Mercurial > hg > sv-dependency-builds

comparison src/fftw-3.3.3/dft/scalar/codelets/q1_4.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 07:36:17 EST 2012 */
23
24 #include "codelet-dft.h"
25
26 #ifdef HAVE_FMA
27
28 /* Generated by: ../../../genfft/gen_twidsq.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -reload-twiddle -dif -n 4 -name q1_4 -include 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 * 76 stack variables, 0 constants, and 64 memory accesses
34 */
35 #include "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 T1X, T1S, T1L, T1Y, T1R;
43 {
44 E T3, Tf, Tv, Ti, Tw, Tx, T6, Tm, Tc, Ts, T1T, T1H, T29, T1W, T2a;
45 E T2b, T1K, T20, T1Q, T26, TN, TB, T13, TQ, T14, T15, TE, TU, TK, T10;
46 E T1l, T19, T1a, T1h, T1B, T1o, T1C, T1b, T1D, T1e, T1c;
47 {
48 E T1I, T1P, T1J, T1M;
49 {
50 E Tb, T4, T5, T8;
51 {
52 E T1, T2, T9, Ta, Tg, Th;
53 T1 = rio[0];
54 T2 = rio[WS(rs, 2)];
55 T9 = iio[0];
56 Ta = iio[WS(rs, 2)];
57 Tg = iio[WS(rs, 1)];
58 T3 = T1 + T2;
59 Tf = T1 - T2;
60 Th = iio[WS(rs, 3)];
61 Tv = T9 + Ta;
62 Tb = T9 - Ta;
63 T4 = rio[WS(rs, 1)];
64 Ti = Tg - Th;
65 Tw = Tg + Th;
66 T5 = rio[WS(rs, 3)];
67 }
68 Tx = Tv - Tw;
69 T8 = T4 - T5;
70 T6 = T4 + T5;
71 {
72 E T1N, T1O, T1F, T1G, T1U, T1V;
73 T1F = rio[WS(vs, 3)];
74 T1G = rio[WS(vs, 3) + WS(rs, 2)];
75 Tm = Tb - T8;
76 Tc = T8 + Tb;
77 Ts = T3 - T6;
78 T1T = T1F - T1G;
79 T1H = T1F + T1G;
80 T1N = iio[WS(vs, 3)];
81 T1O = iio[WS(vs, 3) + WS(rs, 2)];
82 T1U = iio[WS(vs, 3) + WS(rs, 1)];
83 T1V = iio[WS(vs, 3) + WS(rs, 3)];
84 T1I = rio[WS(vs, 3) + WS(rs, 1)];
85 T1P = T1N - T1O;
86 T29 = T1N + T1O;
87 T1W = T1U - T1V;
88 T2a = T1U + T1V;
89 T1J = rio[WS(vs, 3) + WS(rs, 3)];
90 }
91 }
92 T2b = T29 - T2a;
93 T1M = T1I - T1J;
94 T1K = T1I + T1J;
95 {
96 E TC, TJ, TD, TG;
97 {
98 E TH, TI, Tz, TA, TO, TP;
99 Tz = rio[WS(vs, 1)];
100 TA = rio[WS(vs, 1) + WS(rs, 2)];
101 T20 = T1P - T1M;
102 T1Q = T1M + T1P;
103 T26 = T1H - T1K;
104 TN = Tz - TA;
105 TB = Tz + TA;
106 TH = iio[WS(vs, 1)];
107 TI = iio[WS(vs, 1) + WS(rs, 2)];
108 TO = iio[WS(vs, 1) + WS(rs, 1)];
109 TP = iio[WS(vs, 1) + WS(rs, 3)];
110 TC = rio[WS(vs, 1) + WS(rs, 1)];
111 TJ = TH - TI;
112 T13 = TH + TI;
113 TQ = TO - TP;
114 T14 = TO + TP;
115 TD = rio[WS(vs, 1) + WS(rs, 3)];
116 }
117 T15 = T13 - T14;
118 TG = TC - TD;
119 TE = TC + TD;
120 {
121 E T1f, T1g, T17, T18, T1m, T1n;
122 T17 = rio[WS(vs, 2)];
123 T18 = rio[WS(vs, 2) + WS(rs, 2)];
124 TU = TJ - TG;
125 TK = TG + TJ;
126 T10 = TB - TE;
127 T1l = T17 - T18;
128 T19 = T17 + T18;
129 T1f = iio[WS(vs, 2)];
130 T1g = iio[WS(vs, 2) + WS(rs, 2)];
131 T1m = iio[WS(vs, 2) + WS(rs, 1)];
132 T1n = iio[WS(vs, 2) + WS(rs, 3)];
133 T1a = rio[WS(vs, 2) + WS(rs, 1)];
134 T1h = T1f - T1g;
135 T1B = T1f + T1g;
136 T1o = T1m - T1n;
137 T1C = T1m + T1n;
138 T1b = rio[WS(vs, 2) + WS(rs, 3)];
139 }
140 }
141 }
142 T1D = T1B - T1C;
143 T1e = T1a - T1b;
144 T1c = T1a + T1b;
145 {
146 E T1s, T1i, T1y, T28, T27, Tr, Tu;
147 rio[0] = T3 + T6;
148 iio[0] = Tv + Tw;
149 T1s = T1h - T1e;
150 T1i = T1e + T1h;
151 T1y = T19 - T1c;
152 rio[WS(rs, 1)] = TB + TE;
153 iio[WS(rs, 1)] = T13 + T14;
154 rio[WS(rs, 2)] = T19 + T1c;
155 iio[WS(rs, 2)] = T1B + T1C;
156 iio[WS(rs, 3)] = T29 + T2a;
157 rio[WS(rs, 3)] = T1H + T1K;
158 Tr = W[2];
159 Tu = W[3];
160 {
161 E T25, Ty, Tt, T2c;
162 T25 = W[2];
163 T28 = W[3];
164 Ty = Tr * Tx;
165 Tt = Tr * Ts;
166 T2c = T25 * T2b;
167 T27 = T25 * T26;
168 iio[WS(vs, 2)] = FNMS(Tu, Ts, Ty);
169 rio[WS(vs, 2)] = FMA(Tu, Tx, Tt);
170 iio[WS(vs, 2) + WS(rs, 3)] = FNMS(T28, T26, T2c);
171 }
172 rio[WS(vs, 2) + WS(rs, 3)] = FMA(T28, T2b, T27);
173 {
174 E Tp, T1v, T23, T22, T1Z, TR, TM, TF;
175 {
176 E T1A, T1z, TZ, T12;
177 TZ = W[2];
178 T12 = W[3];
179 {
180 E T1x, T16, T11, T1E;
181 T1x = W[2];
182 T1A = W[3];
183 T16 = TZ * T15;
184 T11 = TZ * T10;
185 T1E = T1x * T1D;
186 T1z = T1x * T1y;
187 iio[WS(vs, 2) + WS(rs, 1)] = FNMS(T12, T10, T16);
188 rio[WS(vs, 2) + WS(rs, 1)] = FMA(T12, T15, T11);
189 iio[WS(vs, 2) + WS(rs, 2)] = FNMS(T1A, T1y, T1E);
190 }
191 rio[WS(vs, 2) + WS(rs, 2)] = FMA(T1A, T1D, T1z);
192 {
193 E Tj, Te, T7, T1p, T1k, T1j;
194 Tp = Tf + Ti;
195 Tj = Tf - Ti;
196 Te = W[5];
197 T7 = W[4];
198 {
199 E T1d, T1q, Tk, Td;
200 T1p = T1l - T1o;
201 T1v = T1l + T1o;
202 T1k = W[5];
203 Tk = Te * Tc;
204 Td = T7 * Tc;
205 T1d = W[4];
206 T1q = T1k * T1i;
207 rio[WS(vs, 3)] = FMA(T7, Tj, Tk);
208 iio[WS(vs, 3)] = FNMS(Te, Tj, Td);
209 T1j = T1d * T1i;
210 rio[WS(vs, 3) + WS(rs, 2)] = FMA(T1d, T1p, T1q);
211 }
212 T23 = T1T + T1W;
213 T1X = T1T - T1W;
214 T22 = W[1];
215 iio[WS(vs, 3) + WS(rs, 2)] = FNMS(T1k, T1p, T1j);
216 T1Z = W[0];
217 }
218 }
219 {
220 E TX, TW, TT, TY, TV, T24, T21;
221 TX = TN + TQ;
222 TR = TN - TQ;
223 T24 = T22 * T20;
224 TW = W[1];
225 T21 = T1Z * T20;
226 TT = W[0];
227 rio[WS(vs, 1) + WS(rs, 3)] = FMA(T1Z, T23, T24);
228 TY = TW * TU;
229 iio[WS(vs, 1) + WS(rs, 3)] = FNMS(T22, T23, T21);
230 TV = TT * TU;
231 rio[WS(vs, 1) + WS(rs, 1)] = FMA(TT, TX, TY);
232 TM = W[5];
233 iio[WS(vs, 1) + WS(rs, 1)] = FNMS(TW, TX, TV);
234 TF = W[4];
235 }
236 {
237 E To, Tl, Tq, Tn, TS, TL;
238 TS = TM * TK;
239 To = W[1];
240 TL = TF * TK;
241 Tl = W[0];
242 rio[WS(vs, 3) + WS(rs, 1)] = FMA(TF, TR, TS);
243 Tq = To * Tm;
244 iio[WS(vs, 3) + WS(rs, 1)] = FNMS(TM, TR, TL);
245 Tn = Tl * Tm;
246 {
247 E T1u, T1r, T1w, T1t;
248 rio[WS(vs, 1)] = FMA(Tl, Tp, Tq);
249 T1u = W[1];
250 iio[WS(vs, 1)] = FNMS(To, Tp, Tn);
251 T1r = W[0];
252 T1w = T1u * T1s;
253 T1S = W[5];
254 T1t = T1r * T1s;
255 T1L = W[4];
256 rio[WS(vs, 1) + WS(rs, 2)] = FMA(T1r, T1v, T1w);
257 T1Y = T1S * T1Q;
258 iio[WS(vs, 1) + WS(rs, 2)] = FNMS(T1u, T1v, T1t);
259 T1R = T1L * T1Q;
260 }
261 }
262 }
263 }
264 }
265 rio[WS(vs, 3) + WS(rs, 3)] = FMA(T1L, T1X, T1Y);
266 iio[WS(vs, 3) + WS(rs, 3)] = FNMS(T1S, T1X, T1R);
267 }
268 }
269 }
270
271 static const tw_instr twinstr[] = {
272 {TW_FULL, 0, 4},
273 {TW_NEXT, 1, 0}
274 };
275
276 static const ct_desc desc = { 4, "q1_4", twinstr, &GENUS, {64, 24, 24, 0}, 0, 0, 0 };
277
278 void X(codelet_q1_4) (planner *p) {
279 X(kdft_difsq_register) (p, q1_4, &desc);
280 }
281 #else /* HAVE_FMA */
282
283 /* Generated by: ../../../genfft/gen_twidsq.native -compact -variables 4 -pipeline-latency 4 -reload-twiddle -dif -n 4 -name q1_4 -include q.h */
284
285 /*
286 * This function contains 88 FP additions, 48 FP multiplications,
287 * (or, 64 additions, 24 multiplications, 24 fused multiply/add),
288 * 37 stack variables, 0 constants, and 64 memory accesses
289 */
290 #include "q.h"
291
292 static void q1_4(R *rio, R *iio, const R *W, stride rs, stride vs, INT mb, INT me, INT ms)
293 {
294 {
295 INT m;
296 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)) {
297 E T3, Te, Tb, Tq, T6, T8, Th, Tr, Tv, TG, TD, TS, Ty, TA, TJ;
298 E TT, TX, T18, T15, T1k, T10, T12, T1b, T1l, T1p, T1A, T1x, T1M, T1s, T1u;
299 E T1D, T1N;
300 {
301 E T1, T2, T9, Ta;
302 T1 = rio[0];
303 T2 = rio[WS(rs, 2)];
304 T3 = T1 + T2;
305 Te = T1 - T2;
306 T9 = iio[0];
307 Ta = iio[WS(rs, 2)];
308 Tb = T9 - Ta;
309 Tq = T9 + Ta;
310 }
311 {
312 E T4, T5, Tf, Tg;
313 T4 = rio[WS(rs, 1)];
314 T5 = rio[WS(rs, 3)];
315 T6 = T4 + T5;
316 T8 = T4 - T5;
317 Tf = iio[WS(rs, 1)];
318 Tg = iio[WS(rs, 3)];
319 Th = Tf - Tg;
320 Tr = Tf + Tg;
321 }
322 {
323 E Tt, Tu, TB, TC;
324 Tt = rio[WS(vs, 1)];
325 Tu = rio[WS(vs, 1) + WS(rs, 2)];
326 Tv = Tt + Tu;
327 TG = Tt - Tu;
328 TB = iio[WS(vs, 1)];
329 TC = iio[WS(vs, 1) + WS(rs, 2)];
330 TD = TB - TC;
331 TS = TB + TC;
332 }
333 {
334 E Tw, Tx, TH, TI;
335 Tw = rio[WS(vs, 1) + WS(rs, 1)];
336 Tx = rio[WS(vs, 1) + WS(rs, 3)];
337 Ty = Tw + Tx;
338 TA = Tw - Tx;
339 TH = iio[WS(vs, 1) + WS(rs, 1)];
340 TI = iio[WS(vs, 1) + WS(rs, 3)];
341 TJ = TH - TI;
342 TT = TH + TI;
343 }
344 {
345 E TV, TW, T13, T14;
346 TV = rio[WS(vs, 2)];
347 TW = rio[WS(vs, 2) + WS(rs, 2)];
348 TX = TV + TW;
349 T18 = TV - TW;
350 T13 = iio[WS(vs, 2)];
351 T14 = iio[WS(vs, 2) + WS(rs, 2)];
352 T15 = T13 - T14;
353 T1k = T13 + T14;
354 }
355 {
356 E TY, TZ, T19, T1a;
357 TY = rio[WS(vs, 2) + WS(rs, 1)];
358 TZ = rio[WS(vs, 2) + WS(rs, 3)];
359 T10 = TY + TZ;
360 T12 = TY - TZ;
361 T19 = iio[WS(vs, 2) + WS(rs, 1)];
362 T1a = iio[WS(vs, 2) + WS(rs, 3)];
363 T1b = T19 - T1a;
364 T1l = T19 + T1a;
365 }
366 {
367 E T1n, T1o, T1v, T1w;
368 T1n = rio[WS(vs, 3)];
369 T1o = rio[WS(vs, 3) + WS(rs, 2)];
370 T1p = T1n + T1o;
371 T1A = T1n - T1o;
372 T1v = iio[WS(vs, 3)];
373 T1w = iio[WS(vs, 3) + WS(rs, 2)];
374 T1x = T1v - T1w;
375 T1M = T1v + T1w;
376 }
377 {
378 E T1q, T1r, T1B, T1C;
379 T1q = rio[WS(vs, 3) + WS(rs, 1)];
380 T1r = rio[WS(vs, 3) + WS(rs, 3)];
381 T1s = T1q + T1r;
382 T1u = T1q - T1r;
383 T1B = iio[WS(vs, 3) + WS(rs, 1)];
384 T1C = iio[WS(vs, 3) + WS(rs, 3)];
385 T1D = T1B - T1C;
386 T1N = T1B + T1C;
387 }
388 rio[0] = T3 + T6;
389 iio[0] = Tq + Tr;
390 rio[WS(rs, 1)] = Tv + Ty;
391 iio[WS(rs, 1)] = TS + TT;
392 rio[WS(rs, 2)] = TX + T10;
393 iio[WS(rs, 2)] = T1k + T1l;
394 iio[WS(rs, 3)] = T1M + T1N;
395 rio[WS(rs, 3)] = T1p + T1s;
396 {
397 E Tc, Ti, T7, Td;
398 Tc = T8 + Tb;
399 Ti = Te - Th;
400 T7 = W[4];
401 Td = W[5];
402 iio[WS(vs, 3)] = FNMS(Td, Ti, T7 * Tc);
403 rio[WS(vs, 3)] = FMA(Td, Tc, T7 * Ti);
404 }
405 {
406 E T1K, T1O, T1J, T1L;
407 T1K = T1p - T1s;
408 T1O = T1M - T1N;
409 T1J = W[2];
410 T1L = W[3];
411 rio[WS(vs, 2) + WS(rs, 3)] = FMA(T1J, T1K, T1L * T1O);
412 iio[WS(vs, 2) + WS(rs, 3)] = FNMS(T1L, T1K, T1J * T1O);
413 }
414 {
415 E Tk, Tm, Tj, Tl;
416 Tk = Tb - T8;
417 Tm = Te + Th;
418 Tj = W[0];
419 Tl = W[1];
420 iio[WS(vs, 1)] = FNMS(Tl, Tm, Tj * Tk);
421 rio[WS(vs, 1)] = FMA(Tl, Tk, Tj * Tm);
422 }
423 {
424 E To, Ts, Tn, Tp;
425 To = T3 - T6;
426 Ts = Tq - Tr;
427 Tn = W[2];
428 Tp = W[3];
429 rio[WS(vs, 2)] = FMA(Tn, To, Tp * Ts);
430 iio[WS(vs, 2)] = FNMS(Tp, To, Tn * Ts);
431 }
432 {
433 E T16, T1c, T11, T17;
434 T16 = T12 + T15;
435 T1c = T18 - T1b;
436 T11 = W[4];
437 T17 = W[5];
438 iio[WS(vs, 3) + WS(rs, 2)] = FNMS(T17, T1c, T11 * T16);
439 rio[WS(vs, 3) + WS(rs, 2)] = FMA(T17, T16, T11 * T1c);
440 }
441 {
442 E T1G, T1I, T1F, T1H;
443 T1G = T1x - T1u;
444 T1I = T1A + T1D;
445 T1F = W[0];
446 T1H = W[1];
447 iio[WS(vs, 1) + WS(rs, 3)] = FNMS(T1H, T1I, T1F * T1G);
448 rio[WS(vs, 1) + WS(rs, 3)] = FMA(T1H, T1G, T1F * T1I);
449 }
450 {
451 E TQ, TU, TP, TR;
452 TQ = Tv - Ty;
453 TU = TS - TT;
454 TP = W[2];
455 TR = W[3];
456 rio[WS(vs, 2) + WS(rs, 1)] = FMA(TP, TQ, TR * TU);
457 iio[WS(vs, 2) + WS(rs, 1)] = FNMS(TR, TQ, TP * TU);
458 }
459 {
460 E T1e, T1g, T1d, T1f;
461 T1e = T15 - T12;
462 T1g = T18 + T1b;
463 T1d = W[0];
464 T1f = W[1];
465 iio[WS(vs, 1) + WS(rs, 2)] = FNMS(T1f, T1g, T1d * T1e);
466 rio[WS(vs, 1) + WS(rs, 2)] = FMA(T1f, T1e, T1d * T1g);
467 }
468 {
469 E T1i, T1m, T1h, T1j;
470 T1i = TX - T10;
471 T1m = T1k - T1l;
472 T1h = W[2];
473 T1j = W[3];
474 rio[WS(vs, 2) + WS(rs, 2)] = FMA(T1h, T1i, T1j * T1m);
475 iio[WS(vs, 2) + WS(rs, 2)] = FNMS(T1j, T1i, T1h * T1m);
476 }
477 {
478 E T1y, T1E, T1t, T1z;
479 T1y = T1u + T1x;
480 T1E = T1A - T1D;
481 T1t = W[4];
482 T1z = W[5];
483 iio[WS(vs, 3) + WS(rs, 3)] = FNMS(T1z, T1E, T1t * T1y);
484 rio[WS(vs, 3) + WS(rs, 3)] = FMA(T1z, T1y, T1t * T1E);
485 }
486 {
487 E TM, TO, TL, TN;
488 TM = TD - TA;
489 TO = TG + TJ;
490 TL = W[0];
491 TN = W[1];
492 iio[WS(vs, 1) + WS(rs, 1)] = FNMS(TN, TO, TL * TM);
493 rio[WS(vs, 1) + WS(rs, 1)] = FMA(TN, TM, TL * TO);
494 }
495 {
496 E TE, TK, Tz, TF;
497 TE = TA + TD;
498 TK = TG - TJ;
499 Tz = W[4];
500 TF = W[5];
501 iio[WS(vs, 3) + WS(rs, 1)] = FNMS(TF, TK, Tz * TE);
502 rio[WS(vs, 3) + WS(rs, 1)] = FMA(TF, TE, Tz * TK);
503 }
504 }
505 }
506 }
507
508 static const tw_instr twinstr[] = {
509 {TW_FULL, 0, 4},
510 {TW_NEXT, 1, 0}
511 };
512
513 static const ct_desc desc = { 4, "q1_4", twinstr, &GENUS, {64, 24, 24, 0}, 0, 0, 0 };
514
515 void X(codelet_q1_4) (planner *p) {
516 X(kdft_difsq_register) (p, q1_4, &desc);
517 }
518 #endif /* HAVE_FMA */