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 07:05:37 EST 2020 */
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_twiddle.native -fma -simd -compact -variables 4 -pipeline-latency 8 -n 8 -name t1sv_8 -include dft/simd/ts.h */
29
30 /*
31 * This function contains 66 FP additions, 36 FP multiplications,
32 * (or, 44 additions, 14 multiplications, 22 fused multiply/add),
33 * 34 stack variables, 1 constants, and 32 memory accesses
34 */
35 #include "dft/simd/ts.h"
36
t1sv_8(R * ri,R * ii,const R * W,stride rs,INT mb,INT me,INT ms)37 static void t1sv_8(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms)
38 {
39 DVK(KP707106781, +0.707106781186547524400844362104849039284835938);
40 {
41 INT m;
42 for (m = mb, W = W + (mb * 14); m < me; m = m + (2 * VL), ri = ri + ((2 * VL) * ms), ii = ii + ((2 * VL) * ms), W = W + ((2 * VL) * 14), MAKE_VOLATILE_STRIDE(16, rs)) {
43 V T1, T1m, T7, T1l, Tk, TS, Te, TQ, TF, T14, TL, T16, T12, T17, Ts;
44 V TX, Ty, TZ, TV, T10;
45 T1 = LD(&(ri[0]), ms, &(ri[0]));
46 T1m = LD(&(ii[0]), ms, &(ii[0]));
47 {
48 V T3, T6, T4, T1k, T2, T5;
49 T3 = LD(&(ri[WS(rs, 4)]), ms, &(ri[0]));
50 T6 = LD(&(ii[WS(rs, 4)]), ms, &(ii[0]));
51 T2 = LDW(&(W[TWVL * 6]));
52 T4 = VMUL(T2, T3);
53 T1k = VMUL(T2, T6);
54 T5 = LDW(&(W[TWVL * 7]));
55 T7 = VFMA(T5, T6, T4);
56 T1l = VFNMS(T5, T3, T1k);
57 }
58 {
59 V Tg, Tj, Th, TR, Tf, Ti;
60 Tg = LD(&(ri[WS(rs, 6)]), ms, &(ri[0]));
61 Tj = LD(&(ii[WS(rs, 6)]), ms, &(ii[0]));
62 Tf = LDW(&(W[TWVL * 10]));
63 Th = VMUL(Tf, Tg);
64 TR = VMUL(Tf, Tj);
65 Ti = LDW(&(W[TWVL * 11]));
66 Tk = VFMA(Ti, Tj, Th);
67 TS = VFNMS(Ti, Tg, TR);
68 }
69 {
70 V Ta, Td, Tb, TP, T9, Tc;
71 Ta = LD(&(ri[WS(rs, 2)]), ms, &(ri[0]));
72 Td = LD(&(ii[WS(rs, 2)]), ms, &(ii[0]));
73 T9 = LDW(&(W[TWVL * 2]));
74 Tb = VMUL(T9, Ta);
75 TP = VMUL(T9, Td);
76 Tc = LDW(&(W[TWVL * 3]));
77 Te = VFMA(Tc, Td, Tb);
78 TQ = VFNMS(Tc, Ta, TP);
79 }
80 {
81 V TB, TE, TC, T13, TH, TK, TI, T15, TA, TG, TD, TJ;
82 TB = LD(&(ri[WS(rs, 7)]), ms, &(ri[WS(rs, 1)]));
83 TE = LD(&(ii[WS(rs, 7)]), ms, &(ii[WS(rs, 1)]));
84 TA = LDW(&(W[TWVL * 12]));
85 TC = VMUL(TA, TB);
86 T13 = VMUL(TA, TE);
87 TH = LD(&(ri[WS(rs, 3)]), ms, &(ri[WS(rs, 1)]));
88 TK = LD(&(ii[WS(rs, 3)]), ms, &(ii[WS(rs, 1)]));
89 TG = LDW(&(W[TWVL * 4]));
90 TI = VMUL(TG, TH);
91 T15 = VMUL(TG, TK);
92 TD = LDW(&(W[TWVL * 13]));
93 TF = VFMA(TD, TE, TC);
94 T14 = VFNMS(TD, TB, T13);
95 TJ = LDW(&(W[TWVL * 5]));
96 TL = VFMA(TJ, TK, TI);
97 T16 = VFNMS(TJ, TH, T15);
98 T12 = VSUB(TF, TL);
99 T17 = VSUB(T14, T16);
100 }
101 {
102 V To, Tr, Tp, TW, Tu, Tx, Tv, TY, Tn, Tt, Tq, Tw;
103 To = LD(&(ri[WS(rs, 1)]), ms, &(ri[WS(rs, 1)]));
104 Tr = LD(&(ii[WS(rs, 1)]), ms, &(ii[WS(rs, 1)]));
105 Tn = LDW(&(W[0]));
106 Tp = VMUL(Tn, To);
107 TW = VMUL(Tn, Tr);
108 Tu = LD(&(ri[WS(rs, 5)]), ms, &(ri[WS(rs, 1)]));
109 Tx = LD(&(ii[WS(rs, 5)]), ms, &(ii[WS(rs, 1)]));
110 Tt = LDW(&(W[TWVL * 8]));
111 Tv = VMUL(Tt, Tu);
112 TY = VMUL(Tt, Tx);
113 Tq = LDW(&(W[TWVL * 1]));
114 Ts = VFMA(Tq, Tr, Tp);
115 TX = VFNMS(Tq, To, TW);
116 Tw = LDW(&(W[TWVL * 9]));
117 Ty = VFMA(Tw, Tx, Tv);
118 TZ = VFNMS(Tw, Tu, TY);
119 TV = VSUB(Ts, Ty);
120 T10 = VSUB(TX, TZ);
121 }
122 {
123 V TU, T1a, T1t, T1v, T19, T1w, T1d, T1u;
124 {
125 V TO, TT, T1r, T1s;
126 TO = VSUB(T1, T7);
127 TT = VSUB(TQ, TS);
128 TU = VADD(TO, TT);
129 T1a = VSUB(TO, TT);
130 T1r = VSUB(T1m, T1l);
131 T1s = VSUB(Te, Tk);
132 T1t = VSUB(T1r, T1s);
133 T1v = VADD(T1s, T1r);
134 }
135 {
136 V T11, T18, T1b, T1c;
137 T11 = VADD(TV, T10);
138 T18 = VSUB(T12, T17);
139 T19 = VADD(T11, T18);
140 T1w = VSUB(T18, T11);
141 T1b = VSUB(T10, TV);
142 T1c = VADD(T12, T17);
143 T1d = VSUB(T1b, T1c);
144 T1u = VADD(T1b, T1c);
145 }
146 ST(&(ri[WS(rs, 5)]), VFNMS(LDK(KP707106781), T19, TU), ms, &(ri[WS(rs, 1)]));
147 ST(&(ii[WS(rs, 5)]), VFNMS(LDK(KP707106781), T1u, T1t), ms, &(ii[WS(rs, 1)]));
148 ST(&(ri[WS(rs, 1)]), VFMA(LDK(KP707106781), T19, TU), ms, &(ri[WS(rs, 1)]));
149 ST(&(ii[WS(rs, 1)]), VFMA(LDK(KP707106781), T1u, T1t), ms, &(ii[WS(rs, 1)]));
150 ST(&(ri[WS(rs, 7)]), VFNMS(LDK(KP707106781), T1d, T1a), ms, &(ri[WS(rs, 1)]));
151 ST(&(ii[WS(rs, 7)]), VFNMS(LDK(KP707106781), T1w, T1v), ms, &(ii[WS(rs, 1)]));
152 ST(&(ri[WS(rs, 3)]), VFMA(LDK(KP707106781), T1d, T1a), ms, &(ri[WS(rs, 1)]));
153 ST(&(ii[WS(rs, 3)]), VFMA(LDK(KP707106781), T1w, T1v), ms, &(ii[WS(rs, 1)]));
154 }
155 {
156 V Tm, T1e, T1o, T1q, TN, T1p, T1h, T1i;
157 {
158 V T8, Tl, T1j, T1n;
159 T8 = VADD(T1, T7);
160 Tl = VADD(Te, Tk);
161 Tm = VADD(T8, Tl);
162 T1e = VSUB(T8, Tl);
163 T1j = VADD(TQ, TS);
164 T1n = VADD(T1l, T1m);
165 T1o = VADD(T1j, T1n);
166 T1q = VSUB(T1n, T1j);
167 }
168 {
169 V Tz, TM, T1f, T1g;
170 Tz = VADD(Ts, Ty);
171 TM = VADD(TF, TL);
172 TN = VADD(Tz, TM);
173 T1p = VSUB(TM, Tz);
174 T1f = VADD(TX, TZ);
175 T1g = VADD(T14, T16);
176 T1h = VSUB(T1f, T1g);
177 T1i = VADD(T1f, T1g);
178 }
179 ST(&(ri[WS(rs, 4)]), VSUB(Tm, TN), ms, &(ri[0]));
180 ST(&(ii[WS(rs, 4)]), VSUB(T1o, T1i), ms, &(ii[0]));
181 ST(&(ri[0]), VADD(Tm, TN), ms, &(ri[0]));
182 ST(&(ii[0]), VADD(T1i, T1o), ms, &(ii[0]));
183 ST(&(ri[WS(rs, 6)]), VSUB(T1e, T1h), ms, &(ri[0]));
184 ST(&(ii[WS(rs, 6)]), VSUB(T1q, T1p), ms, &(ii[0]));
185 ST(&(ri[WS(rs, 2)]), VADD(T1e, T1h), ms, &(ri[0]));
186 ST(&(ii[WS(rs, 2)]), VADD(T1p, T1q), ms, &(ii[0]));
187 }
188 }
189 }
190 VLEAVE();
191 }
192
193 static const tw_instr twinstr[] = {
194 VTW(0, 1),
195 VTW(0, 2),
196 VTW(0, 3),
197 VTW(0, 4),
198 VTW(0, 5),
199 VTW(0, 6),
200 VTW(0, 7),
201 { TW_NEXT, (2 * VL), 0 }
202 };
203
204 static const ct_desc desc = { 8, XSIMD_STRING("t1sv_8"), twinstr, &GENUS, { 44, 14, 22, 0 }, 0, 0, 0 };
205
XSIMD(codelet_t1sv_8)206 void XSIMD(codelet_t1sv_8) (planner *p) {
207 X(kdft_dit_register) (p, t1sv_8, &desc);
208 }
209 #else
210
211 /* Generated by: ../../../genfft/gen_twiddle.native -simd -compact -variables 4 -pipeline-latency 8 -n 8 -name t1sv_8 -include dft/simd/ts.h */
212
213 /*
214 * This function contains 66 FP additions, 32 FP multiplications,
215 * (or, 52 additions, 18 multiplications, 14 fused multiply/add),
216 * 28 stack variables, 1 constants, and 32 memory accesses
217 */
218 #include "dft/simd/ts.h"
219
t1sv_8(R * ri,R * ii,const R * W,stride rs,INT mb,INT me,INT ms)220 static void t1sv_8(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms)
221 {
222 DVK(KP707106781, +0.707106781186547524400844362104849039284835938);
223 {
224 INT m;
225 for (m = mb, W = W + (mb * 14); m < me; m = m + (2 * VL), ri = ri + ((2 * VL) * ms), ii = ii + ((2 * VL) * ms), W = W + ((2 * VL) * 14), MAKE_VOLATILE_STRIDE(16, rs)) {
226 V T7, T1e, TH, T19, TF, T13, TR, TU, Ti, T1f, TK, T16, Tu, T12, TM;
227 V TP;
228 {
229 V T1, T18, T6, T17;
230 T1 = LD(&(ri[0]), ms, &(ri[0]));
231 T18 = LD(&(ii[0]), ms, &(ii[0]));
232 {
233 V T3, T5, T2, T4;
234 T3 = LD(&(ri[WS(rs, 4)]), ms, &(ri[0]));
235 T5 = LD(&(ii[WS(rs, 4)]), ms, &(ii[0]));
236 T2 = LDW(&(W[TWVL * 6]));
237 T4 = LDW(&(W[TWVL * 7]));
238 T6 = VFMA(T2, T3, VMUL(T4, T5));
239 T17 = VFNMS(T4, T3, VMUL(T2, T5));
240 }
241 T7 = VADD(T1, T6);
242 T1e = VSUB(T18, T17);
243 TH = VSUB(T1, T6);
244 T19 = VADD(T17, T18);
245 }
246 {
247 V Tz, TS, TE, TT;
248 {
249 V Tw, Ty, Tv, Tx;
250 Tw = LD(&(ri[WS(rs, 7)]), ms, &(ri[WS(rs, 1)]));
251 Ty = LD(&(ii[WS(rs, 7)]), ms, &(ii[WS(rs, 1)]));
252 Tv = LDW(&(W[TWVL * 12]));
253 Tx = LDW(&(W[TWVL * 13]));
254 Tz = VFMA(Tv, Tw, VMUL(Tx, Ty));
255 TS = VFNMS(Tx, Tw, VMUL(Tv, Ty));
256 }
257 {
258 V TB, TD, TA, TC;
259 TB = LD(&(ri[WS(rs, 3)]), ms, &(ri[WS(rs, 1)]));
260 TD = LD(&(ii[WS(rs, 3)]), ms, &(ii[WS(rs, 1)]));
261 TA = LDW(&(W[TWVL * 4]));
262 TC = LDW(&(W[TWVL * 5]));
263 TE = VFMA(TA, TB, VMUL(TC, TD));
264 TT = VFNMS(TC, TB, VMUL(TA, TD));
265 }
266 TF = VADD(Tz, TE);
267 T13 = VADD(TS, TT);
268 TR = VSUB(Tz, TE);
269 TU = VSUB(TS, TT);
270 }
271 {
272 V Tc, TI, Th, TJ;
273 {
274 V T9, Tb, T8, Ta;
275 T9 = LD(&(ri[WS(rs, 2)]), ms, &(ri[0]));
276 Tb = LD(&(ii[WS(rs, 2)]), ms, &(ii[0]));
277 T8 = LDW(&(W[TWVL * 2]));
278 Ta = LDW(&(W[TWVL * 3]));
279 Tc = VFMA(T8, T9, VMUL(Ta, Tb));
280 TI = VFNMS(Ta, T9, VMUL(T8, Tb));
281 }
282 {
283 V Te, Tg, Td, Tf;
284 Te = LD(&(ri[WS(rs, 6)]), ms, &(ri[0]));
285 Tg = LD(&(ii[WS(rs, 6)]), ms, &(ii[0]));
286 Td = LDW(&(W[TWVL * 10]));
287 Tf = LDW(&(W[TWVL * 11]));
288 Th = VFMA(Td, Te, VMUL(Tf, Tg));
289 TJ = VFNMS(Tf, Te, VMUL(Td, Tg));
290 }
291 Ti = VADD(Tc, Th);
292 T1f = VSUB(Tc, Th);
293 TK = VSUB(TI, TJ);
294 T16 = VADD(TI, TJ);
295 }
296 {
297 V To, TN, Tt, TO;
298 {
299 V Tl, Tn, Tk, Tm;
300 Tl = LD(&(ri[WS(rs, 1)]), ms, &(ri[WS(rs, 1)]));
301 Tn = LD(&(ii[WS(rs, 1)]), ms, &(ii[WS(rs, 1)]));
302 Tk = LDW(&(W[0]));
303 Tm = LDW(&(W[TWVL * 1]));
304 To = VFMA(Tk, Tl, VMUL(Tm, Tn));
305 TN = VFNMS(Tm, Tl, VMUL(Tk, Tn));
306 }
307 {
308 V Tq, Ts, Tp, Tr;
309 Tq = LD(&(ri[WS(rs, 5)]), ms, &(ri[WS(rs, 1)]));
310 Ts = LD(&(ii[WS(rs, 5)]), ms, &(ii[WS(rs, 1)]));
311 Tp = LDW(&(W[TWVL * 8]));
312 Tr = LDW(&(W[TWVL * 9]));
313 Tt = VFMA(Tp, Tq, VMUL(Tr, Ts));
314 TO = VFNMS(Tr, Tq, VMUL(Tp, Ts));
315 }
316 Tu = VADD(To, Tt);
317 T12 = VADD(TN, TO);
318 TM = VSUB(To, Tt);
319 TP = VSUB(TN, TO);
320 }
321 {
322 V Tj, TG, T1b, T1c;
323 Tj = VADD(T7, Ti);
324 TG = VADD(Tu, TF);
325 ST(&(ri[WS(rs, 4)]), VSUB(Tj, TG), ms, &(ri[0]));
326 ST(&(ri[0]), VADD(Tj, TG), ms, &(ri[0]));
327 {
328 V T15, T1a, T11, T14;
329 T15 = VADD(T12, T13);
330 T1a = VADD(T16, T19);
331 ST(&(ii[0]), VADD(T15, T1a), ms, &(ii[0]));
332 ST(&(ii[WS(rs, 4)]), VSUB(T1a, T15), ms, &(ii[0]));
333 T11 = VSUB(T7, Ti);
334 T14 = VSUB(T12, T13);
335 ST(&(ri[WS(rs, 6)]), VSUB(T11, T14), ms, &(ri[0]));
336 ST(&(ri[WS(rs, 2)]), VADD(T11, T14), ms, &(ri[0]));
337 }
338 T1b = VSUB(TF, Tu);
339 T1c = VSUB(T19, T16);
340 ST(&(ii[WS(rs, 2)]), VADD(T1b, T1c), ms, &(ii[0]));
341 ST(&(ii[WS(rs, 6)]), VSUB(T1c, T1b), ms, &(ii[0]));
342 {
343 V TX, T1g, T10, T1d, TY, TZ;
344 TX = VSUB(TH, TK);
345 T1g = VSUB(T1e, T1f);
346 TY = VSUB(TP, TM);
347 TZ = VADD(TR, TU);
348 T10 = VMUL(LDK(KP707106781), VSUB(TY, TZ));
349 T1d = VMUL(LDK(KP707106781), VADD(TY, TZ));
350 ST(&(ri[WS(rs, 7)]), VSUB(TX, T10), ms, &(ri[WS(rs, 1)]));
351 ST(&(ii[WS(rs, 5)]), VSUB(T1g, T1d), ms, &(ii[WS(rs, 1)]));
352 ST(&(ri[WS(rs, 3)]), VADD(TX, T10), ms, &(ri[WS(rs, 1)]));
353 ST(&(ii[WS(rs, 1)]), VADD(T1d, T1g), ms, &(ii[WS(rs, 1)]));
354 }
355 {
356 V TL, T1i, TW, T1h, TQ, TV;
357 TL = VADD(TH, TK);
358 T1i = VADD(T1f, T1e);
359 TQ = VADD(TM, TP);
360 TV = VSUB(TR, TU);
361 TW = VMUL(LDK(KP707106781), VADD(TQ, TV));
362 T1h = VMUL(LDK(KP707106781), VSUB(TV, TQ));
363 ST(&(ri[WS(rs, 5)]), VSUB(TL, TW), ms, &(ri[WS(rs, 1)]));
364 ST(&(ii[WS(rs, 7)]), VSUB(T1i, T1h), ms, &(ii[WS(rs, 1)]));
365 ST(&(ri[WS(rs, 1)]), VADD(TL, TW), ms, &(ri[WS(rs, 1)]));
366 ST(&(ii[WS(rs, 3)]), VADD(T1h, T1i), ms, &(ii[WS(rs, 1)]));
367 }
368 }
369 }
370 }
371 VLEAVE();
372 }
373
374 static const tw_instr twinstr[] = {
375 VTW(0, 1),
376 VTW(0, 2),
377 VTW(0, 3),
378 VTW(0, 4),
379 VTW(0, 5),
380 VTW(0, 6),
381 VTW(0, 7),
382 { TW_NEXT, (2 * VL), 0 }
383 };
384
385 static const ct_desc desc = { 8, XSIMD_STRING("t1sv_8"), twinstr, &GENUS, { 52, 18, 14, 0 }, 0, 0, 0 };
386
XSIMD(codelet_t1sv_8)387 void XSIMD(codelet_t1sv_8) (planner *p) {
388 X(kdft_dit_register) (p, t1sv_8, &desc);
389 }
390 #endif
391