falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
65989 Commits
32 Branches
283MB
Tree: e2cd28c926
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'e2cd28c926'
${ noResults }
FFmpeg/libavcodec/fft_template.c

532 lines
15KB
Raw Blame History 

  1. /*

  2.  * FFT/IFFT transforms

  3.  * Copyright (c) 2008 Loren Merritt

  4.  * Copyright (c) 2002 Fabrice Bellard

  5.  * Partly based on libdjbfft by D. J. Bernstein

  6.  *

  7.  * This file is part of FFmpeg.

  8.  *

  9.  * FFmpeg is free software; you can redistribute it and/or

  10.  * modify it under the terms of the GNU Lesser General Public

  11.  * License as published by the Free Software Foundation; either

  12.  * version 2.1 of the License, or (at your option) any later version.

  13.  *

  14.  * FFmpeg is distributed in the hope that it will be useful,

  15.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  16.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  17.  * Lesser General Public License for more details.

  18.  *

  19.  * You should have received a copy of the GNU Lesser General Public

  20.  * License along with FFmpeg; if not, write to the Free Software

  21.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

  22.  */

  23. 

  24. /**

  25.  * @file

  26.  * FFT/IFFT transforms.

  27.  */

  28. 

  29. #include <stdlib.h>

  30. #include <string.h>

  31. #include "libavutil/mathematics.h"

  32. #include "fft.h"

  33. #include "fft-internal.h"

  34. 

  35. #if FFT_FIXED_32

  36. #include "fft_table.h"

  37. #else /* FFT_FIXED_32 */

  38. 

  39. /* cos(2*pi*x/n) for 0<=x<=n/4, followed by its reverse */

  40. #if !CONFIG_HARDCODED_TABLES

  41. COSTABLE(16);

  42. COSTABLE(32);

  43. COSTABLE(64);

  44. COSTABLE(128);

  45. COSTABLE(256);

  46. COSTABLE(512);

  47. COSTABLE(1024);

  48. COSTABLE(2048);

  49. COSTABLE(4096);

  50. COSTABLE(8192);

  51. COSTABLE(16384);

  52. COSTABLE(32768);

  53. COSTABLE(65536);

  54. #endif

  55. COSTABLE_CONST FFTSample * const FFT_NAME(ff_cos_tabs)[] = {

  56.     NULL, NULL, NULL, NULL,

  57.     FFT_NAME(ff_cos_16),

  58.     FFT_NAME(ff_cos_32),

  59.     FFT_NAME(ff_cos_64),

  60.     FFT_NAME(ff_cos_128),

  61.     FFT_NAME(ff_cos_256),

  62.     FFT_NAME(ff_cos_512),

  63.     FFT_NAME(ff_cos_1024),

  64.     FFT_NAME(ff_cos_2048),

  65.     FFT_NAME(ff_cos_4096),

  66.     FFT_NAME(ff_cos_8192),

  67.     FFT_NAME(ff_cos_16384),

  68.     FFT_NAME(ff_cos_32768),

  69.     FFT_NAME(ff_cos_65536),

  70. };

  71. 

  72. #endif /* FFT_FIXED_32 */

  73. 

  74. static void fft_permute_c(FFTContext *s, FFTComplex *z);

  75. static void fft_calc_c(FFTContext *s, FFTComplex *z);

  76. 

  77. static int split_radix_permutation(int i, int n, int inverse)

  78. {

  79.     int m;

  80.     if(n <= 2) return i&1;

  81.     m = n >> 1;

  82.     if(!(i&m))            return split_radix_permutation(i, m, inverse)*2;

  83.     m >>= 1;

  84.     if(inverse == !(i&m)) return split_radix_permutation(i, m, inverse)*4 + 1;

  85.     else                  return split_radix_permutation(i, m, inverse)*4 - 1;

  86. }

  87. 

  88. av_cold void ff_init_ff_cos_tabs(int index)

  89. {

  90. #if (!CONFIG_HARDCODED_TABLES) && (!FFT_FIXED_32)

  91.     int i;

  92.     int m = 1<<index;

  93.     double freq = 2*M_PI/m;

  94.     FFTSample *tab = FFT_NAME(ff_cos_tabs)[index];

  95.     for(i=0; i<=m/4; i++)

  96.         tab[i] = FIX15(cos(i*freq));

  97.     for(i=1; i<m/4; i++)

  98.         tab[m/2-i] = tab[i];

  99. #endif

  100. }

  101. 

  102. static const int avx_tab[] = {

  103.     0, 4, 1, 5, 8, 12, 9, 13, 2, 6, 3, 7, 10, 14, 11, 15

  104. };

  105. 

  106. static int is_second_half_of_fft32(int i, int n)

  107. {

  108.     if (n <= 32)

  109.         return i >= 16;

  110.     else if (i < n/2)

  111.         return is_second_half_of_fft32(i, n/2);

  112.     else if (i < 3*n/4)

  113.         return is_second_half_of_fft32(i - n/2, n/4);

  114.     else

  115.         return is_second_half_of_fft32(i - 3*n/4, n/4);

  116. }

  117. 

  118. static av_cold void fft_perm_avx(FFTContext *s)

  119. {

  120.     int i;

  121.     int n = 1 << s->nbits;

  122. 

  123.     for (i = 0; i < n; i += 16) {

  124.         int k;

  125.         if (is_second_half_of_fft32(i, n)) {

  126.             for (k = 0; k < 16; k++)

  127.                 s->revtab[-split_radix_permutation(i + k, n, s->inverse) & (n - 1)] =

  128.                     i + avx_tab[k];

  129. 

  130.         } else {

  131.             for (k = 0; k < 16; k++) {

  132.                 int j = i + k;

  133.                 j = (j & ~7) | ((j >> 1) & 3) | ((j << 2) & 4);

  134.                 s->revtab[-split_radix_permutation(i + k, n, s->inverse) & (n - 1)] = j;

  135.             }

  136.         }

  137.     }

  138. }

  139. 

  140. av_cold int ff_fft_init(FFTContext *s, int nbits, int inverse)

  141. {

  142.     int i, j, n;

  143. 

  144.     if (nbits < 2 || nbits > 16)

  145.         goto fail;

  146.     s->nbits = nbits;

  147.     n = 1 << nbits;

  148. 

  149.     s->revtab = av_malloc(n * sizeof(uint16_t));

  150.     if (!s->revtab)

  151.         goto fail;

  152.     s->tmp_buf = av_malloc(n * sizeof(FFTComplex));

  153.     if (!s->tmp_buf)

  154.         goto fail;

  155.     s->inverse = inverse;

  156.     s->fft_permutation = FF_FFT_PERM_DEFAULT;

  157. 

  158.     s->fft_permute = fft_permute_c;

  159.     s->fft_calc    = fft_calc_c;

  160. #if CONFIG_MDCT

  161.     s->imdct_calc  = ff_imdct_calc_c;

  162.     s->imdct_half  = ff_imdct_half_c;

  163.     s->mdct_calc   = ff_mdct_calc_c;

  164. #endif

  165. 

  166. #if FFT_FIXED_32

  167.     {

  168.         int n=0;

  169.         ff_fft_lut_init(fft_offsets_lut, 0, 1 << 16, &n);

  170.     }

  171. #else /* FFT_FIXED_32 */

  172. #if FFT_FLOAT

  173.     if (ARCH_AARCH64) ff_fft_init_aarch64(s);

  174.     if (ARCH_ARM)     ff_fft_init_arm(s);

  175.     if (ARCH_PPC)     ff_fft_init_ppc(s);

  176.     if (ARCH_X86)     ff_fft_init_x86(s);

  177.     if (CONFIG_MDCT)  s->mdct_calcw = s->mdct_calc;

  178.     if (HAVE_MIPSFPU) ff_fft_init_mips(s);

  179. #else

  180.     if (CONFIG_MDCT)  s->mdct_calcw = ff_mdct_calcw_c;

  181.     if (ARCH_ARM)     ff_fft_fixed_init_arm(s);

  182. #endif

  183.     for(j=4; j<=nbits; j++) {

  184.         ff_init_ff_cos_tabs(j);

  185.     }

  186. #endif /* FFT_FIXED_32 */

  187. 

  188. 

  189.     if (s->fft_permutation == FF_FFT_PERM_AVX) {

  190.         fft_perm_avx(s);

  191.     } else {

  192.         for(i=0; i<n; i++) {

  193.             j = i;

  194.             if (s->fft_permutation == FF_FFT_PERM_SWAP_LSBS)

  195.                 j = (j&~3) | ((j>>1)&1) | ((j<<1)&2);

  196.             s->revtab[-split_radix_permutation(i, n, s->inverse) & (n-1)] = j;

  197.         }

  198.     }

  199. 

  200.     return 0;

  201.  fail:

  202.     av_freep(&s->revtab);

  203.     av_freep(&s->tmp_buf);

  204.     return -1;

  205. }

  206. 

  207. static void fft_permute_c(FFTContext *s, FFTComplex *z)

  208. {

  209.     int j, np;

  210.     const uint16_t *revtab = s->revtab;

  211.     np = 1 << s->nbits;

  212.     /* TODO: handle split-radix permute in a more optimal way, probably in-place */

  213.     for(j=0;j<np;j++) s->tmp_buf[revtab[j]] = z[j];

  214.     memcpy(z, s->tmp_buf, np * sizeof(FFTComplex));

  215. }

  216. 

  217. av_cold void ff_fft_end(FFTContext *s)

  218. {

  219.     av_freep(&s->revtab);

  220.     av_freep(&s->tmp_buf);

  221. }

  222. 

  223. #if FFT_FIXED_32

  224. 

  225. static void fft_calc_c(FFTContext *s, FFTComplex *z) {

  226. 

  227.     int nbits, i, n, num_transforms, offset, step;

  228.     int n4, n2, n34;

  229.     FFTSample tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;

  230.     FFTComplex *tmpz;

  231.     FFTSample w_re, w_im;

  232.     const FFTSample *w_re_ptr, *w_im_ptr;

  233.     const int fft_size = (1 << s->nbits);

  234.     int64_t accu;

  235. 

  236.     num_transforms = (0x2aab >> (16 - s->nbits)) | 1;

  237. 

  238.     for (n=0; n<num_transforms; n++){

  239.         offset = fft_offsets_lut[n] << 2;

  240.         tmpz = z + offset;

  241. 

  242.         tmp1 = tmpz[0].re + tmpz[1].re;

  243.         tmp5 = tmpz[2].re + tmpz[3].re;

  244.         tmp2 = tmpz[0].im + tmpz[1].im;

  245.         tmp6 = tmpz[2].im + tmpz[3].im;

  246.         tmp3 = tmpz[0].re - tmpz[1].re;

  247.         tmp8 = tmpz[2].im - tmpz[3].im;

  248.         tmp4 = tmpz[0].im - tmpz[1].im;

  249.         tmp7 = tmpz[2].re - tmpz[3].re;

  250. 

  251.         tmpz[0].re = tmp1 + tmp5;

  252.         tmpz[2].re = tmp1 - tmp5;

  253.         tmpz[0].im = tmp2 + tmp6;

  254.         tmpz[2].im = tmp2 - tmp6;

  255.         tmpz[1].re = tmp3 + tmp8;

  256.         tmpz[3].re = tmp3 - tmp8;

  257.         tmpz[1].im = tmp4 - tmp7;

  258.         tmpz[3].im = tmp4 + tmp7;

  259.     }

  260. 

  261.     if (fft_size < 8)

  262.         return;

  263. 

  264.     num_transforms = (num_transforms >> 1) | 1;

  265. 

  266.     for (n=0; n<num_transforms; n++){

  267.         offset = fft_offsets_lut[n] << 3;

  268.         tmpz = z + offset;

  269. 

  270.         tmp1 = tmpz[4].re + tmpz[5].re;

  271.         tmp3 = tmpz[6].re + tmpz[7].re;

  272.         tmp2 = tmpz[4].im + tmpz[5].im;

  273.         tmp4 = tmpz[6].im + tmpz[7].im;

  274.         tmp5 = tmp1 + tmp3;

  275.         tmp7 = tmp1 - tmp3;

  276.         tmp6 = tmp2 + tmp4;

  277.         tmp8 = tmp2 - tmp4;

  278. 

  279.         tmp1 = tmpz[4].re - tmpz[5].re;

  280.         tmp2 = tmpz[4].im - tmpz[5].im;

  281.         tmp3 = tmpz[6].re - tmpz[7].re;

  282.         tmp4 = tmpz[6].im - tmpz[7].im;

  283. 

  284.         tmpz[4].re = tmpz[0].re - tmp5;

  285.         tmpz[0].re = tmpz[0].re + tmp5;

  286.         tmpz[4].im = tmpz[0].im - tmp6;

  287.         tmpz[0].im = tmpz[0].im + tmp6;

  288.         tmpz[6].re = tmpz[2].re - tmp8;

  289.         tmpz[2].re = tmpz[2].re + tmp8;

  290.         tmpz[6].im = tmpz[2].im + tmp7;

  291.         tmpz[2].im = tmpz[2].im - tmp7;

  292. 

  293.         accu = (int64_t)Q31(M_SQRT1_2)*(tmp1 + tmp2);

  294.         tmp5 = (int32_t)((accu + 0x40000000) >> 31);

  295.         accu = (int64_t)Q31(M_SQRT1_2)*(tmp3 - tmp4);

  296.         tmp7 = (int32_t)((accu + 0x40000000) >> 31);

  297.         accu = (int64_t)Q31(M_SQRT1_2)*(tmp2 - tmp1);

  298.         tmp6 = (int32_t)((accu + 0x40000000) >> 31);

  299.         accu = (int64_t)Q31(M_SQRT1_2)*(tmp3 + tmp4);

  300.         tmp8 = (int32_t)((accu + 0x40000000) >> 31);

  301.         tmp1 = tmp5 + tmp7;

  302.         tmp3 = tmp5 - tmp7;

  303.         tmp2 = tmp6 + tmp8;

  304.         tmp4 = tmp6 - tmp8;

  305. 

  306.         tmpz[5].re = tmpz[1].re - tmp1;

  307.         tmpz[1].re = tmpz[1].re + tmp1;

  308.         tmpz[5].im = tmpz[1].im - tmp2;

  309.         tmpz[1].im = tmpz[1].im + tmp2;

  310.         tmpz[7].re = tmpz[3].re - tmp4;

  311.         tmpz[3].re = tmpz[3].re + tmp4;

  312.         tmpz[7].im = tmpz[3].im + tmp3;

  313.         tmpz[3].im = tmpz[3].im - tmp3;

  314.     }

  315. 

  316.     step = 1 << ((MAX_LOG2_NFFT-4) - 4);

  317.     n4 = 4;

  318. 

  319.     for (nbits=4; nbits<=s->nbits; nbits++){

  320.         n2  = 2*n4;

  321.         n34 = 3*n4;

  322.         num_transforms = (num_transforms >> 1) | 1;

  323. 

  324.         for (n=0; n<num_transforms; n++){

  325.             offset = fft_offsets_lut[n] << nbits;

  326.             tmpz = z + offset;

  327. 

  328.             tmp5 = tmpz[ n2].re + tmpz[n34].re;

  329.             tmp1 = tmpz[ n2].re - tmpz[n34].re;

  330.             tmp6 = tmpz[ n2].im + tmpz[n34].im;

  331.             tmp2 = tmpz[ n2].im - tmpz[n34].im;

  332. 

  333.             tmpz[ n2].re = tmpz[ 0].re - tmp5;

  334.             tmpz[  0].re = tmpz[ 0].re + tmp5;

  335.             tmpz[ n2].im = tmpz[ 0].im - tmp6;

  336.             tmpz[  0].im = tmpz[ 0].im + tmp6;

  337.             tmpz[n34].re = tmpz[n4].re - tmp2;

  338.             tmpz[ n4].re = tmpz[n4].re + tmp2;

  339.             tmpz[n34].im = tmpz[n4].im + tmp1;

  340.             tmpz[ n4].im = tmpz[n4].im - tmp1;

  341. 

  342.             w_re_ptr = w_tab_sr + step;

  343.             w_im_ptr = w_tab_sr + MAX_FFT_SIZE/(4*16) - step;

  344. 

  345.             for (i=1; i<n4; i++){

  346.                 w_re = w_re_ptr[0];

  347.                 w_im = w_im_ptr[0];

  348.                 accu  = (int64_t)w_re*tmpz[ n2+i].re;

  349.                 accu += (int64_t)w_im*tmpz[ n2+i].im;

  350.                 tmp1 = (int32_t)((accu + 0x40000000) >> 31);

  351.                 accu  = (int64_t)w_re*tmpz[ n2+i].im;

  352.                 accu -= (int64_t)w_im*tmpz[ n2+i].re;

  353.                 tmp2 = (int32_t)((accu + 0x40000000) >> 31);

  354.                 accu  = (int64_t)w_re*tmpz[n34+i].re;

  355.                 accu -= (int64_t)w_im*tmpz[n34+i].im;

  356.                 tmp3 = (int32_t)((accu + 0x40000000) >> 31);

  357.                 accu  = (int64_t)w_re*tmpz[n34+i].im;

  358.                 accu += (int64_t)w_im*tmpz[n34+i].re;

  359.                 tmp4 = (int32_t)((accu + 0x40000000) >> 31);

  360. 

  361.                 tmp5 = tmp1 + tmp3;

  362.                 tmp1 = tmp1 - tmp3;

  363.                 tmp6 = tmp2 + tmp4;

  364.                 tmp2 = tmp2 - tmp4;

  365. 

  366.                 tmpz[ n2+i].re = tmpz[   i].re - tmp5;

  367.                 tmpz[    i].re = tmpz[   i].re + tmp5;

  368.                 tmpz[ n2+i].im = tmpz[   i].im - tmp6;

  369.                 tmpz[    i].im = tmpz[   i].im + tmp6;

  370.                 tmpz[n34+i].re = tmpz[n4+i].re - tmp2;

  371.                 tmpz[ n4+i].re = tmpz[n4+i].re + tmp2;

  372.                 tmpz[n34+i].im = tmpz[n4+i].im + tmp1;

  373.                 tmpz[ n4+i].im = tmpz[n4+i].im - tmp1;

  374. 

  375.                 w_re_ptr += step;

  376.                 w_im_ptr -= step;

  377.             }

  378.         }

  379.         step >>= 1;

  380.         n4   <<= 1;

  381.     }

  382. }

  383. 

  384. #else /* FFT_FIXED_32 */

  385. 

  386. #define BUTTERFLIES(a0,a1,a2,a3) {\

  387.     BF(t3, t5, t5, t1);\

  388.     BF(a2.re, a0.re, a0.re, t5);\

  389.     BF(a3.im, a1.im, a1.im, t3);\

  390.     BF(t4, t6, t2, t6);\

  391.     BF(a3.re, a1.re, a1.re, t4);\

  392.     BF(a2.im, a0.im, a0.im, t6);\

  393. }

  394. 

  395. // force loading all the inputs before storing any.

  396. // this is slightly slower for small data, but avoids store->load aliasing

  397. // for addresses separated by large powers of 2.

  398. #define BUTTERFLIES_BIG(a0,a1,a2,a3) {\

  399.     FFTSample r0=a0.re, i0=a0.im, r1=a1.re, i1=a1.im;\

  400.     BF(t3, t5, t5, t1);\

  401.     BF(a2.re, a0.re, r0, t5);\

  402.     BF(a3.im, a1.im, i1, t3);\

  403.     BF(t4, t6, t2, t6);\

  404.     BF(a3.re, a1.re, r1, t4);\

  405.     BF(a2.im, a0.im, i0, t6);\

  406. }

  407. 

  408. #define TRANSFORM(a0,a1,a2,a3,wre,wim) {\

  409.     CMUL(t1, t2, a2.re, a2.im, wre, -wim);\

  410.     CMUL(t5, t6, a3.re, a3.im, wre,  wim);\

  411.     BUTTERFLIES(a0,a1,a2,a3)\

  412. }

  413. 

  414. #define TRANSFORM_ZERO(a0,a1,a2,a3) {\

  415.     t1 = a2.re;\

  416.     t2 = a2.im;\

  417.     t5 = a3.re;\

  418.     t6 = a3.im;\

  419.     BUTTERFLIES(a0,a1,a2,a3)\

  420. }

  421. 

  422. /* z[0...8n-1], w[1...2n-1] */

  423. #define PASS(name)\

  424. static void name(FFTComplex *z, const FFTSample *wre, unsigned int n)\

  425. {\

  426.     FFTDouble t1, t2, t3, t4, t5, t6;\

  427.     int o1 = 2*n;\

  428.     int o2 = 4*n;\

  429.     int o3 = 6*n;\

  430.     const FFTSample *wim = wre+o1;\

  431.     n--;\

  432. \

  433.     TRANSFORM_ZERO(z[0],z[o1],z[o2],z[o3]);\

  434.     TRANSFORM(z[1],z[o1+1],z[o2+1],z[o3+1],wre[1],wim[-1]);\

  435.     do {\

  436.         z += 2;\

  437.         wre += 2;\

  438.         wim -= 2;\

  439.         TRANSFORM(z[0],z[o1],z[o2],z[o3],wre[0],wim[0]);\

  440.         TRANSFORM(z[1],z[o1+1],z[o2+1],z[o3+1],wre[1],wim[-1]);\

  441.     } while(--n);\

  442. }

  443. 

  444. PASS(pass)

  445. #undef BUTTERFLIES

  446. #define BUTTERFLIES BUTTERFLIES_BIG

  447. PASS(pass_big)

  448. 

  449. #define DECL_FFT(n,n2,n4)\

  450. static void fft##n(FFTComplex *z)\

  451. {\

  452.     fft##n2(z);\

  453.     fft##n4(z+n4*2);\

  454.     fft##n4(z+n4*3);\

  455.     pass(z,FFT_NAME(ff_cos_##n),n4/2);\

  456. }

  457. 

  458. static void fft4(FFTComplex *z)

  459. {

  460.     FFTDouble t1, t2, t3, t4, t5, t6, t7, t8;

  461. 

  462.     BF(t3, t1, z[0].re, z[1].re);

  463.     BF(t8, t6, z[3].re, z[2].re);

  464.     BF(z[2].re, z[0].re, t1, t6);

  465.     BF(t4, t2, z[0].im, z[1].im);

  466.     BF(t7, t5, z[2].im, z[3].im);

  467.     BF(z[3].im, z[1].im, t4, t8);

  468.     BF(z[3].re, z[1].re, t3, t7);

  469.     BF(z[2].im, z[0].im, t2, t5);

  470. }

  471. 

  472. static void fft8(FFTComplex *z)

  473. {

  474.     FFTDouble t1, t2, t3, t4, t5, t6;

  475. 

  476.     fft4(z);

  477. 

  478.     BF(t1, z[5].re, z[4].re, -z[5].re);

  479.     BF(t2, z[5].im, z[4].im, -z[5].im);

  480.     BF(t5, z[7].re, z[6].re, -z[7].re);

  481.     BF(t6, z[7].im, z[6].im, -z[7].im);

  482. 

  483.     BUTTERFLIES(z[0],z[2],z[4],z[6]);

  484.     TRANSFORM(z[1],z[3],z[5],z[7],sqrthalf,sqrthalf);

  485. }

  486. 

  487. #if !CONFIG_SMALL

  488. static void fft16(FFTComplex *z)

  489. {

  490.     FFTDouble t1, t2, t3, t4, t5, t6;

  491.     FFTSample cos_16_1 = FFT_NAME(ff_cos_16)[1];

  492.     FFTSample cos_16_3 = FFT_NAME(ff_cos_16)[3];

  493. 

  494.     fft8(z);

  495.     fft4(z+8);

  496.     fft4(z+12);

  497. 

  498.     TRANSFORM_ZERO(z[0],z[4],z[8],z[12]);

  499.     TRANSFORM(z[2],z[6],z[10],z[14],sqrthalf,sqrthalf);

  500.     TRANSFORM(z[1],z[5],z[9],z[13],cos_16_1,cos_16_3);

  501.     TRANSFORM(z[3],z[7],z[11],z[15],cos_16_3,cos_16_1);

  502. }

  503. #else

  504. DECL_FFT(16,8,4)

  505. #endif

  506. DECL_FFT(32,16,8)

  507. DECL_FFT(64,32,16)

  508. DECL_FFT(128,64,32)

  509. DECL_FFT(256,128,64)

  510. DECL_FFT(512,256,128)

  511. #if !CONFIG_SMALL

  512. #define pass pass_big

  513. #endif

  514. DECL_FFT(1024,512,256)

  515. DECL_FFT(2048,1024,512)

  516. DECL_FFT(4096,2048,1024)

  517. DECL_FFT(8192,4096,2048)

  518. DECL_FFT(16384,8192,4096)

  519. DECL_FFT(32768,16384,8192)

  520. DECL_FFT(65536,32768,16384)

  521. 

  522. static void (* const fft_dispatch[])(FFTComplex*) = {

  523.     fft4, fft8, fft16, fft32, fft64, fft128, fft256, fft512, fft1024,

  524.     fft2048, fft4096, fft8192, fft16384, fft32768, fft65536,

  525. };

  526. 

  527. static void fft_calc_c(FFTContext *s, FFTComplex *z)

  528. {

  529.     fft_dispatch[s->nbits-2](z);

  530. }

  531. #endif /* FFT_FIXED_32 */