Revert r14786 which resulted in a substantial reduction in accuracy to save a 7.6KB...
[kugel-rb.git] / apps / codecs / libwma / mdct.c
bloba764f47eed10565bee9ae12b97085dc7398a5152
1 /*
2 * WMA compatible decoder
3 * Copyright (c) 2002 The FFmpeg Project.
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 #include "wmafixed.h"
21 #include "mdct.h"
23 /*these are the sin and cos rotations used by the MDCT*/
25 /*accessed too infrequently to give much speedup in IRAM*/
27 fixed32 *tcosarray[5], *tsinarray[5];
28 fixed32 tcos0[1024], tcos1[512], tcos2[256], tcos3[128], tcos4[64];
29 fixed32 tsin0[1024], tsin1[512], tsin2[256], tsin3[128], tsin4[64];
31 uint16_t revtab0[1024];
33 /**
34 * init MDCT or IMDCT computation.
36 int ff_mdct_init(MDCTContext *s, int nbits, int inverse)
38 int n, n4, i;
40 memset(s, 0, sizeof(*s));
41 n = 1 << nbits; //nbits ranges from 12 to 8 inclusive
42 s->nbits = nbits;
43 s->n = n;
44 n4 = n >> 2;
45 s->tcos = tcosarray[12-nbits];
46 s->tsin = tsinarray[12-nbits];
47 for(i=0;i<n4;i++)
50 fixed32 ip = itofix32(i) + 0x2000;
51 ip = ip >> nbits;
53 /*I can't remember why this works, but it seems to agree for ~24 bits, maybe more!*/
54 s->tsin[i] = - fsincos(ip<<16, &(s->tcos[i]));
55 s->tcos[i] *=-1;
58 (&s->fft)->nbits = nbits-2;
60 (&s->fft)->inverse = inverse;
62 return 0;
66 /**
67 * Compute inverse MDCT of size N = 2^nbits
68 * @param output N samples
69 * @param input N/2 samples
70 * @param tmp N/2 samples
72 void ff_imdct_calc(MDCTContext *s,
73 fixed32 *output,
74 fixed32 *input)
76 int k, n8, n4, n2, n, j,scale;
77 const fixed32 *tcos = s->tcos;
78 const fixed32 *tsin = s->tsin;
79 const fixed32 *in1, *in2;
80 FFTComplex *z1 = (FFTComplex *)output;
81 FFTComplex *z2 = (FFTComplex *)input;
82 int revtabshift = 12 - s->nbits;
84 n = 1 << s->nbits;
86 n2 = n >> 1;
87 n4 = n >> 2;
88 n8 = n >> 3;
91 /* pre rotation */
92 in1 = input;
93 in2 = input + n2 - 1;
95 for(k = 0; k < n4; k++)
97 j=revtab0[k<<revtabshift];
98 CMUL(&z1[j].re, &z1[j].im, *in2, *in1, tcos[k], tsin[k]);
99 in1 += 2;
100 in2 -= 2;
103 scale = fft_calc_unscaled(&s->fft, z1);
105 /* post rotation + reordering */
107 for(k = 0; k < n4; k++)
109 CMUL(&z2[k].re, &z2[k].im, (z1[k].re), (z1[k].im), tcos[k], tsin[k]);
112 for(k = 0; k < n8; k++)
114 fixed32 r1,r2,r3,r4,r1n,r2n,r3n;
116 r1 = z2[n8 + k].im;
117 r1n = r1 * -1;
118 r2 = z2[n8-1-k].re;
119 r2n = r2 * -1;
120 r3 = z2[k+n8].re;
121 r3n = r3 * -1;
122 r4 = z2[n8-k-1].im;
124 output[2*k] = r1n;
125 output[n2-1-2*k] = r1;
127 output[2*k+1] = r2;
128 output[n2-1-2*k-1] = r2n;
130 output[n2 + 2*k]= r3n;
131 output[n-1- 2*k]= r3n;
133 output[n2 + 2*k+1]= r4;
134 output[n-2 - 2 * k] = r4;
138 /* init MDCT */
140 int mdct_init_global(void)
142 int i,j,m;
144 /* although seemingly degenerate, these cannot actually be merged together without
145 a substantial increase in error which is unjustified by the tiny memory savings*/
147 tcosarray[0] = tcos0; tcosarray[1] = tcos1; tcosarray[2] = tcos2; tcosarray[3] = tcos3;tcosarray[4] = tcos4;
148 tsinarray[0] = tsin0; tsinarray[1] = tsin1; tsinarray[2] = tsin2; tsinarray[3] = tsin3;tsinarray[4] = tsin4;
150 /* init the MDCT bit reverse table here rather then in fft_init */
152 for(i=0;i<1024;i++) /*hard coded to a 2048 bit rotation*/
153 { /*smaller sizes can reuse the largest*/
154 m=0;
155 for(j=0;j<10;j++)
157 m |= ((i >> j) & 1) << (10-j-1);
160 revtab0[i]=m;
163 fft_init_global();
165 return 0;