Mercurial > hg > sv-dependency-builds

annotate src/fftw-3.3.8/dft/scalar/codelets/n1_5.c @ 84:08ae793730bd

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Add null config files
author Chris Cannam
date Mon, 02 Mar 2020 14:03:47 +0000
parents d0c2a83c1364
children
rev   line source
Chris@82 1 /*
Chris@82 2 * Copyright (c) 2003, 2007-14 Matteo Frigo
Chris@82 3 * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
Chris@82 4 *
Chris@82 5 * This program is free software; you can redistribute it and/or modify
Chris@82 6 * it under the terms of the GNU General Public License as published by
Chris@82 7 * the Free Software Foundation; either version 2 of the License, or
Chris@82 8 * (at your option) any later version.
Chris@82 9 *
Chris@82 10 * This program is distributed in the hope that it will be useful,
Chris@82 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
Chris@82 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
Chris@82 13 * GNU General Public License for more details.
Chris@82 14 *
Chris@82 15 * You should have received a copy of the GNU General Public License
Chris@82 16 * along with this program; if not, write to the Free Software
Chris@82 17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
Chris@82 18 *
Chris@82 19 */
Chris@82 20
Chris@82 21 /* This file was automatically generated --- DO NOT EDIT */
Chris@82 22 /* Generated on Thu May 24 08:04:10 EDT 2018 */
Chris@82 23
Chris@82 24 #include "dft/codelet-dft.h"
Chris@82 25
Chris@82 26 #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA)
Chris@82 27
Chris@82 28 /* Generated by: ../../../genfft/gen_notw.native -fma -compact -variables 4 -pipeline-latency 4 -n 5 -name n1_5 -include dft/scalar/n.h */
Chris@82 29
Chris@82 30 /*
Chris@82 31 * This function contains 32 FP additions, 18 FP multiplications,
Chris@82 32 * (or, 14 additions, 0 multiplications, 18 fused multiply/add),
Chris@82 33 * 21 stack variables, 4 constants, and 20 memory accesses
Chris@82 34 */
Chris@82 35 #include "dft/scalar/n.h"
Chris@82 36
Chris@82 37 static void n1_5(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs)
Chris@82 38 {
Chris@82 39 DK(KP951056516, +0.951056516295153572116439333379382143405698634);
Chris@82 40 DK(KP559016994, +0.559016994374947424102293417182819058860154590);
Chris@82 41 DK(KP250000000, +0.250000000000000000000000000000000000000000000);
Chris@82 42 DK(KP618033988, +0.618033988749894848204586834365638117720309180);
Chris@82 43 {
Chris@82 44 INT i;
Chris@82 45 for (i = v; i > 0; i = i - 1, ri = ri + ivs, ii = ii + ivs, ro = ro + ovs, io = io + ovs, MAKE_VOLATILE_STRIDE(20, is), MAKE_VOLATILE_STRIDE(20, os)) {
Chris@82 46 E T1, Tl, T8, Tt, Ta, Ts, Te, Tq, Th, To;
Chris@82 47 T1 = ri[0];
Chris@82 48 Tl = ii[0];
Chris@82 49 {
Chris@82 50 E T2, T3, T4, T5, T6, T7;
Chris@82 51 T2 = ri[WS(is, 1)];
Chris@82 52 T3 = ri[WS(is, 4)];
Chris@82 53 T4 = T2 + T3;
Chris@82 54 T5 = ri[WS(is, 2)];
Chris@82 55 T6 = ri[WS(is, 3)];
Chris@82 56 T7 = T5 + T6;
Chris@82 57 T8 = T4 + T7;
Chris@82 58 Tt = T5 - T6;
Chris@82 59 Ta = T4 - T7;
Chris@82 60 Ts = T2 - T3;
Chris@82 61 }
Chris@82 62 {
Chris@82 63 E Tc, Td, Tm, Tf, Tg, Tn;
Chris@82 64 Tc = ii[WS(is, 1)];
Chris@82 65 Td = ii[WS(is, 4)];
Chris@82 66 Tm = Tc + Td;
Chris@82 67 Tf = ii[WS(is, 2)];
Chris@82 68 Tg = ii[WS(is, 3)];
Chris@82 69 Tn = Tf + Tg;
Chris@82 70 Te = Tc - Td;
Chris@82 71 Tq = Tm - Tn;
Chris@82 72 Th = Tf - Tg;
Chris@82 73 To = Tm + Tn;
Chris@82 74 }
Chris@82 75 ro[0] = T1 + T8;
Chris@82 76 io[0] = Tl + To;
Chris@82 77 {
Chris@82 78 E Ti, Tk, Tb, Tj, T9;
Chris@82 79 Ti = FMA(KP618033988, Th, Te);
Chris@82 80 Tk = FNMS(KP618033988, Te, Th);
Chris@82 81 T9 = FNMS(KP250000000, T8, T1);
Chris@82 82 Tb = FMA(KP559016994, Ta, T9);
Chris@82 83 Tj = FNMS(KP559016994, Ta, T9);
Chris@82 84 ro[WS(os, 4)] = FNMS(KP951056516, Ti, Tb);
Chris@82 85 ro[WS(os, 3)] = FMA(KP951056516, Tk, Tj);
Chris@82 86 ro[WS(os, 1)] = FMA(KP951056516, Ti, Tb);
Chris@82 87 ro[WS(os, 2)] = FNMS(KP951056516, Tk, Tj);
Chris@82 88 }
Chris@82 89 {
Chris@82 90 E Tu, Tw, Tr, Tv, Tp;
Chris@82 91 Tu = FMA(KP618033988, Tt, Ts);
Chris@82 92 Tw = FNMS(KP618033988, Ts, Tt);
Chris@82 93 Tp = FNMS(KP250000000, To, Tl);
Chris@82 94 Tr = FMA(KP559016994, Tq, Tp);
Chris@82 95 Tv = FNMS(KP559016994, Tq, Tp);
Chris@82 96 io[WS(os, 1)] = FNMS(KP951056516, Tu, Tr);
Chris@82 97 io[WS(os, 3)] = FNMS(KP951056516, Tw, Tv);
Chris@82 98 io[WS(os, 4)] = FMA(KP951056516, Tu, Tr);
Chris@82 99 io[WS(os, 2)] = FMA(KP951056516, Tw, Tv);
Chris@82 100 }
Chris@82 101 }
Chris@82 102 }
Chris@82 103 }
Chris@82 104
Chris@82 105 static const kdft_desc desc = { 5, "n1_5", {14, 0, 18, 0}, &GENUS, 0, 0, 0, 0 };
Chris@82 106
Chris@82 107 void X(codelet_n1_5) (planner *p) {
Chris@82 108 X(kdft_register) (p, n1_5, &desc);
Chris@82 109 }
Chris@82 110
Chris@82 111 #else
Chris@82 112
Chris@82 113 /* Generated by: ../../../genfft/gen_notw.native -compact -variables 4 -pipeline-latency 4 -n 5 -name n1_5 -include dft/scalar/n.h */
Chris@82 114
Chris@82 115 /*
Chris@82 116 * This function contains 32 FP additions, 12 FP multiplications,
Chris@82 117 * (or, 26 additions, 6 multiplications, 6 fused multiply/add),
Chris@82 118 * 21 stack variables, 4 constants, and 20 memory accesses
Chris@82 119 */
Chris@82 120 #include "dft/scalar/n.h"
Chris@82 121
Chris@82 122 static void n1_5(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs)
Chris@82 123 {
Chris@82 124 DK(KP250000000, +0.250000000000000000000000000000000000000000000);
Chris@82 125 DK(KP587785252, +0.587785252292473129168705954639072768597652438);
Chris@82 126 DK(KP951056516, +0.951056516295153572116439333379382143405698634);
Chris@82 127 DK(KP559016994, +0.559016994374947424102293417182819058860154590);
Chris@82 128 {
Chris@82 129 INT i;
Chris@82 130 for (i = v; i > 0; i = i - 1, ri = ri + ivs, ii = ii + ivs, ro = ro + ovs, io = io + ovs, MAKE_VOLATILE_STRIDE(20, is), MAKE_VOLATILE_STRIDE(20, os)) {
Chris@82 131 E T1, To, T8, Tt, T9, Ts, Te, Tp, Th, Tn;
Chris@82 132 T1 = ri[0];
Chris@82 133 To = ii[0];
Chris@82 134 {
Chris@82 135 E T2, T3, T4, T5, T6, T7;
Chris@82 136 T2 = ri[WS(is, 1)];
Chris@82 137 T3 = ri[WS(is, 4)];
Chris@82 138 T4 = T2 + T3;
Chris@82 139 T5 = ri[WS(is, 2)];
Chris@82 140 T6 = ri[WS(is, 3)];
Chris@82 141 T7 = T5 + T6;
Chris@82 142 T8 = T4 + T7;
Chris@82 143 Tt = T5 - T6;
Chris@82 144 T9 = KP559016994 * (T4 - T7);
Chris@82 145 Ts = T2 - T3;
Chris@82 146 }
Chris@82 147 {
Chris@82 148 E Tc, Td, Tl, Tf, Tg, Tm;
Chris@82 149 Tc = ii[WS(is, 1)];
Chris@82 150 Td = ii[WS(is, 4)];
Chris@82 151 Tl = Tc + Td;
Chris@82 152 Tf = ii[WS(is, 2)];
Chris@82 153 Tg = ii[WS(is, 3)];
Chris@82 154 Tm = Tf + Tg;
Chris@82 155 Te = Tc - Td;
Chris@82 156 Tp = Tl + Tm;
Chris@82 157 Th = Tf - Tg;
Chris@82 158 Tn = KP559016994 * (Tl - Tm);
Chris@82 159 }
Chris@82 160 ro[0] = T1 + T8;
Chris@82 161 io[0] = To + Tp;
Chris@82 162 {
Chris@82 163 E Ti, Tk, Tb, Tj, Ta;
Chris@82 164 Ti = FMA(KP951056516, Te, KP587785252 * Th);
Chris@82 165 Tk = FNMS(KP587785252, Te, KP951056516 * Th);
Chris@82 166 Ta = FNMS(KP250000000, T8, T1);
Chris@82 167 Tb = T9 + Ta;
Chris@82 168 Tj = Ta - T9;
Chris@82 169 ro[WS(os, 4)] = Tb - Ti;
Chris@82 170 ro[WS(os, 3)] = Tj + Tk;
Chris@82 171 ro[WS(os, 1)] = Tb + Ti;
Chris@82 172 ro[WS(os, 2)] = Tj - Tk;
Chris@82 173 }
Chris@82 174 {
Chris@82 175 E Tu, Tv, Tr, Tw, Tq;
Chris@82 176 Tu = FMA(KP951056516, Ts, KP587785252 * Tt);
Chris@82 177 Tv = FNMS(KP587785252, Ts, KP951056516 * Tt);
Chris@82 178 Tq = FNMS(KP250000000, Tp, To);
Chris@82 179 Tr = Tn + Tq;
Chris@82 180 Tw = Tq - Tn;
Chris@82 181 io[WS(os, 1)] = Tr - Tu;
Chris@82 182 io[WS(os, 3)] = Tw - Tv;
Chris@82 183 io[WS(os, 4)] = Tu + Tr;
Chris@82 184 io[WS(os, 2)] = Tv + Tw;
Chris@82 185 }
Chris@82 186 }
Chris@82 187 }
Chris@82 188 }
Chris@82 189
Chris@82 190 static const kdft_desc desc = { 5, "n1_5", {26, 6, 6, 0}, &GENUS, 0, 0, 0, 0 };
Chris@82 191
Chris@82 192 void X(codelet_n1_5) (planner *p) {
Chris@82 193 X(kdft_register) (p, n1_5, &desc);
Chris@82 194 }
Chris@82 195
Chris@82 196 #endif