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 07:06:50 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_hc2cdft.native -fma -compact -variables 4 -pipeline-latency 4 -sign 1 -n 8 -dif -name hc2cbdft2_8 -include rdft/scalar/hc2cb.h */
29
30 /*
31 * This function contains 82 FP additions, 36 FP multiplications,
32 * (or, 60 additions, 14 multiplications, 22 fused multiply/add),
33 * 41 stack variables, 1 constants, and 32 memory accesses
34 */
35 #include "rdft/scalar/hc2cb.h"
36
hc2cbdft2_8(R * Rp,R * Ip,R * Rm,R * Im,const R * W,stride rs,INT mb,INT me,INT ms)37 static void hc2cbdft2_8(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms)
38 {
39 DK(KP707106781, +0.707106781186547524400844362104849039284835938);
40 {
41 INT m;
42 for (m = mb, W = W + ((mb - 1) * 14); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 14, MAKE_VOLATILE_STRIDE(32, rs)) {
43 E Tl, T1p, T1g, TM, T1k, TE, TP, T1f, T7, Te, TU, TH, T1l, Tw, T1q;
44 E T1c, T1y;
45 {
46 E T3, TA, Tk, TN, T6, Th, TD, TO, Ta, Tm, Tp, TK, Td, Tr, Tu;
47 E TL, TF, TG;
48 {
49 E T1, T2, Ti, Tj;
50 T1 = Rp[0];
51 T2 = Rm[WS(rs, 3)];
52 T3 = T1 + T2;
53 TA = T1 - T2;
54 Ti = Ip[0];
55 Tj = Im[WS(rs, 3)];
56 Tk = Ti + Tj;
57 TN = Ti - Tj;
58 }
59 {
60 E T4, T5, TB, TC;
61 T4 = Rp[WS(rs, 2)];
62 T5 = Rm[WS(rs, 1)];
63 T6 = T4 + T5;
64 Th = T4 - T5;
65 TB = Ip[WS(rs, 2)];
66 TC = Im[WS(rs, 1)];
67 TD = TB + TC;
68 TO = TB - TC;
69 }
70 {
71 E T8, T9, Tn, To;
72 T8 = Rp[WS(rs, 1)];
73 T9 = Rm[WS(rs, 2)];
74 Ta = T8 + T9;
75 Tm = T8 - T9;
76 Tn = Ip[WS(rs, 1)];
77 To = Im[WS(rs, 2)];
78 Tp = Tn + To;
79 TK = Tn - To;
80 }
81 {
82 E Tb, Tc, Ts, Tt;
83 Tb = Rm[0];
84 Tc = Rp[WS(rs, 3)];
85 Td = Tb + Tc;
86 Tr = Tb - Tc;
87 Ts = Im[0];
88 Tt = Ip[WS(rs, 3)];
89 Tu = Ts + Tt;
90 TL = Tt - Ts;
91 }
92 Tl = Th + Tk;
93 T1p = TA + TD;
94 T1g = TN - TO;
95 TM = TK + TL;
96 T1k = Tk - Th;
97 TE = TA - TD;
98 TP = TN + TO;
99 T1f = Ta - Td;
100 T7 = T3 + T6;
101 Te = Ta + Td;
102 TU = T7 - Te;
103 TF = Tm - Tp;
104 TG = Tr - Tu;
105 TH = TF + TG;
106 T1l = TF - TG;
107 {
108 E Tq, Tv, T1a, T1b;
109 Tq = Tm + Tp;
110 Tv = Tr + Tu;
111 Tw = Tq - Tv;
112 T1q = Tq + Tv;
113 T1a = T3 - T6;
114 T1b = TL - TK;
115 T1c = T1a + T1b;
116 T1y = T1a - T1b;
117 }
118 }
119 {
120 E Tf, TQ, Tx, TI, Ty, TR, Tg, TJ, TS, Tz;
121 Tf = T7 + Te;
122 TQ = TM + TP;
123 Tx = FMA(KP707106781, Tw, Tl);
124 TI = FMA(KP707106781, TH, TE);
125 Tg = W[0];
126 Ty = Tg * Tx;
127 TR = Tg * TI;
128 Tz = W[1];
129 TJ = FMA(Tz, TI, Ty);
130 TS = FNMS(Tz, Tx, TR);
131 Rp[0] = Tf - TJ;
132 Ip[0] = TQ + TS;
133 Rm[0] = Tf + TJ;
134 Im[0] = TS - TQ;
135 }
136 {
137 E T1B, T1A, T1J, T1x, T1z, T1E, T1H, T1F, T1L, T1D;
138 T1B = T1g - T1f;
139 T1A = W[11];
140 T1J = T1A * T1y;
141 T1x = W[10];
142 T1z = T1x * T1y;
143 T1E = FNMS(KP707106781, T1l, T1k);
144 T1H = FMA(KP707106781, T1q, T1p);
145 T1D = W[12];
146 T1F = T1D * T1E;
147 T1L = T1D * T1H;
148 {
149 E T1C, T1K, T1I, T1M, T1G;
150 T1C = FNMS(T1A, T1B, T1z);
151 T1K = FMA(T1x, T1B, T1J);
152 T1G = W[13];
153 T1I = FMA(T1G, T1H, T1F);
154 T1M = FNMS(T1G, T1E, T1L);
155 Rp[WS(rs, 3)] = T1C - T1I;
156 Ip[WS(rs, 3)] = T1K + T1M;
157 Rm[WS(rs, 3)] = T1C + T1I;
158 Im[WS(rs, 3)] = T1M - T1K;
159 }
160 }
161 {
162 E TX, TW, T15, TT, TV, T10, T13, T11, T17, TZ;
163 TX = TP - TM;
164 TW = W[7];
165 T15 = TW * TU;
166 TT = W[6];
167 TV = TT * TU;
168 T10 = FNMS(KP707106781, Tw, Tl);
169 T13 = FNMS(KP707106781, TH, TE);
170 TZ = W[8];
171 T11 = TZ * T10;
172 T17 = TZ * T13;
173 {
174 E TY, T16, T14, T18, T12;
175 TY = FNMS(TW, TX, TV);
176 T16 = FMA(TT, TX, T15);
177 T12 = W[9];
178 T14 = FMA(T12, T13, T11);
179 T18 = FNMS(T12, T10, T17);
180 Rp[WS(rs, 2)] = TY - T14;
181 Ip[WS(rs, 2)] = T16 + T18;
182 Rm[WS(rs, 2)] = TY + T14;
183 Im[WS(rs, 2)] = T18 - T16;
184 }
185 }
186 {
187 E T1h, T1e, T1t, T19, T1d, T1m, T1r, T1n, T1v, T1j;
188 T1h = T1f + T1g;
189 T1e = W[3];
190 T1t = T1e * T1c;
191 T19 = W[2];
192 T1d = T19 * T1c;
193 T1m = FMA(KP707106781, T1l, T1k);
194 T1r = FNMS(KP707106781, T1q, T1p);
195 T1j = W[4];
196 T1n = T1j * T1m;
197 T1v = T1j * T1r;
198 {
199 E T1i, T1u, T1s, T1w, T1o;
200 T1i = FNMS(T1e, T1h, T1d);
201 T1u = FMA(T19, T1h, T1t);
202 T1o = W[5];
203 T1s = FMA(T1o, T1r, T1n);
204 T1w = FNMS(T1o, T1m, T1v);
205 Rp[WS(rs, 1)] = T1i - T1s;
206 Ip[WS(rs, 1)] = T1u + T1w;
207 Rm[WS(rs, 1)] = T1i + T1s;
208 Im[WS(rs, 1)] = T1w - T1u;
209 }
210 }
211 }
212 }
213 }
214
215 static const tw_instr twinstr[] = {
216 { TW_FULL, 1, 8 },
217 { TW_NEXT, 1, 0 }
218 };
219
220 static const hc2c_desc desc = { 8, "hc2cbdft2_8", twinstr, &GENUS, { 60, 14, 22, 0 } };
221
X(codelet_hc2cbdft2_8)222 void X(codelet_hc2cbdft2_8) (planner *p) {
223 X(khc2c_register) (p, hc2cbdft2_8, &desc, HC2C_VIA_DFT);
224 }
225 #else
226
227 /* Generated by: ../../../genfft/gen_hc2cdft.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 8 -dif -name hc2cbdft2_8 -include rdft/scalar/hc2cb.h */
228
229 /*
230 * This function contains 82 FP additions, 32 FP multiplications,
231 * (or, 68 additions, 18 multiplications, 14 fused multiply/add),
232 * 30 stack variables, 1 constants, and 32 memory accesses
233 */
234 #include "rdft/scalar/hc2cb.h"
235
hc2cbdft2_8(R * Rp,R * Ip,R * Rm,R * Im,const R * W,stride rs,INT mb,INT me,INT ms)236 static void hc2cbdft2_8(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms)
237 {
238 DK(KP707106781, +0.707106781186547524400844362104849039284835938);
239 {
240 INT m;
241 for (m = mb, W = W + ((mb - 1) * 14); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 14, MAKE_VOLATILE_STRIDE(32, rs)) {
242 E T7, T1d, T1h, Tl, TG, T14, T19, TO, Te, TL, T18, T15, TB, T1e, Tw;
243 E T1i;
244 {
245 E T3, TC, Tk, TM, T6, Th, TF, TN;
246 {
247 E T1, T2, Ti, Tj;
248 T1 = Rp[0];
249 T2 = Rm[WS(rs, 3)];
250 T3 = T1 + T2;
251 TC = T1 - T2;
252 Ti = Ip[0];
253 Tj = Im[WS(rs, 3)];
254 Tk = Ti + Tj;
255 TM = Ti - Tj;
256 }
257 {
258 E T4, T5, TD, TE;
259 T4 = Rp[WS(rs, 2)];
260 T5 = Rm[WS(rs, 1)];
261 T6 = T4 + T5;
262 Th = T4 - T5;
263 TD = Ip[WS(rs, 2)];
264 TE = Im[WS(rs, 1)];
265 TF = TD + TE;
266 TN = TD - TE;
267 }
268 T7 = T3 + T6;
269 T1d = Tk - Th;
270 T1h = TC + TF;
271 Tl = Th + Tk;
272 TG = TC - TF;
273 T14 = T3 - T6;
274 T19 = TM - TN;
275 TO = TM + TN;
276 }
277 {
278 E Ta, Tm, Tp, TJ, Td, Tr, Tu, TK;
279 {
280 E T8, T9, Tn, To;
281 T8 = Rp[WS(rs, 1)];
282 T9 = Rm[WS(rs, 2)];
283 Ta = T8 + T9;
284 Tm = T8 - T9;
285 Tn = Ip[WS(rs, 1)];
286 To = Im[WS(rs, 2)];
287 Tp = Tn + To;
288 TJ = Tn - To;
289 }
290 {
291 E Tb, Tc, Ts, Tt;
292 Tb = Rm[0];
293 Tc = Rp[WS(rs, 3)];
294 Td = Tb + Tc;
295 Tr = Tb - Tc;
296 Ts = Im[0];
297 Tt = Ip[WS(rs, 3)];
298 Tu = Ts + Tt;
299 TK = Tt - Ts;
300 }
301 Te = Ta + Td;
302 TL = TJ + TK;
303 T18 = Ta - Td;
304 T15 = TK - TJ;
305 {
306 E Tz, TA, Tq, Tv;
307 Tz = Tm - Tp;
308 TA = Tr - Tu;
309 TB = KP707106781 * (Tz + TA);
310 T1e = KP707106781 * (Tz - TA);
311 Tq = Tm + Tp;
312 Tv = Tr + Tu;
313 Tw = KP707106781 * (Tq - Tv);
314 T1i = KP707106781 * (Tq + Tv);
315 }
316 }
317 {
318 E Tf, TP, TI, TQ;
319 Tf = T7 + Te;
320 TP = TL + TO;
321 {
322 E Tx, TH, Tg, Ty;
323 Tx = Tl + Tw;
324 TH = TB + TG;
325 Tg = W[0];
326 Ty = W[1];
327 TI = FMA(Tg, Tx, Ty * TH);
328 TQ = FNMS(Ty, Tx, Tg * TH);
329 }
330 Rp[0] = Tf - TI;
331 Ip[0] = TP + TQ;
332 Rm[0] = Tf + TI;
333 Im[0] = TQ - TP;
334 }
335 {
336 E T1r, T1x, T1w, T1y;
337 {
338 E T1o, T1q, T1n, T1p;
339 T1o = T14 - T15;
340 T1q = T19 - T18;
341 T1n = W[10];
342 T1p = W[11];
343 T1r = FNMS(T1p, T1q, T1n * T1o);
344 T1x = FMA(T1p, T1o, T1n * T1q);
345 }
346 {
347 E T1t, T1v, T1s, T1u;
348 T1t = T1d - T1e;
349 T1v = T1i + T1h;
350 T1s = W[12];
351 T1u = W[13];
352 T1w = FMA(T1s, T1t, T1u * T1v);
353 T1y = FNMS(T1u, T1t, T1s * T1v);
354 }
355 Rp[WS(rs, 3)] = T1r - T1w;
356 Ip[WS(rs, 3)] = T1x + T1y;
357 Rm[WS(rs, 3)] = T1r + T1w;
358 Im[WS(rs, 3)] = T1y - T1x;
359 }
360 {
361 E TV, T11, T10, T12;
362 {
363 E TS, TU, TR, TT;
364 TS = T7 - Te;
365 TU = TO - TL;
366 TR = W[6];
367 TT = W[7];
368 TV = FNMS(TT, TU, TR * TS);
369 T11 = FMA(TT, TS, TR * TU);
370 }
371 {
372 E TX, TZ, TW, TY;
373 TX = Tl - Tw;
374 TZ = TG - TB;
375 TW = W[8];
376 TY = W[9];
377 T10 = FMA(TW, TX, TY * TZ);
378 T12 = FNMS(TY, TX, TW * TZ);
379 }
380 Rp[WS(rs, 2)] = TV - T10;
381 Ip[WS(rs, 2)] = T11 + T12;
382 Rm[WS(rs, 2)] = TV + T10;
383 Im[WS(rs, 2)] = T12 - T11;
384 }
385 {
386 E T1b, T1l, T1k, T1m;
387 {
388 E T16, T1a, T13, T17;
389 T16 = T14 + T15;
390 T1a = T18 + T19;
391 T13 = W[2];
392 T17 = W[3];
393 T1b = FNMS(T17, T1a, T13 * T16);
394 T1l = FMA(T17, T16, T13 * T1a);
395 }
396 {
397 E T1f, T1j, T1c, T1g;
398 T1f = T1d + T1e;
399 T1j = T1h - T1i;
400 T1c = W[4];
401 T1g = W[5];
402 T1k = FMA(T1c, T1f, T1g * T1j);
403 T1m = FNMS(T1g, T1f, T1c * T1j);
404 }
405 Rp[WS(rs, 1)] = T1b - T1k;
406 Ip[WS(rs, 1)] = T1l + T1m;
407 Rm[WS(rs, 1)] = T1b + T1k;
408 Im[WS(rs, 1)] = T1m - T1l;
409 }
410 }
411 }
412 }
413
414 static const tw_instr twinstr[] = {
415 { TW_FULL, 1, 8 },
416 { TW_NEXT, 1, 0 }
417 };
418
419 static const hc2c_desc desc = { 8, "hc2cbdft2_8", twinstr, &GENUS, { 68, 18, 14, 0 } };
420
X(codelet_hc2cbdft2_8)421 void X(codelet_hc2cbdft2_8) (planner *p) {
422 X(khc2c_register) (p, hc2cbdft2_8, &desc, HC2C_VIA_DFT);
423 }
424 #endif
425