2 * Copyright (c) 2001, 2002 Fabrice Bellard
4 * This file is part of FFmpeg.
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.
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.
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
23 #include "libavutil/mem.h"
26 #include "mpegaudiodsp.h"
27 #include "mpegaudio.h"
28 #include "mpegaudiodata.h"
31 #define RENAME(n) n##_float
33 static inline float round_sample(float *sum)
40 #define MACS(rt, ra, rb) rt+=(ra)*(rb)
41 #define MULS(ra, rb) ((ra)*(rb))
42 #define MULH3(x, y, s) ((s)*(y)*(x))
43 #define MLSS(rt, ra, rb) rt-=(ra)*(rb)
44 #define MULLx(x, y, s) ((y)*(x))
45 #define FIXHR(x) ((float)(x))
46 #define FIXR(x) ((float)(x))
47 #define SHR(a,b) ((a)*(1.0f/(1<<(b))))
51 #define RENAME(n) n##_fixed
52 #define OUT_SHIFT (WFRAC_BITS + FRAC_BITS - 15)
54 static inline int round_sample(int64_t *sum)
57 sum1 = (int)((*sum) >> OUT_SHIFT);
58 *sum &= (1<<OUT_SHIFT)-1;
59 return av_clip_int16(sum1);
62 # define MULS(ra, rb) MUL64(ra, rb)
63 # define MACS(rt, ra, rb) MAC64(rt, ra, rb)
64 # define MLSS(rt, ra, rb) MLS64(rt, ra, rb)
65 # define MULH3(x, y, s) MULH((s)*(x), y)
66 # define MULLx(x, y, s) MULL(x,y,s)
67 # define SHR(a,b) ((a)>>(b))
68 # define FIXR(a) ((int)((a) * FRAC_ONE + 0.5))
69 # define FIXHR(a) ((int)((a) * (1LL<<32) + 0.5))
72 DECLARE_ALIGNED(16, MPA_INT, RENAME(ff_mpa_synth_window))[512+256];
74 #define SUM8(op, sum, w, p) \
76 op(sum, (w)[0 * 64], (p)[0 * 64]); \
77 op(sum, (w)[1 * 64], (p)[1 * 64]); \
78 op(sum, (w)[2 * 64], (p)[2 * 64]); \
79 op(sum, (w)[3 * 64], (p)[3 * 64]); \
80 op(sum, (w)[4 * 64], (p)[4 * 64]); \
81 op(sum, (w)[5 * 64], (p)[5 * 64]); \
82 op(sum, (w)[6 * 64], (p)[6 * 64]); \
83 op(sum, (w)[7 * 64], (p)[7 * 64]); \
86 #define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \
90 op1(sum1, (w1)[0 * 64], tmp);\
91 op2(sum2, (w2)[0 * 64], tmp);\
93 op1(sum1, (w1)[1 * 64], tmp);\
94 op2(sum2, (w2)[1 * 64], tmp);\
96 op1(sum1, (w1)[2 * 64], tmp);\
97 op2(sum2, (w2)[2 * 64], tmp);\
99 op1(sum1, (w1)[3 * 64], tmp);\
100 op2(sum2, (w2)[3 * 64], tmp);\
102 op1(sum1, (w1)[4 * 64], tmp);\
103 op2(sum2, (w2)[4 * 64], tmp);\
105 op1(sum1, (w1)[5 * 64], tmp);\
106 op2(sum2, (w2)[5 * 64], tmp);\
108 op1(sum1, (w1)[6 * 64], tmp);\
109 op2(sum2, (w2)[6 * 64], tmp);\
111 op1(sum1, (w1)[7 * 64], tmp);\
112 op2(sum2, (w2)[7 * 64], tmp);\
115 void RENAME(ff_mpadsp_apply_window)(MPA_INT *synth_buf, MPA_INT *window,
116 int *dither_state, OUT_INT *samples,
119 register const MPA_INT *w, *w2, *p;
128 /* copy to avoid wrap */
129 memcpy(synth_buf + 512, synth_buf, 32 * sizeof(*synth_buf));
131 samples2 = samples + 31 * incr;
137 SUM8(MACS, sum, w, p);
139 SUM8(MLSS, sum, w + 32, p);
140 *samples = round_sample(&sum);
144 /* we calculate two samples at the same time to avoid one memory
145 access per two sample */
148 p = synth_buf + 16 + j;
149 SUM8P2(sum, MACS, sum2, MLSS, w, w2, p);
150 p = synth_buf + 48 - j;
151 SUM8P2(sum, MLSS, sum2, MLSS, w + 32, w2 + 32, p);
153 *samples = round_sample(&sum);
156 *samples2 = round_sample(&sum);
163 SUM8(MLSS, sum, w + 32, p);
164 *samples = round_sample(&sum);
168 /* 32 sub band synthesis filter. Input: 32 sub band samples, Output:
170 void RENAME(ff_mpa_synth_filter)(MPADSPContext *s, MPA_INT *synth_buf_ptr,
171 int *synth_buf_offset,
172 MPA_INT *window, int *dither_state,
173 OUT_INT *samples, int incr,
179 offset = *synth_buf_offset;
180 synth_buf = synth_buf_ptr + offset;
182 s->RENAME(dct32)(synth_buf, sb_samples);
183 s->RENAME(apply_window)(synth_buf, window, dither_state, samples, incr);
185 offset = (offset - 32) & 511;
186 *synth_buf_offset = offset;
189 void av_cold RENAME(ff_mpa_synth_init)(MPA_INT *window)
193 /* max = 18760, max sum over all 16 coefs : 44736 */
196 v = ff_mpa_enwindow[i];
198 v *= 1.0 / (1LL<<(16 + FRAC_BITS));
207 // Needed for avoiding shuffles in ASM implementations
209 for(j=0; j < 16; j++)
210 window[512+16*i+j] = window[64*i+32-j];
213 for(j=0; j < 16; j++)
214 window[512+128+16*i+j] = window[64*i+48-j];
218 #define C1 FIXHR(0.98480775301220805936/2)
219 #define C2 FIXHR(0.93969262078590838405/2)
220 #define C3 FIXHR(0.86602540378443864676/2)
221 #define C4 FIXHR(0.76604444311897803520/2)
222 #define C5 FIXHR(0.64278760968653932632/2)
223 #define C6 FIXHR(0.5/2)
224 #define C7 FIXHR(0.34202014332566873304/2)
225 #define C8 FIXHR(0.17364817766693034885/2)
227 /* 0.5 / cos(pi*(2*i+1)/36) */
228 static const INTFLOAT icos36[9] = {
229 FIXR(0.50190991877167369479),
230 FIXR(0.51763809020504152469),
231 FIXR(0.55168895948124587824),
232 FIXR(0.61038729438072803416),
233 FIXR(0.70710678118654752439),
234 FIXR(0.87172339781054900991),
235 FIXR(1.18310079157624925896),
236 FIXR(1.93185165257813657349),
237 FIXR(5.73685662283492756461),
240 /* 0.5 / cos(pi*(2*i+1)/36) */
241 static const INTFLOAT icos36h[9] = {
242 FIXHR(0.50190991877167369479/2),
243 FIXHR(0.51763809020504152469/2),
244 FIXHR(0.55168895948124587824/2),
245 FIXHR(0.61038729438072803416/2),
246 FIXHR(0.70710678118654752439/2),
247 FIXHR(0.87172339781054900991/2),
248 FIXHR(1.18310079157624925896/4),
249 FIXHR(1.93185165257813657349/4),
253 /* using Lee like decomposition followed by hand coded 9 points DCT */
254 void RENAME(ff_imdct36)(INTFLOAT *out, INTFLOAT *buf, INTFLOAT *in,
258 INTFLOAT t0, t1, t2, t3, s0, s1, s2, s3;
259 INTFLOAT tmp[18], *tmp1, *in1;
270 t2 = in1[2*4] + in1[2*8] - in1[2*2];
272 t3 = in1[2*0] + SHR(in1[2*6],1);
273 t1 = in1[2*0] - in1[2*6];
274 tmp1[ 6] = t1 - SHR(t2,1);
277 t0 = MULH3(in1[2*2] + in1[2*4] , C2, 2);
278 t1 = MULH3(in1[2*4] - in1[2*8] , -2*C8, 1);
279 t2 = MULH3(in1[2*2] + in1[2*8] , -C4, 2);
281 tmp1[10] = t3 - t0 - t2;
282 tmp1[ 2] = t3 + t0 + t1;
283 tmp1[14] = t3 + t2 - t1;
285 tmp1[ 4] = MULH3(in1[2*5] + in1[2*7] - in1[2*1], -C3, 2);
286 t2 = MULH3(in1[2*1] + in1[2*5], C1, 2);
287 t3 = MULH3(in1[2*5] - in1[2*7], -2*C7, 1);
288 t0 = MULH3(in1[2*3], C3, 2);
290 t1 = MULH3(in1[2*1] + in1[2*7], -C5, 2);
292 tmp1[ 0] = t2 + t3 + t0;
293 tmp1[12] = t2 + t1 - t0;
294 tmp1[ 8] = t3 - t1 - t0;
306 s1 = MULH3(t3 + t2, icos36h[j], 2);
307 s3 = MULLx(t3 - t2, icos36[8 - j], FRAC_BITS);
311 out[(9 + j)*SBLIMIT] = MULH3(t1, win[9 + j], 1) + buf[9 + j];
312 out[(8 - j)*SBLIMIT] = MULH3(t1, win[8 - j], 1) + buf[8 - j];
313 buf[9 + j] = MULH3(t0, win[20 + 9 + j], 1);
314 buf[8 - j] = MULH3(t0, win[20 + 8 - j], 1);
318 out[(9 + 8 - j)*SBLIMIT] = MULH3(t1, win[9 + 8 - j], 1) + buf[9 + 8 - j];
319 out[( j)*SBLIMIT] = MULH3(t1, win[ j], 1) + buf[ j];
320 buf[9 + 8 - j] = MULH3(t0, win[20 + 9 + 8 - j], 1);
321 buf[ + j] = MULH3(t0, win[20 + j], 1);
326 s1 = MULH3(tmp[17], icos36h[4], 2);
329 out[(9 + 4)*SBLIMIT] = MULH3(t1, win[9 + 4], 1) + buf[9 + 4];
330 out[(8 - 4)*SBLIMIT] = MULH3(t1, win[8 - 4], 1) + buf[8 - 4];
331 buf[9 + 4] = MULH3(t0, win[20 + 9 + 4], 1);
332 buf[8 - 4] = MULH3(t0, win[20 + 8 - 4], 1);