annotate src/fftw-3.3.8/rdft/scalar/r2cb/r2cbIII_4.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 08:07:42 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 4 -name r2cbIII_4 -dft-III -include rdft/scalar/r2cbIII.h */
Chris@82 29
Chris@82 30 /*
Chris@82 31 * This function contains 6 FP additions, 4 FP multiplications,
Chris@82 32 * (or, 6 additions, 4 multiplications, 0 fused multiply/add),
Chris@82 33 * 9 stack variables, 2 constants, and 8 memory accesses
Chris@82 34 */
Chris@82 35 #include "rdft/scalar/r2cbIII.h"
Chris@82 36
Chris@82 37 static void r2cbIII_4(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_414213562, +1.414213562373095048801688724209698078569671875);
Chris@82 40 DK(KP2_000000000, +2.000000000000000000000000000000000000000000000);
Chris@82 41 {
Chris@82 42 INT i;
Chris@82 43 for (i = v; i > 0; i = i - 1, R0 = R0 + ovs, R1 = R1 + ovs, Cr = Cr + ivs, Ci = Ci + ivs, MAKE_VOLATILE_STRIDE(16, rs), MAKE_VOLATILE_STRIDE(16, csr), MAKE_VOLATILE_STRIDE(16, csi)) {
Chris@82 44 E T1, T2, T3, T4, T5, T6;
Chris@82 45 T1 = Cr[0];
Chris@82 46 T2 = Cr[WS(csr, 1)];
Chris@82 47 T3 = T1 - T2;
Chris@82 48 T4 = Ci[0];
Chris@82 49 T5 = Ci[WS(csi, 1)];
Chris@82 50 T6 = T4 + T5;
Chris@82 51 R0[0] = KP2_000000000 * (T1 + T2);
Chris@82 52 R0[WS(rs, 1)] = KP2_000000000 * (T5 - T4);
Chris@82 53 R1[0] = KP1_414213562 * (T3 - T6);
Chris@82 54 R1[WS(rs, 1)] = -(KP1_414213562 * (T3 + T6));
Chris@82 55 }
Chris@82 56 }
Chris@82 57 }
Chris@82 58
Chris@82 59 static const kr2c_desc desc = { 4, "r2cbIII_4", {6, 4, 0, 0}, &GENUS };
Chris@82 60
Chris@82 61 void X(codelet_r2cbIII_4) (planner *p) {
Chris@82 62 X(kr2c_register) (p, r2cbIII_4, &desc);
Chris@82 63 }
Chris@82 64
Chris@82 65 #else
Chris@82 66
Chris@82 67 /* Generated by: ../../../genfft/gen_r2cb.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 4 -name r2cbIII_4 -dft-III -include rdft/scalar/r2cbIII.h */
Chris@82 68
Chris@82 69 /*
Chris@82 70 * This function contains 6 FP additions, 4 FP multiplications,
Chris@82 71 * (or, 6 additions, 4 multiplications, 0 fused multiply/add),
Chris@82 72 * 9 stack variables, 2 constants, and 8 memory accesses
Chris@82 73 */
Chris@82 74 #include "rdft/scalar/r2cbIII.h"
Chris@82 75
Chris@82 76 static void r2cbIII_4(R *R0, R *R1, R *Cr, R *Ci, stride rs, stride csr, stride csi, INT v, INT ivs, INT ovs)
Chris@82 77 {
Chris@82 78 DK(KP1_414213562, +1.414213562373095048801688724209698078569671875);
Chris@82 79 DK(KP2_000000000, +2.000000000000000000000000000000000000000000000);
Chris@82 80 {
Chris@82 81 INT i;
Chris@82 82 for (i = v; i > 0; i = i - 1, R0 = R0 + ovs, R1 = R1 + ovs, Cr = Cr + ivs, Ci = Ci + ivs, MAKE_VOLATILE_STRIDE(16, rs), MAKE_VOLATILE_STRIDE(16, csr), MAKE_VOLATILE_STRIDE(16, csi)) {
Chris@82 83 E T1, T2, T3, T4, T5, T6;
Chris@82 84 T1 = Cr[0];
Chris@82 85 T2 = Cr[WS(csr, 1)];
Chris@82 86 T3 = T1 - T2;
Chris@82 87 T4 = Ci[0];
Chris@82 88 T5 = Ci[WS(csi, 1)];
Chris@82 89 T6 = T4 + T5;
Chris@82 90 R0[0] = KP2_000000000 * (T1 + T2);
Chris@82 91 R0[WS(rs, 1)] = KP2_000000000 * (T5 - T4);
Chris@82 92 R1[0] = KP1_414213562 * (T3 - T6);
Chris@82 93 R1[WS(rs, 1)] = -(KP1_414213562 * (T3 + T6));
Chris@82 94 }
Chris@82 95 }
Chris@82 96 }
Chris@82 97
Chris@82 98 static const kr2c_desc desc = { 4, "r2cbIII_4", {6, 4, 0, 0}, &GENUS };
Chris@82 99
Chris@82 100 void X(codelet_r2cbIII_4) (planner *p) {
Chris@82 101 X(kr2c_register) (p, r2cbIII_4, &desc);
Chris@82 102 }
Chris@82 103
Chris@82 104 #endif