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 07:05:52 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_hc2hc.native -fma -compact -variables 4 -pipeline-latency 4 -n 12 -dit -name hf_12 -include rdft/scalar/hf.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/hf.h"
36
hf_12(R * cr,R * ci,const R * W,stride rs,INT mb,INT me,INT ms)37 static void hf_12(R *cr, R *ci, 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, cr = cr + ms, ci = ci - ms, W = W + 22, MAKE_VOLATILE_STRIDE(24, rs)) {
44 E T1, T2i, Tl, T2e, T10, T1Y, TG, T1S, Ty, T2s, T1s, T2f, T1d, T21, T1H;
45 E T1Z, Te, T2p, T1l, T2h, TT, T1V, T1A, T1T;
46 T1 = cr[0];
47 T2i = ci[0];
48 {
49 E Th, Tk, Ti, T2d, Tg, Tj;
50 Th = cr[WS(rs, 6)];
51 Tk = ci[WS(rs, 6)];
52 Tg = W[10];
53 Ti = Tg * Th;
54 T2d = Tg * Tk;
55 Tj = W[11];
56 Tl = FMA(Tj, Tk, Ti);
57 T2e = FNMS(Tj, Th, T2d);
58 }
59 {
60 E TW, TZ, TX, T1X, TV, TY;
61 TW = cr[WS(rs, 9)];
62 TZ = ci[WS(rs, 9)];
63 TV = W[16];
64 TX = TV * TW;
65 T1X = TV * TZ;
66 TY = W[17];
67 T10 = FMA(TY, TZ, TX);
68 T1Y = FNMS(TY, TW, T1X);
69 }
70 {
71 E TC, TF, TD, T1R, TB, TE;
72 TC = cr[WS(rs, 3)];
73 TF = ci[WS(rs, 3)];
74 TB = W[4];
75 TD = TB * TC;
76 T1R = TB * TF;
77 TE = W[5];
78 TG = FMA(TE, TF, TD);
79 T1S = FNMS(TE, TC, T1R);
80 }
81 {
82 E Tn, Tq, To, T1o, Tt, Tw, Tu, T1q, Tm, Ts;
83 Tn = cr[WS(rs, 10)];
84 Tq = ci[WS(rs, 10)];
85 Tm = W[18];
86 To = Tm * Tn;
87 T1o = Tm * Tq;
88 Tt = cr[WS(rs, 2)];
89 Tw = ci[WS(rs, 2)];
90 Ts = W[2];
91 Tu = Ts * Tt;
92 T1q = Ts * Tw;
93 {
94 E Tr, T1p, Tx, T1r, Tp, Tv;
95 Tp = W[19];
96 Tr = FMA(Tp, Tq, To);
97 T1p = FNMS(Tp, Tn, T1o);
98 Tv = W[3];
99 Tx = FMA(Tv, Tw, Tu);
100 T1r = FNMS(Tv, Tt, T1q);
101 Ty = Tr + Tx;
102 T2s = Tx - Tr;
103 T1s = T1p - T1r;
104 T2f = T1p + T1r;
105 }
106 }
107 {
108 E T12, T15, T13, T1D, T18, T1b, T19, T1F, T11, T17;
109 T12 = cr[WS(rs, 1)];
110 T15 = ci[WS(rs, 1)];
111 T11 = W[0];
112 T13 = T11 * T12;
113 T1D = T11 * T15;
114 T18 = cr[WS(rs, 5)];
115 T1b = ci[WS(rs, 5)];
116 T17 = W[8];
117 T19 = T17 * T18;
118 T1F = T17 * T1b;
119 {
120 E T16, T1E, T1c, T1G, T14, T1a;
121 T14 = W[1];
122 T16 = FMA(T14, T15, T13);
123 T1E = FNMS(T14, T12, T1D);
124 T1a = W[9];
125 T1c = FMA(T1a, T1b, T19);
126 T1G = FNMS(T1a, T18, T1F);
127 T1d = T16 + T1c;
128 T21 = T1c - T16;
129 T1H = T1E - T1G;
130 T1Z = T1E + T1G;
131 }
132 }
133 {
134 E T3, T6, T4, T1h, T9, Tc, Ta, T1j, T2, T8;
135 T3 = cr[WS(rs, 4)];
136 T6 = ci[WS(rs, 4)];
137 T2 = W[6];
138 T4 = T2 * T3;
139 T1h = T2 * T6;
140 T9 = cr[WS(rs, 8)];
141 Tc = ci[WS(rs, 8)];
142 T8 = W[14];
143 Ta = T8 * T9;
144 T1j = T8 * Tc;
145 {
146 E T7, T1i, Td, T1k, T5, Tb;
147 T5 = W[7];
148 T7 = FMA(T5, T6, T4);
149 T1i = FNMS(T5, T3, T1h);
150 Tb = W[15];
151 Td = FMA(Tb, Tc, Ta);
152 T1k = FNMS(Tb, T9, T1j);
153 Te = T7 + Td;
154 T2p = Td - T7;
155 T1l = T1i - T1k;
156 T2h = T1i + T1k;
157 }
158 }
159 {
160 E TI, TL, TJ, T1w, TO, TR, TP, T1y, TH, TN;
161 TI = cr[WS(rs, 7)];
162 TL = ci[WS(rs, 7)];
163 TH = W[12];
164 TJ = TH * TI;
165 T1w = TH * TL;
166 TO = cr[WS(rs, 11)];
167 TR = ci[WS(rs, 11)];
168 TN = W[20];
169 TP = TN * TO;
170 T1y = TN * TR;
171 {
172 E TM, T1x, TS, T1z, TK, TQ;
173 TK = W[13];
174 TM = FMA(TK, TL, TJ);
175 T1x = FNMS(TK, TI, T1w);
176 TQ = W[21];
177 TS = FMA(TQ, TR, TP);
178 T1z = FNMS(TQ, TO, T1y);
179 TT = TM + TS;
180 T1V = TS - TM;
181 T1A = T1x - T1z;
182 T1T = T1x + T1z;
183 }
184 }
185 {
186 E TA, T28, T2k, T2m, T1f, T2l, T2b, T2c;
187 {
188 E Tf, Tz, T2g, T2j;
189 Tf = T1 + Te;
190 Tz = Tl + Ty;
191 TA = Tf + Tz;
192 T28 = Tf - Tz;
193 T2g = T2e + T2f;
194 T2j = T2h + T2i;
195 T2k = T2g + T2j;
196 T2m = T2j - T2g;
197 }
198 {
199 E TU, T1e, T29, T2a;
200 TU = TG + TT;
201 T1e = T10 + T1d;
202 T1f = TU + T1e;
203 T2l = TU - T1e;
204 T29 = T1S + T1T;
205 T2a = T1Y + T1Z;
206 T2b = T29 - T2a;
207 T2c = T29 + T2a;
208 }
209 ci[WS(rs, 5)] = TA - T1f;
210 cr[WS(rs, 9)] = T2l - T2m;
211 ci[WS(rs, 8)] = T2l + T2m;
212 cr[0] = TA + T1f;
213 cr[WS(rs, 3)] = T28 - T2b;
214 cr[WS(rs, 6)] = T2c - T2k;
215 ci[WS(rs, 11)] = T2c + T2k;
216 ci[WS(rs, 2)] = T28 + T2b;
217 }
218 {
219 E T1m, T1K, T2q, T2y, T2t, T2z, T1t, T1L, T1B, T1N, T1W, T25, T22, T26, T1I;
220 E T1O;
221 {
222 E T1g, T2o, T2r, T1n;
223 T1g = FNMS(KP500000000, Te, T1);
224 T1m = FNMS(KP866025403, T1l, T1g);
225 T1K = FMA(KP866025403, T1l, T1g);
226 T2o = FNMS(KP500000000, T2h, T2i);
227 T2q = FNMS(KP866025403, T2p, T2o);
228 T2y = FMA(KP866025403, T2p, T2o);
229 T2r = FNMS(KP500000000, T2f, T2e);
230 T2t = FNMS(KP866025403, T2s, T2r);
231 T2z = FMA(KP866025403, T2s, T2r);
232 T1n = FNMS(KP500000000, Ty, Tl);
233 T1t = FNMS(KP866025403, T1s, T1n);
234 T1L = FMA(KP866025403, T1s, T1n);
235 }
236 {
237 E T1v, T1U, T20, T1C;
238 T1v = FNMS(KP500000000, TT, TG);
239 T1B = FNMS(KP866025403, T1A, T1v);
240 T1N = FMA(KP866025403, T1A, T1v);
241 T1U = FNMS(KP500000000, T1T, T1S);
242 T1W = FNMS(KP866025403, T1V, T1U);
243 T25 = FMA(KP866025403, T1V, T1U);
244 T20 = FNMS(KP500000000, T1Z, T1Y);
245 T22 = FNMS(KP866025403, T21, T20);
246 T26 = FMA(KP866025403, T21, T20);
247 T1C = FNMS(KP500000000, T1d, T10);
248 T1I = FNMS(KP866025403, T1H, T1C);
249 T1O = FMA(KP866025403, T1H, T1C);
250 }
251 {
252 E T1u, T1J, T2v, T2w;
253 T1u = T1m + T1t;
254 T1J = T1B + T1I;
255 cr[WS(rs, 2)] = T1u - T1J;
256 ci[WS(rs, 3)] = T1u + T1J;
257 T2v = T1W + T22;
258 T2w = T2t + T2q;
259 cr[WS(rs, 8)] = -(T2v + T2w);
260 ci[WS(rs, 9)] = T2w - T2v;
261 }
262 {
263 E T2B, T2C, T2x, T2A;
264 T2B = T25 + T26;
265 T2C = T2z + T2y;
266 cr[WS(rs, 10)] = T2B - T2C;
267 ci[WS(rs, 7)] = T2B + T2C;
268 T2x = T1O - T1N;
269 T2A = T2y - T2z;
270 cr[WS(rs, 7)] = T2x - T2A;
271 ci[WS(rs, 10)] = T2x + T2A;
272 }
273 {
274 E T1M, T1P, T24, T27;
275 T1M = T1K + T1L;
276 T1P = T1N + T1O;
277 ci[WS(rs, 1)] = T1M - T1P;
278 cr[WS(rs, 4)] = T1M + T1P;
279 T24 = T1K - T1L;
280 T27 = T25 - T26;
281 ci[WS(rs, 4)] = T24 - T27;
282 cr[WS(rs, 1)] = T24 + T27;
283 }
284 {
285 E T1Q, T23, T2n, T2u;
286 T1Q = T1m - T1t;
287 T23 = T1W - T22;
288 ci[0] = T1Q - T23;
289 cr[WS(rs, 5)] = T1Q + T23;
290 T2n = T1I - T1B;
291 T2u = T2q - T2t;
292 cr[WS(rs, 11)] = T2n - T2u;
293 ci[WS(rs, 6)] = T2n + T2u;
294 }
295 }
296 }
297 }
298 }
299
300 static const tw_instr twinstr[] = {
301 { TW_FULL, 1, 12 },
302 { TW_NEXT, 1, 0 }
303 };
304
305 static const hc2hc_desc desc = { 12, "hf_12", twinstr, &GENUS, { 72, 22, 46, 0 } };
306
X(codelet_hf_12)307 void X(codelet_hf_12) (planner *p) {
308 X(khc2hc_register) (p, hf_12, &desc);
309 }
310 #else
311
312 /* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -n 12 -dit -name hf_12 -include rdft/scalar/hf.h */
313
314 /*
315 * This function contains 118 FP additions, 60 FP multiplications,
316 * (or, 88 additions, 30 multiplications, 30 fused multiply/add),
317 * 47 stack variables, 2 constants, and 48 memory accesses
318 */
319 #include "rdft/scalar/hf.h"
320
hf_12(R * cr,R * ci,const R * W,stride rs,INT mb,INT me,INT ms)321 static void hf_12(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms)
322 {
323 DK(KP500000000, +0.500000000000000000000000000000000000000000000);
324 DK(KP866025403, +0.866025403784438646763723170752936183471402627);
325 {
326 INT m;
327 for (m = mb, W = W + ((mb - 1) * 22); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 22, MAKE_VOLATILE_STRIDE(24, rs)) {
328 E T1, T1W, T18, T23, Tc, T15, T1V, T22, TR, T1E, T1o, T1D, T12, T1l, T1F;
329 E T1G, Ti, T1S, T1d, T26, Tt, T1a, T1T, T25, TA, T1y, T1j, T1B, TL, T1g;
330 E T1z, T1A;
331 {
332 E T6, T16, Tb, T17;
333 T1 = cr[0];
334 T1W = ci[0];
335 {
336 E T3, T5, T2, T4;
337 T3 = cr[WS(rs, 4)];
338 T5 = ci[WS(rs, 4)];
339 T2 = W[6];
340 T4 = W[7];
341 T6 = FMA(T2, T3, T4 * T5);
342 T16 = FNMS(T4, T3, T2 * T5);
343 }
344 {
345 E T8, Ta, T7, T9;
346 T8 = cr[WS(rs, 8)];
347 Ta = ci[WS(rs, 8)];
348 T7 = W[14];
349 T9 = W[15];
350 Tb = FMA(T7, T8, T9 * Ta);
351 T17 = FNMS(T9, T8, T7 * Ta);
352 }
353 T18 = KP866025403 * (T16 - T17);
354 T23 = KP866025403 * (Tb - T6);
355 Tc = T6 + Tb;
356 T15 = FNMS(KP500000000, Tc, T1);
357 T1V = T16 + T17;
358 T22 = FNMS(KP500000000, T1V, T1W);
359 }
360 {
361 E T11, T1n, TW, T1m;
362 {
363 E TO, TQ, TN, TP;
364 TO = cr[WS(rs, 9)];
365 TQ = ci[WS(rs, 9)];
366 TN = W[16];
367 TP = W[17];
368 TR = FMA(TN, TO, TP * TQ);
369 T1E = FNMS(TP, TO, TN * TQ);
370 }
371 {
372 E TY, T10, TX, TZ;
373 TY = cr[WS(rs, 5)];
374 T10 = ci[WS(rs, 5)];
375 TX = W[8];
376 TZ = W[9];
377 T11 = FMA(TX, TY, TZ * T10);
378 T1n = FNMS(TZ, TY, TX * T10);
379 }
380 {
381 E TT, TV, TS, TU;
382 TT = cr[WS(rs, 1)];
383 TV = ci[WS(rs, 1)];
384 TS = W[0];
385 TU = W[1];
386 TW = FMA(TS, TT, TU * TV);
387 T1m = FNMS(TU, TT, TS * TV);
388 }
389 T1o = KP866025403 * (T1m - T1n);
390 T1D = KP866025403 * (T11 - TW);
391 T12 = TW + T11;
392 T1l = FNMS(KP500000000, T12, TR);
393 T1F = T1m + T1n;
394 T1G = FNMS(KP500000000, T1F, T1E);
395 }
396 {
397 E Ts, T1c, Tn, T1b;
398 {
399 E Tf, Th, Te, Tg;
400 Tf = cr[WS(rs, 6)];
401 Th = ci[WS(rs, 6)];
402 Te = W[10];
403 Tg = W[11];
404 Ti = FMA(Te, Tf, Tg * Th);
405 T1S = FNMS(Tg, Tf, Te * Th);
406 }
407 {
408 E Tp, Tr, To, Tq;
409 Tp = cr[WS(rs, 2)];
410 Tr = ci[WS(rs, 2)];
411 To = W[2];
412 Tq = W[3];
413 Ts = FMA(To, Tp, Tq * Tr);
414 T1c = FNMS(Tq, Tp, To * Tr);
415 }
416 {
417 E Tk, Tm, Tj, Tl;
418 Tk = cr[WS(rs, 10)];
419 Tm = ci[WS(rs, 10)];
420 Tj = W[18];
421 Tl = W[19];
422 Tn = FMA(Tj, Tk, Tl * Tm);
423 T1b = FNMS(Tl, Tk, Tj * Tm);
424 }
425 T1d = KP866025403 * (T1b - T1c);
426 T26 = KP866025403 * (Ts - Tn);
427 Tt = Tn + Ts;
428 T1a = FNMS(KP500000000, Tt, Ti);
429 T1T = T1b + T1c;
430 T25 = FNMS(KP500000000, T1T, T1S);
431 }
432 {
433 E TK, T1i, TF, T1h;
434 {
435 E Tx, Tz, Tw, Ty;
436 Tx = cr[WS(rs, 3)];
437 Tz = ci[WS(rs, 3)];
438 Tw = W[4];
439 Ty = W[5];
440 TA = FMA(Tw, Tx, Ty * Tz);
441 T1y = FNMS(Ty, Tx, Tw * Tz);
442 }
443 {
444 E TH, TJ, TG, TI;
445 TH = cr[WS(rs, 11)];
446 TJ = ci[WS(rs, 11)];
447 TG = W[20];
448 TI = W[21];
449 TK = FMA(TG, TH, TI * TJ);
450 T1i = FNMS(TI, TH, TG * TJ);
451 }
452 {
453 E TC, TE, TB, TD;
454 TC = cr[WS(rs, 7)];
455 TE = ci[WS(rs, 7)];
456 TB = W[12];
457 TD = W[13];
458 TF = FMA(TB, TC, TD * TE);
459 T1h = FNMS(TD, TC, TB * TE);
460 }
461 T1j = KP866025403 * (T1h - T1i);
462 T1B = KP866025403 * (TK - TF);
463 TL = TF + TK;
464 T1g = FNMS(KP500000000, TL, TA);
465 T1z = T1h + T1i;
466 T1A = FNMS(KP500000000, T1z, T1y);
467 }
468 {
469 E Tv, T1N, T1Y, T20, T14, T1Z, T1Q, T1R;
470 {
471 E Td, Tu, T1U, T1X;
472 Td = T1 + Tc;
473 Tu = Ti + Tt;
474 Tv = Td + Tu;
475 T1N = Td - Tu;
476 T1U = T1S + T1T;
477 T1X = T1V + T1W;
478 T1Y = T1U + T1X;
479 T20 = T1X - T1U;
480 }
481 {
482 E TM, T13, T1O, T1P;
483 TM = TA + TL;
484 T13 = TR + T12;
485 T14 = TM + T13;
486 T1Z = TM - T13;
487 T1O = T1y + T1z;
488 T1P = T1E + T1F;
489 T1Q = T1O - T1P;
490 T1R = T1O + T1P;
491 }
492 ci[WS(rs, 5)] = Tv - T14;
493 cr[WS(rs, 9)] = T1Z - T20;
494 ci[WS(rs, 8)] = T1Z + T20;
495 cr[0] = Tv + T14;
496 cr[WS(rs, 3)] = T1N - T1Q;
497 cr[WS(rs, 6)] = T1R - T1Y;
498 ci[WS(rs, 11)] = T1R + T1Y;
499 ci[WS(rs, 2)] = T1N + T1Q;
500 }
501 {
502 E T1f, T1x, T28, T2a, T1q, T21, T1I, T29;
503 {
504 E T19, T1e, T24, T27;
505 T19 = T15 - T18;
506 T1e = T1a - T1d;
507 T1f = T19 + T1e;
508 T1x = T19 - T1e;
509 T24 = T22 - T23;
510 T27 = T25 - T26;
511 T28 = T24 - T27;
512 T2a = T27 + T24;
513 }
514 {
515 E T1k, T1p, T1C, T1H;
516 T1k = T1g - T1j;
517 T1p = T1l - T1o;
518 T1q = T1k + T1p;
519 T21 = T1p - T1k;
520 T1C = T1A - T1B;
521 T1H = T1D - T1G;
522 T1I = T1C + T1H;
523 T29 = T1H - T1C;
524 }
525 cr[WS(rs, 2)] = T1f - T1q;
526 cr[WS(rs, 8)] = T29 - T2a;
527 ci[WS(rs, 9)] = T29 + T2a;
528 ci[WS(rs, 3)] = T1f + T1q;
529 ci[0] = T1x - T1I;
530 cr[WS(rs, 11)] = T21 - T28;
531 ci[WS(rs, 6)] = T21 + T28;
532 cr[WS(rs, 5)] = T1x + T1I;
533 }
534 {
535 E T1t, T1J, T2e, T2g, T1w, T2b, T1M, T2f;
536 {
537 E T1r, T1s, T2c, T2d;
538 T1r = T15 + T18;
539 T1s = T1a + T1d;
540 T1t = T1r + T1s;
541 T1J = T1r - T1s;
542 T2c = T23 + T22;
543 T2d = T26 + T25;
544 T2e = T2c - T2d;
545 T2g = T2d + T2c;
546 }
547 {
548 E T1u, T1v, T1K, T1L;
549 T1u = T1g + T1j;
550 T1v = T1l + T1o;
551 T1w = T1u + T1v;
552 T2b = T1v - T1u;
553 T1K = T1B + T1A;
554 T1L = T1D + T1G;
555 T1M = T1K - T1L;
556 T2f = T1K + T1L;
557 }
558 ci[WS(rs, 1)] = T1t - T1w;
559 cr[WS(rs, 1)] = T1J + T1M;
560 cr[WS(rs, 4)] = T1t + T1w;
561 ci[WS(rs, 4)] = T1J - T1M;
562 cr[WS(rs, 7)] = T2b - T2e;
563 ci[WS(rs, 7)] = T2f + T2g;
564 ci[WS(rs, 10)] = T2b + T2e;
565 cr[WS(rs, 10)] = T2f - T2g;
566 }
567 }
568 }
569 }
570
571 static const tw_instr twinstr[] = {
572 { TW_FULL, 1, 12 },
573 { TW_NEXT, 1, 0 }
574 };
575
576 static const hc2hc_desc desc = { 12, "hf_12", twinstr, &GENUS, { 88, 30, 30, 0 } };
577
X(codelet_hf_12)578 void X(codelet_hf_12) (planner *p) {
579 X(khc2hc_register) (p, hf_12, &desc);
580 }
581 #endif
582