Mercurial > hg > sv-dependency-builds

comparison src/fftw-3.3.3/dft/scalar/codelets/q1_3.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 07:36:23 EST 2012 */
23
24 #include "codelet-dft.h"
25
26 #ifdef HAVE_FMA
27
28 /* Generated by: ../../../genfft/gen_twidsq.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -reload-twiddle -dif -n 3 -name q1_3 -include q.h */
29
30 /*
31 * This function contains 48 FP additions, 42 FP multiplications,
32 * (or, 18 additions, 12 multiplications, 30 fused multiply/add),
33 * 56 stack variables, 2 constants, and 36 memory accesses
34 */
35 #include "q.h"
36
37 static void q1_3(R *rio, R *iio, const R *W, stride rs, stride vs, 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 * 4); m < me; m = m + 1, rio = rio + ms, iio = iio + ms, W = W + 4, MAKE_VOLATILE_STRIDE(6, rs), MAKE_VOLATILE_STRIDE(0, vs)) {
44 E Tk, Tn, Tm, To, Tl;
45 {
46 E T1, Td, T4, Tg, Tp, T9, Te, T6, Tf, TB, TE, Ts, TZ, Tu, Tx;
47 E TC, TN, TO, TD, TV, T10, TP, Tq, Tr;
48 {
49 E T2, T3, T7, T8;
50 T1 = rio[0];
51 T2 = rio[WS(rs, 1)];
52 T3 = rio[WS(rs, 2)];
53 Td = iio[0];
54 T7 = iio[WS(rs, 1)];
55 T8 = iio[WS(rs, 2)];
56 T4 = T2 + T3;
57 Tg = T3 - T2;
58 Tp = rio[WS(vs, 1)];
59 T9 = T7 - T8;
60 Te = T7 + T8;
61 T6 = FNMS(KP500000000, T4, T1);
62 Tq = rio[WS(vs, 1) + WS(rs, 1)];
63 Tr = rio[WS(vs, 1) + WS(rs, 2)];
64 Tf = FNMS(KP500000000, Te, Td);
65 }
66 {
67 E Tv, Tw, TT, TU;
68 TB = iio[WS(vs, 1)];
69 Tv = iio[WS(vs, 1) + WS(rs, 1)];
70 TE = Tr - Tq;
71 Ts = Tq + Tr;
72 Tw = iio[WS(vs, 1) + WS(rs, 2)];
73 TZ = iio[WS(vs, 2)];
74 TT = iio[WS(vs, 2) + WS(rs, 1)];
75 Tu = FNMS(KP500000000, Ts, Tp);
76 Tx = Tv - Tw;
77 TC = Tv + Tw;
78 TU = iio[WS(vs, 2) + WS(rs, 2)];
79 TN = rio[WS(vs, 2)];
80 TO = rio[WS(vs, 2) + WS(rs, 1)];
81 TD = FNMS(KP500000000, TC, TB);
82 TV = TT - TU;
83 T10 = TT + TU;
84 TP = rio[WS(vs, 2) + WS(rs, 2)];
85 }
86 {
87 E T11, T12, TS, TQ;
88 rio[0] = T1 + T4;
89 iio[0] = Td + Te;
90 T11 = FNMS(KP500000000, T10, TZ);
91 T12 = TP - TO;
92 TQ = TO + TP;
93 rio[WS(rs, 1)] = Tp + Ts;
94 iio[WS(rs, 1)] = TB + TC;
95 iio[WS(rs, 2)] = TZ + T10;
96 TS = FNMS(KP500000000, TQ, TN);
97 rio[WS(rs, 2)] = TN + TQ;
98 {
99 E TW, T13, Ty, TI, TL, TF, TH, TK;
100 {
101 E Ta, Th, T5, Tc;
102 Tk = FNMS(KP866025403, T9, T6);
103 Ta = FMA(KP866025403, T9, T6);
104 Th = FMA(KP866025403, Tg, Tf);
105 Tn = FNMS(KP866025403, Tg, Tf);
106 T5 = W[0];
107 Tc = W[1];
108 {
109 E T16, T19, T18, T1a, T17, Ti, Tb, T15;
110 TW = FMA(KP866025403, TV, TS);
111 T16 = FNMS(KP866025403, TV, TS);
112 T19 = FNMS(KP866025403, T12, T11);
113 T13 = FMA(KP866025403, T12, T11);
114 Ti = T5 * Th;
115 Tb = T5 * Ta;
116 T15 = W[2];
117 T18 = W[3];
118 iio[WS(vs, 1)] = FNMS(Tc, Ta, Ti);
119 rio[WS(vs, 1)] = FMA(Tc, Th, Tb);
120 T1a = T15 * T19;
121 T17 = T15 * T16;
122 Ty = FMA(KP866025403, Tx, Tu);
123 TI = FNMS(KP866025403, Tx, Tu);
124 TL = FNMS(KP866025403, TE, TD);
125 TF = FMA(KP866025403, TE, TD);
126 iio[WS(vs, 2) + WS(rs, 2)] = FNMS(T18, T16, T1a);
127 rio[WS(vs, 2) + WS(rs, 2)] = FMA(T18, T19, T17);
128 TH = W[2];
129 TK = W[3];
130 }
131 }
132 {
133 E TA, TG, Tz, TM, TJ, Tt;
134 TM = TH * TL;
135 TJ = TH * TI;
136 Tt = W[0];
137 TA = W[1];
138 iio[WS(vs, 2) + WS(rs, 1)] = FNMS(TK, TI, TM);
139 rio[WS(vs, 2) + WS(rs, 1)] = FMA(TK, TL, TJ);
140 TG = Tt * TF;
141 Tz = Tt * Ty;
142 {
143 E TR, TY, T14, TX, Tj;
144 iio[WS(vs, 1) + WS(rs, 1)] = FNMS(TA, Ty, TG);
145 rio[WS(vs, 1) + WS(rs, 1)] = FMA(TA, TF, Tz);
146 TR = W[0];
147 TY = W[1];
148 T14 = TR * T13;
149 TX = TR * TW;
150 Tj = W[2];
151 Tm = W[3];
152 iio[WS(vs, 1) + WS(rs, 2)] = FNMS(TY, TW, T14);
153 rio[WS(vs, 1) + WS(rs, 2)] = FMA(TY, T13, TX);
154 To = Tj * Tn;
155 Tl = Tj * Tk;
156 }
157 }
158 }
159 }
160 }
161 iio[WS(vs, 2)] = FNMS(Tm, Tk, To);
162 rio[WS(vs, 2)] = FMA(Tm, Tn, Tl);
163 }
164 }
165 }
166
167 static const tw_instr twinstr[] = {
168 {TW_FULL, 0, 3},
169 {TW_NEXT, 1, 0}
170 };
171
172 static const ct_desc desc = { 3, "q1_3", twinstr, &GENUS, {18, 12, 30, 0}, 0, 0, 0 };
173
174 void X(codelet_q1_3) (planner *p) {
175 X(kdft_difsq_register) (p, q1_3, &desc);
176 }
177 #else /* HAVE_FMA */
178
179 /* Generated by: ../../../genfft/gen_twidsq.native -compact -variables 4 -pipeline-latency 4 -reload-twiddle -dif -n 3 -name q1_3 -include q.h */
180
181 /*
182 * This function contains 48 FP additions, 36 FP multiplications,
183 * (or, 30 additions, 18 multiplications, 18 fused multiply/add),
184 * 35 stack variables, 2 constants, and 36 memory accesses
185 */
186 #include "q.h"
187
188 static void q1_3(R *rio, R *iio, const R *W, stride rs, stride vs, INT mb, INT me, INT ms)
189 {
190 DK(KP866025403, +0.866025403784438646763723170752936183471402627);
191 DK(KP500000000, +0.500000000000000000000000000000000000000000000);
192 {
193 INT m;
194 for (m = mb, W = W + (mb * 4); m < me; m = m + 1, rio = rio + ms, iio = iio + ms, W = W + 4, MAKE_VOLATILE_STRIDE(6, rs), MAKE_VOLATILE_STRIDE(0, vs)) {
195 E T1, T4, T6, Tc, Td, Te, T9, Tf, Tl, To, Tq, Tw, Tx, Ty, Tt;
196 E Tz, TR, TS, TN, TT, TF, TI, TK, TQ;
197 {
198 E T2, T3, Tr, Ts;
199 T1 = rio[0];
200 T2 = rio[WS(rs, 1)];
201 T3 = rio[WS(rs, 2)];
202 T4 = T2 + T3;
203 T6 = FNMS(KP500000000, T4, T1);
204 Tc = KP866025403 * (T3 - T2);
205 {
206 E T7, T8, Tm, Tn;
207 Td = iio[0];
208 T7 = iio[WS(rs, 1)];
209 T8 = iio[WS(rs, 2)];
210 Te = T7 + T8;
211 T9 = KP866025403 * (T7 - T8);
212 Tf = FNMS(KP500000000, Te, Td);
213 Tl = rio[WS(vs, 1)];
214 Tm = rio[WS(vs, 1) + WS(rs, 1)];
215 Tn = rio[WS(vs, 1) + WS(rs, 2)];
216 To = Tm + Tn;
217 Tq = FNMS(KP500000000, To, Tl);
218 Tw = KP866025403 * (Tn - Tm);
219 }
220 Tx = iio[WS(vs, 1)];
221 Tr = iio[WS(vs, 1) + WS(rs, 1)];
222 Ts = iio[WS(vs, 1) + WS(rs, 2)];
223 Ty = Tr + Ts;
224 Tt = KP866025403 * (Tr - Ts);
225 Tz = FNMS(KP500000000, Ty, Tx);
226 {
227 E TL, TM, TG, TH;
228 TR = iio[WS(vs, 2)];
229 TL = iio[WS(vs, 2) + WS(rs, 1)];
230 TM = iio[WS(vs, 2) + WS(rs, 2)];
231 TS = TL + TM;
232 TN = KP866025403 * (TL - TM);
233 TT = FNMS(KP500000000, TS, TR);
234 TF = rio[WS(vs, 2)];
235 TG = rio[WS(vs, 2) + WS(rs, 1)];
236 TH = rio[WS(vs, 2) + WS(rs, 2)];
237 TI = TG + TH;
238 TK = FNMS(KP500000000, TI, TF);
239 TQ = KP866025403 * (TH - TG);
240 }
241 }
242 rio[0] = T1 + T4;
243 iio[0] = Td + Te;
244 rio[WS(rs, 1)] = Tl + To;
245 iio[WS(rs, 1)] = Tx + Ty;
246 iio[WS(rs, 2)] = TR + TS;
247 rio[WS(rs, 2)] = TF + TI;
248 {
249 E Ta, Tg, T5, Tb;
250 Ta = T6 + T9;
251 Tg = Tc + Tf;
252 T5 = W[0];
253 Tb = W[1];
254 rio[WS(vs, 1)] = FMA(T5, Ta, Tb * Tg);
255 iio[WS(vs, 1)] = FNMS(Tb, Ta, T5 * Tg);
256 }
257 {
258 E TW, TY, TV, TX;
259 TW = TK - TN;
260 TY = TT - TQ;
261 TV = W[2];
262 TX = W[3];
263 rio[WS(vs, 2) + WS(rs, 2)] = FMA(TV, TW, TX * TY);
264 iio[WS(vs, 2) + WS(rs, 2)] = FNMS(TX, TW, TV * TY);
265 }
266 {
267 E TC, TE, TB, TD;
268 TC = Tq - Tt;
269 TE = Tz - Tw;
270 TB = W[2];
271 TD = W[3];
272 rio[WS(vs, 2) + WS(rs, 1)] = FMA(TB, TC, TD * TE);
273 iio[WS(vs, 2) + WS(rs, 1)] = FNMS(TD, TC, TB * TE);
274 }
275 {
276 E Tu, TA, Tp, Tv;
277 Tu = Tq + Tt;
278 TA = Tw + Tz;
279 Tp = W[0];
280 Tv = W[1];
281 rio[WS(vs, 1) + WS(rs, 1)] = FMA(Tp, Tu, Tv * TA);
282 iio[WS(vs, 1) + WS(rs, 1)] = FNMS(Tv, Tu, Tp * TA);
283 }
284 {
285 E TO, TU, TJ, TP;
286 TO = TK + TN;
287 TU = TQ + TT;
288 TJ = W[0];
289 TP = W[1];
290 rio[WS(vs, 1) + WS(rs, 2)] = FMA(TJ, TO, TP * TU);
291 iio[WS(vs, 1) + WS(rs, 2)] = FNMS(TP, TO, TJ * TU);
292 }
293 {
294 E Ti, Tk, Th, Tj;
295 Ti = T6 - T9;
296 Tk = Tf - Tc;
297 Th = W[2];
298 Tj = W[3];
299 rio[WS(vs, 2)] = FMA(Th, Ti, Tj * Tk);
300 iio[WS(vs, 2)] = FNMS(Tj, Ti, Th * Tk);
301 }
302 }
303 }
304 }
305
306 static const tw_instr twinstr[] = {
307 {TW_FULL, 0, 3},
308 {TW_NEXT, 1, 0}
309 };
310
311 static const ct_desc desc = { 3, "q1_3", twinstr, &GENUS, {30, 18, 18, 0}, 0, 0, 0 };
312
313 void X(codelet_q1_3) (planner *p) {
314 X(kdft_difsq_register) (p, q1_3, &desc);
315 }
316 #endif /* HAVE_FMA */