Mercurial > hg > sv-dependency-builds

comparison src/fftw-3.3.3/rdft/scalar/r2cf/hc2cfdft2_8.c @ 10:37bf6b4a2645

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Add FFTW3
author Chris Cannam
date Wed, 20 Mar 2013 15:35:50 +0000
parents
children
comparison
equal deleted inserted replaced
9:c0fb53affa76 10:37bf6b4a2645
1 /*
2 * Copyright (c) 2003, 2007-11 Matteo Frigo
3 * Copyright (c) 2003, 2007-11 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 Sun Nov 25 07:40:50 EST 2012 */
23
24 #include "codelet-rdft.h"
25
26 #ifdef HAVE_FMA
27
28 /* Generated by: ../../../genfft/gen_hc2cdft.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -twiddle-log3 -precompute-twiddles -n 8 -dit -name hc2cfdft2_8 -include hc2cf.h */
29
30 /*
31 * This function contains 90 FP additions, 66 FP multiplications,
32 * (or, 60 additions, 36 multiplications, 30 fused multiply/add),
33 * 68 stack variables, 2 constants, and 32 memory accesses
34 */
35 #include "hc2cf.h"
36
37 static void hc2cfdft2_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 DK(KP500000000, +0.500000000000000000000000000000000000000000000);
41 {
42 INT m;
43 for (m = mb, W = W + ((mb - 1) * 6); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 6, MAKE_VOLATILE_STRIDE(32, rs)) {
44 E T1G, T1F, T1C, T1D, T1N, T1B, T1R, T1L;
45 {
46 E T1, T2, Th, Tj, T4, T3, Ti, Tp, T5;
47 T1 = W[0];
48 T2 = W[2];
49 Th = W[4];
50 Tj = W[5];
51 T4 = W[1];
52 T3 = T1 * T2;
53 Ti = T1 * Th;
54 Tp = T1 * Tj;
55 T5 = W[3];
56 {
57 E Tk, Tq, TI, T1a, T1u, TY, TF, TS, T1s, T1c, Tr, T1n, Tg, T16, Tn;
58 E T13, T1f, Ts, To, T1o;
59 {
60 E T6, Tw, Tc, TB, TQ, TM, TC, TR, Tz, TD, TA;
61 {
62 E TX, TV, TT, TU;
63 {
64 E TG, Tb, TH, TP, TL;
65 TG = Ip[0];
66 Tk = FMA(T4, Tj, Ti);
67 Tq = FNMS(T4, Th, Tp);
68 T6 = FMA(T4, T5, T3);
69 Tw = FNMS(T4, T5, T3);
70 Tb = T1 * T5;
71 TH = Im[0];
72 TT = Rm[0];
73 TP = T6 * Tj;
74 TL = T6 * Th;
75 Tc = FNMS(T4, T2, Tb);
76 TB = FMA(T4, T2, Tb);
77 TX = TG + TH;
78 TI = TG - TH;
79 TU = Rp[0];
80 TQ = FNMS(Tc, Th, TP);
81 TM = FMA(Tc, Tj, TL);
82 }
83 T1a = TU + TT;
84 TV = TT - TU;
85 {
86 E Tx, Ty, T1t, TW;
87 Tx = Ip[WS(rs, 2)];
88 Ty = Im[WS(rs, 2)];
89 T1t = T4 * TV;
90 TW = T1 * TV;
91 TC = Rp[WS(rs, 2)];
92 TR = Tx + Ty;
93 Tz = Tx - Ty;
94 T1u = FMA(T1, TX, T1t);
95 TY = FNMS(T4, TX, TW);
96 TD = Rm[WS(rs, 2)];
97 }
98 TA = Tw * Tz;
99 }
100 {
101 E Td, T9, T12, Te, Ta, T1m;
102 {
103 E T7, T8, TN, TE, TO, T1r, T1b;
104 T7 = Ip[WS(rs, 1)];
105 T8 = Im[WS(rs, 1)];
106 TN = TD - TC;
107 TE = TC + TD;
108 Td = Rp[WS(rs, 1)];
109 T9 = T7 - T8;
110 T12 = T7 + T8;
111 TO = TM * TN;
112 T1r = TQ * TN;
113 T1b = Tw * TE;
114 TF = FNMS(TB, TE, TA);
115 TS = FNMS(TQ, TR, TO);
116 T1s = FMA(TM, TR, T1r);
117 T1c = FMA(TB, Tz, T1b);
118 Te = Rm[WS(rs, 1)];
119 }
120 Ta = T6 * T9;
121 T1m = T2 * T12;
122 {
123 E Tl, T10, Tf, Tm, T11, T1e;
124 Tl = Ip[WS(rs, 3)];
125 T10 = Td - Te;
126 Tf = Td + Te;
127 Tm = Im[WS(rs, 3)];
128 Tr = Rp[WS(rs, 3)];
129 T11 = T2 * T10;
130 T1n = FNMS(T5, T10, T1m);
131 T1e = T6 * Tf;
132 Tg = FNMS(Tc, Tf, Ta);
133 T16 = Tl + Tm;
134 Tn = Tl - Tm;
135 T13 = FMA(T5, T12, T11);
136 T1f = FMA(Tc, T9, T1e);
137 Ts = Rm[WS(rs, 3)];
138 }
139 To = Tk * Tn;
140 T1o = Th * T16;
141 }
142 }
143 {
144 E T1z, T1K, T1y, T1k, T1J, T1A, T1x, T1j;
145 {
146 E T1w, TK, T1l, T19, T1d, T1i;
147 {
148 E TJ, T14, Tt, T1v, T1h;
149 T1z = TI - TF;
150 TJ = TF + TI;
151 T14 = Tr - Ts;
152 Tt = Tr + Ts;
153 T1v = T1s + T1u;
154 T1G = T1u - T1s;
155 {
156 E TZ, T1q, Tv, T18, T15;
157 T1F = TY - TS;
158 TZ = TS + TY;
159 T15 = Th * T14;
160 {
161 E T1p, T1g, Tu, T17;
162 T1p = FNMS(Tj, T14, T1o);
163 T1g = Tk * Tt;
164 Tu = FNMS(Tq, Tt, To);
165 T17 = FMA(Tj, T16, T15);
166 T1C = T1p - T1n;
167 T1q = T1n + T1p;
168 T1h = FMA(Tq, Tn, T1g);
169 T1K = Tg - Tu;
170 Tv = Tg + Tu;
171 T18 = T13 + T17;
172 T1D = T13 - T17;
173 }
174 T1w = T1q - T1v;
175 T1y = T1q + T1v;
176 TK = Tv + TJ;
177 T1l = TJ - Tv;
178 T1k = T18 + TZ;
179 T19 = TZ - T18;
180 }
181 T1J = T1a - T1c;
182 T1d = T1a + T1c;
183 T1i = T1f + T1h;
184 T1A = T1f - T1h;
185 }
186 Ip[0] = KP500000000 * (TK + T19);
187 Im[WS(rs, 3)] = KP500000000 * (T19 - TK);
188 Im[WS(rs, 1)] = KP500000000 * (T1w - T1l);
189 T1x = T1d + T1i;
190 T1j = T1d - T1i;
191 Ip[WS(rs, 2)] = KP500000000 * (T1l + T1w);
192 }
193 Rm[WS(rs, 3)] = KP500000000 * (T1x - T1y);
194 Rp[0] = KP500000000 * (T1x + T1y);
195 Rp[WS(rs, 2)] = KP500000000 * (T1j + T1k);
196 Rm[WS(rs, 1)] = KP500000000 * (T1j - T1k);
197 T1N = T1A + T1z;
198 T1B = T1z - T1A;
199 T1R = T1J + T1K;
200 T1L = T1J - T1K;
201 }
202 }
203 }
204 {
205 E T1E, T1O, T1H, T1P;
206 T1E = T1C + T1D;
207 T1O = T1C - T1D;
208 T1H = T1F - T1G;
209 T1P = T1F + T1G;
210 {
211 E T1S, T1Q, T1I, T1M;
212 T1S = T1O + T1P;
213 T1Q = T1O - T1P;
214 T1I = T1E + T1H;
215 T1M = T1H - T1E;
216 Im[0] = -(KP500000000 * (FNMS(KP707106781, T1Q, T1N)));
217 Ip[WS(rs, 3)] = KP500000000 * (FMA(KP707106781, T1Q, T1N));
218 Rp[WS(rs, 1)] = KP500000000 * (FMA(KP707106781, T1S, T1R));
219 Rm[WS(rs, 2)] = KP500000000 * (FNMS(KP707106781, T1S, T1R));
220 Rp[WS(rs, 3)] = KP500000000 * (FMA(KP707106781, T1M, T1L));
221 Rm[0] = KP500000000 * (FNMS(KP707106781, T1M, T1L));
222 Im[WS(rs, 2)] = -(KP500000000 * (FNMS(KP707106781, T1I, T1B)));
223 Ip[WS(rs, 1)] = KP500000000 * (FMA(KP707106781, T1I, T1B));
224 }
225 }
226 }
227 }
228 }
229
230 static const tw_instr twinstr[] = {
231 {TW_CEXP, 1, 1},
232 {TW_CEXP, 1, 3},
233 {TW_CEXP, 1, 7},
234 {TW_NEXT, 1, 0}
235 };
236
237 static const hc2c_desc desc = { 8, "hc2cfdft2_8", twinstr, &GENUS, {60, 36, 30, 0} };
238
239 void X(codelet_hc2cfdft2_8) (planner *p) {
240 X(khc2c_register) (p, hc2cfdft2_8, &desc, HC2C_VIA_DFT);
241 }
242 #else /* HAVE_FMA */
243
244 /* Generated by: ../../../genfft/gen_hc2cdft.native -compact -variables 4 -pipeline-latency 4 -twiddle-log3 -precompute-twiddles -n 8 -dit -name hc2cfdft2_8 -include hc2cf.h */
245
246 /*
247 * This function contains 90 FP additions, 56 FP multiplications,
248 * (or, 72 additions, 38 multiplications, 18 fused multiply/add),
249 * 51 stack variables, 2 constants, and 32 memory accesses
250 */
251 #include "hc2cf.h"
252
253 static void hc2cfdft2_8(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms)
254 {
255 DK(KP353553390, +0.353553390593273762200422181052424519642417969);
256 DK(KP500000000, +0.500000000000000000000000000000000000000000000);
257 {
258 INT m;
259 for (m = mb, W = W + ((mb - 1) * 6); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 6, MAKE_VOLATILE_STRIDE(32, rs)) {
260 E T1, T4, T2, T5, Tu, Ty, T7, Td, Ti, Tj, Tk, TP, To, TN;
261 {
262 E T3, Tc, T6, Tb;
263 T1 = W[0];
264 T4 = W[1];
265 T2 = W[2];
266 T5 = W[3];
267 T3 = T1 * T2;
268 Tc = T4 * T2;
269 T6 = T4 * T5;
270 Tb = T1 * T5;
271 Tu = T3 - T6;
272 Ty = Tb + Tc;
273 T7 = T3 + T6;
274 Td = Tb - Tc;
275 Ti = W[4];
276 Tj = W[5];
277 Tk = FMA(T1, Ti, T4 * Tj);
278 TP = FNMS(Td, Ti, T7 * Tj);
279 To = FNMS(T4, Ti, T1 * Tj);
280 TN = FMA(T7, Ti, Td * Tj);
281 }
282 {
283 E TF, T11, TC, T12, T1d, T1e, T1q, TM, TR, T1p, Th, Ts, T15, T14, T1a;
284 E T1b, T1m, TV, TY, T1n;
285 {
286 E TD, TE, TL, TI, TJ, TK, Tx, TQ, TB, TO;
287 TD = Ip[0];
288 TE = Im[0];
289 TL = TD + TE;
290 TI = Rm[0];
291 TJ = Rp[0];
292 TK = TI - TJ;
293 {
294 E Tv, Tw, Tz, TA;
295 Tv = Ip[WS(rs, 2)];
296 Tw = Im[WS(rs, 2)];
297 Tx = Tv - Tw;
298 TQ = Tv + Tw;
299 Tz = Rp[WS(rs, 2)];
300 TA = Rm[WS(rs, 2)];
301 TB = Tz + TA;
302 TO = Tz - TA;
303 }
304 TF = TD - TE;
305 T11 = TJ + TI;
306 TC = FNMS(Ty, TB, Tu * Tx);
307 T12 = FMA(Tu, TB, Ty * Tx);
308 T1d = FNMS(TP, TO, TN * TQ);
309 T1e = FMA(T4, TK, T1 * TL);
310 T1q = T1e - T1d;
311 TM = FNMS(T4, TL, T1 * TK);
312 TR = FMA(TN, TO, TP * TQ);
313 T1p = TR + TM;
314 }
315 {
316 E Ta, TU, Tg, TT, Tn, TX, Tr, TW;
317 {
318 E T8, T9, Te, Tf;
319 T8 = Ip[WS(rs, 1)];
320 T9 = Im[WS(rs, 1)];
321 Ta = T8 - T9;
322 TU = T8 + T9;
323 Te = Rp[WS(rs, 1)];
324 Tf = Rm[WS(rs, 1)];
325 Tg = Te + Tf;
326 TT = Te - Tf;
327 }
328 {
329 E Tl, Tm, Tp, Tq;
330 Tl = Ip[WS(rs, 3)];
331 Tm = Im[WS(rs, 3)];
332 Tn = Tl - Tm;
333 TX = Tl + Tm;
334 Tp = Rp[WS(rs, 3)];
335 Tq = Rm[WS(rs, 3)];
336 Tr = Tp + Tq;
337 TW = Tp - Tq;
338 }
339 Th = FNMS(Td, Tg, T7 * Ta);
340 Ts = FNMS(To, Tr, Tk * Tn);
341 T15 = FMA(Tk, Tr, To * Tn);
342 T14 = FMA(T7, Tg, Td * Ta);
343 T1a = FNMS(T5, TT, T2 * TU);
344 T1b = FNMS(Tj, TW, Ti * TX);
345 T1m = T1b - T1a;
346 TV = FMA(T2, TT, T5 * TU);
347 TY = FMA(Ti, TW, Tj * TX);
348 T1n = TV - TY;
349 }
350 {
351 E T1l, T1x, T1A, T1C, T1s, T1w, T1v, T1B;
352 {
353 E T1j, T1k, T1y, T1z;
354 T1j = TF - TC;
355 T1k = T14 - T15;
356 T1l = KP500000000 * (T1j - T1k);
357 T1x = KP500000000 * (T1k + T1j);
358 T1y = T1m - T1n;
359 T1z = T1p + T1q;
360 T1A = KP353553390 * (T1y - T1z);
361 T1C = KP353553390 * (T1y + T1z);
362 }
363 {
364 E T1o, T1r, T1t, T1u;
365 T1o = T1m + T1n;
366 T1r = T1p - T1q;
367 T1s = KP353553390 * (T1o + T1r);
368 T1w = KP353553390 * (T1r - T1o);
369 T1t = T11 - T12;
370 T1u = Th - Ts;
371 T1v = KP500000000 * (T1t - T1u);
372 T1B = KP500000000 * (T1t + T1u);
373 }
374 Ip[WS(rs, 1)] = T1l + T1s;
375 Rp[WS(rs, 1)] = T1B + T1C;
376 Im[WS(rs, 2)] = T1s - T1l;
377 Rm[WS(rs, 2)] = T1B - T1C;
378 Rm[0] = T1v - T1w;
379 Im[0] = T1A - T1x;
380 Rp[WS(rs, 3)] = T1v + T1w;
381 Ip[WS(rs, 3)] = T1x + T1A;
382 }
383 {
384 E TH, T19, T1g, T1i, T10, T18, T17, T1h;
385 {
386 E Tt, TG, T1c, T1f;
387 Tt = Th + Ts;
388 TG = TC + TF;
389 TH = Tt + TG;
390 T19 = TG - Tt;
391 T1c = T1a + T1b;
392 T1f = T1d + T1e;
393 T1g = T1c - T1f;
394 T1i = T1c + T1f;
395 }
396 {
397 E TS, TZ, T13, T16;
398 TS = TM - TR;
399 TZ = TV + TY;
400 T10 = TS - TZ;
401 T18 = TZ + TS;
402 T13 = T11 + T12;
403 T16 = T14 + T15;
404 T17 = T13 - T16;
405 T1h = T13 + T16;
406 }
407 Ip[0] = KP500000000 * (TH + T10);
408 Rp[0] = KP500000000 * (T1h + T1i);
409 Im[WS(rs, 3)] = KP500000000 * (T10 - TH);
410 Rm[WS(rs, 3)] = KP500000000 * (T1h - T1i);
411 Rm[WS(rs, 1)] = KP500000000 * (T17 - T18);
412 Im[WS(rs, 1)] = KP500000000 * (T1g - T19);
413 Rp[WS(rs, 2)] = KP500000000 * (T17 + T18);
414 Ip[WS(rs, 2)] = KP500000000 * (T19 + T1g);
415 }
416 }
417 }
418 }
419 }
420
421 static const tw_instr twinstr[] = {
422 {TW_CEXP, 1, 1},
423 {TW_CEXP, 1, 3},
424 {TW_CEXP, 1, 7},
425 {TW_NEXT, 1, 0}
426 };
427
428 static const hc2c_desc desc = { 8, "hc2cfdft2_8", twinstr, &GENUS, {72, 38, 18, 0} };
429
430 void X(codelet_hc2cfdft2_8) (planner *p) {
431 X(khc2c_register) (p, hc2cfdft2_8, &desc, HC2C_VIA_DFT);
432 }
433 #endif /* HAVE_FMA */