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 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 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 float *ff_sine_windows[5] = {
  57     ff_sine_128, ff_sine_256, ff_sine_512, ff_sine_1024, ff_sine_2048,
  58 };
  59
  60 // Generate a sine window.
  61 void ff_sine_window_init(float *window, int n) {
  62     int i;
  63     for(i = 0; i < n; i++)
  64         window[i] = sinf((i + 0.5) * (M_PI / (2.0 * n)));
  65 }
  66
  67 /**
  68  * init MDCT or IMDCT computation.
  69  */
  70 int ff_mdct_init(MDCTContext *s, int nbits, int inverse)
  71 {
  72     int n, n4, i;
  73     double alpha;
  74
  75     memset(s, 0, sizeof(*s));
  76     n = 1 << nbits;
  77     s->nbits = nbits;
  78     s->n = n;
  79     n4 = n >> 2;
  80     s->tcos = av_malloc(n4 * sizeof(FFTSample));
  81     if (!s->tcos)
  82         goto fail;
  83     s->tsin = av_malloc(n4 * sizeof(FFTSample));
  84     if (!s->tsin)
  85         goto fail;
  86
  87     for(i=0;i<n4;i++) {
  88         alpha = 2 * M_PI * (i + 1.0 / 8.0) / n;
  89         s->tcos[i] = -cos(alpha);
  90         s->tsin[i] = -sin(alpha);
  91     }
  92     if (ff_fft_init(&s->fft, s->nbits - 2, inverse) < 0)
  93         goto fail;
  94     return 0;
  95  fail:
  96     av_freep(&s->tcos);
  97     av_freep(&s->tsin);
  98     return -1;
  99 }
 100
 101 /* complex multiplication: p = a * b */
 102 #define CMUL(pre, pim, are, aim, bre, bim) \
 103 {\
 104     FFTSample _are = (are);\
 105     FFTSample _aim = (aim);\
 106     FFTSample _bre = (bre);\
 107     FFTSample _bim = (bim);\
 108     (pre) = _are * _bre - _aim * _bim;\
 109     (pim) = _are * _bim + _aim * _bre;\
 110 }
 111
 112 /**
 113  * Compute the middle half of the inverse MDCT of size N = 2^nbits,
 114  * thus excluding the parts that can be derived by symmetry
 115  * @param output N/2 samples
 116  * @param input N/2 samples
 117  */
 118 void ff_imdct_half_c(MDCTContext *s, FFTSample *output, const FFTSample *input)
 119 {
 120     int k, n8, n4, n2, n, j;
 121     const uint16_t *revtab = s->fft.revtab;
 122     const FFTSample *tcos = s->tcos;
 123     const FFTSample *tsin = s->tsin;
 124     const FFTSample *in1, *in2;
 125     FFTComplex *z = (FFTComplex *)output;
 126
 127     n = 1 << s->nbits;
 128     n2 = n >> 1;
 129     n4 = n >> 2;
 130     n8 = n >> 3;
 131
 132     /* pre rotation */
 133     in1 = input;
 134     in2 = input + n2 - 1;
 135     for(k = 0; k < n4; k++) {
 136         j=revtab[k];
 137         CMUL(z[j].re, z[j].im, *in2, *in1, tcos[k], tsin[k]);
 138         in1 += 2;
 139         in2 -= 2;
 140     }
 141     ff_fft_calc(&s->fft, z);
 142
 143     /* post rotation + reordering */
 144     output += n4;
 145     for(k = 0; k < n8; k++) {
 146         FFTSample r0, i0, r1, i1;
 147         CMUL(r0, i1, z[n8-k-1].im, z[n8-k-1].re, tsin[n8-k-1], tcos[n8-k-1]);
 148         CMUL(r1, i0, z[n8+k  ].im, z[n8+k  ].re, tsin[n8+k  ], tcos[n8+k  ]);
 149         z[n8-k-1].re = r0;
 150         z[n8-k-1].im = i0;
 151         z[n8+k  ].re = r1;
 152         z[n8+k  ].im = i1;
 153     }
 154 }
 155
 156 /**
 157  * Compute inverse MDCT of size N = 2^nbits
 158  * @param output N samples
 159  * @param input N/2 samples
 160  * @param tmp N/2 samples
 161  */
 162 void ff_imdct_calc_c(MDCTContext *s, FFTSample *output, const FFTSample *input)
 163 {
 164     int k;
 165     int n = 1 << s->nbits;
 166     int n2 = n >> 1;
 167     int n4 = n >> 2;
 168
 169     ff_imdct_half_c(s, output+n4, input);
 170
 171     for(k = 0; k < n4; k++) {
 172         output[k] = -output[n2-k-1];
 173         output[n-k-1] = output[n2+k];
 174     }
 175 }
 176
 177 /**
 178  * Compute MDCT of size N = 2^nbits
 179  * @param input N samples
 180  * @param out N/2 samples
 181  * @param tmp temporary storage of 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 void ff_mdct_end(MDCTContext *s)
 226 {
 227     av_freep(&s->tcos);
 228     av_freep(&s->tsin);
 229     ff_fft_end(&s->fft);
 230 }