furnace/extern/fftw/dft/scalar/codelets/n1_14.c

514 lines
16 KiB
C
Raw Normal View History

/*
* Copyright (c) 2003, 2007-14 Matteo Frigo
* Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*
*/
/* This file was automatically generated --- DO NOT EDIT */
/* Generated on Tue Sep 14 10:44:24 EDT 2021 */
#include "dft/codelet-dft.h"
#if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA)
/* Generated by: ../../../genfft/gen_notw.native -fma -compact -variables 4 -pipeline-latency 4 -n 14 -name n1_14 -include dft/scalar/n.h */
/*
* This function contains 148 FP additions, 84 FP multiplications,
* (or, 64 additions, 0 multiplications, 84 fused multiply/add),
* 67 stack variables, 6 constants, and 56 memory accesses
*/
#include "dft/scalar/n.h"
static void n1_14(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs)
{
DK(KP974927912, +0.974927912181823607018131682993931217232785801);
DK(KP801937735, +0.801937735804838252472204639014890102331838324);
DK(KP554958132, +0.554958132087371191422194871006410481067288862);
DK(KP900968867, +0.900968867902419126236102319507445051165919162);
DK(KP692021471, +0.692021471630095869627814897002069140197260599);
DK(KP356895867, +0.356895867892209443894399510021300583399127187);
{
INT i;
for (i = v; i > 0; i = i - 1, ri = ri + ivs, ii = ii + ivs, ro = ro + ovs, io = io + ovs, MAKE_VOLATILE_STRIDE(56, is), MAKE_VOLATILE_STRIDE(56, os)) {
E T3, Tp, T1b, T1x, T1i, T1L, T1M, T1j, T1k, T1K, Ta, To, Th, Tz, T14;
E TZ, Ts, Ty, Tv, T1Z, T2c, T27, TI, T23, T24, TP, TW, T22, T1c, T1e;
E T1d, T1f, T1s, T1n, T1A, T1G, T1D, T1H, T1U, T1P;
{
E T1, T2, T19, T1a;
T1 = ri[0];
T2 = ri[WS(is, 7)];
T3 = T1 - T2;
Tp = T1 + T2;
T19 = ii[0];
T1a = ii[WS(is, 7)];
T1b = T19 - T1a;
T1x = T19 + T1a;
}
{
E T6, Tq, T9, Tr, Tn, Tx, Tk, Tw, Tg, Tu, Td, Tt;
{
E T4, T5, Ti, Tj;
T4 = ri[WS(is, 2)];
T5 = ri[WS(is, 9)];
T6 = T4 - T5;
Tq = T4 + T5;
{
E T7, T8, Tl, Tm;
T7 = ri[WS(is, 12)];
T8 = ri[WS(is, 5)];
T9 = T7 - T8;
Tr = T7 + T8;
Tl = ri[WS(is, 8)];
Tm = ri[WS(is, 1)];
Tn = Tl - Tm;
Tx = Tl + Tm;
}
Ti = ri[WS(is, 6)];
Tj = ri[WS(is, 13)];
Tk = Ti - Tj;
Tw = Ti + Tj;
{
E Te, Tf, Tb, Tc;
Te = ri[WS(is, 10)];
Tf = ri[WS(is, 3)];
Tg = Te - Tf;
Tu = Te + Tf;
Tb = ri[WS(is, 4)];
Tc = ri[WS(is, 11)];
Td = Tb - Tc;
Tt = Tb + Tc;
}
}
T1i = Tn - Tk;
T1L = Tt - Tu;
T1M = Tr - Tq;
T1j = Tg - Td;
T1k = T9 - T6;
T1K = Tw - Tx;
Ta = T6 + T9;
To = Tk + Tn;
Th = Td + Tg;
Tz = FNMS(KP356895867, Th, Ta);
T14 = FNMS(KP356895867, To, Th);
TZ = FNMS(KP356895867, Ta, To);
Ts = Tq + Tr;
Ty = Tw + Tx;
Tv = Tt + Tu;
T1Z = FNMS(KP356895867, Ts, Ty);
T2c = FNMS(KP356895867, Ty, Tv);
T27 = FNMS(KP356895867, Tv, Ts);
}
{
E TE, T1B, TH, T1C, TV, T1F, TS, T1E, TO, T1z, TL, T1y;
{
E TC, TD, TQ, TR;
TC = ii[WS(is, 4)];
TD = ii[WS(is, 11)];
TE = TC - TD;
T1B = TC + TD;
{
E TF, TG, TT, TU;
TF = ii[WS(is, 10)];
TG = ii[WS(is, 3)];
TH = TF - TG;
T1C = TF + TG;
TT = ii[WS(is, 8)];
TU = ii[WS(is, 1)];
TV = TT - TU;
T1F = TT + TU;
}
TQ = ii[WS(is, 6)];
TR = ii[WS(is, 13)];
TS = TQ - TR;
T1E = TQ + TR;
{
E TM, TN, TJ, TK;
TM = ii[WS(is, 12)];
TN = ii[WS(is, 5)];
TO = TM - TN;
T1z = TM + TN;
TJ = ii[WS(is, 2)];
TK = ii[WS(is, 9)];
TL = TJ - TK;
T1y = TJ + TK;
}
}
TI = TE - TH;
T23 = T1F - T1E;
T24 = T1C - T1B;
TP = TL - TO;
TW = TS - TV;
T22 = T1y - T1z;
T1c = TL + TO;
T1e = TS + TV;
T1d = TE + TH;
T1f = FNMS(KP356895867, T1e, T1d);
T1s = FNMS(KP356895867, T1d, T1c);
T1n = FNMS(KP356895867, T1c, T1e);
T1A = T1y + T1z;
T1G = T1E + T1F;
T1D = T1B + T1C;
T1H = FNMS(KP356895867, T1G, T1D);
T1U = FNMS(KP356895867, T1D, T1A);
T1P = FNMS(KP356895867, T1A, T1G);
}
ro[WS(os, 7)] = T3 + Ta + Th + To;
io[WS(os, 7)] = T1b + T1c + T1d + T1e;
ro[0] = Tp + Ts + Tv + Ty;
io[0] = T1x + T1A + T1D + T1G;
{
E TB, TY, TA, TX;
TA = FNMS(KP692021471, Tz, To);
TB = FNMS(KP900968867, TA, T3);
TX = FMA(KP554958132, TW, TP);
TY = FMA(KP801937735, TX, TI);
ro[WS(os, 13)] = FNMS(KP974927912, TY, TB);
ro[WS(os, 1)] = FMA(KP974927912, TY, TB);
}
{
E T1u, T1w, T1t, T1v;
T1t = FNMS(KP692021471, T1s, T1e);
T1u = FNMS(KP900968867, T1t, T1b);
T1v = FMA(KP554958132, T1i, T1k);
T1w = FMA(KP801937735, T1v, T1j);
io[WS(os, 1)] = FMA(KP974927912, T1w, T1u);
io[WS(os, 13)] = FNMS(KP974927912, T1w, T1u);
}
{
E T11, T13, T10, T12;
T10 = FNMS(KP692021471, TZ, Th);
T11 = FNMS(KP900968867, T10, T3);
T12 = FMA(KP554958132, TI, TW);
T13 = FNMS(KP801937735, T12, TP);
ro[WS(os, 5)] = FNMS(KP974927912, T13, T11);
ro[WS(os, 9)] = FMA(KP974927912, T13, T11);
}
{
E T1p, T1r, T1o, T1q;
T1o = FNMS(KP692021471, T1n, T1d);
T1p = FNMS(KP900968867, T1o, T1b);
T1q = FMA(KP554958132, T1j, T1i);
T1r = FNMS(KP801937735, T1q, T1k);
io[WS(os, 5)] = FNMS(KP974927912, T1r, T1p);
io[WS(os, 9)] = FMA(KP974927912, T1r, T1p);
}
{
E T16, T18, T15, T17;
T15 = FNMS(KP692021471, T14, Ta);
T16 = FNMS(KP900968867, T15, T3);
T17 = FNMS(KP554958132, TP, TI);
T18 = FNMS(KP801937735, T17, TW);
ro[WS(os, 11)] = FNMS(KP974927912, T18, T16);
ro[WS(os, 3)] = FMA(KP974927912, T18, T16);
}
{
E T1h, T1m, T1g, T1l;
T1g = FNMS(KP692021471, T1f, T1c);
T1h = FNMS(KP900968867, T1g, T1b);
T1l = FNMS(KP554958132, T1k, T1j);
T1m = FNMS(KP801937735, T1l, T1i);
io[WS(os, 3)] = FMA(KP974927912, T1m, T1h);
io[WS(os, 11)] = FNMS(KP974927912, T1m, T1h);
}
{
E T1J, T1O, T1I, T1N;
T1I = FNMS(KP692021471, T1H, T1A);
T1J = FNMS(KP900968867, T1I, T1x);
T1N = FMA(KP554958132, T1M, T1L);
T1O = FNMS(KP801937735, T1N, T1K);
io[WS(os, 4)] = FMA(KP974927912, T1O, T1J);
io[WS(os, 10)] = FNMS(KP974927912, T1O, T1J);
}
{
E T2e, T2g, T2d, T2f;
T2d = FNMS(KP692021471, T2c, Ts);
T2e = FNMS(KP900968867, T2d, Tp);
T2f = FMA(KP554958132, T22, T24);
T2g = FNMS(KP801937735, T2f, T23);
ro[WS(os, 10)] = FNMS(KP974927912, T2g, T2e);
ro[WS(os, 4)] = FMA(KP974927912, T2g, T2e);
}
{
E T1R, T1T, T1Q, T1S;
T1Q = FNMS(KP692021471, T1P, T1D);
T1R = FNMS(KP900968867, T1Q, T1x);
T1S = FMA(KP554958132, T1L, T1K);
T1T = FMA(KP801937735, T1S, T1M);
io[WS(os, 2)] = FMA(KP974927912, T1T, T1R);
io[WS(os, 12)] = FNMS(KP974927912, T1T, T1R);
}
{
E T21, T26, T20, T25;
T20 = FNMS(KP692021471, T1Z, Tv);
T21 = FNMS(KP900968867, T20, Tp);
T25 = FMA(KP554958132, T24, T23);
T26 = FMA(KP801937735, T25, T22);
ro[WS(os, 12)] = FNMS(KP974927912, T26, T21);
ro[WS(os, 2)] = FMA(KP974927912, T26, T21);
}
{
E T1W, T1Y, T1V, T1X;
T1V = FNMS(KP692021471, T1U, T1G);
T1W = FNMS(KP900968867, T1V, T1x);
T1X = FNMS(KP554958132, T1K, T1M);
T1Y = FNMS(KP801937735, T1X, T1L);
io[WS(os, 6)] = FMA(KP974927912, T1Y, T1W);
io[WS(os, 8)] = FNMS(KP974927912, T1Y, T1W);
}
{
E T29, T2b, T28, T2a;
T28 = FNMS(KP692021471, T27, Ty);
T29 = FNMS(KP900968867, T28, Tp);
T2a = FNMS(KP554958132, T23, T22);
T2b = FNMS(KP801937735, T2a, T24);
ro[WS(os, 8)] = FNMS(KP974927912, T2b, T29);
ro[WS(os, 6)] = FMA(KP974927912, T2b, T29);
}
}
}
}
static const kdft_desc desc = { 14, "n1_14", { 64, 0, 84, 0 }, &GENUS, 0, 0, 0, 0 };
void X(codelet_n1_14) (planner *p) { X(kdft_register) (p, n1_14, &desc);
}
#else
/* Generated by: ../../../genfft/gen_notw.native -compact -variables 4 -pipeline-latency 4 -n 14 -name n1_14 -include dft/scalar/n.h */
/*
* This function contains 148 FP additions, 72 FP multiplications,
* (or, 100 additions, 24 multiplications, 48 fused multiply/add),
* 43 stack variables, 6 constants, and 56 memory accesses
*/
#include "dft/scalar/n.h"
static void n1_14(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs)
{
DK(KP222520933, +0.222520933956314404288902564496794759466355569);
DK(KP900968867, +0.900968867902419126236102319507445051165919162);
DK(KP623489801, +0.623489801858733530525004884004239810632274731);
DK(KP433883739, +0.433883739117558120475768332848358754609990728);
DK(KP781831482, +0.781831482468029808708444526674057750232334519);
DK(KP974927912, +0.974927912181823607018131682993931217232785801);
{
INT i;
for (i = v; i > 0; i = i - 1, ri = ri + ivs, ii = ii + ivs, ro = ro + ovs, io = io + ovs, MAKE_VOLATILE_STRIDE(56, is), MAKE_VOLATILE_STRIDE(56, os)) {
E T3, Tp, T16, T1f, Ta, T1q, Ts, T10, TG, T1z, T19, T1i, Th, T1s, Tv;
E T12, TU, T1B, T17, T1o, To, T1r, Ty, T11, TN, T1A, T18, T1l;
{
E T1, T2, T14, T15;
T1 = ri[0];
T2 = ri[WS(is, 7)];
T3 = T1 - T2;
Tp = T1 + T2;
T14 = ii[0];
T15 = ii[WS(is, 7)];
T16 = T14 - T15;
T1f = T14 + T15;
}
{
E T6, Tq, T9, Tr;
{
E T4, T5, T7, T8;
T4 = ri[WS(is, 2)];
T5 = ri[WS(is, 9)];
T6 = T4 - T5;
Tq = T4 + T5;
T7 = ri[WS(is, 12)];
T8 = ri[WS(is, 5)];
T9 = T7 - T8;
Tr = T7 + T8;
}
Ta = T6 + T9;
T1q = Tr - Tq;
Ts = Tq + Tr;
T10 = T9 - T6;
}
{
E TC, T1g, TF, T1h;
{
E TA, TB, TD, TE;
TA = ii[WS(is, 2)];
TB = ii[WS(is, 9)];
TC = TA - TB;
T1g = TA + TB;
TD = ii[WS(is, 12)];
TE = ii[WS(is, 5)];
TF = TD - TE;
T1h = TD + TE;
}
TG = TC - TF;
T1z = T1g - T1h;
T19 = TC + TF;
T1i = T1g + T1h;
}
{
E Td, Tt, Tg, Tu;
{
E Tb, Tc, Te, Tf;
Tb = ri[WS(is, 4)];
Tc = ri[WS(is, 11)];
Td = Tb - Tc;
Tt = Tb + Tc;
Te = ri[WS(is, 10)];
Tf = ri[WS(is, 3)];
Tg = Te - Tf;
Tu = Te + Tf;
}
Th = Td + Tg;
T1s = Tt - Tu;
Tv = Tt + Tu;
T12 = Tg - Td;
}
{
E TQ, T1m, TT, T1n;
{
E TO, TP, TR, TS;
TO = ii[WS(is, 4)];
TP = ii[WS(is, 11)];
TQ = TO - TP;
T1m = TO + TP;
TR = ii[WS(is, 10)];
TS = ii[WS(is, 3)];
TT = TR - TS;
T1n = TR + TS;
}
TU = TQ - TT;
T1B = T1n - T1m;
T17 = TQ + TT;
T1o = T1m + T1n;
}
{
E Tk, Tw, Tn, Tx;
{
E Ti, Tj, Tl, Tm;
Ti = ri[WS(is, 6)];
Tj = ri[WS(is, 13)];
Tk = Ti - Tj;
Tw = Ti + Tj;
Tl = ri[WS(is, 8)];
Tm = ri[WS(is, 1)];
Tn = Tl - Tm;
Tx = Tl + Tm;
}
To = Tk + Tn;
T1r = Tw - Tx;
Ty = Tw + Tx;
T11 = Tn - Tk;
}
{
E TJ, T1j, TM, T1k;
{
E TH, TI, TK, TL;
TH = ii[WS(is, 6)];
TI = ii[WS(is, 13)];
TJ = TH - TI;
T1j = TH + TI;
TK = ii[WS(is, 8)];
TL = ii[WS(is, 1)];
TM = TK - TL;
T1k = TK + TL;
}
TN = TJ - TM;
T1A = T1k - T1j;
T18 = TJ + TM;
T1l = T1j + T1k;
}
ro[WS(os, 7)] = T3 + Ta + Th + To;
io[WS(os, 7)] = T16 + T19 + T17 + T18;
ro[0] = Tp + Ts + Tv + Ty;
io[0] = T1f + T1i + T1o + T1l;
{
E TV, Tz, T1e, T1d;
TV = FNMS(KP781831482, TN, KP974927912 * TG) - (KP433883739 * TU);
Tz = FMA(KP623489801, To, T3) + FNMA(KP900968867, Th, KP222520933 * Ta);
ro[WS(os, 5)] = Tz - TV;
ro[WS(os, 9)] = Tz + TV;
T1e = FNMS(KP781831482, T11, KP974927912 * T10) - (KP433883739 * T12);
T1d = FMA(KP623489801, T18, T16) + FNMA(KP900968867, T17, KP222520933 * T19);
io[WS(os, 5)] = T1d - T1e;
io[WS(os, 9)] = T1e + T1d;
}
{
E TX, TW, T1b, T1c;
TX = FMA(KP781831482, TG, KP974927912 * TU) + (KP433883739 * TN);
TW = FMA(KP623489801, Ta, T3) + FNMA(KP900968867, To, KP222520933 * Th);
ro[WS(os, 13)] = TW - TX;
ro[WS(os, 1)] = TW + TX;
T1b = FMA(KP781831482, T10, KP974927912 * T12) + (KP433883739 * T11);
T1c = FMA(KP623489801, T19, T16) + FNMA(KP900968867, T18, KP222520933 * T17);
io[WS(os, 1)] = T1b + T1c;
io[WS(os, 13)] = T1c - T1b;
}
{
E TZ, TY, T13, T1a;
TZ = FMA(KP433883739, TG, KP974927912 * TN) - (KP781831482 * TU);
TY = FMA(KP623489801, Th, T3) + FNMA(KP222520933, To, KP900968867 * Ta);
ro[WS(os, 11)] = TY - TZ;
ro[WS(os, 3)] = TY + TZ;
T13 = FMA(KP433883739, T10, KP974927912 * T11) - (KP781831482 * T12);
T1a = FMA(KP623489801, T17, T16) + FNMA(KP222520933, T18, KP900968867 * T19);
io[WS(os, 3)] = T13 + T1a;
io[WS(os, 11)] = T1a - T13;
}
{
E T1t, T1p, T1C, T1y;
T1t = FNMS(KP433883739, T1r, KP781831482 * T1q) - (KP974927912 * T1s);
T1p = FMA(KP623489801, T1i, T1f) + FNMA(KP900968867, T1l, KP222520933 * T1o);
io[WS(os, 6)] = T1p - T1t;
io[WS(os, 8)] = T1t + T1p;
T1C = FNMS(KP433883739, T1A, KP781831482 * T1z) - (KP974927912 * T1B);
T1y = FMA(KP623489801, Ts, Tp) + FNMA(KP900968867, Ty, KP222520933 * Tv);
ro[WS(os, 6)] = T1y - T1C;
ro[WS(os, 8)] = T1y + T1C;
}
{
E T1v, T1u, T1E, T1D;
T1v = FMA(KP433883739, T1q, KP781831482 * T1s) - (KP974927912 * T1r);
T1u = FMA(KP623489801, T1o, T1f) + FNMA(KP222520933, T1l, KP900968867 * T1i);
io[WS(os, 4)] = T1u - T1v;
io[WS(os, 10)] = T1v + T1u;
T1E = FMA(KP433883739, T1z, KP781831482 * T1B) - (KP974927912 * T1A);
T1D = FMA(KP623489801, Tv, Tp) + FNMA(KP222520933, Ty, KP900968867 * Ts);
ro[WS(os, 4)] = T1D - T1E;
ro[WS(os, 10)] = T1D + T1E;
}
{
E T1w, T1x, T1G, T1F;
T1w = FMA(KP974927912, T1q, KP433883739 * T1s) + (KP781831482 * T1r);
T1x = FMA(KP623489801, T1l, T1f) + FNMA(KP900968867, T1o, KP222520933 * T1i);
io[WS(os, 2)] = T1w + T1x;
io[WS(os, 12)] = T1x - T1w;
T1G = FMA(KP974927912, T1z, KP433883739 * T1B) + (KP781831482 * T1A);
T1F = FMA(KP623489801, Ty, Tp) + FNMA(KP900968867, Tv, KP222520933 * Ts);
ro[WS(os, 12)] = T1F - T1G;
ro[WS(os, 2)] = T1F + T1G;
}
}
}
}
static const kdft_desc desc = { 14, "n1_14", { 100, 24, 48, 0 }, &GENUS, 0, 0, 0, 0 };
void X(codelet_n1_14) (planner *p) { X(kdft_register) (p, n1_14, &desc);
}
#endif