00001
00002
00003
00004
00005
00006
00007
00008
00009
00010
00011
00012
00013
00014
00015
00016
00017
00018
00019
00020
00021
00022
00023
00029 #include <stdlib.h>
00030 #include <string.h>
00031 #include "libavutil/mathematics.h"
00032 #include "fft.h"
00033
00034
00035 #if !CONFIG_HARDCODED_TABLES
00036 COSTABLE(16);
00037 COSTABLE(32);
00038 COSTABLE(64);
00039 COSTABLE(128);
00040 COSTABLE(256);
00041 COSTABLE(512);
00042 COSTABLE(1024);
00043 COSTABLE(2048);
00044 COSTABLE(4096);
00045 COSTABLE(8192);
00046 COSTABLE(16384);
00047 COSTABLE(32768);
00048 COSTABLE(65536);
00049 #endif
00050 COSTABLE_CONST FFTSample * const ff_cos_tabs[] = {
00051 NULL, NULL, NULL, NULL,
00052 ff_cos_16, ff_cos_32, ff_cos_64, ff_cos_128, ff_cos_256, ff_cos_512, ff_cos_1024,
00053 ff_cos_2048, ff_cos_4096, ff_cos_8192, ff_cos_16384, ff_cos_32768, ff_cos_65536,
00054 };
00055
00056 static void ff_fft_permute_c(FFTContext *s, FFTComplex *z);
00057 static void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
00058
00059 static int split_radix_permutation(int i, int n, int inverse)
00060 {
00061 int m;
00062 if(n <= 2) return i&1;
00063 m = n >> 1;
00064 if(!(i&m)) return split_radix_permutation(i, m, inverse)*2;
00065 m >>= 1;
00066 if(inverse == !(i&m)) return split_radix_permutation(i, m, inverse)*4 + 1;
00067 else return split_radix_permutation(i, m, inverse)*4 - 1;
00068 }
00069
00070 av_cold void ff_init_ff_cos_tabs(int index)
00071 {
00072 #if !CONFIG_HARDCODED_TABLES
00073 int i;
00074 int m = 1<<index;
00075 double freq = 2*M_PI/m;
00076 FFTSample *tab = ff_cos_tabs[index];
00077 for(i=0; i<=m/4; i++)
00078 tab[i] = cos(i*freq);
00079 for(i=1; i<m/4; i++)
00080 tab[m/2-i] = tab[i];
00081 #endif
00082 }
00083
00084 av_cold int ff_fft_init(FFTContext *s, int nbits, int inverse)
00085 {
00086 int i, j, n;
00087
00088 if (nbits < 2 || nbits > 16)
00089 goto fail;
00090 s->nbits = nbits;
00091 n = 1 << nbits;
00092
00093 s->revtab = av_malloc(n * sizeof(uint16_t));
00094 if (!s->revtab)
00095 goto fail;
00096 s->tmp_buf = av_malloc(n * sizeof(FFTComplex));
00097 if (!s->tmp_buf)
00098 goto fail;
00099 s->inverse = inverse;
00100 s->fft_permutation = FF_FFT_PERM_DEFAULT;
00101
00102 s->fft_permute = ff_fft_permute_c;
00103 s->fft_calc = ff_fft_calc_c;
00104 #if CONFIG_MDCT
00105 s->imdct_calc = ff_imdct_calc_c;
00106 s->imdct_half = ff_imdct_half_c;
00107 s->mdct_calc = ff_mdct_calc_c;
00108 #endif
00109
00110 if (ARCH_ARM) ff_fft_init_arm(s);
00111 if (HAVE_ALTIVEC) ff_fft_init_altivec(s);
00112 if (HAVE_MMX) ff_fft_init_mmx(s);
00113
00114 for(j=4; j<=nbits; j++) {
00115 ff_init_ff_cos_tabs(j);
00116 }
00117 for(i=0; i<n; i++) {
00118 int j = i;
00119 if (s->fft_permutation == FF_FFT_PERM_SWAP_LSBS)
00120 j = (j&~3) | ((j>>1)&1) | ((j<<1)&2);
00121 s->revtab[-split_radix_permutation(i, n, s->inverse) & (n-1)] = j;
00122 }
00123
00124 return 0;
00125 fail:
00126 av_freep(&s->revtab);
00127 av_freep(&s->tmp_buf);
00128 return -1;
00129 }
00130
00131 static void ff_fft_permute_c(FFTContext *s, FFTComplex *z)
00132 {
00133 int j, np;
00134 const uint16_t *revtab = s->revtab;
00135 np = 1 << s->nbits;
00136
00137 for(j=0;j<np;j++) s->tmp_buf[revtab[j]] = z[j];
00138 memcpy(z, s->tmp_buf, np * sizeof(FFTComplex));
00139 }
00140
00141 av_cold void ff_fft_end(FFTContext *s)
00142 {
00143 av_freep(&s->revtab);
00144 av_freep(&s->tmp_buf);
00145 }
00146
00147 #define sqrthalf (float)M_SQRT1_2
00148
00149 #define BF(x,y,a,b) {\
00150 x = a - b;\
00151 y = a + b;\
00152 }
00153
00154 #define BUTTERFLIES(a0,a1,a2,a3) {\
00155 BF(t3, t5, t5, t1);\
00156 BF(a2.re, a0.re, a0.re, t5);\
00157 BF(a3.im, a1.im, a1.im, t3);\
00158 BF(t4, t6, t2, t6);\
00159 BF(a3.re, a1.re, a1.re, t4);\
00160 BF(a2.im, a0.im, a0.im, t6);\
00161 }
00162
00163
00164
00165
00166 #define BUTTERFLIES_BIG(a0,a1,a2,a3) {\
00167 FFTSample r0=a0.re, i0=a0.im, r1=a1.re, i1=a1.im;\
00168 BF(t3, t5, t5, t1);\
00169 BF(a2.re, a0.re, r0, t5);\
00170 BF(a3.im, a1.im, i1, t3);\
00171 BF(t4, t6, t2, t6);\
00172 BF(a3.re, a1.re, r1, t4);\
00173 BF(a2.im, a0.im, i0, t6);\
00174 }
00175
00176 #define TRANSFORM(a0,a1,a2,a3,wre,wim) {\
00177 t1 = a2.re * wre + a2.im * wim;\
00178 t2 = a2.im * wre - a2.re * wim;\
00179 t5 = a3.re * wre - a3.im * wim;\
00180 t6 = a3.im * wre + a3.re * wim;\
00181 BUTTERFLIES(a0,a1,a2,a3)\
00182 }
00183
00184 #define TRANSFORM_ZERO(a0,a1,a2,a3) {\
00185 t1 = a2.re;\
00186 t2 = a2.im;\
00187 t5 = a3.re;\
00188 t6 = a3.im;\
00189 BUTTERFLIES(a0,a1,a2,a3)\
00190 }
00191
00192
00193 #define PASS(name)\
00194 static void name(FFTComplex *z, const FFTSample *wre, unsigned int n)\
00195 {\
00196 FFTSample t1, t2, t3, t4, t5, t6;\
00197 int o1 = 2*n;\
00198 int o2 = 4*n;\
00199 int o3 = 6*n;\
00200 const FFTSample *wim = wre+o1;\
00201 n--;\
00202 \
00203 TRANSFORM_ZERO(z[0],z[o1],z[o2],z[o3]);\
00204 TRANSFORM(z[1],z[o1+1],z[o2+1],z[o3+1],wre[1],wim[-1]);\
00205 do {\
00206 z += 2;\
00207 wre += 2;\
00208 wim -= 2;\
00209 TRANSFORM(z[0],z[o1],z[o2],z[o3],wre[0],wim[0]);\
00210 TRANSFORM(z[1],z[o1+1],z[o2+1],z[o3+1],wre[1],wim[-1]);\
00211 } while(--n);\
00212 }
00213
00214 PASS(pass)
00215 #undef BUTTERFLIES
00216 #define BUTTERFLIES BUTTERFLIES_BIG
00217 PASS(pass_big)
00218
00219 #define DECL_FFT(n,n2,n4)\
00220 static void fft##n(FFTComplex *z)\
00221 {\
00222 fft##n2(z);\
00223 fft##n4(z+n4*2);\
00224 fft##n4(z+n4*3);\
00225 pass(z,ff_cos_##n,n4/2);\
00226 }
00227
00228 static void fft4(FFTComplex *z)
00229 {
00230 FFTSample t1, t2, t3, t4, t5, t6, t7, t8;
00231
00232 BF(t3, t1, z[0].re, z[1].re);
00233 BF(t8, t6, z[3].re, z[2].re);
00234 BF(z[2].re, z[0].re, t1, t6);
00235 BF(t4, t2, z[0].im, z[1].im);
00236 BF(t7, t5, z[2].im, z[3].im);
00237 BF(z[3].im, z[1].im, t4, t8);
00238 BF(z[3].re, z[1].re, t3, t7);
00239 BF(z[2].im, z[0].im, t2, t5);
00240 }
00241
00242 static void fft8(FFTComplex *z)
00243 {
00244 FFTSample t1, t2, t3, t4, t5, t6, t7, t8;
00245
00246 fft4(z);
00247
00248 BF(t1, z[5].re, z[4].re, -z[5].re);
00249 BF(t2, z[5].im, z[4].im, -z[5].im);
00250 BF(t3, z[7].re, z[6].re, -z[7].re);
00251 BF(t4, z[7].im, z[6].im, -z[7].im);
00252 BF(t8, t1, t3, t1);
00253 BF(t7, t2, t2, t4);
00254 BF(z[4].re, z[0].re, z[0].re, t1);
00255 BF(z[4].im, z[0].im, z[0].im, t2);
00256 BF(z[6].re, z[2].re, z[2].re, t7);
00257 BF(z[6].im, z[2].im, z[2].im, t8);
00258
00259 TRANSFORM(z[1],z[3],z[5],z[7],sqrthalf,sqrthalf);
00260 }
00261
00262 #if !CONFIG_SMALL
00263 static void fft16(FFTComplex *z)
00264 {
00265 FFTSample t1, t2, t3, t4, t5, t6;
00266
00267 fft8(z);
00268 fft4(z+8);
00269 fft4(z+12);
00270
00271 TRANSFORM_ZERO(z[0],z[4],z[8],z[12]);
00272 TRANSFORM(z[2],z[6],z[10],z[14],sqrthalf,sqrthalf);
00273 TRANSFORM(z[1],z[5],z[9],z[13],ff_cos_16[1],ff_cos_16[3]);
00274 TRANSFORM(z[3],z[7],z[11],z[15],ff_cos_16[3],ff_cos_16[1]);
00275 }
00276 #else
00277 DECL_FFT(16,8,4)
00278 #endif
00279 DECL_FFT(32,16,8)
00280 DECL_FFT(64,32,16)
00281 DECL_FFT(128,64,32)
00282 DECL_FFT(256,128,64)
00283 DECL_FFT(512,256,128)
00284 #if !CONFIG_SMALL
00285 #define pass pass_big
00286 #endif
00287 DECL_FFT(1024,512,256)
00288 DECL_FFT(2048,1024,512)
00289 DECL_FFT(4096,2048,1024)
00290 DECL_FFT(8192,4096,2048)
00291 DECL_FFT(16384,8192,4096)
00292 DECL_FFT(32768,16384,8192)
00293 DECL_FFT(65536,32768,16384)
00294
00295 static void (* const fft_dispatch[])(FFTComplex*) = {
00296 fft4, fft8, fft16, fft32, fft64, fft128, fft256, fft512, fft1024,
00297 fft2048, fft4096, fft8192, fft16384, fft32768, fft65536,
00298 };
00299
00300 static void ff_fft_calc_c(FFTContext *s, FFTComplex *z)
00301 {
00302 fft_dispatch[s->nbits-2](z);
00303 }
00304