00001
00002
00003
00004
00005
00006
00007
00008
00009
00010
00011
00012
00013
00014
00015
00016
00017
00018
00019
00020
00021
00022 #include <stdlib.h>
00023 #include <string.h>
00024 #include "libavutil/common.h"
00025 #include "libavutil/mathematics.h"
00026 #include "fft.h"
00027 #include "fft-internal.h"
00028
00034 #if CONFIG_FFT_FLOAT
00035 # define RSCALE(x) (x)
00036 #else
00037 # define RSCALE(x) ((x) >> 1)
00038 #endif
00039
00043 av_cold int ff_mdct_init(FFTContext *s, int nbits, int inverse, double scale)
00044 {
00045 int n, n4, i;
00046 double alpha, theta;
00047 int tstep;
00048
00049 memset(s, 0, sizeof(*s));
00050 n = 1 << nbits;
00051 s->mdct_bits = nbits;
00052 s->mdct_size = n;
00053 n4 = n >> 2;
00054 s->mdct_permutation = FF_MDCT_PERM_NONE;
00055
00056 if (ff_fft_init(s, s->mdct_bits - 2, inverse) < 0)
00057 goto fail;
00058
00059 s->tcos = av_malloc(n/2 * sizeof(FFTSample));
00060 if (!s->tcos)
00061 goto fail;
00062
00063 switch (s->mdct_permutation) {
00064 case FF_MDCT_PERM_NONE:
00065 s->tsin = s->tcos + n4;
00066 tstep = 1;
00067 break;
00068 case FF_MDCT_PERM_INTERLEAVE:
00069 s->tsin = s->tcos + 1;
00070 tstep = 2;
00071 break;
00072 default:
00073 goto fail;
00074 }
00075
00076 theta = 1.0 / 8.0 + (scale < 0 ? n4 : 0);
00077 scale = sqrt(fabs(scale));
00078 for(i=0;i<n4;i++) {
00079 alpha = 2 * M_PI * (i + theta) / n;
00080 s->tcos[i*tstep] = FIX15(-cos(alpha) * scale);
00081 s->tsin[i*tstep] = FIX15(-sin(alpha) * scale);
00082 }
00083 return 0;
00084 fail:
00085 ff_mdct_end(s);
00086 return -1;
00087 }
00088
00095 void ff_imdct_half_c(FFTContext *s, FFTSample *output, const FFTSample *input)
00096 {
00097 int k, n8, n4, n2, n, j;
00098 const uint16_t *revtab = s->revtab;
00099 const FFTSample *tcos = s->tcos;
00100 const FFTSample *tsin = s->tsin;
00101 const FFTSample *in1, *in2;
00102 FFTComplex *z = (FFTComplex *)output;
00103
00104 n = 1 << s->mdct_bits;
00105 n2 = n >> 1;
00106 n4 = n >> 2;
00107 n8 = n >> 3;
00108
00109
00110 in1 = input;
00111 in2 = input + n2 - 1;
00112 for(k = 0; k < n4; k++) {
00113 j=revtab[k];
00114 CMUL(z[j].re, z[j].im, *in2, *in1, tcos[k], tsin[k]);
00115 in1 += 2;
00116 in2 -= 2;
00117 }
00118 s->fft_calc(s, z);
00119
00120
00121 for(k = 0; k < n8; k++) {
00122 FFTSample r0, i0, r1, i1;
00123 CMUL(r0, i1, z[n8-k-1].im, z[n8-k-1].re, tsin[n8-k-1], tcos[n8-k-1]);
00124 CMUL(r1, i0, z[n8+k ].im, z[n8+k ].re, tsin[n8+k ], tcos[n8+k ]);
00125 z[n8-k-1].re = r0;
00126 z[n8-k-1].im = i0;
00127 z[n8+k ].re = r1;
00128 z[n8+k ].im = i1;
00129 }
00130 }
00131
00137 void ff_imdct_calc_c(FFTContext *s, FFTSample *output, const FFTSample *input)
00138 {
00139 int k;
00140 int n = 1 << s->mdct_bits;
00141 int n2 = n >> 1;
00142 int n4 = n >> 2;
00143
00144 ff_imdct_half_c(s, output+n4, input);
00145
00146 for(k = 0; k < n4; k++) {
00147 output[k] = -output[n2-k-1];
00148 output[n-k-1] = output[n2+k];
00149 }
00150 }
00151
00157 void ff_mdct_calc_c(FFTContext *s, FFTSample *out, const FFTSample *input)
00158 {
00159 int i, j, n, n8, n4, n2, n3;
00160 FFTDouble re, im;
00161 const uint16_t *revtab = s->revtab;
00162 const FFTSample *tcos = s->tcos;
00163 const FFTSample *tsin = s->tsin;
00164 FFTComplex *x = (FFTComplex *)out;
00165
00166 n = 1 << s->mdct_bits;
00167 n2 = n >> 1;
00168 n4 = n >> 2;
00169 n8 = n >> 3;
00170 n3 = 3 * n4;
00171
00172
00173 for(i=0;i<n8;i++) {
00174 re = RSCALE(-input[2*i+n3] - input[n3-1-2*i]);
00175 im = RSCALE(-input[n4+2*i] + input[n4-1-2*i]);
00176 j = revtab[i];
00177 CMUL(x[j].re, x[j].im, re, im, -tcos[i], tsin[i]);
00178
00179 re = RSCALE( input[2*i] - input[n2-1-2*i]);
00180 im = RSCALE(-input[n2+2*i] - input[ n-1-2*i]);
00181 j = revtab[n8 + i];
00182 CMUL(x[j].re, x[j].im, re, im, -tcos[n8 + i], tsin[n8 + i]);
00183 }
00184
00185 s->fft_calc(s, x);
00186
00187
00188 for(i=0;i<n8;i++) {
00189 FFTSample r0, i0, r1, i1;
00190 CMUL(i1, r0, x[n8-i-1].re, x[n8-i-1].im, -tsin[n8-i-1], -tcos[n8-i-1]);
00191 CMUL(i0, r1, x[n8+i ].re, x[n8+i ].im, -tsin[n8+i ], -tcos[n8+i ]);
00192 x[n8-i-1].re = r0;
00193 x[n8-i-1].im = i0;
00194 x[n8+i ].re = r1;
00195 x[n8+i ].im = i1;
00196 }
00197 }
00198
00199 av_cold void ff_mdct_end(FFTContext *s)
00200 {
00201 av_freep(&s->tcos);
00202 ff_fft_end(s);
00203 }