make mpa_decode_header() 10 times faster
[FFMpeg-mirror/ordered_chapters.git] / libavcodec / tta.c
blob979a94a742a283870ad4e96256e0d188a0f6ec6e
1 /*
2 * TTA (The Lossless True Audio) decoder
3 * Copyright (c) 2006 Alex Beregszaszi
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 /**
21 * @file tta.c
22 * TTA (The Lossless True Audio) decoder
23 * (www.true-audio.com or tta.corecodec.org)
24 * @author Alex Beregszaszi
28 #define ALT_BITSTREAM_READER_LE
29 //#define DEBUG
30 #include <limits.h>
31 #include "avcodec.h"
32 #include "bitstream.h"
34 #define FORMAT_INT 1
35 #define FORMAT_FLOAT 3
37 typedef struct TTAContext {
38 AVCodecContext *avctx;
39 GetBitContext gb;
41 int flags, channels, bps, is_float, data_length;
42 int frame_length, last_frame_length, total_frames;
44 int32_t *decode_buffer;
45 } TTAContext;
47 #if 0
48 static inline int shift_1(int i)
50 if (i < 32)
51 return 1 << i;
52 else
53 return 0x80000000; // 16 << 31
56 static inline int shift_16(int i)
58 if (i < 28)
59 return 16 << i;
60 else
61 return 0x80000000; // 16 << 27
63 #else
64 static const uint32_t shift_1[] = {
65 0x00000001, 0x00000002, 0x00000004, 0x00000008,
66 0x00000010, 0x00000020, 0x00000040, 0x00000080,
67 0x00000100, 0x00000200, 0x00000400, 0x00000800,
68 0x00001000, 0x00002000, 0x00004000, 0x00008000,
69 0x00010000, 0x00020000, 0x00040000, 0x00080000,
70 0x00100000, 0x00200000, 0x00400000, 0x00800000,
71 0x01000000, 0x02000000, 0x04000000, 0x08000000,
72 0x10000000, 0x20000000, 0x40000000, 0x80000000,
73 0x80000000, 0x80000000, 0x80000000, 0x80000000,
74 0x80000000, 0x80000000, 0x80000000, 0x80000000
77 static const uint32_t *shift_16 = shift_1 + 4;
78 #endif
80 #define MAX_ORDER 16
81 typedef struct TTAFilter {
82 int32_t shift, round, error, mode;
83 int32_t qm[MAX_ORDER];
84 int32_t dx[MAX_ORDER];
85 int32_t dl[MAX_ORDER];
86 } TTAFilter;
88 static int32_t ttafilter_configs[4][2] = {
89 {10, 1},
90 {9, 1},
91 {10, 1},
92 {12, 0}
95 static void ttafilter_init(TTAFilter *c, int32_t shift, int32_t mode) {
96 memset(c, 0, sizeof(TTAFilter));
97 c->shift = shift;
98 c->round = shift_1[shift-1];
99 // c->round = 1 << (shift - 1);
100 c->mode = mode;
103 // FIXME: copy paste from original
104 static inline void memshl(register int32_t *a, register int32_t *b) {
105 *a++ = *b++;
106 *a++ = *b++;
107 *a++ = *b++;
108 *a++ = *b++;
109 *a++ = *b++;
110 *a++ = *b++;
111 *a++ = *b++;
112 *a = *b;
115 // FIXME: copy paste from original
116 // mode=1 encoder, mode=0 decoder
117 static inline void ttafilter_process(TTAFilter *c, int32_t *in, int32_t mode) {
118 register int32_t *dl = c->dl, *qm = c->qm, *dx = c->dx, sum = c->round;
120 if (!c->error) {
121 sum += *dl++ * *qm, qm++;
122 sum += *dl++ * *qm, qm++;
123 sum += *dl++ * *qm, qm++;
124 sum += *dl++ * *qm, qm++;
125 sum += *dl++ * *qm, qm++;
126 sum += *dl++ * *qm, qm++;
127 sum += *dl++ * *qm, qm++;
128 sum += *dl++ * *qm, qm++;
129 dx += 8;
130 } else if(c->error < 0) {
131 sum += *dl++ * (*qm -= *dx++), qm++;
132 sum += *dl++ * (*qm -= *dx++), qm++;
133 sum += *dl++ * (*qm -= *dx++), qm++;
134 sum += *dl++ * (*qm -= *dx++), qm++;
135 sum += *dl++ * (*qm -= *dx++), qm++;
136 sum += *dl++ * (*qm -= *dx++), qm++;
137 sum += *dl++ * (*qm -= *dx++), qm++;
138 sum += *dl++ * (*qm -= *dx++), qm++;
139 } else {
140 sum += *dl++ * (*qm += *dx++), qm++;
141 sum += *dl++ * (*qm += *dx++), qm++;
142 sum += *dl++ * (*qm += *dx++), qm++;
143 sum += *dl++ * (*qm += *dx++), qm++;
144 sum += *dl++ * (*qm += *dx++), qm++;
145 sum += *dl++ * (*qm += *dx++), qm++;
146 sum += *dl++ * (*qm += *dx++), qm++;
147 sum += *dl++ * (*qm += *dx++), qm++;
150 *(dx-0) = ((*(dl-1) >> 30) | 1) << 2;
151 *(dx-1) = ((*(dl-2) >> 30) | 1) << 1;
152 *(dx-2) = ((*(dl-3) >> 30) | 1) << 1;
153 *(dx-3) = ((*(dl-4) >> 30) | 1);
155 // compress
156 if (mode) {
157 *dl = *in;
158 *in -= (sum >> c->shift);
159 c->error = *in;
160 } else {
161 c->error = *in;
162 *in += (sum >> c->shift);
163 *dl = *in;
166 if (c->mode) {
167 *(dl-1) = *dl - *(dl-1);
168 *(dl-2) = *(dl-1) - *(dl-2);
169 *(dl-3) = *(dl-2) - *(dl-3);
172 memshl(c->dl, c->dl + 1);
173 memshl(c->dx, c->dx + 1);
176 typedef struct TTARice {
177 uint32_t k0, k1, sum0, sum1;
178 } TTARice;
180 static void rice_init(TTARice *c, uint32_t k0, uint32_t k1)
182 c->k0 = k0;
183 c->k1 = k1;
184 c->sum0 = shift_16[k0];
185 c->sum1 = shift_16[k1];
188 static int tta_get_unary(GetBitContext *gb)
190 int ret = 0;
192 // count ones
193 while(get_bits1(gb))
194 ret++;
195 return ret;
198 // shamelessly copied from shorten.c
199 static int inline get_le16(GetBitContext *gb)
201 return bswap_16(get_bits_long(gb, 16));
204 static int inline get_le32(GetBitContext *gb)
206 return bswap_32(get_bits_long(gb, 32));
209 static int tta_decode_init(AVCodecContext * avctx)
211 TTAContext *s = avctx->priv_data;
212 int i;
214 s->avctx = avctx;
216 // 30bytes includes a seektable with one frame
217 if (avctx->extradata_size < 30)
218 return -1;
220 init_get_bits(&s->gb, avctx->extradata, avctx->extradata_size);
221 if (show_bits_long(&s->gb, 32) == bswap_32(ff_get_fourcc("TTA1")))
223 /* signature */
224 skip_bits(&s->gb, 32);
225 // if (get_bits_long(&s->gb, 32) != bswap_32(ff_get_fourcc("TTA1"))) {
226 // av_log(s->avctx, AV_LOG_ERROR, "Missing magic\n");
227 // return -1;
228 // }
230 s->flags = get_le16(&s->gb);
231 if (s->flags != 1 && s->flags != 3)
233 av_log(s->avctx, AV_LOG_ERROR, "Invalid flags\n");
234 return -1;
236 s->is_float = (s->flags == FORMAT_FLOAT);
237 avctx->channels = s->channels = get_le16(&s->gb);
238 avctx->bits_per_sample = get_le16(&s->gb);
239 s->bps = (avctx->bits_per_sample + 7) / 8;
240 avctx->sample_rate = get_le32(&s->gb);
241 if(avctx->sample_rate > 1000000){ //prevent FRAME_TIME * avctx->sample_rate from overflowing and sanity check
242 av_log(avctx, AV_LOG_ERROR, "sample_rate too large\n");
243 return -1;
245 s->data_length = get_le32(&s->gb);
246 skip_bits(&s->gb, 32); // CRC32 of header
248 if (s->is_float)
250 avctx->sample_fmt = SAMPLE_FMT_FLT;
251 av_log(s->avctx, AV_LOG_ERROR, "Unsupported sample format. Please contact the developers.\n");
252 return -1;
254 else switch(s->bps) {
255 // case 1: avctx->sample_fmt = SAMPLE_FMT_U8; break;
256 case 2: avctx->sample_fmt = SAMPLE_FMT_S16; break;
257 // case 3: avctx->sample_fmt = SAMPLE_FMT_S24; break;
258 case 4: avctx->sample_fmt = SAMPLE_FMT_S32; break;
259 default:
260 av_log(s->avctx, AV_LOG_ERROR, "Invalid/unsupported sample format. Please contact the developers.\n");
261 return -1;
264 // FIXME: horribly broken, but directly from reference source
265 #define FRAME_TIME 1.04489795918367346939
266 s->frame_length = (int)(FRAME_TIME * avctx->sample_rate);
268 s->last_frame_length = s->data_length % s->frame_length;
269 s->total_frames = s->data_length / s->frame_length +
270 (s->last_frame_length ? 1 : 0);
272 av_log(s->avctx, AV_LOG_DEBUG, "flags: %x chans: %d bps: %d rate: %d block: %d\n",
273 s->flags, avctx->channels, avctx->bits_per_sample, avctx->sample_rate,
274 avctx->block_align);
275 av_log(s->avctx, AV_LOG_DEBUG, "data_length: %d frame_length: %d last: %d total: %d\n",
276 s->data_length, s->frame_length, s->last_frame_length, s->total_frames);
278 // FIXME: seek table
279 for (i = 0; i < s->total_frames; i++)
280 skip_bits(&s->gb, 32);
281 skip_bits(&s->gb, 32); // CRC32 of seektable
283 if(s->frame_length >= UINT_MAX / (s->channels * sizeof(int32_t))){
284 av_log(avctx, AV_LOG_ERROR, "frame_length too large\n");
285 return -1;
288 s->decode_buffer = av_mallocz(sizeof(int32_t)*s->frame_length*s->channels);
289 } else {
290 av_log(avctx, AV_LOG_ERROR, "Wrong extradata present\n");
291 return -1;
294 return 0;
297 static int tta_decode_frame(AVCodecContext *avctx,
298 void *data, int *data_size,
299 uint8_t *buf, int buf_size)
301 TTAContext *s = avctx->priv_data;
302 int i;
304 init_get_bits(&s->gb, buf, buf_size*8);
306 int32_t predictors[s->channels];
307 TTAFilter filters[s->channels];
308 TTARice rices[s->channels];
309 int cur_chan = 0, framelen = s->frame_length;
310 int32_t *p;
312 // FIXME: seeking
313 s->total_frames--;
314 if (!s->total_frames && s->last_frame_length)
315 framelen = s->last_frame_length;
317 // init per channel states
318 for (i = 0; i < s->channels; i++) {
319 predictors[i] = 0;
320 ttafilter_init(&(filters[i]), ttafilter_configs[s->bps-1][0], ttafilter_configs[s->bps-1][1]);
321 rice_init(&(rices[i]), 10, 10);
324 for (p = s->decode_buffer; p < s->decode_buffer + (framelen * s->channels); p++) {
325 int32_t *predictor = &(predictors[cur_chan]);
326 TTAFilter *filter = &(filters[cur_chan]);
327 TTARice *rice = &(rices[cur_chan]);
328 uint32_t unary, depth, k;
329 int32_t value;
331 unary = tta_get_unary(&s->gb);
333 if (unary == 0) {
334 depth = 0;
335 k = rice->k0;
336 } else {
337 depth = 1;
338 k = rice->k1;
339 unary--;
342 if (k)
343 value = (unary << k) + get_bits(&s->gb, k);
344 else
345 value = unary;
347 // FIXME: copy paste from original
348 switch (depth) {
349 case 1:
350 rice->sum1 += value - (rice->sum1 >> 4);
351 if (rice->k1 > 0 && rice->sum1 < shift_16[rice->k1])
352 rice->k1--;
353 else if(rice->sum1 > shift_16[rice->k1 + 1])
354 rice->k1++;
355 value += shift_1[rice->k0];
356 default:
357 rice->sum0 += value - (rice->sum0 >> 4);
358 if (rice->k0 > 0 && rice->sum0 < shift_16[rice->k0])
359 rice->k0--;
360 else if(rice->sum0 > shift_16[rice->k0 + 1])
361 rice->k0++;
364 // extract sign
365 #define SIGN(x) (((x)&1) ? (++(x)>>1) : (-(x)>>1))
366 *p = SIGN(value);
368 // run hybrid filter
369 ttafilter_process(filter, p, 0);
371 // fixed order prediction
372 #define PRED(x, k) (int32_t)((((uint64_t)x << k) - x) >> k)
373 switch (s->bps) {
374 case 1: *p += PRED(*predictor, 4); break;
375 case 2:
376 case 3: *p += PRED(*predictor, 5); break;
377 case 4: *p += *predictor; break;
379 *predictor = *p;
381 #if 0
382 // extract 32bit float from last two int samples
383 if (s->is_float && ((p - data) & 1)) {
384 uint32_t neg = *p & 0x80000000;
385 uint32_t hi = *(p - 1);
386 uint32_t lo = abs(*p) - 1;
388 hi += (hi || lo) ? 0x3f80 : 0;
389 // SWAP16: swap all the 16 bits
390 *(p - 1) = (hi << 16) | SWAP16(lo) | neg;
392 #endif
394 /*if ((get_bits_count(&s->gb)+7)/8 > buf_size)
396 av_log(NULL, AV_LOG_INFO, "overread!!\n");
397 break;
400 // flip channels
401 if (cur_chan < (s->channels-1))
402 cur_chan++;
403 else {
404 // decorrelate in case of stereo integer
405 if (!s->is_float && (s->channels > 1)) {
406 int32_t *r = p - 1;
407 for (*p += *r / 2; r > p - s->channels; r--)
408 *r = *(r + 1) - *r;
410 cur_chan = 0;
414 skip_bits(&s->gb, 32); // frame crc
416 // convert to output buffer
417 switch(s->bps) {
418 case 2: {
419 uint16_t *samples = data;
420 for (p = s->decode_buffer; p < s->decode_buffer + (framelen * s->channels); p++) {
421 // *samples++ = (unsigned char)*p;
422 // *samples++ = (unsigned char)(*p >> 8);
423 *samples++ = *p;
425 *data_size = (uint8_t *)samples - (uint8_t *)data;
426 break;
428 default:
429 av_log(s->avctx, AV_LOG_ERROR, "Error, only 16bit samples supported!\n");
433 // return get_bits_count(&s->gb)+7)/8;
434 return buf_size;
437 static int tta_decode_close(AVCodecContext *avctx) {
438 TTAContext *s = avctx->priv_data;
440 if (s->decode_buffer)
441 av_free(s->decode_buffer);
443 return 0;
446 AVCodec tta_decoder = {
447 "tta",
448 CODEC_TYPE_AUDIO,
449 CODEC_ID_TTA,
450 sizeof(TTAContext),
451 tta_decode_init,
452 NULL,
453 tta_decode_close,
454 tta_decode_frame,