blob: 979b92165b2f30617e1d4daf6b80f66769a66c98
1 | /* |
2 | * WavPack lossless audio encoder |
3 | * |
4 | * This file is part of FFmpeg. |
5 | * |
6 | * FFmpeg is free software; you can redistribute it and/or |
7 | * modify it under the terms of the GNU Lesser General Public |
8 | * License as published by the Free Software Foundation; either |
9 | * version 2.1 of the License, or (at your option) any later version. |
10 | * |
11 | * FFmpeg 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 FFmpeg; if not, write to the Free Software |
18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
19 | */ |
20 | |
21 | #define BITSTREAM_WRITER_LE |
22 | |
23 | #include "libavutil/intreadwrite.h" |
24 | #include "libavutil/opt.h" |
25 | #include "avcodec.h" |
26 | #include "internal.h" |
27 | #include "put_bits.h" |
28 | #include "bytestream.h" |
29 | #include "wavpackenc.h" |
30 | #include "wavpack.h" |
31 | |
32 | #define UPDATE_WEIGHT(weight, delta, source, result) \ |
33 | if ((source) && (result)) { \ |
34 | int32_t s = (int32_t) ((source) ^ (result)) >> 31; \ |
35 | weight = ((delta) ^ s) + ((weight) - s); \ |
36 | } |
37 | |
38 | #define APPLY_WEIGHT_F(weight, sample) ((((((sample) & 0xffff) * (weight)) >> 9) + \ |
39 | ((((sample) & ~0xffff) >> 9) * (weight)) + 1) >> 1) |
40 | |
41 | #define APPLY_WEIGHT_I(weight, sample) (((weight) * (sample) + 512) >> 10) |
42 | |
43 | #define APPLY_WEIGHT(weight, sample) ((sample) != (short) (sample) ? \ |
44 | APPLY_WEIGHT_F(weight, sample) : APPLY_WEIGHT_I (weight, sample)) |
45 | |
46 | #define CLEAR(destin) memset(&destin, 0, sizeof(destin)); |
47 | |
48 | #define SHIFT_LSB 13 |
49 | #define SHIFT_MASK (0x1FU << SHIFT_LSB) |
50 | |
51 | #define MAG_LSB 18 |
52 | #define MAG_MASK (0x1FU << MAG_LSB) |
53 | |
54 | #define SRATE_LSB 23 |
55 | #define SRATE_MASK (0xFU << SRATE_LSB) |
56 | |
57 | #define EXTRA_TRY_DELTAS 1 |
58 | #define EXTRA_ADJUST_DELTAS 2 |
59 | #define EXTRA_SORT_FIRST 4 |
60 | #define EXTRA_BRANCHES 8 |
61 | #define EXTRA_SORT_LAST 16 |
62 | |
63 | typedef struct WavPackExtraInfo { |
64 | struct Decorr dps[MAX_TERMS]; |
65 | int nterms, log_limit, gt16bit; |
66 | uint32_t best_bits; |
67 | } WavPackExtraInfo; |
68 | |
69 | typedef struct WavPackWords { |
70 | int pend_data, holding_one, zeros_acc; |
71 | int holding_zero, pend_count; |
72 | WvChannel c[2]; |
73 | } WavPackWords; |
74 | |
75 | typedef struct WavPackEncodeContext { |
76 | AVClass *class; |
77 | AVCodecContext *avctx; |
78 | PutBitContext pb; |
79 | int block_samples; |
80 | int buffer_size; |
81 | int sample_index; |
82 | int stereo, stereo_in; |
83 | int ch_offset; |
84 | |
85 | int32_t *samples[2]; |
86 | int samples_size[2]; |
87 | |
88 | int32_t *sampleptrs[MAX_TERMS+2][2]; |
89 | int sampleptrs_size[MAX_TERMS+2][2]; |
90 | |
91 | int32_t *temp_buffer[2][2]; |
92 | int temp_buffer_size[2][2]; |
93 | |
94 | int32_t *best_buffer[2]; |
95 | int best_buffer_size[2]; |
96 | |
97 | int32_t *js_left, *js_right; |
98 | int js_left_size, js_right_size; |
99 | |
100 | int32_t *orig_l, *orig_r; |
101 | int orig_l_size, orig_r_size; |
102 | |
103 | unsigned extra_flags; |
104 | int optimize_mono; |
105 | int decorr_filter; |
106 | int joint; |
107 | int num_branches; |
108 | |
109 | uint32_t flags; |
110 | uint32_t crc_x; |
111 | WavPackWords w; |
112 | |
113 | uint8_t int32_sent_bits, int32_zeros, int32_ones, int32_dups; |
114 | uint8_t float_flags, float_shift, float_max_exp, max_exp; |
115 | int32_t shifted_ones, shifted_zeros, shifted_both; |
116 | int32_t false_zeros, neg_zeros, ordata; |
117 | |
118 | int num_terms, shift, joint_stereo, false_stereo; |
119 | int num_decorrs, num_passes, best_decorr, mask_decorr; |
120 | struct Decorr decorr_passes[MAX_TERMS]; |
121 | const WavPackDecorrSpec *decorr_specs; |
122 | float delta_decay; |
123 | } WavPackEncodeContext; |
124 | |
125 | static av_cold int wavpack_encode_init(AVCodecContext *avctx) |
126 | { |
127 | WavPackEncodeContext *s = avctx->priv_data; |
128 | |
129 | s->avctx = avctx; |
130 | |
131 | if (avctx->channels > 255) { |
132 | av_log(avctx, AV_LOG_ERROR, "Invalid channel count: %d\n", avctx->channels); |
133 | return AVERROR(EINVAL); |
134 | } |
135 | |
136 | if (!avctx->frame_size) { |
137 | int block_samples; |
138 | if (!(avctx->sample_rate & 1)) |
139 | block_samples = avctx->sample_rate / 2; |
140 | else |
141 | block_samples = avctx->sample_rate; |
142 | |
143 | while (block_samples * avctx->channels > WV_MAX_SAMPLES) |
144 | block_samples /= 2; |
145 | |
146 | while (block_samples * avctx->channels < 40000) |
147 | block_samples *= 2; |
148 | avctx->frame_size = block_samples; |
149 | } else if (avctx->frame_size && (avctx->frame_size < 128 || |
150 | avctx->frame_size > WV_MAX_SAMPLES)) { |
151 | av_log(avctx, AV_LOG_ERROR, "invalid block size: %d\n", avctx->frame_size); |
152 | return AVERROR(EINVAL); |
153 | } |
154 | |
155 | if (avctx->compression_level != FF_COMPRESSION_DEFAULT) { |
156 | if (avctx->compression_level >= 3) { |
157 | s->decorr_filter = 3; |
158 | s->num_passes = 9; |
159 | if (avctx->compression_level >= 8) { |
160 | s->num_branches = 4; |
161 | s->extra_flags = EXTRA_TRY_DELTAS|EXTRA_ADJUST_DELTAS|EXTRA_SORT_FIRST|EXTRA_SORT_LAST|EXTRA_BRANCHES; |
162 | } else if (avctx->compression_level >= 7) { |
163 | s->num_branches = 3; |
164 | s->extra_flags = EXTRA_TRY_DELTAS|EXTRA_ADJUST_DELTAS|EXTRA_SORT_FIRST|EXTRA_BRANCHES; |
165 | } else if (avctx->compression_level >= 6) { |
166 | s->num_branches = 2; |
167 | s->extra_flags = EXTRA_TRY_DELTAS|EXTRA_ADJUST_DELTAS|EXTRA_SORT_FIRST|EXTRA_BRANCHES; |
168 | } else if (avctx->compression_level >= 5) { |
169 | s->num_branches = 1; |
170 | s->extra_flags = EXTRA_TRY_DELTAS|EXTRA_ADJUST_DELTAS|EXTRA_SORT_FIRST|EXTRA_BRANCHES; |
171 | } else if (avctx->compression_level >= 4) { |
172 | s->num_branches = 1; |
173 | s->extra_flags = EXTRA_TRY_DELTAS|EXTRA_ADJUST_DELTAS|EXTRA_BRANCHES; |
174 | } |
175 | } else if (avctx->compression_level == 2) { |
176 | s->decorr_filter = 2; |
177 | s->num_passes = 4; |
178 | } else if (avctx->compression_level == 1) { |
179 | s->decorr_filter = 1; |
180 | s->num_passes = 2; |
181 | } else if (avctx->compression_level < 1) { |
182 | s->decorr_filter = 0; |
183 | s->num_passes = 0; |
184 | } |
185 | } |
186 | |
187 | s->num_decorrs = decorr_filter_sizes[s->decorr_filter]; |
188 | s->decorr_specs = decorr_filters[s->decorr_filter]; |
189 | |
190 | s->delta_decay = 2.0; |
191 | |
192 | return 0; |
193 | } |
194 | |
195 | static void shift_mono(int32_t *samples, int nb_samples, int shift) |
196 | { |
197 | int i; |
198 | for (i = 0; i < nb_samples; i++) |
199 | samples[i] >>= shift; |
200 | } |
201 | |
202 | static void shift_stereo(int32_t *left, int32_t *right, |
203 | int nb_samples, int shift) |
204 | { |
205 | int i; |
206 | for (i = 0; i < nb_samples; i++) { |
207 | left [i] >>= shift; |
208 | right[i] >>= shift; |
209 | } |
210 | } |
211 | |
212 | #define FLOAT_SHIFT_ONES 1 |
213 | #define FLOAT_SHIFT_SAME 2 |
214 | #define FLOAT_SHIFT_SENT 4 |
215 | #define FLOAT_ZEROS_SENT 8 |
216 | #define FLOAT_NEG_ZEROS 0x10 |
217 | #define FLOAT_EXCEPTIONS 0x20 |
218 | |
219 | #define get_mantissa(f) ((f) & 0x7fffff) |
220 | #define get_exponent(f) (((f) >> 23) & 0xff) |
221 | #define get_sign(f) (((f) >> 31) & 0x1) |
222 | |
223 | static void process_float(WavPackEncodeContext *s, int32_t *sample) |
224 | { |
225 | int32_t shift_count, value, f = *sample; |
226 | |
227 | if (get_exponent(f) == 255) { |
228 | s->float_flags |= FLOAT_EXCEPTIONS; |
229 | value = 0x1000000; |
230 | shift_count = 0; |
231 | } else if (get_exponent(f)) { |
232 | shift_count = s->max_exp - get_exponent(f); |
233 | value = 0x800000 + get_mantissa(f); |
234 | } else { |
235 | shift_count = s->max_exp ? s->max_exp - 1 : 0; |
236 | value = get_mantissa(f); |
237 | } |
238 | |
239 | if (shift_count < 25) |
240 | value >>= shift_count; |
241 | else |
242 | value = 0; |
243 | |
244 | if (!value) { |
245 | if (get_exponent(f) || get_mantissa(f)) |
246 | s->false_zeros++; |
247 | else if (get_sign(f)) |
248 | s->neg_zeros++; |
249 | } else if (shift_count) { |
250 | int32_t mask = (1 << shift_count) - 1; |
251 | |
252 | if (!(get_mantissa(f) & mask)) |
253 | s->shifted_zeros++; |
254 | else if ((get_mantissa(f) & mask) == mask) |
255 | s->shifted_ones++; |
256 | else |
257 | s->shifted_both++; |
258 | } |
259 | |
260 | s->ordata |= value; |
261 | *sample = get_sign(f) ? -value : value; |
262 | } |
263 | |
264 | static int scan_float(WavPackEncodeContext *s, |
265 | int32_t *samples_l, int32_t *samples_r, |
266 | int nb_samples) |
267 | { |
268 | uint32_t crc = 0xffffffffu; |
269 | int i; |
270 | |
271 | s->shifted_ones = s->shifted_zeros = s->shifted_both = s->ordata = 0; |
272 | s->float_shift = s->float_flags = 0; |
273 | s->false_zeros = s->neg_zeros = 0; |
274 | s->max_exp = 0; |
275 | |
276 | if (s->flags & WV_MONO_DATA) { |
277 | for (i = 0; i < nb_samples; i++) { |
278 | int32_t f = samples_l[i]; |
279 | crc = crc * 27 + get_mantissa(f) * 9 + get_exponent(f) * 3 + get_sign(f); |
280 | |
281 | if (get_exponent(f) > s->max_exp && get_exponent(f) < 255) |
282 | s->max_exp = get_exponent(f); |
283 | } |
284 | } else { |
285 | for (i = 0; i < nb_samples; i++) { |
286 | int32_t f; |
287 | |
288 | f = samples_l[i]; |
289 | crc = crc * 27 + get_mantissa(f) * 9 + get_exponent(f) * 3 + get_sign(f); |
290 | if (get_exponent(f) > s->max_exp && get_exponent(f) < 255) |
291 | s->max_exp = get_exponent(f); |
292 | |
293 | f = samples_r[i]; |
294 | crc = crc * 27 + get_mantissa(f) * 9 + get_exponent(f) * 3 + get_sign(f); |
295 | |
296 | if (get_exponent(f) > s->max_exp && get_exponent(f) < 255) |
297 | s->max_exp = get_exponent(f); |
298 | } |
299 | } |
300 | |
301 | s->crc_x = crc; |
302 | |
303 | if (s->flags & WV_MONO_DATA) { |
304 | for (i = 0; i < nb_samples; i++) |
305 | process_float(s, &samples_l[i]); |
306 | } else { |
307 | for (i = 0; i < nb_samples; i++) { |
308 | process_float(s, &samples_l[i]); |
309 | process_float(s, &samples_r[i]); |
310 | } |
311 | } |
312 | |
313 | s->float_max_exp = s->max_exp; |
314 | |
315 | if (s->shifted_both) |
316 | s->float_flags |= FLOAT_SHIFT_SENT; |
317 | else if (s->shifted_ones && !s->shifted_zeros) |
318 | s->float_flags |= FLOAT_SHIFT_ONES; |
319 | else if (s->shifted_ones && s->shifted_zeros) |
320 | s->float_flags |= FLOAT_SHIFT_SAME; |
321 | else if (s->ordata && !(s->ordata & 1)) { |
322 | do { |
323 | s->float_shift++; |
324 | s->ordata >>= 1; |
325 | } while (!(s->ordata & 1)); |
326 | |
327 | if (s->flags & WV_MONO_DATA) |
328 | shift_mono(samples_l, nb_samples, s->float_shift); |
329 | else |
330 | shift_stereo(samples_l, samples_r, nb_samples, s->float_shift); |
331 | } |
332 | |
333 | s->flags &= ~MAG_MASK; |
334 | |
335 | while (s->ordata) { |
336 | s->flags += 1 << MAG_LSB; |
337 | s->ordata >>= 1; |
338 | } |
339 | |
340 | if (s->false_zeros || s->neg_zeros) |
341 | s->float_flags |= FLOAT_ZEROS_SENT; |
342 | |
343 | if (s->neg_zeros) |
344 | s->float_flags |= FLOAT_NEG_ZEROS; |
345 | |
346 | return s->float_flags & (FLOAT_EXCEPTIONS | FLOAT_ZEROS_SENT | |
347 | FLOAT_SHIFT_SENT | FLOAT_SHIFT_SAME); |
348 | } |
349 | |
350 | static void scan_int23(WavPackEncodeContext *s, |
351 | int32_t *samples_l, int32_t *samples_r, |
352 | int nb_samples) |
353 | { |
354 | uint32_t magdata = 0, ordata = 0, xordata = 0, anddata = ~0; |
355 | int i, total_shift = 0; |
356 | |
357 | s->int32_sent_bits = s->int32_zeros = s->int32_ones = s->int32_dups = 0; |
358 | |
359 | if (s->flags & WV_MONO_DATA) { |
360 | for (i = 0; i < nb_samples; i++) { |
361 | int32_t M = samples_l[i]; |
362 | |
363 | magdata |= (M < 0) ? ~M : M; |
364 | xordata |= M ^ -(M & 1); |
365 | anddata &= M; |
366 | ordata |= M; |
367 | |
368 | if ((ordata & 1) && !(anddata & 1) && (xordata & 2)) |
369 | return; |
370 | } |
371 | } else { |
372 | for (i = 0; i < nb_samples; i++) { |
373 | int32_t L = samples_l[i]; |
374 | int32_t R = samples_r[i]; |
375 | |
376 | magdata |= (L < 0) ? ~L : L; |
377 | magdata |= (R < 0) ? ~R : R; |
378 | xordata |= L ^ -(L & 1); |
379 | xordata |= R ^ -(R & 1); |
380 | anddata &= L & R; |
381 | ordata |= L | R; |
382 | |
383 | if ((ordata & 1) && !(anddata & 1) && (xordata & 2)) |
384 | return; |
385 | } |
386 | } |
387 | |
388 | s->flags &= ~MAG_MASK; |
389 | |
390 | while (magdata) { |
391 | s->flags += 1 << MAG_LSB; |
392 | magdata >>= 1; |
393 | } |
394 | |
395 | if (!(s->flags & MAG_MASK)) |
396 | return; |
397 | |
398 | if (!(ordata & 1)) { |
399 | do { |
400 | s->flags -= 1 << MAG_LSB; |
401 | s->int32_zeros++; |
402 | total_shift++; |
403 | ordata >>= 1; |
404 | } while (!(ordata & 1)); |
405 | } else if (anddata & 1) { |
406 | do { |
407 | s->flags -= 1 << MAG_LSB; |
408 | s->int32_ones++; |
409 | total_shift++; |
410 | anddata >>= 1; |
411 | } while (anddata & 1); |
412 | } else if (!(xordata & 2)) { |
413 | do { |
414 | s->flags -= 1 << MAG_LSB; |
415 | s->int32_dups++; |
416 | total_shift++; |
417 | xordata >>= 1; |
418 | } while (!(xordata & 2)); |
419 | } |
420 | |
421 | if (total_shift) { |
422 | s->flags |= WV_INT32_DATA; |
423 | |
424 | if (s->flags & WV_MONO_DATA) |
425 | shift_mono(samples_l, nb_samples, total_shift); |
426 | else |
427 | shift_stereo(samples_l, samples_r, nb_samples, total_shift); |
428 | } |
429 | } |
430 | |
431 | static int scan_int32(WavPackEncodeContext *s, |
432 | int32_t *samples_l, int32_t *samples_r, |
433 | int nb_samples) |
434 | { |
435 | uint32_t magdata = 0, ordata = 0, xordata = 0, anddata = ~0; |
436 | uint32_t crc = 0xffffffffu; |
437 | int i, total_shift = 0; |
438 | |
439 | s->int32_sent_bits = s->int32_zeros = s->int32_ones = s->int32_dups = 0; |
440 | |
441 | if (s->flags & WV_MONO_DATA) { |
442 | for (i = 0; i < nb_samples; i++) { |
443 | int32_t M = samples_l[i]; |
444 | |
445 | crc = crc * 9 + (M & 0xffff) * 3 + ((M >> 16) & 0xffff); |
446 | magdata |= (M < 0) ? ~M : M; |
447 | xordata |= M ^ -(M & 1); |
448 | anddata &= M; |
449 | ordata |= M; |
450 | } |
451 | } else { |
452 | for (i = 0; i < nb_samples; i++) { |
453 | int32_t L = samples_l[i]; |
454 | int32_t R = samples_r[i]; |
455 | |
456 | crc = crc * 9 + (L & 0xffff) * 3 + ((L >> 16) & 0xffff); |
457 | crc = crc * 9 + (R & 0xffff) * 3 + ((R >> 16) & 0xffff); |
458 | magdata |= (L < 0) ? ~L : L; |
459 | magdata |= (R < 0) ? ~R : R; |
460 | xordata |= L ^ -(L & 1); |
461 | xordata |= R ^ -(R & 1); |
462 | anddata &= L & R; |
463 | ordata |= L | R; |
464 | } |
465 | } |
466 | |
467 | s->crc_x = crc; |
468 | s->flags &= ~MAG_MASK; |
469 | |
470 | while (magdata) { |
471 | s->flags += 1 << MAG_LSB; |
472 | magdata >>= 1; |
473 | } |
474 | |
475 | if (!((s->flags & MAG_MASK) >> MAG_LSB)) { |
476 | s->flags &= ~WV_INT32_DATA; |
477 | return 0; |
478 | } |
479 | |
480 | if (!(ordata & 1)) |
481 | do { |
482 | s->flags -= 1 << MAG_LSB; |
483 | s->int32_zeros++; |
484 | total_shift++; |
485 | ordata >>= 1; |
486 | } while (!(ordata & 1)); |
487 | else if (anddata & 1) |
488 | do { |
489 | s->flags -= 1 << MAG_LSB; |
490 | s->int32_ones++; |
491 | total_shift++; |
492 | anddata >>= 1; |
493 | } while (anddata & 1); |
494 | else if (!(xordata & 2)) |
495 | do { |
496 | s->flags -= 1 << MAG_LSB; |
497 | s->int32_dups++; |
498 | total_shift++; |
499 | xordata >>= 1; |
500 | } while (!(xordata & 2)); |
501 | |
502 | if (((s->flags & MAG_MASK) >> MAG_LSB) > 23) { |
503 | s->int32_sent_bits = (uint8_t)(((s->flags & MAG_MASK) >> MAG_LSB) - 23); |
504 | total_shift += s->int32_sent_bits; |
505 | s->flags &= ~MAG_MASK; |
506 | s->flags += 23 << MAG_LSB; |
507 | } |
508 | |
509 | if (total_shift) { |
510 | s->flags |= WV_INT32_DATA; |
511 | |
512 | if (s->flags & WV_MONO_DATA) |
513 | shift_mono(samples_l, nb_samples, total_shift); |
514 | else |
515 | shift_stereo(samples_l, samples_r, nb_samples, total_shift); |
516 | } |
517 | |
518 | return s->int32_sent_bits; |
519 | } |
520 | |
521 | static int8_t store_weight(int weight) |
522 | { |
523 | weight = av_clip(weight, -1024, 1024); |
524 | if (weight > 0) |
525 | weight -= (weight + 64) >> 7; |
526 | |
527 | return (weight + 4) >> 3; |
528 | } |
529 | |
530 | static int restore_weight(int8_t weight) |
531 | { |
532 | int result; |
533 | |
534 | if ((result = (int) weight << 3) > 0) |
535 | result += (result + 64) >> 7; |
536 | |
537 | return result; |
538 | } |
539 | |
540 | static int log2s(int32_t value) |
541 | { |
542 | return (value < 0) ? -wp_log2(-value) : wp_log2(value); |
543 | } |
544 | |
545 | static void decorr_mono(int32_t *in_samples, int32_t *out_samples, |
546 | int nb_samples, struct Decorr *dpp, int dir) |
547 | { |
548 | int m = 0, i; |
549 | |
550 | dpp->sumA = 0; |
551 | |
552 | if (dir < 0) { |
553 | out_samples += (nb_samples - 1); |
554 | in_samples += (nb_samples - 1); |
555 | } |
556 | |
557 | dpp->weightA = restore_weight(store_weight(dpp->weightA)); |
558 | |
559 | for (i = 0; i < MAX_TERM; i++) |
560 | dpp->samplesA[i] = wp_exp2(log2s(dpp->samplesA[i])); |
561 | |
562 | if (dpp->value > MAX_TERM) { |
563 | while (nb_samples--) { |
564 | int32_t left, sam_A; |
565 | |
566 | sam_A = ((3 - (dpp->value & 1)) * dpp->samplesA[0] - dpp->samplesA[1]) >> !(dpp->value & 1); |
567 | |
568 | dpp->samplesA[1] = dpp->samplesA[0]; |
569 | dpp->samplesA[0] = left = in_samples[0]; |
570 | |
571 | left -= APPLY_WEIGHT(dpp->weightA, sam_A); |
572 | UPDATE_WEIGHT(dpp->weightA, dpp->delta, sam_A, left); |
573 | dpp->sumA += dpp->weightA; |
574 | out_samples[0] = left; |
575 | in_samples += dir; |
576 | out_samples += dir; |
577 | } |
578 | } else if (dpp->value > 0) { |
579 | while (nb_samples--) { |
580 | int k = (m + dpp->value) & (MAX_TERM - 1); |
581 | int32_t left, sam_A; |
582 | |
583 | sam_A = dpp->samplesA[m]; |
584 | dpp->samplesA[k] = left = in_samples[0]; |
585 | m = (m + 1) & (MAX_TERM - 1); |
586 | |
587 | left -= APPLY_WEIGHT(dpp->weightA, sam_A); |
588 | UPDATE_WEIGHT(dpp->weightA, dpp->delta, sam_A, left); |
589 | dpp->sumA += dpp->weightA; |
590 | out_samples[0] = left; |
591 | in_samples += dir; |
592 | out_samples += dir; |
593 | } |
594 | } |
595 | |
596 | if (m && dpp->value > 0 && dpp->value <= MAX_TERM) { |
597 | int32_t temp_A[MAX_TERM]; |
598 | |
599 | memcpy(temp_A, dpp->samplesA, sizeof(dpp->samplesA)); |
600 | |
601 | for (i = 0; i < MAX_TERM; i++) { |
602 | dpp->samplesA[i] = temp_A[m]; |
603 | m = (m + 1) & (MAX_TERM - 1); |
604 | } |
605 | } |
606 | } |
607 | |
608 | static void reverse_mono_decorr(struct Decorr *dpp) |
609 | { |
610 | if (dpp->value > MAX_TERM) { |
611 | int32_t sam_A; |
612 | |
613 | if (dpp->value & 1) |
614 | sam_A = 2 * dpp->samplesA[0] - dpp->samplesA[1]; |
615 | else |
616 | sam_A = (3 * dpp->samplesA[0] - dpp->samplesA[1]) >> 1; |
617 | |
618 | dpp->samplesA[1] = dpp->samplesA[0]; |
619 | dpp->samplesA[0] = sam_A; |
620 | |
621 | if (dpp->value & 1) |
622 | sam_A = 2 * dpp->samplesA[0] - dpp->samplesA[1]; |
623 | else |
624 | sam_A = (3 * dpp->samplesA[0] - dpp->samplesA[1]) >> 1; |
625 | |
626 | dpp->samplesA[1] = sam_A; |
627 | } else if (dpp->value > 1) { |
628 | int i, j, k; |
629 | |
630 | for (i = 0, j = dpp->value - 1, k = 0; k < dpp->value / 2; i++, j--, k++) { |
631 | i &= (MAX_TERM - 1); |
632 | j &= (MAX_TERM - 1); |
633 | dpp->samplesA[i] ^= dpp->samplesA[j]; |
634 | dpp->samplesA[j] ^= dpp->samplesA[i]; |
635 | dpp->samplesA[i] ^= dpp->samplesA[j]; |
636 | } |
637 | } |
638 | } |
639 | |
640 | static uint32_t log2sample(uint32_t v, int limit, uint32_t *result) |
641 | { |
642 | uint32_t dbits; |
643 | |
644 | if ((v += v >> 9) < (1 << 8)) { |
645 | dbits = nbits_table[v]; |
646 | *result += (dbits << 8) + wp_log2_table[(v << (9 - dbits)) & 0xff]; |
647 | } else { |
648 | if (v < (1 << 16)) |
649 | dbits = nbits_table[v >> 8] + 8; |
650 | else if (v < (1 << 24)) |
651 | dbits = nbits_table[v >> 16] + 16; |
652 | else |
653 | dbits = nbits_table[v >> 24] + 24; |
654 | |
655 | *result += dbits = (dbits << 8) + wp_log2_table[(v >> (dbits - 9)) & 0xff]; |
656 | |
657 | if (limit && dbits >= limit) |
658 | return 1; |
659 | } |
660 | |
661 | return 0; |
662 | } |
663 | |
664 | static uint32_t log2mono(int32_t *samples, int nb_samples, int limit) |
665 | { |
666 | uint32_t result = 0; |
667 | while (nb_samples--) { |
668 | if (log2sample(abs(*samples++), limit, &result)) |
669 | return UINT32_MAX; |
670 | } |
671 | return result; |
672 | } |
673 | |
674 | static uint32_t log2stereo(int32_t *samples_l, int32_t *samples_r, |
675 | int nb_samples, int limit) |
676 | { |
677 | uint32_t result = 0; |
678 | while (nb_samples--) { |
679 | if (log2sample(abs(*samples_l++), limit, &result) || |
680 | log2sample(abs(*samples_r++), limit, &result)) |
681 | return UINT32_MAX; |
682 | } |
683 | return result; |
684 | } |
685 | |
686 | static void decorr_mono_buffer(int32_t *samples, int32_t *outsamples, |
687 | int nb_samples, struct Decorr *dpp, |
688 | int tindex) |
689 | { |
690 | struct Decorr dp, *dppi = dpp + tindex; |
691 | int delta = dppi->delta, pre_delta, term = dppi->value; |
692 | |
693 | if (delta == 7) |
694 | pre_delta = 7; |
695 | else if (delta < 2) |
696 | pre_delta = 3; |
697 | else |
698 | pre_delta = delta + 1; |
699 | |
700 | CLEAR(dp); |
701 | dp.value = term; |
702 | dp.delta = pre_delta; |
703 | decorr_mono(samples, outsamples, FFMIN(2048, nb_samples), &dp, -1); |
704 | dp.delta = delta; |
705 | |
706 | if (tindex == 0) |
707 | reverse_mono_decorr(&dp); |
708 | else |
709 | CLEAR(dp.samplesA); |
710 | |
711 | memcpy(dppi->samplesA, dp.samplesA, sizeof(dp.samplesA)); |
712 | dppi->weightA = dp.weightA; |
713 | |
714 | if (delta == 0) { |
715 | dp.delta = 1; |
716 | decorr_mono(samples, outsamples, nb_samples, &dp, 1); |
717 | dp.delta = 0; |
718 | memcpy(dp.samplesA, dppi->samplesA, sizeof(dp.samplesA)); |
719 | dppi->weightA = dp.weightA = dp.sumA / nb_samples; |
720 | } |
721 | |
722 | decorr_mono(samples, outsamples, nb_samples, &dp, 1); |
723 | } |
724 | |
725 | static void recurse_mono(WavPackEncodeContext *s, WavPackExtraInfo *info, |
726 | int depth, int delta, uint32_t input_bits) |
727 | { |
728 | int term, branches = s->num_branches - depth; |
729 | int32_t *samples, *outsamples; |
730 | uint32_t term_bits[22], bits; |
731 | |
732 | if (branches < 1 || depth + 1 == info->nterms) |
733 | branches = 1; |
734 | |
735 | CLEAR(term_bits); |
736 | samples = s->sampleptrs[depth][0]; |
737 | outsamples = s->sampleptrs[depth + 1][0]; |
738 | |
739 | for (term = 1; term <= 18; term++) { |
740 | if (term == 17 && branches == 1 && depth + 1 < info->nterms) |
741 | continue; |
742 | |
743 | if (term > 8 && term < 17) |
744 | continue; |
745 | |
746 | if (!s->extra_flags && (term > 4 && term < 17)) |
747 | continue; |
748 | |
749 | info->dps[depth].value = term; |
750 | info->dps[depth].delta = delta; |
751 | decorr_mono_buffer(samples, outsamples, s->block_samples, info->dps, depth); |
752 | bits = log2mono(outsamples, s->block_samples, info->log_limit); |
753 | |
754 | if (bits < info->best_bits) { |
755 | info->best_bits = bits; |
756 | CLEAR(s->decorr_passes); |
757 | memcpy(s->decorr_passes, info->dps, sizeof(info->dps[0]) * (depth + 1)); |
758 | memcpy(s->sampleptrs[info->nterms + 1][0], |
759 | s->sampleptrs[depth + 1][0], s->block_samples * 4); |
760 | } |
761 | |
762 | term_bits[term + 3] = bits; |
763 | } |
764 | |
765 | while (depth + 1 < info->nterms && branches--) { |
766 | uint32_t local_best_bits = input_bits; |
767 | int best_term = 0, i; |
768 | |
769 | for (i = 0; i < 22; i++) |
770 | if (term_bits[i] && term_bits[i] < local_best_bits) { |
771 | local_best_bits = term_bits[i]; |
772 | best_term = i - 3; |
773 | } |
774 | |
775 | if (!best_term) |
776 | break; |
777 | |
778 | term_bits[best_term + 3] = 0; |
779 | |
780 | info->dps[depth].value = best_term; |
781 | info->dps[depth].delta = delta; |
782 | decorr_mono_buffer(samples, outsamples, s->block_samples, info->dps, depth); |
783 | |
784 | recurse_mono(s, info, depth + 1, delta, local_best_bits); |
785 | } |
786 | } |
787 | |
788 | static void sort_mono(WavPackEncodeContext *s, WavPackExtraInfo *info) |
789 | { |
790 | int reversed = 1; |
791 | uint32_t bits; |
792 | |
793 | while (reversed) { |
794 | int ri, i; |
795 | |
796 | memcpy(info->dps, s->decorr_passes, sizeof(s->decorr_passes)); |
797 | reversed = 0; |
798 | |
799 | for (ri = 0; ri < info->nterms && s->decorr_passes[ri].value; ri++) { |
800 | |
801 | if (ri + 1 >= info->nterms || !s->decorr_passes[ri+1].value) |
802 | break; |
803 | |
804 | if (s->decorr_passes[ri].value == s->decorr_passes[ri+1].value) { |
805 | decorr_mono_buffer(s->sampleptrs[ri][0], s->sampleptrs[ri+1][0], |
806 | s->block_samples, info->dps, ri); |
807 | continue; |
808 | } |
809 | |
810 | info->dps[ri ] = s->decorr_passes[ri+1]; |
811 | info->dps[ri+1] = s->decorr_passes[ri ]; |
812 | |
813 | for (i = ri; i < info->nterms && s->decorr_passes[i].value; i++) |
814 | decorr_mono_buffer(s->sampleptrs[i][0], s->sampleptrs[i+1][0], |
815 | s->block_samples, info->dps, i); |
816 | |
817 | bits = log2mono(s->sampleptrs[i][0], s->block_samples, info->log_limit); |
818 | if (bits < info->best_bits) { |
819 | reversed = 1; |
820 | info->best_bits = bits; |
821 | CLEAR(s->decorr_passes); |
822 | memcpy(s->decorr_passes, info->dps, sizeof(info->dps[0]) * i); |
823 | memcpy(s->sampleptrs[info->nterms + 1][0], s->sampleptrs[i][0], |
824 | s->block_samples * 4); |
825 | } else { |
826 | info->dps[ri ] = s->decorr_passes[ri]; |
827 | info->dps[ri+1] = s->decorr_passes[ri+1]; |
828 | decorr_mono_buffer(s->sampleptrs[ri][0], s->sampleptrs[ri+1][0], |
829 | s->block_samples, info->dps, ri); |
830 | } |
831 | } |
832 | } |
833 | } |
834 | |
835 | static void delta_mono(WavPackEncodeContext *s, WavPackExtraInfo *info) |
836 | { |
837 | int lower = 0, delta, d; |
838 | uint32_t bits; |
839 | |
840 | if (!s->decorr_passes[0].value) |
841 | return; |
842 | delta = s->decorr_passes[0].delta; |
843 | |
844 | for (d = delta - 1; d >= 0; d--) { |
845 | int i; |
846 | |
847 | for (i = 0; i < info->nterms && s->decorr_passes[i].value; i++) { |
848 | info->dps[i].value = s->decorr_passes[i].value; |
849 | info->dps[i].delta = d; |
850 | decorr_mono_buffer(s->sampleptrs[i][0], s->sampleptrs[i+1][0], |
851 | s->block_samples, info->dps, i); |
852 | } |
853 | |
854 | bits = log2mono(s->sampleptrs[i][0], s->block_samples, info->log_limit); |
855 | if (bits >= info->best_bits) |
856 | break; |
857 | |
858 | lower = 1; |
859 | info->best_bits = bits; |
860 | CLEAR(s->decorr_passes); |
861 | memcpy(s->decorr_passes, info->dps, sizeof(info->dps[0]) * i); |
862 | memcpy(s->sampleptrs[info->nterms + 1][0], s->sampleptrs[i][0], |
863 | s->block_samples * 4); |
864 | } |
865 | |
866 | for (d = delta + 1; !lower && d <= 7; d++) { |
867 | int i; |
868 | |
869 | for (i = 0; i < info->nterms && s->decorr_passes[i].value; i++) { |
870 | info->dps[i].value = s->decorr_passes[i].value; |
871 | info->dps[i].delta = d; |
872 | decorr_mono_buffer(s->sampleptrs[i][0], s->sampleptrs[i+1][0], |
873 | s->block_samples, info->dps, i); |
874 | } |
875 | |
876 | bits = log2mono(s->sampleptrs[i][0], s->block_samples, info->log_limit); |
877 | if (bits >= info->best_bits) |
878 | break; |
879 | |
880 | info->best_bits = bits; |
881 | CLEAR(s->decorr_passes); |
882 | memcpy(s->decorr_passes, info->dps, sizeof(info->dps[0]) * i); |
883 | memcpy(s->sampleptrs[info->nterms + 1][0], s->sampleptrs[i][0], |
884 | s->block_samples * 4); |
885 | } |
886 | } |
887 | |
888 | static int allocate_buffers2(WavPackEncodeContext *s, int nterms) |
889 | { |
890 | int i; |
891 | |
892 | for (i = 0; i < nterms + 2; i++) { |
893 | av_fast_padded_malloc(&s->sampleptrs[i][0], &s->sampleptrs_size[i][0], |
894 | s->block_samples * 4); |
895 | if (!s->sampleptrs[i][0]) |
896 | return AVERROR(ENOMEM); |
897 | if (!(s->flags & WV_MONO_DATA)) { |
898 | av_fast_padded_malloc(&s->sampleptrs[i][1], &s->sampleptrs_size[i][1], |
899 | s->block_samples * 4); |
900 | if (!s->sampleptrs[i][1]) |
901 | return AVERROR(ENOMEM); |
902 | } |
903 | } |
904 | |
905 | return 0; |
906 | } |
907 | |
908 | static int allocate_buffers(WavPackEncodeContext *s) |
909 | { |
910 | int i; |
911 | |
912 | for (i = 0; i < 2; i++) { |
913 | av_fast_padded_malloc(&s->best_buffer[0], &s->best_buffer_size[0], |
914 | s->block_samples * 4); |
915 | if (!s->best_buffer[0]) |
916 | return AVERROR(ENOMEM); |
917 | |
918 | av_fast_padded_malloc(&s->temp_buffer[i][0], &s->temp_buffer_size[i][0], |
919 | s->block_samples * 4); |
920 | if (!s->temp_buffer[i][0]) |
921 | return AVERROR(ENOMEM); |
922 | if (!(s->flags & WV_MONO_DATA)) { |
923 | av_fast_padded_malloc(&s->best_buffer[1], &s->best_buffer_size[1], |
924 | s->block_samples * 4); |
925 | if (!s->best_buffer[1]) |
926 | return AVERROR(ENOMEM); |
927 | |
928 | av_fast_padded_malloc(&s->temp_buffer[i][1], &s->temp_buffer_size[i][1], |
929 | s->block_samples * 4); |
930 | if (!s->temp_buffer[i][1]) |
931 | return AVERROR(ENOMEM); |
932 | } |
933 | } |
934 | |
935 | return 0; |
936 | } |
937 | |
938 | static void analyze_mono(WavPackEncodeContext *s, int32_t *samples, int do_samples) |
939 | { |
940 | WavPackExtraInfo info; |
941 | int i; |
942 | |
943 | info.log_limit = (((s->flags & MAG_MASK) >> MAG_LSB) + 4) * 256; |
944 | info.log_limit = FFMIN(6912, info.log_limit); |
945 | |
946 | info.nterms = s->num_terms; |
947 | |
948 | if (allocate_buffers2(s, s->num_terms)) |
949 | return; |
950 | |
951 | memcpy(info.dps, s->decorr_passes, sizeof(info.dps)); |
952 | memcpy(s->sampleptrs[0][0], samples, s->block_samples * 4); |
953 | |
954 | for (i = 0; i < info.nterms && info.dps[i].value; i++) |
955 | decorr_mono(s->sampleptrs[i][0], s->sampleptrs[i + 1][0], |
956 | s->block_samples, info.dps + i, 1); |
957 | |
958 | info.best_bits = log2mono(s->sampleptrs[info.nterms][0], s->block_samples, 0) * 1; |
959 | memcpy(s->sampleptrs[info.nterms + 1][0], s->sampleptrs[i][0], s->block_samples * 4); |
960 | |
961 | if (s->extra_flags & EXTRA_BRANCHES) |
962 | recurse_mono(s, &info, 0, (int) floor(s->delta_decay + 0.5), |
963 | log2mono(s->sampleptrs[0][0], s->block_samples, 0)); |
964 | |
965 | if (s->extra_flags & EXTRA_SORT_FIRST) |
966 | sort_mono(s, &info); |
967 | |
968 | if (s->extra_flags & EXTRA_TRY_DELTAS) { |
969 | delta_mono(s, &info); |
970 | |
971 | if ((s->extra_flags & EXTRA_ADJUST_DELTAS) && s->decorr_passes[0].value) |
972 | s->delta_decay = (float)((s->delta_decay * 2.0 + s->decorr_passes[0].delta) / 3.0); |
973 | else |
974 | s->delta_decay = 2.0; |
975 | } |
976 | |
977 | if (s->extra_flags & EXTRA_SORT_LAST) |
978 | sort_mono(s, &info); |
979 | |
980 | if (do_samples) |
981 | memcpy(samples, s->sampleptrs[info.nterms + 1][0], s->block_samples * 4); |
982 | |
983 | for (i = 0; i < info.nterms; i++) |
984 | if (!s->decorr_passes[i].value) |
985 | break; |
986 | |
987 | s->num_terms = i; |
988 | } |
989 | |
990 | static void scan_word(WavPackEncodeContext *s, WvChannel *c, |
991 | int32_t *samples, int nb_samples, int dir) |
992 | { |
993 | if (dir < 0) |
994 | samples += nb_samples - 1; |
995 | |
996 | while (nb_samples--) { |
997 | uint32_t low, value = labs(samples[0]); |
998 | |
999 | if (value < GET_MED(0)) { |
1000 | DEC_MED(0); |
1001 | } else { |
1002 | low = GET_MED(0); |
1003 | INC_MED(0); |
1004 | |
1005 | if (value - low < GET_MED(1)) { |
1006 | DEC_MED(1); |
1007 | } else { |
1008 | low += GET_MED(1); |
1009 | INC_MED(1); |
1010 | |
1011 | if (value - low < GET_MED(2)) { |
1012 | DEC_MED(2); |
1013 | } else { |
1014 | INC_MED(2); |
1015 | } |
1016 | } |
1017 | } |
1018 | samples += dir; |
1019 | } |
1020 | } |
1021 | |
1022 | static int wv_mono(WavPackEncodeContext *s, int32_t *samples, |
1023 | int no_history, int do_samples) |
1024 | { |
1025 | struct Decorr temp_decorr_pass, save_decorr_passes[MAX_TERMS] = {{0}}; |
1026 | int nb_samples = s->block_samples; |
1027 | int buf_size = sizeof(int32_t) * nb_samples; |
1028 | uint32_t best_size = UINT32_MAX, size; |
1029 | int log_limit, pi, i, ret; |
1030 | |
1031 | for (i = 0; i < nb_samples; i++) |
1032 | if (samples[i]) |
1033 | break; |
1034 | |
1035 | if (i == nb_samples) { |
1036 | CLEAR(s->decorr_passes); |
1037 | CLEAR(s->w); |
1038 | s->num_terms = 0; |
1039 | return 0; |
1040 | } |
1041 | |
1042 | log_limit = (((s->flags & MAG_MASK) >> MAG_LSB) + 4) * 256; |
1043 | log_limit = FFMIN(6912, log_limit); |
1044 | |
1045 | if ((ret = allocate_buffers(s)) < 0) |
1046 | return ret; |
1047 | |
1048 | if (no_history || s->num_passes >= 7) |
1049 | s->best_decorr = s->mask_decorr = 0; |
1050 | |
1051 | for (pi = 0; pi < s->num_passes;) { |
1052 | const WavPackDecorrSpec *wpds; |
1053 | int nterms, c, j; |
1054 | |
1055 | if (!pi) { |
1056 | c = s->best_decorr; |
1057 | } else { |
1058 | if (s->mask_decorr == 0) |
1059 | c = 0; |
1060 | else |
1061 | c = (s->best_decorr & (s->mask_decorr - 1)) | s->mask_decorr; |
1062 | |
1063 | if (c == s->best_decorr) { |
1064 | s->mask_decorr = s->mask_decorr ? ((s->mask_decorr << 1) & (s->num_decorrs - 1)) : 1; |
1065 | continue; |
1066 | } |
1067 | } |
1068 | |
1069 | wpds = &s->decorr_specs[c]; |
1070 | nterms = decorr_filter_nterms[s->decorr_filter]; |
1071 | |
1072 | while (1) { |
1073 | memcpy(s->temp_buffer[0][0], samples, buf_size); |
1074 | CLEAR(save_decorr_passes); |
1075 | |
1076 | for (j = 0; j < nterms; j++) { |
1077 | CLEAR(temp_decorr_pass); |
1078 | temp_decorr_pass.delta = wpds->delta; |
1079 | temp_decorr_pass.value = wpds->terms[j]; |
1080 | |
1081 | if (temp_decorr_pass.value < 0) |
1082 | temp_decorr_pass.value = 1; |
1083 | |
1084 | decorr_mono(s->temp_buffer[j&1][0], s->temp_buffer[~j&1][0], |
1085 | FFMIN(nb_samples, 2048), &temp_decorr_pass, -1); |
1086 | |
1087 | if (j) { |
1088 | CLEAR(temp_decorr_pass.samplesA); |
1089 | } else { |
1090 | reverse_mono_decorr(&temp_decorr_pass); |
1091 | } |
1092 | |
1093 | memcpy(save_decorr_passes + j, &temp_decorr_pass, sizeof(struct Decorr)); |
1094 | decorr_mono(s->temp_buffer[j&1][0], s->temp_buffer[~j&1][0], |
1095 | nb_samples, &temp_decorr_pass, 1); |
1096 | } |
1097 | |
1098 | size = log2mono(s->temp_buffer[j&1][0], nb_samples, log_limit); |
1099 | if (size != UINT32_MAX || !nterms) |
1100 | break; |
1101 | nterms >>= 1; |
1102 | } |
1103 | |
1104 | if (size < best_size) { |
1105 | memcpy(s->best_buffer[0], s->temp_buffer[j&1][0], buf_size); |
1106 | memcpy(s->decorr_passes, save_decorr_passes, sizeof(struct Decorr) * MAX_TERMS); |
1107 | s->num_terms = nterms; |
1108 | s->best_decorr = c; |
1109 | best_size = size; |
1110 | } |
1111 | |
1112 | if (pi++) |
1113 | s->mask_decorr = s->mask_decorr ? ((s->mask_decorr << 1) & (s->num_decorrs - 1)) : 1; |
1114 | } |
1115 | |
1116 | if (s->extra_flags) |
1117 | analyze_mono(s, samples, do_samples); |
1118 | else if (do_samples) |
1119 | memcpy(samples, s->best_buffer[0], buf_size); |
1120 | |
1121 | if (no_history || s->extra_flags) { |
1122 | CLEAR(s->w); |
1123 | scan_word(s, &s->w.c[0], s->best_buffer[0], nb_samples, -1); |
1124 | } |
1125 | return 0; |
1126 | } |
1127 | |
1128 | static void decorr_stereo(int32_t *in_left, int32_t *in_right, |
1129 | int32_t *out_left, int32_t *out_right, |
1130 | int nb_samples, struct Decorr *dpp, int dir) |
1131 | { |
1132 | int m = 0, i; |
1133 | |
1134 | dpp->sumA = dpp->sumB = 0; |
1135 | |
1136 | if (dir < 0) { |
1137 | out_left += nb_samples - 1; |
1138 | out_right += nb_samples - 1; |
1139 | in_left += nb_samples - 1; |
1140 | in_right += nb_samples - 1; |
1141 | } |
1142 | |
1143 | dpp->weightA = restore_weight(store_weight(dpp->weightA)); |
1144 | dpp->weightB = restore_weight(store_weight(dpp->weightB)); |
1145 | |
1146 | for (i = 0; i < MAX_TERM; i++) { |
1147 | dpp->samplesA[i] = wp_exp2(log2s(dpp->samplesA[i])); |
1148 | dpp->samplesB[i] = wp_exp2(log2s(dpp->samplesB[i])); |
1149 | } |
1150 | |
1151 | switch (dpp->value) { |
1152 | case 2: |
1153 | while (nb_samples--) { |
1154 | int32_t sam, tmp; |
1155 | |
1156 | sam = dpp->samplesA[0]; |
1157 | dpp->samplesA[0] = dpp->samplesA[1]; |
1158 | out_left[0] = tmp = (dpp->samplesA[1] = in_left[0]) - APPLY_WEIGHT(dpp->weightA, sam); |
1159 | UPDATE_WEIGHT(dpp->weightA, dpp->delta, sam, tmp); |
1160 | dpp->sumA += dpp->weightA; |
1161 | |
1162 | sam = dpp->samplesB[0]; |
1163 | dpp->samplesB[0] = dpp->samplesB[1]; |
1164 | out_right[0] = tmp = (dpp->samplesB[1] = in_right[0]) - APPLY_WEIGHT(dpp->weightB, sam); |
1165 | UPDATE_WEIGHT(dpp->weightB, dpp->delta, sam, tmp); |
1166 | dpp->sumB += dpp->weightB; |
1167 | |
1168 | in_left += dir; |
1169 | out_left += dir; |
1170 | in_right += dir; |
1171 | out_right += dir; |
1172 | } |
1173 | break; |
1174 | case 17: |
1175 | while (nb_samples--) { |
1176 | int32_t sam, tmp; |
1177 | |
1178 | sam = 2 * dpp->samplesA[0] - dpp->samplesA[1]; |
1179 | dpp->samplesA[1] = dpp->samplesA[0]; |
1180 | out_left[0] = tmp = (dpp->samplesA[0] = in_left[0]) - APPLY_WEIGHT(dpp->weightA, sam); |
1181 | UPDATE_WEIGHT(dpp->weightA, dpp->delta, sam, tmp); |
1182 | dpp->sumA += dpp->weightA; |
1183 | |
1184 | sam = 2 * dpp->samplesB[0] - dpp->samplesB[1]; |
1185 | dpp->samplesB[1] = dpp->samplesB[0]; |
1186 | out_right[0] = tmp = (dpp->samplesB[0] = in_right[0]) - APPLY_WEIGHT (dpp->weightB, sam); |
1187 | UPDATE_WEIGHT(dpp->weightB, dpp->delta, sam, tmp); |
1188 | dpp->sumB += dpp->weightB; |
1189 | |
1190 | in_left += dir; |
1191 | out_left += dir; |
1192 | in_right += dir; |
1193 | out_right += dir; |
1194 | } |
1195 | break; |
1196 | case 18: |
1197 | while (nb_samples--) { |
1198 | int32_t sam, tmp; |
1199 | |
1200 | sam = dpp->samplesA[0] + ((dpp->samplesA[0] - dpp->samplesA[1]) >> 1); |
1201 | dpp->samplesA[1] = dpp->samplesA[0]; |
1202 | out_left[0] = tmp = (dpp->samplesA[0] = in_left[0]) - APPLY_WEIGHT(dpp->weightA, sam); |
1203 | UPDATE_WEIGHT(dpp->weightA, dpp->delta, sam, tmp); |
1204 | dpp->sumA += dpp->weightA; |
1205 | |
1206 | sam = dpp->samplesB[0] + ((dpp->samplesB[0] - dpp->samplesB[1]) >> 1); |
1207 | dpp->samplesB[1] = dpp->samplesB[0]; |
1208 | out_right[0] = tmp = (dpp->samplesB[0] = in_right[0]) - APPLY_WEIGHT(dpp->weightB, sam); |
1209 | UPDATE_WEIGHT(dpp->weightB, dpp->delta, sam, tmp); |
1210 | dpp->sumB += dpp->weightB; |
1211 | |
1212 | in_left += dir; |
1213 | out_left += dir; |
1214 | in_right += dir; |
1215 | out_right += dir; |
1216 | } |
1217 | break; |
1218 | default: { |
1219 | int k = dpp->value & (MAX_TERM - 1); |
1220 | |
1221 | while (nb_samples--) { |
1222 | int32_t sam, tmp; |
1223 | |
1224 | sam = dpp->samplesA[m]; |
1225 | out_left[0] = tmp = (dpp->samplesA[k] = in_left[0]) - APPLY_WEIGHT(dpp->weightA, sam); |
1226 | UPDATE_WEIGHT(dpp->weightA, dpp->delta, sam, tmp); |
1227 | dpp->sumA += dpp->weightA; |
1228 | |
1229 | sam = dpp->samplesB[m]; |
1230 | out_right[0] = tmp = (dpp->samplesB[k] = in_right[0]) - APPLY_WEIGHT(dpp->weightB, sam); |
1231 | UPDATE_WEIGHT(dpp->weightB, dpp->delta, sam, tmp); |
1232 | dpp->sumB += dpp->weightB; |
1233 | |
1234 | in_left += dir; |
1235 | out_left += dir; |
1236 | in_right += dir; |
1237 | out_right += dir; |
1238 | m = (m + 1) & (MAX_TERM - 1); |
1239 | k = (k + 1) & (MAX_TERM - 1); |
1240 | } |
1241 | |
1242 | if (m) { |
1243 | int32_t temp_A[MAX_TERM], temp_B[MAX_TERM]; |
1244 | int k; |
1245 | |
1246 | memcpy(temp_A, dpp->samplesA, sizeof(dpp->samplesA)); |
1247 | memcpy(temp_B, dpp->samplesB, sizeof(dpp->samplesB)); |
1248 | |
1249 | for (k = 0; k < MAX_TERM; k++) { |
1250 | dpp->samplesA[k] = temp_A[m]; |
1251 | dpp->samplesB[k] = temp_B[m]; |
1252 | m = (m + 1) & (MAX_TERM - 1); |
1253 | } |
1254 | } |
1255 | break; |
1256 | } |
1257 | case -1: |
1258 | while (nb_samples--) { |
1259 | int32_t sam_A, sam_B, tmp; |
1260 | |
1261 | sam_A = dpp->samplesA[0]; |
1262 | out_left[0] = tmp = (sam_B = in_left[0]) - APPLY_WEIGHT(dpp->weightA, sam_A); |
1263 | UPDATE_WEIGHT_CLIP(dpp->weightA, dpp->delta, sam_A, tmp); |
1264 | dpp->sumA += dpp->weightA; |
1265 | |
1266 | out_right[0] = tmp = (dpp->samplesA[0] = in_right[0]) - APPLY_WEIGHT(dpp->weightB, sam_B); |
1267 | UPDATE_WEIGHT_CLIP(dpp->weightB, dpp->delta, sam_B, tmp); |
1268 | dpp->sumB += dpp->weightB; |
1269 | |
1270 | in_left += dir; |
1271 | out_left += dir; |
1272 | in_right += dir; |
1273 | out_right += dir; |
1274 | } |
1275 | break; |
1276 | case -2: |
1277 | while (nb_samples--) { |
1278 | int32_t sam_A, sam_B, tmp; |
1279 | |
1280 | sam_B = dpp->samplesB[0]; |
1281 | out_right[0] = tmp = (sam_A = in_right[0]) - APPLY_WEIGHT(dpp->weightB, sam_B); |
1282 | UPDATE_WEIGHT_CLIP(dpp->weightB, dpp->delta, sam_B, tmp); |
1283 | dpp->sumB += dpp->weightB; |
1284 | |
1285 | out_left[0] = tmp = (dpp->samplesB[0] = in_left[0]) - APPLY_WEIGHT(dpp->weightA, sam_A); |
1286 | UPDATE_WEIGHT_CLIP(dpp->weightA, dpp->delta, sam_A, tmp); |
1287 | dpp->sumA += dpp->weightA; |
1288 | |
1289 | in_left += dir; |
1290 | out_left += dir; |
1291 | in_right += dir; |
1292 | out_right += dir; |
1293 | } |
1294 | break; |
1295 | case -3: |
1296 | while (nb_samples--) { |
1297 | int32_t sam_A, sam_B, tmp; |
1298 | |
1299 | sam_A = dpp->samplesA[0]; |
1300 | sam_B = dpp->samplesB[0]; |
1301 | |
1302 | dpp->samplesA[0] = tmp = in_right[0]; |
1303 | out_right[0] = tmp -= APPLY_WEIGHT(dpp->weightB, sam_B); |
1304 | UPDATE_WEIGHT_CLIP(dpp->weightB, dpp->delta, sam_B, tmp); |
1305 | dpp->sumB += dpp->weightB; |
1306 | |
1307 | dpp->samplesB[0] = tmp = in_left[0]; |
1308 | out_left[0] = tmp -= APPLY_WEIGHT(dpp->weightA, sam_A); |
1309 | UPDATE_WEIGHT_CLIP(dpp->weightA, dpp->delta, sam_A, tmp); |
1310 | dpp->sumA += dpp->weightA; |
1311 | |
1312 | in_left += dir; |
1313 | out_left += dir; |
1314 | in_right += dir; |
1315 | out_right += dir; |
1316 | } |
1317 | break; |
1318 | } |
1319 | } |
1320 | |
1321 | static void reverse_decorr(struct Decorr *dpp) |
1322 | { |
1323 | if (dpp->value > MAX_TERM) { |
1324 | int32_t sam_A, sam_B; |
1325 | |
1326 | if (dpp->value & 1) { |
1327 | sam_A = 2 * dpp->samplesA[0] - dpp->samplesA[1]; |
1328 | sam_B = 2 * dpp->samplesB[0] - dpp->samplesB[1]; |
1329 | } else { |
1330 | sam_A = (3 * dpp->samplesA[0] - dpp->samplesA[1]) >> 1; |
1331 | sam_B = (3 * dpp->samplesB[0] - dpp->samplesB[1]) >> 1; |
1332 | } |
1333 | |
1334 | dpp->samplesA[1] = dpp->samplesA[0]; |
1335 | dpp->samplesB[1] = dpp->samplesB[0]; |
1336 | dpp->samplesA[0] = sam_A; |
1337 | dpp->samplesB[0] = sam_B; |
1338 | |
1339 | if (dpp->value & 1) { |
1340 | sam_A = 2 * dpp->samplesA[0] - dpp->samplesA[1]; |
1341 | sam_B = 2 * dpp->samplesB[0] - dpp->samplesB[1]; |
1342 | } else { |
1343 | sam_A = (3 * dpp->samplesA[0] - dpp->samplesA[1]) >> 1; |
1344 | sam_B = (3 * dpp->samplesB[0] - dpp->samplesB[1]) >> 1; |
1345 | } |
1346 | |
1347 | dpp->samplesA[1] = sam_A; |
1348 | dpp->samplesB[1] = sam_B; |
1349 | } else if (dpp->value > 1) { |
1350 | int i, j, k; |
1351 | |
1352 | for (i = 0, j = dpp->value - 1, k = 0; k < dpp->value / 2; i++, j--, k++) { |
1353 | i &= (MAX_TERM - 1); |
1354 | j &= (MAX_TERM - 1); |
1355 | dpp->samplesA[i] ^= dpp->samplesA[j]; |
1356 | dpp->samplesA[j] ^= dpp->samplesA[i]; |
1357 | dpp->samplesA[i] ^= dpp->samplesA[j]; |
1358 | dpp->samplesB[i] ^= dpp->samplesB[j]; |
1359 | dpp->samplesB[j] ^= dpp->samplesB[i]; |
1360 | dpp->samplesB[i] ^= dpp->samplesB[j]; |
1361 | } |
1362 | } |
1363 | } |
1364 | |
1365 | static void decorr_stereo_quick(int32_t *in_left, int32_t *in_right, |
1366 | int32_t *out_left, int32_t *out_right, |
1367 | int nb_samples, struct Decorr *dpp) |
1368 | { |
1369 | int m = 0, i; |
1370 | |
1371 | dpp->weightA = restore_weight(store_weight(dpp->weightA)); |
1372 | dpp->weightB = restore_weight(store_weight(dpp->weightB)); |
1373 | |
1374 | for (i = 0; i < MAX_TERM; i++) { |
1375 | dpp->samplesA[i] = wp_exp2(log2s(dpp->samplesA[i])); |
1376 | dpp->samplesB[i] = wp_exp2(log2s(dpp->samplesB[i])); |
1377 | } |
1378 | |
1379 | switch (dpp->value) { |
1380 | case 2: |
1381 | for (i = 0; i < nb_samples; i++) { |
1382 | int32_t sam, tmp; |
1383 | |
1384 | sam = dpp->samplesA[0]; |
1385 | dpp->samplesA[0] = dpp->samplesA[1]; |
1386 | out_left[i] = tmp = (dpp->samplesA[1] = in_left[i]) - APPLY_WEIGHT_I(dpp->weightA, sam); |
1387 | UPDATE_WEIGHT(dpp->weightA, dpp->delta, sam, tmp); |
1388 | |
1389 | sam = dpp->samplesB[0]; |
1390 | dpp->samplesB[0] = dpp->samplesB[1]; |
1391 | out_right[i] = tmp = (dpp->samplesB[1] = in_right[i]) - APPLY_WEIGHT_I(dpp->weightB, sam); |
1392 | UPDATE_WEIGHT(dpp->weightB, dpp->delta, sam, tmp); |
1393 | } |
1394 | break; |
1395 | case 17: |
1396 | for (i = 0; i < nb_samples; i++) { |
1397 | int32_t sam, tmp; |
1398 | |
1399 | sam = 2 * dpp->samplesA[0] - dpp->samplesA[1]; |
1400 | dpp->samplesA[1] = dpp->samplesA[0]; |
1401 | out_left[i] = tmp = (dpp->samplesA[0] = in_left[i]) - APPLY_WEIGHT_I(dpp->weightA, sam); |
1402 | UPDATE_WEIGHT(dpp->weightA, dpp->delta, sam, tmp); |
1403 | |
1404 | sam = 2 * dpp->samplesB[0] - dpp->samplesB[1]; |
1405 | dpp->samplesB[1] = dpp->samplesB[0]; |
1406 | out_right[i] = tmp = (dpp->samplesB[0] = in_right[i]) - APPLY_WEIGHT_I(dpp->weightB, sam); |
1407 | UPDATE_WEIGHT(dpp->weightB, dpp->delta, sam, tmp); |
1408 | } |
1409 | break; |
1410 | case 18: |
1411 | for (i = 0; i < nb_samples; i++) { |
1412 | int32_t sam, tmp; |
1413 | |
1414 | sam = dpp->samplesA[0] + ((dpp->samplesA[0] - dpp->samplesA[1]) >> 1); |
1415 | dpp->samplesA[1] = dpp->samplesA[0]; |
1416 | out_left[i] = tmp = (dpp->samplesA[0] = in_left[i]) - APPLY_WEIGHT_I(dpp->weightA, sam); |
1417 | UPDATE_WEIGHT(dpp->weightA, dpp->delta, sam, tmp); |
1418 | |
1419 | sam = dpp->samplesB[0] + ((dpp->samplesB[0] - dpp->samplesB[1]) >> 1); |
1420 | dpp->samplesB[1] = dpp->samplesB[0]; |
1421 | out_right[i] = tmp = (dpp->samplesB[0] = in_right[i]) - APPLY_WEIGHT_I(dpp->weightB, sam); |
1422 | UPDATE_WEIGHT(dpp->weightB, dpp->delta, sam, tmp); |
1423 | } |
1424 | break; |
1425 | default: { |
1426 | int k = dpp->value & (MAX_TERM - 1); |
1427 | |
1428 | for (i = 0; i < nb_samples; i++) { |
1429 | int32_t sam, tmp; |
1430 | |
1431 | sam = dpp->samplesA[m]; |
1432 | out_left[i] = tmp = (dpp->samplesA[k] = in_left[i]) - APPLY_WEIGHT_I(dpp->weightA, sam); |
1433 | UPDATE_WEIGHT(dpp->weightA, dpp->delta, sam, tmp); |
1434 | |
1435 | sam = dpp->samplesB[m]; |
1436 | out_right[i] = tmp = (dpp->samplesB[k] = in_right[i]) - APPLY_WEIGHT_I(dpp->weightB, sam); |
1437 | UPDATE_WEIGHT(dpp->weightB, dpp->delta, sam, tmp); |
1438 | |
1439 | m = (m + 1) & (MAX_TERM - 1); |
1440 | k = (k + 1) & (MAX_TERM - 1); |
1441 | } |
1442 | |
1443 | if (m) { |
1444 | int32_t temp_A[MAX_TERM], temp_B[MAX_TERM]; |
1445 | int k; |
1446 | |
1447 | memcpy(temp_A, dpp->samplesA, sizeof(dpp->samplesA)); |
1448 | memcpy(temp_B, dpp->samplesB, sizeof(dpp->samplesB)); |
1449 | |
1450 | for (k = 0; k < MAX_TERM; k++) { |
1451 | dpp->samplesA[k] = temp_A[m]; |
1452 | dpp->samplesB[k] = temp_B[m]; |
1453 | m = (m + 1) & (MAX_TERM - 1); |
1454 | } |
1455 | } |
1456 | break; |
1457 | } |
1458 | case -1: |
1459 | for (i = 0; i < nb_samples; i++) { |
1460 | int32_t sam_A, sam_B, tmp; |
1461 | |
1462 | sam_A = dpp->samplesA[0]; |
1463 | out_left[i] = tmp = (sam_B = in_left[i]) - APPLY_WEIGHT_I(dpp->weightA, sam_A); |
1464 | UPDATE_WEIGHT_CLIP(dpp->weightA, dpp->delta, sam_A, tmp); |
1465 | |
1466 | out_right[i] = tmp = (dpp->samplesA[0] = in_right[i]) - APPLY_WEIGHT_I(dpp->weightB, sam_B); |
1467 | UPDATE_WEIGHT_CLIP(dpp->weightB, dpp->delta, sam_B, tmp); |
1468 | } |
1469 | break; |
1470 | case -2: |
1471 | for (i = 0; i < nb_samples; i++) { |
1472 | int32_t sam_A, sam_B, tmp; |
1473 | |
1474 | sam_B = dpp->samplesB[0]; |
1475 | out_right[i] = tmp = (sam_A = in_right[i]) - APPLY_WEIGHT_I(dpp->weightB, sam_B); |
1476 | UPDATE_WEIGHT_CLIP(dpp->weightB, dpp->delta, sam_B, tmp); |
1477 | |
1478 | out_left[i] = tmp = (dpp->samplesB[0] = in_left[i]) - APPLY_WEIGHT_I(dpp->weightA, sam_A); |
1479 | UPDATE_WEIGHT_CLIP(dpp->weightA, dpp->delta, sam_A, tmp); |
1480 | } |
1481 | break; |
1482 | case -3: |
1483 | for (i = 0; i < nb_samples; i++) { |
1484 | int32_t sam_A, sam_B, tmp; |
1485 | |
1486 | sam_A = dpp->samplesA[0]; |
1487 | sam_B = dpp->samplesB[0]; |
1488 | |
1489 | dpp->samplesA[0] = tmp = in_right[i]; |
1490 | out_right[i] = tmp -= APPLY_WEIGHT_I(dpp->weightB, sam_B); |
1491 | UPDATE_WEIGHT_CLIP(dpp->weightB, dpp->delta, sam_B, tmp); |
1492 | |
1493 | dpp->samplesB[0] = tmp = in_left[i]; |
1494 | out_left[i] = tmp -= APPLY_WEIGHT_I(dpp->weightA, sam_A); |
1495 | UPDATE_WEIGHT_CLIP(dpp->weightA, dpp->delta, sam_A, tmp); |
1496 | } |
1497 | break; |
1498 | } |
1499 | } |
1500 | |
1501 | static void decorr_stereo_buffer(WavPackExtraInfo *info, |
1502 | int32_t *in_left, int32_t *in_right, |
1503 | int32_t *out_left, int32_t *out_right, |
1504 | int nb_samples, int tindex) |
1505 | { |
1506 | struct Decorr dp = {0}, *dppi = info->dps + tindex; |
1507 | int delta = dppi->delta, pre_delta; |
1508 | int term = dppi->value; |
1509 | |
1510 | if (delta == 7) |
1511 | pre_delta = 7; |
1512 | else if (delta < 2) |
1513 | pre_delta = 3; |
1514 | else |
1515 | pre_delta = delta + 1; |
1516 | |
1517 | dp.value = term; |
1518 | dp.delta = pre_delta; |
1519 | decorr_stereo(in_left, in_right, out_left, out_right, |
1520 | FFMIN(2048, nb_samples), &dp, -1); |
1521 | dp.delta = delta; |
1522 | |
1523 | if (tindex == 0) { |
1524 | reverse_decorr(&dp); |
1525 | } else { |
1526 | CLEAR(dp.samplesA); |
1527 | CLEAR(dp.samplesB); |
1528 | } |
1529 | |
1530 | memcpy(dppi->samplesA, dp.samplesA, sizeof(dp.samplesA)); |
1531 | memcpy(dppi->samplesB, dp.samplesB, sizeof(dp.samplesB)); |
1532 | dppi->weightA = dp.weightA; |
1533 | dppi->weightB = dp.weightB; |
1534 | |
1535 | if (delta == 0) { |
1536 | dp.delta = 1; |
1537 | decorr_stereo(in_left, in_right, out_left, out_right, nb_samples, &dp, 1); |
1538 | dp.delta = 0; |
1539 | memcpy(dp.samplesA, dppi->samplesA, sizeof(dp.samplesA)); |
1540 | memcpy(dp.samplesB, dppi->samplesB, sizeof(dp.samplesB)); |
1541 | dppi->weightA = dp.weightA = dp.sumA / nb_samples; |
1542 | dppi->weightB = dp.weightB = dp.sumB / nb_samples; |
1543 | } |
1544 | |
1545 | if (info->gt16bit) |
1546 | decorr_stereo(in_left, in_right, out_left, out_right, |
1547 | nb_samples, &dp, 1); |
1548 | else |
1549 | decorr_stereo_quick(in_left, in_right, out_left, out_right, |
1550 | nb_samples, &dp); |
1551 | } |
1552 | |
1553 | static void sort_stereo(WavPackEncodeContext *s, WavPackExtraInfo *info) |
1554 | { |
1555 | int reversed = 1; |
1556 | uint32_t bits; |
1557 | |
1558 | while (reversed) { |
1559 | int ri, i; |
1560 | |
1561 | memcpy(info->dps, s->decorr_passes, sizeof(s->decorr_passes)); |
1562 | reversed = 0; |
1563 | |
1564 | for (ri = 0; ri < info->nterms && s->decorr_passes[ri].value; ri++) { |
1565 | |
1566 | if (ri + 1 >= info->nterms || !s->decorr_passes[ri+1].value) |
1567 | break; |
1568 | |
1569 | if (s->decorr_passes[ri].value == s->decorr_passes[ri+1].value) { |
1570 | decorr_stereo_buffer(info, |
1571 | s->sampleptrs[ri ][0], s->sampleptrs[ri ][1], |
1572 | s->sampleptrs[ri+1][0], s->sampleptrs[ri+1][1], |
1573 | s->block_samples, ri); |
1574 | continue; |
1575 | } |
1576 | |
1577 | info->dps[ri ] = s->decorr_passes[ri+1]; |
1578 | info->dps[ri+1] = s->decorr_passes[ri ]; |
1579 | |
1580 | for (i = ri; i < info->nterms && s->decorr_passes[i].value; i++) |
1581 | decorr_stereo_buffer(info, |
1582 | s->sampleptrs[i ][0], s->sampleptrs[i ][1], |
1583 | s->sampleptrs[i+1][0], s->sampleptrs[i+1][1], |
1584 | s->block_samples, i); |
1585 | |
1586 | bits = log2stereo(s->sampleptrs[i][0], s->sampleptrs[i][1], |
1587 | s->block_samples, info->log_limit); |
1588 | |
1589 | if (bits < info->best_bits) { |
1590 | reversed = 1; |
1591 | info->best_bits = bits; |
1592 | CLEAR(s->decorr_passes); |
1593 | memcpy(s->decorr_passes, info->dps, sizeof(info->dps[0]) * i); |
1594 | memcpy(s->sampleptrs[info->nterms + 1][0], |
1595 | s->sampleptrs[i][0], s->block_samples * 4); |
1596 | memcpy(s->sampleptrs[info->nterms + 1][1], |
1597 | s->sampleptrs[i][1], s->block_samples * 4); |
1598 | } else { |
1599 | info->dps[ri ] = s->decorr_passes[ri ]; |
1600 | info->dps[ri+1] = s->decorr_passes[ri+1]; |
1601 | decorr_stereo_buffer(info, |
1602 | s->sampleptrs[ri ][0], s->sampleptrs[ri ][1], |
1603 | s->sampleptrs[ri+1][0], s->sampleptrs[ri+1][1], |
1604 | s->block_samples, ri); |
1605 | } |
1606 | } |
1607 | } |
1608 | } |
1609 | |
1610 | static void delta_stereo(WavPackEncodeContext *s, WavPackExtraInfo *info) |
1611 | { |
1612 | int lower = 0, delta, d, i; |
1613 | uint32_t bits; |
1614 | |
1615 | if (!s->decorr_passes[0].value) |
1616 | return; |
1617 | delta = s->decorr_passes[0].delta; |
1618 | |
1619 | for (d = delta - 1; d >= 0; d--) { |
1620 | for (i = 0; i < info->nterms && s->decorr_passes[i].value; i++) { |
1621 | info->dps[i].value = s->decorr_passes[i].value; |
1622 | info->dps[i].delta = d; |
1623 | decorr_stereo_buffer(info, |
1624 | s->sampleptrs[i ][0], s->sampleptrs[i ][1], |
1625 | s->sampleptrs[i+1][0], s->sampleptrs[i+1][1], |
1626 | s->block_samples, i); |
1627 | } |
1628 | |
1629 | bits = log2stereo(s->sampleptrs[i][0], s->sampleptrs[i][1], |
1630 | s->block_samples, info->log_limit); |
1631 | if (bits >= info->best_bits) |
1632 | break; |
1633 | lower = 1; |
1634 | info->best_bits = bits; |
1635 | CLEAR(s->decorr_passes); |
1636 | memcpy(s->decorr_passes, info->dps, sizeof(info->dps[0]) * i); |
1637 | memcpy(s->sampleptrs[info->nterms + 1][0], s->sampleptrs[i][0], |
1638 | s->block_samples * 4); |
1639 | memcpy(s->sampleptrs[info->nterms + 1][1], s->sampleptrs[i][1], |
1640 | s->block_samples * 4); |
1641 | } |
1642 | |
1643 | for (d = delta + 1; !lower && d <= 7; d++) { |
1644 | for (i = 0; i < info->nterms && s->decorr_passes[i].value; i++) { |
1645 | info->dps[i].value = s->decorr_passes[i].value; |
1646 | info->dps[i].delta = d; |
1647 | decorr_stereo_buffer(info, |
1648 | s->sampleptrs[i ][0], s->sampleptrs[i ][1], |
1649 | s->sampleptrs[i+1][0], s->sampleptrs[i+1][1], |
1650 | s->block_samples, i); |
1651 | } |
1652 | |
1653 | bits = log2stereo(s->sampleptrs[i][0], s->sampleptrs[i][1], |
1654 | s->block_samples, info->log_limit); |
1655 | |
1656 | if (bits < info->best_bits) { |
1657 | info->best_bits = bits; |
1658 | CLEAR(s->decorr_passes); |
1659 | memcpy(s->decorr_passes, info->dps, sizeof(info->dps[0]) * i); |
1660 | memcpy(s->sampleptrs[info->nterms + 1][0], |
1661 | s->sampleptrs[i][0], s->block_samples * 4); |
1662 | memcpy(s->sampleptrs[info->nterms + 1][1], |
1663 | s->sampleptrs[i][1], s->block_samples * 4); |
1664 | } |
1665 | else |
1666 | break; |
1667 | } |
1668 | } |
1669 | |
1670 | static void recurse_stereo(WavPackEncodeContext *s, WavPackExtraInfo *info, |
1671 | int depth, int delta, uint32_t input_bits) |
1672 | { |
1673 | int term, branches = s->num_branches - depth; |
1674 | int32_t *in_left, *in_right, *out_left, *out_right; |
1675 | uint32_t term_bits[22], bits; |
1676 | |
1677 | if (branches < 1 || depth + 1 == info->nterms) |
1678 | branches = 1; |
1679 | |
1680 | CLEAR(term_bits); |
1681 | in_left = s->sampleptrs[depth ][0]; |
1682 | in_right = s->sampleptrs[depth ][1]; |
1683 | out_left = s->sampleptrs[depth + 1][0]; |
1684 | out_right = s->sampleptrs[depth + 1][1]; |
1685 | |
1686 | for (term = -3; term <= 18; term++) { |
1687 | if (!term || (term > 8 && term < 17)) |
1688 | continue; |
1689 | |
1690 | if (term == 17 && branches == 1 && depth + 1 < info->nterms) |
1691 | continue; |
1692 | |
1693 | if (term == -1 || term == -2) |
1694 | if (!(s->flags & WV_CROSS_DECORR)) |
1695 | continue; |
1696 | |
1697 | if (!s->extra_flags && (term > 4 && term < 17)) |
1698 | continue; |
1699 | |
1700 | info->dps[depth].value = term; |
1701 | info->dps[depth].delta = delta; |
1702 | decorr_stereo_buffer(info, in_left, in_right, out_left, out_right, |
1703 | s->block_samples, depth); |
1704 | bits = log2stereo(out_left, out_right, s->block_samples, info->log_limit); |
1705 | |
1706 | if (bits < info->best_bits) { |
1707 | info->best_bits = bits; |
1708 | CLEAR(s->decorr_passes); |
1709 | memcpy(s->decorr_passes, info->dps, sizeof(info->dps[0]) * (depth + 1)); |
1710 | memcpy(s->sampleptrs[info->nterms + 1][0], s->sampleptrs[depth + 1][0], |
1711 | s->block_samples * 4); |
1712 | memcpy(s->sampleptrs[info->nterms + 1][1], s->sampleptrs[depth + 1][1], |
1713 | s->block_samples * 4); |
1714 | } |
1715 | |
1716 | term_bits[term + 3] = bits; |
1717 | } |
1718 | |
1719 | while (depth + 1 < info->nterms && branches--) { |
1720 | uint32_t local_best_bits = input_bits; |
1721 | int best_term = 0, i; |
1722 | |
1723 | for (i = 0; i < 22; i++) |
1724 | if (term_bits[i] && term_bits[i] < local_best_bits) { |
1725 | local_best_bits = term_bits[i]; |
1726 | best_term = i - 3; |
1727 | } |
1728 | |
1729 | if (!best_term) |
1730 | break; |
1731 | |
1732 | term_bits[best_term + 3] = 0; |
1733 | |
1734 | info->dps[depth].value = best_term; |
1735 | info->dps[depth].delta = delta; |
1736 | decorr_stereo_buffer(info, in_left, in_right, out_left, out_right, |
1737 | s->block_samples, depth); |
1738 | |
1739 | recurse_stereo(s, info, depth + 1, delta, local_best_bits); |
1740 | } |
1741 | } |
1742 | |
1743 | static void analyze_stereo(WavPackEncodeContext *s, |
1744 | int32_t *in_left, int32_t *in_right, |
1745 | int do_samples) |
1746 | { |
1747 | WavPackExtraInfo info; |
1748 | int i; |
1749 | |
1750 | info.gt16bit = ((s->flags & MAG_MASK) >> MAG_LSB) >= 16; |
1751 | |
1752 | info.log_limit = (((s->flags & MAG_MASK) >> MAG_LSB) + 4) * 256; |
1753 | info.log_limit = FFMIN(6912, info.log_limit); |
1754 | |
1755 | info.nterms = s->num_terms; |
1756 | |
1757 | if (allocate_buffers2(s, s->num_terms)) |
1758 | return; |
1759 | |
1760 | memcpy(info.dps, s->decorr_passes, sizeof(info.dps)); |
1761 | memcpy(s->sampleptrs[0][0], in_left, s->block_samples * 4); |
1762 | memcpy(s->sampleptrs[0][1], in_right, s->block_samples * 4); |
1763 | |
1764 | for (i = 0; i < info.nterms && info.dps[i].value; i++) |
1765 | if (info.gt16bit) |
1766 | decorr_stereo(s->sampleptrs[i ][0], s->sampleptrs[i ][1], |
1767 | s->sampleptrs[i + 1][0], s->sampleptrs[i + 1][1], |
1768 | s->block_samples, info.dps + i, 1); |
1769 | else |
1770 | decorr_stereo_quick(s->sampleptrs[i ][0], s->sampleptrs[i ][1], |
1771 | s->sampleptrs[i + 1][0], s->sampleptrs[i + 1][1], |
1772 | s->block_samples, info.dps + i); |
1773 | |
1774 | info.best_bits = log2stereo(s->sampleptrs[info.nterms][0], s->sampleptrs[info.nterms][1], |
1775 | s->block_samples, 0); |
1776 | |
1777 | memcpy(s->sampleptrs[info.nterms + 1][0], s->sampleptrs[i][0], s->block_samples * 4); |
1778 | memcpy(s->sampleptrs[info.nterms + 1][1], s->sampleptrs[i][1], s->block_samples * 4); |
1779 | |
1780 | if (s->extra_flags & EXTRA_BRANCHES) |
1781 | recurse_stereo(s, &info, 0, (int) floor(s->delta_decay + 0.5), |
1782 | log2stereo(s->sampleptrs[0][0], s->sampleptrs[0][1], |
1783 | s->block_samples, 0)); |
1784 | |
1785 | if (s->extra_flags & EXTRA_SORT_FIRST) |
1786 | sort_stereo(s, &info); |
1787 | |
1788 | if (s->extra_flags & EXTRA_TRY_DELTAS) { |
1789 | delta_stereo(s, &info); |
1790 | |
1791 | if ((s->extra_flags & EXTRA_ADJUST_DELTAS) && s->decorr_passes[0].value) |
1792 | s->delta_decay = (float)((s->delta_decay * 2.0 + s->decorr_passes[0].delta) / 3.0); |
1793 | else |
1794 | s->delta_decay = 2.0; |
1795 | } |
1796 | |
1797 | if (s->extra_flags & EXTRA_SORT_LAST) |
1798 | sort_stereo(s, &info); |
1799 | |
1800 | if (do_samples) { |
1801 | memcpy(in_left, s->sampleptrs[info.nterms + 1][0], s->block_samples * 4); |
1802 | memcpy(in_right, s->sampleptrs[info.nterms + 1][1], s->block_samples * 4); |
1803 | } |
1804 | |
1805 | for (i = 0; i < info.nterms; i++) |
1806 | if (!s->decorr_passes[i].value) |
1807 | break; |
1808 | |
1809 | s->num_terms = i; |
1810 | } |
1811 | |
1812 | static int wv_stereo(WavPackEncodeContext *s, |
1813 | int32_t *samples_l, int32_t *samples_r, |
1814 | int no_history, int do_samples) |
1815 | { |
1816 | struct Decorr temp_decorr_pass, save_decorr_passes[MAX_TERMS] = {{0}}; |
1817 | int nb_samples = s->block_samples, ret; |
1818 | int buf_size = sizeof(int32_t) * nb_samples; |
1819 | int log_limit, force_js = 0, force_ts = 0, got_js = 0, pi, i; |
1820 | uint32_t best_size = UINT32_MAX, size; |
1821 | |
1822 | for (i = 0; i < nb_samples; i++) |
1823 | if (samples_l[i] || samples_r[i]) |
1824 | break; |
1825 | |
1826 | if (i == nb_samples) { |
1827 | s->flags &= ~((uint32_t) WV_JOINT_STEREO); |
1828 | CLEAR(s->decorr_passes); |
1829 | CLEAR(s->w); |
1830 | s->num_terms = 0; |
1831 | return 0; |
1832 | } |
1833 | |
1834 | log_limit = (((s->flags & MAG_MASK) >> MAG_LSB) + 4) * 256; |
1835 | log_limit = FFMIN(6912, log_limit); |
1836 | |
1837 | if (s->joint != -1) { |
1838 | force_js = s->joint; |
1839 | force_ts = !s->joint; |
1840 | } |
1841 | |
1842 | if ((ret = allocate_buffers(s)) < 0) |
1843 | return ret; |
1844 | |
1845 | if (no_history || s->num_passes >= 7) |
1846 | s->best_decorr = s->mask_decorr = 0; |
1847 | |
1848 | for (pi = 0; pi < s->num_passes;) { |
1849 | const WavPackDecorrSpec *wpds; |
1850 | int nterms, c, j; |
1851 | |
1852 | if (!pi) |
1853 | c = s->best_decorr; |
1854 | else { |
1855 | if (s->mask_decorr == 0) |
1856 | c = 0; |
1857 | else |
1858 | c = (s->best_decorr & (s->mask_decorr - 1)) | s->mask_decorr; |
1859 | |
1860 | if (c == s->best_decorr) { |
1861 | s->mask_decorr = s->mask_decorr ? ((s->mask_decorr << 1) & (s->num_decorrs - 1)) : 1; |
1862 | continue; |
1863 | } |
1864 | } |
1865 | |
1866 | wpds = &s->decorr_specs[c]; |
1867 | nterms = decorr_filter_nterms[s->decorr_filter]; |
1868 | |
1869 | while (1) { |
1870 | if (force_js || (wpds->joint_stereo && !force_ts)) { |
1871 | if (!got_js) { |
1872 | av_fast_padded_malloc(&s->js_left, &s->js_left_size, buf_size); |
1873 | av_fast_padded_malloc(&s->js_right, &s->js_right_size, buf_size); |
1874 | memcpy(s->js_left, samples_l, buf_size); |
1875 | memcpy(s->js_right, samples_r, buf_size); |
1876 | |
1877 | for (i = 0; i < nb_samples; i++) |
1878 | s->js_right[i] += ((s->js_left[i] -= s->js_right[i]) >> 1); |
1879 | got_js = 1; |
1880 | } |
1881 | |
1882 | memcpy(s->temp_buffer[0][0], s->js_left, buf_size); |
1883 | memcpy(s->temp_buffer[0][1], s->js_right, buf_size); |
1884 | } else { |
1885 | memcpy(s->temp_buffer[0][0], samples_l, buf_size); |
1886 | memcpy(s->temp_buffer[0][1], samples_r, buf_size); |
1887 | } |
1888 | |
1889 | CLEAR(save_decorr_passes); |
1890 | |
1891 | for (j = 0; j < nterms; j++) { |
1892 | CLEAR(temp_decorr_pass); |
1893 | temp_decorr_pass.delta = wpds->delta; |
1894 | temp_decorr_pass.value = wpds->terms[j]; |
1895 | |
1896 | if (temp_decorr_pass.value < 0 && !(s->flags & WV_CROSS_DECORR)) |
1897 | temp_decorr_pass.value = -3; |
1898 | |
1899 | decorr_stereo(s->temp_buffer[ j&1][0], s->temp_buffer[ j&1][1], |
1900 | s->temp_buffer[~j&1][0], s->temp_buffer[~j&1][1], |
1901 | FFMIN(2048, nb_samples), &temp_decorr_pass, -1); |
1902 | |
1903 | if (j) { |
1904 | CLEAR(temp_decorr_pass.samplesA); |
1905 | CLEAR(temp_decorr_pass.samplesB); |
1906 | } else { |
1907 | reverse_decorr(&temp_decorr_pass); |
1908 | } |
1909 | |
1910 | memcpy(save_decorr_passes + j, &temp_decorr_pass, sizeof(struct Decorr)); |
1911 | |
1912 | if (((s->flags & MAG_MASK) >> MAG_LSB) >= 16) |
1913 | decorr_stereo(s->temp_buffer[ j&1][0], s->temp_buffer[ j&1][1], |
1914 | s->temp_buffer[~j&1][0], s->temp_buffer[~j&1][1], |
1915 | nb_samples, &temp_decorr_pass, 1); |
1916 | else |
1917 | decorr_stereo_quick(s->temp_buffer[ j&1][0], s->temp_buffer[ j&1][1], |
1918 | s->temp_buffer[~j&1][0], s->temp_buffer[~j&1][1], |
1919 | nb_samples, &temp_decorr_pass); |
1920 | } |
1921 | |
1922 | size = log2stereo(s->temp_buffer[j&1][0], s->temp_buffer[j&1][1], |
1923 | nb_samples, log_limit); |
1924 | if (size != UINT32_MAX || !nterms) |
1925 | break; |
1926 | nterms >>= 1; |
1927 | } |
1928 | |
1929 | if (size < best_size) { |
1930 | memcpy(s->best_buffer[0], s->temp_buffer[j&1][0], buf_size); |
1931 | memcpy(s->best_buffer[1], s->temp_buffer[j&1][1], buf_size); |
1932 | memcpy(s->decorr_passes, save_decorr_passes, sizeof(struct Decorr) * MAX_TERMS); |
1933 | s->num_terms = nterms; |
1934 | s->best_decorr = c; |
1935 | best_size = size; |
1936 | } |
1937 | |
1938 | if (pi++) |
1939 | s->mask_decorr = s->mask_decorr ? ((s->mask_decorr << 1) & (s->num_decorrs - 1)) : 1; |
1940 | } |
1941 | |
1942 | if (force_js || (s->decorr_specs[s->best_decorr].joint_stereo && !force_ts)) |
1943 | s->flags |= WV_JOINT_STEREO; |
1944 | else |
1945 | s->flags &= ~((uint32_t) WV_JOINT_STEREO); |
1946 | |
1947 | if (s->extra_flags) { |
1948 | if (s->flags & WV_JOINT_STEREO) { |
1949 | analyze_stereo(s, s->js_left, s->js_right, do_samples); |
1950 | |
1951 | if (do_samples) { |
1952 | memcpy(samples_l, s->js_left, buf_size); |
1953 | memcpy(samples_r, s->js_right, buf_size); |
1954 | } |
1955 | } else |
1956 | analyze_stereo(s, samples_l, samples_r, do_samples); |
1957 | } else if (do_samples) { |
1958 | memcpy(samples_l, s->best_buffer[0], buf_size); |
1959 | memcpy(samples_r, s->best_buffer[1], buf_size); |
1960 | } |
1961 | |
1962 | if (s->extra_flags || no_history || |
1963 | s->joint_stereo != s->decorr_specs[s->best_decorr].joint_stereo) { |
1964 | s->joint_stereo = s->decorr_specs[s->best_decorr].joint_stereo; |
1965 | CLEAR(s->w); |
1966 | scan_word(s, &s->w.c[0], s->best_buffer[0], nb_samples, -1); |
1967 | scan_word(s, &s->w.c[1], s->best_buffer[1], nb_samples, -1); |
1968 | } |
1969 | return 0; |
1970 | } |
1971 | |
1972 | #define count_bits(av) ( \ |
1973 | (av) < (1 << 8) ? nbits_table[av] : \ |
1974 | ( \ |
1975 | (av) < (1 << 16) ? nbits_table[(av) >> 8] + 8 : \ |
1976 | ((av) < (1 << 24) ? nbits_table[(av) >> 16] + 16 : nbits_table[(av) >> 24] + 24) \ |
1977 | ) \ |
1978 | ) |
1979 | |
1980 | static void encode_flush(WavPackEncodeContext *s) |
1981 | { |
1982 | WavPackWords *w = &s->w; |
1983 | PutBitContext *pb = &s->pb; |
1984 | |
1985 | if (w->zeros_acc) { |
1986 | int cbits = count_bits(w->zeros_acc); |
1987 | |
1988 | do { |
1989 | if (cbits > 31) { |
1990 | put_bits(pb, 31, 0x7FFFFFFF); |
1991 | cbits -= 31; |
1992 | } else { |
1993 | put_bits(pb, cbits, (1 << cbits) - 1); |
1994 | cbits = 0; |
1995 | } |
1996 | } while (cbits); |
1997 | |
1998 | put_bits(pb, 1, 0); |
1999 | |
2000 | while (w->zeros_acc > 1) { |
2001 | put_bits(pb, 1, w->zeros_acc & 1); |
2002 | w->zeros_acc >>= 1; |
2003 | } |
2004 | |
2005 | w->zeros_acc = 0; |
2006 | } |
2007 | |
2008 | if (w->holding_one) { |
2009 | if (w->holding_one >= 16) { |
2010 | int cbits; |
2011 | |
2012 | put_bits(pb, 16, (1 << 16) - 1); |
2013 | put_bits(pb, 1, 0); |
2014 | w->holding_one -= 16; |
2015 | cbits = count_bits(w->holding_one); |
2016 | |
2017 | do { |
2018 | if (cbits > 31) { |
2019 | put_bits(pb, 31, 0x7FFFFFFF); |
2020 | cbits -= 31; |
2021 | } else { |
2022 | put_bits(pb, cbits, (1 << cbits) - 1); |
2023 | cbits = 0; |
2024 | } |
2025 | } while (cbits); |
2026 | |
2027 | put_bits(pb, 1, 0); |
2028 | |
2029 | while (w->holding_one > 1) { |
2030 | put_bits(pb, 1, w->holding_one & 1); |
2031 | w->holding_one >>= 1; |
2032 | } |
2033 | |
2034 | w->holding_zero = 0; |
2035 | } else { |
2036 | put_bits(pb, w->holding_one, (1 << w->holding_one) - 1); |
2037 | } |
2038 | |
2039 | w->holding_one = 0; |
2040 | } |
2041 | |
2042 | if (w->holding_zero) { |
2043 | put_bits(pb, 1, 0); |
2044 | w->holding_zero = 0; |
2045 | } |
2046 | |
2047 | if (w->pend_count) { |
2048 | put_bits(pb, w->pend_count, w->pend_data); |
2049 | w->pend_data = w->pend_count = 0; |
2050 | } |
2051 | } |
2052 | |
2053 | static void wavpack_encode_sample(WavPackEncodeContext *s, WvChannel *c, int32_t sample) |
2054 | { |
2055 | WavPackWords *w = &s->w; |
2056 | uint32_t ones_count, low, high; |
2057 | int sign = sample < 0; |
2058 | |
2059 | if (s->w.c[0].median[0] < 2 && !s->w.holding_zero && s->w.c[1].median[0] < 2) { |
2060 | if (w->zeros_acc) { |
2061 | if (sample) |
2062 | encode_flush(s); |
2063 | else { |
2064 | w->zeros_acc++; |
2065 | return; |
2066 | } |
2067 | } else if (sample) { |
2068 | put_bits(&s->pb, 1, 0); |
2069 | } else { |
2070 | CLEAR(s->w.c[0].median); |
2071 | CLEAR(s->w.c[1].median); |
2072 | w->zeros_acc = 1; |
2073 | return; |
2074 | } |
2075 | } |
2076 | |
2077 | if (sign) |
2078 | sample = ~sample; |
2079 | |
2080 | if (sample < (int32_t) GET_MED(0)) { |
2081 | ones_count = low = 0; |
2082 | high = GET_MED(0) - 1; |
2083 | DEC_MED(0); |
2084 | } else { |
2085 | low = GET_MED(0); |
2086 | INC_MED(0); |
2087 | |
2088 | if (sample - low < GET_MED(1)) { |
2089 | ones_count = 1; |
2090 | high = low + GET_MED(1) - 1; |
2091 | DEC_MED(1); |
2092 | } else { |
2093 | low += GET_MED(1); |
2094 | INC_MED(1); |
2095 | |
2096 | if (sample - low < GET_MED(2)) { |
2097 | ones_count = 2; |
2098 | high = low + GET_MED(2) - 1; |
2099 | DEC_MED(2); |
2100 | } else { |
2101 | ones_count = 2 + (sample - low) / GET_MED(2); |
2102 | low += (ones_count - 2) * GET_MED(2); |
2103 | high = low + GET_MED(2) - 1; |
2104 | INC_MED(2); |
2105 | } |
2106 | } |
2107 | } |
2108 | |
2109 | if (w->holding_zero) { |
2110 | if (ones_count) |
2111 | w->holding_one++; |
2112 | |
2113 | encode_flush(s); |
2114 | |
2115 | if (ones_count) { |
2116 | w->holding_zero = 1; |
2117 | ones_count--; |
2118 | } else |
2119 | w->holding_zero = 0; |
2120 | } else |
2121 | w->holding_zero = 1; |
2122 | |
2123 | w->holding_one = ones_count * 2; |
2124 | |
2125 | if (high != low) { |
2126 | uint32_t maxcode = high - low, code = sample - low; |
2127 | int bitcount = count_bits(maxcode); |
2128 | uint32_t extras = (1 << bitcount) - maxcode - 1; |
2129 | |
2130 | if (code < extras) { |
2131 | w->pend_data |= code << w->pend_count; |
2132 | w->pend_count += bitcount - 1; |
2133 | } else { |
2134 | w->pend_data |= ((code + extras) >> 1) << w->pend_count; |
2135 | w->pend_count += bitcount - 1; |
2136 | w->pend_data |= ((code + extras) & 1) << w->pend_count++; |
2137 | } |
2138 | } |
2139 | |
2140 | w->pend_data |= ((int32_t) sign << w->pend_count++); |
2141 | |
2142 | if (!w->holding_zero) |
2143 | encode_flush(s); |
2144 | } |
2145 | |
2146 | static void pack_int32(WavPackEncodeContext *s, |
2147 | int32_t *samples_l, int32_t *samples_r, |
2148 | int nb_samples) |
2149 | { |
2150 | const int sent_bits = s->int32_sent_bits; |
2151 | PutBitContext *pb = &s->pb; |
2152 | int i, pre_shift; |
2153 | |
2154 | pre_shift = s->int32_zeros + s->int32_ones + s->int32_dups; |
2155 | |
2156 | if (!sent_bits) |
2157 | return; |
2158 | |
2159 | if (s->flags & WV_MONO_DATA) { |
2160 | for (i = 0; i < nb_samples; i++) { |
2161 | put_sbits(pb, sent_bits, samples_l[i] >> pre_shift); |
2162 | } |
2163 | } else { |
2164 | for (i = 0; i < nb_samples; i++) { |
2165 | put_sbits(pb, sent_bits, samples_l[i] >> pre_shift); |
2166 | put_sbits(pb, sent_bits, samples_r[i] >> pre_shift); |
2167 | } |
2168 | } |
2169 | } |
2170 | |
2171 | static void pack_float_sample(WavPackEncodeContext *s, int32_t *sample) |
2172 | { |
2173 | const int max_exp = s->float_max_exp; |
2174 | PutBitContext *pb = &s->pb; |
2175 | int32_t value, shift_count; |
2176 | |
2177 | if (get_exponent(*sample) == 255) { |
2178 | if (get_mantissa(*sample)) { |
2179 | put_bits(pb, 1, 1); |
2180 | put_bits(pb, 23, get_mantissa(*sample)); |
2181 | } else { |
2182 | put_bits(pb, 1, 0); |
2183 | } |
2184 | |
2185 | value = 0x1000000; |
2186 | shift_count = 0; |
2187 | } else if (get_exponent(*sample)) { |
2188 | shift_count = max_exp - get_exponent(*sample); |
2189 | value = 0x800000 + get_mantissa(*sample); |
2190 | } else { |
2191 | shift_count = max_exp ? max_exp - 1 : 0; |
2192 | value = get_mantissa(*sample); |
2193 | } |
2194 | |
2195 | if (shift_count < 25) |
2196 | value >>= shift_count; |
2197 | else |
2198 | value = 0; |
2199 | |
2200 | if (!value) { |
2201 | if (s->float_flags & FLOAT_ZEROS_SENT) { |
2202 | if (get_exponent(*sample) || get_mantissa(*sample)) { |
2203 | put_bits(pb, 1, 1); |
2204 | put_bits(pb, 23, get_mantissa(*sample)); |
2205 | |
2206 | if (max_exp >= 25) |
2207 | put_bits(pb, 8, get_exponent(*sample)); |
2208 | |
2209 | put_bits(pb, 1, get_sign(*sample)); |
2210 | } else { |
2211 | put_bits(pb, 1, 0); |
2212 | |
2213 | if (s->float_flags & FLOAT_NEG_ZEROS) |
2214 | put_bits(pb, 1, get_sign(*sample)); |
2215 | } |
2216 | } |
2217 | } else if (shift_count) { |
2218 | if (s->float_flags & FLOAT_SHIFT_SENT) { |
2219 | put_sbits(pb, shift_count, get_mantissa(*sample)); |
2220 | } else if (s->float_flags & FLOAT_SHIFT_SAME) { |
2221 | put_bits(pb, 1, get_mantissa(*sample) & 1); |
2222 | } |
2223 | } |
2224 | } |
2225 | |
2226 | static void pack_float(WavPackEncodeContext *s, |
2227 | int32_t *samples_l, int32_t *samples_r, |
2228 | int nb_samples) |
2229 | { |
2230 | int i; |
2231 | |
2232 | if (s->flags & WV_MONO_DATA) { |
2233 | for (i = 0; i < nb_samples; i++) |
2234 | pack_float_sample(s, &samples_l[i]); |
2235 | } else { |
2236 | for (i = 0; i < nb_samples; i++) { |
2237 | pack_float_sample(s, &samples_l[i]); |
2238 | pack_float_sample(s, &samples_r[i]); |
2239 | } |
2240 | } |
2241 | } |
2242 | |
2243 | static void decorr_stereo_pass2(struct Decorr *dpp, |
2244 | int32_t *samples_l, int32_t *samples_r, |
2245 | int nb_samples) |
2246 | { |
2247 | int i, m, k; |
2248 | |
2249 | switch (dpp->value) { |
2250 | case 17: |
2251 | for (i = 0; i < nb_samples; i++) { |
2252 | int32_t sam, tmp; |
2253 | |
2254 | sam = 2 * dpp->samplesA[0] - dpp->samplesA[1]; |
2255 | dpp->samplesA[1] = dpp->samplesA[0]; |
2256 | samples_l[i] = tmp = (dpp->samplesA[0] = samples_l[i]) - APPLY_WEIGHT(dpp->weightA, sam); |
2257 | UPDATE_WEIGHT(dpp->weightA, dpp->delta, sam, tmp); |
2258 | |
2259 | sam = 2 * dpp->samplesB[0] - dpp->samplesB[1]; |
2260 | dpp->samplesB[1] = dpp->samplesB[0]; |
2261 | samples_r[i] = tmp = (dpp->samplesB[0] = samples_r[i]) - APPLY_WEIGHT(dpp->weightB, sam); |
2262 | UPDATE_WEIGHT(dpp->weightB, dpp->delta, sam, tmp); |
2263 | } |
2264 | break; |
2265 | case 18: |
2266 | for (i = 0; i < nb_samples; i++) { |
2267 | int32_t sam, tmp; |
2268 | |
2269 | sam = dpp->samplesA[0] + ((dpp->samplesA[0] - dpp->samplesA[1]) >> 1); |
2270 | dpp->samplesA[1] = dpp->samplesA[0]; |
2271 | samples_l[i] = tmp = (dpp->samplesA[0] = samples_l[i]) - APPLY_WEIGHT(dpp->weightA, sam); |
2272 | UPDATE_WEIGHT(dpp->weightA, dpp->delta, sam, tmp); |
2273 | |
2274 | sam = dpp->samplesB[0] + ((dpp->samplesB[0] - dpp->samplesB[1]) >> 1); |
2275 | dpp->samplesB[1] = dpp->samplesB[0]; |
2276 | samples_r[i] = tmp = (dpp->samplesB[0] = samples_r[i]) - APPLY_WEIGHT(dpp->weightB, sam); |
2277 | UPDATE_WEIGHT(dpp->weightB, dpp->delta, sam, tmp); |
2278 | } |
2279 | break; |
2280 | default: |
2281 | for (m = 0, k = dpp->value & (MAX_TERM - 1), i = 0; i < nb_samples; i++) { |
2282 | int32_t sam, tmp; |
2283 | |
2284 | sam = dpp->samplesA[m]; |
2285 | samples_l[i] = tmp = (dpp->samplesA[k] = samples_l[i]) - APPLY_WEIGHT(dpp->weightA, sam); |
2286 | UPDATE_WEIGHT(dpp->weightA, dpp->delta, sam, tmp); |
2287 | |
2288 | sam = dpp->samplesB[m]; |
2289 | samples_r[i] = tmp = (dpp->samplesB[k] = samples_r[i]) - APPLY_WEIGHT(dpp->weightB, sam); |
2290 | UPDATE_WEIGHT(dpp->weightB, dpp->delta, sam, tmp); |
2291 | |
2292 | m = (m + 1) & (MAX_TERM - 1); |
2293 | k = (k + 1) & (MAX_TERM - 1); |
2294 | } |
2295 | if (m) { |
2296 | int32_t temp_A[MAX_TERM], temp_B[MAX_TERM]; |
2297 | |
2298 | memcpy(temp_A, dpp->samplesA, sizeof (dpp->samplesA)); |
2299 | memcpy(temp_B, dpp->samplesB, sizeof (dpp->samplesB)); |
2300 | |
2301 | for (k = 0; k < MAX_TERM; k++) { |
2302 | dpp->samplesA[k] = temp_A[m]; |
2303 | dpp->samplesB[k] = temp_B[m]; |
2304 | m = (m + 1) & (MAX_TERM - 1); |
2305 | } |
2306 | } |
2307 | break; |
2308 | case -1: |
2309 | for (i = 0; i < nb_samples; i++) { |
2310 | int32_t sam_A, sam_B, tmp; |
2311 | |
2312 | sam_A = dpp->samplesA[0]; |
2313 | samples_l[i] = tmp = (sam_B = samples_l[i]) - APPLY_WEIGHT(dpp->weightA, sam_A); |
2314 | UPDATE_WEIGHT_CLIP(dpp->weightA, dpp->delta, sam_A, tmp); |
2315 | |
2316 | samples_r[i] = tmp = (dpp->samplesA[0] = samples_r[i]) - APPLY_WEIGHT(dpp->weightB, sam_B); |
2317 | UPDATE_WEIGHT_CLIP(dpp->weightB, dpp->delta, sam_B, tmp); |
2318 | } |
2319 | break; |
2320 | case -2: |
2321 | for (i = 0; i < nb_samples; i++) { |
2322 | int32_t sam_A, sam_B, tmp; |
2323 | |
2324 | sam_B = dpp->samplesB[0]; |
2325 | samples_r[i] = tmp = (sam_A = samples_r[i]) - APPLY_WEIGHT(dpp->weightB, sam_B); |
2326 | UPDATE_WEIGHT_CLIP(dpp->weightB, dpp->delta, sam_B, tmp); |
2327 | |
2328 | samples_l[i] = tmp = (dpp->samplesB[0] = samples_l[i]) - APPLY_WEIGHT(dpp->weightA, sam_A); |
2329 | UPDATE_WEIGHT_CLIP(dpp->weightA, dpp->delta, sam_A, tmp); |
2330 | } |
2331 | break; |
2332 | case -3: |
2333 | for (i = 0; i < nb_samples; i++) { |
2334 | int32_t sam_A, sam_B, tmp; |
2335 | |
2336 | sam_A = dpp->samplesA[0]; |
2337 | sam_B = dpp->samplesB[0]; |
2338 | |
2339 | dpp->samplesA[0] = tmp = samples_r[i]; |
2340 | samples_r[i] = tmp -= APPLY_WEIGHT(dpp->weightB, sam_B); |
2341 | UPDATE_WEIGHT_CLIP(dpp->weightB, dpp->delta, sam_B, tmp); |
2342 | |
2343 | dpp->samplesB[0] = tmp = samples_l[i]; |
2344 | samples_l[i] = tmp -= APPLY_WEIGHT(dpp->weightA, sam_A); |
2345 | UPDATE_WEIGHT_CLIP(dpp->weightA, dpp->delta, sam_A, tmp); |
2346 | } |
2347 | break; |
2348 | } |
2349 | } |
2350 | |
2351 | #define update_weight_d2(weight, delta, source, result) \ |
2352 | if (source && result) \ |
2353 | weight -= (((source ^ result) >> 29) & 4) - 2; |
2354 | |
2355 | #define update_weight_clip_d2(weight, delta, source, result) \ |
2356 | if (source && result) { \ |
2357 | const int32_t s = (source ^ result) >> 31; \ |
2358 | if ((weight = (weight ^ s) + (2 - s)) > 1024) weight = 1024; \ |
2359 | weight = (weight ^ s) - s; \ |
2360 | } |
2361 | |
2362 | static void decorr_stereo_pass_id2(struct Decorr *dpp, |
2363 | int32_t *samples_l, int32_t *samples_r, |
2364 | int nb_samples) |
2365 | { |
2366 | int i, m, k; |
2367 | |
2368 | switch (dpp->value) { |
2369 | case 17: |
2370 | for (i = 0; i < nb_samples; i++) { |
2371 | int32_t sam, tmp; |
2372 | |
2373 | sam = 2 * dpp->samplesA[0] - dpp->samplesA[1]; |
2374 | dpp->samplesA[1] = dpp->samplesA[0]; |
2375 | samples_l[i] = tmp = (dpp->samplesA[0] = samples_l[i]) - APPLY_WEIGHT_I(dpp->weightA, sam); |
2376 | update_weight_d2(dpp->weightA, dpp->delta, sam, tmp); |
2377 | |
2378 | sam = 2 * dpp->samplesB[0] - dpp->samplesB[1]; |
2379 | dpp->samplesB[1] = dpp->samplesB[0]; |
2380 | samples_r[i] = tmp = (dpp->samplesB[0] = samples_r[i]) - APPLY_WEIGHT_I(dpp->weightB, sam); |
2381 | update_weight_d2(dpp->weightB, dpp->delta, sam, tmp); |
2382 | } |
2383 | break; |
2384 | case 18: |
2385 | for (i = 0; i < nb_samples; i++) { |
2386 | int32_t sam, tmp; |
2387 | |
2388 | sam = dpp->samplesA[0] + ((dpp->samplesA[0] - dpp->samplesA[1]) >> 1); |
2389 | dpp->samplesA[1] = dpp->samplesA[0]; |
2390 | samples_l[i] = tmp = (dpp->samplesA[0] = samples_l[i]) - APPLY_WEIGHT_I(dpp->weightA, sam); |
2391 | update_weight_d2(dpp->weightA, dpp->delta, sam, tmp); |
2392 | |
2393 | sam = dpp->samplesB[0] + ((dpp->samplesB[0] - dpp->samplesB[1]) >> 1); |
2394 | dpp->samplesB[1] = dpp->samplesB[0]; |
2395 | samples_r[i] = tmp = (dpp->samplesB[0] = samples_r[i]) - APPLY_WEIGHT_I(dpp->weightB, sam); |
2396 | update_weight_d2(dpp->weightB, dpp->delta, sam, tmp); |
2397 | } |
2398 | break; |
2399 | default: |
2400 | for (m = 0, k = dpp->value & (MAX_TERM - 1), i = 0; i < nb_samples; i++) { |
2401 | int32_t sam, tmp; |
2402 | |
2403 | sam = dpp->samplesA[m]; |
2404 | samples_l[i] = tmp = (dpp->samplesA[k] = samples_l[i]) - APPLY_WEIGHT_I(dpp->weightA, sam); |
2405 | update_weight_d2(dpp->weightA, dpp->delta, sam, tmp); |
2406 | |
2407 | sam = dpp->samplesB[m]; |
2408 | samples_r[i] = tmp = (dpp->samplesB[k] = samples_r[i]) - APPLY_WEIGHT_I(dpp->weightB, sam); |
2409 | update_weight_d2(dpp->weightB, dpp->delta, sam, tmp); |
2410 | |
2411 | m = (m + 1) & (MAX_TERM - 1); |
2412 | k = (k + 1) & (MAX_TERM - 1); |
2413 | } |
2414 | |
2415 | if (m) { |
2416 | int32_t temp_A[MAX_TERM], temp_B[MAX_TERM]; |
2417 | |
2418 | memcpy(temp_A, dpp->samplesA, sizeof(dpp->samplesA)); |
2419 | memcpy(temp_B, dpp->samplesB, sizeof(dpp->samplesB)); |
2420 | |
2421 | for (k = 0; k < MAX_TERM; k++) { |
2422 | dpp->samplesA[k] = temp_A[m]; |
2423 | dpp->samplesB[k] = temp_B[m]; |
2424 | m = (m + 1) & (MAX_TERM - 1); |
2425 | } |
2426 | } |
2427 | break; |
2428 | case -1: |
2429 | for (i = 0; i < nb_samples; i++) { |
2430 | int32_t sam_A, sam_B, tmp; |
2431 | |
2432 | sam_A = dpp->samplesA[0]; |
2433 | samples_l[i] = tmp = (sam_B = samples_l[i]) - APPLY_WEIGHT_I(dpp->weightA, sam_A); |
2434 | update_weight_clip_d2(dpp->weightA, dpp->delta, sam_A, tmp); |
2435 | |
2436 | samples_r[i] = tmp = (dpp->samplesA[0] = samples_r[i]) - APPLY_WEIGHT_I(dpp->weightB, sam_B); |
2437 | update_weight_clip_d2(dpp->weightB, dpp->delta, sam_B, tmp); |
2438 | } |
2439 | break; |
2440 | case -2: |
2441 | for (i = 0; i < nb_samples; i++) { |
2442 | int32_t sam_A, sam_B, tmp; |
2443 | |
2444 | sam_B = dpp->samplesB[0]; |
2445 | samples_r[i] = tmp = (sam_A = samples_r[i]) - APPLY_WEIGHT_I(dpp->weightB, sam_B); |
2446 | update_weight_clip_d2(dpp->weightB, dpp->delta, sam_B, tmp); |
2447 | |
2448 | samples_l[i] = tmp = (dpp->samplesB[0] = samples_l[i]) - APPLY_WEIGHT_I(dpp->weightA, sam_A); |
2449 | update_weight_clip_d2(dpp->weightA, dpp->delta, sam_A, tmp); |
2450 | } |
2451 | break; |
2452 | case -3: |
2453 | for (i = 0; i < nb_samples; i++) { |
2454 | int32_t sam_A, sam_B, tmp; |
2455 | |
2456 | sam_A = dpp->samplesA[0]; |
2457 | sam_B = dpp->samplesB[0]; |
2458 | |
2459 | dpp->samplesA[0] = tmp = samples_r[i]; |
2460 | samples_r[i] = tmp -= APPLY_WEIGHT_I(dpp->weightB, sam_B); |
2461 | update_weight_clip_d2(dpp->weightB, dpp->delta, sam_B, tmp); |
2462 | |
2463 | dpp->samplesB[0] = tmp = samples_l[i]; |
2464 | samples_l[i] = tmp -= APPLY_WEIGHT_I(dpp->weightA, sam_A); |
2465 | update_weight_clip_d2(dpp->weightA, dpp->delta, sam_A, tmp); |
2466 | } |
2467 | break; |
2468 | } |
2469 | } |
2470 | |
2471 | static void put_metadata_block(PutByteContext *pb, int flags, int size) |
2472 | { |
2473 | if (size & 1) |
2474 | flags |= WP_IDF_ODD; |
2475 | |
2476 | bytestream2_put_byte(pb, flags); |
2477 | bytestream2_put_byte(pb, (size + 1) >> 1); |
2478 | } |
2479 | |
2480 | static int wavpack_encode_block(WavPackEncodeContext *s, |
2481 | int32_t *samples_l, int32_t *samples_r, |
2482 | uint8_t *out, int out_size) |
2483 | { |
2484 | int block_size, start, end, data_size, tcount, temp, m = 0; |
2485 | int i, j, ret = 0, got_extra = 0, nb_samples = s->block_samples; |
2486 | uint32_t crc = 0xffffffffu; |
2487 | struct Decorr *dpp; |
2488 | PutByteContext pb; |
2489 | |
2490 | if (s->flags & WV_MONO_DATA) { |
2491 | CLEAR(s->w); |
2492 | } |
2493 | if (!(s->flags & WV_MONO) && s->optimize_mono) { |
2494 | int32_t lor = 0, diff = 0; |
2495 | |
2496 | for (i = 0; i < nb_samples; i++) { |
2497 | lor |= samples_l[i] | samples_r[i]; |
2498 | diff |= samples_l[i] - samples_r[i]; |
2499 | |
2500 | if (lor && diff) |
2501 | break; |
2502 | } |
2503 | |
2504 | if (i == nb_samples && lor && !diff) { |
2505 | s->flags &= ~(WV_JOINT_STEREO | WV_CROSS_DECORR); |
2506 | s->flags |= WV_FALSE_STEREO; |
2507 | |
2508 | if (!s->false_stereo) { |
2509 | s->false_stereo = 1; |
2510 | s->num_terms = 0; |
2511 | CLEAR(s->w); |
2512 | } |
2513 | } else if (s->false_stereo) { |
2514 | s->false_stereo = 0; |
2515 | s->num_terms = 0; |
2516 | CLEAR(s->w); |
2517 | } |
2518 | } |
2519 | |
2520 | if (s->flags & SHIFT_MASK) { |
2521 | int shift = (s->flags & SHIFT_MASK) >> SHIFT_LSB; |
2522 | int mag = (s->flags & MAG_MASK) >> MAG_LSB; |
2523 | |
2524 | if (s->flags & WV_MONO_DATA) |
2525 | shift_mono(samples_l, nb_samples, shift); |
2526 | else |
2527 | shift_stereo(samples_l, samples_r, nb_samples, shift); |
2528 | |
2529 | if ((mag -= shift) < 0) |
2530 | s->flags &= ~MAG_MASK; |
2531 | else |
2532 | s->flags -= (1 << MAG_LSB) * shift; |
2533 | } |
2534 | |
2535 | if ((s->flags & WV_FLOAT_DATA) || (s->flags & MAG_MASK) >> MAG_LSB >= 24) { |
2536 | av_fast_padded_malloc(&s->orig_l, &s->orig_l_size, sizeof(int32_t) * nb_samples); |
2537 | memcpy(s->orig_l, samples_l, sizeof(int32_t) * nb_samples); |
2538 | if (!(s->flags & WV_MONO_DATA)) { |
2539 | av_fast_padded_malloc(&s->orig_r, &s->orig_r_size, sizeof(int32_t) * nb_samples); |
2540 | memcpy(s->orig_r, samples_r, sizeof(int32_t) * nb_samples); |
2541 | } |
2542 | |
2543 | if (s->flags & WV_FLOAT_DATA) |
2544 | got_extra = scan_float(s, samples_l, samples_r, nb_samples); |
2545 | else |
2546 | got_extra = scan_int32(s, samples_l, samples_r, nb_samples); |
2547 | s->num_terms = 0; |
2548 | } else { |
2549 | scan_int23(s, samples_l, samples_r, nb_samples); |
2550 | if (s->shift != s->int32_zeros + s->int32_ones + s->int32_dups) { |
2551 | s->shift = s->int32_zeros + s->int32_ones + s->int32_dups; |
2552 | s->num_terms = 0; |
2553 | } |
2554 | } |
2555 | |
2556 | if (!s->num_passes && !s->num_terms) { |
2557 | s->num_passes = 1; |
2558 | |
2559 | if (s->flags & WV_MONO_DATA) |
2560 | ret = wv_mono(s, samples_l, 1, 0); |
2561 | else |
2562 | ret = wv_stereo(s, samples_l, samples_r, 1, 0); |
2563 | |
2564 | s->num_passes = 0; |
2565 | } |
2566 | if (s->flags & WV_MONO_DATA) { |
2567 | for (i = 0; i < nb_samples; i++) |
2568 | crc += (crc << 1) + samples_l[i]; |
2569 | |
2570 | if (s->num_passes) |
2571 | ret = wv_mono(s, samples_l, !s->num_terms, 1); |
2572 | } else { |
2573 | for (i = 0; i < nb_samples; i++) |
2574 | crc += (crc << 3) + (samples_l[i] << 1) + samples_l[i] + samples_r[i]; |
2575 | |
2576 | if (s->num_passes) |
2577 | ret = wv_stereo(s, samples_l, samples_r, !s->num_terms, 1); |
2578 | } |
2579 | if (ret < 0) |
2580 | return ret; |
2581 | |
2582 | if (!s->ch_offset) |
2583 | s->flags |= WV_INITIAL_BLOCK; |
2584 | |
2585 | s->ch_offset += 1 + !(s->flags & WV_MONO); |
2586 | |
2587 | if (s->ch_offset == s->avctx->channels) |
2588 | s->flags |= WV_FINAL_BLOCK; |
2589 | |
2590 | bytestream2_init_writer(&pb, out, out_size); |
2591 | bytestream2_put_le32(&pb, MKTAG('w', 'v', 'p', 'k')); |
2592 | bytestream2_put_le32(&pb, 0); |
2593 | bytestream2_put_le16(&pb, 0x410); |
2594 | bytestream2_put_le16(&pb, 0); |
2595 | bytestream2_put_le32(&pb, 0); |
2596 | bytestream2_put_le32(&pb, s->sample_index); |
2597 | bytestream2_put_le32(&pb, nb_samples); |
2598 | bytestream2_put_le32(&pb, s->flags); |
2599 | bytestream2_put_le32(&pb, crc); |
2600 | |
2601 | if (s->flags & WV_INITIAL_BLOCK && |
2602 | s->avctx->channel_layout != AV_CH_LAYOUT_MONO && |
2603 | s->avctx->channel_layout != AV_CH_LAYOUT_STEREO) { |
2604 | put_metadata_block(&pb, WP_ID_CHANINFO, 5); |
2605 | bytestream2_put_byte(&pb, s->avctx->channels); |
2606 | bytestream2_put_le32(&pb, s->avctx->channel_layout); |
2607 | bytestream2_put_byte(&pb, 0); |
2608 | } |
2609 | |
2610 | if ((s->flags & SRATE_MASK) == SRATE_MASK) { |
2611 | put_metadata_block(&pb, WP_ID_SAMPLE_RATE, 3); |
2612 | bytestream2_put_le24(&pb, s->avctx->sample_rate); |
2613 | bytestream2_put_byte(&pb, 0); |
2614 | } |
2615 | |
2616 | put_metadata_block(&pb, WP_ID_DECTERMS, s->num_terms); |
2617 | for (i = 0; i < s->num_terms; i++) { |
2618 | struct Decorr *dpp = &s->decorr_passes[i]; |
2619 | bytestream2_put_byte(&pb, ((dpp->value + 5) & 0x1f) | ((dpp->delta << 5) & 0xe0)); |
2620 | } |
2621 | if (s->num_terms & 1) |
2622 | bytestream2_put_byte(&pb, 0); |
2623 | |
2624 | #define WRITE_DECWEIGHT(type) do { \ |
2625 | temp = store_weight(type); \ |
2626 | bytestream2_put_byte(&pb, temp); \ |
2627 | type = restore_weight(temp); \ |
2628 | } while (0) |
2629 | |
2630 | bytestream2_put_byte(&pb, WP_ID_DECWEIGHTS); |
2631 | bytestream2_put_byte(&pb, 0); |
2632 | start = bytestream2_tell_p(&pb); |
2633 | for (i = s->num_terms - 1; i >= 0; --i) { |
2634 | struct Decorr *dpp = &s->decorr_passes[i]; |
2635 | |
2636 | if (store_weight(dpp->weightA) || |
2637 | (!(s->flags & WV_MONO_DATA) && store_weight(dpp->weightB))) |
2638 | break; |
2639 | } |
2640 | tcount = i + 1; |
2641 | for (i = 0; i < s->num_terms; i++) { |
2642 | struct Decorr *dpp = &s->decorr_passes[i]; |
2643 | if (i < tcount) { |
2644 | WRITE_DECWEIGHT(dpp->weightA); |
2645 | if (!(s->flags & WV_MONO_DATA)) |
2646 | WRITE_DECWEIGHT(dpp->weightB); |
2647 | } else { |
2648 | dpp->weightA = dpp->weightB = 0; |
2649 | } |
2650 | } |
2651 | end = bytestream2_tell_p(&pb); |
2652 | out[start - 2] = WP_ID_DECWEIGHTS | (((end - start) & 1) ? WP_IDF_ODD: 0); |
2653 | out[start - 1] = (end - start + 1) >> 1; |
2654 | if ((end - start) & 1) |
2655 | bytestream2_put_byte(&pb, 0); |
2656 | |
2657 | #define WRITE_DECSAMPLE(type) do { \ |
2658 | temp = log2s(type); \ |
2659 | type = wp_exp2(temp); \ |
2660 | bytestream2_put_le16(&pb, temp); \ |
2661 | } while (0) |
2662 | |
2663 | bytestream2_put_byte(&pb, WP_ID_DECSAMPLES); |
2664 | bytestream2_put_byte(&pb, 0); |
2665 | start = bytestream2_tell_p(&pb); |
2666 | for (i = 0; i < s->num_terms; i++) { |
2667 | struct Decorr *dpp = &s->decorr_passes[i]; |
2668 | if (i == 0) { |
2669 | if (dpp->value > MAX_TERM) { |
2670 | WRITE_DECSAMPLE(dpp->samplesA[0]); |
2671 | WRITE_DECSAMPLE(dpp->samplesA[1]); |
2672 | if (!(s->flags & WV_MONO_DATA)) { |
2673 | WRITE_DECSAMPLE(dpp->samplesB[0]); |
2674 | WRITE_DECSAMPLE(dpp->samplesB[1]); |
2675 | } |
2676 | } else if (dpp->value < 0) { |
2677 | WRITE_DECSAMPLE(dpp->samplesA[0]); |
2678 | WRITE_DECSAMPLE(dpp->samplesB[0]); |
2679 | } else { |
2680 | for (j = 0; j < dpp->value; j++) { |
2681 | WRITE_DECSAMPLE(dpp->samplesA[j]); |
2682 | if (!(s->flags & WV_MONO_DATA)) |
2683 | WRITE_DECSAMPLE(dpp->samplesB[j]); |
2684 | } |
2685 | } |
2686 | } else { |
2687 | CLEAR(dpp->samplesA); |
2688 | CLEAR(dpp->samplesB); |
2689 | } |
2690 | } |
2691 | end = bytestream2_tell_p(&pb); |
2692 | out[start - 1] = (end - start) >> 1; |
2693 | |
2694 | #define WRITE_CHAN_ENTROPY(chan) do { \ |
2695 | for (i = 0; i < 3; i++) { \ |
2696 | temp = wp_log2(s->w.c[chan].median[i]); \ |
2697 | bytestream2_put_le16(&pb, temp); \ |
2698 | s->w.c[chan].median[i] = wp_exp2(temp); \ |
2699 | } \ |
2700 | } while (0) |
2701 | |
2702 | put_metadata_block(&pb, WP_ID_ENTROPY, 6 * (1 + (!(s->flags & WV_MONO_DATA)))); |
2703 | WRITE_CHAN_ENTROPY(0); |
2704 | if (!(s->flags & WV_MONO_DATA)) |
2705 | WRITE_CHAN_ENTROPY(1); |
2706 | |
2707 | if (s->flags & WV_FLOAT_DATA) { |
2708 | put_metadata_block(&pb, WP_ID_FLOATINFO, 4); |
2709 | bytestream2_put_byte(&pb, s->float_flags); |
2710 | bytestream2_put_byte(&pb, s->float_shift); |
2711 | bytestream2_put_byte(&pb, s->float_max_exp); |
2712 | bytestream2_put_byte(&pb, 127); |
2713 | } |
2714 | |
2715 | if (s->flags & WV_INT32_DATA) { |
2716 | put_metadata_block(&pb, WP_ID_INT32INFO, 4); |
2717 | bytestream2_put_byte(&pb, s->int32_sent_bits); |
2718 | bytestream2_put_byte(&pb, s->int32_zeros); |
2719 | bytestream2_put_byte(&pb, s->int32_ones); |
2720 | bytestream2_put_byte(&pb, s->int32_dups); |
2721 | } |
2722 | |
2723 | if (s->flags & WV_MONO_DATA && !s->num_passes) { |
2724 | for (i = 0; i < nb_samples; i++) { |
2725 | int32_t code = samples_l[i]; |
2726 | |
2727 | for (tcount = s->num_terms, dpp = s->decorr_passes; tcount--; dpp++) { |
2728 | int32_t sam; |
2729 | |
2730 | if (dpp->value > MAX_TERM) { |
2731 | if (dpp->value & 1) |
2732 | sam = 2 * dpp->samplesA[0] - dpp->samplesA[1]; |
2733 | else |
2734 | sam = (3 * dpp->samplesA[0] - dpp->samplesA[1]) >> 1; |
2735 | |
2736 | dpp->samplesA[1] = dpp->samplesA[0]; |
2737 | dpp->samplesA[0] = code; |
2738 | } else { |
2739 | sam = dpp->samplesA[m]; |
2740 | dpp->samplesA[(m + dpp->value) & (MAX_TERM - 1)] = code; |
2741 | } |
2742 | |
2743 | code -= APPLY_WEIGHT(dpp->weightA, sam); |
2744 | UPDATE_WEIGHT(dpp->weightA, dpp->delta, sam, code); |
2745 | } |
2746 | |
2747 | m = (m + 1) & (MAX_TERM - 1); |
2748 | samples_l[i] = code; |
2749 | } |
2750 | if (m) { |
2751 | for (tcount = s->num_terms, dpp = s->decorr_passes; tcount--; dpp++) |
2752 | if (dpp->value > 0 && dpp->value <= MAX_TERM) { |
2753 | int32_t temp_A[MAX_TERM], temp_B[MAX_TERM]; |
2754 | int k; |
2755 | |
2756 | memcpy(temp_A, dpp->samplesA, sizeof(dpp->samplesA)); |
2757 | memcpy(temp_B, dpp->samplesB, sizeof(dpp->samplesB)); |
2758 | |
2759 | for (k = 0; k < MAX_TERM; k++) { |
2760 | dpp->samplesA[k] = temp_A[m]; |
2761 | dpp->samplesB[k] = temp_B[m]; |
2762 | m = (m + 1) & (MAX_TERM - 1); |
2763 | } |
2764 | } |
2765 | } |
2766 | } else if (!s->num_passes) { |
2767 | if (s->flags & WV_JOINT_STEREO) { |
2768 | for (i = 0; i < nb_samples; i++) |
2769 | samples_r[i] += ((samples_l[i] -= samples_r[i]) >> 1); |
2770 | } |
2771 | |
2772 | for (i = 0; i < s->num_terms; i++) { |
2773 | struct Decorr *dpp = &s->decorr_passes[i]; |
2774 | if (((s->flags & MAG_MASK) >> MAG_LSB) >= 16 || dpp->delta != 2) |
2775 | decorr_stereo_pass2(dpp, samples_l, samples_r, nb_samples); |
2776 | else |
2777 | decorr_stereo_pass_id2(dpp, samples_l, samples_r, nb_samples); |
2778 | } |
2779 | } |
2780 | |
2781 | bytestream2_put_byte(&pb, WP_ID_DATA | WP_IDF_LONG); |
2782 | init_put_bits(&s->pb, pb.buffer + 3, bytestream2_get_bytes_left_p(&pb)); |
2783 | if (s->flags & WV_MONO_DATA) { |
2784 | for (i = 0; i < nb_samples; i++) |
2785 | wavpack_encode_sample(s, &s->w.c[0], s->samples[0][i]); |
2786 | } else { |
2787 | for (i = 0; i < nb_samples; i++) { |
2788 | wavpack_encode_sample(s, &s->w.c[0], s->samples[0][i]); |
2789 | wavpack_encode_sample(s, &s->w.c[1], s->samples[1][i]); |
2790 | } |
2791 | } |
2792 | encode_flush(s); |
2793 | flush_put_bits(&s->pb); |
2794 | data_size = put_bits_count(&s->pb) >> 3; |
2795 | bytestream2_put_le24(&pb, (data_size + 1) >> 1); |
2796 | bytestream2_skip_p(&pb, data_size); |
2797 | if (data_size & 1) |
2798 | bytestream2_put_byte(&pb, 0); |
2799 | |
2800 | if (got_extra) { |
2801 | bytestream2_put_byte(&pb, WP_ID_EXTRABITS | WP_IDF_LONG); |
2802 | init_put_bits(&s->pb, pb.buffer + 7, bytestream2_get_bytes_left_p(&pb)); |
2803 | if (s->flags & WV_FLOAT_DATA) |
2804 | pack_float(s, s->orig_l, s->orig_r, nb_samples); |
2805 | else |
2806 | pack_int32(s, s->orig_l, s->orig_r, nb_samples); |
2807 | flush_put_bits(&s->pb); |
2808 | data_size = put_bits_count(&s->pb) >> 3; |
2809 | bytestream2_put_le24(&pb, (data_size + 5) >> 1); |
2810 | bytestream2_put_le32(&pb, s->crc_x); |
2811 | bytestream2_skip_p(&pb, data_size); |
2812 | if (data_size & 1) |
2813 | bytestream2_put_byte(&pb, 0); |
2814 | } |
2815 | |
2816 | block_size = bytestream2_tell_p(&pb); |
2817 | AV_WL32(out + 4, block_size - 8); |
2818 | |
2819 | av_assert0(!bytestream2_get_eof(&pb)); |
2820 | |
2821 | return block_size; |
2822 | } |
2823 | |
2824 | static void fill_buffer(WavPackEncodeContext *s, |
2825 | const int8_t *src, int32_t *dst, |
2826 | int nb_samples) |
2827 | { |
2828 | int i; |
2829 | |
2830 | #define COPY_SAMPLES(type, offset, shift) do { \ |
2831 | const type *sptr = (const type *)src; \ |
2832 | for (i = 0; i < nb_samples; i++) \ |
2833 | dst[i] = (sptr[i] - offset) >> shift; \ |
2834 | } while (0) |
2835 | |
2836 | switch (s->avctx->sample_fmt) { |
2837 | case AV_SAMPLE_FMT_U8P: |
2838 | COPY_SAMPLES(int8_t, 0x80, 0); |
2839 | break; |
2840 | case AV_SAMPLE_FMT_S16P: |
2841 | COPY_SAMPLES(int16_t, 0, 0); |
2842 | break; |
2843 | case AV_SAMPLE_FMT_S32P: |
2844 | if (s->avctx->bits_per_raw_sample <= 24) { |
2845 | COPY_SAMPLES(int32_t, 0, 8); |
2846 | break; |
2847 | } |
2848 | case AV_SAMPLE_FMT_FLTP: |
2849 | memcpy(dst, src, nb_samples * 4); |
2850 | } |
2851 | } |
2852 | |
2853 | static void set_samplerate(WavPackEncodeContext *s) |
2854 | { |
2855 | int i; |
2856 | |
2857 | for (i = 0; i < 15; i++) { |
2858 | if (wv_rates[i] == s->avctx->sample_rate) |
2859 | break; |
2860 | } |
2861 | |
2862 | s->flags = i << SRATE_LSB; |
2863 | } |
2864 | |
2865 | static int wavpack_encode_frame(AVCodecContext *avctx, AVPacket *avpkt, |
2866 | const AVFrame *frame, int *got_packet_ptr) |
2867 | { |
2868 | WavPackEncodeContext *s = avctx->priv_data; |
2869 | int buf_size, ret; |
2870 | uint8_t *buf; |
2871 | |
2872 | s->block_samples = frame->nb_samples; |
2873 | av_fast_padded_malloc(&s->samples[0], &s->samples_size[0], |
2874 | sizeof(int32_t) * s->block_samples); |
2875 | if (!s->samples[0]) |
2876 | return AVERROR(ENOMEM); |
2877 | if (avctx->channels > 1) { |
2878 | av_fast_padded_malloc(&s->samples[1], &s->samples_size[1], |
2879 | sizeof(int32_t) * s->block_samples); |
2880 | if (!s->samples[1]) |
2881 | return AVERROR(ENOMEM); |
2882 | } |
2883 | |
2884 | buf_size = s->block_samples * avctx->channels * 8 |
2885 | + 200 * avctx->channels /* for headers */; |
2886 | if ((ret = ff_alloc_packet2(avctx, avpkt, buf_size, 0)) < 0) |
2887 | return ret; |
2888 | buf = avpkt->data; |
2889 | |
2890 | for (s->ch_offset = 0; s->ch_offset < avctx->channels;) { |
2891 | set_samplerate(s); |
2892 | |
2893 | switch (s->avctx->sample_fmt) { |
2894 | case AV_SAMPLE_FMT_S16P: s->flags |= 1; break; |
2895 | case AV_SAMPLE_FMT_S32P: s->flags |= 3 - (s->avctx->bits_per_raw_sample <= 24); break; |
2896 | case AV_SAMPLE_FMT_FLTP: s->flags |= 3 | WV_FLOAT_DATA; |
2897 | } |
2898 | |
2899 | fill_buffer(s, frame->extended_data[s->ch_offset], s->samples[0], s->block_samples); |
2900 | if (avctx->channels - s->ch_offset == 1) { |
2901 | s->flags |= WV_MONO; |
2902 | } else { |
2903 | s->flags |= WV_CROSS_DECORR; |
2904 | fill_buffer(s, frame->extended_data[s->ch_offset + 1], s->samples[1], s->block_samples); |
2905 | } |
2906 | |
2907 | s->flags += (1 << MAG_LSB) * ((s->flags & 3) * 8 + 7); |
2908 | |
2909 | if ((ret = wavpack_encode_block(s, s->samples[0], s->samples[1], |
2910 | buf, buf_size)) < 0) |
2911 | return ret; |
2912 | |
2913 | buf += ret; |
2914 | buf_size -= ret; |
2915 | } |
2916 | s->sample_index += frame->nb_samples; |
2917 | |
2918 | avpkt->pts = frame->pts; |
2919 | avpkt->size = buf - avpkt->data; |
2920 | avpkt->duration = ff_samples_to_time_base(avctx, frame->nb_samples); |
2921 | *got_packet_ptr = 1; |
2922 | return 0; |
2923 | } |
2924 | |
2925 | static av_cold int wavpack_encode_close(AVCodecContext *avctx) |
2926 | { |
2927 | WavPackEncodeContext *s = avctx->priv_data; |
2928 | int i; |
2929 | |
2930 | for (i = 0; i < MAX_TERMS + 2; i++) { |
2931 | av_freep(&s->sampleptrs[i][0]); |
2932 | av_freep(&s->sampleptrs[i][1]); |
2933 | s->sampleptrs_size[i][0] = s->sampleptrs_size[i][1] = 0; |
2934 | } |
2935 | |
2936 | for (i = 0; i < 2; i++) { |
2937 | av_freep(&s->samples[i]); |
2938 | s->samples_size[i] = 0; |
2939 | |
2940 | av_freep(&s->best_buffer[i]); |
2941 | s->best_buffer_size[i] = 0; |
2942 | |
2943 | av_freep(&s->temp_buffer[i][0]); |
2944 | av_freep(&s->temp_buffer[i][1]); |
2945 | s->temp_buffer_size[i][0] = s->temp_buffer_size[i][1] = 0; |
2946 | } |
2947 | |
2948 | av_freep(&s->js_left); |
2949 | av_freep(&s->js_right); |
2950 | s->js_left_size = s->js_right_size = 0; |
2951 | |
2952 | av_freep(&s->orig_l); |
2953 | av_freep(&s->orig_r); |
2954 | s->orig_l_size = s->orig_r_size = 0; |
2955 | |
2956 | return 0; |
2957 | } |
2958 | |
2959 | #define OFFSET(x) offsetof(WavPackEncodeContext, x) |
2960 | #define FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM |
2961 | static const AVOption options[] = { |
2962 | { "joint_stereo", "", OFFSET(joint), AV_OPT_TYPE_BOOL, {.i64=-1}, -1, 1, FLAGS }, |
2963 | { "optimize_mono", "", OFFSET(optimize_mono), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS }, |
2964 | { NULL }, |
2965 | }; |
2966 | |
2967 | static const AVClass wavpack_encoder_class = { |
2968 | .class_name = "WavPack encoder", |
2969 | .item_name = av_default_item_name, |
2970 | .option = options, |
2971 | .version = LIBAVUTIL_VERSION_INT, |
2972 | }; |
2973 | |
2974 | AVCodec ff_wavpack_encoder = { |
2975 | .name = "wavpack", |
2976 | .long_name = NULL_IF_CONFIG_SMALL("WavPack"), |
2977 | .type = AVMEDIA_TYPE_AUDIO, |
2978 | .id = AV_CODEC_ID_WAVPACK, |
2979 | .priv_data_size = sizeof(WavPackEncodeContext), |
2980 | .priv_class = &wavpack_encoder_class, |
2981 | .init = wavpack_encode_init, |
2982 | .encode2 = wavpack_encode_frame, |
2983 | .close = wavpack_encode_close, |
2984 | .capabilities = AV_CODEC_CAP_SMALL_LAST_FRAME, |
2985 | .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_U8P, |
2986 | AV_SAMPLE_FMT_S16P, |
2987 | AV_SAMPLE_FMT_S32P, |
2988 | AV_SAMPLE_FMT_FLTP, |
2989 | AV_SAMPLE_FMT_NONE }, |
2990 | }; |
2991 |