Mercurial > hg > sv-dependency-builds

annotate src/fftw-3.3.5/dft/scalar/codelets/q1_3.c @ 84:08ae793730bd

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Add null config files
author Chris Cannam
date Mon, 02 Mar 2020 14:03:47 +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 16:37:56 EDT 2016 */
Chris@42 23
Chris@42 24 #include "codelet-dft.h"
Chris@42 25
Chris@42 26 #ifdef HAVE_FMA
Chris@42 27
Chris@42 28 /* Generated by: ../../../genfft/gen_twidsq.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -reload-twiddle -dif -n 3 -name q1_3 -include q.h */
Chris@42 29
Chris@42 30 /*
Chris@42 31 * This function contains 48 FP additions, 42 FP multiplications,
Chris@42 32 * (or, 18 additions, 12 multiplications, 30 fused multiply/add),
Chris@42 33 * 56 stack variables, 2 constants, and 36 memory accesses
Chris@42 34 */
Chris@42 35 #include "q.h"
Chris@42 36
Chris@42 37 static void q1_3(R *rio, R *iio, const R *W, stride rs, stride vs, INT mb, INT me, INT ms)
Chris@42 38 {
Chris@42 39 DK(KP866025403, +0.866025403784438646763723170752936183471402627);
Chris@42 40 DK(KP500000000, +0.500000000000000000000000000000000000000000000);
Chris@42 41 {
Chris@42 42 INT m;
Chris@42 43 for (m = mb, W = W + (mb * 4); m < me; m = m + 1, rio = rio + ms, iio = iio + ms, W = W + 4, MAKE_VOLATILE_STRIDE(6, rs), MAKE_VOLATILE_STRIDE(0, vs)) {
Chris@42 44 E Tk, Tn, Tm, To, Tl;
Chris@42 45 {
Chris@42 46 E T1, Td, T4, Tg, Tp, T9, Te, T6, Tf, TB, TE, Ts, TZ, Tu, Tx;
Chris@42 47 E TC, TN, TO, TD, TV, T10, TP, Tq, Tr;
Chris@42 48 {
Chris@42 49 E T2, T3, T7, T8;
Chris@42 50 T1 = rio[0];
Chris@42 51 T2 = rio[WS(rs, 1)];
Chris@42 52 T3 = rio[WS(rs, 2)];
Chris@42 53 Td = iio[0];
Chris@42 54 T7 = iio[WS(rs, 1)];
Chris@42 55 T8 = iio[WS(rs, 2)];
Chris@42 56 T4 = T2 + T3;
Chris@42 57 Tg = T3 - T2;
Chris@42 58 Tp = rio[WS(vs, 1)];
Chris@42 59 T9 = T7 - T8;
Chris@42 60 Te = T7 + T8;
Chris@42 61 T6 = FNMS(KP500000000, T4, T1);
Chris@42 62 Tq = rio[WS(vs, 1) + WS(rs, 1)];
Chris@42 63 Tr = rio[WS(vs, 1) + WS(rs, 2)];
Chris@42 64 Tf = FNMS(KP500000000, Te, Td);
Chris@42 65 }
Chris@42 66 {
Chris@42 67 E Tv, Tw, TT, TU;
Chris@42 68 TB = iio[WS(vs, 1)];
Chris@42 69 Tv = iio[WS(vs, 1) + WS(rs, 1)];
Chris@42 70 TE = Tr - Tq;
Chris@42 71 Ts = Tq + Tr;
Chris@42 72 Tw = iio[WS(vs, 1) + WS(rs, 2)];
Chris@42 73 TZ = iio[WS(vs, 2)];
Chris@42 74 TT = iio[WS(vs, 2) + WS(rs, 1)];
Chris@42 75 Tu = FNMS(KP500000000, Ts, Tp);
Chris@42 76 Tx = Tv - Tw;
Chris@42 77 TC = Tv + Tw;
Chris@42 78 TU = iio[WS(vs, 2) + WS(rs, 2)];
Chris@42 79 TN = rio[WS(vs, 2)];
Chris@42 80 TO = rio[WS(vs, 2) + WS(rs, 1)];
Chris@42 81 TD = FNMS(KP500000000, TC, TB);
Chris@42 82 TV = TT - TU;
Chris@42 83 T10 = TT + TU;
Chris@42 84 TP = rio[WS(vs, 2) + WS(rs, 2)];
Chris@42 85 }
Chris@42 86 {
Chris@42 87 E T11, T12, TS, TQ;
Chris@42 88 rio[0] = T1 + T4;
Chris@42 89 iio[0] = Td + Te;
Chris@42 90 T11 = FNMS(KP500000000, T10, TZ);
Chris@42 91 T12 = TP - TO;
Chris@42 92 TQ = TO + TP;
Chris@42 93 rio[WS(rs, 1)] = Tp + Ts;
Chris@42 94 iio[WS(rs, 1)] = TB + TC;
Chris@42 95 iio[WS(rs, 2)] = TZ + T10;
Chris@42 96 TS = FNMS(KP500000000, TQ, TN);
Chris@42 97 rio[WS(rs, 2)] = TN + TQ;
Chris@42 98 {
Chris@42 99 E TW, T13, Ty, TI, TL, TF, TH, TK;
Chris@42 100 {
Chris@42 101 E Ta, Th, T5, Tc;
Chris@42 102 Tk = FNMS(KP866025403, T9, T6);
Chris@42 103 Ta = FMA(KP866025403, T9, T6);
Chris@42 104 Th = FMA(KP866025403, Tg, Tf);
Chris@42 105 Tn = FNMS(KP866025403, Tg, Tf);
Chris@42 106 T5 = W[0];
Chris@42 107 Tc = W[1];
Chris@42 108 {
Chris@42 109 E T16, T19, T18, T1a, T17, Ti, Tb, T15;
Chris@42 110 TW = FMA(KP866025403, TV, TS);
Chris@42 111 T16 = FNMS(KP866025403, TV, TS);
Chris@42 112 T19 = FNMS(KP866025403, T12, T11);
Chris@42 113 T13 = FMA(KP866025403, T12, T11);
Chris@42 114 Ti = T5 * Th;
Chris@42 115 Tb = T5 * Ta;
Chris@42 116 T15 = W[2];
Chris@42 117 T18 = W[3];
Chris@42 118 iio[WS(vs, 1)] = FNMS(Tc, Ta, Ti);
Chris@42 119 rio[WS(vs, 1)] = FMA(Tc, Th, Tb);
Chris@42 120 T1a = T15 * T19;
Chris@42 121 T17 = T15 * T16;
Chris@42 122 Ty = FMA(KP866025403, Tx, Tu);
Chris@42 123 TI = FNMS(KP866025403, Tx, Tu);
Chris@42 124 TL = FNMS(KP866025403, TE, TD);
Chris@42 125 TF = FMA(KP866025403, TE, TD);
Chris@42 126 iio[WS(vs, 2) + WS(rs, 2)] = FNMS(T18, T16, T1a);
Chris@42 127 rio[WS(vs, 2) + WS(rs, 2)] = FMA(T18, T19, T17);
Chris@42 128 TH = W[2];
Chris@42 129 TK = W[3];
Chris@42 130 }
Chris@42 131 }
Chris@42 132 {
Chris@42 133 E TA, TG, Tz, TM, TJ, Tt;
Chris@42 134 TM = TH * TL;
Chris@42 135 TJ = TH * TI;
Chris@42 136 Tt = W[0];
Chris@42 137 TA = W[1];
Chris@42 138 iio[WS(vs, 2) + WS(rs, 1)] = FNMS(TK, TI, TM);
Chris@42 139 rio[WS(vs, 2) + WS(rs, 1)] = FMA(TK, TL, TJ);
Chris@42 140 TG = Tt * TF;
Chris@42 141 Tz = Tt * Ty;
Chris@42 142 {
Chris@42 143 E TR, TY, T14, TX, Tj;
Chris@42 144 iio[WS(vs, 1) + WS(rs, 1)] = FNMS(TA, Ty, TG);
Chris@42 145 rio[WS(vs, 1) + WS(rs, 1)] = FMA(TA, TF, Tz);
Chris@42 146 TR = W[0];
Chris@42 147 TY = W[1];
Chris@42 148 T14 = TR * T13;
Chris@42 149 TX = TR * TW;
Chris@42 150 Tj = W[2];
Chris@42 151 Tm = W[3];
Chris@42 152 iio[WS(vs, 1) + WS(rs, 2)] = FNMS(TY, TW, T14);
Chris@42 153 rio[WS(vs, 1) + WS(rs, 2)] = FMA(TY, T13, TX);
Chris@42 154 To = Tj * Tn;
Chris@42 155 Tl = Tj * Tk;
Chris@42 156 }
Chris@42 157 }
Chris@42 158 }
Chris@42 159 }
Chris@42 160 }
Chris@42 161 iio[WS(vs, 2)] = FNMS(Tm, Tk, To);
Chris@42 162 rio[WS(vs, 2)] = FMA(Tm, Tn, Tl);
Chris@42 163 }
Chris@42 164 }
Chris@42 165 }
Chris@42 166
Chris@42 167 static const tw_instr twinstr[] = {
Chris@42 168 {TW_FULL, 0, 3},
Chris@42 169 {TW_NEXT, 1, 0}
Chris@42 170 };
Chris@42 171
Chris@42 172 static const ct_desc desc = { 3, "q1_3", twinstr, &GENUS, {18, 12, 30, 0}, 0, 0, 0 };
Chris@42 173
Chris@42 174 void X(codelet_q1_3) (planner *p) {
Chris@42 175 X(kdft_difsq_register) (p, q1_3, &desc);
Chris@42 176 }
Chris@42 177 #else /* HAVE_FMA */
Chris@42 178
Chris@42 179 /* Generated by: ../../../genfft/gen_twidsq.native -compact -variables 4 -pipeline-latency 4 -reload-twiddle -dif -n 3 -name q1_3 -include q.h */
Chris@42 180
Chris@42 181 /*
Chris@42 182 * This function contains 48 FP additions, 36 FP multiplications,
Chris@42 183 * (or, 30 additions, 18 multiplications, 18 fused multiply/add),
Chris@42 184 * 35 stack variables, 2 constants, and 36 memory accesses
Chris@42 185 */
Chris@42 186 #include "q.h"
Chris@42 187
Chris@42 188 static void q1_3(R *rio, R *iio, const R *W, stride rs, stride vs, INT mb, INT me, INT ms)
Chris@42 189 {
Chris@42 190 DK(KP866025403, +0.866025403784438646763723170752936183471402627);
Chris@42 191 DK(KP500000000, +0.500000000000000000000000000000000000000000000);
Chris@42 192 {
Chris@42 193 INT m;
Chris@42 194 for (m = mb, W = W + (mb * 4); m < me; m = m + 1, rio = rio + ms, iio = iio + ms, W = W + 4, MAKE_VOLATILE_STRIDE(6, rs), MAKE_VOLATILE_STRIDE(0, vs)) {
Chris@42 195 E T1, T4, T6, Tc, Td, Te, T9, Tf, Tl, To, Tq, Tw, Tx, Ty, Tt;
Chris@42 196 E Tz, TR, TS, TN, TT, TF, TI, TK, TQ;
Chris@42 197 {
Chris@42 198 E T2, T3, Tr, Ts;
Chris@42 199 T1 = rio[0];
Chris@42 200 T2 = rio[WS(rs, 1)];
Chris@42 201 T3 = rio[WS(rs, 2)];
Chris@42 202 T4 = T2 + T3;
Chris@42 203 T6 = FNMS(KP500000000, T4, T1);
Chris@42 204 Tc = KP866025403 * (T3 - T2);
Chris@42 205 {
Chris@42 206 E T7, T8, Tm, Tn;
Chris@42 207 Td = iio[0];
Chris@42 208 T7 = iio[WS(rs, 1)];
Chris@42 209 T8 = iio[WS(rs, 2)];
Chris@42 210 Te = T7 + T8;
Chris@42 211 T9 = KP866025403 * (T7 - T8);
Chris@42 212 Tf = FNMS(KP500000000, Te, Td);
Chris@42 213 Tl = rio[WS(vs, 1)];
Chris@42 214 Tm = rio[WS(vs, 1) + WS(rs, 1)];
Chris@42 215 Tn = rio[WS(vs, 1) + WS(rs, 2)];
Chris@42 216 To = Tm + Tn;
Chris@42 217 Tq = FNMS(KP500000000, To, Tl);
Chris@42 218 Tw = KP866025403 * (Tn - Tm);
Chris@42 219 }
Chris@42 220 Tx = iio[WS(vs, 1)];
Chris@42 221 Tr = iio[WS(vs, 1) + WS(rs, 1)];
Chris@42 222 Ts = iio[WS(vs, 1) + WS(rs, 2)];
Chris@42 223 Ty = Tr + Ts;
Chris@42 224 Tt = KP866025403 * (Tr - Ts);
Chris@42 225 Tz = FNMS(KP500000000, Ty, Tx);
Chris@42 226 {
Chris@42 227 E TL, TM, TG, TH;
Chris@42 228 TR = iio[WS(vs, 2)];
Chris@42 229 TL = iio[WS(vs, 2) + WS(rs, 1)];
Chris@42 230 TM = iio[WS(vs, 2) + WS(rs, 2)];
Chris@42 231 TS = TL + TM;
Chris@42 232 TN = KP866025403 * (TL - TM);
Chris@42 233 TT = FNMS(KP500000000, TS, TR);
Chris@42 234 TF = rio[WS(vs, 2)];
Chris@42 235 TG = rio[WS(vs, 2) + WS(rs, 1)];
Chris@42 236 TH = rio[WS(vs, 2) + WS(rs, 2)];
Chris@42 237 TI = TG + TH;
Chris@42 238 TK = FNMS(KP500000000, TI, TF);
Chris@42 239 TQ = KP866025403 * (TH - TG);
Chris@42 240 }
Chris@42 241 }
Chris@42 242 rio[0] = T1 + T4;
Chris@42 243 iio[0] = Td + Te;
Chris@42 244 rio[WS(rs, 1)] = Tl + To;
Chris@42 245 iio[WS(rs, 1)] = Tx + Ty;
Chris@42 246 iio[WS(rs, 2)] = TR + TS;
Chris@42 247 rio[WS(rs, 2)] = TF + TI;
Chris@42 248 {
Chris@42 249 E Ta, Tg, T5, Tb;
Chris@42 250 Ta = T6 + T9;
Chris@42 251 Tg = Tc + Tf;
Chris@42 252 T5 = W[0];
Chris@42 253 Tb = W[1];
Chris@42 254 rio[WS(vs, 1)] = FMA(T5, Ta, Tb * Tg);
Chris@42 255 iio[WS(vs, 1)] = FNMS(Tb, Ta, T5 * Tg);
Chris@42 256 }
Chris@42 257 {
Chris@42 258 E TW, TY, TV, TX;
Chris@42 259 TW = TK - TN;
Chris@42 260 TY = TT - TQ;
Chris@42 261 TV = W[2];
Chris@42 262 TX = W[3];
Chris@42 263 rio[WS(vs, 2) + WS(rs, 2)] = FMA(TV, TW, TX * TY);
Chris@42 264 iio[WS(vs, 2) + WS(rs, 2)] = FNMS(TX, TW, TV * TY);
Chris@42 265 }
Chris@42 266 {
Chris@42 267 E TC, TE, TB, TD;
Chris@42 268 TC = Tq - Tt;
Chris@42 269 TE = Tz - Tw;
Chris@42 270 TB = W[2];
Chris@42 271 TD = W[3];
Chris@42 272 rio[WS(vs, 2) + WS(rs, 1)] = FMA(TB, TC, TD * TE);
Chris@42 273 iio[WS(vs, 2) + WS(rs, 1)] = FNMS(TD, TC, TB * TE);
Chris@42 274 }
Chris@42 275 {
Chris@42 276 E Tu, TA, Tp, Tv;
Chris@42 277 Tu = Tq + Tt;
Chris@42 278 TA = Tw + Tz;
Chris@42 279 Tp = W[0];
Chris@42 280 Tv = W[1];
Chris@42 281 rio[WS(vs, 1) + WS(rs, 1)] = FMA(Tp, Tu, Tv * TA);
Chris@42 282 iio[WS(vs, 1) + WS(rs, 1)] = FNMS(Tv, Tu, Tp * TA);
Chris@42 283 }
Chris@42 284 {
Chris@42 285 E TO, TU, TJ, TP;
Chris@42 286 TO = TK + TN;
Chris@42 287 TU = TQ + TT;
Chris@42 288 TJ = W[0];
Chris@42 289 TP = W[1];
Chris@42 290 rio[WS(vs, 1) + WS(rs, 2)] = FMA(TJ, TO, TP * TU);
Chris@42 291 iio[WS(vs, 1) + WS(rs, 2)] = FNMS(TP, TO, TJ * TU);
Chris@42 292 }
Chris@42 293 {
Chris@42 294 E Ti, Tk, Th, Tj;
Chris@42 295 Ti = T6 - T9;
Chris@42 296 Tk = Tf - Tc;
Chris@42 297 Th = W[2];
Chris@42 298 Tj = W[3];
Chris@42 299 rio[WS(vs, 2)] = FMA(Th, Ti, Tj * Tk);
Chris@42 300 iio[WS(vs, 2)] = FNMS(Tj, Ti, Th * Tk);
Chris@42 301 }
Chris@42 302 }
Chris@42 303 }
Chris@42 304 }
Chris@42 305
Chris@42 306 static const tw_instr twinstr[] = {
Chris@42 307 {TW_FULL, 0, 3},
Chris@42 308 {TW_NEXT, 1, 0}
Chris@42 309 };
Chris@42 310
Chris@42 311 static const ct_desc desc = { 3, "q1_3", twinstr, &GENUS, {30, 18, 18, 0}, 0, 0, 0 };
Chris@42 312
Chris@42 313 void X(codelet_q1_3) (planner *p) {
Chris@42 314 X(kdft_difsq_register) (p, q1_3, &desc);
Chris@42 315 }
Chris@42 316 #endif /* HAVE_FMA */