Replace LDLATEFLAGS hackery by proper LDFLAGS tests.
[FFMpeg-mirror/ffmpeg-vdpau.git] / libavcodec / mdct.c
blob6a3b69a01473bc26ab9ba8693c3dd3fb1fff3474
1 /*
2 * MDCT/IMDCT transforms
3 * Copyright (c) 2002 Fabrice Bellard.
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "dsputil.h"
23 /**
24 * @file mdct.c
25 * MDCT/IMDCT transforms.
28 // Generate a Kaiser-Bessel Derived Window.
29 #define BESSEL_I0_ITER 50 // default: 50 iterations of Bessel I0 approximation
30 void ff_kbd_window_init(float *window, float alpha, int n)
32 int i, j;
33 double sum = 0.0, bessel, tmp;
34 double local_window[n];
35 double alpha2 = (alpha * M_PI / n) * (alpha * M_PI / n);
37 for (i = 0; i < n; i++) {
38 tmp = i * (n - i) * alpha2;
39 bessel = 1.0;
40 for (j = BESSEL_I0_ITER; j > 0; j--)
41 bessel = bessel * tmp / (j * j) + 1;
42 sum += bessel;
43 local_window[i] = sum;
46 sum++;
47 for (i = 0; i < n; i++)
48 window[i] = sqrt(local_window[i] / sum);
51 // Generate a sine window.
52 void ff_sine_window_init(float *window, int n) {
53 int i;
54 for(i = 0; i < n; i++)
55 window[i] = sin((i + 0.5) / (2 * n) * M_PI);
58 /**
59 * init MDCT or IMDCT computation.
61 int ff_mdct_init(MDCTContext *s, int nbits, int inverse)
63 int n, n4, i;
64 double alpha;
66 memset(s, 0, sizeof(*s));
67 n = 1 << nbits;
68 s->nbits = nbits;
69 s->n = n;
70 n4 = n >> 2;
71 s->tcos = av_malloc(n4 * sizeof(FFTSample));
72 if (!s->tcos)
73 goto fail;
74 s->tsin = av_malloc(n4 * sizeof(FFTSample));
75 if (!s->tsin)
76 goto fail;
78 for(i=0;i<n4;i++) {
79 alpha = 2 * M_PI * (i + 1.0 / 8.0) / n;
80 s->tcos[i] = -cos(alpha);
81 s->tsin[i] = -sin(alpha);
83 if (ff_fft_init(&s->fft, s->nbits - 2, inverse) < 0)
84 goto fail;
85 return 0;
86 fail:
87 av_freep(&s->tcos);
88 av_freep(&s->tsin);
89 return -1;
92 /* complex multiplication: p = a * b */
93 #define CMUL(pre, pim, are, aim, bre, bim) \
95 double _are = (are);\
96 double _aim = (aim);\
97 double _bre = (bre);\
98 double _bim = (bim);\
99 (pre) = _are * _bre - _aim * _bim;\
100 (pim) = _are * _bim + _aim * _bre;\
103 static void imdct_c(MDCTContext *s, const FFTSample *input, FFTSample *tmp)
105 int k, n4, n2, n, j;
106 const uint16_t *revtab = s->fft.revtab;
107 const FFTSample *tcos = s->tcos;
108 const FFTSample *tsin = s->tsin;
109 const FFTSample *in1, *in2;
110 FFTComplex *z = (FFTComplex *)tmp;
112 n = 1 << s->nbits;
113 n2 = n >> 1;
114 n4 = n >> 2;
116 /* pre rotation */
117 in1 = input;
118 in2 = input + n2 - 1;
119 for(k = 0; k < n4; k++) {
120 j=revtab[k];
121 CMUL(z[j].re, z[j].im, *in2, *in1, tcos[k], tsin[k]);
122 in1 += 2;
123 in2 -= 2;
125 ff_fft_calc(&s->fft, z);
127 /* post rotation + reordering */
128 /* XXX: optimize */
129 for(k = 0; k < n4; k++) {
130 CMUL(z[k].re, z[k].im, z[k].re, z[k].im, tcos[k], tsin[k]);
135 * Compute inverse MDCT of size N = 2^nbits
136 * @param output N samples
137 * @param input N/2 samples
138 * @param tmp N/2 samples
140 void ff_imdct_calc(MDCTContext *s, FFTSample *output,
141 const FFTSample *input, FFTSample *tmp)
143 int k, n8, n2, n;
144 FFTComplex *z = (FFTComplex *)tmp;
145 n = 1 << s->nbits;
146 n2 = n >> 1;
147 n8 = n >> 3;
149 imdct_c(s, input, tmp);
151 for(k = 0; k < n8; k++) {
152 output[2*k] = -z[n8 + k].im;
153 output[n2-1-2*k] = z[n8 + k].im;
155 output[2*k+1] = z[n8-1-k].re;
156 output[n2-1-2*k-1] = -z[n8-1-k].re;
158 output[n2 + 2*k]=-z[k+n8].re;
159 output[n-1- 2*k]=-z[k+n8].re;
161 output[n2 + 2*k+1]=z[n8-k-1].im;
162 output[n-2 - 2 * k] = z[n8-k-1].im;
167 * Compute the middle half of the inverse MDCT of size N = 2^nbits,
168 * thus excluding the parts that can be derived by symmetry
169 * @param output N/2 samples
170 * @param input N/2 samples
171 * @param tmp N/2 samples
173 void ff_imdct_half(MDCTContext *s, FFTSample *output,
174 const FFTSample *input, FFTSample *tmp)
176 int k, n8, n4, n;
177 FFTComplex *z = (FFTComplex *)tmp;
178 n = 1 << s->nbits;
179 n4 = n >> 2;
180 n8 = n >> 3;
182 imdct_c(s, input, tmp);
184 for(k = 0; k < n8; k++) {
185 output[n4-1-2*k] = z[n8+k].im;
186 output[n4-1-2*k-1] = -z[n8-k-1].re;
187 output[n4 + 2*k] = -z[n8+k].re;
188 output[n4 + 2*k+1] = z[n8-k-1].im;
193 * Compute MDCT of size N = 2^nbits
194 * @param input N samples
195 * @param out N/2 samples
196 * @param tmp temporary storage of N/2 samples
198 void ff_mdct_calc(MDCTContext *s, FFTSample *out,
199 const FFTSample *input, FFTSample *tmp)
201 int i, j, n, n8, n4, n2, n3;
202 FFTSample re, im, re1, im1;
203 const uint16_t *revtab = s->fft.revtab;
204 const FFTSample *tcos = s->tcos;
205 const FFTSample *tsin = s->tsin;
206 FFTComplex *x = (FFTComplex *)tmp;
208 n = 1 << s->nbits;
209 n2 = n >> 1;
210 n4 = n >> 2;
211 n8 = n >> 3;
212 n3 = 3 * n4;
214 /* pre rotation */
215 for(i=0;i<n8;i++) {
216 re = -input[2*i+3*n4] - input[n3-1-2*i];
217 im = -input[n4+2*i] + input[n4-1-2*i];
218 j = revtab[i];
219 CMUL(x[j].re, x[j].im, re, im, -tcos[i], tsin[i]);
221 re = input[2*i] - input[n2-1-2*i];
222 im = -(input[n2+2*i] + input[n-1-2*i]);
223 j = revtab[n8 + i];
224 CMUL(x[j].re, x[j].im, re, im, -tcos[n8 + i], tsin[n8 + i]);
227 ff_fft_calc(&s->fft, x);
229 /* post rotation */
230 for(i=0;i<n4;i++) {
231 re = x[i].re;
232 im = x[i].im;
233 CMUL(re1, im1, re, im, -tsin[i], -tcos[i]);
234 out[2*i] = im1;
235 out[n2-1-2*i] = re1;
239 void ff_mdct_end(MDCTContext *s)
241 av_freep(&s->tcos);
242 av_freep(&s->tsin);
243 ff_fft_end(&s->fft);