annotate src/fftw-3.3.5/dft/scalar/codelets/n1_3.c @ 127:7867fa7e1b6b

Current fftw source
author Chris Cannam <cannam@all-day-breakfast.com>
date Tue, 18 Oct 2016 13:40:26 +0100
parents
children
rev   line source
cannam@127 1 /*
cannam@127 2 * Copyright (c) 2003, 2007-14 Matteo Frigo
cannam@127 3 * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
cannam@127 4 *
cannam@127 5 * This program is free software; you can redistribute it and/or modify
cannam@127 6 * it under the terms of the GNU General Public License as published by
cannam@127 7 * the Free Software Foundation; either version 2 of the License, or
cannam@127 8 * (at your option) any later version.
cannam@127 9 *
cannam@127 10 * This program is distributed in the hope that it will be useful,
cannam@127 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
cannam@127 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
cannam@127 13 * GNU General Public License for more details.
cannam@127 14 *
cannam@127 15 * You should have received a copy of the GNU General Public License
cannam@127 16 * along with this program; if not, write to the Free Software
cannam@127 17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
cannam@127 18 *
cannam@127 19 */
cannam@127 20
cannam@127 21 /* This file was automatically generated --- DO NOT EDIT */
cannam@127 22 /* Generated on Sat Jul 30 16:35:50 EDT 2016 */
cannam@127 23
cannam@127 24 #include "codelet-dft.h"
cannam@127 25
cannam@127 26 #ifdef HAVE_FMA
cannam@127 27
cannam@127 28 /* Generated by: ../../../genfft/gen_notw.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 3 -name n1_3 -include n.h */
cannam@127 29
cannam@127 30 /*
cannam@127 31 * This function contains 12 FP additions, 6 FP multiplications,
cannam@127 32 * (or, 6 additions, 0 multiplications, 6 fused multiply/add),
cannam@127 33 * 15 stack variables, 2 constants, and 12 memory accesses
cannam@127 34 */
cannam@127 35 #include "n.h"
cannam@127 36
cannam@127 37 static void n1_3(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs)
cannam@127 38 {
cannam@127 39 DK(KP866025403, +0.866025403784438646763723170752936183471402627);
cannam@127 40 DK(KP500000000, +0.500000000000000000000000000000000000000000000);
cannam@127 41 {
cannam@127 42 INT i;
cannam@127 43 for (i = v; i > 0; i = i - 1, ri = ri + ivs, ii = ii + ivs, ro = ro + ovs, io = io + ovs, MAKE_VOLATILE_STRIDE(12, is), MAKE_VOLATILE_STRIDE(12, os)) {
cannam@127 44 E T1, T9, T2, T3, T6, T7;
cannam@127 45 T1 = ri[0];
cannam@127 46 T9 = ii[0];
cannam@127 47 T2 = ri[WS(is, 1)];
cannam@127 48 T3 = ri[WS(is, 2)];
cannam@127 49 T6 = ii[WS(is, 1)];
cannam@127 50 T7 = ii[WS(is, 2)];
cannam@127 51 {
cannam@127 52 E T4, Tc, T8, Ta, T5, Tb;
cannam@127 53 T4 = T2 + T3;
cannam@127 54 Tc = T3 - T2;
cannam@127 55 T8 = T6 - T7;
cannam@127 56 Ta = T6 + T7;
cannam@127 57 T5 = FNMS(KP500000000, T4, T1);
cannam@127 58 ro[0] = T1 + T4;
cannam@127 59 Tb = FNMS(KP500000000, Ta, T9);
cannam@127 60 io[0] = T9 + Ta;
cannam@127 61 ro[WS(os, 1)] = FMA(KP866025403, T8, T5);
cannam@127 62 ro[WS(os, 2)] = FNMS(KP866025403, T8, T5);
cannam@127 63 io[WS(os, 2)] = FNMS(KP866025403, Tc, Tb);
cannam@127 64 io[WS(os, 1)] = FMA(KP866025403, Tc, Tb);
cannam@127 65 }
cannam@127 66 }
cannam@127 67 }
cannam@127 68 }
cannam@127 69
cannam@127 70 static const kdft_desc desc = { 3, "n1_3", {6, 0, 6, 0}, &GENUS, 0, 0, 0, 0 };
cannam@127 71
cannam@127 72 void X(codelet_n1_3) (planner *p) {
cannam@127 73 X(kdft_register) (p, n1_3, &desc);
cannam@127 74 }
cannam@127 75
cannam@127 76 #else /* HAVE_FMA */
cannam@127 77
cannam@127 78 /* Generated by: ../../../genfft/gen_notw.native -compact -variables 4 -pipeline-latency 4 -n 3 -name n1_3 -include n.h */
cannam@127 79
cannam@127 80 /*
cannam@127 81 * This function contains 12 FP additions, 4 FP multiplications,
cannam@127 82 * (or, 10 additions, 2 multiplications, 2 fused multiply/add),
cannam@127 83 * 15 stack variables, 2 constants, and 12 memory accesses
cannam@127 84 */
cannam@127 85 #include "n.h"
cannam@127 86
cannam@127 87 static void n1_3(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs)
cannam@127 88 {
cannam@127 89 DK(KP500000000, +0.500000000000000000000000000000000000000000000);
cannam@127 90 DK(KP866025403, +0.866025403784438646763723170752936183471402627);
cannam@127 91 {
cannam@127 92 INT i;
cannam@127 93 for (i = v; i > 0; i = i - 1, ri = ri + ivs, ii = ii + ivs, ro = ro + ovs, io = io + ovs, MAKE_VOLATILE_STRIDE(12, is), MAKE_VOLATILE_STRIDE(12, os)) {
cannam@127 94 E T1, Ta, T4, T9, T8, Tb, T5, Tc;
cannam@127 95 T1 = ri[0];
cannam@127 96 Ta = ii[0];
cannam@127 97 {
cannam@127 98 E T2, T3, T6, T7;
cannam@127 99 T2 = ri[WS(is, 1)];
cannam@127 100 T3 = ri[WS(is, 2)];
cannam@127 101 T4 = T2 + T3;
cannam@127 102 T9 = KP866025403 * (T3 - T2);
cannam@127 103 T6 = ii[WS(is, 1)];
cannam@127 104 T7 = ii[WS(is, 2)];
cannam@127 105 T8 = KP866025403 * (T6 - T7);
cannam@127 106 Tb = T6 + T7;
cannam@127 107 }
cannam@127 108 ro[0] = T1 + T4;
cannam@127 109 io[0] = Ta + Tb;
cannam@127 110 T5 = FNMS(KP500000000, T4, T1);
cannam@127 111 ro[WS(os, 2)] = T5 - T8;
cannam@127 112 ro[WS(os, 1)] = T5 + T8;
cannam@127 113 Tc = FNMS(KP500000000, Tb, Ta);
cannam@127 114 io[WS(os, 1)] = T9 + Tc;
cannam@127 115 io[WS(os, 2)] = Tc - T9;
cannam@127 116 }
cannam@127 117 }
cannam@127 118 }
cannam@127 119
cannam@127 120 static const kdft_desc desc = { 3, "n1_3", {10, 2, 2, 0}, &GENUS, 0, 0, 0, 0 };
cannam@127 121
cannam@127 122 void X(codelet_n1_3) (planner *p) {
cannam@127 123 X(kdft_register) (p, n1_3, &desc);
cannam@127 124 }
cannam@127 125
cannam@127 126 #endif /* HAVE_FMA */