Mercurial > hg > sv-dependency-builds

annotate src/fftw-3.3.5/rdft/scalar/r2cf/r2cf_8.c @ 83:ae30d91d2ffe

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Replace these with versions built using an older toolset (so as to avoid ABI compatibilities when linking on Ubuntu 14.04 for packaging purposes)
author Chris Cannam
date Fri, 07 Feb 2020 11:51:13 +0000
parents 2cd0e3b3e1fd
children
rev   line source
Chris@42 1 /*
Chris@42 2 * Copyright (c) 2003, 2007-14 Matteo Frigo
Chris@42 3 * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
Chris@42 4 *
Chris@42 5 * This program is free software; you can redistribute it and/or modify
Chris@42 6 * it under the terms of the GNU General Public License as published by
Chris@42 7 * the Free Software Foundation; either version 2 of the License, or
Chris@42 8 * (at your option) any later version.
Chris@42 9 *
Chris@42 10 * This program is distributed in the hope that it will be useful,
Chris@42 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
Chris@42 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
Chris@42 13 * GNU General Public License for more details.
Chris@42 14 *
Chris@42 15 * You should have received a copy of the GNU General Public License
Chris@42 16 * along with this program; if not, write to the Free Software
Chris@42 17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
Chris@42 18 *
Chris@42 19 */
Chris@42 20
Chris@42 21 /* This file was automatically generated --- DO NOT EDIT */
Chris@42 22 /* Generated on Sat Jul 30 16:46:03 EDT 2016 */
Chris@42 23
Chris@42 24 #include "codelet-rdft.h"
Chris@42 25
Chris@42 26 #ifdef HAVE_FMA
Chris@42 27
Chris@42 28 /* Generated by: ../../../genfft/gen_r2cf.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 8 -name r2cf_8 -include r2cf.h */
Chris@42 29
Chris@42 30 /*
Chris@42 31 * This function contains 20 FP additions, 4 FP multiplications,
Chris@42 32 * (or, 16 additions, 0 multiplications, 4 fused multiply/add),
Chris@42 33 * 18 stack variables, 1 constants, and 16 memory accesses
Chris@42 34 */
Chris@42 35 #include "r2cf.h"
Chris@42 36
Chris@42 37 static void r2cf_8(R *R0, R *R1, R *Cr, R *Ci, stride rs, stride csr, stride csi, INT v, INT ivs, INT ovs)
Chris@42 38 {
Chris@42 39 DK(KP707106781, +0.707106781186547524400844362104849039284835938);
Chris@42 40 {
Chris@42 41 INT i;
Chris@42 42 for (i = v; i > 0; i = i - 1, R0 = R0 + ivs, R1 = R1 + ivs, Cr = Cr + ovs, Ci = Ci + ovs, MAKE_VOLATILE_STRIDE(32, rs), MAKE_VOLATILE_STRIDE(32, csr), MAKE_VOLATILE_STRIDE(32, csi)) {
Chris@42 43 E T4, T7, T3, Tj, Td, T5, T8, T9;
Chris@42 44 {
Chris@42 45 E T1, T2, Tb, Tc;
Chris@42 46 T1 = R0[0];
Chris@42 47 T2 = R0[WS(rs, 2)];
Chris@42 48 Tb = R1[WS(rs, 3)];
Chris@42 49 Tc = R1[WS(rs, 1)];
Chris@42 50 T4 = R0[WS(rs, 1)];
Chris@42 51 T7 = T1 - T2;
Chris@42 52 T3 = T1 + T2;
Chris@42 53 Tj = Tb + Tc;
Chris@42 54 Td = Tb - Tc;
Chris@42 55 T5 = R0[WS(rs, 3)];
Chris@42 56 T8 = R1[0];
Chris@42 57 T9 = R1[WS(rs, 2)];
Chris@42 58 }
Chris@42 59 {
Chris@42 60 E T6, Tf, Ta, Ti;
Chris@42 61 T6 = T4 + T5;
Chris@42 62 Tf = T4 - T5;
Chris@42 63 Ta = T8 - T9;
Chris@42 64 Ti = T8 + T9;
Chris@42 65 {
Chris@42 66 E Th, Tk, Te, Tg;
Chris@42 67 Th = T3 + T6;
Chris@42 68 Cr[WS(csr, 2)] = T3 - T6;
Chris@42 69 Tk = Ti + Tj;
Chris@42 70 Ci[WS(csi, 2)] = Tj - Ti;
Chris@42 71 Te = Ta + Td;
Chris@42 72 Tg = Td - Ta;
Chris@42 73 Cr[0] = Th + Tk;
Chris@42 74 Cr[WS(csr, 4)] = Th - Tk;
Chris@42 75 Ci[WS(csi, 3)] = FMA(KP707106781, Tg, Tf);
Chris@42 76 Ci[WS(csi, 1)] = FMS(KP707106781, Tg, Tf);
Chris@42 77 Cr[WS(csr, 1)] = FMA(KP707106781, Te, T7);
Chris@42 78 Cr[WS(csr, 3)] = FNMS(KP707106781, Te, T7);
Chris@42 79 }
Chris@42 80 }
Chris@42 81 }
Chris@42 82 }
Chris@42 83 }
Chris@42 84
Chris@42 85 static const kr2c_desc desc = { 8, "r2cf_8", {16, 0, 4, 0}, &GENUS };
Chris@42 86
Chris@42 87 void X(codelet_r2cf_8) (planner *p) {
Chris@42 88 X(kr2c_register) (p, r2cf_8, &desc);
Chris@42 89 }
Chris@42 90
Chris@42 91 #else /* HAVE_FMA */
Chris@42 92
Chris@42 93 /* Generated by: ../../../genfft/gen_r2cf.native -compact -variables 4 -pipeline-latency 4 -n 8 -name r2cf_8 -include r2cf.h */
Chris@42 94
Chris@42 95 /*
Chris@42 96 * This function contains 20 FP additions, 2 FP multiplications,
Chris@42 97 * (or, 20 additions, 2 multiplications, 0 fused multiply/add),
Chris@42 98 * 14 stack variables, 1 constants, and 16 memory accesses
Chris@42 99 */
Chris@42 100 #include "r2cf.h"
Chris@42 101
Chris@42 102 static void r2cf_8(R *R0, R *R1, R *Cr, R *Ci, stride rs, stride csr, stride csi, INT v, INT ivs, INT ovs)
Chris@42 103 {
Chris@42 104 DK(KP707106781, +0.707106781186547524400844362104849039284835938);
Chris@42 105 {
Chris@42 106 INT i;
Chris@42 107 for (i = v; i > 0; i = i - 1, R0 = R0 + ivs, R1 = R1 + ivs, Cr = Cr + ovs, Ci = Ci + ovs, MAKE_VOLATILE_STRIDE(32, rs), MAKE_VOLATILE_STRIDE(32, csr), MAKE_VOLATILE_STRIDE(32, csi)) {
Chris@42 108 E T3, T7, Td, Tj, T6, Tg, Ta, Ti;
Chris@42 109 {
Chris@42 110 E T1, T2, Tb, Tc;
Chris@42 111 T1 = R0[0];
Chris@42 112 T2 = R0[WS(rs, 2)];
Chris@42 113 T3 = T1 + T2;
Chris@42 114 T7 = T1 - T2;
Chris@42 115 Tb = R1[WS(rs, 3)];
Chris@42 116 Tc = R1[WS(rs, 1)];
Chris@42 117 Td = Tb - Tc;
Chris@42 118 Tj = Tb + Tc;
Chris@42 119 }
Chris@42 120 {
Chris@42 121 E T4, T5, T8, T9;
Chris@42 122 T4 = R0[WS(rs, 1)];
Chris@42 123 T5 = R0[WS(rs, 3)];
Chris@42 124 T6 = T4 + T5;
Chris@42 125 Tg = T4 - T5;
Chris@42 126 T8 = R1[0];
Chris@42 127 T9 = R1[WS(rs, 2)];
Chris@42 128 Ta = T8 - T9;
Chris@42 129 Ti = T8 + T9;
Chris@42 130 }
Chris@42 131 Cr[WS(csr, 2)] = T3 - T6;
Chris@42 132 Ci[WS(csi, 2)] = Tj - Ti;
Chris@42 133 {
Chris@42 134 E Te, Tf, Th, Tk;
Chris@42 135 Te = KP707106781 * (Ta + Td);
Chris@42 136 Cr[WS(csr, 3)] = T7 - Te;
Chris@42 137 Cr[WS(csr, 1)] = T7 + Te;
Chris@42 138 Tf = KP707106781 * (Td - Ta);
Chris@42 139 Ci[WS(csi, 1)] = Tf - Tg;
Chris@42 140 Ci[WS(csi, 3)] = Tg + Tf;
Chris@42 141 Th = T3 + T6;
Chris@42 142 Tk = Ti + Tj;
Chris@42 143 Cr[WS(csr, 4)] = Th - Tk;
Chris@42 144 Cr[0] = Th + Tk;
Chris@42 145 }
Chris@42 146 }
Chris@42 147 }
Chris@42 148 }
Chris@42 149
Chris@42 150 static const kr2c_desc desc = { 8, "r2cf_8", {20, 2, 0, 0}, &GENUS };
Chris@42 151
Chris@42 152 void X(codelet_r2cf_8) (planner *p) {
Chris@42 153 X(kr2c_register) (p, r2cf_8, &desc);
Chris@42 154 }
Chris@42 155
Chris@42 156 #endif /* HAVE_FMA */