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 Dec 10 07:06:44 EST 2020 */
23
24 #include "rdft/codelet-rdft.h"
25
26 #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA)
27
28 /* Generated by: ../../../genfft/gen_hc2c.native -fma -compact -variables 4 -pipeline-latency 4 -sign 1 -n 12 -dif -name hc2cb_12 -include rdft/scalar/hc2cb.h */
29
30 /*
31 * This function contains 118 FP additions, 68 FP multiplications,
32 * (or, 72 additions, 22 multiplications, 46 fused multiply/add),
33 * 47 stack variables, 2 constants, and 48 memory accesses
34 */
35 #include "rdft/scalar/hc2cb.h"
36
hc2cb_12(R * Rp,R * Ip,R * Rm,R * Im,const R * W,stride rs,INT mb,INT me,INT ms)37 static void hc2cb_12(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms)
38 {
39 DK(KP866025403, +0.866025403784438646763723170752936183471402627);
40 DK(KP500000000, +0.500000000000000000000000000000000000000000000);
41 {
42 INT m;
43 for (m = mb, W = W + ((mb - 1) * 22); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 22, MAKE_VOLATILE_STRIDE(48, rs)) {
44 E T18, T20, T1b, T21, T1s, T2a, T1p, T29, TI, TN, TO, Tb, To, T1f, T23;
45 E T1i, T24, T1z, T2d, T1w, T2c, Tt, Ty, Tz, Tm, TD;
46 {
47 E T1, TE, T6, TM, T4, T1o, TH, T17, T9, T1r, TL, T1a;
48 T1 = Rp[0];
49 TE = Ip[0];
50 T6 = Rm[WS(rs, 5)];
51 TM = Im[WS(rs, 5)];
52 {
53 E T2, T3, TF, TG;
54 T2 = Rp[WS(rs, 4)];
55 T3 = Rm[WS(rs, 3)];
56 T4 = T2 + T3;
57 T1o = T2 - T3;
58 TF = Ip[WS(rs, 4)];
59 TG = Im[WS(rs, 3)];
60 TH = TF - TG;
61 T17 = TF + TG;
62 }
63 {
64 E T7, T8, TJ, TK;
65 T7 = Rm[WS(rs, 1)];
66 T8 = Rp[WS(rs, 2)];
67 T9 = T7 + T8;
68 T1r = T7 - T8;
69 TJ = Ip[WS(rs, 2)];
70 TK = Im[WS(rs, 1)];
71 TL = TJ - TK;
72 T1a = TJ + TK;
73 }
74 {
75 E T16, T19, T1q, T1n, T5, Ta;
76 T16 = FNMS(KP500000000, T4, T1);
77 T18 = FNMS(KP866025403, T17, T16);
78 T20 = FMA(KP866025403, T17, T16);
79 T19 = FNMS(KP500000000, T9, T6);
80 T1b = FMA(KP866025403, T1a, T19);
81 T21 = FNMS(KP866025403, T1a, T19);
82 T1q = FMA(KP500000000, TL, TM);
83 T1s = FNMS(KP866025403, T1r, T1q);
84 T2a = FMA(KP866025403, T1r, T1q);
85 T1n = FNMS(KP500000000, TH, TE);
86 T1p = FMA(KP866025403, T1o, T1n);
87 T29 = FNMS(KP866025403, T1o, T1n);
88 TI = TE + TH;
89 TN = TL - TM;
90 TO = TI - TN;
91 T5 = T1 + T4;
92 Ta = T6 + T9;
93 Tb = T5 + Ta;
94 To = T5 - Ta;
95 }
96 }
97 {
98 E Tc, Tp, Th, Tx, Tf, T1v, Ts, T1e, Tk, T1y, Tw, T1h;
99 Tc = Rp[WS(rs, 3)];
100 Tp = Ip[WS(rs, 3)];
101 Th = Rm[WS(rs, 2)];
102 Tx = Im[WS(rs, 2)];
103 {
104 E Td, Te, Tq, Tr;
105 Td = Rm[WS(rs, 4)];
106 Te = Rm[0];
107 Tf = Td + Te;
108 T1v = Td - Te;
109 Tq = Im[WS(rs, 4)];
110 Tr = Im[0];
111 Ts = Tq + Tr;
112 T1e = Tq - Tr;
113 }
114 {
115 E Ti, Tj, Tu, Tv;
116 Ti = Rp[WS(rs, 1)];
117 Tj = Rp[WS(rs, 5)];
118 Tk = Ti + Tj;
119 T1y = Ti - Tj;
120 Tu = Ip[WS(rs, 1)];
121 Tv = Ip[WS(rs, 5)];
122 Tw = Tu + Tv;
123 T1h = Tv - Tu;
124 }
125 {
126 E T1d, T1g, T1x, T1u, Tg, Tl;
127 T1d = FNMS(KP500000000, Tf, Tc);
128 T1f = FMA(KP866025403, T1e, T1d);
129 T23 = FNMS(KP866025403, T1e, T1d);
130 T1g = FNMS(KP500000000, Tk, Th);
131 T1i = FMA(KP866025403, T1h, T1g);
132 T24 = FNMS(KP866025403, T1h, T1g);
133 T1x = FMA(KP500000000, Tw, Tx);
134 T1z = FNMS(KP866025403, T1y, T1x);
135 T2d = FMA(KP866025403, T1y, T1x);
136 T1u = FMA(KP500000000, Ts, Tp);
137 T1w = FMA(KP866025403, T1v, T1u);
138 T2c = FNMS(KP866025403, T1v, T1u);
139 Tt = Tp - Ts;
140 Ty = Tw - Tx;
141 Tz = Tt - Ty;
142 Tg = Tc + Tf;
143 Tl = Th + Tk;
144 Tm = Tg + Tl;
145 TD = Tg - Tl;
146 }
147 }
148 Rp[0] = Tb + Tm;
149 {
150 E TA, TP, TB, TQ, Tn, TC;
151 TA = To - Tz;
152 TP = TD + TO;
153 Tn = W[16];
154 TB = Tn * TA;
155 TQ = Tn * TP;
156 TC = W[17];
157 Ip[WS(rs, 4)] = FNMS(TC, TP, TB);
158 Im[WS(rs, 4)] = FMA(TC, TA, TQ);
159 }
160 {
161 E TS, TV, TT, TW, TR, TU;
162 TS = To + Tz;
163 TV = TO - TD;
164 TR = W[4];
165 TT = TR * TS;
166 TW = TR * TV;
167 TU = W[5];
168 Ip[WS(rs, 1)] = FNMS(TU, TV, TT);
169 Im[WS(rs, 1)] = FMA(TU, TS, TW);
170 }
171 {
172 E T11, T12, T13, TX, TZ, T10, T14, TY;
173 T11 = TI + TN;
174 T12 = Tt + Ty;
175 T13 = T11 - T12;
176 TY = Tb - Tm;
177 TX = W[10];
178 TZ = TX * TY;
179 T10 = W[11];
180 T14 = T10 * TY;
181 Rm[0] = T11 + T12;
182 Rm[WS(rs, 3)] = FMA(TX, T13, T14);
183 Rp[WS(rs, 3)] = FNMS(T10, T13, TZ);
184 }
185 {
186 E T1k, T1E, T1B, T1H;
187 {
188 E T1c, T1j, T1t, T1A;
189 T1c = T18 + T1b;
190 T1j = T1f + T1i;
191 T1k = T1c - T1j;
192 T1E = T1c + T1j;
193 T1t = T1p - T1s;
194 T1A = T1w - T1z;
195 T1B = T1t - T1A;
196 T1H = T1t + T1A;
197 }
198 {
199 E T15, T1l, T1m, T1C;
200 T15 = W[18];
201 T1l = T15 * T1k;
202 T1m = W[19];
203 T1C = T1m * T1k;
204 Rp[WS(rs, 5)] = FNMS(T1m, T1B, T1l);
205 Rm[WS(rs, 5)] = FMA(T15, T1B, T1C);
206 }
207 {
208 E T1D, T1F, T1G, T1I;
209 T1D = W[6];
210 T1F = T1D * T1E;
211 T1G = W[7];
212 T1I = T1G * T1E;
213 Rp[WS(rs, 2)] = FNMS(T1G, T1H, T1F);
214 Rm[WS(rs, 2)] = FMA(T1D, T1H, T1I);
215 }
216 }
217 {
218 E T26, T2i, T2f, T2l;
219 {
220 E T22, T25, T2b, T2e;
221 T22 = T20 + T21;
222 T25 = T23 + T24;
223 T26 = T22 - T25;
224 T2i = T22 + T25;
225 T2b = T29 - T2a;
226 T2e = T2c - T2d;
227 T2f = T2b - T2e;
228 T2l = T2b + T2e;
229 }
230 {
231 E T1Z, T27, T28, T2g;
232 T1Z = W[2];
233 T27 = T1Z * T26;
234 T28 = W[3];
235 T2g = T28 * T26;
236 Rp[WS(rs, 1)] = FNMS(T28, T2f, T27);
237 Rm[WS(rs, 1)] = FMA(T1Z, T2f, T2g);
238 }
239 {
240 E T2h, T2j, T2k, T2m;
241 T2h = W[14];
242 T2j = T2h * T2i;
243 T2k = W[15];
244 T2m = T2k * T2i;
245 Rp[WS(rs, 4)] = FNMS(T2k, T2l, T2j);
246 Rm[WS(rs, 4)] = FMA(T2h, T2l, T2m);
247 }
248 }
249 {
250 E T2q, T2y, T2v, T2B;
251 {
252 E T2o, T2p, T2t, T2u;
253 T2o = T20 - T21;
254 T2p = T2c + T2d;
255 T2q = T2o - T2p;
256 T2y = T2o + T2p;
257 T2t = T29 + T2a;
258 T2u = T23 - T24;
259 T2v = T2t + T2u;
260 T2B = T2t - T2u;
261 }
262 {
263 E T2r, T2w, T2n, T2s;
264 T2n = W[8];
265 T2r = T2n * T2q;
266 T2w = T2n * T2v;
267 T2s = W[9];
268 Ip[WS(rs, 2)] = FNMS(T2s, T2v, T2r);
269 Im[WS(rs, 2)] = FMA(T2s, T2q, T2w);
270 }
271 {
272 E T2z, T2C, T2x, T2A;
273 T2x = W[20];
274 T2z = T2x * T2y;
275 T2C = T2x * T2B;
276 T2A = W[21];
277 Ip[WS(rs, 5)] = FNMS(T2A, T2B, T2z);
278 Im[WS(rs, 5)] = FMA(T2A, T2y, T2C);
279 }
280 }
281 {
282 E T1M, T1U, T1R, T1X;
283 {
284 E T1K, T1L, T1P, T1Q;
285 T1K = T18 - T1b;
286 T1L = T1w + T1z;
287 T1M = T1K - T1L;
288 T1U = T1K + T1L;
289 T1P = T1p + T1s;
290 T1Q = T1f - T1i;
291 T1R = T1P + T1Q;
292 T1X = T1P - T1Q;
293 }
294 {
295 E T1N, T1S, T1J, T1O;
296 T1J = W[0];
297 T1N = T1J * T1M;
298 T1S = T1J * T1R;
299 T1O = W[1];
300 Ip[0] = FNMS(T1O, T1R, T1N);
301 Im[0] = FMA(T1O, T1M, T1S);
302 }
303 {
304 E T1V, T1Y, T1T, T1W;
305 T1T = W[12];
306 T1V = T1T * T1U;
307 T1Y = T1T * T1X;
308 T1W = W[13];
309 Ip[WS(rs, 3)] = FNMS(T1W, T1X, T1V);
310 Im[WS(rs, 3)] = FMA(T1W, T1U, T1Y);
311 }
312 }
313 }
314 }
315 }
316
317 static const tw_instr twinstr[] = {
318 { TW_FULL, 1, 12 },
319 { TW_NEXT, 1, 0 }
320 };
321
322 static const hc2c_desc desc = { 12, "hc2cb_12", twinstr, &GENUS, { 72, 22, 46, 0 } };
323
X(codelet_hc2cb_12)324 void X(codelet_hc2cb_12) (planner *p) {
325 X(khc2c_register) (p, hc2cb_12, &desc, HC2C_VIA_RDFT);
326 }
327 #else
328
329 /* Generated by: ../../../genfft/gen_hc2c.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 12 -dif -name hc2cb_12 -include rdft/scalar/hc2cb.h */
330
331 /*
332 * This function contains 118 FP additions, 60 FP multiplications,
333 * (or, 88 additions, 30 multiplications, 30 fused multiply/add),
334 * 39 stack variables, 2 constants, and 48 memory accesses
335 */
336 #include "rdft/scalar/hc2cb.h"
337
hc2cb_12(R * Rp,R * Ip,R * Rm,R * Im,const R * W,stride rs,INT mb,INT me,INT ms)338 static void hc2cb_12(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms)
339 {
340 DK(KP500000000, +0.500000000000000000000000000000000000000000000);
341 DK(KP866025403, +0.866025403784438646763723170752936183471402627);
342 {
343 INT m;
344 for (m = mb, W = W + ((mb - 1) * 22); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 22, MAKE_VOLATILE_STRIDE(48, rs)) {
345 E T5, TH, T12, T1M, T1i, T1U, Tl, Ty, T1c, T1Y, T1s, T1Q, Ta, TM, T15;
346 E T1N, T1l, T1V, Tg, Tt, T19, T1X, T1p, T1P;
347 {
348 E T1, TD, T4, T1g, TG, T11, T10, T1h;
349 T1 = Rp[0];
350 TD = Ip[0];
351 {
352 E T2, T3, TE, TF;
353 T2 = Rp[WS(rs, 4)];
354 T3 = Rm[WS(rs, 3)];
355 T4 = T2 + T3;
356 T1g = KP866025403 * (T2 - T3);
357 TE = Ip[WS(rs, 4)];
358 TF = Im[WS(rs, 3)];
359 TG = TE - TF;
360 T11 = KP866025403 * (TE + TF);
361 }
362 T5 = T1 + T4;
363 TH = TD + TG;
364 T10 = FNMS(KP500000000, T4, T1);
365 T12 = T10 - T11;
366 T1M = T10 + T11;
367 T1h = FNMS(KP500000000, TG, TD);
368 T1i = T1g + T1h;
369 T1U = T1h - T1g;
370 }
371 {
372 E Th, Tx, Tk, T1a, Tw, T1r, T1b, T1q;
373 Th = Rm[WS(rs, 2)];
374 Tx = Im[WS(rs, 2)];
375 {
376 E Ti, Tj, Tu, Tv;
377 Ti = Rp[WS(rs, 1)];
378 Tj = Rp[WS(rs, 5)];
379 Tk = Ti + Tj;
380 T1a = KP866025403 * (Ti - Tj);
381 Tu = Ip[WS(rs, 1)];
382 Tv = Ip[WS(rs, 5)];
383 Tw = Tu + Tv;
384 T1r = KP866025403 * (Tv - Tu);
385 }
386 Tl = Th + Tk;
387 Ty = Tw - Tx;
388 T1b = FMA(KP500000000, Tw, Tx);
389 T1c = T1a - T1b;
390 T1Y = T1a + T1b;
391 T1q = FNMS(KP500000000, Tk, Th);
392 T1s = T1q + T1r;
393 T1Q = T1q - T1r;
394 }
395 {
396 E T6, TL, T9, T1j, TK, T14, T13, T1k;
397 T6 = Rm[WS(rs, 5)];
398 TL = Im[WS(rs, 5)];
399 {
400 E T7, T8, TI, TJ;
401 T7 = Rm[WS(rs, 1)];
402 T8 = Rp[WS(rs, 2)];
403 T9 = T7 + T8;
404 T1j = KP866025403 * (T7 - T8);
405 TI = Ip[WS(rs, 2)];
406 TJ = Im[WS(rs, 1)];
407 TK = TI - TJ;
408 T14 = KP866025403 * (TI + TJ);
409 }
410 Ta = T6 + T9;
411 TM = TK - TL;
412 T13 = FNMS(KP500000000, T9, T6);
413 T15 = T13 + T14;
414 T1N = T13 - T14;
415 T1k = FMA(KP500000000, TK, TL);
416 T1l = T1j - T1k;
417 T1V = T1j + T1k;
418 }
419 {
420 E Tc, Tp, Tf, T17, Ts, T1o, T18, T1n;
421 Tc = Rp[WS(rs, 3)];
422 Tp = Ip[WS(rs, 3)];
423 {
424 E Td, Te, Tq, Tr;
425 Td = Rm[WS(rs, 4)];
426 Te = Rm[0];
427 Tf = Td + Te;
428 T17 = KP866025403 * (Td - Te);
429 Tq = Im[WS(rs, 4)];
430 Tr = Im[0];
431 Ts = Tq + Tr;
432 T1o = KP866025403 * (Tq - Tr);
433 }
434 Tg = Tc + Tf;
435 Tt = Tp - Ts;
436 T18 = FMA(KP500000000, Ts, Tp);
437 T19 = T17 + T18;
438 T1X = T18 - T17;
439 T1n = FNMS(KP500000000, Tf, Tc);
440 T1p = T1n + T1o;
441 T1P = T1n - T1o;
442 }
443 {
444 E Tb, Tm, TU, TW, TX, TY, TT, TV;
445 Tb = T5 + Ta;
446 Tm = Tg + Tl;
447 TU = Tb - Tm;
448 TW = TH + TM;
449 TX = Tt + Ty;
450 TY = TW - TX;
451 Rp[0] = Tb + Tm;
452 Rm[0] = TW + TX;
453 TT = W[10];
454 TV = W[11];
455 Rp[WS(rs, 3)] = FNMS(TV, TY, TT * TU);
456 Rm[WS(rs, 3)] = FMA(TV, TU, TT * TY);
457 }
458 {
459 E TA, TQ, TO, TS;
460 {
461 E To, Tz, TC, TN;
462 To = T5 - Ta;
463 Tz = Tt - Ty;
464 TA = To - Tz;
465 TQ = To + Tz;
466 TC = Tg - Tl;
467 TN = TH - TM;
468 TO = TC + TN;
469 TS = TN - TC;
470 }
471 {
472 E Tn, TB, TP, TR;
473 Tn = W[16];
474 TB = W[17];
475 Ip[WS(rs, 4)] = FNMS(TB, TO, Tn * TA);
476 Im[WS(rs, 4)] = FMA(Tn, TO, TB * TA);
477 TP = W[4];
478 TR = W[5];
479 Ip[WS(rs, 1)] = FNMS(TR, TS, TP * TQ);
480 Im[WS(rs, 1)] = FMA(TP, TS, TR * TQ);
481 }
482 }
483 {
484 E T28, T2e, T2c, T2g;
485 {
486 E T26, T27, T2a, T2b;
487 T26 = T1M - T1N;
488 T27 = T1X + T1Y;
489 T28 = T26 - T27;
490 T2e = T26 + T27;
491 T2a = T1U + T1V;
492 T2b = T1P - T1Q;
493 T2c = T2a + T2b;
494 T2g = T2a - T2b;
495 }
496 {
497 E T25, T29, T2d, T2f;
498 T25 = W[8];
499 T29 = W[9];
500 Ip[WS(rs, 2)] = FNMS(T29, T2c, T25 * T28);
501 Im[WS(rs, 2)] = FMA(T25, T2c, T29 * T28);
502 T2d = W[20];
503 T2f = W[21];
504 Ip[WS(rs, 5)] = FNMS(T2f, T2g, T2d * T2e);
505 Im[WS(rs, 5)] = FMA(T2d, T2g, T2f * T2e);
506 }
507 }
508 {
509 E T1S, T22, T20, T24;
510 {
511 E T1O, T1R, T1W, T1Z;
512 T1O = T1M + T1N;
513 T1R = T1P + T1Q;
514 T1S = T1O - T1R;
515 T22 = T1O + T1R;
516 T1W = T1U - T1V;
517 T1Z = T1X - T1Y;
518 T20 = T1W - T1Z;
519 T24 = T1W + T1Z;
520 }
521 {
522 E T1L, T1T, T21, T23;
523 T1L = W[2];
524 T1T = W[3];
525 Rp[WS(rs, 1)] = FNMS(T1T, T20, T1L * T1S);
526 Rm[WS(rs, 1)] = FMA(T1T, T1S, T1L * T20);
527 T21 = W[14];
528 T23 = W[15];
529 Rp[WS(rs, 4)] = FNMS(T23, T24, T21 * T22);
530 Rm[WS(rs, 4)] = FMA(T23, T22, T21 * T24);
531 }
532 }
533 {
534 E T1C, T1I, T1G, T1K;
535 {
536 E T1A, T1B, T1E, T1F;
537 T1A = T12 + T15;
538 T1B = T1p + T1s;
539 T1C = T1A - T1B;
540 T1I = T1A + T1B;
541 T1E = T1i + T1l;
542 T1F = T19 + T1c;
543 T1G = T1E - T1F;
544 T1K = T1E + T1F;
545 }
546 {
547 E T1z, T1D, T1H, T1J;
548 T1z = W[18];
549 T1D = W[19];
550 Rp[WS(rs, 5)] = FNMS(T1D, T1G, T1z * T1C);
551 Rm[WS(rs, 5)] = FMA(T1D, T1C, T1z * T1G);
552 T1H = W[6];
553 T1J = W[7];
554 Rp[WS(rs, 2)] = FNMS(T1J, T1K, T1H * T1I);
555 Rm[WS(rs, 2)] = FMA(T1J, T1I, T1H * T1K);
556 }
557 }
558 {
559 E T1e, T1w, T1u, T1y;
560 {
561 E T16, T1d, T1m, T1t;
562 T16 = T12 - T15;
563 T1d = T19 - T1c;
564 T1e = T16 - T1d;
565 T1w = T16 + T1d;
566 T1m = T1i - T1l;
567 T1t = T1p - T1s;
568 T1u = T1m + T1t;
569 T1y = T1m - T1t;
570 }
571 {
572 E TZ, T1f, T1v, T1x;
573 TZ = W[0];
574 T1f = W[1];
575 Ip[0] = FNMS(T1f, T1u, TZ * T1e);
576 Im[0] = FMA(TZ, T1u, T1f * T1e);
577 T1v = W[12];
578 T1x = W[13];
579 Ip[WS(rs, 3)] = FNMS(T1x, T1y, T1v * T1w);
580 Im[WS(rs, 3)] = FMA(T1v, T1y, T1x * T1w);
581 }
582 }
583 }
584 }
585 }
586
587 static const tw_instr twinstr[] = {
588 { TW_FULL, 1, 12 },
589 { TW_NEXT, 1, 0 }
590 };
591
592 static const hc2c_desc desc = { 12, "hc2cb_12", twinstr, &GENUS, { 88, 30, 30, 0 } };
593
X(codelet_hc2cb_12)594 void X(codelet_hc2cb_12) (planner *p) {
595 X(khc2c_register) (p, hc2cb_12, &desc, HC2C_VIA_RDFT);
596 }
597 #endif
598