Mercurial > hg > sv-dependency-builds

comparison src/fftw-3.3.3/dft/simd/common/t2sv_16.c @ 10:37bf6b4a2645

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Add FFTW3
author Chris Cannam
date Wed, 20 Mar 2013 15:35:50 +0000
parents
children
comparison
equal deleted inserted replaced
9:c0fb53affa76 10:37bf6b4a2645
1 /*
2 * Copyright (c) 2003, 2007-11 Matteo Frigo
3 * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
18 *
19 */
20
21 /* This file was automatically generated --- DO NOT EDIT */
22 /* Generated on Sun Nov 25 07:39:26 EST 2012 */
23
24 #include "codelet-dft.h"
25
26 #ifdef HAVE_FMA
27
28 /* Generated by: ../../../genfft/gen_twiddle.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -twiddle-log3 -precompute-twiddles -n 16 -name t2sv_16 -include ts.h */
29
30 /*
31 * This function contains 196 FP additions, 134 FP multiplications,
32 * (or, 104 additions, 42 multiplications, 92 fused multiply/add),
33 * 120 stack variables, 3 constants, and 64 memory accesses
34 */
35 #include "ts.h"
36
37 static void t2sv_16(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms)
38 {
39 DVK(KP923879532, +0.923879532511286756128183189396788286822416626);
40 DVK(KP414213562, +0.414213562373095048801688724209698078569671875);
41 DVK(KP707106781, +0.707106781186547524400844362104849039284835938);
42 {
43 INT m;
44 for (m = mb, W = W + (mb * 8); m < me; m = m + (2 * VL), ri = ri + ((2 * VL) * ms), ii = ii + ((2 * VL) * ms), W = W + ((2 * VL) * 8), MAKE_VOLATILE_STRIDE(32, rs)) {
45 V T34, T30, T2N, T2v, T2M, T2g, T3V, T3X, T32, T2U, T33, T2X, T2O, T2K, T3P;
46 V T3R;
47 {
48 V T2, Tf, TM, TO, T3, T6, T5, Th;
49 T2 = LDW(&(W[0]));
50 Tf = LDW(&(W[TWVL * 2]));
51 TM = LDW(&(W[TWVL * 6]));
52 TO = LDW(&(W[TWVL * 7]));
53 T3 = LDW(&(W[TWVL * 4]));
54 T6 = LDW(&(W[TWVL * 5]));
55 T5 = LDW(&(W[TWVL * 1]));
56 Th = LDW(&(W[TWVL * 3]));
57 {
58 V TW, TZ, Te, T1U, T3A, T3L, T2D, T1G, T3h, T2A, T2B, T1R, T3i, T2I, Tx;
59 V T3M, T1Z, T3w, TL, T26, T25, T37, T1l, T2q, T1d, T2o, T2l, T3c, T1r, T2s;
60 V TX, T10, TV, T2a;
61 {
62 V Tz, TP, TT, Tq, TF, Tu, TI, Tm, TC, T1j, T1p, T1m, T1f, T1O, T1M;
63 V T1K, T2F, Tj, Tn, T1Q, T2G, Tk, T1V, Tr, Tv;
64 {
65 V T1, Ti, Tb, T3z, T8, Tc, T1u, T1D, T1L, T1z, T9, T3x, T1v, T1w, T1A;
66 V T1E;
67 {
68 V T7, T1i, T1e, T1C, T1y;
69 T1 = LD(&(ri[0]), ms, &(ri[0]));
70 {
71 V Tg, TN, TS, Tp;
72 Tg = VMUL(T2, Tf);
73 TN = VMUL(T2, TM);
74 TS = VMUL(T2, TO);
75 Tp = VMUL(Tf, T3);
76 {
77 V T4, Tt, Ta, Tl;
78 T4 = VMUL(T2, T3);
79 Tt = VMUL(Tf, T6);
80 Ta = VMUL(T2, T6);
81 Tl = VMUL(T2, Th);
82 Ti = VFNMS(T5, Th, Tg);
83 Tz = VFMA(T5, Th, Tg);
84 TP = VFMA(T5, TO, TN);
85 TT = VFNMS(T5, TM, TS);
86 TW = VFMA(Th, T6, Tp);
87 Tq = VFNMS(Th, T6, Tp);
88 TF = VFNMS(T5, T6, T4);
89 T7 = VFMA(T5, T6, T4);
90 Tu = VFMA(Th, T3, Tt);
91 TZ = VFNMS(Th, T3, Tt);
92 TI = VFMA(T5, T3, Ta);
93 Tb = VFNMS(T5, T3, Ta);
94 Tm = VFMA(T5, Tf, Tl);
95 TC = VFNMS(T5, Tf, Tl);
96 T1i = VMUL(Ti, T6);
97 T1e = VMUL(Ti, T3);
98 T1C = VMUL(Tz, T6);
99 T1y = VMUL(Tz, T3);
100 T3z = LD(&(ii[0]), ms, &(ii[0]));
101 }
102 }
103 T8 = LD(&(ri[WS(rs, 8)]), ms, &(ri[0]));
104 Tc = LD(&(ii[WS(rs, 8)]), ms, &(ii[0]));
105 T1u = LD(&(ri[WS(rs, 15)]), ms, &(ri[WS(rs, 1)]));
106 T1j = VFNMS(Tm, T3, T1i);
107 T1p = VFMA(Tm, T3, T1i);
108 T1m = VFNMS(Tm, T6, T1e);
109 T1f = VFMA(Tm, T6, T1e);
110 T1D = VFNMS(TC, T3, T1C);
111 T1O = VFMA(TC, T3, T1C);
112 T1L = VFNMS(TC, T6, T1y);
113 T1z = VFMA(TC, T6, T1y);
114 T9 = VMUL(T7, T8);
115 T3x = VMUL(T7, Tc);
116 T1v = VMUL(TM, T1u);
117 T1w = LD(&(ii[WS(rs, 15)]), ms, &(ii[WS(rs, 1)]));
118 T1A = LD(&(ri[WS(rs, 7)]), ms, &(ri[WS(rs, 1)]));
119 T1E = LD(&(ii[WS(rs, 7)]), ms, &(ii[WS(rs, 1)]));
120 }
121 {
122 V T1x, T2x, T1F, T2z, T1N, T1P;
123 {
124 V T1H, T1J, T1I, T2E;
125 {
126 V Td, T3y, T2w, T1B, T2y;
127 T1H = LD(&(ri[WS(rs, 3)]), ms, &(ri[WS(rs, 1)]));
128 T1J = LD(&(ii[WS(rs, 3)]), ms, &(ii[WS(rs, 1)]));
129 Td = VFMA(Tb, Tc, T9);
130 T3y = VFNMS(Tb, T8, T3x);
131 T1M = LD(&(ri[WS(rs, 11)]), ms, &(ri[WS(rs, 1)]));
132 T1x = VFMA(TO, T1w, T1v);
133 T2w = VMUL(TM, T1w);
134 T1B = VMUL(T1z, T1A);
135 T2y = VMUL(T1z, T1E);
136 T1I = VMUL(Tf, T1H);
137 T2E = VMUL(Tf, T1J);
138 Te = VADD(T1, Td);
139 T1U = VSUB(T1, Td);
140 T3A = VADD(T3y, T3z);
141 T3L = VSUB(T3z, T3y);
142 T2x = VFNMS(TO, T1u, T2w);
143 T1F = VFMA(T1D, T1E, T1B);
144 T2z = VFNMS(T1D, T1A, T2y);
145 T1N = VMUL(T1L, T1M);
146 T1P = LD(&(ii[WS(rs, 11)]), ms, &(ii[WS(rs, 1)]));
147 }
148 T1K = VFMA(Th, T1J, T1I);
149 T2F = VFNMS(Th, T1H, T2E);
150 }
151 Tj = LD(&(ri[WS(rs, 4)]), ms, &(ri[0]));
152 Tn = LD(&(ii[WS(rs, 4)]), ms, &(ii[0]));
153 T2D = VSUB(T1x, T1F);
154 T1G = VADD(T1x, T1F);
155 T3h = VADD(T2x, T2z);
156 T2A = VSUB(T2x, T2z);
157 T1Q = VFMA(T1O, T1P, T1N);
158 T2G = VMUL(T1L, T1P);
159 Tk = VMUL(Ti, Tj);
160 T1V = VMUL(Ti, Tn);
161 Tr = LD(&(ri[WS(rs, 12)]), ms, &(ri[0]));
162 Tv = LD(&(ii[WS(rs, 12)]), ms, &(ii[0]));
163 }
164 }
165 {
166 V TE, T22, T15, T17, TK, T16, T2h, T24, T19, T1b;
167 {
168 V To, T1W, TG, TJ, Tw, T1Y, TH, T23;
169 {
170 V TA, TD, TB, T21, T2H, Ts, T1X;
171 TA = LD(&(ri[WS(rs, 2)]), ms, &(ri[0]));
172 TD = LD(&(ii[WS(rs, 2)]), ms, &(ii[0]));
173 T2B = VSUB(T1K, T1Q);
174 T1R = VADD(T1K, T1Q);
175 T2H = VFNMS(T1O, T1M, T2G);
176 To = VFMA(Tm, Tn, Tk);
177 T1W = VFNMS(Tm, Tj, T1V);
178 Ts = VMUL(Tq, Tr);
179 T1X = VMUL(Tq, Tv);
180 TB = VMUL(Tz, TA);
181 T21 = VMUL(Tz, TD);
182 TG = LD(&(ri[WS(rs, 10)]), ms, &(ri[0]));
183 T3i = VADD(T2F, T2H);
184 T2I = VSUB(T2F, T2H);
185 TJ = LD(&(ii[WS(rs, 10)]), ms, &(ii[0]));
186 Tw = VFMA(Tu, Tv, Ts);
187 T1Y = VFNMS(Tu, Tr, T1X);
188 TE = VFMA(TC, TD, TB);
189 T22 = VFNMS(TC, TA, T21);
190 TH = VMUL(TF, TG);
191 }
192 T15 = LD(&(ri[WS(rs, 1)]), ms, &(ri[WS(rs, 1)]));
193 T17 = LD(&(ii[WS(rs, 1)]), ms, &(ii[WS(rs, 1)]));
194 T23 = VMUL(TF, TJ);
195 Tx = VADD(To, Tw);
196 T3M = VSUB(To, Tw);
197 T1Z = VSUB(T1W, T1Y);
198 T3w = VADD(T1W, T1Y);
199 TK = VFMA(TI, TJ, TH);
200 T16 = VMUL(T2, T15);
201 T2h = VMUL(T2, T17);
202 T24 = VFNMS(TI, TG, T23);
203 T19 = LD(&(ri[WS(rs, 9)]), ms, &(ri[WS(rs, 1)]));
204 T1b = LD(&(ii[WS(rs, 9)]), ms, &(ii[WS(rs, 1)]));
205 }
206 {
207 V T1g, T1k, T18, T2i, T1a, T2j, T1h, T2p, T1n, T1q;
208 T1g = LD(&(ri[WS(rs, 5)]), ms, &(ri[WS(rs, 1)]));
209 T1k = LD(&(ii[WS(rs, 5)]), ms, &(ii[WS(rs, 1)]));
210 TL = VADD(TE, TK);
211 T26 = VSUB(TE, TK);
212 T18 = VFMA(T5, T17, T16);
213 T2i = VFNMS(T5, T15, T2h);
214 T25 = VSUB(T22, T24);
215 T37 = VADD(T22, T24);
216 T1a = VMUL(T3, T19);
217 T2j = VMUL(T3, T1b);
218 T1h = VMUL(T1f, T1g);
219 T2p = VMUL(T1f, T1k);
220 T1n = LD(&(ri[WS(rs, 13)]), ms, &(ri[WS(rs, 1)]));
221 T1q = LD(&(ii[WS(rs, 13)]), ms, &(ii[WS(rs, 1)]));
222 {
223 V TQ, TU, TR, T29;
224 {
225 V T1c, T2k, T1o, T2r;
226 TQ = LD(&(ri[WS(rs, 14)]), ms, &(ri[0]));
227 TU = LD(&(ii[WS(rs, 14)]), ms, &(ii[0]));
228 T1c = VFMA(T6, T1b, T1a);
229 T2k = VFNMS(T6, T19, T2j);
230 T1l = VFMA(T1j, T1k, T1h);
231 T2q = VFNMS(T1j, T1g, T2p);
232 T1o = VMUL(T1m, T1n);
233 T2r = VMUL(T1m, T1q);
234 TR = VMUL(TP, TQ);
235 T29 = VMUL(TP, TU);
236 T1d = VADD(T18, T1c);
237 T2o = VSUB(T18, T1c);
238 T2l = VSUB(T2i, T2k);
239 T3c = VADD(T2i, T2k);
240 T1r = VFMA(T1p, T1q, T1o);
241 T2s = VFNMS(T1p, T1n, T2r);
242 TX = LD(&(ri[WS(rs, 6)]), ms, &(ri[0]));
243 T10 = LD(&(ii[WS(rs, 6)]), ms, &(ii[0]));
244 }
245 TV = VFMA(TT, TU, TR);
246 T2a = VFNMS(TT, TQ, T29);
247 }
248 }
249 }
250 }
251 {
252 V T36, Ty, T3B, T3G, T1s, T2m, T2t, T3d, TY, T2b, T3g, T1S, T3s, T3j;
253 T36 = VSUB(Te, Tx);
254 Ty = VADD(Te, Tx);
255 T3B = VADD(T3w, T3A);
256 T3G = VSUB(T3A, T3w);
257 T1s = VADD(T1l, T1r);
258 T2m = VSUB(T1l, T1r);
259 T2t = VSUB(T2q, T2s);
260 T3d = VADD(T2q, T2s);
261 TY = VMUL(TW, TX);
262 T2b = VMUL(TW, T10);
263 T3g = VSUB(T1G, T1R);
264 T1S = VADD(T1G, T1R);
265 T3s = VADD(T3h, T3i);
266 T3j = VSUB(T3h, T3i);
267 {
268 V T3D, T1T, T3u, T3t, T28, T12, T38, T2d, T3n, T3f;
269 {
270 V T1t, T3b, T3e, T3r, T11, T2c;
271 T1t = VADD(T1d, T1s);
272 T3b = VSUB(T1d, T1s);
273 T3e = VSUB(T3c, T3d);
274 T3r = VADD(T3c, T3d);
275 T11 = VFMA(TZ, T10, TY);
276 T2c = VFNMS(TZ, TX, T2b);
277 T3D = VSUB(T1S, T1t);
278 T1T = VADD(T1t, T1S);
279 T3u = VADD(T3r, T3s);
280 T3t = VSUB(T3r, T3s);
281 T28 = VSUB(TV, T11);
282 T12 = VADD(TV, T11);
283 T38 = VADD(T2a, T2c);
284 T2d = VSUB(T2a, T2c);
285 T3n = VSUB(T3e, T3b);
286 T3f = VADD(T3b, T3e);
287 }
288 {
289 V T2Q, T20, T3N, T3T, T2J, T2C, T2W, T2V, T3O, T2f, T3U, T2T;
290 {
291 V T2R, T27, T2e, T2S, T13, T3F;
292 T2Q = VADD(T1U, T1Z);
293 T20 = VSUB(T1U, T1Z);
294 T3N = VSUB(T3L, T3M);
295 T3T = VADD(T3M, T3L);
296 T13 = VADD(TL, T12);
297 T3F = VSUB(T12, TL);
298 {
299 V T3v, T39, T3o, T3k;
300 T3v = VADD(T37, T38);
301 T39 = VSUB(T37, T38);
302 T3o = VADD(T3g, T3j);
303 T3k = VSUB(T3g, T3j);
304 {
305 V T3H, T3J, T14, T3q;
306 T3H = VADD(T3F, T3G);
307 T3J = VSUB(T3G, T3F);
308 T14 = VADD(Ty, T13);
309 T3q = VSUB(Ty, T13);
310 {
311 V T3a, T3m, T3C, T3E;
312 T3a = VADD(T36, T39);
313 T3m = VSUB(T36, T39);
314 T3C = VADD(T3v, T3B);
315 T3E = VSUB(T3B, T3v);
316 {
317 V T3I, T3p, T3l, T3K;
318 T3I = VADD(T3n, T3o);
319 T3p = VSUB(T3n, T3o);
320 T3l = VADD(T3f, T3k);
321 T3K = VSUB(T3k, T3f);
322 ST(&(ri[WS(rs, 4)]), VADD(T3q, T3t), ms, &(ri[0]));
323 ST(&(ri[WS(rs, 12)]), VSUB(T3q, T3t), ms, &(ri[0]));
324 ST(&(ri[0]), VADD(T14, T1T), ms, &(ri[0]));
325 ST(&(ri[WS(rs, 8)]), VSUB(T14, T1T), ms, &(ri[0]));
326 ST(&(ii[WS(rs, 4)]), VADD(T3D, T3E), ms, &(ii[0]));
327 ST(&(ii[WS(rs, 12)]), VSUB(T3E, T3D), ms, &(ii[0]));
328 ST(&(ii[0]), VADD(T3u, T3C), ms, &(ii[0]));
329 ST(&(ii[WS(rs, 8)]), VSUB(T3C, T3u), ms, &(ii[0]));
330 ST(&(ri[WS(rs, 6)]), VFMA(LDK(KP707106781), T3p, T3m), ms, &(ri[0]));
331 ST(&(ri[WS(rs, 14)]), VFNMS(LDK(KP707106781), T3p, T3m), ms, &(ri[0]));
332 ST(&(ii[WS(rs, 10)]), VFNMS(LDK(KP707106781), T3I, T3H), ms, &(ii[0]));
333 ST(&(ii[WS(rs, 2)]), VFMA(LDK(KP707106781), T3I, T3H), ms, &(ii[0]));
334 ST(&(ii[WS(rs, 14)]), VFNMS(LDK(KP707106781), T3K, T3J), ms, &(ii[0]));
335 ST(&(ii[WS(rs, 6)]), VFMA(LDK(KP707106781), T3K, T3J), ms, &(ii[0]));
336 ST(&(ri[WS(rs, 2)]), VFMA(LDK(KP707106781), T3l, T3a), ms, &(ri[0]));
337 ST(&(ri[WS(rs, 10)]), VFNMS(LDK(KP707106781), T3l, T3a), ms, &(ri[0]));
338 T2R = VADD(T26, T25);
339 T27 = VSUB(T25, T26);
340 T2e = VADD(T28, T2d);
341 T2S = VSUB(T28, T2d);
342 }
343 }
344 }
345 }
346 {
347 V T2Y, T2Z, T2n, T2u;
348 T2J = VSUB(T2D, T2I);
349 T2Y = VADD(T2D, T2I);
350 T2Z = VSUB(T2A, T2B);
351 T2C = VADD(T2A, T2B);
352 T2W = VSUB(T2l, T2m);
353 T2n = VADD(T2l, T2m);
354 T2u = VSUB(T2o, T2t);
355 T2V = VADD(T2o, T2t);
356 T3O = VADD(T27, T2e);
357 T2f = VSUB(T27, T2e);
358 T34 = VFMA(LDK(KP414213562), T2Y, T2Z);
359 T30 = VFNMS(LDK(KP414213562), T2Z, T2Y);
360 T3U = VSUB(T2S, T2R);
361 T2T = VADD(T2R, T2S);
362 T2N = VFNMS(LDK(KP414213562), T2n, T2u);
363 T2v = VFMA(LDK(KP414213562), T2u, T2n);
364 }
365 }
366 T2M = VFNMS(LDK(KP707106781), T2f, T20);
367 T2g = VFMA(LDK(KP707106781), T2f, T20);
368 T3V = VFMA(LDK(KP707106781), T3U, T3T);
369 T3X = VFNMS(LDK(KP707106781), T3U, T3T);
370 T32 = VFNMS(LDK(KP707106781), T2T, T2Q);
371 T2U = VFMA(LDK(KP707106781), T2T, T2Q);
372 T33 = VFNMS(LDK(KP414213562), T2V, T2W);
373 T2X = VFMA(LDK(KP414213562), T2W, T2V);
374 T2O = VFMA(LDK(KP414213562), T2C, T2J);
375 T2K = VFNMS(LDK(KP414213562), T2J, T2C);
376 T3P = VFMA(LDK(KP707106781), T3O, T3N);
377 T3R = VFNMS(LDK(KP707106781), T3O, T3N);
378 }
379 }
380 }
381 }
382 }
383 {
384 V T3Q, T35, T31, T3S;
385 T3Q = VADD(T33, T34);
386 T35 = VSUB(T33, T34);
387 T31 = VADD(T2X, T30);
388 T3S = VSUB(T30, T2X);
389 {
390 V T3W, T2P, T2L, T3Y;
391 T3W = VSUB(T2O, T2N);
392 T2P = VADD(T2N, T2O);
393 T2L = VSUB(T2v, T2K);
394 T3Y = VADD(T2v, T2K);
395 ST(&(ri[WS(rs, 5)]), VFMA(LDK(KP923879532), T35, T32), ms, &(ri[WS(rs, 1)]));
396 ST(&(ri[WS(rs, 13)]), VFNMS(LDK(KP923879532), T35, T32), ms, &(ri[WS(rs, 1)]));
397 ST(&(ii[WS(rs, 9)]), VFNMS(LDK(KP923879532), T3Q, T3P), ms, &(ii[WS(rs, 1)]));
398 ST(&(ii[WS(rs, 1)]), VFMA(LDK(KP923879532), T3Q, T3P), ms, &(ii[WS(rs, 1)]));
399 ST(&(ii[WS(rs, 13)]), VFNMS(LDK(KP923879532), T3S, T3R), ms, &(ii[WS(rs, 1)]));
400 ST(&(ii[WS(rs, 5)]), VFMA(LDK(KP923879532), T3S, T3R), ms, &(ii[WS(rs, 1)]));
401 ST(&(ri[WS(rs, 1)]), VFMA(LDK(KP923879532), T31, T2U), ms, &(ri[WS(rs, 1)]));
402 ST(&(ri[WS(rs, 9)]), VFNMS(LDK(KP923879532), T31, T2U), ms, &(ri[WS(rs, 1)]));
403 ST(&(ri[WS(rs, 15)]), VFMA(LDK(KP923879532), T2P, T2M), ms, &(ri[WS(rs, 1)]));
404 ST(&(ri[WS(rs, 7)]), VFNMS(LDK(KP923879532), T2P, T2M), ms, &(ri[WS(rs, 1)]));
405 ST(&(ii[WS(rs, 11)]), VFNMS(LDK(KP923879532), T3W, T3V), ms, &(ii[WS(rs, 1)]));
406 ST(&(ii[WS(rs, 3)]), VFMA(LDK(KP923879532), T3W, T3V), ms, &(ii[WS(rs, 1)]));
407 ST(&(ii[WS(rs, 15)]), VFMA(LDK(KP923879532), T3Y, T3X), ms, &(ii[WS(rs, 1)]));
408 ST(&(ii[WS(rs, 7)]), VFNMS(LDK(KP923879532), T3Y, T3X), ms, &(ii[WS(rs, 1)]));
409 ST(&(ri[WS(rs, 3)]), VFMA(LDK(KP923879532), T2L, T2g), ms, &(ri[WS(rs, 1)]));
410 ST(&(ri[WS(rs, 11)]), VFNMS(LDK(KP923879532), T2L, T2g), ms, &(ri[WS(rs, 1)]));
411 }
412 }
413 }
414 }
415 VLEAVE();
416 }
417
418 static const tw_instr twinstr[] = {
419 VTW(0, 1),
420 VTW(0, 3),
421 VTW(0, 9),
422 VTW(0, 15),
423 {TW_NEXT, (2 * VL), 0}
424 };
425
426 static const ct_desc desc = { 16, XSIMD_STRING("t2sv_16"), twinstr, &GENUS, {104, 42, 92, 0}, 0, 0, 0 };
427
428 void XSIMD(codelet_t2sv_16) (planner *p) {
429 X(kdft_dit_register) (p, t2sv_16, &desc);
430 }
431 #else /* HAVE_FMA */
432
433 /* Generated by: ../../../genfft/gen_twiddle.native -simd -compact -variables 4 -pipeline-latency 8 -twiddle-log3 -precompute-twiddles -n 16 -name t2sv_16 -include ts.h */
434
435 /*
436 * This function contains 196 FP additions, 108 FP multiplications,
437 * (or, 156 additions, 68 multiplications, 40 fused multiply/add),
438 * 82 stack variables, 3 constants, and 64 memory accesses
439 */
440 #include "ts.h"
441
442 static void t2sv_16(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms)
443 {
444 DVK(KP382683432, +0.382683432365089771728459984030398866761344562);
445 DVK(KP923879532, +0.923879532511286756128183189396788286822416626);
446 DVK(KP707106781, +0.707106781186547524400844362104849039284835938);
447 {
448 INT m;
449 for (m = mb, W = W + (mb * 8); m < me; m = m + (2 * VL), ri = ri + ((2 * VL) * ms), ii = ii + ((2 * VL) * ms), W = W + ((2 * VL) * 8), MAKE_VOLATILE_STRIDE(32, rs)) {
450 V T2, T5, Tg, Ti, Tk, To, TE, TC, T6, T3, T8, TW, TJ, Tt, TU;
451 V Tc, Tx, TH, TN, TO, TP, TR, T1f, T1k, T1b, T1i, T1y, T1H, T1u, T1F;
452 {
453 V T7, Tv, Ta, Ts, T4, Tw, Tb, Tr;
454 {
455 V Th, Tn, Tj, Tm;
456 T2 = LDW(&(W[0]));
457 T5 = LDW(&(W[TWVL * 1]));
458 Tg = LDW(&(W[TWVL * 2]));
459 Ti = LDW(&(W[TWVL * 3]));
460 Th = VMUL(T2, Tg);
461 Tn = VMUL(T5, Tg);
462 Tj = VMUL(T5, Ti);
463 Tm = VMUL(T2, Ti);
464 Tk = VSUB(Th, Tj);
465 To = VADD(Tm, Tn);
466 TE = VSUB(Tm, Tn);
467 TC = VADD(Th, Tj);
468 T6 = LDW(&(W[TWVL * 5]));
469 T7 = VMUL(T5, T6);
470 Tv = VMUL(Tg, T6);
471 Ta = VMUL(T2, T6);
472 Ts = VMUL(Ti, T6);
473 T3 = LDW(&(W[TWVL * 4]));
474 T4 = VMUL(T2, T3);
475 Tw = VMUL(Ti, T3);
476 Tb = VMUL(T5, T3);
477 Tr = VMUL(Tg, T3);
478 }
479 T8 = VADD(T4, T7);
480 TW = VSUB(Tv, Tw);
481 TJ = VADD(Ta, Tb);
482 Tt = VSUB(Tr, Ts);
483 TU = VADD(Tr, Ts);
484 Tc = VSUB(Ta, Tb);
485 Tx = VADD(Tv, Tw);
486 TH = VSUB(T4, T7);
487 TN = LDW(&(W[TWVL * 6]));
488 TO = LDW(&(W[TWVL * 7]));
489 TP = VFMA(T2, TN, VMUL(T5, TO));
490 TR = VFNMS(T5, TN, VMUL(T2, TO));
491 {
492 V T1d, T1e, T19, T1a;
493 T1d = VMUL(Tk, T6);
494 T1e = VMUL(To, T3);
495 T1f = VSUB(T1d, T1e);
496 T1k = VADD(T1d, T1e);
497 T19 = VMUL(Tk, T3);
498 T1a = VMUL(To, T6);
499 T1b = VADD(T19, T1a);
500 T1i = VSUB(T19, T1a);
501 }
502 {
503 V T1w, T1x, T1s, T1t;
504 T1w = VMUL(TC, T6);
505 T1x = VMUL(TE, T3);
506 T1y = VSUB(T1w, T1x);
507 T1H = VADD(T1w, T1x);
508 T1s = VMUL(TC, T3);
509 T1t = VMUL(TE, T6);
510 T1u = VADD(T1s, T1t);
511 T1F = VSUB(T1s, T1t);
512 }
513 }
514 {
515 V Tf, T3r, T1N, T3e, TA, T3s, T1Q, T3b, TM, T2M, T1W, T2w, TZ, T2N, T21;
516 V T2x, T1B, T1K, T2V, T2W, T2X, T2Y, T2j, T2D, T2o, T2E, T18, T1n, T2Q, T2R;
517 V T2S, T2T, T28, T2A, T2d, T2B;
518 {
519 V T1, T3d, Te, T3c, T9, Td;
520 T1 = LD(&(ri[0]), ms, &(ri[0]));
521 T3d = LD(&(ii[0]), ms, &(ii[0]));
522 T9 = LD(&(ri[WS(rs, 8)]), ms, &(ri[0]));
523 Td = LD(&(ii[WS(rs, 8)]), ms, &(ii[0]));
524 Te = VFMA(T8, T9, VMUL(Tc, Td));
525 T3c = VFNMS(Tc, T9, VMUL(T8, Td));
526 Tf = VADD(T1, Te);
527 T3r = VSUB(T3d, T3c);
528 T1N = VSUB(T1, Te);
529 T3e = VADD(T3c, T3d);
530 }
531 {
532 V Tq, T1O, Tz, T1P;
533 {
534 V Tl, Tp, Tu, Ty;
535 Tl = LD(&(ri[WS(rs, 4)]), ms, &(ri[0]));
536 Tp = LD(&(ii[WS(rs, 4)]), ms, &(ii[0]));
537 Tq = VFMA(Tk, Tl, VMUL(To, Tp));
538 T1O = VFNMS(To, Tl, VMUL(Tk, Tp));
539 Tu = LD(&(ri[WS(rs, 12)]), ms, &(ri[0]));
540 Ty = LD(&(ii[WS(rs, 12)]), ms, &(ii[0]));
541 Tz = VFMA(Tt, Tu, VMUL(Tx, Ty));
542 T1P = VFNMS(Tx, Tu, VMUL(Tt, Ty));
543 }
544 TA = VADD(Tq, Tz);
545 T3s = VSUB(Tq, Tz);
546 T1Q = VSUB(T1O, T1P);
547 T3b = VADD(T1O, T1P);
548 }
549 {
550 V TG, T1S, TL, T1T, T1U, T1V;
551 {
552 V TD, TF, TI, TK;
553 TD = LD(&(ri[WS(rs, 2)]), ms, &(ri[0]));
554 TF = LD(&(ii[WS(rs, 2)]), ms, &(ii[0]));
555 TG = VFMA(TC, TD, VMUL(TE, TF));
556 T1S = VFNMS(TE, TD, VMUL(TC, TF));
557 TI = LD(&(ri[WS(rs, 10)]), ms, &(ri[0]));
558 TK = LD(&(ii[WS(rs, 10)]), ms, &(ii[0]));
559 TL = VFMA(TH, TI, VMUL(TJ, TK));
560 T1T = VFNMS(TJ, TI, VMUL(TH, TK));
561 }
562 TM = VADD(TG, TL);
563 T2M = VADD(T1S, T1T);
564 T1U = VSUB(T1S, T1T);
565 T1V = VSUB(TG, TL);
566 T1W = VSUB(T1U, T1V);
567 T2w = VADD(T1V, T1U);
568 }
569 {
570 V TT, T1Y, TY, T1Z, T1X, T20;
571 {
572 V TQ, TS, TV, TX;
573 TQ = LD(&(ri[WS(rs, 14)]), ms, &(ri[0]));
574 TS = LD(&(ii[WS(rs, 14)]), ms, &(ii[0]));
575 TT = VFMA(TP, TQ, VMUL(TR, TS));
576 T1Y = VFNMS(TR, TQ, VMUL(TP, TS));
577 TV = LD(&(ri[WS(rs, 6)]), ms, &(ri[0]));
578 TX = LD(&(ii[WS(rs, 6)]), ms, &(ii[0]));
579 TY = VFMA(TU, TV, VMUL(TW, TX));
580 T1Z = VFNMS(TW, TV, VMUL(TU, TX));
581 }
582 TZ = VADD(TT, TY);
583 T2N = VADD(T1Y, T1Z);
584 T1X = VSUB(TT, TY);
585 T20 = VSUB(T1Y, T1Z);
586 T21 = VADD(T1X, T20);
587 T2x = VSUB(T1X, T20);
588 }
589 {
590 V T1r, T2k, T1J, T2h, T1A, T2l, T1E, T2g;
591 {
592 V T1p, T1q, T1G, T1I;
593 T1p = LD(&(ri[WS(rs, 15)]), ms, &(ri[WS(rs, 1)]));
594 T1q = LD(&(ii[WS(rs, 15)]), ms, &(ii[WS(rs, 1)]));
595 T1r = VFMA(TN, T1p, VMUL(TO, T1q));
596 T2k = VFNMS(TO, T1p, VMUL(TN, T1q));
597 T1G = LD(&(ri[WS(rs, 11)]), ms, &(ri[WS(rs, 1)]));
598 T1I = LD(&(ii[WS(rs, 11)]), ms, &(ii[WS(rs, 1)]));
599 T1J = VFMA(T1F, T1G, VMUL(T1H, T1I));
600 T2h = VFNMS(T1H, T1G, VMUL(T1F, T1I));
601 }
602 {
603 V T1v, T1z, T1C, T1D;
604 T1v = LD(&(ri[WS(rs, 7)]), ms, &(ri[WS(rs, 1)]));
605 T1z = LD(&(ii[WS(rs, 7)]), ms, &(ii[WS(rs, 1)]));
606 T1A = VFMA(T1u, T1v, VMUL(T1y, T1z));
607 T2l = VFNMS(T1y, T1v, VMUL(T1u, T1z));
608 T1C = LD(&(ri[WS(rs, 3)]), ms, &(ri[WS(rs, 1)]));
609 T1D = LD(&(ii[WS(rs, 3)]), ms, &(ii[WS(rs, 1)]));
610 T1E = VFMA(Tg, T1C, VMUL(Ti, T1D));
611 T2g = VFNMS(Ti, T1C, VMUL(Tg, T1D));
612 }
613 T1B = VADD(T1r, T1A);
614 T1K = VADD(T1E, T1J);
615 T2V = VSUB(T1B, T1K);
616 T2W = VADD(T2k, T2l);
617 T2X = VADD(T2g, T2h);
618 T2Y = VSUB(T2W, T2X);
619 {
620 V T2f, T2i, T2m, T2n;
621 T2f = VSUB(T1r, T1A);
622 T2i = VSUB(T2g, T2h);
623 T2j = VSUB(T2f, T2i);
624 T2D = VADD(T2f, T2i);
625 T2m = VSUB(T2k, T2l);
626 T2n = VSUB(T1E, T1J);
627 T2o = VADD(T2m, T2n);
628 T2E = VSUB(T2m, T2n);
629 }
630 }
631 {
632 V T14, T24, T1m, T2b, T17, T25, T1h, T2a;
633 {
634 V T12, T13, T1j, T1l;
635 T12 = LD(&(ri[WS(rs, 1)]), ms, &(ri[WS(rs, 1)]));
636 T13 = LD(&(ii[WS(rs, 1)]), ms, &(ii[WS(rs, 1)]));
637 T14 = VFMA(T2, T12, VMUL(T5, T13));
638 T24 = VFNMS(T5, T12, VMUL(T2, T13));
639 T1j = LD(&(ri[WS(rs, 13)]), ms, &(ri[WS(rs, 1)]));
640 T1l = LD(&(ii[WS(rs, 13)]), ms, &(ii[WS(rs, 1)]));
641 T1m = VFMA(T1i, T1j, VMUL(T1k, T1l));
642 T2b = VFNMS(T1k, T1j, VMUL(T1i, T1l));
643 }
644 {
645 V T15, T16, T1c, T1g;
646 T15 = LD(&(ri[WS(rs, 9)]), ms, &(ri[WS(rs, 1)]));
647 T16 = LD(&(ii[WS(rs, 9)]), ms, &(ii[WS(rs, 1)]));
648 T17 = VFMA(T3, T15, VMUL(T6, T16));
649 T25 = VFNMS(T6, T15, VMUL(T3, T16));
650 T1c = LD(&(ri[WS(rs, 5)]), ms, &(ri[WS(rs, 1)]));
651 T1g = LD(&(ii[WS(rs, 5)]), ms, &(ii[WS(rs, 1)]));
652 T1h = VFMA(T1b, T1c, VMUL(T1f, T1g));
653 T2a = VFNMS(T1f, T1c, VMUL(T1b, T1g));
654 }
655 T18 = VADD(T14, T17);
656 T1n = VADD(T1h, T1m);
657 T2Q = VSUB(T18, T1n);
658 T2R = VADD(T24, T25);
659 T2S = VADD(T2a, T2b);
660 T2T = VSUB(T2R, T2S);
661 {
662 V T26, T27, T29, T2c;
663 T26 = VSUB(T24, T25);
664 T27 = VSUB(T1h, T1m);
665 T28 = VADD(T26, T27);
666 T2A = VSUB(T26, T27);
667 T29 = VSUB(T14, T17);
668 T2c = VSUB(T2a, T2b);
669 T2d = VSUB(T29, T2c);
670 T2B = VADD(T29, T2c);
671 }
672 }
673 {
674 V T23, T2r, T3A, T3C, T2q, T3B, T2u, T3x;
675 {
676 V T1R, T22, T3y, T3z;
677 T1R = VSUB(T1N, T1Q);
678 T22 = VMUL(LDK(KP707106781), VSUB(T1W, T21));
679 T23 = VADD(T1R, T22);
680 T2r = VSUB(T1R, T22);
681 T3y = VMUL(LDK(KP707106781), VSUB(T2x, T2w));
682 T3z = VADD(T3s, T3r);
683 T3A = VADD(T3y, T3z);
684 T3C = VSUB(T3z, T3y);
685 }
686 {
687 V T2e, T2p, T2s, T2t;
688 T2e = VFMA(LDK(KP923879532), T28, VMUL(LDK(KP382683432), T2d));
689 T2p = VFNMS(LDK(KP923879532), T2o, VMUL(LDK(KP382683432), T2j));
690 T2q = VADD(T2e, T2p);
691 T3B = VSUB(T2p, T2e);
692 T2s = VFNMS(LDK(KP923879532), T2d, VMUL(LDK(KP382683432), T28));
693 T2t = VFMA(LDK(KP382683432), T2o, VMUL(LDK(KP923879532), T2j));
694 T2u = VSUB(T2s, T2t);
695 T3x = VADD(T2s, T2t);
696 }
697 ST(&(ri[WS(rs, 11)]), VSUB(T23, T2q), ms, &(ri[WS(rs, 1)]));
698 ST(&(ii[WS(rs, 11)]), VSUB(T3A, T3x), ms, &(ii[WS(rs, 1)]));
699 ST(&(ri[WS(rs, 3)]), VADD(T23, T2q), ms, &(ri[WS(rs, 1)]));
700 ST(&(ii[WS(rs, 3)]), VADD(T3x, T3A), ms, &(ii[WS(rs, 1)]));
701 ST(&(ri[WS(rs, 15)]), VSUB(T2r, T2u), ms, &(ri[WS(rs, 1)]));
702 ST(&(ii[WS(rs, 15)]), VSUB(T3C, T3B), ms, &(ii[WS(rs, 1)]));
703 ST(&(ri[WS(rs, 7)]), VADD(T2r, T2u), ms, &(ri[WS(rs, 1)]));
704 ST(&(ii[WS(rs, 7)]), VADD(T3B, T3C), ms, &(ii[WS(rs, 1)]));
705 }
706 {
707 V T2P, T31, T3m, T3o, T30, T3n, T34, T3j;
708 {
709 V T2L, T2O, T3k, T3l;
710 T2L = VSUB(Tf, TA);
711 T2O = VSUB(T2M, T2N);
712 T2P = VADD(T2L, T2O);
713 T31 = VSUB(T2L, T2O);
714 T3k = VSUB(TZ, TM);
715 T3l = VSUB(T3e, T3b);
716 T3m = VADD(T3k, T3l);
717 T3o = VSUB(T3l, T3k);
718 }
719 {
720 V T2U, T2Z, T32, T33;
721 T2U = VADD(T2Q, T2T);
722 T2Z = VSUB(T2V, T2Y);
723 T30 = VMUL(LDK(KP707106781), VADD(T2U, T2Z));
724 T3n = VMUL(LDK(KP707106781), VSUB(T2Z, T2U));
725 T32 = VSUB(T2T, T2Q);
726 T33 = VADD(T2V, T2Y);
727 T34 = VMUL(LDK(KP707106781), VSUB(T32, T33));
728 T3j = VMUL(LDK(KP707106781), VADD(T32, T33));
729 }
730 ST(&(ri[WS(rs, 10)]), VSUB(T2P, T30), ms, &(ri[0]));
731 ST(&(ii[WS(rs, 10)]), VSUB(T3m, T3j), ms, &(ii[0]));
732 ST(&(ri[WS(rs, 2)]), VADD(T2P, T30), ms, &(ri[0]));
733 ST(&(ii[WS(rs, 2)]), VADD(T3j, T3m), ms, &(ii[0]));
734 ST(&(ri[WS(rs, 14)]), VSUB(T31, T34), ms, &(ri[0]));
735 ST(&(ii[WS(rs, 14)]), VSUB(T3o, T3n), ms, &(ii[0]));
736 ST(&(ri[WS(rs, 6)]), VADD(T31, T34), ms, &(ri[0]));
737 ST(&(ii[WS(rs, 6)]), VADD(T3n, T3o), ms, &(ii[0]));
738 }
739 {
740 V T2z, T2H, T3u, T3w, T2G, T3v, T2K, T3p;
741 {
742 V T2v, T2y, T3q, T3t;
743 T2v = VADD(T1N, T1Q);
744 T2y = VMUL(LDK(KP707106781), VADD(T2w, T2x));
745 T2z = VADD(T2v, T2y);
746 T2H = VSUB(T2v, T2y);
747 T3q = VMUL(LDK(KP707106781), VADD(T1W, T21));
748 T3t = VSUB(T3r, T3s);
749 T3u = VADD(T3q, T3t);
750 T3w = VSUB(T3t, T3q);
751 }
752 {
753 V T2C, T2F, T2I, T2J;
754 T2C = VFMA(LDK(KP382683432), T2A, VMUL(LDK(KP923879532), T2B));
755 T2F = VFNMS(LDK(KP382683432), T2E, VMUL(LDK(KP923879532), T2D));
756 T2G = VADD(T2C, T2F);
757 T3v = VSUB(T2F, T2C);
758 T2I = VFNMS(LDK(KP382683432), T2B, VMUL(LDK(KP923879532), T2A));
759 T2J = VFMA(LDK(KP923879532), T2E, VMUL(LDK(KP382683432), T2D));
760 T2K = VSUB(T2I, T2J);
761 T3p = VADD(T2I, T2J);
762 }
763 ST(&(ri[WS(rs, 9)]), VSUB(T2z, T2G), ms, &(ri[WS(rs, 1)]));
764 ST(&(ii[WS(rs, 9)]), VSUB(T3u, T3p), ms, &(ii[WS(rs, 1)]));
765 ST(&(ri[WS(rs, 1)]), VADD(T2z, T2G), ms, &(ri[WS(rs, 1)]));
766 ST(&(ii[WS(rs, 1)]), VADD(T3p, T3u), ms, &(ii[WS(rs, 1)]));
767 ST(&(ri[WS(rs, 13)]), VSUB(T2H, T2K), ms, &(ri[WS(rs, 1)]));
768 ST(&(ii[WS(rs, 13)]), VSUB(T3w, T3v), ms, &(ii[WS(rs, 1)]));
769 ST(&(ri[WS(rs, 5)]), VADD(T2H, T2K), ms, &(ri[WS(rs, 1)]));
770 ST(&(ii[WS(rs, 5)]), VADD(T3v, T3w), ms, &(ii[WS(rs, 1)]));
771 }
772 {
773 V T11, T35, T3g, T3i, T1M, T3h, T38, T39;
774 {
775 V TB, T10, T3a, T3f;
776 TB = VADD(Tf, TA);
777 T10 = VADD(TM, TZ);
778 T11 = VADD(TB, T10);
779 T35 = VSUB(TB, T10);
780 T3a = VADD(T2M, T2N);
781 T3f = VADD(T3b, T3e);
782 T3g = VADD(T3a, T3f);
783 T3i = VSUB(T3f, T3a);
784 }
785 {
786 V T1o, T1L, T36, T37;
787 T1o = VADD(T18, T1n);
788 T1L = VADD(T1B, T1K);
789 T1M = VADD(T1o, T1L);
790 T3h = VSUB(T1L, T1o);
791 T36 = VADD(T2R, T2S);
792 T37 = VADD(T2W, T2X);
793 T38 = VSUB(T36, T37);
794 T39 = VADD(T36, T37);
795 }
796 ST(&(ri[WS(rs, 8)]), VSUB(T11, T1M), ms, &(ri[0]));
797 ST(&(ii[WS(rs, 8)]), VSUB(T3g, T39), ms, &(ii[0]));
798 ST(&(ri[0]), VADD(T11, T1M), ms, &(ri[0]));
799 ST(&(ii[0]), VADD(T39, T3g), ms, &(ii[0]));
800 ST(&(ri[WS(rs, 12)]), VSUB(T35, T38), ms, &(ri[0]));
801 ST(&(ii[WS(rs, 12)]), VSUB(T3i, T3h), ms, &(ii[0]));
802 ST(&(ri[WS(rs, 4)]), VADD(T35, T38), ms, &(ri[0]));
803 ST(&(ii[WS(rs, 4)]), VADD(T3h, T3i), ms, &(ii[0]));
804 }
805 }
806 }
807 }
808 VLEAVE();
809 }
810
811 static const tw_instr twinstr[] = {
812 VTW(0, 1),
813 VTW(0, 3),
814 VTW(0, 9),
815 VTW(0, 15),
816 {TW_NEXT, (2 * VL), 0}
817 };
818
819 static const ct_desc desc = { 16, XSIMD_STRING("t2sv_16"), twinstr, &GENUS, {156, 68, 40, 0}, 0, 0, 0 };
820
821 void XSIMD(codelet_t2sv_16) (planner *p) {
822 X(kdft_dit_register) (p, t2sv_16, &desc);
823 }
824 #endif /* HAVE_FMA */