1 /*
2 * Copyright (c) 2013
3 * MIPS Technologies, Inc., California.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its
14 * contributors may be used to endorse or promote products derived from
15 * this software without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * AAC decoder fixed-point implementation
30 *
31 * Copyright (c) 2005-2006 Oded Shimon ( ods15 ods15 dyndns org )
32 * Copyright (c) 2006-2007 Maxim Gavrilov ( maxim.gavrilov gmail com )
33 *
34 * This file is part of FFmpeg.
35 *
36 * FFmpeg is free software; you can redistribute it and/or
37 * modify it under the terms of the GNU Lesser General Public
38 * License as published by the Free Software Foundation; either
39 * version 2.1 of the License, or (at your option) any later version.
40 *
41 * FFmpeg is distributed in the hope that it will be useful,
42 * but WITHOUT ANY WARRANTY; without even the implied warranty of
43 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
44 * Lesser General Public License for more details.
45 *
46 * You should have received a copy of the GNU Lesser General Public
47 * License along with FFmpeg; if not, write to the Free Software
48 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
49 */
50
51 /**
52 * @file
53 * AAC decoder
54 * @author Oded Shimon ( ods15 ods15 dyndns org )
55 * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
56 *
57 * Fixed point implementation
58 * @author Stanislav Ocovaj ( stanislav.ocovaj imgtec com )
59 */
60
61 #define FFT_FLOAT 0
62 #define FFT_FIXED_32 1
63 #define USE_FIXED 1
64
65 #include "libavutil/fixed_dsp.h"
66 #include "libavutil/opt.h"
67 #include "avcodec.h"
68 #include "internal.h"
69 #include "get_bits.h"
70 #include "fft.h"
71 #include "lpc.h"
72 #include "kbdwin.h"
73 #include "sinewin.h"
74
75 #include "aac.h"
76 #include "aactab.h"
77 #include "aacdectab.h"
78 #include "adts_header.h"
79 #include "cbrt_data.h"
80 #include "sbr.h"
81 #include "aacsbr.h"
82 #include "mpeg4audio.h"
83 #include "profiles.h"
84 #include "libavutil/intfloat.h"
85
86 #include <math.h>
87 #include <string.h>
88
reset_predict_state(PredictorState * ps)89 static av_always_inline void reset_predict_state(PredictorState *ps)
90 {
91 ps->r0.mant = 0;
92 ps->r0.exp = 0;
93 ps->r1.mant = 0;
94 ps->r1.exp = 0;
95 ps->cor0.mant = 0;
96 ps->cor0.exp = 0;
97 ps->cor1.mant = 0;
98 ps->cor1.exp = 0;
99 ps->var0.mant = 0x20000000;
100 ps->var0.exp = 1;
101 ps->var1.mant = 0x20000000;
102 ps->var1.exp = 1;
103 }
104
105 static const int exp2tab[4] = { Q31(1.0000000000/2), Q31(1.1892071150/2), Q31(1.4142135624/2), Q31(1.6817928305/2) }; // 2^0, 2^0.25, 2^0.5, 2^0.75
106
DEC_SPAIR(int * dst,unsigned idx)107 static inline int *DEC_SPAIR(int *dst, unsigned idx)
108 {
109 dst[0] = (idx & 15) - 4;
110 dst[1] = (idx >> 4 & 15) - 4;
111
112 return dst + 2;
113 }
114
DEC_SQUAD(int * dst,unsigned idx)115 static inline int *DEC_SQUAD(int *dst, unsigned idx)
116 {
117 dst[0] = (idx & 3) - 1;
118 dst[1] = (idx >> 2 & 3) - 1;
119 dst[2] = (idx >> 4 & 3) - 1;
120 dst[3] = (idx >> 6 & 3) - 1;
121
122 return dst + 4;
123 }
124
DEC_UPAIR(int * dst,unsigned idx,unsigned sign)125 static inline int *DEC_UPAIR(int *dst, unsigned idx, unsigned sign)
126 {
127 dst[0] = (idx & 15) * (1 - (sign & 0xFFFFFFFE));
128 dst[1] = (idx >> 4 & 15) * (1 - ((sign & 1) * 2));
129
130 return dst + 2;
131 }
132
DEC_UQUAD(int * dst,unsigned idx,unsigned sign)133 static inline int *DEC_UQUAD(int *dst, unsigned idx, unsigned sign)
134 {
135 unsigned nz = idx >> 12;
136
137 dst[0] = (idx & 3) * (1 + (((int)sign >> 31) * 2));
138 sign <<= nz & 1;
139 nz >>= 1;
140 dst[1] = (idx >> 2 & 3) * (1 + (((int)sign >> 31) * 2));
141 sign <<= nz & 1;
142 nz >>= 1;
143 dst[2] = (idx >> 4 & 3) * (1 + (((int)sign >> 31) * 2));
144 sign <<= nz & 1;
145 nz >>= 1;
146 dst[3] = (idx >> 6 & 3) * (1 + (((int)sign >> 31) * 2));
147
148 return dst + 4;
149 }
150
vector_pow43(int * coefs,int len)151 static void vector_pow43(int *coefs, int len)
152 {
153 int i, coef;
154
155 for (i=0; i<len; i++) {
156 coef = coefs[i];
157 if (coef < 0)
158 coef = -(int)ff_cbrt_tab_fixed[(-coef) & 8191];
159 else
160 coef = (int)ff_cbrt_tab_fixed[ coef & 8191];
161 coefs[i] = coef;
162 }
163 }
164
subband_scale(int * dst,int * src,int scale,int offset,int len,void * log_context)165 static void subband_scale(int *dst, int *src, int scale, int offset, int len, void *log_context)
166 {
167 int ssign = scale < 0 ? -1 : 1;
168 int s = FFABS(scale);
169 unsigned int round;
170 int i, out, c = exp2tab[s & 3];
171
172 s = offset - (s >> 2);
173
174 if (s > 31) {
175 for (i=0; i<len; i++) {
176 dst[i] = 0;
177 }
178 } else if (s > 0) {
179 round = 1 << (s-1);
180 for (i=0; i<len; i++) {
181 out = (int)(((int64_t)src[i] * c) >> 32);
182 dst[i] = ((int)(out+round) >> s) * ssign;
183 }
184 } else if (s > -32) {
185 s = s + 32;
186 round = 1U << (s-1);
187 for (i=0; i<len; i++) {
188 out = (int)((int64_t)((int64_t)src[i] * c + round) >> s);
189 dst[i] = out * (unsigned)ssign;
190 }
191 } else {
192 av_log(log_context, AV_LOG_ERROR, "Overflow in subband_scale()\n");
193 }
194 }
195
noise_scale(int * coefs,int scale,int band_energy,int len)196 static void noise_scale(int *coefs, int scale, int band_energy, int len)
197 {
198 int s = -scale;
199 unsigned int round;
200 int i, out, c = exp2tab[s & 3];
201 int nlz = 0;
202
203 av_assert0(s >= 0);
204 while (band_energy > 0x7fff) {
205 band_energy >>= 1;
206 nlz++;
207 }
208 c /= band_energy;
209 s = 21 + nlz - (s >> 2);
210
211 if (s > 31) {
212 for (i=0; i<len; i++) {
213 coefs[i] = 0;
214 }
215 } else if (s >= 0) {
216 round = s ? 1 << (s-1) : 0;
217 for (i=0; i<len; i++) {
218 out = (int)(((int64_t)coefs[i] * c) >> 32);
219 coefs[i] = -((int)(out+round) >> s);
220 }
221 }
222 else {
223 s = s + 32;
224 if (s > 0) {
225 round = 1 << (s-1);
226 for (i=0; i<len; i++) {
227 out = (int)((int64_t)((int64_t)coefs[i] * c + round) >> s);
228 coefs[i] = -out;
229 }
230 } else {
231 for (i=0; i<len; i++)
232 coefs[i] = -(int64_t)coefs[i] * c * (1 << -s);
233 }
234 }
235 }
236
flt16_round(SoftFloat pf)237 static av_always_inline SoftFloat flt16_round(SoftFloat pf)
238 {
239 SoftFloat tmp;
240 int s;
241
242 tmp.exp = pf.exp;
243 s = pf.mant >> 31;
244 tmp.mant = (pf.mant ^ s) - s;
245 tmp.mant = (tmp.mant + 0x00200000U) & 0xFFC00000U;
246 tmp.mant = (tmp.mant ^ s) - s;
247
248 return tmp;
249 }
250
flt16_even(SoftFloat pf)251 static av_always_inline SoftFloat flt16_even(SoftFloat pf)
252 {
253 SoftFloat tmp;
254 int s;
255
256 tmp.exp = pf.exp;
257 s = pf.mant >> 31;
258 tmp.mant = (pf.mant ^ s) - s;
259 tmp.mant = (tmp.mant + 0x001FFFFFU + (tmp.mant & 0x00400000U >> 16)) & 0xFFC00000U;
260 tmp.mant = (tmp.mant ^ s) - s;
261
262 return tmp;
263 }
264
flt16_trunc(SoftFloat pf)265 static av_always_inline SoftFloat flt16_trunc(SoftFloat pf)
266 {
267 SoftFloat pun;
268 int s;
269
270 pun.exp = pf.exp;
271 s = pf.mant >> 31;
272 pun.mant = (pf.mant ^ s) - s;
273 pun.mant = pun.mant & 0xFFC00000U;
274 pun.mant = (pun.mant ^ s) - s;
275
276 return pun;
277 }
278
predict(PredictorState * ps,int * coef,int output_enable)279 static av_always_inline void predict(PredictorState *ps, int *coef,
280 int output_enable)
281 {
282 const SoftFloat a = { 1023410176, 0 }; // 61.0 / 64
283 const SoftFloat alpha = { 973078528, 0 }; // 29.0 / 32
284 SoftFloat e0, e1;
285 SoftFloat pv;
286 SoftFloat k1, k2;
287 SoftFloat r0 = ps->r0, r1 = ps->r1;
288 SoftFloat cor0 = ps->cor0, cor1 = ps->cor1;
289 SoftFloat var0 = ps->var0, var1 = ps->var1;
290 SoftFloat tmp;
291
292 if (var0.exp > 1 || (var0.exp == 1 && var0.mant > 0x20000000)) {
293 k1 = av_mul_sf(cor0, flt16_even(av_div_sf(a, var0)));
294 }
295 else {
296 k1.mant = 0;
297 k1.exp = 0;
298 }
299
300 if (var1.exp > 1 || (var1.exp == 1 && var1.mant > 0x20000000)) {
301 k2 = av_mul_sf(cor1, flt16_even(av_div_sf(a, var1)));
302 }
303 else {
304 k2.mant = 0;
305 k2.exp = 0;
306 }
307
308 tmp = av_mul_sf(k1, r0);
309 pv = flt16_round(av_add_sf(tmp, av_mul_sf(k2, r1)));
310 if (output_enable) {
311 int shift = 28 - pv.exp;
312
313 if (shift < 31) {
314 if (shift > 0) {
315 *coef += (unsigned)((pv.mant + (1 << (shift - 1))) >> shift);
316 } else
317 *coef += (unsigned)pv.mant << -shift;
318 }
319 }
320
321 e0 = av_int2sf(*coef, 2);
322 e1 = av_sub_sf(e0, tmp);
323
324 ps->cor1 = flt16_trunc(av_add_sf(av_mul_sf(alpha, cor1), av_mul_sf(r1, e1)));
325 tmp = av_add_sf(av_mul_sf(r1, r1), av_mul_sf(e1, e1));
326 tmp.exp--;
327 ps->var1 = flt16_trunc(av_add_sf(av_mul_sf(alpha, var1), tmp));
328 ps->cor0 = flt16_trunc(av_add_sf(av_mul_sf(alpha, cor0), av_mul_sf(r0, e0)));
329 tmp = av_add_sf(av_mul_sf(r0, r0), av_mul_sf(e0, e0));
330 tmp.exp--;
331 ps->var0 = flt16_trunc(av_add_sf(av_mul_sf(alpha, var0), tmp));
332
333 ps->r1 = flt16_trunc(av_mul_sf(a, av_sub_sf(r0, av_mul_sf(k1, e0))));
334 ps->r0 = flt16_trunc(av_mul_sf(a, e0));
335 }
336
337
338 static const int cce_scale_fixed[8] = {
339 Q30(1.0), //2^(0/8)
340 Q30(1.0905077327), //2^(1/8)
341 Q30(1.1892071150), //2^(2/8)
342 Q30(1.2968395547), //2^(3/8)
343 Q30(1.4142135624), //2^(4/8)
344 Q30(1.5422108254), //2^(5/8)
345 Q30(1.6817928305), //2^(6/8)
346 Q30(1.8340080864), //2^(7/8)
347 };
348
349 /**
350 * Apply dependent channel coupling (applied before IMDCT).
351 *
352 * @param index index into coupling gain array
353 */
apply_dependent_coupling_fixed(AACContext * ac,SingleChannelElement * target,ChannelElement * cce,int index)354 static void apply_dependent_coupling_fixed(AACContext *ac,
355 SingleChannelElement *target,
356 ChannelElement *cce, int index)
357 {
358 IndividualChannelStream *ics = &cce->ch[0].ics;
359 const uint16_t *offsets = ics->swb_offset;
360 int *dest = target->coeffs;
361 const int *src = cce->ch[0].coeffs;
362 int g, i, group, k, idx = 0;
363 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
364 av_log(ac->avctx, AV_LOG_ERROR,
365 "Dependent coupling is not supported together with LTP\n");
366 return;
367 }
368 for (g = 0; g < ics->num_window_groups; g++) {
369 for (i = 0; i < ics->max_sfb; i++, idx++) {
370 if (cce->ch[0].band_type[idx] != ZERO_BT) {
371 const int gain = cce->coup.gain[index][idx];
372 int shift, round, c, tmp;
373
374 if (gain < 0) {
375 c = -cce_scale_fixed[-gain & 7];
376 shift = (-gain-1024) >> 3;
377 }
378 else {
379 c = cce_scale_fixed[gain & 7];
380 shift = (gain-1024) >> 3;
381 }
382
383 if (shift < -31) {
384 // Nothing to do
385 } else if (shift < 0) {
386 shift = -shift;
387 round = 1 << (shift - 1);
388
389 for (group = 0; group < ics->group_len[g]; group++) {
390 for (k = offsets[i]; k < offsets[i + 1]; k++) {
391 tmp = (int)(((int64_t)src[group * 128 + k] * c + \
392 (int64_t)0x1000000000) >> 37);
393 dest[group * 128 + k] += (tmp + (int64_t)round) >> shift;
394 }
395 }
396 }
397 else {
398 for (group = 0; group < ics->group_len[g]; group++) {
399 for (k = offsets[i]; k < offsets[i + 1]; k++) {
400 tmp = (int)(((int64_t)src[group * 128 + k] * c + \
401 (int64_t)0x1000000000) >> 37);
402 dest[group * 128 + k] += tmp * (1U << shift);
403 }
404 }
405 }
406 }
407 }
408 dest += ics->group_len[g] * 128;
409 src += ics->group_len[g] * 128;
410 }
411 }
412
413 /**
414 * Apply independent channel coupling (applied after IMDCT).
415 *
416 * @param index index into coupling gain array
417 */
apply_independent_coupling_fixed(AACContext * ac,SingleChannelElement * target,ChannelElement * cce,int index)418 static void apply_independent_coupling_fixed(AACContext *ac,
419 SingleChannelElement *target,
420 ChannelElement *cce, int index)
421 {
422 int i, c, shift, round, tmp;
423 const int gain = cce->coup.gain[index][0];
424 const int *src = cce->ch[0].ret;
425 unsigned int *dest = target->ret;
426 const int len = 1024 << (ac->oc[1].m4ac.sbr == 1);
427
428 c = cce_scale_fixed[gain & 7];
429 shift = (gain-1024) >> 3;
430 if (shift < -31) {
431 return;
432 } else if (shift < 0) {
433 shift = -shift;
434 round = 1 << (shift - 1);
435
436 for (i = 0; i < len; i++) {
437 tmp = (int)(((int64_t)src[i] * c + (int64_t)0x1000000000) >> 37);
438 dest[i] += (tmp + round) >> shift;
439 }
440 }
441 else {
442 for (i = 0; i < len; i++) {
443 tmp = (int)(((int64_t)src[i] * c + (int64_t)0x1000000000) >> 37);
444 dest[i] += tmp * (1U << shift);
445 }
446 }
447 }
448
449 #include "aacdec_template.c"
450
451 AVCodec ff_aac_fixed_decoder = {
452 .name = "aac_fixed",
453 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
454 .type = AVMEDIA_TYPE_AUDIO,
455 .id = AV_CODEC_ID_AAC,
456 .priv_data_size = sizeof(AACContext),
457 .init = aac_decode_init,
458 .close = aac_decode_close,
459 .decode = aac_decode_frame,
460 .sample_fmts = (const enum AVSampleFormat[]) {
461 AV_SAMPLE_FMT_S32P, AV_SAMPLE_FMT_NONE
462 },
463 .capabilities = AV_CODEC_CAP_CHANNEL_CONF | AV_CODEC_CAP_DR1,
464 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP,
465 .channel_layouts = aac_channel_layout,
466 .profiles = NULL_IF_CONFIG_SMALL(ff_aac_profiles),
467 .flush = flush,
468 };
469