Mercurial > hg > sv-dependency-builds

annotate src/fftw-3.3.8/dft/simd/common/t3fv_8.c @ 82:d0c2a83c1364

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
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 08:05:51 EDT 2018 */
Chris@82 23
Chris@82 24 #include "dft/codelet-dft.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_twiddle_c.native -fma -simd -compact -variables 4 -pipeline-latency 8 -twiddle-log3 -precompute-twiddles -no-generate-bytw -n 8 -name t3fv_8 -include dft/simd/t3f.h */
Chris@82 29
Chris@82 30 /*
Chris@82 31 * This function contains 37 FP additions, 32 FP multiplications,
Chris@82 32 * (or, 27 additions, 22 multiplications, 10 fused multiply/add),
Chris@82 33 * 31 stack variables, 1 constants, and 16 memory accesses
Chris@82 34 */
Chris@82 35 #include "dft/simd/t3f.h"
Chris@82 36
Chris@82 37 static void t3fv_8(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms)
Chris@82 38 {
Chris@82 39 DVK(KP707106781, +0.707106781186547524400844362104849039284835938);
Chris@82 40 {
Chris@82 41 INT m;
Chris@82 42 R *x;
Chris@82 43 x = ri;
Chris@82 44 for (m = mb, W = W + (mb * ((TWVL / VL) * 6)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 6), MAKE_VOLATILE_STRIDE(8, rs)) {
Chris@82 45 V T2, T3, Ta, T4, Tb, Tc, Tp;
Chris@82 46 T2 = LDW(&(W[0]));
Chris@82 47 T3 = LDW(&(W[TWVL * 2]));
Chris@82 48 Ta = VZMULJ(T2, T3);
Chris@82 49 T4 = VZMUL(T2, T3);
Chris@82 50 Tb = LDW(&(W[TWVL * 4]));
Chris@82 51 Tc = VZMULJ(Ta, Tb);
Chris@82 52 Tp = VZMULJ(T2, Tb);
Chris@82 53 {
Chris@82 54 V T7, Tx, Ts, Ty, Tf, TA, Tk, TB, T1, T6, T5;
Chris@82 55 T1 = LD(&(x[0]), ms, &(x[0]));
Chris@82 56 T5 = LD(&(x[WS(rs, 4)]), ms, &(x[0]));
Chris@82 57 T6 = VZMULJ(T4, T5);
Chris@82 58 T7 = VSUB(T1, T6);
Chris@82 59 Tx = VADD(T1, T6);
Chris@82 60 {
Chris@82 61 V To, Tr, Tn, Tq;
Chris@82 62 Tn = LD(&(x[WS(rs, 2)]), ms, &(x[0]));
Chris@82 63 To = VZMULJ(Ta, Tn);
Chris@82 64 Tq = LD(&(x[WS(rs, 6)]), ms, &(x[0]));
Chris@82 65 Tr = VZMULJ(Tp, Tq);
Chris@82 66 Ts = VSUB(To, Tr);
Chris@82 67 Ty = VADD(To, Tr);
Chris@82 68 }
Chris@82 69 {
Chris@82 70 V T9, Te, T8, Td;
Chris@82 71 T8 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)]));
Chris@82 72 T9 = VZMULJ(T2, T8);
Chris@82 73 Td = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)]));
Chris@82 74 Te = VZMULJ(Tc, Td);
Chris@82 75 Tf = VSUB(T9, Te);
Chris@82 76 TA = VADD(T9, Te);
Chris@82 77 }
Chris@82 78 {
Chris@82 79 V Th, Tj, Tg, Ti;
Chris@82 80 Tg = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)]));
Chris@82 81 Th = VZMULJ(Tb, Tg);
Chris@82 82 Ti = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)]));
Chris@82 83 Tj = VZMULJ(T3, Ti);
Chris@82 84 Tk = VSUB(Th, Tj);
Chris@82 85 TB = VADD(Th, Tj);
Chris@82 86 }
Chris@82 87 {
Chris@82 88 V Tz, TC, TD, TE;
Chris@82 89 Tz = VADD(Tx, Ty);
Chris@82 90 TC = VADD(TA, TB);
Chris@82 91 ST(&(x[WS(rs, 4)]), VSUB(Tz, TC), ms, &(x[0]));
Chris@82 92 ST(&(x[0]), VADD(Tz, TC), ms, &(x[0]));
Chris@82 93 TD = VSUB(Tx, Ty);
Chris@82 94 TE = VSUB(TB, TA);
Chris@82 95 ST(&(x[WS(rs, 6)]), VFNMSI(TE, TD), ms, &(x[0]));
Chris@82 96 ST(&(x[WS(rs, 2)]), VFMAI(TE, TD), ms, &(x[0]));
Chris@82 97 {
Chris@82 98 V Tm, Tv, Tu, Tw, Tl, Tt;
Chris@82 99 Tl = VADD(Tf, Tk);
Chris@82 100 Tm = VFMA(LDK(KP707106781), Tl, T7);
Chris@82 101 Tv = VFNMS(LDK(KP707106781), Tl, T7);
Chris@82 102 Tt = VSUB(Tk, Tf);
Chris@82 103 Tu = VFNMS(LDK(KP707106781), Tt, Ts);
Chris@82 104 Tw = VFMA(LDK(KP707106781), Tt, Ts);
Chris@82 105 ST(&(x[WS(rs, 1)]), VFNMSI(Tu, Tm), ms, &(x[WS(rs, 1)]));
Chris@82 106 ST(&(x[WS(rs, 3)]), VFMAI(Tw, Tv), ms, &(x[WS(rs, 1)]));
Chris@82 107 ST(&(x[WS(rs, 7)]), VFMAI(Tu, Tm), ms, &(x[WS(rs, 1)]));
Chris@82 108 ST(&(x[WS(rs, 5)]), VFNMSI(Tw, Tv), ms, &(x[WS(rs, 1)]));
Chris@82 109 }
Chris@82 110 }
Chris@82 111 }
Chris@82 112 }
Chris@82 113 }
Chris@82 114 VLEAVE();
Chris@82 115 }
Chris@82 116
Chris@82 117 static const tw_instr twinstr[] = {
Chris@82 118 VTW(0, 1),
Chris@82 119 VTW(0, 3),
Chris@82 120 VTW(0, 7),
Chris@82 121 {TW_NEXT, VL, 0}
Chris@82 122 };
Chris@82 123
Chris@82 124 static const ct_desc desc = { 8, XSIMD_STRING("t3fv_8"), twinstr, &GENUS, {27, 22, 10, 0}, 0, 0, 0 };
Chris@82 125
Chris@82 126 void XSIMD(codelet_t3fv_8) (planner *p) {
Chris@82 127 X(kdft_dit_register) (p, t3fv_8, &desc);
Chris@82 128 }
Chris@82 129 #else
Chris@82 130
Chris@82 131 /* Generated by: ../../../genfft/gen_twiddle_c.native -simd -compact -variables 4 -pipeline-latency 8 -twiddle-log3 -precompute-twiddles -no-generate-bytw -n 8 -name t3fv_8 -include dft/simd/t3f.h */
Chris@82 132
Chris@82 133 /*
Chris@82 134 * This function contains 37 FP additions, 24 FP multiplications,
Chris@82 135 * (or, 37 additions, 24 multiplications, 0 fused multiply/add),
Chris@82 136 * 31 stack variables, 1 constants, and 16 memory accesses
Chris@82 137 */
Chris@82 138 #include "dft/simd/t3f.h"
Chris@82 139
Chris@82 140 static void t3fv_8(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms)
Chris@82 141 {
Chris@82 142 DVK(KP707106781, +0.707106781186547524400844362104849039284835938);
Chris@82 143 {
Chris@82 144 INT m;
Chris@82 145 R *x;
Chris@82 146 x = ri;
Chris@82 147 for (m = mb, W = W + (mb * ((TWVL / VL) * 6)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 6), MAKE_VOLATILE_STRIDE(8, rs)) {
Chris@82 148 V T2, T3, Ta, T4, Tb, Tc, Tq;
Chris@82 149 T2 = LDW(&(W[0]));
Chris@82 150 T3 = LDW(&(W[TWVL * 2]));
Chris@82 151 Ta = VZMULJ(T2, T3);
Chris@82 152 T4 = VZMUL(T2, T3);
Chris@82 153 Tb = LDW(&(W[TWVL * 4]));
Chris@82 154 Tc = VZMULJ(Ta, Tb);
Chris@82 155 Tq = VZMULJ(T2, Tb);
Chris@82 156 {
Chris@82 157 V T7, Tx, Tt, Ty, Tf, TA, Tk, TB, T1, T6, T5;
Chris@82 158 T1 = LD(&(x[0]), ms, &(x[0]));
Chris@82 159 T5 = LD(&(x[WS(rs, 4)]), ms, &(x[0]));
Chris@82 160 T6 = VZMULJ(T4, T5);
Chris@82 161 T7 = VSUB(T1, T6);
Chris@82 162 Tx = VADD(T1, T6);
Chris@82 163 {
Chris@82 164 V Tp, Ts, To, Tr;
Chris@82 165 To = LD(&(x[WS(rs, 2)]), ms, &(x[0]));
Chris@82 166 Tp = VZMULJ(Ta, To);
Chris@82 167 Tr = LD(&(x[WS(rs, 6)]), ms, &(x[0]));
Chris@82 168 Ts = VZMULJ(Tq, Tr);
Chris@82 169 Tt = VSUB(Tp, Ts);
Chris@82 170 Ty = VADD(Tp, Ts);
Chris@82 171 }
Chris@82 172 {
Chris@82 173 V T9, Te, T8, Td;
Chris@82 174 T8 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)]));
Chris@82 175 T9 = VZMULJ(T2, T8);
Chris@82 176 Td = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)]));
Chris@82 177 Te = VZMULJ(Tc, Td);
Chris@82 178 Tf = VSUB(T9, Te);
Chris@82 179 TA = VADD(T9, Te);
Chris@82 180 }
Chris@82 181 {
Chris@82 182 V Th, Tj, Tg, Ti;
Chris@82 183 Tg = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)]));
Chris@82 184 Th = VZMULJ(Tb, Tg);
Chris@82 185 Ti = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)]));
Chris@82 186 Tj = VZMULJ(T3, Ti);
Chris@82 187 Tk = VSUB(Th, Tj);
Chris@82 188 TB = VADD(Th, Tj);
Chris@82 189 }
Chris@82 190 {
Chris@82 191 V Tz, TC, TD, TE;
Chris@82 192 Tz = VADD(Tx, Ty);
Chris@82 193 TC = VADD(TA, TB);
Chris@82 194 ST(&(x[WS(rs, 4)]), VSUB(Tz, TC), ms, &(x[0]));
Chris@82 195 ST(&(x[0]), VADD(Tz, TC), ms, &(x[0]));
Chris@82 196 TD = VSUB(Tx, Ty);
Chris@82 197 TE = VBYI(VSUB(TB, TA));
Chris@82 198 ST(&(x[WS(rs, 6)]), VSUB(TD, TE), ms, &(x[0]));
Chris@82 199 ST(&(x[WS(rs, 2)]), VADD(TD, TE), ms, &(x[0]));
Chris@82 200 {
Chris@82 201 V Tm, Tv, Tu, Tw, Tl, Tn;
Chris@82 202 Tl = VMUL(LDK(KP707106781), VADD(Tf, Tk));
Chris@82 203 Tm = VADD(T7, Tl);
Chris@82 204 Tv = VSUB(T7, Tl);
Chris@82 205 Tn = VMUL(LDK(KP707106781), VSUB(Tk, Tf));
Chris@82 206 Tu = VBYI(VSUB(Tn, Tt));
Chris@82 207 Tw = VBYI(VADD(Tt, Tn));
Chris@82 208 ST(&(x[WS(rs, 7)]), VSUB(Tm, Tu), ms, &(x[WS(rs, 1)]));
Chris@82 209 ST(&(x[WS(rs, 3)]), VADD(Tv, Tw), ms, &(x[WS(rs, 1)]));
Chris@82 210 ST(&(x[WS(rs, 1)]), VADD(Tm, Tu), ms, &(x[WS(rs, 1)]));
Chris@82 211 ST(&(x[WS(rs, 5)]), VSUB(Tv, Tw), ms, &(x[WS(rs, 1)]));
Chris@82 212 }
Chris@82 213 }
Chris@82 214 }
Chris@82 215 }
Chris@82 216 }
Chris@82 217 VLEAVE();
Chris@82 218 }
Chris@82 219
Chris@82 220 static const tw_instr twinstr[] = {
Chris@82 221 VTW(0, 1),
Chris@82 222 VTW(0, 3),
Chris@82 223 VTW(0, 7),
Chris@82 224 {TW_NEXT, VL, 0}
Chris@82 225 };
Chris@82 226
Chris@82 227 static const ct_desc desc = { 8, XSIMD_STRING("t3fv_8"), twinstr, &GENUS, {37, 24, 0, 0}, 0, 0, 0 };
Chris@82 228
Chris@82 229 void XSIMD(codelet_t3fv_8) (planner *p) {
Chris@82 230 X(kdft_dit_register) (p, t3fv_8, &desc);
Chris@82 231 }
Chris@82 232 #endif