FFmpeg coverage


Directory: ../../../ffmpeg/
File: src/libavcodec/fft_template.c
Date: 2021-09-23 20:34:37
Exec Total Coverage
Lines: 257 277 92.8%
Branches: 60 78 76.9%

Line Branch Exec Source
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 "libavutil/thread.h"
33 #include "fft.h"
34 #include "fft-internal.h"
35
36 #if !FFT_FLOAT
37 #include "fft_table.h"
38 #else /* !FFT_FLOAT */
39
40 /* cos(2*pi*x/n) for 0<=x<=n/4, followed by its reverse */
41 #if !CONFIG_HARDCODED_TABLES
42 COSTABLE(16);
43 COSTABLE(32);
44 COSTABLE(64);
45 COSTABLE(128);
46 COSTABLE(256);
47 COSTABLE(512);
48 COSTABLE(1024);
49 COSTABLE(2048);
50 COSTABLE(4096);
51 COSTABLE(8192);
52 COSTABLE(16384);
53 COSTABLE(32768);
54 COSTABLE(65536);
55 COSTABLE(131072);
56
57 2462 static av_cold void init_ff_cos_tabs(int index)
58 {
59 int i;
60 2462 int m = 1<<index;
61 2462 double freq = 2*M_PI/m;
62 2462 FFTSample *tab = FFT_NAME(ff_cos_tabs)[index];
63
2/2
✓ Branch 0 taken 179250 times.
✓ Branch 1 taken 2462 times.
181712 for(i=0; i<=m/4; i++)
64 179250 tab[i] = FIX15(cos(i*freq));
65
2/2
✓ Branch 0 taken 174326 times.
✓ Branch 1 taken 2462 times.
176788 for(i=1; i<m/4; i++)
66 174326 tab[m/2-i] = tab[i];
67 2462 }
68
69 typedef struct CosTabsInitOnce {
70 void (*func)(void);
71 AVOnce control;
72 } CosTabsInitOnce;
73
74 #define INIT_FF_COS_TABS_FUNC(index, size) \
75 static av_cold void init_ff_cos_tabs_ ## size (void)\
76 { \
77 init_ff_cos_tabs(index); \
78 }
79
80 485 INIT_FF_COS_TABS_FUNC(4, 16)
81 468 INIT_FF_COS_TABS_FUNC(5, 32)
82 394 INIT_FF_COS_TABS_FUNC(6, 64)
83 370 INIT_FF_COS_TABS_FUNC(7, 128)
84 314 INIT_FF_COS_TABS_FUNC(8, 256)
85 290 INIT_FF_COS_TABS_FUNC(9, 512)
86 68 INIT_FF_COS_TABS_FUNC(10, 1024)
87 37 INIT_FF_COS_TABS_FUNC(11, 2048)
88 24 INIT_FF_COS_TABS_FUNC(12, 4096)
89 7 INIT_FF_COS_TABS_FUNC(13, 8192)
90 5 INIT_FF_COS_TABS_FUNC(14, 16384)
91 INIT_FF_COS_TABS_FUNC(15, 32768)
92 INIT_FF_COS_TABS_FUNC(16, 65536)
93 INIT_FF_COS_TABS_FUNC(17, 131072)
94
95 static CosTabsInitOnce cos_tabs_init_once[] = {
96 { NULL },
97 { NULL },
98 { NULL },
99 { NULL },
100 { init_ff_cos_tabs_16, AV_ONCE_INIT },
101 { init_ff_cos_tabs_32, AV_ONCE_INIT },
102 { init_ff_cos_tabs_64, AV_ONCE_INIT },
103 { init_ff_cos_tabs_128, AV_ONCE_INIT },
104 { init_ff_cos_tabs_256, AV_ONCE_INIT },
105 { init_ff_cos_tabs_512, AV_ONCE_INIT },
106 { init_ff_cos_tabs_1024, AV_ONCE_INIT },
107 { init_ff_cos_tabs_2048, AV_ONCE_INIT },
108 { init_ff_cos_tabs_4096, AV_ONCE_INIT },
109 { init_ff_cos_tabs_8192, AV_ONCE_INIT },
110 { init_ff_cos_tabs_16384, AV_ONCE_INIT },
111 { init_ff_cos_tabs_32768, AV_ONCE_INIT },
112 { init_ff_cos_tabs_65536, AV_ONCE_INIT },
113 { init_ff_cos_tabs_131072, AV_ONCE_INIT },
114 };
115
116 10653 av_cold void ff_init_ff_cos_tabs(int index)
117 {
118 10653 ff_thread_once(&cos_tabs_init_once[index].control, cos_tabs_init_once[index].func);
119 10653 }
120 #endif
121 COSTABLE_CONST FFTSample * const FFT_NAME(ff_cos_tabs)[] = {
122 NULL, NULL, NULL, NULL,
123 FFT_NAME(ff_cos_16),
124 FFT_NAME(ff_cos_32),
125 FFT_NAME(ff_cos_64),
126 FFT_NAME(ff_cos_128),
127 FFT_NAME(ff_cos_256),
128 FFT_NAME(ff_cos_512),
129 FFT_NAME(ff_cos_1024),
130 FFT_NAME(ff_cos_2048),
131 FFT_NAME(ff_cos_4096),
132 FFT_NAME(ff_cos_8192),
133 FFT_NAME(ff_cos_16384),
134 FFT_NAME(ff_cos_32768),
135 FFT_NAME(ff_cos_65536),
136 FFT_NAME(ff_cos_131072),
137 };
138
139 #endif /* FFT_FLOAT */
140
141 static void fft_permute_c(FFTContext *s, FFTComplex *z);
142 static void fft_calc_c(FFTContext *s, FFTComplex *z);
143
144 5189724 static int split_radix_permutation(int i, int n, int inverse)
145 {
146 int m;
147
2/2
✓ Branch 0 taken 767612 times.
✓ Branch 1 taken 4422112 times.
5189724 if(n <= 2) return i&1;
148 4422112 m = n >> 1;
149
2/2
✓ Branch 0 taken 2211056 times.
✓ Branch 1 taken 2211056 times.
4422112 if(!(i&m)) return split_radix_permutation(i, m, inverse)*2;
150 2211056 m >>= 1;
151
2/2
✓ Branch 0 taken 1105528 times.
✓ Branch 1 taken 1105528 times.
2211056 if(inverse == !(i&m)) return split_radix_permutation(i, m, inverse)*4 + 1;
152 1105528 else return split_radix_permutation(i, m, inverse)*4 - 1;
153 }
154
155
156 static const int avx_tab[] = {
157 0, 4, 1, 5, 8, 12, 9, 13, 2, 6, 3, 7, 10, 14, 11, 15
158 };
159
160 16036 static int is_second_half_of_fft32(int i, int n)
161 {
162
2/2
✓ Branch 0 taken 4572 times.
✓ Branch 1 taken 11464 times.
16036 if (n <= 32)
163 4572 return i >= 16;
164
2/2
✓ Branch 0 taken 5732 times.
✓ Branch 1 taken 5732 times.
11464 else if (i < n/2)
165 5732 return is_second_half_of_fft32(i, n/2);
166
2/2
✓ Branch 0 taken 2866 times.
✓ Branch 1 taken 2866 times.
5732 else if (i < 3*n/4)
167 2866 return is_second_half_of_fft32(i - n/2, n/4);
168 else
169 2866 return is_second_half_of_fft32(i - 3*n/4, n/4);
170 }
171
172 360 static av_cold void fft_perm_avx(FFTContext *s)
173 {
174 int i;
175 360 int n = 1 << s->nbits;
176
177
2/2
✓ Branch 0 taken 4572 times.
✓ Branch 1 taken 360 times.
4932 for (i = 0; i < n; i += 16) {
178 int k;
179
2/2
✓ Branch 1 taken 1568 times.
✓ Branch 2 taken 3004 times.
4572 if (is_second_half_of_fft32(i, n)) {
180
2/2
✓ Branch 0 taken 25088 times.
✓ Branch 1 taken 1568 times.
26656 for (k = 0; k < 16; k++)
181 50176 s->revtab[-split_radix_permutation(i + k, n, s->inverse) & (n - 1)] =
182 25088 i + avx_tab[k];
183
184 } else {
185
2/2
✓ Branch 0 taken 48064 times.
✓ Branch 1 taken 3004 times.
51068 for (k = 0; k < 16; k++) {
186 48064 int j = i + k;
187 48064 j = (j & ~7) | ((j >> 1) & 3) | ((j << 2) & 4);
188 48064 s->revtab[-split_radix_permutation(i + k, n, s->inverse) & (n - 1)] = j;
189 }
190 }
191 }
192 360 }
193
194 3666 av_cold int ff_fft_init(FFTContext *s, int nbits, int inverse)
195 {
196 int i, j, n;
197
198 3666 s->revtab = NULL;
199 3666 s->revtab32 = NULL;
200
201
2/4
✓ Branch 0 taken 3666 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 3666 times.
3666 if (nbits < 2 || nbits > 17)
202 goto fail;
203 3666 s->nbits = nbits;
204 3666 n = 1 << nbits;
205
206
1/2
✓ Branch 0 taken 3666 times.
✗ Branch 1 not taken.
3666 if (nbits <= 16) {
207 3666 s->revtab = av_malloc(n * sizeof(uint16_t));
208
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 3666 times.
3666 if (!s->revtab)
209 goto fail;
210 } else {
211 s->revtab32 = av_malloc(n * sizeof(uint32_t));
212 if (!s->revtab32)
213 goto fail;
214 }
215 3666 s->tmp_buf = av_malloc(n * sizeof(FFTComplex));
216
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 3666 times.
3666 if (!s->tmp_buf)
217 goto fail;
218 3666 s->inverse = inverse;
219 3666 s->fft_permutation = FF_FFT_PERM_DEFAULT;
220
221 3666 s->fft_permute = fft_permute_c;
222 3666 s->fft_calc = fft_calc_c;
223 #if CONFIG_MDCT
224 3666 s->imdct_calc = ff_imdct_calc_c;
225 3666 s->imdct_half = ff_imdct_half_c;
226 3666 s->mdct_calc = ff_mdct_calc_c;
227 #endif
228
229 #if FFT_FLOAT
230 if (ARCH_AARCH64) ff_fft_init_aarch64(s);
231 if (ARCH_ARM) ff_fft_init_arm(s);
232 if (ARCH_PPC) ff_fft_init_ppc(s);
233 3478 if (ARCH_X86) ff_fft_init_x86(s);
234 if (HAVE_MIPSFPU) ff_fft_init_mips(s);
235
2/2
✓ Branch 0 taken 10474 times.
✓ Branch 1 taken 3478 times.
13952 for(j=4; j<=nbits; j++) {
236 10474 ff_init_ff_cos_tabs(j);
237 }
238 #else /* FFT_FLOAT */
239 188 ff_fft_lut_init();
240 #endif
241
242
243
2/2
✓ Branch 0 taken 360 times.
✓ Branch 1 taken 3118 times.
3478 if (ARCH_X86 && FFT_FLOAT && s->fft_permutation == FF_FFT_PERM_AVX) {
244 360 fft_perm_avx(s);
245 } else {
246 #define PROCESS_FFT_PERM_SWAP_LSBS(num) do {\
247 for(i = 0; i < n; i++) {\
248 int k;\
249 j = i;\
250 j = (j & ~3) | ((j >> 1) & 1) | ((j << 1) & 2);\
251 k = -split_radix_permutation(i, n, s->inverse) & (n - 1);\
252 s->revtab##num[k] = j;\
253 } \
254 } while(0);
255
256 #define PROCESS_FFT_PERM_DEFAULT(num) do {\
257 for(i = 0; i < n; i++) {\
258 int k;\
259 j = i;\
260 k = -split_radix_permutation(i, n, s->inverse) & (n - 1);\
261 s->revtab##num[k] = j;\
262 } \
263 } while(0);
264
265 #define SPLIT_RADIX_PERMUTATION(num) do { \
266 if (s->fft_permutation == FF_FFT_PERM_SWAP_LSBS) {\
267 PROCESS_FFT_PERM_SWAP_LSBS(num) \
268 } else {\
269 PROCESS_FFT_PERM_DEFAULT(num) \
270 }\
271 } while(0);
272
273
1/2
✓ Branch 0 taken 3306 times.
✗ Branch 1 not taken.
3306 if (s->revtab)
274
6/6
✓ Branch 0 taken 132 times.
✓ Branch 1 taken 3174 times.
✓ Branch 3 taken 1352 times.
✓ Branch 4 taken 132 times.
✓ Branch 6 taken 693108 times.
✓ Branch 7 taken 3174 times.
697766 SPLIT_RADIX_PERMUTATION()
275
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 3306 times.
3306 if (s->revtab32)
276 SPLIT_RADIX_PERMUTATION(32)
277
278 #undef PROCESS_FFT_PERM_DEFAULT
279 #undef PROCESS_FFT_PERM_SWAP_LSBS
280 #undef SPLIT_RADIX_PERMUTATION
281 }
282
283 3666 return 0;
284 fail:
285 av_freep(&s->revtab);
286 av_freep(&s->revtab32);
287 av_freep(&s->tmp_buf);
288 return -1;
289 }
290
291 48062 static void fft_permute_c(FFTContext *s, FFTComplex *z)
292 {
293 int j, np;
294 48062 const uint16_t *revtab = s->revtab;
295 48062 const uint32_t *revtab32 = s->revtab32;
296 48062 np = 1 << s->nbits;
297 /* TODO: handle split-radix permute in a more optimal way, probably in-place */
298
1/2
✓ Branch 0 taken 48062 times.
✗ Branch 1 not taken.
48062 if (revtab) {
299
2/2
✓ Branch 0 taken 12857920 times.
✓ Branch 1 taken 48062 times.
12905982 for(j=0;j<np;j++) s->tmp_buf[revtab[j]] = z[j];
300 } else
301 for(j=0;j<np;j++) s->tmp_buf[revtab32[j]] = z[j];
302
303 48062 memcpy(z, s->tmp_buf, np * sizeof(FFTComplex));
304 48062 }
305
306 3760 av_cold void ff_fft_end(FFTContext *s)
307 {
308 3760 av_freep(&s->revtab);
309 3760 av_freep(&s->revtab32);
310 3760 av_freep(&s->tmp_buf);
311 3760 }
312
313 #if !FFT_FLOAT
314
315 770822 static void fft_calc_c(FFTContext *s, FFTComplex *z) {
316
317 int nbits, i, n, num_transforms, offset, step;
318 int n4, n2, n34;
319 unsigned tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
320 FFTComplex *tmpz;
321 770822 const int fft_size = (1 << s->nbits);
322 int64_t accu;
323
324 770822 num_transforms = (0x2aab >> (16 - s->nbits)) | 1;
325
326
2/2
✓ Branch 0 taken 5842712 times.
✓ Branch 1 taken 770822 times.
6613534 for (n=0; n<num_transforms; n++){
327 5842712 offset = ff_fft_offsets_lut[n] << 2;
328 5842712 tmpz = z + offset;
329
330 5842712 tmp1 = tmpz[0].re + (unsigned)tmpz[1].re;
331 5842712 tmp5 = tmpz[2].re + (unsigned)tmpz[3].re;
332 5842712 tmp2 = tmpz[0].im + (unsigned)tmpz[1].im;
333 5842712 tmp6 = tmpz[2].im + (unsigned)tmpz[3].im;
334 5842712 tmp3 = tmpz[0].re - (unsigned)tmpz[1].re;
335 5842712 tmp8 = tmpz[2].im - (unsigned)tmpz[3].im;
336 5842712 tmp4 = tmpz[0].im - (unsigned)tmpz[1].im;
337 5842712 tmp7 = tmpz[2].re - (unsigned)tmpz[3].re;
338
339 5842712 tmpz[0].re = tmp1 + tmp5;
340 5842712 tmpz[2].re = tmp1 - tmp5;
341 5842712 tmpz[0].im = tmp2 + tmp6;
342 5842712 tmpz[2].im = tmp2 - tmp6;
343 5842712 tmpz[1].re = tmp3 + tmp8;
344 5842712 tmpz[3].re = tmp3 - tmp8;
345 5842712 tmpz[1].im = tmp4 - tmp7;
346 5842712 tmpz[3].im = tmp4 + tmp7;
347 }
348
349
2/2
✓ Branch 0 taken 1 times.
✓ Branch 1 taken 770821 times.
770822 if (fft_size < 8)
350 1 return;
351
352 770821 num_transforms = (num_transforms >> 1) | 1;
353
354
2/2
✓ Branch 0 taken 3289151 times.
✓ Branch 1 taken 770821 times.
4059972 for (n=0; n<num_transforms; n++){
355 3289151 offset = ff_fft_offsets_lut[n] << 3;
356 3289151 tmpz = z + offset;
357
358 3289151 tmp1 = tmpz[4].re + (unsigned)tmpz[5].re;
359 3289151 tmp3 = tmpz[6].re + (unsigned)tmpz[7].re;
360 3289151 tmp2 = tmpz[4].im + (unsigned)tmpz[5].im;
361 3289151 tmp4 = tmpz[6].im + (unsigned)tmpz[7].im;
362 3289151 tmp5 = tmp1 + tmp3;
363 3289151 tmp7 = tmp1 - tmp3;
364 3289151 tmp6 = tmp2 + tmp4;
365 3289151 tmp8 = tmp2 - tmp4;
366
367 3289151 tmp1 = tmpz[4].re - (unsigned)tmpz[5].re;
368 3289151 tmp2 = tmpz[4].im - (unsigned)tmpz[5].im;
369 3289151 tmp3 = tmpz[6].re - (unsigned)tmpz[7].re;
370 3289151 tmp4 = tmpz[6].im - (unsigned)tmpz[7].im;
371
372 3289151 tmpz[4].re = tmpz[0].re - tmp5;
373 3289151 tmpz[0].re = tmpz[0].re + tmp5;
374 3289151 tmpz[4].im = tmpz[0].im - tmp6;
375 3289151 tmpz[0].im = tmpz[0].im + tmp6;
376 3289151 tmpz[6].re = tmpz[2].re - tmp8;
377 3289151 tmpz[2].re = tmpz[2].re + tmp8;
378 3289151 tmpz[6].im = tmpz[2].im + tmp7;
379 3289151 tmpz[2].im = tmpz[2].im - tmp7;
380
381 3289151 accu = (int64_t)Q31(M_SQRT1_2)*(int)(tmp1 + tmp2);
382 3289151 tmp5 = (int32_t)((accu + 0x40000000) >> 31);
383 3289151 accu = (int64_t)Q31(M_SQRT1_2)*(int)(tmp3 - tmp4);
384 3289151 tmp7 = (int32_t)((accu + 0x40000000) >> 31);
385 3289151 accu = (int64_t)Q31(M_SQRT1_2)*(int)(tmp2 - tmp1);
386 3289151 tmp6 = (int32_t)((accu + 0x40000000) >> 31);
387 3289151 accu = (int64_t)Q31(M_SQRT1_2)*(int)(tmp3 + tmp4);
388 3289151 tmp8 = (int32_t)((accu + 0x40000000) >> 31);
389 3289151 tmp1 = tmp5 + tmp7;
390 3289151 tmp3 = tmp5 - tmp7;
391 3289151 tmp2 = tmp6 + tmp8;
392 3289151 tmp4 = tmp6 - tmp8;
393
394 3289151 tmpz[5].re = tmpz[1].re - tmp1;
395 3289151 tmpz[1].re = tmpz[1].re + tmp1;
396 3289151 tmpz[5].im = tmpz[1].im - tmp2;
397 3289151 tmpz[1].im = tmpz[1].im + tmp2;
398 3289151 tmpz[7].re = tmpz[3].re - tmp4;
399 3289151 tmpz[3].re = tmpz[3].re + tmp4;
400 3289151 tmpz[7].im = tmpz[3].im + tmp3;
401 3289151 tmpz[3].im = tmpz[3].im - tmp3;
402 }
403
404 770821 step = 1 << ((MAX_LOG2_NFFT-4) - 4);
405 770821 n4 = 4;
406
407
2/2
✓ Branch 0 taken 1670377 times.
✓ Branch 1 taken 770821 times.
2441198 for (nbits=4; nbits<=s->nbits; nbits++){
408 1670377 n2 = 2*n4;
409 1670377 n34 = 3*n4;
410 1670377 num_transforms = (num_transforms >> 1) | 1;
411
412
2/2
✓ Branch 0 taken 2535945 times.
✓ Branch 1 taken 1670377 times.
4206322 for (n=0; n<num_transforms; n++){
413 2535945 const FFTSample *w_re_ptr = ff_w_tab_sr + step;
414 2535945 const FFTSample *w_im_ptr = ff_w_tab_sr + MAX_FFT_SIZE/(4*16) - step;
415 2535945 offset = ff_fft_offsets_lut[n] << nbits;
416 2535945 tmpz = z + offset;
417
418 2535945 tmp5 = tmpz[ n2].re + (unsigned)tmpz[n34].re;
419 2535945 tmp1 = tmpz[ n2].re - (unsigned)tmpz[n34].re;
420 2535945 tmp6 = tmpz[ n2].im + (unsigned)tmpz[n34].im;
421 2535945 tmp2 = tmpz[ n2].im - (unsigned)tmpz[n34].im;
422
423 2535945 tmpz[ n2].re = tmpz[ 0].re - tmp5;
424 2535945 tmpz[ 0].re = tmpz[ 0].re + tmp5;
425 2535945 tmpz[ n2].im = tmpz[ 0].im - tmp6;
426 2535945 tmpz[ 0].im = tmpz[ 0].im + tmp6;
427 2535945 tmpz[n34].re = tmpz[n4].re - tmp2;
428 2535945 tmpz[ n4].re = tmpz[n4].re + tmp2;
429 2535945 tmpz[n34].im = tmpz[n4].im + tmp1;
430 2535945 tmpz[ n4].im = tmpz[n4].im - tmp1;
431
432
2/2
✓ Branch 0 taken 19082127 times.
✓ Branch 1 taken 2535945 times.
21618072 for (i=1; i<n4; i++){
433 19082127 FFTSample w_re = w_re_ptr[0];
434 19082127 FFTSample w_im = w_im_ptr[0];
435 19082127 accu = (int64_t)w_re*tmpz[ n2+i].re;
436 19082127 accu += (int64_t)w_im*tmpz[ n2+i].im;
437 19082127 tmp1 = (int32_t)((accu + 0x40000000) >> 31);
438 19082127 accu = (int64_t)w_re*tmpz[ n2+i].im;
439 19082127 accu -= (int64_t)w_im*tmpz[ n2+i].re;
440 19082127 tmp2 = (int32_t)((accu + 0x40000000) >> 31);
441 19082127 accu = (int64_t)w_re*tmpz[n34+i].re;
442 19082127 accu -= (int64_t)w_im*tmpz[n34+i].im;
443 19082127 tmp3 = (int32_t)((accu + 0x40000000) >> 31);
444 19082127 accu = (int64_t)w_re*tmpz[n34+i].im;
445 19082127 accu += (int64_t)w_im*tmpz[n34+i].re;
446 19082127 tmp4 = (int32_t)((accu + 0x40000000) >> 31);
447
448 19082127 tmp5 = tmp1 + tmp3;
449 19082127 tmp1 = tmp1 - tmp3;
450 19082127 tmp6 = tmp2 + tmp4;
451 19082127 tmp2 = tmp2 - tmp4;
452
453 19082127 tmpz[ n2+i].re = tmpz[ i].re - tmp5;
454 19082127 tmpz[ i].re = tmpz[ i].re + tmp5;
455 19082127 tmpz[ n2+i].im = tmpz[ i].im - tmp6;
456 19082127 tmpz[ i].im = tmpz[ i].im + tmp6;
457 19082127 tmpz[n34+i].re = tmpz[n4+i].re - tmp2;
458 19082127 tmpz[ n4+i].re = tmpz[n4+i].re + tmp2;
459 19082127 tmpz[n34+i].im = tmpz[n4+i].im + tmp1;
460 19082127 tmpz[ n4+i].im = tmpz[n4+i].im - tmp1;
461
462 19082127 w_re_ptr += step;
463 19082127 w_im_ptr -= step;
464 }
465 }
466 1670377 step >>= 1;
467 1670377 n4 <<= 1;
468 }
469 }
470
471 #else /* !FFT_FLOAT */
472
473 #define BUTTERFLIES(a0,a1,a2,a3) {\
474 BF(t3, t5, t5, t1);\
475 BF(a2.re, a0.re, a0.re, t5);\
476 BF(a3.im, a1.im, a1.im, t3);\
477 BF(t4, t6, t2, t6);\
478 BF(a3.re, a1.re, a1.re, t4);\
479 BF(a2.im, a0.im, a0.im, t6);\
480 }
481
482 // force loading all the inputs before storing any.
483 // this is slightly slower for small data, but avoids store->load aliasing
484 // for addresses separated by large powers of 2.
485 #define BUTTERFLIES_BIG(a0,a1,a2,a3) {\
486 FFTSample r0=a0.re, i0=a0.im, r1=a1.re, i1=a1.im;\
487 BF(t3, t5, t5, t1);\
488 BF(a2.re, a0.re, r0, t5);\
489 BF(a3.im, a1.im, i1, t3);\
490 BF(t4, t6, t2, t6);\
491 BF(a3.re, a1.re, r1, t4);\
492 BF(a2.im, a0.im, i0, t6);\
493 }
494
495 #define TRANSFORM(a0,a1,a2,a3,wre,wim) {\
496 CMUL(t1, t2, a2.re, a2.im, wre, -wim);\
497 CMUL(t5, t6, a3.re, a3.im, wre, wim);\
498 BUTTERFLIES(a0,a1,a2,a3)\
499 }
500
501 #define TRANSFORM_ZERO(a0,a1,a2,a3) {\
502 t1 = a2.re;\
503 t2 = a2.im;\
504 t5 = a3.re;\
505 t6 = a3.im;\
506 BUTTERFLIES(a0,a1,a2,a3)\
507 }
508
509 /* z[0...8n-1], w[1...2n-1] */
510 #define PASS(name)\
511 static void name(FFTComplex *z, const FFTSample *wre, unsigned int n)\
512 {\
513 FFTDouble t1, t2, t3, t4, t5, t6;\
514 int o1 = 2*n;\
515 int o2 = 4*n;\
516 int o3 = 6*n;\
517 const FFTSample *wim = wre+o1;\
518 n--;\
519 \
520 TRANSFORM_ZERO(z[0],z[o1],z[o2],z[o3]);\
521 TRANSFORM(z[1],z[o1+1],z[o2+1],z[o3+1],wre[1],wim[-1]);\
522 do {\
523 z += 2;\
524 wre += 2;\
525 wim -= 2;\
526 TRANSFORM(z[0],z[o1],z[o2],z[o3],wre[0],wim[0]);\
527 TRANSFORM(z[1],z[o1+1],z[o2+1],z[o3+1],wre[1],wim[-1]);\
528 } while(--n);\
529 }
530
531
2/2
✓ Branch 0 taken 22005242 times.
✓ Branch 1 taken 3744877 times.
25750119 PASS(pass)
532 #if !CONFIG_SMALL
533 #undef BUTTERFLIES
534 #define BUTTERFLIES BUTTERFLIES_BIG
535
2/2
✓ Branch 0 taken 3050042 times.
✓ Branch 1 taken 13219 times.
3063261 PASS(pass_big)
536 #endif
537
538 #define DECL_FFT(n,n2,n4)\
539 static void fft##n(FFTComplex *z)\
540 {\
541 fft##n2(z);\
542 fft##n4(z+n4*2);\
543 fft##n4(z+n4*3);\
544 pass(z,FFT_NAME(ff_cos_##n),n4/2);\
545 }
546
547 18921989 static void fft4(FFTComplex *z)
548 {
549 FFTDouble t1, t2, t3, t4, t5, t6, t7, t8;
550
551 18921989 BF(t3, t1, z[0].re, z[1].re);
552 18921989 BF(t8, t6, z[3].re, z[2].re);
553 18921989 BF(z[2].re, z[0].re, t1, t6);
554 18921989 BF(t4, t2, z[0].im, z[1].im);
555 18921989 BF(t7, t5, z[2].im, z[3].im);
556 18921989 BF(z[3].im, z[1].im, t4, t8);
557 18921989 BF(z[3].re, z[1].re, t3, t7);
558 18921989 BF(z[2].im, z[0].im, t2, t5);
559 18921989 }
560
561 9628585 static void fft8(FFTComplex *z)
562 {
563 FFTDouble t1, t2, t3, t4, t5, t6;
564
565 9628585 fft4(z);
566
567 9628585 BF(t1, z[5].re, z[4].re, -z[5].re);
568 9628585 BF(t2, z[5].im, z[4].im, -z[5].im);
569 9628585 BF(t5, z[7].re, z[6].re, -z[7].re);
570 9628585 BF(t6, z[7].im, z[6].im, -z[7].im);
571
572 9628585 BUTTERFLIES(z[0],z[2],z[4],z[6]);
573 9628585 TRANSFORM(z[1],z[3],z[5],z[7],sqrthalf,sqrthalf);
574 9628585 }
575
576 #if !CONFIG_SMALL
577 4134945 static void fft16(FFTComplex *z)
578 {
579 FFTDouble t1, t2, t3, t4, t5, t6;
580 4134945 FFTSample cos_16_1 = FFT_NAME(ff_cos_16)[1];
581 4134945 FFTSample cos_16_3 = FFT_NAME(ff_cos_16)[3];
582
583 4134945 fft8(z);
584 4134945 fft4(z+8);
585 4134945 fft4(z+12);
586
587 4134945 TRANSFORM_ZERO(z[0],z[4],z[8],z[12]);
588 4134945 TRANSFORM(z[2],z[6],z[10],z[14],sqrthalf,sqrthalf);
589 4134945 TRANSFORM(z[1],z[5],z[9],z[13],cos_16_1,cos_16_3);
590 4134945 TRANSFORM(z[3],z[7],z[11],z[15],cos_16_3,cos_16_1);
591 4134945 }
592 #else
593 DECL_FFT(16,8,4)
594 #endif
595 2592995 DECL_FFT(32,16,8)
596 631397 DECL_FFT(64,32,16)
597 324274 DECL_FFT(128,64,32)
598 114814 DECL_FFT(256,128,64)
599 81397 DECL_FFT(512,256,128)
600 #if !CONFIG_SMALL
601 #define pass pass_big
602 #endif
603 8553 DECL_FFT(1024,512,256)
604 2969 DECL_FFT(2048,1024,512)
605 1008 DECL_FFT(4096,2048,1024)
606 689 DECL_FFT(8192,4096,2048)
607 DECL_FFT(16384,8192,4096)
608 DECL_FFT(32768,16384,8192)
609 DECL_FFT(65536,32768,16384)
610 DECL_FFT(131072,65536,32768)
611
612 static void (* const fft_dispatch[])(FFTComplex*) = {
613 fft4, fft8, fft16, fft32, fft64, fft128, fft256, fft512, fft1024,
614 fft2048, fft4096, fft8192, fft16384, fft32768, fft65536, fft131072
615 };
616
617 3135907 static void fft_calc_c(FFTContext *s, FFTComplex *z)
618 {
619 3135907 fft_dispatch[s->nbits-2](z);
620 3135907 }
621 #endif /* !FFT_FLOAT */
622