Reorder nut specific codec tags and add a comment for marking them as
[ffmpeg.git] / libavcodec / shorten.c
1 /*
2  * Shorten decoder
3  * Copyright (c) 2005 Jeff Muizelaar
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21
22 /**
23  * @file
24  * Shorten decoder
25  * @author Jeff Muizelaar
26  *
27  */
28
29 #define DEBUG
30 #include <limits.h>
31 #include "avcodec.h"
32 #include "get_bits.h"
33 #include "golomb.h"
34
35 #define MAX_CHANNELS 8
36 #define MAX_BLOCKSIZE 65535
37
38 #define OUT_BUFFER_SIZE 16384
39
40 #define ULONGSIZE 2
41
42 #define WAVE_FORMAT_PCM 0x0001
43
44 #define DEFAULT_BLOCK_SIZE 256
45
46 #define TYPESIZE 4
47 #define CHANSIZE 0
48 #define LPCQSIZE 2
49 #define ENERGYSIZE 3
50 #define BITSHIFTSIZE 2
51
52 #define TYPE_S16HL 3
53 #define TYPE_S16LH 5
54
55 #define NWRAP 3
56 #define NSKIPSIZE 1
57
58 #define LPCQUANT 5
59 #define V2LPCQOFFSET (1 << LPCQUANT)
60
61 #define FNSIZE 2
62 #define FN_DIFF0        0
63 #define FN_DIFF1        1
64 #define FN_DIFF2        2
65 #define FN_DIFF3        3
66 #define FN_QUIT         4
67 #define FN_BLOCKSIZE    5
68 #define FN_BITSHIFT     6
69 #define FN_QLPC         7
70 #define FN_ZERO         8
71 #define FN_VERBATIM     9
72
73 #define VERBATIM_CKSIZE_SIZE 5
74 #define VERBATIM_BYTE_SIZE 8
75 #define CANONICAL_HEADER_SIZE 44
76
77 typedef struct ShortenContext {
78     AVCodecContext *avctx;
79     GetBitContext gb;
80
81     int min_framesize, max_framesize;
82     int channels;
83
84     int32_t *decoded[MAX_CHANNELS];
85     int32_t *offset[MAX_CHANNELS];
86     uint8_t *bitstream;
87     int bitstream_size;
88     int bitstream_index;
89     unsigned int allocated_bitstream_size;
90     int header_size;
91     uint8_t header[OUT_BUFFER_SIZE];
92     int version;
93     int cur_chan;
94     int bitshift;
95     int nmean;
96     int internal_ftype;
97     int nwrap;
98     int blocksize;
99     int bitindex;
100     int32_t lpcqoffset;
101 } ShortenContext;
102
103 static av_cold int shorten_decode_init(AVCodecContext * avctx)
104 {
105     ShortenContext *s = avctx->priv_data;
106     s->avctx = avctx;
107     avctx->sample_fmt = SAMPLE_FMT_S16;
108
109     return 0;
110 }
111
112 static int allocate_buffers(ShortenContext *s)
113 {
114     int i, chan;
115     for (chan=0; chan<s->channels; chan++) {
116         if(FFMAX(1, s->nmean) >= UINT_MAX/sizeof(int32_t)){
117             av_log(s->avctx, AV_LOG_ERROR, "nmean too large\n");
118             return -1;
119         }
120         if(s->blocksize + s->nwrap >= UINT_MAX/sizeof(int32_t) || s->blocksize + s->nwrap <= (unsigned)s->nwrap){
121             av_log(s->avctx, AV_LOG_ERROR, "s->blocksize + s->nwrap too large\n");
122             return -1;
123         }
124
125         s->offset[chan] = av_realloc(s->offset[chan], sizeof(int32_t)*FFMAX(1, s->nmean));
126
127         s->decoded[chan] = av_realloc(s->decoded[chan], sizeof(int32_t)*(s->blocksize + s->nwrap));
128         for (i=0; i<s->nwrap; i++)
129             s->decoded[chan][i] = 0;
130         s->decoded[chan] += s->nwrap;
131     }
132     return 0;
133 }
134
135
136 static inline unsigned int get_uint(ShortenContext *s, int k)
137 {
138     if (s->version != 0)
139         k = get_ur_golomb_shorten(&s->gb, ULONGSIZE);
140     return get_ur_golomb_shorten(&s->gb, k);
141 }
142
143
144 static void fix_bitshift(ShortenContext *s, int32_t *buffer)
145 {
146     int i;
147
148     if (s->bitshift != 0)
149         for (i = 0; i < s->blocksize; i++)
150             buffer[s->nwrap + i] <<= s->bitshift;
151 }
152
153
154 static void init_offset(ShortenContext *s)
155 {
156     int32_t mean = 0;
157     int  chan, i;
158     int nblock = FFMAX(1, s->nmean);
159     /* initialise offset */
160     switch (s->internal_ftype)
161     {
162         case TYPE_S16HL:
163         case TYPE_S16LH:
164             mean = 0;
165             break;
166         default:
167             av_log(s->avctx, AV_LOG_ERROR, "unknown audio type");
168             abort();
169     }
170
171     for (chan = 0; chan < s->channels; chan++)
172         for (i = 0; i < nblock; i++)
173             s->offset[chan][i] = mean;
174 }
175
176 static inline int get_le32(GetBitContext *gb)
177 {
178     return bswap_32(get_bits_long(gb, 32));
179 }
180
181 static inline short get_le16(GetBitContext *gb)
182 {
183     return bswap_16(get_bits_long(gb, 16));
184 }
185
186 static int decode_wave_header(AVCodecContext *avctx, uint8_t *header, int header_size)
187 {
188     GetBitContext hb;
189     int len;
190     int chunk_size;
191     short wave_format;
192
193     init_get_bits(&hb, header, header_size*8);
194     if (get_le32(&hb) != MKTAG('R','I','F','F')) {
195         av_log(avctx, AV_LOG_ERROR, "missing RIFF tag\n");
196         return -1;
197     }
198
199     chunk_size = get_le32(&hb);
200
201     if (get_le32(&hb) != MKTAG('W','A','V','E')) {
202         av_log(avctx, AV_LOG_ERROR, "missing WAVE tag\n");
203         return -1;
204     }
205
206     while (get_le32(&hb) != MKTAG('f','m','t',' ')) {
207         len = get_le32(&hb);
208         skip_bits(&hb, 8*len);
209     }
210     len = get_le32(&hb);
211
212     if (len < 16) {
213         av_log(avctx, AV_LOG_ERROR, "fmt chunk was too short\n");
214         return -1;
215     }
216
217     wave_format = get_le16(&hb);
218
219     switch (wave_format) {
220         case WAVE_FORMAT_PCM:
221             break;
222         default:
223             av_log(avctx, AV_LOG_ERROR, "unsupported wave format\n");
224             return -1;
225     }
226
227     avctx->channels = get_le16(&hb);
228     avctx->sample_rate = get_le32(&hb);
229     avctx->bit_rate = get_le32(&hb) * 8;
230     avctx->block_align = get_le16(&hb);
231     avctx->bits_per_coded_sample = get_le16(&hb);
232
233     if (avctx->bits_per_coded_sample != 16) {
234         av_log(avctx, AV_LOG_ERROR, "unsupported number of bits per sample\n");
235         return -1;
236     }
237
238     len -= 16;
239     if (len > 0)
240         av_log(avctx, AV_LOG_INFO, "%d header bytes unparsed\n", len);
241
242     return 0;
243 }
244
245 static int16_t * interleave_buffer(int16_t *samples, int nchan, int blocksize, int32_t **buffer) {
246     int i, chan;
247     for (i=0; i<blocksize; i++)
248         for (chan=0; chan < nchan; chan++)
249             *samples++ = FFMIN(buffer[chan][i], 32768);
250     return samples;
251 }
252
253 static void decode_subframe_lpc(ShortenContext *s, int channel, int residual_size, int pred_order)
254 {
255     int sum, i, j;
256     int coeffs[pred_order];
257
258     for (i=0; i<pred_order; i++)
259         coeffs[i] = get_sr_golomb_shorten(&s->gb, LPCQUANT);
260
261     for (i=0; i < s->blocksize; i++) {
262         sum = s->lpcqoffset;
263         for (j=0; j<pred_order; j++)
264             sum += coeffs[j] * s->decoded[channel][i-j-1];
265         s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) + (sum >> LPCQUANT);
266     }
267 }
268
269
270 static int shorten_decode_frame(AVCodecContext *avctx,
271         void *data, int *data_size,
272         AVPacket *avpkt)
273 {
274     const uint8_t *buf = avpkt->data;
275     int buf_size = avpkt->size;
276     ShortenContext *s = avctx->priv_data;
277     int i, input_buf_size = 0;
278     int16_t *samples = data;
279     if(s->max_framesize == 0){
280         s->max_framesize= 1024; // should hopefully be enough for the first header
281         s->bitstream= av_fast_realloc(s->bitstream, &s->allocated_bitstream_size, s->max_framesize);
282     }
283
284     if(1 && s->max_framesize){//FIXME truncated
285         buf_size= FFMIN(buf_size, s->max_framesize - s->bitstream_size);
286         input_buf_size= buf_size;
287
288         if(s->bitstream_index + s->bitstream_size + buf_size > s->allocated_bitstream_size){
289             //                printf("memmove\n");
290             memmove(s->bitstream, &s->bitstream[s->bitstream_index], s->bitstream_size);
291             s->bitstream_index=0;
292         }
293         memcpy(&s->bitstream[s->bitstream_index + s->bitstream_size], buf, buf_size);
294         buf= &s->bitstream[s->bitstream_index];
295         buf_size += s->bitstream_size;
296         s->bitstream_size= buf_size;
297
298         if(buf_size < s->max_framesize){
299             //dprintf(avctx, "wanna more data ... %d\n", buf_size);
300             *data_size = 0;
301             return input_buf_size;
302         }
303     }
304     init_get_bits(&s->gb, buf, buf_size*8);
305     skip_bits(&s->gb, s->bitindex);
306     if (!s->blocksize)
307     {
308         int maxnlpc = 0;
309         /* shorten signature */
310         if (get_bits_long(&s->gb, 32) != AV_RB32("ajkg")) {
311             av_log(s->avctx, AV_LOG_ERROR, "missing shorten magic 'ajkg'\n");
312             return -1;
313         }
314
315         s->lpcqoffset = 0;
316         s->blocksize = DEFAULT_BLOCK_SIZE;
317         s->channels = 1;
318         s->nmean = -1;
319         s->version = get_bits(&s->gb, 8);
320         s->internal_ftype = get_uint(s, TYPESIZE);
321
322         s->channels = get_uint(s, CHANSIZE);
323         if (s->channels > MAX_CHANNELS) {
324             av_log(s->avctx, AV_LOG_ERROR, "too many channels: %d\n", s->channels);
325             return -1;
326         }
327
328         /* get blocksize if version > 0 */
329         if (s->version > 0) {
330             int skip_bytes;
331             s->blocksize = get_uint(s, av_log2(DEFAULT_BLOCK_SIZE));
332             maxnlpc = get_uint(s, LPCQSIZE);
333             s->nmean = get_uint(s, 0);
334
335             skip_bytes = get_uint(s, NSKIPSIZE);
336             for (i=0; i<skip_bytes; i++) {
337                 skip_bits(&s->gb, 8);
338             }
339         }
340         s->nwrap = FFMAX(NWRAP, maxnlpc);
341
342         if (allocate_buffers(s))
343             return -1;
344
345         init_offset(s);
346
347         if (s->version > 1)
348             s->lpcqoffset = V2LPCQOFFSET;
349
350         if (get_ur_golomb_shorten(&s->gb, FNSIZE) != FN_VERBATIM) {
351             av_log(s->avctx, AV_LOG_ERROR, "missing verbatim section at beginning of stream\n");
352             return -1;
353         }
354
355         s->header_size = get_ur_golomb_shorten(&s->gb, VERBATIM_CKSIZE_SIZE);
356         if (s->header_size >= OUT_BUFFER_SIZE || s->header_size < CANONICAL_HEADER_SIZE) {
357             av_log(s->avctx, AV_LOG_ERROR, "header is wrong size: %d\n", s->header_size);
358             return -1;
359         }
360
361         for (i=0; i<s->header_size; i++)
362             s->header[i] = (char)get_ur_golomb_shorten(&s->gb, VERBATIM_BYTE_SIZE);
363
364         if (decode_wave_header(avctx, s->header, s->header_size) < 0)
365             return -1;
366
367         s->cur_chan = 0;
368         s->bitshift = 0;
369     }
370     else
371     {
372         int cmd;
373         int len;
374         cmd = get_ur_golomb_shorten(&s->gb, FNSIZE);
375         switch (cmd) {
376             case FN_ZERO:
377             case FN_DIFF0:
378             case FN_DIFF1:
379             case FN_DIFF2:
380             case FN_DIFF3:
381             case FN_QLPC:
382                 {
383                     int residual_size = 0;
384                     int channel = s->cur_chan;
385                     int32_t coffset;
386                     if (cmd != FN_ZERO) {
387                         residual_size = get_ur_golomb_shorten(&s->gb, ENERGYSIZE);
388                         /* this is a hack as version 0 differed in defintion of get_sr_golomb_shorten */
389                         if (s->version == 0)
390                             residual_size--;
391                     }
392
393                     if (s->nmean == 0)
394                         coffset = s->offset[channel][0];
395                     else {
396                         int32_t sum = (s->version < 2) ? 0 : s->nmean / 2;
397                         for (i=0; i<s->nmean; i++)
398                             sum += s->offset[channel][i];
399                         coffset = sum / s->nmean;
400                         if (s->version >= 2)
401                             coffset >>= FFMIN(1, s->bitshift);
402                     }
403                     switch (cmd) {
404                         case FN_ZERO:
405                             for (i=0; i<s->blocksize; i++)
406                                 s->decoded[channel][i] = 0;
407                             break;
408                         case FN_DIFF0:
409                             for (i=0; i<s->blocksize; i++)
410                                 s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) + coffset;
411                             break;
412                         case FN_DIFF1:
413                             for (i=0; i<s->blocksize; i++)
414                                 s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) + s->decoded[channel][i - 1];
415                             break;
416                         case FN_DIFF2:
417                             for (i=0; i<s->blocksize; i++)
418                                 s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) + 2*s->decoded[channel][i-1]
419                                                                                                       -   s->decoded[channel][i-2];
420                             break;
421                         case FN_DIFF3:
422                             for (i=0; i<s->blocksize; i++)
423                                 s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) + 3*s->decoded[channel][i-1]
424                                                                                                       - 3*s->decoded[channel][i-2]
425                                                                                                       +   s->decoded[channel][i-3];
426                             break;
427                         case FN_QLPC:
428                             {
429                                 int pred_order = get_ur_golomb_shorten(&s->gb, LPCQSIZE);
430                                 for (i=0; i<pred_order; i++)
431                                     s->decoded[channel][i - pred_order] -= coffset;
432                                 decode_subframe_lpc(s, channel, residual_size, pred_order);
433                                 if (coffset != 0)
434                                     for (i=0; i < s->blocksize; i++)
435                                         s->decoded[channel][i] += coffset;
436                             }
437                     }
438                     if (s->nmean > 0) {
439                         int32_t sum = (s->version < 2) ? 0 : s->blocksize / 2;
440                         for (i=0; i<s->blocksize; i++)
441                             sum += s->decoded[channel][i];
442
443                         for (i=1; i<s->nmean; i++)
444                             s->offset[channel][i-1] = s->offset[channel][i];
445
446                         if (s->version < 2)
447                             s->offset[channel][s->nmean - 1] = sum / s->blocksize;
448                         else
449                             s->offset[channel][s->nmean - 1] = (sum / s->blocksize) << s->bitshift;
450                     }
451                     for (i=-s->nwrap; i<0; i++)
452                         s->decoded[channel][i] = s->decoded[channel][i + s->blocksize];
453
454                     fix_bitshift(s, s->decoded[channel]);
455
456                     s->cur_chan++;
457                     if (s->cur_chan == s->channels) {
458                         samples = interleave_buffer(samples, s->channels, s->blocksize, s->decoded);
459                         s->cur_chan = 0;
460                         goto frame_done;
461                     }
462                     break;
463                 }
464                 break;
465             case FN_VERBATIM:
466                 len = get_ur_golomb_shorten(&s->gb, VERBATIM_CKSIZE_SIZE);
467                 while (len--) {
468                     get_ur_golomb_shorten(&s->gb, VERBATIM_BYTE_SIZE);
469                 }
470                 break;
471             case FN_BITSHIFT:
472                 s->bitshift = get_ur_golomb_shorten(&s->gb, BITSHIFTSIZE);
473                 break;
474             case FN_BLOCKSIZE:
475                 s->blocksize = get_uint(s, av_log2(s->blocksize));
476                 break;
477             case FN_QUIT:
478                 *data_size = 0;
479                 return buf_size;
480                 break;
481             default:
482                 av_log(avctx, AV_LOG_ERROR, "unknown shorten function %d\n", cmd);
483                 return -1;
484                 break;
485         }
486     }
487 frame_done:
488     *data_size = (int8_t *)samples - (int8_t *)data;
489
490     //    s->last_blocksize = s->blocksize;
491     s->bitindex = get_bits_count(&s->gb) - 8*((get_bits_count(&s->gb))/8);
492     i= (get_bits_count(&s->gb))/8;
493     if (i > buf_size) {
494         av_log(s->avctx, AV_LOG_ERROR, "overread: %d\n", i - buf_size);
495         s->bitstream_size=0;
496         s->bitstream_index=0;
497         return -1;
498     }
499     if (s->bitstream_size) {
500         s->bitstream_index += i;
501         s->bitstream_size  -= i;
502         return input_buf_size;
503     } else
504         return i;
505 }
506
507 static av_cold int shorten_decode_close(AVCodecContext *avctx)
508 {
509     ShortenContext *s = avctx->priv_data;
510     int i;
511
512     for (i = 0; i < s->channels; i++) {
513         s->decoded[i] -= s->nwrap;
514         av_freep(&s->decoded[i]);
515         av_freep(&s->offset[i]);
516     }
517     av_freep(&s->bitstream);
518     return 0;
519 }
520
521 static void shorten_flush(AVCodecContext *avctx){
522     ShortenContext *s = avctx->priv_data;
523
524     s->bitstream_size=
525         s->bitstream_index= 0;
526 }
527
528 AVCodec shorten_decoder = {
529     "shorten",
530     AVMEDIA_TYPE_AUDIO,
531     CODEC_ID_SHORTEN,
532     sizeof(ShortenContext),
533     shorten_decode_init,
534     NULL,
535     shorten_decode_close,
536     shorten_decode_frame,
537     .flush= shorten_flush,
538     .long_name= NULL_IF_CONFIG_SMALL("Shorten"),
539 };