Line data Source code
1 : /*
2 : * http://www.kurims.kyoto-u.ac.jp/~ooura/fft.html
3 : * Copyright Takuya OOURA, 1996-2001
4 : *
5 : * You may use, copy, modify and distribute this code for any purpose (include
6 : * commercial use) and without fee. Please refer to this package when you modify
7 : * this code.
8 : *
9 : * Changes by the WebRTC authors:
10 : * - Trivial type modifications.
11 : * - Minimal code subset to do rdft of length 128.
12 : * - Optimizations because of known length.
13 : * - Removed the global variables by moving the code in to a class in order
14 : * to make it thread safe.
15 : *
16 : * All changes are covered by the WebRTC license and IP grant:
17 : * Use of this source code is governed by a BSD-style license
18 : * that can be found in the LICENSE file in the root of the source
19 : * tree. An additional intellectual property rights grant can be found
20 : * in the file PATENTS. All contributing project authors may
21 : * be found in the AUTHORS file in the root of the source tree.
22 : */
23 :
24 : #include "webrtc/modules/audio_processing/utility/ooura_fft.h"
25 :
26 : #include <math.h>
27 :
28 : #include "webrtc/modules/audio_processing/utility/ooura_fft_tables_common.h"
29 : #include "webrtc/system_wrappers/include/cpu_features_wrapper.h"
30 : #include "webrtc/typedefs.h"
31 :
32 : namespace webrtc {
33 :
34 : namespace {
35 :
36 : #if !(defined(MIPS_FPU_LE) || defined(WEBRTC_HAS_NEON))
37 0 : static void cft1st_128_C(float* a) {
38 0 : const int n = 128;
39 : int j, k1, k2;
40 : float wk1r, wk1i, wk2r, wk2i, wk3r, wk3i;
41 : float x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i;
42 :
43 : // The processing of the first set of elements was simplified in C to avoid
44 : // some operations (multiplication by zero or one, addition of two elements
45 : // multiplied by the same weight, ...).
46 0 : x0r = a[0] + a[2];
47 0 : x0i = a[1] + a[3];
48 0 : x1r = a[0] - a[2];
49 0 : x1i = a[1] - a[3];
50 0 : x2r = a[4] + a[6];
51 0 : x2i = a[5] + a[7];
52 0 : x3r = a[4] - a[6];
53 0 : x3i = a[5] - a[7];
54 0 : a[0] = x0r + x2r;
55 0 : a[1] = x0i + x2i;
56 0 : a[4] = x0r - x2r;
57 0 : a[5] = x0i - x2i;
58 0 : a[2] = x1r - x3i;
59 0 : a[3] = x1i + x3r;
60 0 : a[6] = x1r + x3i;
61 0 : a[7] = x1i - x3r;
62 0 : wk1r = rdft_w[2];
63 0 : x0r = a[8] + a[10];
64 0 : x0i = a[9] + a[11];
65 0 : x1r = a[8] - a[10];
66 0 : x1i = a[9] - a[11];
67 0 : x2r = a[12] + a[14];
68 0 : x2i = a[13] + a[15];
69 0 : x3r = a[12] - a[14];
70 0 : x3i = a[13] - a[15];
71 0 : a[8] = x0r + x2r;
72 0 : a[9] = x0i + x2i;
73 0 : a[12] = x2i - x0i;
74 0 : a[13] = x0r - x2r;
75 0 : x0r = x1r - x3i;
76 0 : x0i = x1i + x3r;
77 0 : a[10] = wk1r * (x0r - x0i);
78 0 : a[11] = wk1r * (x0r + x0i);
79 0 : x0r = x3i + x1r;
80 0 : x0i = x3r - x1i;
81 0 : a[14] = wk1r * (x0i - x0r);
82 0 : a[15] = wk1r * (x0i + x0r);
83 0 : k1 = 0;
84 0 : for (j = 16; j < n; j += 16) {
85 0 : k1 += 2;
86 0 : k2 = 2 * k1;
87 0 : wk2r = rdft_w[k1 + 0];
88 0 : wk2i = rdft_w[k1 + 1];
89 0 : wk1r = rdft_w[k2 + 0];
90 0 : wk1i = rdft_w[k2 + 1];
91 0 : wk3r = rdft_wk3ri_first[k1 + 0];
92 0 : wk3i = rdft_wk3ri_first[k1 + 1];
93 0 : x0r = a[j + 0] + a[j + 2];
94 0 : x0i = a[j + 1] + a[j + 3];
95 0 : x1r = a[j + 0] - a[j + 2];
96 0 : x1i = a[j + 1] - a[j + 3];
97 0 : x2r = a[j + 4] + a[j + 6];
98 0 : x2i = a[j + 5] + a[j + 7];
99 0 : x3r = a[j + 4] - a[j + 6];
100 0 : x3i = a[j + 5] - a[j + 7];
101 0 : a[j + 0] = x0r + x2r;
102 0 : a[j + 1] = x0i + x2i;
103 0 : x0r -= x2r;
104 0 : x0i -= x2i;
105 0 : a[j + 4] = wk2r * x0r - wk2i * x0i;
106 0 : a[j + 5] = wk2r * x0i + wk2i * x0r;
107 0 : x0r = x1r - x3i;
108 0 : x0i = x1i + x3r;
109 0 : a[j + 2] = wk1r * x0r - wk1i * x0i;
110 0 : a[j + 3] = wk1r * x0i + wk1i * x0r;
111 0 : x0r = x1r + x3i;
112 0 : x0i = x1i - x3r;
113 0 : a[j + 6] = wk3r * x0r - wk3i * x0i;
114 0 : a[j + 7] = wk3r * x0i + wk3i * x0r;
115 0 : wk1r = rdft_w[k2 + 2];
116 0 : wk1i = rdft_w[k2 + 3];
117 0 : wk3r = rdft_wk3ri_second[k1 + 0];
118 0 : wk3i = rdft_wk3ri_second[k1 + 1];
119 0 : x0r = a[j + 8] + a[j + 10];
120 0 : x0i = a[j + 9] + a[j + 11];
121 0 : x1r = a[j + 8] - a[j + 10];
122 0 : x1i = a[j + 9] - a[j + 11];
123 0 : x2r = a[j + 12] + a[j + 14];
124 0 : x2i = a[j + 13] + a[j + 15];
125 0 : x3r = a[j + 12] - a[j + 14];
126 0 : x3i = a[j + 13] - a[j + 15];
127 0 : a[j + 8] = x0r + x2r;
128 0 : a[j + 9] = x0i + x2i;
129 0 : x0r -= x2r;
130 0 : x0i -= x2i;
131 0 : a[j + 12] = -wk2i * x0r - wk2r * x0i;
132 0 : a[j + 13] = -wk2i * x0i + wk2r * x0r;
133 0 : x0r = x1r - x3i;
134 0 : x0i = x1i + x3r;
135 0 : a[j + 10] = wk1r * x0r - wk1i * x0i;
136 0 : a[j + 11] = wk1r * x0i + wk1i * x0r;
137 0 : x0r = x1r + x3i;
138 0 : x0i = x1i - x3r;
139 0 : a[j + 14] = wk3r * x0r - wk3i * x0i;
140 0 : a[j + 15] = wk3r * x0i + wk3i * x0r;
141 : }
142 0 : }
143 :
144 0 : static void cftmdl_128_C(float* a) {
145 0 : const int l = 8;
146 0 : const int n = 128;
147 0 : const int m = 32;
148 : int j0, j1, j2, j3, k, k1, k2, m2;
149 : float wk1r, wk1i, wk2r, wk2i, wk3r, wk3i;
150 : float x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i;
151 :
152 0 : for (j0 = 0; j0 < l; j0 += 2) {
153 0 : j1 = j0 + 8;
154 0 : j2 = j0 + 16;
155 0 : j3 = j0 + 24;
156 0 : x0r = a[j0 + 0] + a[j1 + 0];
157 0 : x0i = a[j0 + 1] + a[j1 + 1];
158 0 : x1r = a[j0 + 0] - a[j1 + 0];
159 0 : x1i = a[j0 + 1] - a[j1 + 1];
160 0 : x2r = a[j2 + 0] + a[j3 + 0];
161 0 : x2i = a[j2 + 1] + a[j3 + 1];
162 0 : x3r = a[j2 + 0] - a[j3 + 0];
163 0 : x3i = a[j2 + 1] - a[j3 + 1];
164 0 : a[j0 + 0] = x0r + x2r;
165 0 : a[j0 + 1] = x0i + x2i;
166 0 : a[j2 + 0] = x0r - x2r;
167 0 : a[j2 + 1] = x0i - x2i;
168 0 : a[j1 + 0] = x1r - x3i;
169 0 : a[j1 + 1] = x1i + x3r;
170 0 : a[j3 + 0] = x1r + x3i;
171 0 : a[j3 + 1] = x1i - x3r;
172 : }
173 0 : wk1r = rdft_w[2];
174 0 : for (j0 = m; j0 < l + m; j0 += 2) {
175 0 : j1 = j0 + 8;
176 0 : j2 = j0 + 16;
177 0 : j3 = j0 + 24;
178 0 : x0r = a[j0 + 0] + a[j1 + 0];
179 0 : x0i = a[j0 + 1] + a[j1 + 1];
180 0 : x1r = a[j0 + 0] - a[j1 + 0];
181 0 : x1i = a[j0 + 1] - a[j1 + 1];
182 0 : x2r = a[j2 + 0] + a[j3 + 0];
183 0 : x2i = a[j2 + 1] + a[j3 + 1];
184 0 : x3r = a[j2 + 0] - a[j3 + 0];
185 0 : x3i = a[j2 + 1] - a[j3 + 1];
186 0 : a[j0 + 0] = x0r + x2r;
187 0 : a[j0 + 1] = x0i + x2i;
188 0 : a[j2 + 0] = x2i - x0i;
189 0 : a[j2 + 1] = x0r - x2r;
190 0 : x0r = x1r - x3i;
191 0 : x0i = x1i + x3r;
192 0 : a[j1 + 0] = wk1r * (x0r - x0i);
193 0 : a[j1 + 1] = wk1r * (x0r + x0i);
194 0 : x0r = x3i + x1r;
195 0 : x0i = x3r - x1i;
196 0 : a[j3 + 0] = wk1r * (x0i - x0r);
197 0 : a[j3 + 1] = wk1r * (x0i + x0r);
198 : }
199 0 : k1 = 0;
200 0 : m2 = 2 * m;
201 0 : for (k = m2; k < n; k += m2) {
202 0 : k1 += 2;
203 0 : k2 = 2 * k1;
204 0 : wk2r = rdft_w[k1 + 0];
205 0 : wk2i = rdft_w[k1 + 1];
206 0 : wk1r = rdft_w[k2 + 0];
207 0 : wk1i = rdft_w[k2 + 1];
208 0 : wk3r = rdft_wk3ri_first[k1 + 0];
209 0 : wk3i = rdft_wk3ri_first[k1 + 1];
210 0 : for (j0 = k; j0 < l + k; j0 += 2) {
211 0 : j1 = j0 + 8;
212 0 : j2 = j0 + 16;
213 0 : j3 = j0 + 24;
214 0 : x0r = a[j0 + 0] + a[j1 + 0];
215 0 : x0i = a[j0 + 1] + a[j1 + 1];
216 0 : x1r = a[j0 + 0] - a[j1 + 0];
217 0 : x1i = a[j0 + 1] - a[j1 + 1];
218 0 : x2r = a[j2 + 0] + a[j3 + 0];
219 0 : x2i = a[j2 + 1] + a[j3 + 1];
220 0 : x3r = a[j2 + 0] - a[j3 + 0];
221 0 : x3i = a[j2 + 1] - a[j3 + 1];
222 0 : a[j0 + 0] = x0r + x2r;
223 0 : a[j0 + 1] = x0i + x2i;
224 0 : x0r -= x2r;
225 0 : x0i -= x2i;
226 0 : a[j2 + 0] = wk2r * x0r - wk2i * x0i;
227 0 : a[j2 + 1] = wk2r * x0i + wk2i * x0r;
228 0 : x0r = x1r - x3i;
229 0 : x0i = x1i + x3r;
230 0 : a[j1 + 0] = wk1r * x0r - wk1i * x0i;
231 0 : a[j1 + 1] = wk1r * x0i + wk1i * x0r;
232 0 : x0r = x1r + x3i;
233 0 : x0i = x1i - x3r;
234 0 : a[j3 + 0] = wk3r * x0r - wk3i * x0i;
235 0 : a[j3 + 1] = wk3r * x0i + wk3i * x0r;
236 : }
237 0 : wk1r = rdft_w[k2 + 2];
238 0 : wk1i = rdft_w[k2 + 3];
239 0 : wk3r = rdft_wk3ri_second[k1 + 0];
240 0 : wk3i = rdft_wk3ri_second[k1 + 1];
241 0 : for (j0 = k + m; j0 < l + (k + m); j0 += 2) {
242 0 : j1 = j0 + 8;
243 0 : j2 = j0 + 16;
244 0 : j3 = j0 + 24;
245 0 : x0r = a[j0 + 0] + a[j1 + 0];
246 0 : x0i = a[j0 + 1] + a[j1 + 1];
247 0 : x1r = a[j0 + 0] - a[j1 + 0];
248 0 : x1i = a[j0 + 1] - a[j1 + 1];
249 0 : x2r = a[j2 + 0] + a[j3 + 0];
250 0 : x2i = a[j2 + 1] + a[j3 + 1];
251 0 : x3r = a[j2 + 0] - a[j3 + 0];
252 0 : x3i = a[j2 + 1] - a[j3 + 1];
253 0 : a[j0 + 0] = x0r + x2r;
254 0 : a[j0 + 1] = x0i + x2i;
255 0 : x0r -= x2r;
256 0 : x0i -= x2i;
257 0 : a[j2 + 0] = -wk2i * x0r - wk2r * x0i;
258 0 : a[j2 + 1] = -wk2i * x0i + wk2r * x0r;
259 0 : x0r = x1r - x3i;
260 0 : x0i = x1i + x3r;
261 0 : a[j1 + 0] = wk1r * x0r - wk1i * x0i;
262 0 : a[j1 + 1] = wk1r * x0i + wk1i * x0r;
263 0 : x0r = x1r + x3i;
264 0 : x0i = x1i - x3r;
265 0 : a[j3 + 0] = wk3r * x0r - wk3i * x0i;
266 0 : a[j3 + 1] = wk3r * x0i + wk3i * x0r;
267 : }
268 : }
269 0 : }
270 :
271 0 : static void rftfsub_128_C(float* a) {
272 0 : const float* c = rdft_w + 32;
273 : int j1, j2, k1, k2;
274 : float wkr, wki, xr, xi, yr, yi;
275 :
276 0 : for (j1 = 1, j2 = 2; j2 < 64; j1 += 1, j2 += 2) {
277 0 : k2 = 128 - j2;
278 0 : k1 = 32 - j1;
279 0 : wkr = 0.5f - c[k1];
280 0 : wki = c[j1];
281 0 : xr = a[j2 + 0] - a[k2 + 0];
282 0 : xi = a[j2 + 1] + a[k2 + 1];
283 0 : yr = wkr * xr - wki * xi;
284 0 : yi = wkr * xi + wki * xr;
285 0 : a[j2 + 0] -= yr;
286 0 : a[j2 + 1] -= yi;
287 0 : a[k2 + 0] += yr;
288 0 : a[k2 + 1] -= yi;
289 : }
290 0 : }
291 :
292 0 : static void rftbsub_128_C(float* a) {
293 0 : const float* c = rdft_w + 32;
294 : int j1, j2, k1, k2;
295 : float wkr, wki, xr, xi, yr, yi;
296 :
297 0 : a[1] = -a[1];
298 0 : for (j1 = 1, j2 = 2; j2 < 64; j1 += 1, j2 += 2) {
299 0 : k2 = 128 - j2;
300 0 : k1 = 32 - j1;
301 0 : wkr = 0.5f - c[k1];
302 0 : wki = c[j1];
303 0 : xr = a[j2 + 0] - a[k2 + 0];
304 0 : xi = a[j2 + 1] + a[k2 + 1];
305 0 : yr = wkr * xr + wki * xi;
306 0 : yi = wkr * xi - wki * xr;
307 0 : a[j2 + 0] = a[j2 + 0] - yr;
308 0 : a[j2 + 1] = yi - a[j2 + 1];
309 0 : a[k2 + 0] = yr + a[k2 + 0];
310 0 : a[k2 + 1] = yi - a[k2 + 1];
311 : }
312 0 : a[65] = -a[65];
313 0 : }
314 : #endif
315 :
316 :
317 : } // namespace
318 :
319 0 : OouraFft::OouraFft() {
320 : #if defined(WEBRTC_ARCH_X86_FAMILY)
321 0 : use_sse2_ = (WebRtc_GetCPUInfo(kSSE2) != 0);
322 : #else
323 : use_sse2_ = false;
324 : #endif
325 0 : }
326 :
327 : OouraFft::~OouraFft() = default;
328 :
329 0 : void OouraFft::Fft(float* a) const {
330 : float xi;
331 0 : bitrv2_128(a);
332 0 : cftfsub_128(a);
333 0 : rftfsub_128(a);
334 0 : xi = a[0] - a[1];
335 0 : a[0] += a[1];
336 0 : a[1] = xi;
337 0 : }
338 0 : void OouraFft::InverseFft(float* a) const {
339 0 : a[1] = 0.5f * (a[0] - a[1]);
340 0 : a[0] -= a[1];
341 0 : rftbsub_128(a);
342 0 : bitrv2_128(a);
343 0 : cftbsub_128(a);
344 0 : }
345 :
346 0 : void OouraFft::cft1st_128(float* a) const {
347 : #if defined(MIPS_FPU_LE)
348 : cft1st_128_mips(a);
349 : #elif defined(WEBRTC_HAS_NEON)
350 : cft1st_128_neon(a);
351 : #elif defined(WEBRTC_ARCH_X86_FAMILY)
352 0 : if (use_sse2_) {
353 0 : cft1st_128_SSE2(a);
354 : } else {
355 0 : cft1st_128_C(a);
356 : }
357 : #else
358 : cft1st_128_C(a);
359 : #endif
360 0 : }
361 0 : void OouraFft::cftmdl_128(float* a) const {
362 : #if defined(MIPS_FPU_LE)
363 : cftmdl_128_mips(a);
364 : #elif defined(WEBRTC_HAS_NEON)
365 : cftmdl_128_neon(a);
366 : #elif defined(WEBRTC_ARCH_X86_FAMILY)
367 0 : if (use_sse2_) {
368 0 : cftmdl_128_SSE2(a);
369 : } else {
370 0 : cftmdl_128_C(a);
371 : }
372 : #else
373 : cftmdl_128_C(a);
374 : #endif
375 0 : }
376 0 : void OouraFft::rftfsub_128(float* a) const {
377 : #if defined(MIPS_FPU_LE)
378 : rftfsub_128_mips(a);
379 : #elif defined(WEBRTC_HAS_NEON)
380 : rftfsub_128_neon(a);
381 : #elif defined(WEBRTC_ARCH_X86_FAMILY)
382 0 : if (use_sse2_) {
383 0 : rftfsub_128_SSE2(a);
384 : } else {
385 0 : rftfsub_128_C(a);
386 : }
387 : #else
388 : rftfsub_128_C(a);
389 : #endif
390 0 : }
391 :
392 0 : void OouraFft::rftbsub_128(float* a) const {
393 : #if defined(MIPS_FPU_LE)
394 : rftbsub_128_mips(a);
395 : #elif defined(WEBRTC_HAS_NEON)
396 : rftbsub_128_neon(a);
397 : #elif defined(WEBRTC_ARCH_X86_FAMILY)
398 0 : if (use_sse2_) {
399 0 : rftbsub_128_SSE2(a);
400 : } else {
401 0 : rftbsub_128_C(a);
402 : }
403 : #else
404 : rftbsub_128_C(a);
405 : #endif
406 0 : }
407 :
408 0 : void OouraFft::cftbsub_128(float* a) const {
409 : int j, j1, j2, j3, l;
410 : float x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i;
411 :
412 0 : cft1st_128(a);
413 0 : cftmdl_128(a);
414 0 : l = 32;
415 :
416 0 : for (j = 0; j < l; j += 2) {
417 0 : j1 = j + l;
418 0 : j2 = j1 + l;
419 0 : j3 = j2 + l;
420 0 : x0r = a[j] + a[j1];
421 0 : x0i = -a[j + 1] - a[j1 + 1];
422 0 : x1r = a[j] - a[j1];
423 0 : x1i = -a[j + 1] + a[j1 + 1];
424 0 : x2r = a[j2] + a[j3];
425 0 : x2i = a[j2 + 1] + a[j3 + 1];
426 0 : x3r = a[j2] - a[j3];
427 0 : x3i = a[j2 + 1] - a[j3 + 1];
428 0 : a[j] = x0r + x2r;
429 0 : a[j + 1] = x0i - x2i;
430 0 : a[j2] = x0r - x2r;
431 0 : a[j2 + 1] = x0i + x2i;
432 0 : a[j1] = x1r - x3i;
433 0 : a[j1 + 1] = x1i - x3r;
434 0 : a[j3] = x1r + x3i;
435 0 : a[j3 + 1] = x1i + x3r;
436 : }
437 0 : }
438 :
439 0 : void OouraFft::cftfsub_128(float* a) const {
440 : int j, j1, j2, j3, l;
441 : float x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i;
442 :
443 0 : cft1st_128(a);
444 0 : cftmdl_128(a);
445 0 : l = 32;
446 0 : for (j = 0; j < l; j += 2) {
447 0 : j1 = j + l;
448 0 : j2 = j1 + l;
449 0 : j3 = j2 + l;
450 0 : x0r = a[j] + a[j1];
451 0 : x0i = a[j + 1] + a[j1 + 1];
452 0 : x1r = a[j] - a[j1];
453 0 : x1i = a[j + 1] - a[j1 + 1];
454 0 : x2r = a[j2] + a[j3];
455 0 : x2i = a[j2 + 1] + a[j3 + 1];
456 0 : x3r = a[j2] - a[j3];
457 0 : x3i = a[j2 + 1] - a[j3 + 1];
458 0 : a[j] = x0r + x2r;
459 0 : a[j + 1] = x0i + x2i;
460 0 : a[j2] = x0r - x2r;
461 0 : a[j2 + 1] = x0i - x2i;
462 0 : a[j1] = x1r - x3i;
463 0 : a[j1 + 1] = x1i + x3r;
464 0 : a[j3] = x1r + x3i;
465 0 : a[j3 + 1] = x1i - x3r;
466 : }
467 0 : }
468 :
469 0 : void OouraFft::bitrv2_128(float* a) const {
470 : /*
471 : Following things have been attempted but are no faster:
472 : (a) Storing the swap indexes in a LUT (index calculations are done
473 : for 'free' while waiting on memory/L1).
474 : (b) Consolidate the load/store of two consecutive floats by a 64 bit
475 : integer (execution is memory/L1 bound).
476 : (c) Do a mix of floats and 64 bit integer to maximize register
477 : utilization (execution is memory/L1 bound).
478 : (d) Replacing ip[i] by ((k<<31)>>25) + ((k >> 1)<<5).
479 : (e) Hard-coding of the offsets to completely eliminates index
480 : calculations.
481 : */
482 :
483 : unsigned int j, j1, k, k1;
484 : float xr, xi, yr, yi;
485 :
486 0 : const int ip[4] = {0, 64, 32, 96};
487 0 : for (k = 0; k < 4; k++) {
488 0 : for (j = 0; j < k; j++) {
489 0 : j1 = 2 * j + ip[k];
490 0 : k1 = 2 * k + ip[j];
491 0 : xr = a[j1 + 0];
492 0 : xi = a[j1 + 1];
493 0 : yr = a[k1 + 0];
494 0 : yi = a[k1 + 1];
495 0 : a[j1 + 0] = yr;
496 0 : a[j1 + 1] = yi;
497 0 : a[k1 + 0] = xr;
498 0 : a[k1 + 1] = xi;
499 0 : j1 += 8;
500 0 : k1 += 16;
501 0 : xr = a[j1 + 0];
502 0 : xi = a[j1 + 1];
503 0 : yr = a[k1 + 0];
504 0 : yi = a[k1 + 1];
505 0 : a[j1 + 0] = yr;
506 0 : a[j1 + 1] = yi;
507 0 : a[k1 + 0] = xr;
508 0 : a[k1 + 1] = xi;
509 0 : j1 += 8;
510 0 : k1 -= 8;
511 0 : xr = a[j1 + 0];
512 0 : xi = a[j1 + 1];
513 0 : yr = a[k1 + 0];
514 0 : yi = a[k1 + 1];
515 0 : a[j1 + 0] = yr;
516 0 : a[j1 + 1] = yi;
517 0 : a[k1 + 0] = xr;
518 0 : a[k1 + 1] = xi;
519 0 : j1 += 8;
520 0 : k1 += 16;
521 0 : xr = a[j1 + 0];
522 0 : xi = a[j1 + 1];
523 0 : yr = a[k1 + 0];
524 0 : yi = a[k1 + 1];
525 0 : a[j1 + 0] = yr;
526 0 : a[j1 + 1] = yi;
527 0 : a[k1 + 0] = xr;
528 0 : a[k1 + 1] = xi;
529 : }
530 0 : j1 = 2 * k + 8 + ip[k];
531 0 : k1 = j1 + 8;
532 0 : xr = a[j1 + 0];
533 0 : xi = a[j1 + 1];
534 0 : yr = a[k1 + 0];
535 0 : yi = a[k1 + 1];
536 0 : a[j1 + 0] = yr;
537 0 : a[j1 + 1] = yi;
538 0 : a[k1 + 0] = xr;
539 0 : a[k1 + 1] = xi;
540 : }
541 0 : }
542 :
543 : } // namespace webrtc
|
