blob: acb8178337de305cec811c7db3a03a421b33f615
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 "cbrt_data.h" |
79 | #include "sbr.h" |
80 | #include "aacsbr.h" |
81 | #include "mpeg4audio.h" |
82 | #include "aacadtsdec.h" |
83 | #include "profiles.h" |
84 | #include "libavutil/intfloat.h" |
85 | |
86 | #include <math.h> |
87 | #include <string.h> |
88 | |
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 | |
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 | |
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 | |
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) << 1)); |
129 | |
130 | return dst + 2; |
131 | } |
132 | |
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) << 1)); |
138 | sign <<= nz & 1; |
139 | nz >>= 1; |
140 | dst[1] = (idx >> 2 & 3) * (1 + (((int)sign >> 31) << 1)); |
141 | sign <<= nz & 1; |
142 | nz >>= 1; |
143 | dst[2] = (idx >> 4 & 3) * (1 + (((int)sign >> 31) << 1)); |
144 | sign <<= nz & 1; |
145 | nz >>= 1; |
146 | dst[3] = (idx >> 6 & 3) * (1 + (((int)sign >> 31) << 1)); |
147 | |
148 | return dst + 4; |
149 | } |
150 | |
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]; |
159 | else |
160 | coef = (int)ff_cbrt_tab_fixed[coef]; |
161 | coefs[i] = coef; |
162 | } |
163 | } |
164 | |
165 | static void subband_scale(int *dst, int *src, int scale, int offset, int len) |
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 > 0) { |
175 | round = 1 << (s-1); |
176 | for (i=0; i<len; i++) { |
177 | out = (int)(((int64_t)src[i] * c) >> 32); |
178 | dst[i] = ((int)(out+round) >> s) * ssign; |
179 | } |
180 | } |
181 | else { |
182 | s = s + 32; |
183 | round = 1 << (s-1); |
184 | for (i=0; i<len; i++) { |
185 | out = (int)((int64_t)((int64_t)src[i] * c + round) >> s); |
186 | dst[i] = out * ssign; |
187 | } |
188 | } |
189 | } |
190 | |
191 | static void noise_scale(int *coefs, int scale, int band_energy, int len) |
192 | { |
193 | int ssign = scale < 0 ? -1 : 1; |
194 | int s = FFABS(scale); |
195 | unsigned int round; |
196 | int i, out, c = exp2tab[s & 3]; |
197 | int nlz = 0; |
198 | |
199 | while (band_energy > 0x7fff) { |
200 | band_energy >>= 1; |
201 | nlz++; |
202 | } |
203 | c /= band_energy; |
204 | s = 21 + nlz - (s >> 2); |
205 | |
206 | if (s > 0) { |
207 | round = 1 << (s-1); |
208 | for (i=0; i<len; i++) { |
209 | out = (int)(((int64_t)coefs[i] * c) >> 32); |
210 | coefs[i] = ((int)(out+round) >> s) * ssign; |
211 | } |
212 | } |
213 | else { |
214 | s = s + 32; |
215 | round = 1 << (s-1); |
216 | for (i=0; i<len; i++) { |
217 | out = (int)((int64_t)((int64_t)coefs[i] * c + round) >> s); |
218 | coefs[i] = out * ssign; |
219 | } |
220 | } |
221 | } |
222 | |
223 | static av_always_inline SoftFloat flt16_round(SoftFloat pf) |
224 | { |
225 | SoftFloat tmp; |
226 | int s; |
227 | |
228 | tmp.exp = pf.exp; |
229 | s = pf.mant >> 31; |
230 | tmp.mant = (pf.mant ^ s) - s; |
231 | tmp.mant = (tmp.mant + 0x00200000U) & 0xFFC00000U; |
232 | tmp.mant = (tmp.mant ^ s) - s; |
233 | |
234 | return tmp; |
235 | } |
236 | |
237 | static av_always_inline SoftFloat flt16_even(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 + 0x001FFFFFU + (tmp.mant & 0x00400000U >> 16)) & 0xFFC00000U; |
246 | tmp.mant = (tmp.mant ^ s) - s; |
247 | |
248 | return tmp; |
249 | } |
250 | |
251 | static av_always_inline SoftFloat flt16_trunc(SoftFloat pf) |
252 | { |
253 | SoftFloat pun; |
254 | int s; |
255 | |
256 | pun.exp = pf.exp; |
257 | s = pf.mant >> 31; |
258 | pun.mant = (pf.mant ^ s) - s; |
259 | pun.mant = pun.mant & 0xFFC00000U; |
260 | pun.mant = (pun.mant ^ s) - s; |
261 | |
262 | return pun; |
263 | } |
264 | |
265 | static av_always_inline void predict(PredictorState *ps, int *coef, |
266 | int output_enable) |
267 | { |
268 | const SoftFloat a = { 1023410176, 0 }; // 61.0 / 64 |
269 | const SoftFloat alpha = { 973078528, 0 }; // 29.0 / 32 |
270 | SoftFloat e0, e1; |
271 | SoftFloat pv; |
272 | SoftFloat k1, k2; |
273 | SoftFloat r0 = ps->r0, r1 = ps->r1; |
274 | SoftFloat cor0 = ps->cor0, cor1 = ps->cor1; |
275 | SoftFloat var0 = ps->var0, var1 = ps->var1; |
276 | SoftFloat tmp; |
277 | |
278 | if (var0.exp > 1 || (var0.exp == 1 && var0.mant > 0x20000000)) { |
279 | k1 = av_mul_sf(cor0, flt16_even(av_div_sf(a, var0))); |
280 | } |
281 | else { |
282 | k1.mant = 0; |
283 | k1.exp = 0; |
284 | } |
285 | |
286 | if (var1.exp > 1 || (var1.exp == 1 && var1.mant > 0x20000000)) { |
287 | k2 = av_mul_sf(cor1, flt16_even(av_div_sf(a, var1))); |
288 | } |
289 | else { |
290 | k2.mant = 0; |
291 | k2.exp = 0; |
292 | } |
293 | |
294 | tmp = av_mul_sf(k1, r0); |
295 | pv = flt16_round(av_add_sf(tmp, av_mul_sf(k2, r1))); |
296 | if (output_enable) { |
297 | int shift = 28 - pv.exp; |
298 | |
299 | if (shift < 31) |
300 | *coef += (pv.mant + (1 << (shift - 1))) >> shift; |
301 | } |
302 | |
303 | e0 = av_int2sf(*coef, 2); |
304 | e1 = av_sub_sf(e0, tmp); |
305 | |
306 | ps->cor1 = flt16_trunc(av_add_sf(av_mul_sf(alpha, cor1), av_mul_sf(r1, e1))); |
307 | tmp = av_add_sf(av_mul_sf(r1, r1), av_mul_sf(e1, e1)); |
308 | tmp.exp--; |
309 | ps->var1 = flt16_trunc(av_add_sf(av_mul_sf(alpha, var1), tmp)); |
310 | ps->cor0 = flt16_trunc(av_add_sf(av_mul_sf(alpha, cor0), av_mul_sf(r0, e0))); |
311 | tmp = av_add_sf(av_mul_sf(r0, r0), av_mul_sf(e0, e0)); |
312 | tmp.exp--; |
313 | ps->var0 = flt16_trunc(av_add_sf(av_mul_sf(alpha, var0), tmp)); |
314 | |
315 | ps->r1 = flt16_trunc(av_mul_sf(a, av_sub_sf(r0, av_mul_sf(k1, e0)))); |
316 | ps->r0 = flt16_trunc(av_mul_sf(a, e0)); |
317 | } |
318 | |
319 | |
320 | static const int cce_scale_fixed[8] = { |
321 | Q30(1.0), //2^(0/8) |
322 | Q30(1.0905077327), //2^(1/8) |
323 | Q30(1.1892071150), //2^(2/8) |
324 | Q30(1.2968395547), //2^(3/8) |
325 | Q30(1.4142135624), //2^(4/8) |
326 | Q30(1.5422108254), //2^(5/8) |
327 | Q30(1.6817928305), //2^(6/8) |
328 | Q30(1.8340080864), //2^(7/8) |
329 | }; |
330 | |
331 | /** |
332 | * Apply dependent channel coupling (applied before IMDCT). |
333 | * |
334 | * @param index index into coupling gain array |
335 | */ |
336 | static void apply_dependent_coupling_fixed(AACContext *ac, |
337 | SingleChannelElement *target, |
338 | ChannelElement *cce, int index) |
339 | { |
340 | IndividualChannelStream *ics = &cce->ch[0].ics; |
341 | const uint16_t *offsets = ics->swb_offset; |
342 | int *dest = target->coeffs; |
343 | const int *src = cce->ch[0].coeffs; |
344 | int g, i, group, k, idx = 0; |
345 | if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) { |
346 | av_log(ac->avctx, AV_LOG_ERROR, |
347 | "Dependent coupling is not supported together with LTP\n"); |
348 | return; |
349 | } |
350 | for (g = 0; g < ics->num_window_groups; g++) { |
351 | for (i = 0; i < ics->max_sfb; i++, idx++) { |
352 | if (cce->ch[0].band_type[idx] != ZERO_BT) { |
353 | const int gain = cce->coup.gain[index][idx]; |
354 | int shift, round, c, tmp; |
355 | |
356 | if (gain < 0) { |
357 | c = -cce_scale_fixed[-gain & 7]; |
358 | shift = (-gain-1024) >> 3; |
359 | } |
360 | else { |
361 | c = cce_scale_fixed[gain & 7]; |
362 | shift = (gain-1024) >> 3; |
363 | } |
364 | |
365 | if (shift < 0) { |
366 | shift = -shift; |
367 | round = 1 << (shift - 1); |
368 | |
369 | for (group = 0; group < ics->group_len[g]; group++) { |
370 | for (k = offsets[i]; k < offsets[i + 1]; k++) { |
371 | tmp = (int)(((int64_t)src[group * 128 + k] * c + \ |
372 | (int64_t)0x1000000000) >> 37); |
373 | dest[group * 128 + k] += (tmp + round) >> shift; |
374 | } |
375 | } |
376 | } |
377 | else { |
378 | for (group = 0; group < ics->group_len[g]; group++) { |
379 | for (k = offsets[i]; k < offsets[i + 1]; k++) { |
380 | tmp = (int)(((int64_t)src[group * 128 + k] * c + \ |
381 | (int64_t)0x1000000000) >> 37); |
382 | dest[group * 128 + k] += tmp << shift; |
383 | } |
384 | } |
385 | } |
386 | } |
387 | } |
388 | dest += ics->group_len[g] * 128; |
389 | src += ics->group_len[g] * 128; |
390 | } |
391 | } |
392 | |
393 | /** |
394 | * Apply independent channel coupling (applied after IMDCT). |
395 | * |
396 | * @param index index into coupling gain array |
397 | */ |
398 | static void apply_independent_coupling_fixed(AACContext *ac, |
399 | SingleChannelElement *target, |
400 | ChannelElement *cce, int index) |
401 | { |
402 | int i, c, shift, round, tmp; |
403 | const int gain = cce->coup.gain[index][0]; |
404 | const int *src = cce->ch[0].ret; |
405 | int *dest = target->ret; |
406 | const int len = 1024 << (ac->oc[1].m4ac.sbr == 1); |
407 | |
408 | c = cce_scale_fixed[gain & 7]; |
409 | shift = (gain-1024) >> 3; |
410 | if (shift < 0) { |
411 | shift = -shift; |
412 | round = 1 << (shift - 1); |
413 | |
414 | for (i = 0; i < len; i++) { |
415 | tmp = (int)(((int64_t)src[i] * c + (int64_t)0x1000000000) >> 37); |
416 | dest[i] += (tmp + round) >> shift; |
417 | } |
418 | } |
419 | else { |
420 | for (i = 0; i < len; i++) { |
421 | tmp = (int)(((int64_t)src[i] * c + (int64_t)0x1000000000) >> 37); |
422 | dest[i] += tmp << shift; |
423 | } |
424 | } |
425 | } |
426 | |
427 | #include "aacdec_template.c" |
428 | |
429 | AVCodec ff_aac_fixed_decoder = { |
430 | .name = "aac_fixed", |
431 | .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"), |
432 | .type = AVMEDIA_TYPE_AUDIO, |
433 | .id = AV_CODEC_ID_AAC, |
434 | .priv_data_size = sizeof(AACContext), |
435 | .init = aac_decode_init, |
436 | .close = aac_decode_close, |
437 | .decode = aac_decode_frame, |
438 | .sample_fmts = (const enum AVSampleFormat[]) { |
439 | AV_SAMPLE_FMT_S32P, AV_SAMPLE_FMT_NONE |
440 | }, |
441 | .capabilities = AV_CODEC_CAP_CHANNEL_CONF | AV_CODEC_CAP_DR1, |
442 | .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE, |
443 | .channel_layouts = aac_channel_layout, |
444 | .profiles = NULL_IF_CONFIG_SMALL(ff_aac_profiles), |
445 | .flush = flush, |
446 | }; |
447 |