOAT ,{.dbl=1                     }, 0      , INT_MAX   , PARAM },
89 {"max_soft_comp"        , "Maximum factor by which data is stretched/squeezed to make it match the timestamps."
90                                                    , OFFSET(max_soft_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=0                     }, INT_MIN, INT_MAX   , PARAM },
91 { "matrix_encoding"     , "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" },
92     { "none",  "None",               0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_NONE  }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
93     { "dolby", "Dolby",              0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_DOLBY }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
94     { "dplii", "Dolby Pro Logic II", 0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_DPLII }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
95 { "filter_type"         , "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" },
96     { "cubic"           , "Cubic"                       , 0                      , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_CUBIC            }, INT_MIN, INT_MAX, PARAM, "filter_type" },
97     { "blackman_nuttall", "Blackman Nuttall Windowed Sinc", 0                    , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_BLACKMAN_NUTTALL }, INT_MIN, INT_MAX, PARAM, "filter_type" },
98     { "kaiser"          , "Kaiser Windowed Sinc"        , 0                      , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_KAISER           }, INT_MIN, INT_MAX, PARAM, "filter_type" },
99 { "kaiser_beta"         , "Kaiser Window Beta"          ,OFFSET(kaiser_beta)     , AV_OPT_TYPE_INT  , {.i64=9                     }, 2      , 16        , PARAM },
100
101 {0}
102 };
103
104 static const char* context_to_name(void* ptr) {
105     return "SWR";
106 }
107
108 static const AVClass av_class = {
109     .class_name                = "SWResampler",
110     .item_name                 = context_to_name,
111     .option                    = options,
112     .version                   = LIBAVUTIL_VERSION_INT,
113     .log_level_offset_offset   = OFFSET(log_level_offset),
114     .parent_log_context_offset = OFFSET(log_ctx),
115     .category                  = AV_CLASS_CATEGORY_SWRESAMPLER,
116 };
117
118 unsigned swresample_version(void)
119 {
120     av_assert0(LIBSWRESAMPLE_VERSION_MICRO >= 100);
121     return LIBSWRESAMPLE_VERSION_INT;
122 }
123
124 const char *swresample_configuration(void)
125 {
126     return FFMPEG_CONFIGURATION;
127 }
128
129 const char *swresample_license(void)
130 {
131 #define LICENSE_PREFIX "libswresample license: "
132     return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
133 }
134
135 int swr_set_channel_mapping(struct SwrContext *s, const int *channel_map){
136     if(!s || s->in_convert) // s needs to be allocated but not initialized
137         return AVERROR(EINVAL);
138     s->channel_map = channel_map;
139     return 0;
140 }
141
142 const AVClass *swr_get_class(void)
143 {
144     return &av_class;
145 }
146
147 struct SwrContext *swr_alloc(void){
148     SwrContext *s= av_mallocz(sizeof(SwrContext));
149     if(s){
150         s->av_class= &av_class;
151         av_opt_set_defaults(s);
152     }
153     return s;
154 }
155
156 struct SwrContext *swr_alloc_set_opts(struct SwrContext *s,
157                                       int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
158                                       int64_t  in_ch_layout, enum AVSampleFormat  in_sample_fmt, int  in_sample_rate,
159                                       int log_offset, void *log_ctx){
160     if(!s) s= swr_alloc();
161     if(!s) return NULL;
162
163     s->log_level_offset= log_offset;
164     s->log_ctx= log_ctx;
165
166     av_opt_set_int(s, "ocl", out_ch_layout,   0);
167     av_opt_set_int(s, "osf", out_sample_fmt,  0);
168     av_opt_set_int(s, "osr", out_sample_rate, 0);
169     av_opt_set_int(s, "icl", in_ch_layout,    0);
170     av_opt_set_int(s, "isf", in_sample_fmt,   0);
171     av_opt_set_int(s, "isr", in_sample_rate,  0);
172     av_opt_set_int(s, "tsf", AV_SAMPLE_FMT_NONE,   0);
173     av_opt_set_int(s, "ich", av_get_channel_layout_nb_channels(s-> in_ch_layout), 0);
174     av_opt_set_int(s, "och", av_get_channel_layout_nb_channels(s->out_ch_layout), 0);
175     av_opt_set_int(s, "uch", 0, 0);
176     return s;
177 }
178
179 static void set_audiodata_fmt(AudioData *a, enum AVSampleFormat fmt){
180     a->fmt   = fmt;
181     a->bps   = av_get_bytes_per_sample(fmt);
182     a->planar= av_sample_fmt_is_planar(fmt);
183 }
184
185 static void free_temp(AudioData *a){
186     av_free(a->data);
187     memset(a, 0, sizeof(*a));
188 }
189
190 void swr_free(SwrContext **ss){
191     SwrContext *s= *ss;
192     if(s){
193         free_temp(&s->postin);
194         free_temp(&s->midbuf);
195         free_temp(&s->preout);
196         free_temp(&s->in_buffer);
197         free_temp(&s->dither);
198         swri_audio_convert_free(&s-> in_convert);
199         swri_audio_convert_free(&s->out_convert);
200         swri_audio_convert_free(&s->full_convert);
201         swri_resample_free(&s->resample);
202         swri_rematrix_free(s);
203     }
204
205     av_freep(ss);
206 }
207
208 int swr_init(struct SwrContext *s){
209     s->in_buffer_index= 0;
210     s->in_buffer_count= 0;
211     s->resample_in_constraint= 0;
212     free_temp(&s->postin);
213     free_temp(&s->midbuf);
214     free_temp(&s->preout);
215     free_temp(&s->in_buffer);
216     free_temp(&s->dither);
217     swri_audio_convert_free(&s-> in_convert);
218     swri_audio_convert_free(&s->out_convert);
219     swri_audio_convert_free(&s->full_convert);
220     swri_rematrix_free(s);
221
222     s->flushed = 0;
223
224     if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
225         av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
226         return AVERROR(EINVAL);
227     }
228     if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
229         av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
230         return AVERROR(EINVAL);
231     }
232
233     if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
234         if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_S16P){
235             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
236         }else if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_FLTP){
237             s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
238         }else{
239             av_log(s, AV_LOG_DEBUG, "Using double precision mode\n");
240             s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
241         }
242     }
243
244     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16P
245         &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
246         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
247         &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
248         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));
249         return AVERROR(EINVAL);
250     }
251
252     set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
253     set_audiodata_fmt(&s->out, s->out_sample_fmt);
254
255     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
256         s->resample = swri_resample_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);
257     }else
258         swri_resample_free(&s->resample);
259     if(    s->int_sample_fmt != AV_SAMPLE_FMT_S16P
260         && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
261         && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
262         && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
263         && s->resample){
264         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16/s32/flt/dbl\n");
265         return -1;
266     }
267
268     if(!s->used_ch_count)
269         s->used_ch_count= s->in.ch_count;
270
271     if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
272         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
273         s-> in_ch_layout= 0;
274     }
275
276     if(!s-> in_ch_layout)
277         s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
278     if(!s->out_ch_layout)
279         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
280
281     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
282                  s->rematrix_custom;
283
284 #define RSC 1 //FIXME finetune
285     if(!s-> in.ch_count)
286         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
287     if(!s->used_ch_count)
288         s->used_ch_count= s->in.ch_count;
289     if(!s->out.ch_count)
290         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
291
292     if(!s-> in.ch_count){
293         av_assert0(!s->in_ch_layout);
294         av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
295         return -1;
296     }
297
298     if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
299         av_log(s, AV_LOG_ERROR, "Rematrix is needed but there is not enough information to do it\n");
300         return -1;
301     }
302
303 av_assert0(s->used_ch_count);
304 av_assert0(s->out.ch_count);
305     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
306
307     s->in_buffer= s->in;
308
309     if(!s->resample && !s->rematrix && !s->channel_map && !s->dither_method){
310         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
311                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
312         return 0;
313     }
314
315     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
316                                              s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
317     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
318                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
319
320
321     s->postin= s->in;
322     s->preout= s->out;
323     s->midbuf= s->in;
324
325     if(s->channel_map){
326         s->postin.ch_count=
327         s->midbuf.ch_count= s->used_ch_count;
328         if(s->resample)
329             s->in_buffer.ch_count= s->used_ch_count;
330     }
331     if(!s->resample_first){
332         s->midbuf.ch_count= s->out.ch_count;
333         if(s->resample)
334             s->in_buffer.ch_count = s->out.ch_count;
335     }
336
337     set_audiodata_fmt(&s->postin, s->int_sample_fmt);
338     set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
339     set_audiodata_fmt(&s->preout, s->int_sample_fmt);
340
341     if(s->resample){
342         set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
343     }
344
345     s->dither = s->preout;
346
347     if(s->rematrix || s->dither_method)
348         return swri_rematrix_init(s);
349
350     return 0;
351 }
352
353 static int realloc_audio(AudioData *a, int count){
354     int i, countb;
355     AudioData old;
356
357     if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
358         return AVERROR(EINVAL);
359
360     if(a->count >= count)
361         return 0;
362
363     count*=2;
364
365     countb= FFALIGN(count*a->bps, ALIGN);
366     old= *a;
367
368     av_assert0(a->bps);
369     av_assert0(a->ch_count);
370
371     a->data= av_mallocz(countb*a->ch_count);
372     if(!a->data)
373         return AVERROR(ENOMEM);
374     for(i=0; i<a->ch_count; i++){
375         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
376         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
377     }
378     if(!a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
379     av_free(old.data);
380     a->count= count;
381
382     return 1;
383 }
384
385 static void copy(AudioData *out, AudioData *in,
386                  int count){
387     av_assert0(out->planar == in->planar);
388     av_assert0(out->bps == in->bps);
389     av_assert0(out->ch_count == in->ch_count);
390     if(out->planar){
391         int ch;
392         for(ch=0; ch<out->ch_count; ch++)
393             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
394     }else
395         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
396 }
397
398 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
399     int i;
400     if(!in_arg){
401         memset(out->ch, 0, sizeof(out->ch));
402     }else if(out->planar){
403         for(i=0; i<out->ch_count; i++)
404             out->ch[i]= in_arg[i];
405     }else{
406         for(i=0; i<out->ch_count; i++)
407             out->ch[i]= in_arg[0] + i*out->bps;
408     }
409 }
410
411 static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
412     int i;
413     if(out->planar){
414         for(i=0; i<out->ch_count; i++)
415             in_arg[i]= out->ch[i];
416     }else{
417         in_arg[0]= out->ch[0];
418     }
419 }
420
421 /**
422  *
423  * out may be equal in.
424  */
425 static void buf_set(AudioData *out, AudioData *in, int count){
426     int ch;
427     if(in->planar){
428         for(ch=0; ch<out->ch_count; ch++)
429             out->ch[ch]= in->ch[ch] + count*out->bps;
430     }else{
431         for(ch=out->ch_count-1; ch>=0; ch--)
432             out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
433     }
434 }
435
436 /**
437  *
438  * @return number of samples output per channel
439  */
440 static int resample(SwrContext *s, AudioData *out_param, int out_count,
441                              const AudioData * in_param, int in_count){
442     AudioData in, out, tmp;
443     int ret_sum=0;
444     int border=0;
445
446     av_assert1(s->in_buffer.ch_count == in_param->ch_count);
447     av_assert1(s->in_buffer.planar   == in_param->planar);
448     av_assert1(s->in_buffer.fmt      == in_param->fmt);
449
450     tmp=out=*out_param;
451     in =  *in_param;
452
453     do{
454         int ret, size, consumed;
455         if(!s->resample_in_constraint && s->in_buffer_count){
456             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
457             ret= swri_multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
458             out_count -= ret;
459             ret_sum += ret;
460             buf_set(&out, &out, ret);
461             s->in_buffer_count -= consumed;
462             s->in_buffer_index += consumed;
463
464             if(!in_count)
465                 break;
466             if(s->in_buffer_count <= border){
467                 buf_set(&in, &in, -s->in_buffer_count);
468                 in_count += s->in_buffer_count;
469                 s->in_buffer_count=0;
470                 s->in_buffer_index=0;
471                 border = 0;
472             }
473         }
474
475         if(in_count && !s->in_buffer_count){
476             s->in_buffer_index=0;
477             ret= swri_multiple_resample(s->resample, &out, out_count, &in, in_count, &consumed);
478             out_count -= ret;
479             ret_sum += ret;
480             buf_set(&out, &out, ret);
481             in_count -= consumed;
482             buf_set(&in, &in, consumed);
483         }
484
485         //TODO is this check sane considering the advanced copy avoidance below
486         size= s->in_buffer_index + s->in_buffer_count + in_count;
487         if(   size > s->in_buffer.count
488            && s->in_buffer_count + in_count <= s->in_buffer_index){
489             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
490             copy(&s->in_buffer, &tmp, s->in_buffer_count);
491             s->in_buffer_index=0;
492         }else
493             if((ret=realloc_audio(&s->in_buffer, size)) < 0)
494                 return ret;
495
496         if(in_count){
497             int count= in_count;
498             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
499
500             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
501             copy(&tmp, &in, /*in_*/count);
502             s->in_buffer_count += count;
503             in_count -= count;
504             border += count;
505             buf_set(&in, &in, count);
506             s->resample_in_constraint= 0;
507             if(s->in_buffer_count != count || in_count)
508                 continue;
509         }
510         break;
511     }while(1);
512
513     s->resample_in_constraint= !!out_count;
514
515     return ret_sum;
516 }
517
518 static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
519                                                       AudioData *in , int  in_count){
520     AudioData *postin, *midbuf, *preout;
521     int ret/*, in_max*/;
522     AudioData preout_tmp, midbuf_tmp;
523
524     if(s->full_convert){
525         av_assert0(!s->resample);
526         swri_audio_convert(s->full_convert, out, in, in_count);
527         return out_count;
528     }
529
530 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
531 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
532
533     if((ret=realloc_audio(&s->postin, in_count))<0)
534         return ret;
535     if(s->resample_first){
536         av_assert0(s->midbuf.ch_count == s->used_ch_count);
537         if((ret=realloc_audio(&s->midbuf, out_count))<0)
538             return ret;
539     }else{
540         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
541         if((ret=realloc_audio(&s->midbuf,  in_count))<0)
542             return ret;
543     }
544     if((ret=realloc_audio(&s->preout, out_count))<0)
545         return ret;
546
547     postin= &s->postin;
548
549     midbuf_tmp= s->midbuf;
550     midbuf= &midbuf_tmp;
551     preout_tmp= s->preout;
552     preout= &preout_tmp;
553
554     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
555         postin= in;
556
557     if(s->resample_first ? !s->resample : !s->rematrix)
558         midbuf= postin;
559
560     if(s->resample_first ? !s->rematrix : !s->resample)
561         preout= midbuf;
562
563     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar){
564         if(preout==in){
565             out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
566             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
567             copy(out, in, out_count);
568             return out_count;
569         }
570         else if(preout==postin) preout= midbuf= postin= out;
571         else if(preout==midbuf) preout= midbuf= out;
572         else                    preout= out;
573     }
574
575     if(in != postin){
576         swri_audio_convert(s->in_convert, postin, in, in_count);
577     }
578
579     if(s->resample_first){
580         if(postin != midbuf)
581             out_count= resample(s, midbuf, out_count, postin, in_count);
582         if(midbuf != preout)
583             swri_rematrix(s, preout, midbuf, out_count, preout==out);
584     }else{
585         if(postin != midbuf)
586             swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
587         if(midbuf != preout)
588             out_count= resample(s, preout, out_count, midbuf, in_count);
589     }
590
591     if(preout != out && out_count){
592         if(s->dither_method){
593             int ch;
594             int dither_count= FFMAX(out_count, 1<<16);
595             av_assert0(preout != in);
596
597             if((ret=realloc_audio(&s->dither, dither_count))<0)
598                 return ret;
599             if(ret)
600                 for(ch=0; ch<s->dither.ch_count; ch++)
601                     swri_get_dither(s, s->dither.ch[ch], s->dither.count, 12345678913579<<ch, s->out_sample_fmt, s->int_sample_fmt);
602             av_assert0(s->dither.ch_count == preout->ch_count);
603
604             if(s->dither_pos + out_count > s->dither.count)
605                 s->dither_pos = 0;
606
607             for(ch=0; ch<preout->ch_count; ch++)
608                 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);
609
610             s->dither_pos += out_count;
611         }
612 //FIXME packed doesnt need more than 1 chan here!
613         swri_audio_convert(s->out_convert, out, preout, out_count);
614     }
615     return out_count;
616 }
617
618 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
619                                 const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
620     AudioData * in= &s->in;
621     AudioData *out= &s->out;
622
623     if(s->drop_output > 0){
624         int ret;
625         AudioData tmp = s->out;
626         uint8_t *tmp_arg[SWR_CH_MAX];
627         tmp.count = 0;
628         tmp.data  = NULL;
629         if((ret=realloc_audio(&tmp, s->drop_output))<0)
630             return ret;
631
632         reversefill_audiodata(&tmp, tmp_arg);
633         s->drop_output *= -1; //FIXME find a less hackish solution
634         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
635         s->drop_output *= -1;
636         if(ret>0)
637             s->drop_output -= ret;
638
639         av_freep(&tmp.data);
640         if(s->drop_output || !out_arg)
641             return 0;
642         in_count = 0;
643     }
644
645     if(!in_arg){
646         if(s->in_buffer_count){
647             if (s->resample && !s->flushed) {
648                 AudioData *a= &s->in_buffer;
649                 int i, j, ret;
650                 if((ret=realloc_audio(a, s->in_buffer_index + 2*s->in_buffer_count)) < 0)
651                     return ret;
652                 av_assert0(a->planar);
653                 for(i=0; i<a->ch_count; i++){
654                     for(j=0; j<s->in_buffer_count; j++){
655                         memcpy(a->ch[i] + (s->in_buffer_index+s->in_buffer_count+j  )*a->bps,
656                             a->ch[i] + (s->in_buffer_index+s->in_buffer_count-j-1)*a->bps, a->bps);
657                     }
658                 }
659                 s->in_buffer_count += (s->in_buffer_count+1)/2;
660                 s->resample_in_constraint = 0;
661                 s->flushed = 1;
662             }
663         }else{
664             return 0;
665         }
666     }else
667         fill_audiodata(in ,  (void*)in_arg);
668
669     fill_audiodata(out, out_arg);
670
671     if(s->resample){
672         int ret = swr_convert_internal(s, out, out_count, in, in_count);
673         if(ret>0 && !s->drop_output)
674             s->outpts += ret * (int64_t)s->in_sample_rate;
675         return ret;
676     }else{
677         AudioData tmp= *in;
678         int ret2=0;
679         int ret, size;
680         size = FFMIN(out_count, s->in_buffer_count);
681         if(size){
682             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
683             ret= swr_convert_internal(s, out, size, &tmp, size);
684             if(ret<0)
685                 return ret;
686             ret2= ret;
687             s->in_buffer_count -= ret;
688             s->in_buffer_index += ret;
689             buf_set(out, out, ret);
690             out_count -= ret;
691             if(!s->in_buffer_count)
692                 s->in_buffer_index = 0;
693         }
694
695         if(in_count){
696             size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
697
698             if(in_count > out_count) { //FIXME move after swr_convert_internal
699                 if(   size > s->in_buffer.count
700                 && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
701                     buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
702                     copy(&s->in_buffer, &tmp, s->in_buffer_count);
703                     s->in_buffer_index=0;
704                 }else
705                     if((ret=realloc_audio(&s->in_buffer, size)) < 0)
706                         return ret;
707             }
708
709             if(out_count){
710                 size = FFMIN(in_count, out_count);
711                 ret= swr_convert_internal(s, out, size, in, size);
712                 if(ret<0)
713                     return ret;
714                 buf_set(in, in, ret);
715                 in_count -= ret;
716                 ret2 += ret;
717             }
718             if(in_count){
719                 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
720                 copy(&tmp, in, in_count);
721                 s->in_buffer_count += in_count;
722             }
723         }
724         if(ret2>0 && !s->drop_output)
725             s->outpts += ret2 * (int64_t)s->in_sample_rate;
726         return ret2;
727     }
728 }
729
730 int swr_drop_output(struct SwrContext *s, int count){
731     s->drop_output += count;
732
733     if(s->drop_output <= 0)
734         return 0;
735
736     av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
737     return swr_convert(s, NULL, s->drop_output, NULL, 0);
738 }
739
740 int swr_inject_silence(struct SwrContext *s, int count){
741     int ret, i;
742     AudioData silence = s->in;
743     uint8_t *tmp_arg[SWR_CH_MAX];
744
745     if(count <= 0)
746         return 0;
747
748     silence.count = 0;
749     silence.data  = NULL;
750     if((ret=realloc_audio(&silence, count))<0)
751         return ret;
752
753     if(silence.planar) for(i=0; i<silence.ch_count; i++) {
754         memset(silence.ch[i], silence.bps==1 ? 0x80 : 0, count*silence.bps);
755     } else
756         memset(silence.ch[0], silence.bps==1 ? 0x80 : 0, count*silence.bps*silence.ch_count);
757
758     reversefill_audiodata(&silence, tmp_arg);
759     av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
760     ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
761     av_freep(&silence.data);
762     return ret;
763 }
764
765 int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
766     if(pts == INT64_MIN)
767         return s->outpts;
768     if(s->min_compensation >= FLT_MAX) {
769         return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
770     } else {
771         int64_t delta = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate) - s->outpts;
772         double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
773
774         if(fabs(fdelta) > s->min_compensation) {
775             if(!s->outpts || fabs(fdelta) > s->min_hard_compensation){
776                 int ret;
777                 if(delta > 0) ret = swr_inject_silence(s,  delta / s->out_sample_rate);
778                 else          ret = swr_drop_output   (s, -delta / s-> in_sample_rate);
779                 if(ret<0){
780                     av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
781                 }
782             } else if(s->soft_compensation_duration && s->max_soft_compensation) {
783                 int duration = s->out_sample_rate * s->soft_compensation_duration;
784                 double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
785                 int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
786                 av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
787                 swr_set_compensation(s, comp, duration);
788             }
789         }
790
791         return s->outpts;
792     }
793 }