Mercurial > hg > sv-dependency-builds

comparison src/fftw-3.3.8/dft/simd/common/t3bv_16.c @ 167:bd3cc4d1df30

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Add FFTW 3.3.8 source, and a Linux build
author Chris Cannam <cannam@all-day-breakfast.com>
date Tue, 19 Nov 2019 14:52:55 +0000
parents
children
comparison
equal deleted inserted replaced
166:cbd6d7e562c7 167:bd3cc4d1df30
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 May 24 08:06:06 EDT 2018 */
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_c.native -fma -simd -compact -variables 4 -pipeline-latency 8 -twiddle-log3 -precompute-twiddles -no-generate-bytw -n 16 -name t3bv_16 -include dft/simd/t3b.h -sign 1 */
29
30 /*
31 * This function contains 98 FP additions, 86 FP multiplications,
32 * (or, 64 additions, 52 multiplications, 34 fused multiply/add),
33 * 51 stack variables, 3 constants, and 32 memory accesses
34 */
35 #include "dft/simd/t3b.h"
36
37 static void t3bv_16(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms)
38 {
39 DVK(KP923879532, +0.923879532511286756128183189396788286822416626);
40 DVK(KP707106781, +0.707106781186547524400844362104849039284835938);
41 DVK(KP414213562, +0.414213562373095048801688724209698078569671875);
42 {
43 INT m;
44 R *x;
45 x = ii;
46 for (m = mb, W = W + (mb * ((TWVL / VL) * 8)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 8), MAKE_VOLATILE_STRIDE(16, rs)) {
47 V T2, T8, T9, Tx, Tu, TR, T3, T4, TN, TU, Tc, Tm, Ty, TE, Tp;
48 T2 = LDW(&(W[0]));
49 T8 = LDW(&(W[TWVL * 2]));
50 T9 = VZMUL(T2, T8);
51 Tx = VZMULJ(T2, T8);
52 Tu = LDW(&(W[TWVL * 6]));
53 TR = VZMULJ(T2, Tu);
54 T3 = LDW(&(W[TWVL * 4]));
55 T4 = VZMULJ(T2, T3);
56 TN = VZMUL(T2, T3);
57 TU = VZMULJ(T8, T3);
58 Tc = VZMUL(T8, T3);
59 Tm = VZMULJ(T9, T3);
60 Ty = VZMULJ(Tx, T3);
61 TE = VZMUL(Tx, T3);
62 Tp = VZMUL(T9, T3);
63 {
64 V T7, T1b, Tf, T1o, TQ, TX, T1e, T1p, Tl, Ts, Tt, T1i, T1r, TB, TH;
65 V TI, T1l, T1s, T1, T6, T5;
66 T1 = LD(&(x[0]), ms, &(x[0]));
67 T5 = LD(&(x[WS(rs, 8)]), ms, &(x[0]));
68 T6 = VZMUL(T4, T5);
69 T7 = VADD(T1, T6);
70 T1b = VSUB(T1, T6);
71 {
72 V Tb, Te, Ta, Td;
73 Ta = LD(&(x[WS(rs, 4)]), ms, &(x[0]));
74 Tb = VZMUL(T9, Ta);
75 Td = LD(&(x[WS(rs, 12)]), ms, &(x[0]));
76 Te = VZMUL(Tc, Td);
77 Tf = VADD(Tb, Te);
78 T1o = VSUB(Tb, Te);
79 }
80 {
81 V TM, TW, TP, TT, T1c, T1d;
82 {
83 V TL, TV, TO, TS;
84 TL = LD(&(x[WS(rs, 2)]), ms, &(x[0]));
85 TM = VZMUL(Tx, TL);
86 TV = LD(&(x[WS(rs, 6)]), ms, &(x[0]));
87 TW = VZMUL(TU, TV);
88 TO = LD(&(x[WS(rs, 10)]), ms, &(x[0]));
89 TP = VZMUL(TN, TO);
90 TS = LD(&(x[WS(rs, 14)]), ms, &(x[0]));
91 TT = VZMUL(TR, TS);
92 }
93 TQ = VADD(TM, TP);
94 TX = VADD(TT, TW);
95 T1c = VSUB(TM, TP);
96 T1d = VSUB(TT, TW);
97 T1e = VADD(T1c, T1d);
98 T1p = VSUB(T1c, T1d);
99 }
100 {
101 V Ti, Tr, Tk, To, T1g, T1h;
102 {
103 V Th, Tq, Tj, Tn;
104 Th = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)]));
105 Ti = VZMUL(T2, Th);
106 Tq = LD(&(x[WS(rs, 13)]), ms, &(x[WS(rs, 1)]));
107 Tr = VZMUL(Tp, Tq);
108 Tj = LD(&(x[WS(rs, 9)]), ms, &(x[WS(rs, 1)]));
109 Tk = VZMUL(T3, Tj);
110 Tn = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)]));
111 To = VZMUL(Tm, Tn);
112 }
113 Tl = VADD(Ti, Tk);
114 Ts = VADD(To, Tr);
115 Tt = VSUB(Tl, Ts);
116 T1g = VSUB(Ti, Tk);
117 T1h = VSUB(To, Tr);
118 T1i = VFNMS(LDK(KP414213562), T1h, T1g);
119 T1r = VFMA(LDK(KP414213562), T1g, T1h);
120 }
121 {
122 V Tw, TG, TA, TD, T1j, T1k;
123 {
124 V Tv, TF, Tz, TC;
125 Tv = LD(&(x[WS(rs, 15)]), ms, &(x[WS(rs, 1)]));
126 Tw = VZMUL(Tu, Tv);
127 TF = LD(&(x[WS(rs, 11)]), ms, &(x[WS(rs, 1)]));
128 TG = VZMUL(TE, TF);
129 Tz = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)]));
130 TA = VZMUL(Ty, Tz);
131 TC = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)]));
132 TD = VZMUL(T8, TC);
133 }
134 TB = VADD(Tw, TA);
135 TH = VADD(TD, TG);
136 TI = VSUB(TB, TH);
137 T1j = VSUB(Tw, TA);
138 T1k = VSUB(TG, TD);
139 T1l = VFNMS(LDK(KP414213562), T1k, T1j);
140 T1s = VFMA(LDK(KP414213562), T1j, T1k);
141 }
142 {
143 V TK, T11, T10, T12;
144 {
145 V Tg, TJ, TY, TZ;
146 Tg = VSUB(T7, Tf);
147 TJ = VADD(Tt, TI);
148 TK = VFNMS(LDK(KP707106781), TJ, Tg);
149 T11 = VFMA(LDK(KP707106781), TJ, Tg);
150 TY = VSUB(TQ, TX);
151 TZ = VSUB(Tt, TI);
152 T10 = VFNMS(LDK(KP707106781), TZ, TY);
153 T12 = VFMA(LDK(KP707106781), TZ, TY);
154 }
155 ST(&(x[WS(rs, 6)]), VFNMSI(T10, TK), ms, &(x[0]));
156 ST(&(x[WS(rs, 14)]), VFNMSI(T12, T11), ms, &(x[0]));
157 ST(&(x[WS(rs, 10)]), VFMAI(T10, TK), ms, &(x[0]));
158 ST(&(x[WS(rs, 2)]), VFMAI(T12, T11), ms, &(x[0]));
159 }
160 {
161 V T1z, T1D, T1C, T1E;
162 {
163 V T1x, T1y, T1A, T1B;
164 T1x = VFNMS(LDK(KP707106781), T1e, T1b);
165 T1y = VADD(T1r, T1s);
166 T1z = VFNMS(LDK(KP923879532), T1y, T1x);
167 T1D = VFMA(LDK(KP923879532), T1y, T1x);
168 T1A = VFNMS(LDK(KP707106781), T1p, T1o);
169 T1B = VSUB(T1i, T1l);
170 T1C = VFMA(LDK(KP923879532), T1B, T1A);
171 T1E = VFNMS(LDK(KP923879532), T1B, T1A);
172 }
173 ST(&(x[WS(rs, 5)]), VFMAI(T1C, T1z), ms, &(x[WS(rs, 1)]));
174 ST(&(x[WS(rs, 13)]), VFMAI(T1E, T1D), ms, &(x[WS(rs, 1)]));
175 ST(&(x[WS(rs, 11)]), VFNMSI(T1C, T1z), ms, &(x[WS(rs, 1)]));
176 ST(&(x[WS(rs, 3)]), VFNMSI(T1E, T1D), ms, &(x[WS(rs, 1)]));
177 }
178 {
179 V T15, T19, T18, T1a;
180 {
181 V T13, T14, T16, T17;
182 T13 = VADD(T7, Tf);
183 T14 = VADD(TQ, TX);
184 T15 = VSUB(T13, T14);
185 T19 = VADD(T13, T14);
186 T16 = VADD(Tl, Ts);
187 T17 = VADD(TB, TH);
188 T18 = VSUB(T16, T17);
189 T1a = VADD(T16, T17);
190 }
191 ST(&(x[WS(rs, 12)]), VFNMSI(T18, T15), ms, &(x[0]));
192 ST(&(x[0]), VADD(T19, T1a), ms, &(x[0]));
193 ST(&(x[WS(rs, 4)]), VFMAI(T18, T15), ms, &(x[0]));
194 ST(&(x[WS(rs, 8)]), VSUB(T19, T1a), ms, &(x[0]));
195 }
196 {
197 V T1n, T1v, T1u, T1w;
198 {
199 V T1f, T1m, T1q, T1t;
200 T1f = VFMA(LDK(KP707106781), T1e, T1b);
201 T1m = VADD(T1i, T1l);
202 T1n = VFNMS(LDK(KP923879532), T1m, T1f);
203 T1v = VFMA(LDK(KP923879532), T1m, T1f);
204 T1q = VFMA(LDK(KP707106781), T1p, T1o);
205 T1t = VSUB(T1r, T1s);
206 T1u = VFNMS(LDK(KP923879532), T1t, T1q);
207 T1w = VFMA(LDK(KP923879532), T1t, T1q);
208 }
209 ST(&(x[WS(rs, 7)]), VFNMSI(T1u, T1n), ms, &(x[WS(rs, 1)]));
210 ST(&(x[WS(rs, 1)]), VFMAI(T1w, T1v), ms, &(x[WS(rs, 1)]));
211 ST(&(x[WS(rs, 9)]), VFMAI(T1u, T1n), ms, &(x[WS(rs, 1)]));
212 ST(&(x[WS(rs, 15)]), VFNMSI(T1w, T1v), ms, &(x[WS(rs, 1)]));
213 }
214 }
215 }
216 }
217 VLEAVE();
218 }
219
220 static const tw_instr twinstr[] = {
221 VTW(0, 1),
222 VTW(0, 3),
223 VTW(0, 9),
224 VTW(0, 15),
225 {TW_NEXT, VL, 0}
226 };
227
228 static const ct_desc desc = { 16, XSIMD_STRING("t3bv_16"), twinstr, &GENUS, {64, 52, 34, 0}, 0, 0, 0 };
229
230 void XSIMD(codelet_t3bv_16) (planner *p) {
231 X(kdft_dit_register) (p, t3bv_16, &desc);
232 }
233 #else
234
235 /* Generated by: ../../../genfft/gen_twiddle_c.native -simd -compact -variables 4 -pipeline-latency 8 -twiddle-log3 -precompute-twiddles -no-generate-bytw -n 16 -name t3bv_16 -include dft/simd/t3b.h -sign 1 */
236
237 /*
238 * This function contains 98 FP additions, 64 FP multiplications,
239 * (or, 94 additions, 60 multiplications, 4 fused multiply/add),
240 * 51 stack variables, 3 constants, and 32 memory accesses
241 */
242 #include "dft/simd/t3b.h"
243
244 static void t3bv_16(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms)
245 {
246 DVK(KP382683432, +0.382683432365089771728459984030398866761344562);
247 DVK(KP923879532, +0.923879532511286756128183189396788286822416626);
248 DVK(KP707106781, +0.707106781186547524400844362104849039284835938);
249 {
250 INT m;
251 R *x;
252 x = ii;
253 for (m = mb, W = W + (mb * ((TWVL / VL) * 8)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 8), MAKE_VOLATILE_STRIDE(16, rs)) {
254 V T1, T8, T9, Tl, Ti, TE, T4, Ta, TO, TV, Td, Tm, TA, TH, Ts;
255 T1 = LDW(&(W[0]));
256 T8 = LDW(&(W[TWVL * 2]));
257 T9 = VZMUL(T1, T8);
258 Tl = VZMULJ(T1, T8);
259 Ti = LDW(&(W[TWVL * 6]));
260 TE = VZMULJ(T1, Ti);
261 T4 = LDW(&(W[TWVL * 4]));
262 Ta = VZMULJ(T9, T4);
263 TO = VZMUL(T8, T4);
264 TV = VZMULJ(T1, T4);
265 Td = VZMUL(T9, T4);
266 Tm = VZMULJ(Tl, T4);
267 TA = VZMUL(T1, T4);
268 TH = VZMULJ(T8, T4);
269 Ts = VZMUL(Tl, T4);
270 {
271 V TY, T1q, TR, T1r, T1m, T1n, TL, TZ, T1f, T1g, T1h, Th, T11, T1i, T1j;
272 V T1k, Tw, T12, TU, TX, TW;
273 TU = LD(&(x[0]), ms, &(x[0]));
274 TW = LD(&(x[WS(rs, 8)]), ms, &(x[0]));
275 TX = VZMUL(TV, TW);
276 TY = VSUB(TU, TX);
277 T1q = VADD(TU, TX);
278 {
279 V TN, TQ, TM, TP;
280 TM = LD(&(x[WS(rs, 4)]), ms, &(x[0]));
281 TN = VZMUL(T9, TM);
282 TP = LD(&(x[WS(rs, 12)]), ms, &(x[0]));
283 TQ = VZMUL(TO, TP);
284 TR = VSUB(TN, TQ);
285 T1r = VADD(TN, TQ);
286 }
287 {
288 V Tz, TJ, TC, TG, TD, TK;
289 {
290 V Ty, TI, TB, TF;
291 Ty = LD(&(x[WS(rs, 2)]), ms, &(x[0]));
292 Tz = VZMUL(Tl, Ty);
293 TI = LD(&(x[WS(rs, 6)]), ms, &(x[0]));
294 TJ = VZMUL(TH, TI);
295 TB = LD(&(x[WS(rs, 10)]), ms, &(x[0]));
296 TC = VZMUL(TA, TB);
297 TF = LD(&(x[WS(rs, 14)]), ms, &(x[0]));
298 TG = VZMUL(TE, TF);
299 }
300 T1m = VADD(Tz, TC);
301 T1n = VADD(TG, TJ);
302 TD = VSUB(Tz, TC);
303 TK = VSUB(TG, TJ);
304 TL = VMUL(LDK(KP707106781), VSUB(TD, TK));
305 TZ = VMUL(LDK(KP707106781), VADD(TD, TK));
306 }
307 {
308 V T3, Tf, T6, Tc, T7, Tg;
309 {
310 V T2, Te, T5, Tb;
311 T2 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)]));
312 T3 = VZMUL(T1, T2);
313 Te = LD(&(x[WS(rs, 13)]), ms, &(x[WS(rs, 1)]));
314 Tf = VZMUL(Td, Te);
315 T5 = LD(&(x[WS(rs, 9)]), ms, &(x[WS(rs, 1)]));
316 T6 = VZMUL(T4, T5);
317 Tb = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)]));
318 Tc = VZMUL(Ta, Tb);
319 }
320 T1f = VADD(T3, T6);
321 T1g = VADD(Tc, Tf);
322 T1h = VSUB(T1f, T1g);
323 T7 = VSUB(T3, T6);
324 Tg = VSUB(Tc, Tf);
325 Th = VFNMS(LDK(KP382683432), Tg, VMUL(LDK(KP923879532), T7));
326 T11 = VFMA(LDK(KP382683432), T7, VMUL(LDK(KP923879532), Tg));
327 }
328 {
329 V Tk, Tu, To, Tr, Tp, Tv;
330 {
331 V Tj, Tt, Tn, Tq;
332 Tj = LD(&(x[WS(rs, 15)]), ms, &(x[WS(rs, 1)]));
333 Tk = VZMUL(Ti, Tj);
334 Tt = LD(&(x[WS(rs, 11)]), ms, &(x[WS(rs, 1)]));
335 Tu = VZMUL(Ts, Tt);
336 Tn = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)]));
337 To = VZMUL(Tm, Tn);
338 Tq = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)]));
339 Tr = VZMUL(T8, Tq);
340 }
341 T1i = VADD(Tk, To);
342 T1j = VADD(Tr, Tu);
343 T1k = VSUB(T1i, T1j);
344 Tp = VSUB(Tk, To);
345 Tv = VSUB(Tr, Tu);
346 Tw = VFMA(LDK(KP923879532), Tp, VMUL(LDK(KP382683432), Tv));
347 T12 = VFNMS(LDK(KP382683432), Tp, VMUL(LDK(KP923879532), Tv));
348 }
349 {
350 V T1p, T1v, T1u, T1w;
351 {
352 V T1l, T1o, T1s, T1t;
353 T1l = VMUL(LDK(KP707106781), VSUB(T1h, T1k));
354 T1o = VSUB(T1m, T1n);
355 T1p = VBYI(VSUB(T1l, T1o));
356 T1v = VBYI(VADD(T1o, T1l));
357 T1s = VSUB(T1q, T1r);
358 T1t = VMUL(LDK(KP707106781), VADD(T1h, T1k));
359 T1u = VSUB(T1s, T1t);
360 T1w = VADD(T1s, T1t);
361 }
362 ST(&(x[WS(rs, 6)]), VADD(T1p, T1u), ms, &(x[0]));
363 ST(&(x[WS(rs, 14)]), VSUB(T1w, T1v), ms, &(x[0]));
364 ST(&(x[WS(rs, 10)]), VSUB(T1u, T1p), ms, &(x[0]));
365 ST(&(x[WS(rs, 2)]), VADD(T1v, T1w), ms, &(x[0]));
366 }
367 {
368 V T1z, T1D, T1C, T1E;
369 {
370 V T1x, T1y, T1A, T1B;
371 T1x = VADD(T1q, T1r);
372 T1y = VADD(T1m, T1n);
373 T1z = VSUB(T1x, T1y);
374 T1D = VADD(T1x, T1y);
375 T1A = VADD(T1f, T1g);
376 T1B = VADD(T1i, T1j);
377 T1C = VBYI(VSUB(T1A, T1B));
378 T1E = VADD(T1A, T1B);
379 }
380 ST(&(x[WS(rs, 12)]), VSUB(T1z, T1C), ms, &(x[0]));
381 ST(&(x[0]), VADD(T1D, T1E), ms, &(x[0]));
382 ST(&(x[WS(rs, 4)]), VADD(T1z, T1C), ms, &(x[0]));
383 ST(&(x[WS(rs, 8)]), VSUB(T1D, T1E), ms, &(x[0]));
384 }
385 {
386 V TT, T15, T14, T16;
387 {
388 V Tx, TS, T10, T13;
389 Tx = VSUB(Th, Tw);
390 TS = VSUB(TL, TR);
391 TT = VBYI(VSUB(Tx, TS));
392 T15 = VBYI(VADD(TS, Tx));
393 T10 = VSUB(TY, TZ);
394 T13 = VSUB(T11, T12);
395 T14 = VSUB(T10, T13);
396 T16 = VADD(T10, T13);
397 }
398 ST(&(x[WS(rs, 5)]), VADD(TT, T14), ms, &(x[WS(rs, 1)]));
399 ST(&(x[WS(rs, 13)]), VSUB(T16, T15), ms, &(x[WS(rs, 1)]));
400 ST(&(x[WS(rs, 11)]), VSUB(T14, TT), ms, &(x[WS(rs, 1)]));
401 ST(&(x[WS(rs, 3)]), VADD(T15, T16), ms, &(x[WS(rs, 1)]));
402 }
403 {
404 V T19, T1d, T1c, T1e;
405 {
406 V T17, T18, T1a, T1b;
407 T17 = VADD(TY, TZ);
408 T18 = VADD(Th, Tw);
409 T19 = VADD(T17, T18);
410 T1d = VSUB(T17, T18);
411 T1a = VADD(TR, TL);
412 T1b = VADD(T11, T12);
413 T1c = VBYI(VADD(T1a, T1b));
414 T1e = VBYI(VSUB(T1b, T1a));
415 }
416 ST(&(x[WS(rs, 15)]), VSUB(T19, T1c), ms, &(x[WS(rs, 1)]));
417 ST(&(x[WS(rs, 7)]), VADD(T1d, T1e), ms, &(x[WS(rs, 1)]));
418 ST(&(x[WS(rs, 1)]), VADD(T19, T1c), ms, &(x[WS(rs, 1)]));
419 ST(&(x[WS(rs, 9)]), VSUB(T1d, T1e), ms, &(x[WS(rs, 1)]));
420 }
421 }
422 }
423 }
424 VLEAVE();
425 }
426
427 static const tw_instr twinstr[] = {
428 VTW(0, 1),
429 VTW(0, 3),
430 VTW(0, 9),
431 VTW(0, 15),
432 {TW_NEXT, VL, 0}
433 };
434
435 static const ct_desc desc = { 16, XSIMD_STRING("t3bv_16"), twinstr, &GENUS, {94, 60, 4, 0}, 0, 0, 0 };
436
437 void XSIMD(codelet_t3bv_16) (planner *p) {
438 X(kdft_dit_register) (p, t3bv_16, &desc);
439 }
440 #endif