• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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_fixed_tablegen.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 
89 DECLARE_ALIGNED(32, static int, AAC_RENAME2(aac_kbd_long_1024))[1024];
90 DECLARE_ALIGNED(32, static int, AAC_RENAME2(aac_kbd_short_128))[128];
91 
reset_predict_state(PredictorState * ps)92 static av_always_inline void reset_predict_state(PredictorState *ps)
93 {
94     ps->r0.mant   = 0;
95     ps->r0.exp   = 0;
96     ps->r1.mant   = 0;
97     ps->r1.exp   = 0;
98     ps->cor0.mant = 0;
99     ps->cor0.exp = 0;
100     ps->cor1.mant = 0;
101     ps->cor1.exp = 0;
102     ps->var0.mant = 0x20000000;
103     ps->var0.exp = 1;
104     ps->var1.mant = 0x20000000;
105     ps->var1.exp = 1;
106 }
107 
108 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
109 
DEC_SPAIR(int * dst,unsigned idx)110 static inline int *DEC_SPAIR(int *dst, unsigned idx)
111 {
112     dst[0] = (idx & 15) - 4;
113     dst[1] = (idx >> 4 & 15) - 4;
114 
115     return dst + 2;
116 }
117 
DEC_SQUAD(int * dst,unsigned idx)118 static inline int *DEC_SQUAD(int *dst, unsigned idx)
119 {
120     dst[0] = (idx & 3) - 1;
121     dst[1] = (idx >> 2 & 3) - 1;
122     dst[2] = (idx >> 4 & 3) - 1;
123     dst[3] = (idx >> 6 & 3) - 1;
124 
125     return dst + 4;
126 }
127 
DEC_UPAIR(int * dst,unsigned idx,unsigned sign)128 static inline int *DEC_UPAIR(int *dst, unsigned idx, unsigned sign)
129 {
130     dst[0] = (idx & 15) * (1 - (sign & 0xFFFFFFFE));
131     dst[1] = (idx >> 4 & 15) * (1 - ((sign & 1) * 2));
132 
133     return dst + 2;
134 }
135 
DEC_UQUAD(int * dst,unsigned idx,unsigned sign)136 static inline int *DEC_UQUAD(int *dst, unsigned idx, unsigned sign)
137 {
138     unsigned nz = idx >> 12;
139 
140     dst[0] = (idx & 3) * (1 + (((int)sign >> 31) * 2));
141     sign <<= nz & 1;
142     nz >>= 1;
143     dst[1] = (idx >> 2 & 3) * (1 + (((int)sign >> 31) * 2));
144     sign <<= nz & 1;
145     nz >>= 1;
146     dst[2] = (idx >> 4 & 3) * (1 + (((int)sign >> 31) * 2));
147     sign <<= nz & 1;
148     nz >>= 1;
149     dst[3] = (idx >> 6 & 3) * (1 + (((int)sign >> 31) * 2));
150 
151     return dst + 4;
152 }
153 
vector_pow43(int * coefs,int len)154 static void vector_pow43(int *coefs, int len)
155 {
156     int i, coef;
157 
158     for (i=0; i<len; i++) {
159         coef = coefs[i];
160         if (coef < 0)
161             coef = -(int)ff_cbrt_tab_fixed[(-coef) & 8191];
162         else
163             coef =  (int)ff_cbrt_tab_fixed[  coef  & 8191];
164         coefs[i] = coef;
165     }
166 }
167 
subband_scale(int * dst,int * src,int scale,int offset,int len,void * log_context)168 static void subband_scale(int *dst, int *src, int scale, int offset, int len, void *log_context)
169 {
170     int ssign = scale < 0 ? -1 : 1;
171     int s = FFABS(scale);
172     unsigned int round;
173     int i, out, c = exp2tab[s & 3];
174 
175     s = offset - (s >> 2);
176 
177     if (s > 31) {
178         for (i=0; i<len; i++) {
179             dst[i] = 0;
180         }
181     } else if (s > 0) {
182         round = 1 << (s-1);
183         for (i=0; i<len; i++) {
184             out = (int)(((int64_t)src[i] * c) >> 32);
185             dst[i] = ((int)(out+round) >> s) * ssign;
186         }
187     } else if (s > -32) {
188         s = s + 32;
189         round = 1U << (s-1);
190         for (i=0; i<len; i++) {
191             out = (int)((int64_t)((int64_t)src[i] * c + round) >> s);
192             dst[i] = out * (unsigned)ssign;
193         }
194     } else {
195         av_log(log_context, AV_LOG_ERROR, "Overflow in subband_scale()\n");
196     }
197 }
198 
noise_scale(int * coefs,int scale,int band_energy,int len)199 static void noise_scale(int *coefs, int scale, int band_energy, int len)
200 {
201     int s = -scale;
202     unsigned int round;
203     int i, out, c = exp2tab[s & 3];
204     int nlz = 0;
205 
206     av_assert0(s >= 0);
207     while (band_energy > 0x7fff) {
208         band_energy >>= 1;
209         nlz++;
210     }
211     c /= band_energy;
212     s = 21 + nlz - (s >> 2);
213 
214     if (s > 31) {
215         for (i=0; i<len; i++) {
216             coefs[i] = 0;
217         }
218     } else if (s >= 0) {
219         round = s ? 1 << (s-1) : 0;
220         for (i=0; i<len; i++) {
221             out = (int)(((int64_t)coefs[i] * c) >> 32);
222             coefs[i] = -((int)(out+round) >> s);
223         }
224     }
225     else {
226         s = s + 32;
227         if (s > 0) {
228             round = 1 << (s-1);
229             for (i=0; i<len; i++) {
230                 out = (int)((int64_t)((int64_t)coefs[i] * c + round) >> s);
231                 coefs[i] = -out;
232             }
233         } else {
234             for (i=0; i<len; i++)
235                 coefs[i] = -(int64_t)coefs[i] * c * (1 << -s);
236         }
237     }
238 }
239 
flt16_round(SoftFloat pf)240 static av_always_inline SoftFloat flt16_round(SoftFloat pf)
241 {
242     SoftFloat tmp;
243     int s;
244 
245     tmp.exp = pf.exp;
246     s = pf.mant >> 31;
247     tmp.mant = (pf.mant ^ s) - s;
248     tmp.mant = (tmp.mant + 0x00200000U) & 0xFFC00000U;
249     tmp.mant = (tmp.mant ^ s) - s;
250 
251     return tmp;
252 }
253 
flt16_even(SoftFloat pf)254 static av_always_inline SoftFloat flt16_even(SoftFloat pf)
255 {
256     SoftFloat tmp;
257     int s;
258 
259     tmp.exp = pf.exp;
260     s = pf.mant >> 31;
261     tmp.mant = (pf.mant ^ s) - s;
262     tmp.mant = (tmp.mant + 0x001FFFFFU + (tmp.mant & 0x00400000U >> 16)) & 0xFFC00000U;
263     tmp.mant = (tmp.mant ^ s) - s;
264 
265     return tmp;
266 }
267 
flt16_trunc(SoftFloat pf)268 static av_always_inline SoftFloat flt16_trunc(SoftFloat pf)
269 {
270     SoftFloat pun;
271     int s;
272 
273     pun.exp = pf.exp;
274     s = pf.mant >> 31;
275     pun.mant = (pf.mant ^ s) - s;
276     pun.mant = pun.mant & 0xFFC00000U;
277     pun.mant = (pun.mant ^ s) - s;
278 
279     return pun;
280 }
281 
predict(PredictorState * ps,int * coef,int output_enable)282 static av_always_inline void predict(PredictorState *ps, int *coef,
283                                      int output_enable)
284 {
285     const SoftFloat a     = { 1023410176, 0 };  // 61.0 / 64
286     const SoftFloat alpha = {  973078528, 0 };  // 29.0 / 32
287     SoftFloat e0, e1;
288     SoftFloat pv;
289     SoftFloat k1, k2;
290     SoftFloat   r0 = ps->r0,     r1 = ps->r1;
291     SoftFloat cor0 = ps->cor0, cor1 = ps->cor1;
292     SoftFloat var0 = ps->var0, var1 = ps->var1;
293     SoftFloat tmp;
294 
295     if (var0.exp > 1 || (var0.exp == 1 && var0.mant > 0x20000000)) {
296         k1 = av_mul_sf(cor0, flt16_even(av_div_sf(a, var0)));
297     }
298     else {
299         k1.mant = 0;
300         k1.exp = 0;
301     }
302 
303     if (var1.exp > 1 || (var1.exp == 1 && var1.mant > 0x20000000)) {
304         k2 = av_mul_sf(cor1, flt16_even(av_div_sf(a, var1)));
305     }
306     else {
307         k2.mant = 0;
308         k2.exp = 0;
309     }
310 
311     tmp = av_mul_sf(k1, r0);
312     pv = flt16_round(av_add_sf(tmp, av_mul_sf(k2, r1)));
313     if (output_enable) {
314         int shift = 28 - pv.exp;
315 
316         if (shift < 31) {
317             if (shift > 0) {
318                 *coef += (unsigned)((pv.mant + (1 << (shift - 1))) >> shift);
319             } else
320                 *coef += (unsigned)pv.mant << -shift;
321         }
322     }
323 
324     e0 = av_int2sf(*coef, 2);
325     e1 = av_sub_sf(e0, tmp);
326 
327     ps->cor1 = flt16_trunc(av_add_sf(av_mul_sf(alpha, cor1), av_mul_sf(r1, e1)));
328     tmp = av_add_sf(av_mul_sf(r1, r1), av_mul_sf(e1, e1));
329     tmp.exp--;
330     ps->var1 = flt16_trunc(av_add_sf(av_mul_sf(alpha, var1), tmp));
331     ps->cor0 = flt16_trunc(av_add_sf(av_mul_sf(alpha, cor0), av_mul_sf(r0, e0)));
332     tmp = av_add_sf(av_mul_sf(r0, r0), av_mul_sf(e0, e0));
333     tmp.exp--;
334     ps->var0 = flt16_trunc(av_add_sf(av_mul_sf(alpha, var0), tmp));
335 
336     ps->r1 = flt16_trunc(av_mul_sf(a, av_sub_sf(r0, av_mul_sf(k1, e0))));
337     ps->r0 = flt16_trunc(av_mul_sf(a, e0));
338 }
339 
340 
341 static const int cce_scale_fixed[8] = {
342     Q30(1.0),          //2^(0/8)
343     Q30(1.0905077327), //2^(1/8)
344     Q30(1.1892071150), //2^(2/8)
345     Q30(1.2968395547), //2^(3/8)
346     Q30(1.4142135624), //2^(4/8)
347     Q30(1.5422108254), //2^(5/8)
348     Q30(1.6817928305), //2^(6/8)
349     Q30(1.8340080864), //2^(7/8)
350 };
351 
352 /**
353  * Apply dependent channel coupling (applied before IMDCT).
354  *
355  * @param   index   index into coupling gain array
356  */
apply_dependent_coupling_fixed(AACContext * ac,SingleChannelElement * target,ChannelElement * cce,int index)357 static void apply_dependent_coupling_fixed(AACContext *ac,
358                                      SingleChannelElement *target,
359                                      ChannelElement *cce, int index)
360 {
361     IndividualChannelStream *ics = &cce->ch[0].ics;
362     const uint16_t *offsets = ics->swb_offset;
363     int *dest = target->coeffs;
364     const int *src = cce->ch[0].coeffs;
365     int g, i, group, k, idx = 0;
366     if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
367         av_log(ac->avctx, AV_LOG_ERROR,
368                "Dependent coupling is not supported together with LTP\n");
369         return;
370     }
371     for (g = 0; g < ics->num_window_groups; g++) {
372         for (i = 0; i < ics->max_sfb; i++, idx++) {
373             if (cce->ch[0].band_type[idx] != ZERO_BT) {
374                 const int gain = cce->coup.gain[index][idx];
375                 int shift, round, c, tmp;
376 
377                 if (gain < 0) {
378                     c = -cce_scale_fixed[-gain & 7];
379                     shift = (-gain-1024) >> 3;
380                 }
381                 else {
382                     c = cce_scale_fixed[gain & 7];
383                     shift = (gain-1024) >> 3;
384                 }
385 
386                 if (shift < -31) {
387                     // Nothing to do
388                 } else if (shift < 0) {
389                     shift = -shift;
390                     round = 1 << (shift - 1);
391 
392                     for (group = 0; group < ics->group_len[g]; group++) {
393                         for (k = offsets[i]; k < offsets[i + 1]; k++) {
394                             tmp = (int)(((int64_t)src[group * 128 + k] * c + \
395                                        (int64_t)0x1000000000) >> 37);
396                             dest[group * 128 + k] += (tmp + (int64_t)round) >> shift;
397                         }
398                     }
399                 }
400                 else {
401                     for (group = 0; group < ics->group_len[g]; group++) {
402                         for (k = offsets[i]; k < offsets[i + 1]; k++) {
403                             tmp = (int)(((int64_t)src[group * 128 + k] * c + \
404                                         (int64_t)0x1000000000) >> 37);
405                             dest[group * 128 + k] += tmp * (1U << shift);
406                         }
407                     }
408                 }
409             }
410         }
411         dest += ics->group_len[g] * 128;
412         src  += ics->group_len[g] * 128;
413     }
414 }
415 
416 /**
417  * Apply independent channel coupling (applied after IMDCT).
418  *
419  * @param   index   index into coupling gain array
420  */
apply_independent_coupling_fixed(AACContext * ac,SingleChannelElement * target,ChannelElement * cce,int index)421 static void apply_independent_coupling_fixed(AACContext *ac,
422                                        SingleChannelElement *target,
423                                        ChannelElement *cce, int index)
424 {
425     int i, c, shift, round, tmp;
426     const int gain = cce->coup.gain[index][0];
427     const int *src = cce->ch[0].ret;
428     unsigned int *dest = target->ret;
429     const int len = 1024 << (ac->oc[1].m4ac.sbr == 1);
430 
431     c = cce_scale_fixed[gain & 7];
432     shift = (gain-1024) >> 3;
433     if (shift < -31) {
434         return;
435     } else if (shift < 0) {
436         shift = -shift;
437         round = 1 << (shift - 1);
438 
439         for (i = 0; i < len; i++) {
440             tmp = (int)(((int64_t)src[i] * c + (int64_t)0x1000000000) >> 37);
441             dest[i] += (tmp + round) >> shift;
442         }
443     }
444     else {
445       for (i = 0; i < len; i++) {
446           tmp = (int)(((int64_t)src[i] * c + (int64_t)0x1000000000) >> 37);
447           dest[i] += tmp * (1U << shift);
448       }
449     }
450 }
451 
452 #include "aacdec_template.c"
453 
454 AVCodec ff_aac_fixed_decoder = {
455     .name            = "aac_fixed",
456     .long_name       = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
457     .type            = AVMEDIA_TYPE_AUDIO,
458     .id              = AV_CODEC_ID_AAC,
459     .priv_data_size  = sizeof(AACContext),
460     .init            = aac_decode_init,
461     .close           = aac_decode_close,
462     .decode          = aac_decode_frame,
463     .sample_fmts     = (const enum AVSampleFormat[]) {
464         AV_SAMPLE_FMT_S32P, AV_SAMPLE_FMT_NONE
465     },
466     .capabilities    = AV_CODEC_CAP_CHANNEL_CONF | AV_CODEC_CAP_DR1,
467     .caps_internal   = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP,
468     .channel_layouts = aac_channel_layout,
469     .profiles        = NULL_IF_CONFIG_SMALL(ff_aac_profiles),
470     .flush = flush,
471 };
472