libavcodec/mdct.c

   1 /*
   2  * MDCT/IMDCT transforms
   3  * Copyright (c) 2002 Fabrice Bellard
   4  *
   5  * This file is part of FFmpeg.
   6  *
   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.
  11  *
  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.
  16  *
  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
  20  */
  21 #include "dsputil.h"
  22
  23 /**
  24  * @file libavcodec/mdct.c
  25  * MDCT/IMDCT transforms.
  26  */
  27
  28 // Generate a Kaiser-Bessel Derived Window.
  29 #define BESSEL_I0_ITER 50 // default: 50 iterations of Bessel I0 approximation
  30 av_cold void ff_kbd_window_init(float *window, float alpha, int n)
  31 {
  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);
  36
  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;
  44    }
  45
  46    sum++;
  47    for (i = 0; i < n; i++)
  48        window[i] = sqrt(local_window[i] / sum);
  49 }
  50
  51 DECLARE_ALIGNED(16, float, ff_sine_32  [  32]);
  52 DECLARE_ALIGNED(16, float, ff_sine_64  [  64]);
  53 DECLARE_ALIGNED(16, float, ff_sine_128 [ 128]);
  54 DECLARE_ALIGNED(16, float, ff_sine_256 [ 256]);
  55 DECLARE_ALIGNED(16, float, ff_sine_512 [ 512]);
  56 DECLARE_ALIGNED(16, float, ff_sine_1024[1024]);
  57 DECLARE_ALIGNED(16, float, ff_sine_2048[2048]);
  58 DECLARE_ALIGNED(16, float, ff_sine_4096[4096]);
  59 float * const ff_sine_windows[] = {
  60     NULL, NULL, NULL, NULL, NULL, // unused
  61     ff_sine_32 , ff_sine_64 ,
  62     ff_sine_128, ff_sine_256, ff_sine_512, ff_sine_1024, ff_sine_2048, ff_sine_4096
  63 };
  64
  65 // Generate a sine window.
  66 av_cold void ff_sine_window_init(float *window, int n) {
  67     int i;
  68     for(i = 0; i < n; i++)
  69         window[i] = sinf((i + 0.5) * (M_PI / (2.0 * n)));
  70 }
  71
  72 /**
  73  * init MDCT or IMDCT computation.
  74  */
  75 av_cold int ff_mdct_init(MDCTContext *s, int nbits, int inverse, double scale)
  76 {
  77     int n, n4, i;
  78     double alpha, theta;
  79
  80     memset(s, 0, sizeof(*s));
  81     n = 1 << nbits;
  82     s->nbits = nbits;
  83     s->n = n;
  84     n4 = n >> 2;
  85     s->tcos = av_malloc(n4 * sizeof(FFTSample));
  86     if (!s->tcos)
  87         goto fail;
  88     s->tsin = av_malloc(n4 * sizeof(FFTSample));
  89     if (!s->tsin)
  90         goto fail;
  91
  92     theta = 1.0 / 8.0 + (scale < 0 ? n4 : 0);
  93     scale = sqrt(fabs(scale));
  94     for(i=0;i<n4;i++) {
  95         alpha = 2 * M_PI * (i + theta) / n;
  96         s->tcos[i] = -cos(alpha) * scale;
  97         s->tsin[i] = -sin(alpha) * scale;
  98     }
  99     if (ff_fft_init(&s->fft, s->nbits - 2, inverse) < 0)
 100         goto fail;
 101     return 0;
 102  fail:
 103     av_freep(&s->tcos);
 104     av_freep(&s->tsin);
 105     return -1;
 106 }
 107
 108 /* complex multiplication: p = a * b */
 109 #define CMUL(pre, pim, are, aim, bre, bim) \
 110 {\
 111     FFTSample _are = (are);\
 112     FFTSample _aim = (aim);\
 113     FFTSample _bre = (bre);\
 114     FFTSample _bim = (bim);\
 115     (pre) = _are * _bre - _aim * _bim;\
 116     (pim) = _are * _bim + _aim * _bre;\
 117 }
 118
 119 /**
 120  * Compute the middle half of the inverse MDCT of size N = 2^nbits,
 121  * thus excluding the parts that can be derived by symmetry
 122  * @param output N/2 samples
 123  * @param input N/2 samples
 124  */
 125 void ff_imdct_half_c(MDCTContext *s, FFTSample *output, const FFTSample *input)
 126 {
 127     int k, n8, n4, n2, n, j;
 128     const uint16_t *revtab = s->fft.revtab;
 129     const FFTSample *tcos = s->tcos;
 130     const FFTSample *tsin = s->tsin;
 131     const FFTSample *in1, *in2;
 132     FFTComplex *z = (FFTComplex *)output;
 133
 134     n = 1 << s->nbits;
 135     n2 = n >> 1;
 136     n4 = n >> 2;
 137     n8 = n >> 3;
 138
 139     /* pre rotation */
 140     in1 = input;
 141     in2 = input + n2 - 1;
 142     for(k = 0; k < n4; k++) {
 143         j=revtab[k];
 144         CMUL(z[j].re, z[j].im, *in2, *in1, tcos[k], tsin[k]);
 145         in1 += 2;
 146         in2 -= 2;
 147     }
 148     ff_fft_calc(&s->fft, z);
 149
 150     /* post rotation + reordering */
 151     for(k = 0; k < n8; k++) {
 152         FFTSample r0, i0, r1, i1;
 153         CMUL(r0, i1, z[n8-k-1].im, z[n8-k-1].re, tsin[n8-k-1], tcos[n8-k-1]);
 154         CMUL(r1, i0, z[n8+k  ].im, z[n8+k  ].re, tsin[n8+k  ], tcos[n8+k  ]);
 155         z[n8-k-1].re = r0;
 156         z[n8-k-1].im = i0;
 157         z[n8+k  ].re = r1;
 158         z[n8+k  ].im = i1;
 159     }
 160 }
 161
 162 /**
 163  * Compute inverse MDCT of size N = 2^nbits
 164  * @param output N samples
 165  * @param input N/2 samples
 166  */
 167 void ff_imdct_calc_c(MDCTContext *s, FFTSample *output, const FFTSample *input)
 168 {
 169     int k;
 170     int n = 1 << s->nbits;
 171     int n2 = n >> 1;
 172     int n4 = n >> 2;
 173
 174     ff_imdct_half_c(s, output+n4, input);
 175
 176     for(k = 0; k < n4; k++) {
 177         output[k] = -output[n2-k-1];
 178         output[n-k-1] = output[n2+k];
 179     }
 180 }
 181
 182 /**
 183  * Compute MDCT of size N = 2^nbits
 184  * @param input N samples
 185  * @param out N/2 samples
 186  */
 187 void ff_mdct_calc_c(MDCTContext *s, FFTSample *out, const FFTSample *input)
 188 {
 189     int i, j, n, n8, n4, n2, n3;
 190     FFTSample re, im;
 191     const uint16_t *revtab = s->fft.revtab;
 192     const FFTSample *tcos = s->tcos;
 193     const FFTSample *tsin = s->tsin;
 194     FFTComplex *x = (FFTComplex *)out;
 195
 196     n = 1 << s->nbits;
 197     n2 = n >> 1;
 198     n4 = n >> 2;
 199     n8 = n >> 3;
 200     n3 = 3 * n4;
 201
 202     /* pre rotation */
 203     for(i=0;i<n8;i++) {
 204         re = -input[2*i+3*n4] - input[n3-1-2*i];
 205         im = -input[n4+2*i] + input[n4-1-2*i];
 206         j = revtab[i];
 207         CMUL(x[j].re, x[j].im, re, im, -tcos[i], tsin[i]);
 208
 209         re = input[2*i] - input[n2-1-2*i];
 210         im = -(input[n2+2*i] + input[n-1-2*i]);
 211         j = revtab[n8 + i];
 212         CMUL(x[j].re, x[j].im, re, im, -tcos[n8 + i], tsin[n8 + i]);
 213     }
 214
 215     ff_fft_calc(&s->fft, x);
 216
 217     /* post rotation */
 218     for(i=0;i<n8;i++) {
 219         FFTSample r0, i0, r1, i1;
 220         CMUL(i1, r0, x[n8-i-1].re, x[n8-i-1].im, -tsin[n8-i-1], -tcos[n8-i-1]);
 221         CMUL(i0, r1, x[n8+i  ].re, x[n8+i  ].im, -tsin[n8+i  ], -tcos[n8+i  ]);
 222         x[n8-i-1].re = r0;
 223         x[n8-i-1].im = i0;
 224         x[n8+i  ].re = r1;
 225         x[n8+i  ].im = i1;
 226     }
 227 }
 228
 229 av_cold void ff_mdct_end(MDCTContext *s)
 230 {
 231     av_freep(&s->tcos);
 232     av_freep(&s->tsin);
 233     ff_fft_end(&s->fft);
 234 }