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_128 [ 128]);
  52 DECLARE_ALIGNED(16, float, ff_sine_256 [ 256]);
  53 DECLARE_ALIGNED(16, float, ff_sine_512 [ 512]);
  54 DECLARE_ALIGNED(16, float, ff_sine_1024[1024]);
  55 DECLARE_ALIGNED(16, float, ff_sine_2048[2048]);
  56 DECLARE_ALIGNED(16, float, ff_sine_4096[4096]);
  57 float *ff_sine_windows[6] = {
  58     ff_sine_128, ff_sine_256, ff_sine_512, ff_sine_1024, ff_sine_2048, ff_sine_4096
  59 };
  60
  61 // Generate a sine window.
  62 av_cold void ff_sine_window_init(float *window, int n) {
  63     int i;
  64     for(i = 0; i < n; i++)
  65         window[i] = sinf((i + 0.5) * (M_PI / (2.0 * n)));
  66 }
  67
  68 /**
  69  * init MDCT or IMDCT computation.
  70  */
  71 av_cold int ff_mdct_init(MDCTContext *s, int nbits, int inverse, double scale)
  72 {
  73     int n, n4, i;
  74     double alpha, theta;
  75
  76     memset(s, 0, sizeof(*s));
  77     n = 1 << nbits;
  78     s->nbits = nbits;
  79     s->n = n;
  80     n4 = n >> 2;
  81     s->tcos = av_malloc(n4 * sizeof(FFTSample));
  82     if (!s->tcos)
  83         goto fail;
  84     s->tsin = av_malloc(n4 * sizeof(FFTSample));
  85     if (!s->tsin)
  86         goto fail;
  87
  88     theta = 1.0 / 8.0 + (scale < 0 ? n4 : 0);
  89     scale = sqrt(fabs(scale));
  90     for(i=0;i<n4;i++) {
  91         alpha = 2 * M_PI * (i + theta) / n;
  92         s->tcos[i] = -cos(alpha) * scale;
  93         s->tsin[i] = -sin(alpha) * scale;
  94     }
  95     if (ff_fft_init(&s->fft, s->nbits - 2, inverse) < 0)
  96         goto fail;
  97     return 0;
  98  fail:
  99     av_freep(&s->tcos);
 100     av_freep(&s->tsin);
 101     return -1;
 102 }
 103
 104 /* complex multiplication: p = a * b */
 105 #define CMUL(pre, pim, are, aim, bre, bim) \
 106 {\
 107     FFTSample _are = (are);\
 108     FFTSample _aim = (aim);\
 109     FFTSample _bre = (bre);\
 110     FFTSample _bim = (bim);\
 111     (pre) = _are * _bre - _aim * _bim;\
 112     (pim) = _are * _bim + _aim * _bre;\
 113 }
 114
 115 /**
 116  * Compute the middle half of the inverse MDCT of size N = 2^nbits,
 117  * thus excluding the parts that can be derived by symmetry
 118  * @param output N/2 samples
 119  * @param input N/2 samples
 120  */
 121 void ff_imdct_half_c(MDCTContext *s, FFTSample *output, const FFTSample *input)
 122 {
 123     int k, n8, n4, n2, n, j;
 124     const uint16_t *revtab = s->fft.revtab;
 125     const FFTSample *tcos = s->tcos;
 126     const FFTSample *tsin = s->tsin;
 127     const FFTSample *in1, *in2;
 128     FFTComplex *z = (FFTComplex *)output;
 129
 130     n = 1 << s->nbits;
 131     n2 = n >> 1;
 132     n4 = n >> 2;
 133     n8 = n >> 3;
 134
 135     /* pre rotation */
 136     in1 = input;
 137     in2 = input + n2 - 1;
 138     for(k = 0; k < n4; k++) {
 139         j=revtab[k];
 140         CMUL(z[j].re, z[j].im, *in2, *in1, tcos[k], tsin[k]);
 141         in1 += 2;
 142         in2 -= 2;
 143     }
 144     ff_fft_calc(&s->fft, z);
 145
 146     /* post rotation + reordering */
 147     for(k = 0; k < n8; k++) {
 148         FFTSample r0, i0, r1, i1;
 149         CMUL(r0, i1, z[n8-k-1].im, z[n8-k-1].re, tsin[n8-k-1], tcos[n8-k-1]);
 150         CMUL(r1, i0, z[n8+k  ].im, z[n8+k  ].re, tsin[n8+k  ], tcos[n8+k  ]);
 151         z[n8-k-1].re = r0;
 152         z[n8-k-1].im = i0;
 153         z[n8+k  ].re = r1;
 154         z[n8+k  ].im = i1;
 155     }
 156 }
 157
 158 /**
 159  * Compute inverse MDCT of size N = 2^nbits
 160  * @param output N samples
 161  * @param input N/2 samples
 162  */
 163 void ff_imdct_calc_c(MDCTContext *s, FFTSample *output, const FFTSample *input)
 164 {
 165     int k;
 166     int n = 1 << s->nbits;
 167     int n2 = n >> 1;
 168     int n4 = n >> 2;
 169
 170     ff_imdct_half_c(s, output+n4, input);
 171
 172     for(k = 0; k < n4; k++) {
 173         output[k] = -output[n2-k-1];
 174         output[n-k-1] = output[n2+k];
 175     }
 176 }
 177
 178 /**
 179  * Compute MDCT of size N = 2^nbits
 180  * @param input N samples
 181  * @param out N/2 samples
 182  */
 183 void ff_mdct_calc(MDCTContext *s, FFTSample *out, const FFTSample *input)
 184 {
 185     int i, j, n, n8, n4, n2, n3;
 186     FFTSample re, im;
 187     const uint16_t *revtab = s->fft.revtab;
 188     const FFTSample *tcos = s->tcos;
 189     const FFTSample *tsin = s->tsin;
 190     FFTComplex *x = (FFTComplex *)out;
 191
 192     n = 1 << s->nbits;
 193     n2 = n >> 1;
 194     n4 = n >> 2;
 195     n8 = n >> 3;
 196     n3 = 3 * n4;
 197
 198     /* pre rotation */
 199     for(i=0;i<n8;i++) {
 200         re = -input[2*i+3*n4] - input[n3-1-2*i];
 201         im = -input[n4+2*i] + input[n4-1-2*i];
 202         j = revtab[i];
 203         CMUL(x[j].re, x[j].im, re, im, -tcos[i], tsin[i]);
 204
 205         re = input[2*i] - input[n2-1-2*i];
 206         im = -(input[n2+2*i] + input[n-1-2*i]);
 207         j = revtab[n8 + i];
 208         CMUL(x[j].re, x[j].im, re, im, -tcos[n8 + i], tsin[n8 + i]);
 209     }
 210
 211     ff_fft_calc(&s->fft, x);
 212
 213     /* post rotation */
 214     for(i=0;i<n8;i++) {
 215         FFTSample r0, i0, r1, i1;
 216         CMUL(i1, r0, x[n8-i-1].re, x[n8-i-1].im, -tsin[n8-i-1], -tcos[n8-i-1]);
 217         CMUL(i0, r1, x[n8+i  ].re, x[n8+i  ].im, -tsin[n8+i  ], -tcos[n8+i  ]);
 218         x[n8-i-1].re = r0;
 219         x[n8-i-1].im = i0;
 220         x[n8+i  ].re = r1;
 221         x[n8+i  ].im = i1;
 222     }
 223 }
 224
 225 av_cold void ff_mdct_end(MDCTContext *s)
 226 {
 227     av_freep(&s->tcos);
 228     av_freep(&s->tsin);
 229     ff_fft_end(&s->fft);
 230 }