vorbis.h -> vorbis_data.c
[ffmpeg.git] / libavcodec / vorbis.c
1 /**
2  * @file vorbis.c
3  * Vorbis I decoder
4  * @author Denes Balatoni  ( dbalatoni programozo hu )
5
6  * This library 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 of the License, or (at your option) any later version.
10  *
11  * This library 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 this library; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19  *
20  */
21
22 #undef V_DEBUG
23 //#define V_DEBUG
24 //#define AV_DEBUG(...) av_log(NULL, AV_LOG_INFO, __VA_ARGS__)
25
26 #include <math.h>
27
28 #define ALT_BITSTREAM_READER_LE
29 #include "avcodec.h"
30 #include "bitstream.h"
31 #include "dsputil.h"
32
33 #include "vorbis.h"
34
35 #define V_NB_BITS 8
36 #define V_NB_BITS2 11
37 #define V_MAX_VLCS (1<<16)
38
39 #ifndef V_DEBUG
40 #define AV_DEBUG(...)
41 #endif
42
43 #undef NDEBUG
44 #include <assert.h>
45
46 typedef struct {
47     uint_fast8_t dimensions;
48     uint_fast8_t lookup_type;
49     uint_fast8_t maxdepth;
50     VLC vlc;
51     float *codevectors;
52     unsigned int nb_bits;
53 } vorbis_codebook;
54
55 typedef union vorbis_floor_u vorbis_floor_data;
56 typedef struct vorbis_floor0_s vorbis_floor0;
57 typedef struct vorbis_floor1_s vorbis_floor1;
58 struct vorbis_context_s;
59 typedef
60 uint_fast8_t (* vorbis_floor_decode_func)
61              (struct vorbis_context_s *, vorbis_floor_data *, float *);
62 typedef struct {
63     uint_fast8_t floor_type;
64     vorbis_floor_decode_func decode;
65     union vorbis_floor_u
66     {
67         struct vorbis_floor0_s
68         {
69             uint_fast8_t order;
70             uint_fast16_t rate;
71             uint_fast16_t bark_map_size;
72             int_fast32_t * map[2];
73             uint_fast32_t map_size[2];
74             uint_fast8_t amplitude_bits;
75             uint_fast8_t amplitude_offset;
76             uint_fast8_t num_books;
77             uint_fast8_t * book_list;
78             float * lsp;
79         } t0;
80         struct vorbis_floor1_s
81         {
82             uint_fast8_t partitions;
83             uint_fast8_t maximum_class;
84             uint_fast8_t partition_class[32];
85             uint_fast8_t class_dimensions[16];
86             uint_fast8_t class_subclasses[16];
87             uint_fast8_t class_masterbook[16];
88             int_fast16_t subclass_books[16][8];
89             uint_fast8_t multiplier;
90             uint_fast16_t x_list_dim;
91             uint_fast16_t *x_list;
92             uint_fast16_t *x_list_order;
93             uint_fast16_t *low_neighbour;
94             uint_fast16_t *high_neighbour;
95         } t1;
96     } data;
97 } vorbis_floor;
98
99 typedef struct {
100     uint_fast16_t type;
101     uint_fast32_t begin;
102     uint_fast32_t end;
103     uint_fast32_t partition_size;
104     uint_fast8_t classifications;
105     uint_fast8_t classbook;
106     int_fast16_t books[64][8];
107     uint_fast8_t maxpass;
108 } vorbis_residue;
109
110 typedef struct {
111     uint_fast8_t submaps;
112     uint_fast16_t coupling_steps;
113     uint_fast8_t *magnitude;
114     uint_fast8_t *angle;
115     uint_fast8_t *mux;
116     uint_fast8_t submap_floor[16];
117     uint_fast8_t submap_residue[16];
118 } vorbis_mapping;
119
120 typedef struct {
121     uint_fast8_t blockflag;
122     uint_fast16_t windowtype;
123     uint_fast16_t transformtype;
124     uint_fast8_t mapping;
125 } vorbis_mode;
126
127 typedef struct vorbis_context_s {
128     AVCodecContext *avccontext;
129     GetBitContext gb;
130     DSPContext dsp;
131
132     MDCTContext mdct[2];
133     uint_fast8_t first_frame;
134     uint_fast32_t version;
135     uint_fast8_t audio_channels;
136     uint_fast32_t audio_samplerate;
137     uint_fast32_t bitrate_maximum;
138     uint_fast32_t bitrate_nominal;
139     uint_fast32_t bitrate_minimum;
140     uint_fast32_t blocksize[2];
141     const float * win[2];
142     uint_fast16_t codebook_count;
143     vorbis_codebook *codebooks;
144     uint_fast8_t floor_count;
145     vorbis_floor *floors;
146     uint_fast8_t residue_count;
147     vorbis_residue *residues;
148     uint_fast8_t mapping_count;
149     vorbis_mapping *mappings;
150     uint_fast8_t mode_count;
151     vorbis_mode *modes;
152     uint_fast8_t mode_number; // mode number for the current packet
153     float *channel_residues;
154     float *channel_floors;
155     float *saved;
156     uint_fast16_t saved_start;
157     float *ret;
158     float *buf;
159     float *buf_tmp;
160     uint_fast32_t add_bias; // for float->int conversion
161     uint_fast32_t exp_bias;
162 } vorbis_context;
163
164 /* Helper functions */
165
166 #define ilog(i) av_log2(2*(i))
167
168 #define BARK(x) \
169     (13.1f*atan(0.00074f*(x))+2.24f*atan(1.85e-8f*(x)*(x))+1e-4f*(x))
170
171 static unsigned int nth_root(unsigned int x, unsigned int n) {   // x^(1/n)
172     unsigned int ret=0, i, j;
173
174     do {
175         ++ret;
176         for(i=0,j=ret;i<n-1;i++) j*=ret;
177     } while (j<=x);
178
179     return (ret-1);
180 }
181
182 static float vorbisfloat2float(uint_fast32_t val) {
183     double mant=val&0x1fffff;
184     long exp=(val&0x7fe00000L)>>21;
185     if (val&0x80000000) mant=-mant;
186     return(ldexp(mant, exp-20-768));
187 }
188
189
190 // Generate vlc codes from vorbis huffman code lengths
191
192 static int vorbis_len2vlc(vorbis_context *vc, uint_fast8_t *bits, uint_fast32_t *codes, uint_fast32_t num) {
193     uint_fast32_t exit_at_level[33]={404,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
194         0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0};
195
196     uint_fast8_t i,j;
197     uint_fast32_t code,p;
198
199 #ifdef V_DEBUG
200     GetBitContext gb;
201 #endif
202
203     for(p=0;(bits[p]==0) && (p<num);++p);
204     if (p==num) {
205 //        av_log(vc->avccontext, AV_LOG_INFO, "An empty codebook. Heh?! \n");
206         return 0;
207     }
208
209     codes[p]=0;
210     for(i=0;i<bits[p];++i) {
211         exit_at_level[i+1]=1<<i;
212     }
213
214 #ifdef V_DEBUG
215     av_log(vc->avccontext, AV_LOG_INFO, " %d. of %d code len %d code %d - ", p, num, bits[p], codes[p]);
216     init_get_bits(&gb, (uint_fast8_t *)&codes[p], bits[p]);
217     for(i=0;i<bits[p];++i) {
218         av_log(vc->avccontext, AV_LOG_INFO, "%s", get_bits1(&gb) ? "1" : "0");
219     }
220     av_log(vc->avccontext, AV_LOG_INFO, "\n");
221 #endif
222
223     ++p;
224
225     for(;p<num;++p) {
226         if (bits[p]==0) continue;
227         // find corresponding exit(node which the tree can grow further from)
228         for(i=bits[p];i>0;--i) {
229             if (exit_at_level[i]) break;
230         }
231         if (!i) return 1; // overspecified tree
232         code=exit_at_level[i];
233         exit_at_level[i]=0;
234         // construct code (append 0s to end) and introduce new exits
235         for(j=i+1;j<=bits[p];++j) {
236             exit_at_level[j]=code+(1<<(j-1));
237         }
238         codes[p]=code;
239
240 #ifdef V_DEBUG
241         av_log(vc->avccontext, AV_LOG_INFO, " %d. code len %d code %d - ", p, bits[p], codes[p]);
242         init_get_bits(&gb, (uint_fast8_t *)&codes[p], bits[p]);
243         for(i=0;i<bits[p];++i) {
244             av_log(vc->avccontext, AV_LOG_INFO, "%s", get_bits1(&gb) ? "1" : "0");
245         }
246         av_log(vc->avccontext, AV_LOG_INFO, "\n");
247 #endif
248
249     }
250
251     //FIXME no exits should be left (underspecified tree - ie. unused valid vlcs - not allowed by SPEC)
252
253     return 0;
254 }
255
256 // Free all allocated memory -----------------------------------------
257
258 static void vorbis_free(vorbis_context *vc) {
259     int_fast16_t i;
260
261     av_freep(&vc->channel_residues);
262     av_freep(&vc->channel_floors);
263     av_freep(&vc->saved);
264     av_freep(&vc->ret);
265     av_freep(&vc->buf);
266     av_freep(&vc->buf_tmp);
267
268     av_freep(&vc->residues);
269     av_freep(&vc->modes);
270
271     ff_mdct_end(&vc->mdct[0]);
272     ff_mdct_end(&vc->mdct[1]);
273
274     for(i=0;i<vc->codebook_count;++i) {
275         av_free(vc->codebooks[i].codevectors);
276         free_vlc(&vc->codebooks[i].vlc);
277     }
278     av_freep(&vc->codebooks);
279
280     for(i=0;i<vc->floor_count;++i) {
281         if(vc->floors[i].floor_type==0) {
282             av_free(vc->floors[i].data.t0.map[0]);
283             av_free(vc->floors[i].data.t0.map[1]);
284             av_free(vc->floors[i].data.t0.book_list);
285             av_free(vc->floors[i].data.t0.lsp);
286         }
287         else {
288             av_free(vc->floors[i].data.t1.x_list);
289             av_free(vc->floors[i].data.t1.x_list_order);
290             av_free(vc->floors[i].data.t1.low_neighbour);
291             av_free(vc->floors[i].data.t1.high_neighbour);
292         }
293     }
294     av_freep(&vc->floors);
295
296     for(i=0;i<vc->mapping_count;++i) {
297         av_free(vc->mappings[i].magnitude);
298         av_free(vc->mappings[i].angle);
299         av_free(vc->mappings[i].mux);
300     }
301     av_freep(&vc->mappings);
302
303     if(vc->exp_bias){
304         av_freep(&vc->win[0]);
305         av_freep(&vc->win[1]);
306     }
307 }
308
309 // Parse setup header -------------------------------------------------
310
311 // Process codebooks part
312
313 static int vorbis_parse_setup_hdr_codebooks(vorbis_context *vc) {
314     uint_fast16_t cb;
315     uint_fast8_t *tmp_vlc_bits;
316     uint_fast32_t *tmp_vlc_codes;
317     GetBitContext *gb=&vc->gb;
318
319     vc->codebook_count=get_bits(gb,8)+1;
320
321     AV_DEBUG(" Codebooks: %d \n", vc->codebook_count);
322
323     vc->codebooks=(vorbis_codebook *)av_mallocz(vc->codebook_count * sizeof(vorbis_codebook));
324     tmp_vlc_bits=(uint_fast8_t *)av_mallocz(V_MAX_VLCS * sizeof(uint_fast8_t));
325     tmp_vlc_codes=(uint_fast32_t *)av_mallocz(V_MAX_VLCS * sizeof(uint_fast32_t));
326
327     for(cb=0;cb<vc->codebook_count;++cb) {
328         vorbis_codebook *codebook_setup=&vc->codebooks[cb];
329         uint_fast8_t ordered;
330         uint_fast32_t t, used_entries=0;
331         uint_fast32_t entries;
332
333         AV_DEBUG(" %d. Codebook \n", cb);
334
335         if (get_bits(gb, 24)!=0x564342) {
336             av_log(vc->avccontext, AV_LOG_ERROR, " %"PRIdFAST16". Codebook setup data corrupt. \n", cb);
337             goto error;
338         }
339
340         codebook_setup->dimensions=get_bits(gb, 16);
341         if (codebook_setup->dimensions>16) {
342             av_log(vc->avccontext, AV_LOG_ERROR, " %"PRIdFAST16". Codebook's dimension is too large (%d). \n", cb, codebook_setup->dimensions);
343             goto error;
344         }
345         entries=get_bits(gb, 24);
346         if (entries>V_MAX_VLCS) {
347             av_log(vc->avccontext, AV_LOG_ERROR, " %"PRIdFAST16". Codebook has too many entries (%"PRIdFAST32"). \n", cb, entries);
348             goto error;
349         }
350
351         ordered=get_bits1(gb);
352
353         AV_DEBUG(" codebook_dimensions %d, codebook_entries %d \n", codebook_setup->dimensions, entries);
354
355         if (!ordered) {
356             uint_fast16_t ce;
357             uint_fast8_t flag;
358             uint_fast8_t sparse=get_bits1(gb);
359
360             AV_DEBUG(" not ordered \n");
361
362             if (sparse) {
363                 AV_DEBUG(" sparse \n");
364
365                 used_entries=0;
366                 for(ce=0;ce<entries;++ce) {
367                     flag=get_bits1(gb);
368                     if (flag) {
369                         tmp_vlc_bits[ce]=get_bits(gb, 5)+1;
370                         ++used_entries;
371                     }
372                     else tmp_vlc_bits[ce]=0;
373                 }
374             } else {
375                 AV_DEBUG(" not sparse \n");
376
377                 used_entries=entries;
378                 for(ce=0;ce<entries;++ce) {
379                     tmp_vlc_bits[ce]=get_bits(gb, 5)+1;
380                 }
381             }
382         } else {
383             uint_fast16_t current_entry=0;
384             uint_fast8_t current_length=get_bits(gb, 5)+1;
385
386             AV_DEBUG(" ordered, current length: %d \n", current_length);  //FIXME
387
388             used_entries=entries;
389             for(;current_entry<used_entries;++current_length) {
390                 uint_fast16_t i, number;
391
392                 AV_DEBUG(" number bits: %d ", ilog(entries - current_entry));
393
394                 number=get_bits(gb, ilog(entries - current_entry));
395
396                 AV_DEBUG(" number: %d \n", number);
397
398                 for(i=current_entry;i<number+current_entry;++i) {
399                     if (i<used_entries) tmp_vlc_bits[i]=current_length;
400                 }
401
402                 current_entry+=number;
403             }
404             if (current_entry>used_entries) {
405                 av_log(vc->avccontext, AV_LOG_ERROR, " More codelengths than codes in codebook. \n");
406                 goto error;
407             }
408         }
409
410         codebook_setup->lookup_type=get_bits(gb, 4);
411
412         AV_DEBUG(" lookup type: %d : %s \n", codebook_setup->lookup_type, codebook_setup->lookup_type ? "vq" : "no lookup" );
413
414 // If the codebook is used for (inverse) VQ, calculate codevectors.
415
416         if (codebook_setup->lookup_type==1) {
417             uint_fast16_t i, j, k;
418             uint_fast16_t codebook_lookup_values=nth_root(entries, codebook_setup->dimensions);
419             uint_fast16_t codebook_multiplicands[codebook_lookup_values];
420
421             float codebook_minimum_value=vorbisfloat2float(get_bits_long(gb, 32));
422             float codebook_delta_value=vorbisfloat2float(get_bits_long(gb, 32));
423             uint_fast8_t codebook_value_bits=get_bits(gb, 4)+1;
424             uint_fast8_t codebook_sequence_p=get_bits1(gb);
425
426             AV_DEBUG(" We expect %d numbers for building the codevectors. \n", codebook_lookup_values);
427             AV_DEBUG("  delta %f minmum %f \n", codebook_delta_value, codebook_minimum_value);
428
429             for(i=0;i<codebook_lookup_values;++i) {
430                 codebook_multiplicands[i]=get_bits(gb, codebook_value_bits);
431
432                 AV_DEBUG(" multiplicands*delta+minmum : %e \n", (float)codebook_multiplicands[i]*codebook_delta_value+codebook_minimum_value);
433                 AV_DEBUG(" multiplicand %d \n", codebook_multiplicands[i]);
434             }
435
436 // Weed out unused vlcs and build codevector vector
437             codebook_setup->codevectors=(float *)av_mallocz(used_entries*codebook_setup->dimensions * sizeof(float));
438             for(j=0, i=0;i<entries;++i) {
439                 uint_fast8_t dim=codebook_setup->dimensions;
440
441                 if (tmp_vlc_bits[i]) {
442                     float last=0.0;
443                     uint_fast32_t lookup_offset=i;
444
445 #ifdef V_DEBUG
446                     av_log(vc->avccontext, AV_LOG_INFO, "Lookup offset %d ,", i);
447 #endif
448
449                     for(k=0;k<dim;++k) {
450                         uint_fast32_t multiplicand_offset = lookup_offset % codebook_lookup_values;
451                         codebook_setup->codevectors[j*dim+k]=codebook_multiplicands[multiplicand_offset]*codebook_delta_value+codebook_minimum_value+last;
452                         if (codebook_sequence_p) {
453                             last=codebook_setup->codevectors[j*dim+k];
454                         }
455                         lookup_offset/=codebook_lookup_values;
456                     }
457                     tmp_vlc_bits[j]=tmp_vlc_bits[i];
458
459 #ifdef V_DEBUG
460                     av_log(vc->avccontext, AV_LOG_INFO, "real lookup offset %d, vector: ", j);
461                     for(k=0;k<dim;++k) {
462                         av_log(vc->avccontext, AV_LOG_INFO, " %f ", codebook_setup->codevectors[j*dim+k]);
463                     }
464                     av_log(vc->avccontext, AV_LOG_INFO, "\n");
465 #endif
466
467                     ++j;
468                 }
469             }
470             if (j!=used_entries) {
471                 av_log(vc->avccontext, AV_LOG_ERROR, "Bug in codevector vector building code. \n");
472                 goto error;
473             }
474             entries=used_entries;
475         }
476         else if (codebook_setup->lookup_type>=2) {
477             av_log(vc->avccontext, AV_LOG_ERROR, "Codebook lookup type not supported. \n");
478             goto error;
479         }
480
481 // Initialize VLC table
482         if (vorbis_len2vlc(vc, tmp_vlc_bits, tmp_vlc_codes, entries)) {
483             av_log(vc->avccontext, AV_LOG_ERROR, " Invalid code lengths while generating vlcs. \n");
484             goto error;
485         }
486         codebook_setup->maxdepth=0;
487         for(t=0;t<entries;++t)
488             if (tmp_vlc_bits[t]>=codebook_setup->maxdepth) codebook_setup->maxdepth=tmp_vlc_bits[t];
489
490         if(codebook_setup->maxdepth > 3*V_NB_BITS) codebook_setup->nb_bits=V_NB_BITS2;
491         else                                       codebook_setup->nb_bits=V_NB_BITS;
492
493         codebook_setup->maxdepth=(codebook_setup->maxdepth+codebook_setup->nb_bits-1)/codebook_setup->nb_bits;
494
495         if (init_vlc(&codebook_setup->vlc, codebook_setup->nb_bits, entries, tmp_vlc_bits, sizeof(*tmp_vlc_bits), sizeof(*tmp_vlc_bits), tmp_vlc_codes, sizeof(*tmp_vlc_codes), sizeof(*tmp_vlc_codes), INIT_VLC_LE)) {
496             av_log(vc->avccontext, AV_LOG_ERROR, " Error generating vlc tables. \n");
497             goto error;
498         }
499     }
500
501     av_free(tmp_vlc_bits);
502     av_free(tmp_vlc_codes);
503     return 0;
504
505 // Error:
506 error:
507     av_free(tmp_vlc_bits);
508     av_free(tmp_vlc_codes);
509     return 1;
510 }
511
512 // Process time domain transforms part (unused in Vorbis I)
513
514 static int vorbis_parse_setup_hdr_tdtransforms(vorbis_context *vc) {
515     GetBitContext *gb=&vc->gb;
516     uint_fast8_t i;
517     uint_fast8_t vorbis_time_count=get_bits(gb, 6)+1;
518
519     for(i=0;i<vorbis_time_count;++i) {
520         uint_fast16_t vorbis_tdtransform=get_bits(gb, 16);
521
522         AV_DEBUG(" Vorbis time domain transform %d: %d \n", vorbis_time_count, vorbis_tdtransform);
523
524         if (vorbis_tdtransform) {
525             av_log(vc->avccontext, AV_LOG_ERROR, "Vorbis time domain transform data nonzero. \n");
526             return 1;
527         }
528     }
529     return 0;
530 }
531
532 // Process floors part
533
534 static uint_fast8_t vorbis_floor0_decode(vorbis_context *vc,
535                                          vorbis_floor_data *vfu, float *vec);
536 static void create_map( vorbis_context * vc, uint_fast8_t floor_number );
537 static uint_fast8_t vorbis_floor1_decode(vorbis_context *vc,
538                                          vorbis_floor_data *vfu, float *vec);
539 static int vorbis_parse_setup_hdr_floors(vorbis_context *vc) {
540     GetBitContext *gb=&vc->gb;
541     uint_fast16_t i,j,k;
542
543     vc->floor_count=get_bits(gb, 6)+1;
544
545     vc->floors=(vorbis_floor *)av_mallocz(vc->floor_count * sizeof(vorbis_floor));
546
547     for (i=0;i<vc->floor_count;++i) {
548         vorbis_floor *floor_setup=&vc->floors[i];
549
550         floor_setup->floor_type=get_bits(gb, 16);
551
552         AV_DEBUG(" %d. floor type %d \n", i, floor_setup->floor_type);
553
554         if (floor_setup->floor_type==1) {
555             uint_fast8_t maximum_class=0;
556             uint_fast8_t rangebits;
557             uint_fast16_t floor1_values=2;
558
559             floor_setup->decode=vorbis_floor1_decode;
560
561             floor_setup->data.t1.partitions=get_bits(gb, 5);
562
563             AV_DEBUG(" %d.floor: %d partitions \n", i, floor_setup->data.t1.partitions);
564
565             for(j=0;j<floor_setup->data.t1.partitions;++j) {
566                 floor_setup->data.t1.partition_class[j]=get_bits(gb, 4);
567                 if (floor_setup->data.t1.partition_class[j]>maximum_class) maximum_class=floor_setup->data.t1.partition_class[j];
568
569                 AV_DEBUG(" %d. floor %d partition class %d \n", i, j, floor_setup->data.t1.partition_class[j]);
570
571             }
572
573             AV_DEBUG(" maximum class %d \n", maximum_class);
574
575             floor_setup->data.t1.maximum_class=maximum_class;
576
577             for(j=0;j<=maximum_class;++j) {
578                 floor_setup->data.t1.class_dimensions[j]=get_bits(gb, 3)+1;
579                 floor_setup->data.t1.class_subclasses[j]=get_bits(gb, 2);
580
581                 AV_DEBUG(" %d floor %d class dim: %d subclasses %d \n", i, j, floor_setup->data.t1.class_dimensions[j], floor_setup->data.t1.class_subclasses[j]);
582
583                 if (floor_setup->data.t1.class_subclasses[j]) {
584                     floor_setup->data.t1.class_masterbook[j]=get_bits(gb, 8);
585
586                     AV_DEBUG("   masterbook: %d \n", floor_setup->data.t1.class_masterbook[j]);
587                 }
588
589                 for(k=0;k<(1<<floor_setup->data.t1.class_subclasses[j]);++k) {
590                     floor_setup->data.t1.subclass_books[j][k]=(int16_t)get_bits(gb, 8)-1;
591
592                     AV_DEBUG("    book %d. : %d \n", k, floor_setup->data.t1.subclass_books[j][k]);
593                 }
594             }
595
596             floor_setup->data.t1.multiplier=get_bits(gb, 2)+1;
597             floor_setup->data.t1.x_list_dim=2;
598
599             for(j=0;j<floor_setup->data.t1.partitions;++j) {
600                 floor_setup->data.t1.x_list_dim+=floor_setup->data.t1.class_dimensions[floor_setup->data.t1.partition_class[j]];
601             }
602
603             floor_setup->data.t1.x_list=(uint_fast16_t *)av_mallocz(floor_setup->data.t1.x_list_dim * sizeof(uint_fast16_t));
604             floor_setup->data.t1.x_list_order=(uint_fast16_t *)av_mallocz(floor_setup->data.t1.x_list_dim * sizeof(uint_fast16_t));
605             floor_setup->data.t1.low_neighbour=(uint_fast16_t *)av_mallocz(floor_setup->data.t1.x_list_dim * sizeof(uint_fast16_t));
606             floor_setup->data.t1.high_neighbour=(uint_fast16_t *)av_mallocz(floor_setup->data.t1.x_list_dim * sizeof(uint_fast16_t));
607
608
609             rangebits=get_bits(gb, 4);
610             floor_setup->data.t1.x_list[0] = 0;
611             floor_setup->data.t1.x_list[1] = (1<<rangebits);
612
613             for(j=0;j<floor_setup->data.t1.partitions;++j) {
614                 for(k=0;k<floor_setup->data.t1.class_dimensions[floor_setup->data.t1.partition_class[j]];++k,++floor1_values) {
615                     floor_setup->data.t1.x_list[floor1_values]=get_bits(gb, rangebits);
616
617                     AV_DEBUG(" %d. floor1 Y coord. %d \n", floor1_values, floor_setup->data.t1.x_list[floor1_values]);
618                 }
619             }
620
621 // Precalculate order of x coordinates - needed for decode
622
623             for(k=0;k<floor_setup->data.t1.x_list_dim;++k) {
624                 floor_setup->data.t1.x_list_order[k]=k;
625             }
626
627             for(k=0;k<floor_setup->data.t1.x_list_dim-1;++k) {   // FIXME optimize sorting ?
628                 for(j=k+1;j<floor_setup->data.t1.x_list_dim;++j) {
629                     if(floor_setup->data.t1.x_list[floor_setup->data.t1.x_list_order[k]]>floor_setup->data.t1.x_list[floor_setup->data.t1.x_list_order[j]]) {
630                         uint_fast16_t tmp=floor_setup->data.t1.x_list_order[k];
631                         floor_setup->data.t1.x_list_order[k]=floor_setup->data.t1.x_list_order[j];
632                         floor_setup->data.t1.x_list_order[j]=tmp;
633                     }
634                 }
635             }
636
637 // Precalculate low and high neighbours
638
639             for(k=2;k<floor_setup->data.t1.x_list_dim;++k) {
640                 floor_setup->data.t1.low_neighbour[k]=0;
641                 floor_setup->data.t1.high_neighbour[k]=1;  // correct according to SPEC requirements
642
643                 for (j=0;j<k;++j) {
644                     if ((floor_setup->data.t1.x_list[j]<floor_setup->data.t1.x_list[k]) &&
645                       (floor_setup->data.t1.x_list[j]>floor_setup->data.t1.x_list[floor_setup->data.t1.low_neighbour[k]])) {
646                         floor_setup->data.t1.low_neighbour[k]=j;
647                     }
648                     if ((floor_setup->data.t1.x_list[j]>floor_setup->data.t1.x_list[k]) &&
649                       (floor_setup->data.t1.x_list[j]<floor_setup->data.t1.x_list[floor_setup->data.t1.high_neighbour[k]])) {
650                         floor_setup->data.t1.high_neighbour[k]=j;
651                     }
652                 }
653             }
654         }
655         else if(floor_setup->floor_type==0) {
656             uint_fast8_t max_codebook_dim=0;
657
658             floor_setup->decode=vorbis_floor0_decode;
659
660             floor_setup->data.t0.order=get_bits(gb, 8);
661             floor_setup->data.t0.rate=get_bits(gb, 16);
662             floor_setup->data.t0.bark_map_size=get_bits(gb, 16);
663             floor_setup->data.t0.amplitude_bits=get_bits(gb, 6);
664             /* zero would result in a div by zero later *
665              * 2^0 - 1 == 0                             */
666             if (floor_setup->data.t0.amplitude_bits == 0) {
667               av_log(vc->avccontext, AV_LOG_ERROR,
668                      "Floor 0 amplitude bits is 0.\n");
669               return 1;
670             }
671             floor_setup->data.t0.amplitude_offset=get_bits(gb, 8);
672             floor_setup->data.t0.num_books=get_bits(gb, 4)+1;
673
674             /* allocate mem for booklist */
675             floor_setup->data.t0.book_list=
676                 av_malloc(floor_setup->data.t0.num_books);
677             if(!floor_setup->data.t0.book_list) { return 1; }
678             /* read book indexes */
679             {
680                 int idx;
681                 uint_fast8_t book_idx;
682                 for (idx=0;idx<floor_setup->data.t0.num_books;++idx) {
683                     book_idx=get_bits(gb, 8);
684                     floor_setup->data.t0.book_list[idx]=book_idx;
685                     if (vc->codebooks[book_idx].dimensions > max_codebook_dim)
686                         max_codebook_dim=vc->codebooks[book_idx].dimensions;
687
688                     if (floor_setup->data.t0.book_list[idx]>vc->codebook_count)
689                         return 1;
690                 }
691             }
692
693             create_map( vc, i );
694
695             /* allocate mem for lsp coefficients */
696             {
697                 /* codebook dim is for padding if codebook dim doesn't *
698                  * divide order+1 then we need to read more data       */
699                 floor_setup->data.t0.lsp=
700                     av_malloc((floor_setup->data.t0.order+1 + max_codebook_dim)
701                               * sizeof(float));
702                 if(!floor_setup->data.t0.lsp) { return 1; }
703             }
704
705 #ifdef V_DEBUG /* debug output parsed headers */
706             AV_DEBUG("floor0 order: %u\n", floor_setup->data.t0.order);
707             AV_DEBUG("floor0 rate: %u\n", floor_setup->data.t0.rate);
708             AV_DEBUG("floor0 bark map size: %u\n",
709               floor_setup->data.t0.bark_map_size);
710             AV_DEBUG("floor0 amplitude bits: %u\n",
711               floor_setup->data.t0.amplitude_bits);
712             AV_DEBUG("floor0 amplitude offset: %u\n",
713               floor_setup->data.t0.amplitude_offset);
714             AV_DEBUG("floor0 number of books: %u\n",
715               floor_setup->data.t0.num_books);
716             AV_DEBUG("floor0 book list pointer: %p\n",
717               floor_setup->data.t0.book_list);
718             {
719               int idx;
720               for (idx=0;idx<floor_setup->data.t0.num_books;++idx) {
721                 AV_DEBUG( "  Book %d: %u\n",
722                   idx+1,
723                   floor_setup->data.t0.book_list[idx] );
724               }
725             }
726 #endif
727         }
728         else {
729             av_log(vc->avccontext, AV_LOG_ERROR, "Invalid floor type!\n");
730             return 1;
731         }
732     }
733     return 0;
734 }
735
736 // Process residues part
737
738 static int vorbis_parse_setup_hdr_residues(vorbis_context *vc){
739     GetBitContext *gb=&vc->gb;
740     uint_fast8_t i, j, k;
741
742     vc->residue_count=get_bits(gb, 6)+1;
743     vc->residues=(vorbis_residue *)av_mallocz(vc->residue_count * sizeof(vorbis_residue));
744
745     AV_DEBUG(" There are %d residues. \n", vc->residue_count);
746
747     for(i=0;i<vc->residue_count;++i) {
748         vorbis_residue *res_setup=&vc->residues[i];
749         uint_fast8_t cascade[64];
750         uint_fast8_t high_bits;
751         uint_fast8_t low_bits;
752
753         res_setup->type=get_bits(gb, 16);
754
755         AV_DEBUG(" %d. residue type %d \n", i, res_setup->type);
756
757         res_setup->begin=get_bits(gb, 24);
758         res_setup->end=get_bits(gb, 24);
759         res_setup->partition_size=get_bits(gb, 24)+1;
760         res_setup->classifications=get_bits(gb, 6)+1;
761         res_setup->classbook=get_bits(gb, 8);
762
763         AV_DEBUG("    begin %d end %d part.size %d classif.s %d classbook %d \n", res_setup->begin, res_setup->end, res_setup->partition_size,
764           res_setup->classifications, res_setup->classbook);
765
766         for(j=0;j<res_setup->classifications;++j) {
767             high_bits=0;
768             low_bits=get_bits(gb, 3);
769             if (get_bits1(gb)) {
770                 high_bits=get_bits(gb, 5);
771             }
772             cascade[j]=(high_bits<<3)+low_bits;
773
774             AV_DEBUG("     %d class casscade depth: %d \n", j, ilog(cascade[j]));
775         }
776
777         res_setup->maxpass=0;
778         for(j=0;j<res_setup->classifications;++j) {
779             for(k=0;k<8;++k) {
780                 if (cascade[j]&(1<<k)) {
781                         res_setup->books[j][k]=get_bits(gb, 8);
782
783                     AV_DEBUG("     %d class casscade depth %d book: %d \n", j, k, res_setup->books[j][k]);
784
785                     if (k>res_setup->maxpass) {
786                         res_setup->maxpass=k;
787                     }
788                 } else {
789                     res_setup->books[j][k]=-1;
790                 }
791             }
792         }
793     }
794     return 0;
795 }
796
797 // Process mappings part
798
799 static int vorbis_parse_setup_hdr_mappings(vorbis_context *vc) {
800     GetBitContext *gb=&vc->gb;
801     uint_fast8_t i, j;
802
803     vc->mapping_count=get_bits(gb, 6)+1;
804     vc->mappings=(vorbis_mapping *)av_mallocz(vc->mapping_count * sizeof(vorbis_mapping));
805
806     AV_DEBUG(" There are %d mappings. \n", vc->mapping_count);
807
808     for(i=0;i<vc->mapping_count;++i) {
809         vorbis_mapping *mapping_setup=&vc->mappings[i];
810
811         if (get_bits(gb, 16)) {
812             av_log(vc->avccontext, AV_LOG_ERROR, "Other mappings than type 0 are not compliant with the Vorbis I specification. \n");
813             return 1;
814         }
815         if (get_bits1(gb)) {
816             mapping_setup->submaps=get_bits(gb, 4)+1;
817         } else {
818             mapping_setup->submaps=1;
819         }
820
821         if (get_bits1(gb)) {
822             mapping_setup->coupling_steps=get_bits(gb, 8)+1;
823             mapping_setup->magnitude=(uint_fast8_t *)av_mallocz(mapping_setup->coupling_steps * sizeof(uint_fast8_t));
824             mapping_setup->angle=(uint_fast8_t *)av_mallocz(mapping_setup->coupling_steps * sizeof(uint_fast8_t));
825             for(j=0;j<mapping_setup->coupling_steps;++j) {
826                 mapping_setup->magnitude[j]=get_bits(gb, ilog(vc->audio_channels-1));
827                 mapping_setup->angle[j]=get_bits(gb, ilog(vc->audio_channels-1));
828                 // FIXME: sanity checks
829             }
830         } else {
831             mapping_setup->coupling_steps=0;
832         }
833
834         AV_DEBUG("   %d mapping coupling steps: %d \n", i, mapping_setup->coupling_steps);
835
836         if(get_bits(gb, 2)) {
837             av_log(vc->avccontext, AV_LOG_ERROR, "%d. mapping setup data invalid. \n", i);
838             return 1; // following spec.
839         }
840
841         if (mapping_setup->submaps>1) {
842             mapping_setup->mux=(uint_fast8_t *)av_mallocz(vc->audio_channels * sizeof(uint_fast8_t));
843             for(j=0;j<vc->audio_channels;++j) {
844                 mapping_setup->mux[j]=get_bits(gb, 4);
845             }
846         }
847
848         for(j=0;j<mapping_setup->submaps;++j) {
849             get_bits(gb, 8); // FIXME check?
850             mapping_setup->submap_floor[j]=get_bits(gb, 8);
851             mapping_setup->submap_residue[j]=get_bits(gb, 8);
852
853             AV_DEBUG("   %d mapping %d submap : floor %d, residue %d \n", i, j, mapping_setup->submap_floor[j], mapping_setup->submap_residue[j]);
854         }
855     }
856     return 0;
857 }
858
859 // Process modes part
860
861 static void create_map( vorbis_context * vc, uint_fast8_t floor_number )
862 {
863     vorbis_floor * floors=vc->floors;
864     vorbis_floor0 * vf;
865     int idx;
866     int_fast8_t blockflag;
867     int_fast32_t * map;
868     int_fast32_t n; //TODO: could theoretically be smaller?
869
870     for (blockflag=0;blockflag<2;++blockflag)
871     {
872     n=vc->blocksize[blockflag]/2;
873     floors[floor_number].data.t0.map[blockflag]=
874         av_malloc((n+1) * sizeof(int_fast32_t)); // n+sentinel
875
876     map=floors[floor_number].data.t0.map[blockflag];
877     vf=&floors[floor_number].data.t0;
878
879     for (idx=0; idx<n;++idx) {
880         map[idx]=floor( BARK((vf->rate*idx)/(2.0f*n)) *
881                               ((vf->bark_map_size)/
882                                BARK(vf->rate/2.0f )) );
883         if (vf->bark_map_size-1 < map[idx]) {
884             map[idx]=vf->bark_map_size-1;
885         }
886     }
887     map[n]=-1;
888     vf->map_size[blockflag]=n;
889     }
890
891 #   ifdef V_DEBUG
892     for(idx=0;idx<=n;++idx) {
893         AV_DEBUG("floor0 map: map at pos %d is %d\n",
894                  idx, map[idx]);
895     }
896 #   endif
897 }
898
899 static int vorbis_parse_setup_hdr_modes(vorbis_context *vc) {
900     GetBitContext *gb=&vc->gb;
901     uint_fast8_t i;
902
903     vc->mode_count=get_bits(gb, 6)+1;
904     vc->modes=(vorbis_mode *)av_mallocz(vc->mode_count * sizeof(vorbis_mode));
905
906     AV_DEBUG(" There are %d modes.\n", vc->mode_count);
907
908     for(i=0;i<vc->mode_count;++i) {
909         vorbis_mode *mode_setup=&vc->modes[i];
910
911         mode_setup->blockflag=get_bits(gb, 1);
912         mode_setup->windowtype=get_bits(gb, 16); //FIXME check
913         mode_setup->transformtype=get_bits(gb, 16); //FIXME check
914         mode_setup->mapping=get_bits(gb, 8); //FIXME check
915
916         AV_DEBUG(" %d mode: blockflag %d, windowtype %d, transformtype %d, mapping %d \n", i, mode_setup->blockflag, mode_setup->windowtype, mode_setup->transformtype, mode_setup->mapping);
917     }
918     return 0;
919 }
920
921 // Process the whole setup header using the functions above
922
923 static int vorbis_parse_setup_hdr(vorbis_context *vc) {
924     GetBitContext *gb=&vc->gb;
925
926     if ((get_bits(gb, 8)!='v') || (get_bits(gb, 8)!='o') ||
927     (get_bits(gb, 8)!='r') || (get_bits(gb, 8)!='b') ||
928     (get_bits(gb, 8)!='i') || (get_bits(gb, 8)!='s')) {
929         av_log(vc->avccontext, AV_LOG_ERROR, " Vorbis setup header packet corrupt (no vorbis signature). \n");
930         return 1;
931     }
932
933     if (vorbis_parse_setup_hdr_codebooks(vc)) {
934         av_log(vc->avccontext, AV_LOG_ERROR, " Vorbis setup header packet corrupt (codebooks). \n");
935         return 2;
936     }
937     if (vorbis_parse_setup_hdr_tdtransforms(vc)) {
938         av_log(vc->avccontext, AV_LOG_ERROR, " Vorbis setup header packet corrupt (time domain transforms). \n");
939         return 3;
940     }
941     if (vorbis_parse_setup_hdr_floors(vc)) {
942         av_log(vc->avccontext, AV_LOG_ERROR, " Vorbis setup header packet corrupt (floors). \n");
943         return 4;
944     }
945     if (vorbis_parse_setup_hdr_residues(vc)) {
946         av_log(vc->avccontext, AV_LOG_ERROR, " Vorbis setup header packet corrupt (residues). \n");
947         return 5;
948     }
949     if (vorbis_parse_setup_hdr_mappings(vc)) {
950         av_log(vc->avccontext, AV_LOG_ERROR, " Vorbis setup header packet corrupt (mappings). \n");
951         return 6;
952     }
953     if (vorbis_parse_setup_hdr_modes(vc)) {
954         av_log(vc->avccontext, AV_LOG_ERROR, " Vorbis setup header packet corrupt (modes). \n");
955         return 7;
956     }
957     if (!get_bits1(gb)) {
958         av_log(vc->avccontext, AV_LOG_ERROR, " Vorbis setup header packet corrupt (framing flag). \n");
959         return 8; // framing flag bit unset error
960     }
961
962     return 0;
963 }
964
965 // Process the identification header
966
967 static int vorbis_parse_id_hdr(vorbis_context *vc){
968     GetBitContext *gb=&vc->gb;
969     uint_fast8_t bl0, bl1;
970
971     if ((get_bits(gb, 8)!='v') || (get_bits(gb, 8)!='o') ||
972     (get_bits(gb, 8)!='r') || (get_bits(gb, 8)!='b') ||
973     (get_bits(gb, 8)!='i') || (get_bits(gb, 8)!='s')) {
974         av_log(vc->avccontext, AV_LOG_ERROR, " Vorbis id header packet corrupt (no vorbis signature). \n");
975         return 1;
976     }
977
978     vc->version=get_bits_long(gb, 32);    //FIXME check 0
979     vc->audio_channels=get_bits(gb, 8);   //FIXME check >0
980     vc->audio_samplerate=get_bits_long(gb, 32);   //FIXME check >0
981     vc->bitrate_maximum=get_bits_long(gb, 32);
982     vc->bitrate_nominal=get_bits_long(gb, 32);
983     vc->bitrate_minimum=get_bits_long(gb, 32);
984     bl0=get_bits(gb, 4);
985     bl1=get_bits(gb, 4);
986     vc->blocksize[0]=(1<<bl0);
987     vc->blocksize[1]=(1<<bl1);
988     if (bl0>13 || bl0<6 || bl1>13 || bl1<6 || bl1<bl0) {
989         av_log(vc->avccontext, AV_LOG_ERROR, " Vorbis id header packet corrupt (illegal blocksize). \n");
990         return 3;
991     }
992     // output format int16
993     if (vc->blocksize[1]/2 * vc->audio_channels * 2 >
994                                              AVCODEC_MAX_AUDIO_FRAME_SIZE) {
995         av_log(vc->avccontext, AV_LOG_ERROR, "Vorbis channel count makes "
996                "output packets too large.\n");
997         return 4;
998     }
999     vc->win[0]=ff_vorbis_vwin[bl0-6];
1000     vc->win[1]=ff_vorbis_vwin[bl1-6];
1001
1002     if(vc->exp_bias){
1003         int i, j;
1004         for(j=0; j<2; j++){
1005             float *win = av_malloc(vc->blocksize[j]/2 * sizeof(float));
1006             for(i=0; i<vc->blocksize[j]/2; i++)
1007                 win[i] = vc->win[j][i] * (1<<15);
1008             vc->win[j] = win;
1009         }
1010     }
1011
1012     if ((get_bits1(gb)) == 0) {
1013         av_log(vc->avccontext, AV_LOG_ERROR, " Vorbis id header packet corrupt (framing flag not set). \n");
1014         return 2;
1015     }
1016
1017     vc->channel_residues=(float *)av_malloc((vc->blocksize[1]/2)*vc->audio_channels * sizeof(float));
1018     vc->channel_floors=(float *)av_malloc((vc->blocksize[1]/2)*vc->audio_channels * sizeof(float));
1019     vc->saved=(float *)av_malloc((vc->blocksize[1]/2)*vc->audio_channels * sizeof(float));
1020     vc->ret=(float *)av_malloc((vc->blocksize[1]/2)*vc->audio_channels * sizeof(float));
1021     vc->buf=(float *)av_malloc(vc->blocksize[1] * sizeof(float));
1022     vc->buf_tmp=(float *)av_malloc(vc->blocksize[1] * sizeof(float));
1023     vc->saved_start=0;
1024
1025     ff_mdct_init(&vc->mdct[0], bl0, 1);
1026     ff_mdct_init(&vc->mdct[1], bl1, 1);
1027
1028     AV_DEBUG(" vorbis version %d \n audio_channels %d \n audio_samplerate %d \n bitrate_max %d \n bitrate_nom %d \n bitrate_min %d \n blk_0 %d blk_1 %d \n ",
1029             vc->version, vc->audio_channels, vc->audio_samplerate, vc->bitrate_maximum, vc->bitrate_nominal, vc->bitrate_minimum, vc->blocksize[0], vc->blocksize[1]);
1030
1031 /*
1032     BLK=vc->blocksize[0];
1033     for(i=0;i<BLK/2;++i) {
1034         vc->win[0][i]=sin(0.5*3.14159265358*(sin(((float)i+0.5)/(float)BLK*3.14159265358))*(sin(((float)i+0.5)/(float)BLK*3.14159265358)));
1035     }
1036 */
1037
1038     return 0;
1039 }
1040
1041 // Process the extradata using the functions above (identification header, setup header)
1042
1043 static int vorbis_decode_init(AVCodecContext *avccontext) {
1044     vorbis_context *vc = avccontext->priv_data ;
1045     uint8_t *headers = avccontext->extradata;
1046     int headers_len=avccontext->extradata_size;
1047     uint8_t *header_start[3];
1048     int header_len[3];
1049     GetBitContext *gb = &(vc->gb);
1050     int i, j, hdr_type;
1051
1052     vc->avccontext = avccontext;
1053     dsputil_init(&vc->dsp, avccontext);
1054
1055     if(vc->dsp.float_to_int16 == ff_float_to_int16_c) {
1056         vc->add_bias = 385;
1057         vc->exp_bias = 0;
1058     } else {
1059         vc->add_bias = 0;
1060         vc->exp_bias = 15<<23;
1061     }
1062
1063     if (!headers_len) {
1064         av_log(avccontext, AV_LOG_ERROR, "Extradata corrupt.\n");
1065         return -1;
1066     }
1067
1068     if(headers[0] == 0 && headers[1] == 30) {
1069         for(i = 0; i < 3; i++){
1070             header_len[i] = *headers++ << 8;
1071             header_len[i] += *headers++;
1072             header_start[i] = headers;
1073             headers += header_len[i];
1074         }
1075     } else if(headers[0] == 2) {
1076         for(j=1,i=0;i<2;++i, ++j) {
1077             header_len[i]=0;
1078             while(j<headers_len && headers[j]==0xff) {
1079                 header_len[i]+=0xff;
1080                 ++j;
1081             }
1082             if (j>=headers_len) {
1083                 av_log(avccontext, AV_LOG_ERROR, "Extradata corrupt.\n");
1084                 return -1;
1085             }
1086             header_len[i]+=headers[j];
1087         }
1088         header_len[2]=headers_len-header_len[0]-header_len[1]-j;
1089         headers+=j;
1090         header_start[0] = headers;
1091         header_start[1] = header_start[0] + header_len[0];
1092         header_start[2] = header_start[1] + header_len[1];
1093     } else {
1094         av_log(avccontext, AV_LOG_ERROR, "Extradata corrupt.\n");
1095         return -1;
1096     }
1097
1098     init_get_bits(gb, header_start[0], header_len[0]*8);
1099     hdr_type=get_bits(gb, 8);
1100     if (hdr_type!=1) {
1101         av_log(avccontext, AV_LOG_ERROR, "First header is not the id header.\n");
1102         return -1;
1103     }
1104     if (vorbis_parse_id_hdr(vc)) {
1105         av_log(avccontext, AV_LOG_ERROR, "Id header corrupt.\n");
1106         vorbis_free(vc);
1107         return -1;
1108     }
1109
1110     init_get_bits(gb, header_start[2], header_len[2]*8);
1111     hdr_type=get_bits(gb, 8);
1112     if (hdr_type!=5) {
1113         av_log(avccontext, AV_LOG_ERROR, "Third header is not the setup header.\n");
1114         return -1;
1115     }
1116     if (vorbis_parse_setup_hdr(vc)) {
1117         av_log(avccontext, AV_LOG_ERROR, "Setup header corrupt.\n");
1118         vorbis_free(vc);
1119         return -1;
1120     }
1121
1122     avccontext->channels = vc->audio_channels;
1123     avccontext->sample_rate = vc->audio_samplerate;
1124
1125     return 0 ;
1126 }
1127
1128 // Decode audiopackets -------------------------------------------------
1129
1130 // Read and decode floor
1131
1132 static uint_fast8_t vorbis_floor0_decode(vorbis_context *vc,
1133                                          vorbis_floor_data *vfu, float *vec) {
1134     vorbis_floor0 * vf=&vfu->t0;
1135     float * lsp=vf->lsp;
1136     uint_fast32_t amplitude;
1137     uint_fast32_t book_idx;
1138     uint_fast8_t blockflag=vc->modes[vc->mode_number].blockflag;
1139
1140     amplitude=get_bits(&vc->gb, vf->amplitude_bits);
1141     if (amplitude>0) {
1142         float last = 0;
1143         uint_fast16_t lsp_len = 0;
1144         uint_fast16_t idx;
1145         vorbis_codebook codebook;
1146
1147         book_idx=get_bits(&vc->gb, ilog(vf->num_books));
1148         if ( book_idx >= vf->num_books ) {
1149             av_log( vc->avccontext, AV_LOG_ERROR,
1150                     "floor0 dec: booknumber too high!\n" );
1151             //FIXME: look above
1152         }
1153         AV_DEBUG( "floor0 dec: booknumber: %u\n", book_idx );
1154         codebook=vc->codebooks[vf->book_list[book_idx]];
1155
1156         while (lsp_len<vf->order) {
1157             int vec_off;
1158
1159             AV_DEBUG( "floor0 dec: book dimension: %d\n", codebook.dimensions );
1160             AV_DEBUG( "floor0 dec: maximum depth: %d\n", codebook.maxdepth );
1161             /* read temp vector */
1162             vec_off=get_vlc2(&vc->gb,
1163                              codebook.vlc.table,
1164                              codebook.nb_bits,
1165                              codebook.maxdepth ) *
1166                              codebook.dimensions;
1167             AV_DEBUG( "floor0 dec: vector offset: %d\n", vec_off );
1168             /* copy each vector component and add last to it */
1169             for (idx=0; idx<codebook.dimensions; ++idx) {
1170                 lsp[lsp_len+idx]=codebook.codevectors[vec_off+idx]+last;
1171             }
1172             last=lsp[lsp_len+idx-1]; /* set last to last vector component */
1173
1174             lsp_len += codebook.dimensions;
1175         }
1176 #ifdef V_DEBUG
1177         /* DEBUG: output lsp coeffs */
1178         {
1179             int idx;
1180             for ( idx = 0; idx < lsp_len; ++idx )
1181                 AV_DEBUG("floor0 dec: coeff at %d is %f\n", idx, lsp[idx] );
1182         }
1183 #endif
1184
1185         /* synthesize floor output vector */
1186         {
1187             int i;
1188             int order=vf->order;
1189             float wstep=M_PI/vf->bark_map_size;
1190
1191             for(i=0;i<order;i++) { lsp[i]=2.0f*cos(lsp[i]); }
1192
1193             AV_DEBUG("floor0 synth: map_size=%d; m=%d; wstep=%f\n",
1194                      vf->map_size, order, wstep);
1195
1196             i=0;
1197             while(i<vf->map_size[blockflag]) {
1198                 int j, iter_cond=vf->map[blockflag][i];
1199                 float p=0.5f;
1200                 float q=0.5f;
1201                 float two_cos_w=2.0f*cos(wstep*iter_cond); // needed all times
1202
1203                 /* similar part for the q and p products */
1204                 for(j=0;j<order;j+=2) {
1205                     q *= lsp[j]  -two_cos_w;
1206                     p *= lsp[j+1]-two_cos_w;
1207                 }
1208                 if(j==order) { // even order
1209                     p *= p*(2.0f-two_cos_w);
1210                     q *= q*(2.0f+two_cos_w);
1211                 }
1212                 else { // odd order
1213                     q *= two_cos_w-lsp[j]; // one more time for q
1214
1215                     /* final step and square */
1216                     p *= p*(4.f-two_cos_w*two_cos_w);
1217                     q *= q;
1218                 }
1219
1220                 /* calculate linear floor value */
1221                 {
1222                     q=exp( (
1223                              ( (amplitude*vf->amplitude_offset)/
1224                                (((1<<vf->amplitude_bits)-1) * sqrt(p+q)) )
1225                              - vf->amplitude_offset ) * .11512925f
1226                          );
1227                 }
1228
1229                 /* fill vector */
1230                 do { vec[i]=q; ++i; }while(vf->map[blockflag][i]==iter_cond);
1231             }
1232         }
1233     }
1234     else {
1235         /* this channel is unused */
1236         return 1;
1237     }
1238
1239     AV_DEBUG(" Floor0 decoded\n");
1240
1241     return 0;
1242 }
1243 static uint_fast8_t vorbis_floor1_decode(vorbis_context *vc, vorbis_floor_data *vfu, float *vec) {
1244     vorbis_floor1 * vf=&vfu->t1;
1245     GetBitContext *gb=&vc->gb;
1246     uint_fast16_t range_v[4]={ 256, 128, 86, 64 };
1247     uint_fast16_t range=range_v[vf->multiplier-1];
1248     uint_fast16_t floor1_Y[vf->x_list_dim];
1249     uint_fast16_t floor1_Y_final[vf->x_list_dim];
1250     uint_fast8_t floor1_flag[vf->x_list_dim];
1251     uint_fast8_t class_;
1252     uint_fast8_t cdim;
1253     uint_fast8_t cbits;
1254     uint_fast8_t csub;
1255     uint_fast8_t cval;
1256     int_fast16_t book;
1257     uint_fast16_t offset;
1258     uint_fast16_t i,j;
1259     uint_fast16_t *floor_x_sort=vf->x_list_order;
1260     /*u*/int_fast16_t adx, ady, off, predicted; // WTF ? dy/adx= (unsigned)dy/adx ?
1261     int_fast16_t dy, err;
1262     uint_fast16_t lx,hx, ly, hy=0;
1263
1264
1265     if (!get_bits1(gb)) return 1; // silence
1266
1267 // Read values (or differences) for the floor's points
1268
1269     floor1_Y[0]=get_bits(gb, ilog(range-1));
1270     floor1_Y[1]=get_bits(gb, ilog(range-1));
1271
1272     AV_DEBUG("floor 0 Y %d floor 1 Y %d \n", floor1_Y[0], floor1_Y[1]);
1273
1274     offset=2;
1275     for(i=0;i<vf->partitions;++i) {
1276         class_=vf->partition_class[i];
1277         cdim=vf->class_dimensions[class_];
1278         cbits=vf->class_subclasses[class_];
1279         csub=(1<<cbits)-1;
1280         cval=0;
1281
1282         AV_DEBUG("Cbits %d \n", cbits);
1283
1284         if (cbits) { // this reads all subclasses for this partition's class
1285             cval=get_vlc2(gb, vc->codebooks[vf->class_masterbook[class_]].vlc.table,
1286             vc->codebooks[vf->class_masterbook[class_]].nb_bits, 3);
1287         }
1288
1289         for(j=0;j<cdim;++j) {
1290             book=vf->subclass_books[class_][cval & csub];
1291
1292             AV_DEBUG("book %d Cbits %d cval %d  bits:%d \n", book, cbits, cval, get_bits_count(gb));
1293
1294             cval=cval>>cbits;
1295             if (book>-1) {
1296                 floor1_Y[offset+j]=get_vlc2(gb, vc->codebooks[book].vlc.table,
1297                 vc->codebooks[book].nb_bits, 3);
1298             } else {
1299                 floor1_Y[offset+j]=0;
1300             }
1301
1302             AV_DEBUG(" floor(%d) = %d \n", vf->x_list[offset+j], floor1_Y[offset+j]);
1303         }
1304         offset+=cdim;
1305     }
1306
1307 // Amplitude calculation from the differences
1308
1309     floor1_flag[0]=1;
1310     floor1_flag[1]=1;
1311     floor1_Y_final[0]=floor1_Y[0];
1312     floor1_Y_final[1]=floor1_Y[1];
1313
1314     for(i=2;i<vf->x_list_dim;++i) {
1315         uint_fast16_t val, highroom, lowroom, room;
1316         uint_fast16_t high_neigh_offs;
1317         uint_fast16_t low_neigh_offs;
1318
1319         low_neigh_offs=vf->low_neighbour[i];
1320         high_neigh_offs=vf->high_neighbour[i];
1321         dy=floor1_Y_final[high_neigh_offs]-floor1_Y_final[low_neigh_offs];  // render_point begin
1322         adx=vf->x_list[high_neigh_offs]-vf->x_list[low_neigh_offs];
1323         ady= ABS(dy);
1324         err=ady*(vf->x_list[i]-vf->x_list[low_neigh_offs]);
1325         off=(int16_t)err/(int16_t)adx;
1326         if (dy<0) {
1327             predicted=floor1_Y_final[low_neigh_offs]-off;
1328         } else {
1329             predicted=floor1_Y_final[low_neigh_offs]+off;
1330         } // render_point end
1331
1332         val=floor1_Y[i];
1333         highroom=range-predicted;
1334         lowroom=predicted;
1335         if (highroom < lowroom) {
1336             room=highroom*2;
1337         } else {
1338             room=lowroom*2;   // SPEC mispelling
1339         }
1340         if (val) {
1341             floor1_flag[low_neigh_offs]=1;
1342             floor1_flag[high_neigh_offs]=1;
1343             floor1_flag[i]=1;
1344             if (val>=room) {
1345                 if (highroom > lowroom) {
1346                     floor1_Y_final[i]=val-lowroom+predicted;
1347                 } else {
1348                     floor1_Y_final[i]=predicted-val+highroom-1;
1349                 }
1350             } else {
1351                 if (val & 1) {
1352                     floor1_Y_final[i]=predicted-(val+1)/2;
1353                 } else {
1354                     floor1_Y_final[i]=predicted+val/2;
1355                 }
1356             }
1357         } else {
1358             floor1_flag[i]=0;
1359             floor1_Y_final[i]=predicted;
1360         }
1361
1362         AV_DEBUG(" Decoded floor(%d) = %d / val %d \n", vf->x_list[i], floor1_Y_final[i], val);
1363     }
1364
1365 // Curve synth - connect the calculated dots and convert from dB scale FIXME optimize ?
1366
1367     hx=0;
1368     lx=0;
1369     ly=floor1_Y_final[0]*vf->multiplier;  // conforms to SPEC
1370
1371     vec[0]=ff_vorbis_floor1_inverse_db_table[ly];
1372
1373     for(i=1;i<vf->x_list_dim;++i) {
1374         AV_DEBUG(" Looking at post %d \n", i);
1375
1376         if (floor1_flag[floor_x_sort[i]]) {   // SPEC mispelled
1377             int_fast16_t x, y, dy, base, sy; // if uncommented: dy = -32 adx = 2  base = 2blablabla ?????
1378
1379             hy=floor1_Y_final[floor_x_sort[i]]*vf->multiplier;
1380             hx=vf->x_list[floor_x_sort[i]];
1381
1382             dy=hy-ly;
1383             adx=hx-lx;
1384             ady= (dy<0) ? -dy:dy;//ABS(dy);
1385             base=(int16_t)dy/(int16_t)adx;
1386
1387             AV_DEBUG(" dy %d  adx %d base %d = %d \n", dy, adx, base, dy/adx);
1388
1389             x=lx;
1390             y=ly;
1391             err=0;
1392             if (dy<0) {
1393                 sy=base-1;
1394             } else {
1395                 sy=base+1;
1396             }
1397             ady=ady-(base<0 ? -base : base)*adx;
1398             vec[x]=ff_vorbis_floor1_inverse_db_table[y];
1399
1400             AV_DEBUG(" vec[ %d ] = %d \n", x, y);
1401
1402             for(x=lx+1;(x<hx) && (x<vf->x_list[1]);++x) {
1403                 err+=ady;
1404                 if (err>=adx) {
1405                     err-=adx;
1406                     y+=sy;
1407                 } else {
1408                     y+=base;
1409                 }
1410                 vec[x]=ff_vorbis_floor1_inverse_db_table[y];
1411
1412                 AV_DEBUG(" vec[ %d ] = %d \n", x, y);
1413             }
1414
1415 /*            for(j=1;j<hx-lx+1;++j) {  // iterating render_point
1416                 dy=hy-ly;
1417                 adx=hx-lx;
1418                 ady= dy<0 ? -dy : dy;
1419                 err=ady*j;
1420                 off=err/adx;
1421                 if (dy<0) {
1422                     predicted=ly-off;
1423                 } else {
1424                     predicted=ly+off;
1425                 }
1426                 if (lx+j < vf->x_list[1]) {
1427                     vec[lx+j]=ff_vorbis_floor1_inverse_db_table[predicted];
1428                 }
1429             }*/
1430
1431             lx=hx;
1432             ly=hy;
1433         }
1434     }
1435
1436     if (hx<vf->x_list[1]) {
1437         for(i=hx;i<vf->x_list[1];++i) {
1438             vec[i]=ff_vorbis_floor1_inverse_db_table[hy];
1439         }
1440     }
1441
1442     AV_DEBUG(" Floor decoded\n");
1443
1444     return 0;
1445 }
1446
1447 // Read and decode residue
1448
1449 static int vorbis_residue_decode(vorbis_context *vc, vorbis_residue *vr, uint_fast8_t ch, uint_fast8_t *do_not_decode, float *vec, uint_fast16_t vlen) {
1450     GetBitContext *gb=&vc->gb;
1451     uint_fast8_t c_p_c=vc->codebooks[vr->classbook].dimensions;
1452     uint_fast16_t n_to_read=vr->end-vr->begin;
1453     uint_fast16_t ptns_to_read=n_to_read/vr->partition_size;
1454     uint_fast8_t classifs[ptns_to_read*vc->audio_channels];
1455     uint_fast8_t pass;
1456     uint_fast8_t ch_used;
1457     uint_fast8_t i,j,l;
1458     uint_fast16_t k;
1459
1460     if (vr->type==2) {
1461         for(j=1;j<ch;++j) {
1462                 do_not_decode[0]&=do_not_decode[j];  // FIXME - clobbering input
1463         }
1464         if (do_not_decode[0]) return 0;
1465         ch_used=1;
1466     } else {
1467         ch_used=ch;
1468     }
1469
1470     AV_DEBUG(" residue type 0/1/2 decode begin, ch: %d  cpc %d  \n", ch, c_p_c);
1471
1472     for(pass=0;pass<=vr->maxpass;++pass) { // FIXME OPTIMIZE?
1473         uint_fast16_t voffset;
1474         uint_fast16_t partition_count;
1475         uint_fast16_t j_times_ptns_to_read;
1476
1477         voffset=vr->begin;
1478         for(partition_count=0;partition_count<ptns_to_read;) {  // SPEC        error
1479             if (!pass) {
1480                 uint_fast32_t inverse_class = inverse[vr->classifications];
1481                 for(j_times_ptns_to_read=0, j=0;j<ch_used;++j) {
1482                     if (!do_not_decode[j]) {
1483                         uint_fast32_t temp=get_vlc2(gb, vc->codebooks[vr->classbook].vlc.table,
1484                         vc->codebooks[vr->classbook].nb_bits, 3);
1485
1486                         AV_DEBUG("Classword: %d \n", temp);
1487
1488                         assert(vr->classifications > 1 && temp<=65536); //needed for inverse[]
1489                         for(i=0;i<c_p_c;++i) {
1490                             uint_fast32_t temp2;
1491
1492                             temp2=(((uint_fast64_t)temp) * inverse_class)>>32;
1493                             if (partition_count+c_p_c-1-i < ptns_to_read) {
1494                                 classifs[j_times_ptns_to_read+partition_count+c_p_c-1-i]=temp-temp2*vr->classifications;
1495                             }
1496                             temp=temp2;
1497                         }
1498                     }
1499                     j_times_ptns_to_read+=ptns_to_read;
1500                 }
1501             }
1502             for(i=0;(i<c_p_c) && (partition_count<ptns_to_read);++i) {
1503                 for(j_times_ptns_to_read=0, j=0;j<ch_used;++j) {
1504                     uint_fast16_t voffs;
1505
1506                     if (!do_not_decode[j]) {
1507                         uint_fast8_t vqclass=classifs[j_times_ptns_to_read+partition_count];
1508                         int_fast16_t vqbook=vr->books[vqclass][pass];
1509
1510                         if (vqbook>=0) {
1511                             uint_fast16_t coffs;
1512                             unsigned dim= vc->codebooks[vqbook].dimensions; // not uint_fast8_t: 64bit is slower here on amd64
1513                             uint_fast16_t step= dim==1 ? vr->partition_size
1514                                               : FASTDIV(vr->partition_size, dim);
1515                             vorbis_codebook codebook= vc->codebooks[vqbook];
1516
1517                             if (vr->type==0) {
1518
1519                                 voffs=voffset+j*vlen;
1520                                 for(k=0;k<step;++k) {
1521                                     coffs=get_vlc2(gb, codebook.vlc.table, codebook.nb_bits, 3) * dim;
1522                                     for(l=0;l<dim;++l) {
1523                                         vec[voffs+k+l*step]+=codebook.codevectors[coffs+l];  // FPMATH
1524                                     }
1525                                 }
1526                             }
1527                             else if (vr->type==1) {
1528                                 voffs=voffset+j*vlen;
1529                                 for(k=0;k<step;++k) {
1530                                     coffs=get_vlc2(gb, codebook.vlc.table, codebook.nb_bits, 3) * dim;
1531                                     for(l=0;l<dim;++l, ++voffs) {
1532                                         vec[voffs]+=codebook.codevectors[coffs+l];  // FPMATH
1533
1534                                         AV_DEBUG(" pass %d offs: %d curr: %f change: %f cv offs.: %d  \n", pass, voffs, vec[voffs], codebook.codevectors[coffs+l], coffs);
1535                                     }
1536                                 }
1537                             }
1538                             else if (vr->type==2 && ch==2 && (voffset&1)==0 && (dim&1)==0) { // most frequent case optimized
1539                                 voffs=voffset>>1;
1540
1541                                 if(dim==2) {
1542                                     for(k=0;k<step;++k) {
1543                                         coffs=get_vlc2(gb, codebook.vlc.table, codebook.nb_bits, 3) * 2;
1544                                         vec[voffs+k     ]+=codebook.codevectors[coffs  ];  // FPMATH
1545                                         vec[voffs+k+vlen]+=codebook.codevectors[coffs+1];  // FPMATH
1546                                     }
1547                                 } else
1548                                 for(k=0;k<step;++k) {
1549                                     coffs=get_vlc2(gb, codebook.vlc.table, codebook.nb_bits, 3) * dim;
1550                                     for(l=0;l<dim;l+=2, voffs++) {
1551                                         vec[voffs     ]+=codebook.codevectors[coffs+l  ];  // FPMATH
1552                                         vec[voffs+vlen]+=codebook.codevectors[coffs+l+1];  // FPMATH
1553
1554                                         AV_DEBUG(" pass %d offs: %d curr: %f change: %f cv offs.: %d+%d  \n", pass, voffset/ch+(voffs%ch)*vlen, vec[voffset/ch+(voffs%ch)*vlen], codebook.codevectors[coffs+l], coffs, l);
1555                                     }
1556                                 }
1557
1558                             }
1559                             else if (vr->type==2) {
1560                                 voffs=voffset;
1561
1562                                 for(k=0;k<step;++k) {
1563                                     coffs=get_vlc2(gb, codebook.vlc.table, codebook.nb_bits, 3) * dim;
1564                                     for(l=0;l<dim;++l, ++voffs) {
1565                                         vec[voffs/ch+(voffs%ch)*vlen]+=codebook.codevectors[coffs+l];  // FPMATH FIXME use if and counter instead of / and %
1566
1567                                         AV_DEBUG(" pass %d offs: %d curr: %f change: %f cv offs.: %d+%d  \n", pass, voffset/ch+(voffs%ch)*vlen, vec[voffset/ch+(voffs%ch)*vlen], codebook.codevectors[coffs+l], coffs, l);
1568                                     }
1569                                 }
1570                             } else {
1571                                 av_log(vc->avccontext, AV_LOG_ERROR, " Invalid residue type while residue decode?! \n");
1572                                 return 1;
1573                             }
1574                         }
1575                     }
1576                     j_times_ptns_to_read+=ptns_to_read;
1577                 }
1578                 ++partition_count;
1579                 voffset+=vr->partition_size;
1580             }
1581         }
1582     }
1583     return 0;
1584 }
1585
1586 void vorbis_inverse_coupling(float *mag, float *ang, int blocksize)
1587 {
1588     int i;
1589     for(i=0; i<blocksize; i++)
1590     {
1591         if (mag[i]>0.0) {
1592             if (ang[i]>0.0) {
1593                 ang[i]=mag[i]-ang[i];
1594             } else {
1595                 float temp=ang[i];
1596                 ang[i]=mag[i];
1597                 mag[i]+=temp;
1598             }
1599         } else {
1600             if (ang[i]>0.0) {
1601                 ang[i]+=mag[i];
1602             } else {
1603                 float temp=ang[i];
1604                 ang[i]=mag[i];
1605                 mag[i]-=temp;
1606             }
1607         }
1608     }
1609 }
1610
1611 // Decode the audio packet using the functions above
1612
1613 static int vorbis_parse_audio_packet(vorbis_context *vc) {
1614     GetBitContext *gb=&vc->gb;
1615
1616     uint_fast8_t previous_window=0,next_window=0;
1617     uint_fast8_t mode_number;
1618     uint_fast16_t blocksize;
1619     int_fast32_t i,j;
1620     uint_fast8_t no_residue[vc->audio_channels];
1621     uint_fast8_t do_not_decode[vc->audio_channels];
1622     vorbis_mapping *mapping;
1623     float *ch_res_ptr=vc->channel_residues;
1624     float *ch_floor_ptr=vc->channel_floors;
1625     uint_fast8_t res_chan[vc->audio_channels];
1626     uint_fast8_t res_num=0;
1627     int_fast16_t retlen=0;
1628     uint_fast16_t saved_start=0;
1629     float fadd_bias = vc->add_bias;
1630
1631     if (get_bits1(gb)) {
1632         av_log(vc->avccontext, AV_LOG_ERROR, "Not a Vorbis I audio packet.\n");
1633         return -1; // packet type not audio
1634     }
1635
1636     if (vc->mode_count==1) {
1637         mode_number=0;
1638     } else {
1639         mode_number=get_bits(gb, ilog(vc->mode_count-1));
1640     }
1641     vc->mode_number=mode_number;
1642     mapping=&vc->mappings[vc->modes[mode_number].mapping];
1643
1644     AV_DEBUG(" Mode number: %d , mapping: %d , blocktype %d \n", mode_number, vc->modes[mode_number].mapping, vc->modes[mode_number].blockflag);
1645
1646     if (vc->modes[mode_number].blockflag) {
1647         previous_window=get_bits1(gb);
1648         next_window=get_bits1(gb);
1649     }
1650
1651     blocksize=vc->blocksize[vc->modes[mode_number].blockflag];
1652     memset(ch_res_ptr, 0, sizeof(float)*vc->audio_channels*blocksize/2); //FIXME can this be removed ?
1653     memset(ch_floor_ptr, 0, sizeof(float)*vc->audio_channels*blocksize/2); //FIXME can this be removed ?
1654
1655 // Decode floor
1656
1657     for(i=0;i<vc->audio_channels;++i) {
1658         vorbis_floor *floor;
1659         if (mapping->submaps>1) {
1660             floor=&vc->floors[mapping->submap_floor[mapping->mux[i]]];
1661         } else {
1662             floor=&vc->floors[mapping->submap_floor[0]];
1663         }
1664
1665         no_residue[i]=floor->decode(vc, &floor->data, ch_floor_ptr);
1666         ch_floor_ptr+=blocksize/2;
1667     }
1668
1669 // Nonzero vector propagate
1670
1671     for(i=mapping->coupling_steps-1;i>=0;--i) {
1672         if (!(no_residue[mapping->magnitude[i]] & no_residue[mapping->angle[i]])) {
1673             no_residue[mapping->magnitude[i]]=0;
1674             no_residue[mapping->angle[i]]=0;
1675         }
1676     }
1677
1678 // Decode residue
1679
1680     for(i=0;i<mapping->submaps;++i) {
1681         vorbis_residue *residue;
1682         uint_fast8_t ch=0;
1683
1684         for(j=0;j<vc->audio_channels;++j) {
1685             if ((mapping->submaps==1) || (i=mapping->mux[j])) {
1686                 res_chan[j]=res_num;
1687                 if (no_residue[j]) {
1688                     do_not_decode[ch]=1;
1689                 } else {
1690                     do_not_decode[ch]=0;
1691                 }
1692                 ++ch;
1693                 ++res_num;
1694             }
1695         }
1696         residue=&vc->residues[mapping->submap_residue[i]];
1697         vorbis_residue_decode(vc, residue, ch, do_not_decode, ch_res_ptr, blocksize/2);
1698
1699         ch_res_ptr+=ch*blocksize/2;
1700     }
1701
1702 // Inverse coupling
1703
1704     for(i=mapping->coupling_steps-1;i>=0;--i) { //warning: i has to be signed
1705         float *mag, *ang;
1706
1707         mag=vc->channel_residues+res_chan[mapping->magnitude[i]]*blocksize/2;
1708         ang=vc->channel_residues+res_chan[mapping->angle[i]]*blocksize/2;
1709         vc->dsp.vorbis_inverse_coupling(mag, ang, blocksize/2);
1710     }
1711
1712 // Dotproduct
1713
1714     for(j=0, ch_floor_ptr=vc->channel_floors;j<vc->audio_channels;++j,ch_floor_ptr+=blocksize/2) {
1715         ch_res_ptr=vc->channel_residues+res_chan[j]*blocksize/2;
1716         vc->dsp.vector_fmul(ch_floor_ptr, ch_res_ptr, blocksize/2);
1717     }
1718
1719 // MDCT, overlap/add, save data for next overlapping  FPMATH
1720
1721     for(j=0;j<vc->audio_channels;++j) {
1722         uint_fast8_t step=vc->audio_channels;
1723         uint_fast16_t k;
1724         float *saved=vc->saved+j*vc->blocksize[1]/2;
1725         float *ret=vc->ret;
1726         const float *lwin=vc->win[1];
1727         const float *swin=vc->win[0];
1728         float *buf=vc->buf;
1729         float *buf_tmp=vc->buf_tmp;
1730
1731         ch_floor_ptr=vc->channel_floors+j*blocksize/2;
1732
1733         saved_start=vc->saved_start;
1734
1735         vc->mdct[0].fft.imdct_calc(&vc->mdct[vc->modes[mode_number].blockflag], buf, ch_floor_ptr, buf_tmp);
1736
1737         //FIXME process channels together, to allow faster simd vector_fmul_add_add?
1738         if (vc->modes[mode_number].blockflag) {
1739             // -- overlap/add
1740             if (previous_window) {
1741                 vc->dsp.vector_fmul_add_add(ret+j, buf, lwin, saved, vc->add_bias, vc->blocksize[1]/2, step);
1742                 retlen=vc->blocksize[1]/2;
1743             } else {
1744                 int len = (vc->blocksize[1]-vc->blocksize[0])/4;
1745                 buf += len;
1746                 vc->dsp.vector_fmul_add_add(ret+j, buf, swin, saved, vc->add_bias, vc->blocksize[0]/2, step);
1747                 k = vc->blocksize[0]/2*step + j;
1748                 buf += vc->blocksize[0]/2;
1749                 if(vc->exp_bias){
1750                     for(i=0; i<len; i++, k+=step)
1751                         ((uint32_t*)ret)[k] = ((uint32_t*)buf)[i] + vc->exp_bias; // ret[k]=buf[i]*(1<<bias)
1752                 } else {
1753                     for(i=0; i<len; i++, k+=step)
1754                         ret[k] = buf[i] + fadd_bias;
1755                 }
1756                 buf=vc->buf;
1757                 retlen=vc->blocksize[0]/2+len;
1758             }
1759             // -- save
1760             if (next_window) {
1761                 buf += vc->blocksize[1]/2;
1762                 vc->dsp.vector_fmul_reverse(saved, buf, lwin, vc->blocksize[1]/2);
1763                 saved_start=0;
1764             } else {
1765                 saved_start=(vc->blocksize[1]-vc->blocksize[0])/4;
1766                 buf += vc->blocksize[1]/2;
1767                 for(i=0; i<saved_start; i++)
1768                     ((uint32_t*)saved)[i] = ((uint32_t*)buf)[i] + vc->exp_bias;
1769                 vc->dsp.vector_fmul_reverse(saved+saved_start, buf+saved_start, swin, vc->blocksize[0]/2);
1770             }
1771         } else {
1772             // --overlap/add
1773             if(vc->add_bias) {
1774                 for(k=j, i=0;i<saved_start;++i, k+=step)
1775                     ret[k] = saved[i] + fadd_bias;
1776             } else {
1777                 for(k=j, i=0;i<saved_start;++i, k+=step)
1778                     ret[k] = saved[i];
1779             }
1780             vc->dsp.vector_fmul_add_add(ret+k, buf, swin, saved+saved_start, vc->add_bias, vc->blocksize[0]/2, step);
1781             retlen=saved_start+vc->blocksize[0]/2;
1782             // -- save
1783             buf += vc->blocksize[0]/2;
1784             vc->dsp.vector_fmul_reverse(saved, buf, swin, vc->blocksize[0]/2);
1785             saved_start=0;
1786         }
1787     }
1788     vc->saved_start=saved_start;
1789
1790     return retlen*vc->audio_channels;
1791 }
1792
1793 // Return the decoded audio packet through the standard api
1794
1795 static int vorbis_decode_frame(AVCodecContext *avccontext,
1796                         void *data, int *data_size,
1797                         uint8_t *buf, int buf_size)
1798 {
1799     vorbis_context *vc = avccontext->priv_data ;
1800     GetBitContext *gb = &(vc->gb);
1801
1802     int_fast16_t len;
1803
1804     if(!buf_size){
1805         return 0;
1806     }
1807
1808     AV_DEBUG("packet length %d \n", buf_size);
1809
1810     init_get_bits(gb, buf, buf_size*8);
1811
1812     len=vorbis_parse_audio_packet(vc);
1813
1814     if (len<=0) {
1815         *data_size=0;
1816         return buf_size;
1817     }
1818
1819     if (!vc->first_frame) {
1820         vc->first_frame=1;
1821         *data_size=0;
1822         return buf_size ;
1823     }
1824
1825     AV_DEBUG("parsed %d bytes %d bits, returned %d samples (*ch*bits) \n", get_bits_count(gb)/8, get_bits_count(gb)%8, len);
1826
1827     vc->dsp.float_to_int16(data, vc->ret, len);
1828     *data_size=len*2;
1829
1830     return buf_size ;
1831 }
1832
1833 // Close decoder
1834
1835 static int vorbis_decode_close(AVCodecContext *avccontext) {
1836     vorbis_context *vc = avccontext->priv_data;
1837
1838     vorbis_free(vc);
1839
1840     return 0 ;
1841 }
1842
1843 AVCodec vorbis_decoder = {
1844     "vorbis",
1845     CODEC_TYPE_AUDIO,
1846     CODEC_ID_VORBIS,
1847     sizeof(vorbis_context),
1848     vorbis_decode_init,
1849     NULL,
1850     vorbis_decode_close,
1851     vorbis_decode_frame,
1852 };
1853