Mercurial > hg > sv-dependency-builds

annotate src/fftw-3.3.8/rdft/scalar/r2cb/r2cbIII_7.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 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 08:07:42 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_r2cb.native -fma -compact -variables 4 -pipeline-latency 4 -sign 1 -n 7 -name r2cbIII_7 -dft-III -include rdft/scalar/r2cbIII.h */
Chris@82 29
Chris@82 30 /*
Chris@82 31 * This function contains 24 FP additions, 22 FP multiplications,
Chris@82 32 * (or, 2 additions, 0 multiplications, 22 fused multiply/add),
Chris@82 33 * 27 stack variables, 7 constants, and 14 memory accesses
Chris@82 34 */
Chris@82 35 #include "rdft/scalar/r2cbIII.h"
Chris@82 36
Chris@82 37 static void r2cbIII_7(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(KP1_949855824, +1.949855824363647214036263365987862434465571601);
Chris@82 40 DK(KP801937735, +0.801937735804838252472204639014890102331838324);
Chris@82 41 DK(KP1_801937735, +1.801937735804838252472204639014890102331838324);
Chris@82 42 DK(KP2_000000000, +2.000000000000000000000000000000000000000000000);
Chris@82 43 DK(KP692021471, +0.692021471630095869627814897002069140197260599);
Chris@82 44 DK(KP356895867, +0.356895867892209443894399510021300583399127187);
Chris@82 45 DK(KP554958132, +0.554958132087371191422194871006410481067288862);
Chris@82 46 {
Chris@82 47 INT i;
Chris@82 48 for (i = v; i > 0; i = i - 1, R0 = R0 + ovs, R1 = R1 + ovs, Cr = Cr + ivs, Ci = Ci + ivs, MAKE_VOLATILE_STRIDE(28, rs), MAKE_VOLATILE_STRIDE(28, csr), MAKE_VOLATILE_STRIDE(28, csi)) {
Chris@82 49 E T1, T9, Tb, Ta, Tc, Tm, Th, T7, Tk, Tf, T5, Tl, Tn;
Chris@82 50 T1 = Cr[WS(csr, 3)];
Chris@82 51 T9 = Ci[WS(csi, 1)];
Chris@82 52 Tb = Ci[0];
Chris@82 53 Ta = Ci[WS(csi, 2)];
Chris@82 54 Tc = FMA(KP554958132, Tb, Ta);
Chris@82 55 Tm = FNMS(KP554958132, Ta, T9);
Chris@82 56 Th = FMA(KP554958132, T9, Tb);
Chris@82 57 {
Chris@82 58 E T2, T4, T3, T6, Tj, Te;
Chris@82 59 T2 = Cr[WS(csr, 2)];
Chris@82 60 T4 = Cr[0];
Chris@82 61 T3 = Cr[WS(csr, 1)];
Chris@82 62 T6 = FNMS(KP356895867, T3, T2);
Chris@82 63 Tj = FNMS(KP356895867, T4, T3);
Chris@82 64 Te = FNMS(KP356895867, T2, T4);
Chris@82 65 T7 = FNMS(KP692021471, T6, T4);
Chris@82 66 Tk = FNMS(KP692021471, Tj, T2);
Chris@82 67 Tf = FNMS(KP692021471, Te, T3);
Chris@82 68 T5 = T2 + T3 + T4;
Chris@82 69 }
Chris@82 70 R0[0] = FMA(KP2_000000000, T5, T1);
Chris@82 71 Tl = FNMS(KP1_801937735, Tk, T1);
Chris@82 72 Tn = FNMS(KP801937735, Tm, Tb);
Chris@82 73 R1[WS(rs, 1)] = -(FMA(KP1_949855824, Tn, Tl));
Chris@82 74 R0[WS(rs, 2)] = FNMS(KP1_949855824, Tn, Tl);
Chris@82 75 {
Chris@82 76 E T8, Td, Tg, Ti;
Chris@82 77 T8 = FNMS(KP1_801937735, T7, T1);
Chris@82 78 Td = FMA(KP801937735, Tc, T9);
Chris@82 79 R1[0] = -(FMA(KP1_949855824, Td, T8));
Chris@82 80 R0[WS(rs, 3)] = FNMS(KP1_949855824, Td, T8);
Chris@82 81 Tg = FNMS(KP1_801937735, Tf, T1);
Chris@82 82 Ti = FNMS(KP801937735, Th, Ta);
Chris@82 83 R0[WS(rs, 1)] = FMA(KP1_949855824, Ti, Tg);
Chris@82 84 R1[WS(rs, 2)] = FMS(KP1_949855824, Ti, Tg);
Chris@82 85 }
Chris@82 86 }
Chris@82 87 }
Chris@82 88 }
Chris@82 89
Chris@82 90 static const kr2c_desc desc = { 7, "r2cbIII_7", {2, 0, 22, 0}, &GENUS };
Chris@82 91
Chris@82 92 void X(codelet_r2cbIII_7) (planner *p) {
Chris@82 93 X(kr2c_register) (p, r2cbIII_7, &desc);
Chris@82 94 }
Chris@82 95
Chris@82 96 #else
Chris@82 97
Chris@82 98 /* Generated by: ../../../genfft/gen_r2cb.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 7 -name r2cbIII_7 -dft-III -include rdft/scalar/r2cbIII.h */
Chris@82 99
Chris@82 100 /*
Chris@82 101 * This function contains 24 FP additions, 19 FP multiplications,
Chris@82 102 * (or, 9 additions, 4 multiplications, 15 fused multiply/add),
Chris@82 103 * 21 stack variables, 7 constants, and 14 memory accesses
Chris@82 104 */
Chris@82 105 #include "rdft/scalar/r2cbIII.h"
Chris@82 106
Chris@82 107 static void r2cbIII_7(R *R0, R *R1, R *Cr, R *Ci, stride rs, stride csr, stride csi, INT v, INT ivs, INT ovs)
Chris@82 108 {
Chris@82 109 DK(KP2_000000000, +2.000000000000000000000000000000000000000000000);
Chris@82 110 DK(KP1_246979603, +1.246979603717467061050009768008479621264549462);
Chris@82 111 DK(KP1_801937735, +1.801937735804838252472204639014890102331838324);
Chris@82 112 DK(KP445041867, +0.445041867912628808577805128993589518932711138);
Chris@82 113 DK(KP867767478, +0.867767478235116240951536665696717509219981456);
Chris@82 114 DK(KP1_949855824, +1.949855824363647214036263365987862434465571601);
Chris@82 115 DK(KP1_563662964, +1.563662964936059617416889053348115500464669037);
Chris@82 116 {
Chris@82 117 INT i;
Chris@82 118 for (i = v; i > 0; i = i - 1, R0 = R0 + ovs, R1 = R1 + ovs, Cr = Cr + ivs, Ci = Ci + ivs, MAKE_VOLATILE_STRIDE(28, rs), MAKE_VOLATILE_STRIDE(28, csr), MAKE_VOLATILE_STRIDE(28, csi)) {
Chris@82 119 E T9, Td, Tb, T1, T4, T2, T3, T5, Tc, Ta, T6, T8, T7;
Chris@82 120 T6 = Ci[WS(csi, 2)];
Chris@82 121 T8 = Ci[0];
Chris@82 122 T7 = Ci[WS(csi, 1)];
Chris@82 123 T9 = FMA(KP1_563662964, T6, KP1_949855824 * T7) + (KP867767478 * T8);
Chris@82 124 Td = FNMS(KP1_949855824, T8, KP1_563662964 * T7) - (KP867767478 * T6);
Chris@82 125 Tb = FNMS(KP1_563662964, T8, KP1_949855824 * T6) - (KP867767478 * T7);
Chris@82 126 T1 = Cr[WS(csr, 3)];
Chris@82 127 T4 = Cr[0];
Chris@82 128 T2 = Cr[WS(csr, 2)];
Chris@82 129 T3 = Cr[WS(csr, 1)];
Chris@82 130 T5 = FMA(KP445041867, T3, KP1_801937735 * T4) + FNMA(KP1_246979603, T2, T1);
Chris@82 131 Tc = FMA(KP1_801937735, T2, KP445041867 * T4) + FNMA(KP1_246979603, T3, T1);
Chris@82 132 Ta = FMA(KP1_246979603, T4, T1) + FNMA(KP1_801937735, T3, KP445041867 * T2);
Chris@82 133 R1[0] = T5 - T9;
Chris@82 134 R0[WS(rs, 3)] = -(T5 + T9);
Chris@82 135 R0[WS(rs, 2)] = Td - Tc;
Chris@82 136 R1[WS(rs, 1)] = Tc + Td;
Chris@82 137 R1[WS(rs, 2)] = Tb - Ta;
Chris@82 138 R0[WS(rs, 1)] = Ta + Tb;
Chris@82 139 R0[0] = FMA(KP2_000000000, T2 + T3 + T4, T1);
Chris@82 140 }
Chris@82 141 }
Chris@82 142 }
Chris@82 143
Chris@82 144 static const kr2c_desc desc = { 7, "r2cbIII_7", {9, 4, 15, 0}, &GENUS };
Chris@82 145
Chris@82 146 void X(codelet_r2cbIII_7) (planner *p) {
Chris@82 147 X(kr2c_register) (p, r2cbIII_7, &desc);
Chris@82 148 }
Chris@82 149
Chris@82 150 #endif