Directory: | ../../../ffmpeg/ |
---|---|
File: | src/libavcodec/fft_template.c |
Date: | 2022-07-04 00:18:54 |
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 | 2558 | static av_cold void init_ff_cos_tabs(int index) | |
58 | { | ||
59 | int i; | ||
60 | 2558 | int m = 1<<index; | |
61 | 2558 | double freq = 2*M_PI/m; | |
62 | 2558 | FFTSample *tab = FFT_NAME(ff_cos_tabs)[index]; | |
63 |
2/2✓ Branch 0 taken 182898 times.
✓ Branch 1 taken 2558 times.
|
185456 | for(i=0; i<=m/4; i++) |
64 | 182898 | tab[i] = FIX15(cos(i*freq)); | |
65 |
2/2✓ Branch 0 taken 177782 times.
✓ Branch 1 taken 2558 times.
|
180340 | for(i=1; i<m/4; i++) |
66 | 177782 | tab[m/2-i] = tab[i]; | |
67 | 2558 | } | |
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 | 505 | INIT_FF_COS_TABS_FUNC(4, 16) | |
81 | 488 | INIT_FF_COS_TABS_FUNC(5, 32) | |
82 | 413 | INIT_FF_COS_TABS_FUNC(6, 64) | |
83 | 386 | INIT_FF_COS_TABS_FUNC(7, 128) | |
84 | 323 | INIT_FF_COS_TABS_FUNC(8, 256) | |
85 | 299 | INIT_FF_COS_TABS_FUNC(9, 512) | |
86 | 71 | 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 | 10999 | av_cold void ff_init_ff_cos_tabs(int index) | |
117 | { | ||
118 | 10999 | ff_thread_once(&cos_tabs_init_once[index].control, cos_tabs_init_once[index].func); | |
119 | 10999 | } | |
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 | 5316040 | static int split_radix_permutation(int i, int n, int inverse) | |
145 | { | ||
146 | int m; | ||
147 |
2/2✓ Branch 0 taken 788304 times.
✓ Branch 1 taken 4527736 times.
|
5316040 | if(n <= 2) return i&1; |
148 | 4527736 | m = n >> 1; | |
149 |
2/2✓ Branch 0 taken 2263868 times.
✓ Branch 1 taken 2263868 times.
|
4527736 | if(!(i&m)) return split_radix_permutation(i, m, inverse)*2; |
150 | 2263868 | m >>= 1; | |
151 |
2/2✓ Branch 0 taken 1131934 times.
✓ Branch 1 taken 1131934 times.
|
2263868 | if(inverse == !(i&m)) return split_radix_permutation(i, m, inverse)*4 + 1; |
152 | 1131934 | 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 | 17786 | static int is_second_half_of_fft32(int i, int n) | |
161 | { | ||
162 |
2/2✓ Branch 0 taken 5066 times.
✓ Branch 1 taken 12720 times.
|
17786 | if (n <= 32) |
163 | 5066 | return i >= 16; | |
164 |
2/2✓ Branch 0 taken 6360 times.
✓ Branch 1 taken 6360 times.
|
12720 | else if (i < n/2) |
165 | 6360 | return is_second_half_of_fft32(i, n/2); | |
166 |
2/2✓ Branch 0 taken 3180 times.
✓ Branch 1 taken 3180 times.
|
6360 | else if (i < 3*n/4) |
167 | 3180 | return is_second_half_of_fft32(i - n/2, n/4); | |
168 | else | ||
169 | 3180 | return is_second_half_of_fft32(i - 3*n/4, n/4); | |
170 | } | ||
171 | |||
172 | 399 | static av_cold void fft_perm_avx(FFTContext *s) | |
173 | { | ||
174 | int i; | ||
175 | 399 | int n = 1 << s->nbits; | |
176 | |||
177 |
2/2✓ Branch 0 taken 5066 times.
✓ Branch 1 taken 399 times.
|
5465 | for (i = 0; i < n; i += 16) { |
178 | int k; | ||
179 |
2/2✓ Branch 1 taken 1735 times.
✓ Branch 2 taken 3331 times.
|
5066 | if (is_second_half_of_fft32(i, n)) { |
180 |
2/2✓ Branch 0 taken 27760 times.
✓ Branch 1 taken 1735 times.
|
29495 | for (k = 0; k < 16; k++) |
181 | 55520 | s->revtab[-split_radix_permutation(i + k, n, s->inverse) & (n - 1)] = | |
182 | 27760 | i + avx_tab[k]; | |
183 | |||
184 | } else { | ||
185 |
2/2✓ Branch 0 taken 53296 times.
✓ Branch 1 taken 3331 times.
|
56627 | for (k = 0; k < 16; k++) { |
186 | 53296 | int j = i + k; | |
187 | 53296 | j = (j & ~7) | ((j >> 1) & 3) | ((j << 2) & 4); | |
188 | 53296 | s->revtab[-split_radix_permutation(i + k, n, s->inverse) & (n - 1)] = j; | |
189 | } | ||
190 | } | ||
191 | } | ||
192 | 399 | } | |
193 | |||
194 | 3790 | av_cold int ff_fft_init(FFTContext *s, int nbits, int inverse) | |
195 | { | ||
196 | int i, j, n; | ||
197 | |||
198 | 3790 | s->revtab = NULL; | |
199 | 3790 | s->revtab32 = NULL; | |
200 | |||
201 |
2/4✓ Branch 0 taken 3790 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 3790 times.
|
3790 | if (nbits < 2 || nbits > 17) |
202 | ✗ | goto fail; | |
203 | 3790 | s->nbits = nbits; | |
204 | 3790 | n = 1 << nbits; | |
205 | |||
206 |
1/2✓ Branch 0 taken 3790 times.
✗ Branch 1 not taken.
|
3790 | if (nbits <= 16) { |
207 | 3790 | s->revtab = av_malloc(n * sizeof(uint16_t)); | |
208 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 3790 times.
|
3790 | 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 | 3790 | s->tmp_buf = av_malloc(n * sizeof(FFTComplex)); | |
216 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 3790 times.
|
3790 | if (!s->tmp_buf) |
217 | ✗ | goto fail; | |
218 | 3790 | s->inverse = inverse; | |
219 | 3790 | s->fft_permutation = FF_FFT_PERM_DEFAULT; | |
220 | |||
221 | 3790 | s->fft_permute = fft_permute_c; | |
222 | 3790 | s->fft_calc = fft_calc_c; | |
223 | #if CONFIG_MDCT | ||
224 | 3790 | s->imdct_calc = ff_imdct_calc_c; | |
225 | 3790 | s->imdct_half = ff_imdct_half_c; | |
226 | 3790 | s->mdct_calc = ff_mdct_calc_c; | |
227 | #endif | ||
228 | |||
229 | #if FFT_FLOAT | ||
230 | #if ARCH_AARCH64 | ||
231 | ff_fft_init_aarch64(s); | ||
232 | #elif ARCH_ARM | ||
233 | ff_fft_init_arm(s); | ||
234 | #elif ARCH_PPC | ||
235 | ff_fft_init_ppc(s); | ||
236 | #elif ARCH_X86 | ||
237 | 3586 | ff_fft_init_x86(s); | |
238 | #endif | ||
239 | #if HAVE_MIPSFPU | ||
240 | ff_fft_init_mips(s); | ||
241 | #endif | ||
242 |
2/2✓ Branch 0 taken 10818 times.
✓ Branch 1 taken 3586 times.
|
14404 | for(j=4; j<=nbits; j++) { |
243 | 10818 | ff_init_ff_cos_tabs(j); | |
244 | } | ||
245 | #else /* FFT_FLOAT */ | ||
246 | 204 | ff_fft_lut_init(); | |
247 | #endif | ||
248 | |||
249 | |||
250 |
2/2✓ Branch 0 taken 399 times.
✓ Branch 1 taken 3187 times.
|
3586 | if (ARCH_X86 && FFT_FLOAT && s->fft_permutation == FF_FFT_PERM_AVX) { |
251 | 399 | fft_perm_avx(s); | |
252 | } else { | ||
253 | #define PROCESS_FFT_PERM_SWAP_LSBS(num) do {\ | ||
254 | for(i = 0; i < n; i++) {\ | ||
255 | int k;\ | ||
256 | j = i;\ | ||
257 | j = (j & ~3) | ((j >> 1) & 1) | ((j << 1) & 2);\ | ||
258 | k = -split_radix_permutation(i, n, s->inverse) & (n - 1);\ | ||
259 | s->revtab##num[k] = j;\ | ||
260 | } \ | ||
261 | } while(0); | ||
262 | |||
263 | #define PROCESS_FFT_PERM_DEFAULT(num) do {\ | ||
264 | for(i = 0; i < n; i++) {\ | ||
265 | int k;\ | ||
266 | j = i;\ | ||
267 | k = -split_radix_permutation(i, n, s->inverse) & (n - 1);\ | ||
268 | s->revtab##num[k] = j;\ | ||
269 | } \ | ||
270 | } while(0); | ||
271 | |||
272 | #define SPLIT_RADIX_PERMUTATION(num) do { \ | ||
273 | if (s->fft_permutation == FF_FFT_PERM_SWAP_LSBS) {\ | ||
274 | PROCESS_FFT_PERM_SWAP_LSBS(num) \ | ||
275 | } else {\ | ||
276 | PROCESS_FFT_PERM_DEFAULT(num) \ | ||
277 | }\ | ||
278 | } while(0); | ||
279 | |||
280 |
1/2✓ Branch 0 taken 3391 times.
✗ Branch 1 not taken.
|
3391 | if (s->revtab) |
281 |
6/6✓ Branch 0 taken 140 times.
✓ Branch 1 taken 3251 times.
✓ Branch 3 taken 1428 times.
✓ Branch 4 taken 140 times.
✓ Branch 6 taken 705820 times.
✓ Branch 7 taken 3251 times.
|
710639 | SPLIT_RADIX_PERMUTATION() |
282 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 3391 times.
|
3391 | if (s->revtab32) |
283 | ✗ | SPLIT_RADIX_PERMUTATION(32) | |
284 | |||
285 | #undef PROCESS_FFT_PERM_DEFAULT | ||
286 | #undef PROCESS_FFT_PERM_SWAP_LSBS | ||
287 | #undef SPLIT_RADIX_PERMUTATION | ||
288 | } | ||
289 | |||
290 | 3790 | return 0; | |
291 | ✗ | fail: | |
292 | ✗ | av_freep(&s->revtab); | |
293 | ✗ | av_freep(&s->revtab32); | |
294 | ✗ | av_freep(&s->tmp_buf); | |
295 | ✗ | return -1; | |
296 | } | ||
297 | |||
298 | 48126 | static void fft_permute_c(FFTContext *s, FFTComplex *z) | |
299 | { | ||
300 | int j, np; | ||
301 | 48126 | const uint16_t *revtab = s->revtab; | |
302 | 48126 | const uint32_t *revtab32 = s->revtab32; | |
303 | 48126 | np = 1 << s->nbits; | |
304 | /* TODO: handle split-radix permute in a more optimal way, probably in-place */ | ||
305 |
1/2✓ Branch 0 taken 48126 times.
✗ Branch 1 not taken.
|
48126 | if (revtab) { |
306 |
2/2✓ Branch 0 taken 12874304 times.
✓ Branch 1 taken 48126 times.
|
12922430 | for(j=0;j<np;j++) s->tmp_buf[revtab[j]] = z[j]; |
307 | } else | ||
308 | ✗ | for(j=0;j<np;j++) s->tmp_buf[revtab32[j]] = z[j]; | |
309 | |||
310 | 48126 | memcpy(z, s->tmp_buf, np * sizeof(FFTComplex)); | |
311 | 48126 | } | |
312 | |||
313 | 3884 | av_cold void ff_fft_end(FFTContext *s) | |
314 | { | ||
315 | 3884 | av_freep(&s->revtab); | |
316 | 3884 | av_freep(&s->revtab32); | |
317 | 3884 | av_freep(&s->tmp_buf); | |
318 | 3884 | } | |
319 | |||
320 | #if !FFT_FLOAT | ||
321 | |||
322 | 772645 | static void fft_calc_c(FFTContext *s, FFTComplex *z) { | |
323 | |||
324 | int nbits, i, n, num_transforms, offset, step; | ||
325 | int n4, n2, n34; | ||
326 | unsigned tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8; | ||
327 | FFTComplex *tmpz; | ||
328 | 772645 | const int fft_size = (1 << s->nbits); | |
329 | int64_t accu; | ||
330 | |||
331 | 772645 | num_transforms = (0x2aab >> (16 - s->nbits)) | 1; | |
332 | |||
333 |
2/2✓ Branch 0 taken 5890291 times.
✓ Branch 1 taken 772645 times.
|
6662936 | for (n=0; n<num_transforms; n++){ |
334 | 5890291 | offset = ff_fft_offsets_lut[n] << 2; | |
335 | 5890291 | tmpz = z + offset; | |
336 | |||
337 | 5890291 | tmp1 = tmpz[0].re + (unsigned)tmpz[1].re; | |
338 | 5890291 | tmp5 = tmpz[2].re + (unsigned)tmpz[3].re; | |
339 | 5890291 | tmp2 = tmpz[0].im + (unsigned)tmpz[1].im; | |
340 | 5890291 | tmp6 = tmpz[2].im + (unsigned)tmpz[3].im; | |
341 | 5890291 | tmp3 = tmpz[0].re - (unsigned)tmpz[1].re; | |
342 | 5890291 | tmp8 = tmpz[2].im - (unsigned)tmpz[3].im; | |
343 | 5890291 | tmp4 = tmpz[0].im - (unsigned)tmpz[1].im; | |
344 | 5890291 | tmp7 = tmpz[2].re - (unsigned)tmpz[3].re; | |
345 | |||
346 | 5890291 | tmpz[0].re = tmp1 + tmp5; | |
347 | 5890291 | tmpz[2].re = tmp1 - tmp5; | |
348 | 5890291 | tmpz[0].im = tmp2 + tmp6; | |
349 | 5890291 | tmpz[2].im = tmp2 - tmp6; | |
350 | 5890291 | tmpz[1].re = tmp3 + tmp8; | |
351 | 5890291 | tmpz[3].re = tmp3 - tmp8; | |
352 | 5890291 | tmpz[1].im = tmp4 - tmp7; | |
353 | 5890291 | tmpz[3].im = tmp4 + tmp7; | |
354 | } | ||
355 | |||
356 |
2/2✓ Branch 0 taken 1 times.
✓ Branch 1 taken 772644 times.
|
772645 | if (fft_size < 8) |
357 | 1 | return; | |
358 | |||
359 | 772644 | num_transforms = (num_transforms >> 1) | 1; | |
360 | |||
361 |
2/2✓ Branch 0 taken 3313828 times.
✓ Branch 1 taken 772644 times.
|
4086472 | for (n=0; n<num_transforms; n++){ |
362 | 3313828 | offset = ff_fft_offsets_lut[n] << 3; | |
363 | 3313828 | tmpz = z + offset; | |
364 | |||
365 | 3313828 | tmp1 = tmpz[4].re + (unsigned)tmpz[5].re; | |
366 | 3313828 | tmp3 = tmpz[6].re + (unsigned)tmpz[7].re; | |
367 | 3313828 | tmp2 = tmpz[4].im + (unsigned)tmpz[5].im; | |
368 | 3313828 | tmp4 = tmpz[6].im + (unsigned)tmpz[7].im; | |
369 | 3313828 | tmp5 = tmp1 + tmp3; | |
370 | 3313828 | tmp7 = tmp1 - tmp3; | |
371 | 3313828 | tmp6 = tmp2 + tmp4; | |
372 | 3313828 | tmp8 = tmp2 - tmp4; | |
373 | |||
374 | 3313828 | tmp1 = tmpz[4].re - (unsigned)tmpz[5].re; | |
375 | 3313828 | tmp2 = tmpz[4].im - (unsigned)tmpz[5].im; | |
376 | 3313828 | tmp3 = tmpz[6].re - (unsigned)tmpz[7].re; | |
377 | 3313828 | tmp4 = tmpz[6].im - (unsigned)tmpz[7].im; | |
378 | |||
379 | 3313828 | tmpz[4].re = tmpz[0].re - tmp5; | |
380 | 3313828 | tmpz[0].re = tmpz[0].re + tmp5; | |
381 | 3313828 | tmpz[4].im = tmpz[0].im - tmp6; | |
382 | 3313828 | tmpz[0].im = tmpz[0].im + tmp6; | |
383 | 3313828 | tmpz[6].re = tmpz[2].re - tmp8; | |
384 | 3313828 | tmpz[2].re = tmpz[2].re + tmp8; | |
385 | 3313828 | tmpz[6].im = tmpz[2].im + tmp7; | |
386 | 3313828 | tmpz[2].im = tmpz[2].im - tmp7; | |
387 | |||
388 | 3313828 | accu = (int64_t)Q31(M_SQRT1_2)*(int)(tmp1 + tmp2); | |
389 | 3313828 | tmp5 = (int32_t)((accu + 0x40000000) >> 31); | |
390 | 3313828 | accu = (int64_t)Q31(M_SQRT1_2)*(int)(tmp3 - tmp4); | |
391 | 3313828 | tmp7 = (int32_t)((accu + 0x40000000) >> 31); | |
392 | 3313828 | accu = (int64_t)Q31(M_SQRT1_2)*(int)(tmp2 - tmp1); | |
393 | 3313828 | tmp6 = (int32_t)((accu + 0x40000000) >> 31); | |
394 | 3313828 | accu = (int64_t)Q31(M_SQRT1_2)*(int)(tmp3 + tmp4); | |
395 | 3313828 | tmp8 = (int32_t)((accu + 0x40000000) >> 31); | |
396 | 3313828 | tmp1 = tmp5 + tmp7; | |
397 | 3313828 | tmp3 = tmp5 - tmp7; | |
398 | 3313828 | tmp2 = tmp6 + tmp8; | |
399 | 3313828 | tmp4 = tmp6 - tmp8; | |
400 | |||
401 | 3313828 | tmpz[5].re = tmpz[1].re - tmp1; | |
402 | 3313828 | tmpz[1].re = tmpz[1].re + tmp1; | |
403 | 3313828 | tmpz[5].im = tmpz[1].im - tmp2; | |
404 | 3313828 | tmpz[1].im = tmpz[1].im + tmp2; | |
405 | 3313828 | tmpz[7].re = tmpz[3].re - tmp4; | |
406 | 3313828 | tmpz[3].re = tmpz[3].re + tmp4; | |
407 | 3313828 | tmpz[7].im = tmpz[3].im + tmp3; | |
408 | 3313828 | tmpz[3].im = tmpz[3].im - tmp3; | |
409 | } | ||
410 | |||
411 | 772644 | step = 1 << ((MAX_LOG2_NFFT-4) - 4); | |
412 | 772644 | n4 = 4; | |
413 | |||
414 |
2/2✓ Branch 0 taken 1677943 times.
✓ Branch 1 taken 772644 times.
|
2450587 | for (nbits=4; nbits<=s->nbits; nbits++){ |
415 | 1677943 | n2 = 2*n4; | |
416 | 1677943 | n34 = 3*n4; | |
417 | 1677943 | num_transforms = (num_transforms >> 1) | 1; | |
418 | |||
419 |
2/2✓ Branch 0 taken 2558823 times.
✓ Branch 1 taken 1677943 times.
|
4236766 | for (n=0; n<num_transforms; n++){ |
420 | 2558823 | const FFTSample *w_re_ptr = ff_w_tab_sr + step; | |
421 | 2558823 | const FFTSample *w_im_ptr = ff_w_tab_sr + MAX_FFT_SIZE/(4*16) - step; | |
422 | 2558823 | offset = ff_fft_offsets_lut[n] << nbits; | |
423 | 2558823 | tmpz = z + offset; | |
424 | |||
425 | 2558823 | tmp5 = tmpz[ n2].re + (unsigned)tmpz[n34].re; | |
426 | 2558823 | tmp1 = tmpz[ n2].re - (unsigned)tmpz[n34].re; | |
427 | 2558823 | tmp6 = tmpz[ n2].im + (unsigned)tmpz[n34].im; | |
428 | 2558823 | tmp2 = tmpz[ n2].im - (unsigned)tmpz[n34].im; | |
429 | |||
430 | 2558823 | tmpz[ n2].re = tmpz[ 0].re - tmp5; | |
431 | 2558823 | tmpz[ 0].re = tmpz[ 0].re + tmp5; | |
432 | 2558823 | tmpz[ n2].im = tmpz[ 0].im - tmp6; | |
433 | 2558823 | tmpz[ 0].im = tmpz[ 0].im + tmp6; | |
434 | 2558823 | tmpz[n34].re = tmpz[n4].re - tmp2; | |
435 | 2558823 | tmpz[ n4].re = tmpz[n4].re + tmp2; | |
436 | 2558823 | tmpz[n34].im = tmpz[n4].im + tmp1; | |
437 | 2558823 | tmpz[ n4].im = tmpz[n4].im - tmp1; | |
438 | |||
439 |
2/2✓ Branch 0 taken 19292373 times.
✓ Branch 1 taken 2558823 times.
|
21851196 | for (i=1; i<n4; i++){ |
440 | 19292373 | FFTSample w_re = w_re_ptr[0]; | |
441 | 19292373 | FFTSample w_im = w_im_ptr[0]; | |
442 | 19292373 | accu = (int64_t)w_re*tmpz[ n2+i].re; | |
443 | 19292373 | accu += (int64_t)w_im*tmpz[ n2+i].im; | |
444 | 19292373 | tmp1 = (int32_t)((accu + 0x40000000) >> 31); | |
445 | 19292373 | accu = (int64_t)w_re*tmpz[ n2+i].im; | |
446 | 19292373 | accu -= (int64_t)w_im*tmpz[ n2+i].re; | |
447 | 19292373 | tmp2 = (int32_t)((accu + 0x40000000) >> 31); | |
448 | 19292373 | accu = (int64_t)w_re*tmpz[n34+i].re; | |
449 | 19292373 | accu -= (int64_t)w_im*tmpz[n34+i].im; | |
450 | 19292373 | tmp3 = (int32_t)((accu + 0x40000000) >> 31); | |
451 | 19292373 | accu = (int64_t)w_re*tmpz[n34+i].im; | |
452 | 19292373 | accu += (int64_t)w_im*tmpz[n34+i].re; | |
453 | 19292373 | tmp4 = (int32_t)((accu + 0x40000000) >> 31); | |
454 | |||
455 | 19292373 | tmp5 = tmp1 + tmp3; | |
456 | 19292373 | tmp1 = tmp1 - tmp3; | |
457 | 19292373 | tmp6 = tmp2 + tmp4; | |
458 | 19292373 | tmp2 = tmp2 - tmp4; | |
459 | |||
460 | 19292373 | tmpz[ n2+i].re = tmpz[ i].re - tmp5; | |
461 | 19292373 | tmpz[ i].re = tmpz[ i].re + tmp5; | |
462 | 19292373 | tmpz[ n2+i].im = tmpz[ i].im - tmp6; | |
463 | 19292373 | tmpz[ i].im = tmpz[ i].im + tmp6; | |
464 | 19292373 | tmpz[n34+i].re = tmpz[n4+i].re - tmp2; | |
465 | 19292373 | tmpz[ n4+i].re = tmpz[n4+i].re + tmp2; | |
466 | 19292373 | tmpz[n34+i].im = tmpz[n4+i].im + tmp1; | |
467 | 19292373 | tmpz[ n4+i].im = tmpz[n4+i].im - tmp1; | |
468 | |||
469 | 19292373 | w_re_ptr += step; | |
470 | 19292373 | w_im_ptr -= step; | |
471 | } | ||
472 | } | ||
473 | 1677943 | step >>= 1; | |
474 | 1677943 | n4 <<= 1; | |
475 | } | ||
476 | } | ||
477 | |||
478 | #else /* !FFT_FLOAT */ | ||
479 | |||
480 | #define BUTTERFLIES(a0,a1,a2,a3) {\ | ||
481 | BF(t3, t5, t5, t1);\ | ||
482 | BF(a2.re, a0.re, a0.re, t5);\ | ||
483 | BF(a3.im, a1.im, a1.im, t3);\ | ||
484 | BF(t4, t6, t2, t6);\ | ||
485 | BF(a3.re, a1.re, a1.re, t4);\ | ||
486 | BF(a2.im, a0.im, a0.im, t6);\ | ||
487 | } | ||
488 | |||
489 | // force loading all the inputs before storing any. | ||
490 | // this is slightly slower for small data, but avoids store->load aliasing | ||
491 | // for addresses separated by large powers of 2. | ||
492 | #define BUTTERFLIES_BIG(a0,a1,a2,a3) {\ | ||
493 | FFTSample r0=a0.re, i0=a0.im, r1=a1.re, i1=a1.im;\ | ||
494 | BF(t3, t5, t5, t1);\ | ||
495 | BF(a2.re, a0.re, r0, t5);\ | ||
496 | BF(a3.im, a1.im, i1, t3);\ | ||
497 | BF(t4, t6, t2, t6);\ | ||
498 | BF(a3.re, a1.re, r1, t4);\ | ||
499 | BF(a2.im, a0.im, i0, t6);\ | ||
500 | } | ||
501 | |||
502 | #define TRANSFORM(a0,a1,a2,a3,wre,wim) {\ | ||
503 | CMUL(t1, t2, a2.re, a2.im, wre, -wim);\ | ||
504 | CMUL(t5, t6, a3.re, a3.im, wre, wim);\ | ||
505 | BUTTERFLIES(a0,a1,a2,a3)\ | ||
506 | } | ||
507 | |||
508 | #define TRANSFORM_ZERO(a0,a1,a2,a3) {\ | ||
509 | t1 = a2.re;\ | ||
510 | t2 = a2.im;\ | ||
511 | t5 = a3.re;\ | ||
512 | t6 = a3.im;\ | ||
513 | BUTTERFLIES(a0,a1,a2,a3)\ | ||
514 | } | ||
515 | |||
516 | /* z[0...8n-1], w[1...2n-1] */ | ||
517 | #define PASS(name)\ | ||
518 | static void name(FFTComplex *z, const FFTSample *wre, unsigned int n)\ | ||
519 | {\ | ||
520 | FFTDouble t1, t2, t3, t4, t5, t6;\ | ||
521 | int o1 = 2*n;\ | ||
522 | int o2 = 4*n;\ | ||
523 | int o3 = 6*n;\ | ||
524 | const FFTSample *wim = wre+o1;\ | ||
525 | n--;\ | ||
526 | \ | ||
527 | TRANSFORM_ZERO(z[0],z[o1],z[o2],z[o3]);\ | ||
528 | TRANSFORM(z[1],z[o1+1],z[o2+1],z[o3+1],wre[1],wim[-1]);\ | ||
529 | do {\ | ||
530 | z += 2;\ | ||
531 | wre += 2;\ | ||
532 | wim -= 2;\ | ||
533 | TRANSFORM(z[0],z[o1],z[o2],z[o3],wre[0],wim[0]);\ | ||
534 | TRANSFORM(z[1],z[o1+1],z[o2+1],z[o3+1],wre[1],wim[-1]);\ | ||
535 | } while(--n);\ | ||
536 | } | ||
537 | |||
538 |
2/2✓ Branch 0 taken 22013484 times.
✓ Branch 1 taken 3746142 times.
|
25759626 | PASS(pass) |
539 | #if !CONFIG_SMALL | ||
540 | #undef BUTTERFLIES | ||
541 | #define BUTTERFLIES BUTTERFLIES_BIG | ||
542 |
2/2✓ Branch 0 taken 3050042 times.
✓ Branch 1 taken 13219 times.
|
3063261 | PASS(pass_big) |
543 | #endif | ||
544 | |||
545 | #define DECL_FFT(n,n2,n4)\ | ||
546 | static void fft##n(FFTComplex *z)\ | ||
547 | {\ | ||
548 | fft##n2(z);\ | ||
549 | fft##n4(z+n4*2);\ | ||
550 | fft##n4(z+n4*3);\ | ||
551 | pass(z,FFT_NAME(ff_cos_##n),n4/2);\ | ||
552 | } | ||
553 | |||
554 | 18927350 | static void fft4(FFTComplex *z) | |
555 | { | ||
556 | FFTDouble t1, t2, t3, t4, t5, t6, t7, t8; | ||
557 | |||
558 | 18927350 | BF(t3, t1, z[0].re, z[1].re); | |
559 | 18927350 | BF(t8, t6, z[3].re, z[2].re); | |
560 | 18927350 | BF(z[2].re, z[0].re, t1, t6); | |
561 | 18927350 | BF(t4, t2, z[0].im, z[1].im); | |
562 | 18927350 | BF(t7, t5, z[2].im, z[3].im); | |
563 | 18927350 | BF(z[3].im, z[1].im, t4, t8); | |
564 | 18927350 | BF(z[3].re, z[1].re, t3, t7); | |
565 | 18927350 | BF(z[2].im, z[0].im, t2, t5); | |
566 | 18927350 | } | |
567 | |||
568 | 9631288 | static void fft8(FFTComplex *z) | |
569 | { | ||
570 | FFTDouble t1, t2, t3, t4, t5, t6; | ||
571 | |||
572 | 9631288 | fft4(z); | |
573 | |||
574 | 9631288 | BF(t1, z[5].re, z[4].re, -z[5].re); | |
575 | 9631288 | BF(t2, z[5].im, z[4].im, -z[5].im); | |
576 | 9631288 | BF(t5, z[7].re, z[6].re, -z[7].re); | |
577 | 9631288 | BF(t6, z[7].im, z[6].im, -z[7].im); | |
578 | |||
579 | 9631288 | BUTTERFLIES(z[0],z[2],z[4],z[6]); | |
580 | 9631288 | TRANSFORM(z[1],z[3],z[5],z[7],sqrthalf,sqrthalf); | |
581 | 9631288 | } | |
582 | |||
583 | #if !CONFIG_SMALL | ||
584 | 4136274 | static void fft16(FFTComplex *z) | |
585 | { | ||
586 | FFTDouble t1, t2, t3, t4, t5, t6; | ||
587 | 4136274 | FFTSample cos_16_1 = FFT_NAME(ff_cos_16)[1]; | |
588 | 4136274 | FFTSample cos_16_3 = FFT_NAME(ff_cos_16)[3]; | |
589 | |||
590 | 4136274 | fft8(z); | |
591 | 4136274 | fft4(z+8); | |
592 | 4136274 | fft4(z+12); | |
593 | |||
594 | 4136274 | TRANSFORM_ZERO(z[0],z[4],z[8],z[12]); | |
595 | 4136274 | TRANSFORM(z[2],z[6],z[10],z[14],sqrthalf,sqrthalf); | |
596 | 4136274 | TRANSFORM(z[1],z[5],z[9],z[13],cos_16_1,cos_16_3); | |
597 | 4136274 | TRANSFORM(z[3],z[7],z[11],z[15],cos_16_3,cos_16_1); | |
598 | 4136274 | } | |
599 | #else | ||
600 | DECL_FFT(16,8,4) | ||
601 | #endif | ||
602 | 2593682 | DECL_FFT(32,16,8) | |
603 | 631718 | DECL_FFT(64,32,16) | |
604 | 324457 | DECL_FFT(128,64,32) | |
605 | 114883 | DECL_FFT(256,128,64) | |
606 | 81402 | DECL_FFT(512,256,128) | |
607 | #if !CONFIG_SMALL | ||
608 | #define pass pass_big | ||
609 | #endif | ||
610 | 8553 | DECL_FFT(1024,512,256) | |
611 | 2969 | DECL_FFT(2048,1024,512) | |
612 | 1008 | DECL_FFT(4096,2048,1024) | |
613 | 689 | DECL_FFT(8192,4096,2048) | |
614 | ✗ | DECL_FFT(16384,8192,4096) | |
615 | ✗ | DECL_FFT(32768,16384,8192) | |
616 | ✗ | DECL_FFT(65536,32768,16384) | |
617 | ✗ | DECL_FFT(131072,65536,32768) | |
618 | |||
619 | static void (* const fft_dispatch[])(FFTComplex*) = { | ||
620 | fft4, fft8, fft16, fft32, fft64, fft128, fft256, fft512, fft1024, | ||
621 | fft2048, fft4096, fft8192, fft16384, fft32768, fft65536, fft131072 | ||
622 | }; | ||
623 | |||
624 | 3136080 | static void fft_calc_c(FFTContext *s, FFTComplex *z) | |
625 | { | ||
626 | 3136080 | fft_dispatch[s->nbits-2](z); | |
627 | 3136080 | } | |
628 | #endif /* !FFT_FLOAT */ | ||
629 |