annotate src/fftw-3.3.8/rdft/scalar/r2cf/r2cf_8.c @ 82:d0c2a83c1364

Add FFTW 3.3.8 source, and a Linux build
author Chris Cannam
date Tue, 19 Nov 2019 14:52:55 +0000
parents
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 08:06:26 EDT 2018 */
Chris@82 23
Chris@82 24 #include "rdft/codelet-rdft.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_r2cf.native -fma -compact -variables 4 -pipeline-latency 4 -n 8 -name r2cf_8 -include rdft/scalar/r2cf.h */
Chris@82 29
Chris@82 30 /*
Chris@82 31 * This function contains 20 FP additions, 4 FP multiplications,
Chris@82 32 * (or, 16 additions, 0 multiplications, 4 fused multiply/add),
Chris@82 33 * 14 stack variables, 1 constants, and 16 memory accesses
Chris@82 34 */
Chris@82 35 #include "rdft/scalar/r2cf.h"
Chris@82 36
Chris@82 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@82 38 {
Chris@82 39 DK(KP707106781, +0.707106781186547524400844362104849039284835938);
Chris@82 40 {
Chris@82 41 INT i;
Chris@82 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@82 43 E T3, T7, Td, Tj, T6, Tf, Ta, Ti;
Chris@82 44 {
Chris@82 45 E T1, T2, Tb, Tc;
Chris@82 46 T1 = R0[0];
Chris@82 47 T2 = R0[WS(rs, 2)];
Chris@82 48 T3 = T1 + T2;
Chris@82 49 T7 = T1 - T2;
Chris@82 50 Tb = R1[WS(rs, 3)];
Chris@82 51 Tc = R1[WS(rs, 1)];
Chris@82 52 Td = Tb - Tc;
Chris@82 53 Tj = Tb + Tc;
Chris@82 54 }
Chris@82 55 {
Chris@82 56 E T4, T5, T8, T9;
Chris@82 57 T4 = R0[WS(rs, 1)];
Chris@82 58 T5 = R0[WS(rs, 3)];
Chris@82 59 T6 = T4 + T5;
Chris@82 60 Tf = T4 - T5;
Chris@82 61 T8 = R1[0];
Chris@82 62 T9 = R1[WS(rs, 2)];
Chris@82 63 Ta = T8 - T9;
Chris@82 64 Ti = T8 + T9;
Chris@82 65 }
Chris@82 66 Cr[WS(csr, 2)] = T3 - T6;
Chris@82 67 Ci[WS(csi, 2)] = Tj - Ti;
Chris@82 68 {
Chris@82 69 E Te, Tg, Th, Tk;
Chris@82 70 Te = Ta + Td;
Chris@82 71 Cr[WS(csr, 3)] = FNMS(KP707106781, Te, T7);
Chris@82 72 Cr[WS(csr, 1)] = FMA(KP707106781, Te, T7);
Chris@82 73 Tg = Td - Ta;
Chris@82 74 Ci[WS(csi, 1)] = FMS(KP707106781, Tg, Tf);
Chris@82 75 Ci[WS(csi, 3)] = FMA(KP707106781, Tg, Tf);
Chris@82 76 Th = T3 + T6;
Chris@82 77 Tk = Ti + Tj;
Chris@82 78 Cr[WS(csr, 4)] = Th - Tk;
Chris@82 79 Cr[0] = Th + Tk;
Chris@82 80 }
Chris@82 81 }
Chris@82 82 }
Chris@82 83 }
Chris@82 84
Chris@82 85 static const kr2c_desc desc = { 8, "r2cf_8", {16, 0, 4, 0}, &GENUS };
Chris@82 86
Chris@82 87 void X(codelet_r2cf_8) (planner *p) {
Chris@82 88 X(kr2c_register) (p, r2cf_8, &desc);
Chris@82 89 }
Chris@82 90
Chris@82 91 #else
Chris@82 92
Chris@82 93 /* Generated by: ../../../genfft/gen_r2cf.native -compact -variables 4 -pipeline-latency 4 -n 8 -name r2cf_8 -include rdft/scalar/r2cf.h */
Chris@82 94
Chris@82 95 /*
Chris@82 96 * This function contains 20 FP additions, 2 FP multiplications,
Chris@82 97 * (or, 20 additions, 2 multiplications, 0 fused multiply/add),
Chris@82 98 * 14 stack variables, 1 constants, and 16 memory accesses
Chris@82 99 */
Chris@82 100 #include "rdft/scalar/r2cf.h"
Chris@82 101
Chris@82 102 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@82 103 {
Chris@82 104 DK(KP707106781, +0.707106781186547524400844362104849039284835938);
Chris@82 105 {
Chris@82 106 INT i;
Chris@82 107 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@82 108 E T3, T7, Td, Tj, T6, Tg, Ta, Ti;
Chris@82 109 {
Chris@82 110 E T1, T2, Tb, Tc;
Chris@82 111 T1 = R0[0];
Chris@82 112 T2 = R0[WS(rs, 2)];
Chris@82 113 T3 = T1 + T2;
Chris@82 114 T7 = T1 - T2;
Chris@82 115 Tb = R1[WS(rs, 3)];
Chris@82 116 Tc = R1[WS(rs, 1)];
Chris@82 117 Td = Tb - Tc;
Chris@82 118 Tj = Tb + Tc;
Chris@82 119 }
Chris@82 120 {
Chris@82 121 E T4, T5, T8, T9;
Chris@82 122 T4 = R0[WS(rs, 1)];
Chris@82 123 T5 = R0[WS(rs, 3)];
Chris@82 124 T6 = T4 + T5;
Chris@82 125 Tg = T4 - T5;
Chris@82 126 T8 = R1[0];
Chris@82 127 T9 = R1[WS(rs, 2)];
Chris@82 128 Ta = T8 - T9;
Chris@82 129 Ti = T8 + T9;
Chris@82 130 }
Chris@82 131 Cr[WS(csr, 2)] = T3 - T6;
Chris@82 132 Ci[WS(csi, 2)] = Tj - Ti;
Chris@82 133 {
Chris@82 134 E Te, Tf, Th, Tk;
Chris@82 135 Te = KP707106781 * (Ta + Td);
Chris@82 136 Cr[WS(csr, 3)] = T7 - Te;
Chris@82 137 Cr[WS(csr, 1)] = T7 + Te;
Chris@82 138 Tf = KP707106781 * (Td - Ta);
Chris@82 139 Ci[WS(csi, 1)] = Tf - Tg;
Chris@82 140 Ci[WS(csi, 3)] = Tg + Tf;
Chris@82 141 Th = T3 + T6;
Chris@82 142 Tk = Ti + Tj;
Chris@82 143 Cr[WS(csr, 4)] = Th - Tk;
Chris@82 144 Cr[0] = Th + Tk;
Chris@82 145 }
Chris@82 146 }
Chris@82 147 }
Chris@82 148 }
Chris@82 149
Chris@82 150 static const kr2c_desc desc = { 8, "r2cf_8", {20, 2, 0, 0}, &GENUS };
Chris@82 151
Chris@82 152 void X(codelet_r2cf_8) (planner *p) {
Chris@82 153 X(kr2c_register) (p, r2cf_8, &desc);
Chris@82 154 }
Chris@82 155
Chris@82 156 #endif