Mercurial > hg > sv-dependency-builds

comparison src/fftw-3.3.8/dft/scalar/codelets/q1_4.c @ 82:d0c2a83c1364

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Add FFTW 3.3.8 source, and a Linux build
author Chris Cannam
date Tue, 19 Nov 2019 14:52:55 +0000
parents
children
comparison
equal deleted inserted replaced
81:7029a4916348 82:d0c2a83c1364
1 /*
2 * Copyright (c) 2003, 2007-14 Matteo Frigo
3 * Copyright (c) 2003, 2007-14 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 Thu May 24 08:04:29 EDT 2018 */
23
24 #include "dft/codelet-dft.h"
25
26 #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA)
27
28 /* Generated by: ../../../genfft/gen_twidsq.native -fma -compact -variables 4 -pipeline-latency 4 -reload-twiddle -dif -n 4 -name q1_4 -include dft/scalar/q.h */
29
30 /*
31 * This function contains 88 FP additions, 48 FP multiplications,
32 * (or, 64 additions, 24 multiplications, 24 fused multiply/add),
33 * 51 stack variables, 0 constants, and 64 memory accesses
34 */
35 #include "dft/scalar/q.h"
36
37 static void q1_4(R *rio, R *iio, const R *W, stride rs, stride vs, INT mb, INT me, INT ms)
38 {
39 {
40 INT m;
41 for (m = mb, W = W + (mb * 6); m < me; m = m + 1, rio = rio + ms, iio = iio + ms, W = W + 6, MAKE_VOLATILE_STRIDE(8, rs), MAKE_VOLATILE_STRIDE(0, vs)) {
42 E T3, Tv, Tw, T6, Tc, Tf, Tx, Ts, Tm, Ti, T1H, T29, T2a, T1K, T1Q;
43 E T1T, T2b, T26, T20, T1W, TB, T13, T14, TE, TK, TN, T15, T10, TU, TQ;
44 E T19, T1B, T1C, T1c, T1i, T1l, T1D, T1y, T1s, T1o;
45 {
46 E T1, T2, Tb, Tg, Th, T8;
47 {
48 E T9, Ta, T4, T5;
49 T1 = rio[0];
50 T2 = rio[WS(rs, 2)];
51 T3 = T1 + T2;
52 T9 = iio[0];
53 Ta = iio[WS(rs, 2)];
54 Tb = T9 - Ta;
55 Tv = T9 + Ta;
56 Tg = iio[WS(rs, 1)];
57 Th = iio[WS(rs, 3)];
58 Tw = Tg + Th;
59 T4 = rio[WS(rs, 1)];
60 T5 = rio[WS(rs, 3)];
61 T6 = T4 + T5;
62 T8 = T4 - T5;
63 }
64 Tc = T8 + Tb;
65 Tf = T1 - T2;
66 Tx = Tv - Tw;
67 Ts = T3 - T6;
68 Tm = Tb - T8;
69 Ti = Tg - Th;
70 }
71 {
72 E T1F, T1G, T1P, T1U, T1V, T1M;
73 {
74 E T1N, T1O, T1I, T1J;
75 T1F = rio[WS(vs, 3)];
76 T1G = rio[WS(vs, 3) + WS(rs, 2)];
77 T1H = T1F + T1G;
78 T1N = iio[WS(vs, 3)];
79 T1O = iio[WS(vs, 3) + WS(rs, 2)];
80 T1P = T1N - T1O;
81 T29 = T1N + T1O;
82 T1U = iio[WS(vs, 3) + WS(rs, 1)];
83 T1V = iio[WS(vs, 3) + WS(rs, 3)];
84 T2a = T1U + T1V;
85 T1I = rio[WS(vs, 3) + WS(rs, 1)];
86 T1J = rio[WS(vs, 3) + WS(rs, 3)];
87 T1K = T1I + T1J;
88 T1M = T1I - T1J;
89 }
90 T1Q = T1M + T1P;
91 T1T = T1F - T1G;
92 T2b = T29 - T2a;
93 T26 = T1H - T1K;
94 T20 = T1P - T1M;
95 T1W = T1U - T1V;
96 }
97 {
98 E Tz, TA, TJ, TO, TP, TG;
99 {
100 E TH, TI, TC, TD;
101 Tz = rio[WS(vs, 1)];
102 TA = rio[WS(vs, 1) + WS(rs, 2)];
103 TB = Tz + TA;
104 TH = iio[WS(vs, 1)];
105 TI = iio[WS(vs, 1) + WS(rs, 2)];
106 TJ = TH - TI;
107 T13 = TH + TI;
108 TO = iio[WS(vs, 1) + WS(rs, 1)];
109 TP = iio[WS(vs, 1) + WS(rs, 3)];
110 T14 = TO + TP;
111 TC = rio[WS(vs, 1) + WS(rs, 1)];
112 TD = rio[WS(vs, 1) + WS(rs, 3)];
113 TE = TC + TD;
114 TG = TC - TD;
115 }
116 TK = TG + TJ;
117 TN = Tz - TA;
118 T15 = T13 - T14;
119 T10 = TB - TE;
120 TU = TJ - TG;
121 TQ = TO - TP;
122 }
123 {
124 E T17, T18, T1h, T1m, T1n, T1e;
125 {
126 E T1f, T1g, T1a, T1b;
127 T17 = rio[WS(vs, 2)];
128 T18 = rio[WS(vs, 2) + WS(rs, 2)];
129 T19 = T17 + T18;
130 T1f = iio[WS(vs, 2)];
131 T1g = iio[WS(vs, 2) + WS(rs, 2)];
132 T1h = T1f - T1g;
133 T1B = T1f + T1g;
134 T1m = iio[WS(vs, 2) + WS(rs, 1)];
135 T1n = iio[WS(vs, 2) + WS(rs, 3)];
136 T1C = T1m + T1n;
137 T1a = rio[WS(vs, 2) + WS(rs, 1)];
138 T1b = rio[WS(vs, 2) + WS(rs, 3)];
139 T1c = T1a + T1b;
140 T1e = T1a - T1b;
141 }
142 T1i = T1e + T1h;
143 T1l = T17 - T18;
144 T1D = T1B - T1C;
145 T1y = T19 - T1c;
146 T1s = T1h - T1e;
147 T1o = T1m - T1n;
148 }
149 rio[0] = T3 + T6;
150 iio[0] = Tv + Tw;
151 rio[WS(rs, 1)] = TB + TE;
152 iio[WS(rs, 1)] = T13 + T14;
153 rio[WS(rs, 2)] = T19 + T1c;
154 iio[WS(rs, 2)] = T1B + T1C;
155 iio[WS(rs, 3)] = T29 + T2a;
156 rio[WS(rs, 3)] = T1H + T1K;
157 {
158 E Tt, Ty, Tr, Tu;
159 Tr = W[2];
160 Tt = Tr * Ts;
161 Ty = Tr * Tx;
162 Tu = W[3];
163 rio[WS(vs, 2)] = FMA(Tu, Tx, Tt);
164 iio[WS(vs, 2)] = FNMS(Tu, Ts, Ty);
165 }
166 {
167 E T27, T2c, T25, T28;
168 T25 = W[2];
169 T27 = T25 * T26;
170 T2c = T25 * T2b;
171 T28 = W[3];
172 rio[WS(vs, 2) + WS(rs, 3)] = FMA(T28, T2b, T27);
173 iio[WS(vs, 2) + WS(rs, 3)] = FNMS(T28, T26, T2c);
174 }
175 {
176 E T11, T16, TZ, T12;
177 TZ = W[2];
178 T11 = TZ * T10;
179 T16 = TZ * T15;
180 T12 = W[3];
181 rio[WS(vs, 2) + WS(rs, 1)] = FMA(T12, T15, T11);
182 iio[WS(vs, 2) + WS(rs, 1)] = FNMS(T12, T10, T16);
183 }
184 {
185 E T1z, T1E, T1x, T1A;
186 T1x = W[2];
187 T1z = T1x * T1y;
188 T1E = T1x * T1D;
189 T1A = W[3];
190 rio[WS(vs, 2) + WS(rs, 2)] = FMA(T1A, T1D, T1z);
191 iio[WS(vs, 2) + WS(rs, 2)] = FNMS(T1A, T1y, T1E);
192 }
193 {
194 E Tj, Te, Tk, T7, Td;
195 Tj = Tf - Ti;
196 Te = W[5];
197 Tk = Te * Tc;
198 T7 = W[4];
199 Td = T7 * Tc;
200 iio[WS(vs, 3)] = FNMS(Te, Tj, Td);
201 rio[WS(vs, 3)] = FMA(T7, Tj, Tk);
202 }
203 {
204 E T1p, T1k, T1q, T1d, T1j;
205 T1p = T1l - T1o;
206 T1k = W[5];
207 T1q = T1k * T1i;
208 T1d = W[4];
209 T1j = T1d * T1i;
210 iio[WS(vs, 3) + WS(rs, 2)] = FNMS(T1k, T1p, T1j);
211 rio[WS(vs, 3) + WS(rs, 2)] = FMA(T1d, T1p, T1q);
212 }
213 {
214 E T23, T22, T24, T1Z, T21;
215 T23 = T1T + T1W;
216 T22 = W[1];
217 T24 = T22 * T20;
218 T1Z = W[0];
219 T21 = T1Z * T20;
220 iio[WS(vs, 1) + WS(rs, 3)] = FNMS(T22, T23, T21);
221 rio[WS(vs, 1) + WS(rs, 3)] = FMA(T1Z, T23, T24);
222 }
223 {
224 E TX, TW, TY, TT, TV;
225 TX = TN + TQ;
226 TW = W[1];
227 TY = TW * TU;
228 TT = W[0];
229 TV = TT * TU;
230 iio[WS(vs, 1) + WS(rs, 1)] = FNMS(TW, TX, TV);
231 rio[WS(vs, 1) + WS(rs, 1)] = FMA(TT, TX, TY);
232 }
233 {
234 E TR, TM, TS, TF, TL;
235 TR = TN - TQ;
236 TM = W[5];
237 TS = TM * TK;
238 TF = W[4];
239 TL = TF * TK;
240 iio[WS(vs, 3) + WS(rs, 1)] = FNMS(TM, TR, TL);
241 rio[WS(vs, 3) + WS(rs, 1)] = FMA(TF, TR, TS);
242 }
243 {
244 E Tp, To, Tq, Tl, Tn;
245 Tp = Tf + Ti;
246 To = W[1];
247 Tq = To * Tm;
248 Tl = W[0];
249 Tn = Tl * Tm;
250 iio[WS(vs, 1)] = FNMS(To, Tp, Tn);
251 rio[WS(vs, 1)] = FMA(Tl, Tp, Tq);
252 }
253 {
254 E T1v, T1u, T1w, T1r, T1t;
255 T1v = T1l + T1o;
256 T1u = W[1];
257 T1w = T1u * T1s;
258 T1r = W[0];
259 T1t = T1r * T1s;
260 iio[WS(vs, 1) + WS(rs, 2)] = FNMS(T1u, T1v, T1t);
261 rio[WS(vs, 1) + WS(rs, 2)] = FMA(T1r, T1v, T1w);
262 }
263 {
264 E T1X, T1S, T1Y, T1L, T1R;
265 T1X = T1T - T1W;
266 T1S = W[5];
267 T1Y = T1S * T1Q;
268 T1L = W[4];
269 T1R = T1L * T1Q;
270 iio[WS(vs, 3) + WS(rs, 3)] = FNMS(T1S, T1X, T1R);
271 rio[WS(vs, 3) + WS(rs, 3)] = FMA(T1L, T1X, T1Y);
272 }
273 }
274 }
275 }
276
277 static const tw_instr twinstr[] = {
278 {TW_FULL, 0, 4},
279 {TW_NEXT, 1, 0}
280 };
281
282 static const ct_desc desc = { 4, "q1_4", twinstr, &GENUS, {64, 24, 24, 0}, 0, 0, 0 };
283
284 void X(codelet_q1_4) (planner *p) {
285 X(kdft_difsq_register) (p, q1_4, &desc);
286 }
287 #else
288
289 /* Generated by: ../../../genfft/gen_twidsq.native -compact -variables 4 -pipeline-latency 4 -reload-twiddle -dif -n 4 -name q1_4 -include dft/scalar/q.h */
290
291 /*
292 * This function contains 88 FP additions, 48 FP multiplications,
293 * (or, 64 additions, 24 multiplications, 24 fused multiply/add),
294 * 37 stack variables, 0 constants, and 64 memory accesses
295 */
296 #include "dft/scalar/q.h"
297
298 static void q1_4(R *rio, R *iio, const R *W, stride rs, stride vs, INT mb, INT me, INT ms)
299 {
300 {
301 INT m;
302 for (m = mb, W = W + (mb * 6); m < me; m = m + 1, rio = rio + ms, iio = iio + ms, W = W + 6, MAKE_VOLATILE_STRIDE(8, rs), MAKE_VOLATILE_STRIDE(0, vs)) {
303 E T3, Te, Tb, Tq, T6, T8, Th, Tr, Tv, TG, TD, TS, Ty, TA, TJ;
304 E TT, TX, T18, T15, T1k, T10, T12, T1b, T1l, T1p, T1A, T1x, T1M, T1s, T1u;
305 E T1D, T1N;
306 {
307 E T1, T2, T9, Ta;
308 T1 = rio[0];
309 T2 = rio[WS(rs, 2)];
310 T3 = T1 + T2;
311 Te = T1 - T2;
312 T9 = iio[0];
313 Ta = iio[WS(rs, 2)];
314 Tb = T9 - Ta;
315 Tq = T9 + Ta;
316 }
317 {
318 E T4, T5, Tf, Tg;
319 T4 = rio[WS(rs, 1)];
320 T5 = rio[WS(rs, 3)];
321 T6 = T4 + T5;
322 T8 = T4 - T5;
323 Tf = iio[WS(rs, 1)];
324 Tg = iio[WS(rs, 3)];
325 Th = Tf - Tg;
326 Tr = Tf + Tg;
327 }
328 {
329 E Tt, Tu, TB, TC;
330 Tt = rio[WS(vs, 1)];
331 Tu = rio[WS(vs, 1) + WS(rs, 2)];
332 Tv = Tt + Tu;
333 TG = Tt - Tu;
334 TB = iio[WS(vs, 1)];
335 TC = iio[WS(vs, 1) + WS(rs, 2)];
336 TD = TB - TC;
337 TS = TB + TC;
338 }
339 {
340 E Tw, Tx, TH, TI;
341 Tw = rio[WS(vs, 1) + WS(rs, 1)];
342 Tx = rio[WS(vs, 1) + WS(rs, 3)];
343 Ty = Tw + Tx;
344 TA = Tw - Tx;
345 TH = iio[WS(vs, 1) + WS(rs, 1)];
346 TI = iio[WS(vs, 1) + WS(rs, 3)];
347 TJ = TH - TI;
348 TT = TH + TI;
349 }
350 {
351 E TV, TW, T13, T14;
352 TV = rio[WS(vs, 2)];
353 TW = rio[WS(vs, 2) + WS(rs, 2)];
354 TX = TV + TW;
355 T18 = TV - TW;
356 T13 = iio[WS(vs, 2)];
357 T14 = iio[WS(vs, 2) + WS(rs, 2)];
358 T15 = T13 - T14;
359 T1k = T13 + T14;
360 }
361 {
362 E TY, TZ, T19, T1a;
363 TY = rio[WS(vs, 2) + WS(rs, 1)];
364 TZ = rio[WS(vs, 2) + WS(rs, 3)];
365 T10 = TY + TZ;
366 T12 = TY - TZ;
367 T19 = iio[WS(vs, 2) + WS(rs, 1)];
368 T1a = iio[WS(vs, 2) + WS(rs, 3)];
369 T1b = T19 - T1a;
370 T1l = T19 + T1a;
371 }
372 {
373 E T1n, T1o, T1v, T1w;
374 T1n = rio[WS(vs, 3)];
375 T1o = rio[WS(vs, 3) + WS(rs, 2)];
376 T1p = T1n + T1o;
377 T1A = T1n - T1o;
378 T1v = iio[WS(vs, 3)];
379 T1w = iio[WS(vs, 3) + WS(rs, 2)];
380 T1x = T1v - T1w;
381 T1M = T1v + T1w;
382 }
383 {
384 E T1q, T1r, T1B, T1C;
385 T1q = rio[WS(vs, 3) + WS(rs, 1)];
386 T1r = rio[WS(vs, 3) + WS(rs, 3)];
387 T1s = T1q + T1r;
388 T1u = T1q - T1r;
389 T1B = iio[WS(vs, 3) + WS(rs, 1)];
390 T1C = iio[WS(vs, 3) + WS(rs, 3)];
391 T1D = T1B - T1C;
392 T1N = T1B + T1C;
393 }
394 rio[0] = T3 + T6;
395 iio[0] = Tq + Tr;
396 rio[WS(rs, 1)] = Tv + Ty;
397 iio[WS(rs, 1)] = TS + TT;
398 rio[WS(rs, 2)] = TX + T10;
399 iio[WS(rs, 2)] = T1k + T1l;
400 iio[WS(rs, 3)] = T1M + T1N;
401 rio[WS(rs, 3)] = T1p + T1s;
402 {
403 E Tc, Ti, T7, Td;
404 Tc = T8 + Tb;
405 Ti = Te - Th;
406 T7 = W[4];
407 Td = W[5];
408 iio[WS(vs, 3)] = FNMS(Td, Ti, T7 * Tc);
409 rio[WS(vs, 3)] = FMA(Td, Tc, T7 * Ti);
410 }
411 {
412 E T1K, T1O, T1J, T1L;
413 T1K = T1p - T1s;
414 T1O = T1M - T1N;
415 T1J = W[2];
416 T1L = W[3];
417 rio[WS(vs, 2) + WS(rs, 3)] = FMA(T1J, T1K, T1L * T1O);
418 iio[WS(vs, 2) + WS(rs, 3)] = FNMS(T1L, T1K, T1J * T1O);
419 }
420 {
421 E Tk, Tm, Tj, Tl;
422 Tk = Tb - T8;
423 Tm = Te + Th;
424 Tj = W[0];
425 Tl = W[1];
426 iio[WS(vs, 1)] = FNMS(Tl, Tm, Tj * Tk);
427 rio[WS(vs, 1)] = FMA(Tl, Tk, Tj * Tm);
428 }
429 {
430 E To, Ts, Tn, Tp;
431 To = T3 - T6;
432 Ts = Tq - Tr;
433 Tn = W[2];
434 Tp = W[3];
435 rio[WS(vs, 2)] = FMA(Tn, To, Tp * Ts);
436 iio[WS(vs, 2)] = FNMS(Tp, To, Tn * Ts);
437 }
438 {
439 E T16, T1c, T11, T17;
440 T16 = T12 + T15;
441 T1c = T18 - T1b;
442 T11 = W[4];
443 T17 = W[5];
444 iio[WS(vs, 3) + WS(rs, 2)] = FNMS(T17, T1c, T11 * T16);
445 rio[WS(vs, 3) + WS(rs, 2)] = FMA(T17, T16, T11 * T1c);
446 }
447 {
448 E T1G, T1I, T1F, T1H;
449 T1G = T1x - T1u;
450 T1I = T1A + T1D;
451 T1F = W[0];
452 T1H = W[1];
453 iio[WS(vs, 1) + WS(rs, 3)] = FNMS(T1H, T1I, T1F * T1G);
454 rio[WS(vs, 1) + WS(rs, 3)] = FMA(T1H, T1G, T1F * T1I);
455 }
456 {
457 E TQ, TU, TP, TR;
458 TQ = Tv - Ty;
459 TU = TS - TT;
460 TP = W[2];
461 TR = W[3];
462 rio[WS(vs, 2) + WS(rs, 1)] = FMA(TP, TQ, TR * TU);
463 iio[WS(vs, 2) + WS(rs, 1)] = FNMS(TR, TQ, TP * TU);
464 }
465 {
466 E T1e, T1g, T1d, T1f;
467 T1e = T15 - T12;
468 T1g = T18 + T1b;
469 T1d = W[0];
470 T1f = W[1];
471 iio[WS(vs, 1) + WS(rs, 2)] = FNMS(T1f, T1g, T1d * T1e);
472 rio[WS(vs, 1) + WS(rs, 2)] = FMA(T1f, T1e, T1d * T1g);
473 }
474 {
475 E T1i, T1m, T1h, T1j;
476 T1i = TX - T10;
477 T1m = T1k - T1l;
478 T1h = W[2];
479 T1j = W[3];
480 rio[WS(vs, 2) + WS(rs, 2)] = FMA(T1h, T1i, T1j * T1m);
481 iio[WS(vs, 2) + WS(rs, 2)] = FNMS(T1j, T1i, T1h * T1m);
482 }
483 {
484 E T1y, T1E, T1t, T1z;
485 T1y = T1u + T1x;
486 T1E = T1A - T1D;
487 T1t = W[4];
488 T1z = W[5];
489 iio[WS(vs, 3) + WS(rs, 3)] = FNMS(T1z, T1E, T1t * T1y);
490 rio[WS(vs, 3) + WS(rs, 3)] = FMA(T1z, T1y, T1t * T1E);
491 }
492 {
493 E TM, TO, TL, TN;
494 TM = TD - TA;
495 TO = TG + TJ;
496 TL = W[0];
497 TN = W[1];
498 iio[WS(vs, 1) + WS(rs, 1)] = FNMS(TN, TO, TL * TM);
499 rio[WS(vs, 1) + WS(rs, 1)] = FMA(TN, TM, TL * TO);
500 }
501 {
502 E TE, TK, Tz, TF;
503 TE = TA + TD;
504 TK = TG - TJ;
505 Tz = W[4];
506 TF = W[5];
507 iio[WS(vs, 3) + WS(rs, 1)] = FNMS(TF, TK, Tz * TE);
508 rio[WS(vs, 3) + WS(rs, 1)] = FMA(TF, TE, Tz * TK);
509 }
510 }
511 }
512 }
513
514 static const tw_instr twinstr[] = {
515 {TW_FULL, 0, 4},
516 {TW_NEXT, 1, 0}
517 };
518
519 static const ct_desc desc = { 4, "q1_4", twinstr, &GENUS, {64, 24, 24, 0}, 0, 0, 0 };
520
521 void X(codelet_q1_4) (planner *p) {
522 X(kdft_difsq_register) (p, q1_4, &desc);
523 }
524 #endif