mirror of
https://github.com/tildearrow/furnace.git
synced 2024-11-14 16:55:06 +00:00
555 lines
17 KiB
C
555 lines
17 KiB
C
|
/*
|
||
|
* 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 15 -name n1_15 -include dft/scalar/n.h */
|
||
|
|
||
|
/*
|
||
|
* This function contains 156 FP additions, 84 FP multiplications,
|
||
|
* (or, 72 additions, 0 multiplications, 84 fused multiply/add),
|
||
|
* 69 stack variables, 6 constants, and 60 memory accesses
|
||
|
*/
|
||
|
#include "dft/scalar/n.h"
|
||
|
|
||
|
static void n1_15(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs)
|
||
|
{
|
||
|
DK(KP951056516, +0.951056516295153572116439333379382143405698634);
|
||
|
DK(KP559016994, +0.559016994374947424102293417182819058860154590);
|
||
|
DK(KP618033988, +0.618033988749894848204586834365638117720309180);
|
||
|
DK(KP250000000, +0.250000000000000000000000000000000000000000000);
|
||
|
DK(KP866025403, +0.866025403784438646763723170752936183471402627);
|
||
|
DK(KP500000000, +0.500000000000000000000000000000000000000000000);
|
||
|
{
|
||
|
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(60, is), MAKE_VOLATILE_STRIDE(60, os)) {
|
||
|
E T5, T2l, Tx, TV, T1z, T1X, Tl, Tq, Tr, TN, TS, TT, T2c, T2d, T2n;
|
||
|
E T1O, T1P, T1Z, T1l, T1q, T1B, TZ, T10, T11, Ta, Tf, Tg, TC, TH, TI;
|
||
|
E T2f, T2g, T2m, T1R, T1S, T1Y, T1a, T1f, T1A, TW, TX, TY;
|
||
|
{
|
||
|
E T1, T1v, T4, T1y, Tw, T1w, Tt, T1x;
|
||
|
T1 = ri[0];
|
||
|
T1v = ii[0];
|
||
|
{
|
||
|
E T2, T3, Tu, Tv;
|
||
|
T2 = ri[WS(is, 5)];
|
||
|
T3 = ri[WS(is, 10)];
|
||
|
T4 = T2 + T3;
|
||
|
T1y = T3 - T2;
|
||
|
Tu = ii[WS(is, 5)];
|
||
|
Tv = ii[WS(is, 10)];
|
||
|
Tw = Tu - Tv;
|
||
|
T1w = Tu + Tv;
|
||
|
}
|
||
|
T5 = T1 + T4;
|
||
|
T2l = T1v + T1w;
|
||
|
Tt = FNMS(KP500000000, T4, T1);
|
||
|
Tx = FNMS(KP866025403, Tw, Tt);
|
||
|
TV = FMA(KP866025403, Tw, Tt);
|
||
|
T1x = FNMS(KP500000000, T1w, T1v);
|
||
|
T1z = FMA(KP866025403, T1y, T1x);
|
||
|
T1X = FNMS(KP866025403, T1y, T1x);
|
||
|
}
|
||
|
{
|
||
|
E Th, Tk, TJ, T1k, T1h, T1i, TM, T1j, Tm, Tp, TO, T1p, T1m, T1n, TR;
|
||
|
E T1o;
|
||
|
{
|
||
|
E Ti, Tj, TK, TL;
|
||
|
Th = ri[WS(is, 6)];
|
||
|
Ti = ri[WS(is, 11)];
|
||
|
Tj = ri[WS(is, 1)];
|
||
|
Tk = Ti + Tj;
|
||
|
TJ = FNMS(KP500000000, Tk, Th);
|
||
|
T1k = Tj - Ti;
|
||
|
T1h = ii[WS(is, 6)];
|
||
|
TK = ii[WS(is, 11)];
|
||
|
TL = ii[WS(is, 1)];
|
||
|
T1i = TK + TL;
|
||
|
TM = TK - TL;
|
||
|
T1j = FNMS(KP500000000, T1i, T1h);
|
||
|
}
|
||
|
{
|
||
|
E Tn, To, TP, TQ;
|
||
|
Tm = ri[WS(is, 9)];
|
||
|
Tn = ri[WS(is, 14)];
|
||
|
To = ri[WS(is, 4)];
|
||
|
Tp = Tn + To;
|
||
|
TO = FNMS(KP500000000, Tp, Tm);
|
||
|
T1p = To - Tn;
|
||
|
T1m = ii[WS(is, 9)];
|
||
|
TP = ii[WS(is, 14)];
|
||
|
TQ = ii[WS(is, 4)];
|
||
|
T1n = TP + TQ;
|
||
|
TR = TP - TQ;
|
||
|
T1o = FNMS(KP500000000, T1n, T1m);
|
||
|
}
|
||
|
Tl = Th + Tk;
|
||
|
Tq = Tm + Tp;
|
||
|
Tr = Tl + Tq;
|
||
|
TN = FNMS(KP866025403, TM, TJ);
|
||
|
TS = FNMS(KP866025403, TR, TO);
|
||
|
TT = TN + TS;
|
||
|
T2c = T1h + T1i;
|
||
|
T2d = T1m + T1n;
|
||
|
T2n = T2c + T2d;
|
||
|
T1O = FNMS(KP866025403, T1k, T1j);
|
||
|
T1P = FNMS(KP866025403, T1p, T1o);
|
||
|
T1Z = T1O + T1P;
|
||
|
T1l = FMA(KP866025403, T1k, T1j);
|
||
|
T1q = FMA(KP866025403, T1p, T1o);
|
||
|
T1B = T1l + T1q;
|
||
|
TZ = FMA(KP866025403, TM, TJ);
|
||
|
T10 = FMA(KP866025403, TR, TO);
|
||
|
T11 = TZ + T10;
|
||
|
}
|
||
|
{
|
||
|
E T6, T9, Ty, T19, T16, T17, TB, T18, Tb, Te, TD, T1e, T1b, T1c, TG;
|
||
|
E T1d;
|
||
|
{
|
||
|
E T7, T8, Tz, TA;
|
||
|
T6 = ri[WS(is, 3)];
|
||
|
T7 = ri[WS(is, 8)];
|
||
|
T8 = ri[WS(is, 13)];
|
||
|
T9 = T7 + T8;
|
||
|
Ty = FNMS(KP500000000, T9, T6);
|
||
|
T19 = T8 - T7;
|
||
|
T16 = ii[WS(is, 3)];
|
||
|
Tz = ii[WS(is, 8)];
|
||
|
TA = ii[WS(is, 13)];
|
||
|
T17 = Tz + TA;
|
||
|
TB = Tz - TA;
|
||
|
T18 = FNMS(KP500000000, T17, T16);
|
||
|
}
|
||
|
{
|
||
|
E Tc, Td, TE, TF;
|
||
|
Tb = ri[WS(is, 12)];
|
||
|
Tc = ri[WS(is, 2)];
|
||
|
Td = ri[WS(is, 7)];
|
||
|
Te = Tc + Td;
|
||
|
TD = FNMS(KP500000000, Te, Tb);
|
||
|
T1e = Td - Tc;
|
||
|
T1b = ii[WS(is, 12)];
|
||
|
TE = ii[WS(is, 2)];
|
||
|
TF = ii[WS(is, 7)];
|
||
|
T1c = TE + TF;
|
||
|
TG = TE - TF;
|
||
|
T1d = FNMS(KP500000000, T1c, T1b);
|
||
|
}
|
||
|
Ta = T6 + T9;
|
||
|
Tf = Tb + Te;
|
||
|
Tg = Ta + Tf;
|
||
|
TC = FNMS(KP866025403, TB, Ty);
|
||
|
TH = FNMS(KP866025403, TG, TD);
|
||
|
TI = TC + TH;
|
||
|
T2f = T16 + T17;
|
||
|
T2g = T1b + T1c;
|
||
|
T2m = T2f + T2g;
|
||
|
T1R = FNMS(KP866025403, T19, T18);
|
||
|
T1S = FNMS(KP866025403, T1e, T1d);
|
||
|
T1Y = T1R + T1S;
|
||
|
T1a = FMA(KP866025403, T19, T18);
|
||
|
T1f = FMA(KP866025403, T1e, T1d);
|
||
|
T1A = T1a + T1f;
|
||
|
TW = FMA(KP866025403, TB, Ty);
|
||
|
TX = FMA(KP866025403, TG, TD);
|
||
|
TY = TW + TX;
|
||
|
}
|
||
|
{
|
||
|
E T2a, Ts, T29, T2i, T2k, T2e, T2h, T2j, T2b;
|
||
|
T2a = Tg - Tr;
|
||
|
Ts = Tg + Tr;
|
||
|
T29 = FNMS(KP250000000, Ts, T5);
|
||
|
T2e = T2c - T2d;
|
||
|
T2h = T2f - T2g;
|
||
|
T2i = FNMS(KP618033988, T2h, T2e);
|
||
|
T2k = FMA(KP618033988, T2e, T2h);
|
||
|
ro[0] = T5 + Ts;
|
||
|
T2j = FMA(KP559016994, T2a, T29);
|
||
|
ro[WS(os, 9)] = FNMS(KP951056516, T2k, T2j);
|
||
|
ro[WS(os, 6)] = FMA(KP951056516, T2k, T2j);
|
||
|
T2b = FNMS(KP559016994, T2a, T29);
|
||
|
ro[WS(os, 12)] = FNMS(KP951056516, T2i, T2b);
|
||
|
ro[WS(os, 3)] = FMA(KP951056516, T2i, T2b);
|
||
|
}
|
||
|
{
|
||
|
E T2q, T2o, T2p, T2u, T2w, T2s, T2t, T2v, T2r;
|
||
|
T2q = T2m - T2n;
|
||
|
T2o = T2m + T2n;
|
||
|
T2p = FNMS(KP250000000, T2o, T2l);
|
||
|
T2s = Tl - Tq;
|
||
|
T2t = Ta - Tf;
|
||
|
T2u = FNMS(KP618033988, T2t, T2s);
|
||
|
T2w = FMA(KP618033988, T2s, T2t);
|
||
|
io[0] = T2l + T2o;
|
||
|
T2v = FMA(KP559016994, T2q, T2p);
|
||
|
io[WS(os, 6)] = FNMS(KP951056516, T2w, T2v);
|
||
|
io[WS(os, 9)] = FMA(KP951056516, T2w, T2v);
|
||
|
T2r = FNMS(KP559016994, T2q, T2p);
|
||
|
io[WS(os, 3)] = FNMS(KP951056516, T2u, T2r);
|
||
|
io[WS(os, 12)] = FMA(KP951056516, T2u, T2r);
|
||
|
}
|
||
|
{
|
||
|
E T1M, TU, T1L, T1U, T1W, T1Q, T1T, T1V, T1N;
|
||
|
T1M = TI - TT;
|
||
|
TU = TI + TT;
|
||
|
T1L = FNMS(KP250000000, TU, Tx);
|
||
|
T1Q = T1O - T1P;
|
||
|
T1T = T1R - T1S;
|
||
|
T1U = FNMS(KP618033988, T1T, T1Q);
|
||
|
T1W = FMA(KP618033988, T1Q, T1T);
|
||
|
ro[WS(os, 5)] = Tx + TU;
|
||
|
T1V = FMA(KP559016994, T1M, T1L);
|
||
|
ro[WS(os, 14)] = FNMS(KP951056516, T1W, T1V);
|
||
|
ro[WS(os, 11)] = FMA(KP951056516, T1W, T1V);
|
||
|
T1N = FNMS(KP559016994, T1M, T1L);
|
||
|
ro[WS(os, 2)] = FNMS(KP951056516, T1U, T1N);
|
||
|
ro[WS(os, 8)] = FMA(KP951056516, T1U, T1N);
|
||
|
}
|
||
|
{
|
||
|
E T22, T20, T21, T26, T28, T24, T25, T27, T23;
|
||
|
T22 = T1Y - T1Z;
|
||
|
T20 = T1Y + T1Z;
|
||
|
T21 = FNMS(KP250000000, T20, T1X);
|
||
|
T24 = TN - TS;
|
||
|
T25 = TC - TH;
|
||
|
T26 = FNMS(KP618033988, T25, T24);
|
||
|
T28 = FMA(KP618033988, T24, T25);
|
||
|
io[WS(os, 5)] = T1X + T20;
|
||
|
T27 = FMA(KP559016994, T22, T21);
|
||
|
io[WS(os, 11)] = FNMS(KP951056516, T28, T27);
|
||
|
io[WS(os, 14)] = FMA(KP951056516, T28, T27);
|
||
|
T23 = FNMS(KP559016994, T22, T21);
|
||
|
io[WS(os, 2)] = FMA(KP951056516, T26, T23);
|
||
|
io[WS(os, 8)] = FNMS(KP951056516, T26, T23);
|
||
|
}
|
||
|
{
|
||
|
E T1E, T1C, T1D, T1I, T1K, T1G, T1H, T1J, T1F;
|
||
|
T1E = T1A - T1B;
|
||
|
T1C = T1A + T1B;
|
||
|
T1D = FNMS(KP250000000, T1C, T1z);
|
||
|
T1G = TW - TX;
|
||
|
T1H = TZ - T10;
|
||
|
T1I = FMA(KP618033988, T1H, T1G);
|
||
|
T1K = FNMS(KP618033988, T1G, T1H);
|
||
|
io[WS(os, 10)] = T1z + T1C;
|
||
|
T1J = FNMS(KP559016994, T1E, T1D);
|
||
|
io[WS(os, 7)] = FMA(KP951056516, T1K, T1J);
|
||
|
io[WS(os, 13)] = FNMS(KP951056516, T1K, T1J);
|
||
|
T1F = FMA(KP559016994, T1E, T1D);
|
||
|
io[WS(os, 1)] = FNMS(KP951056516, T1I, T1F);
|
||
|
io[WS(os, 4)] = FMA(KP951056516, T1I, T1F);
|
||
|
}
|
||
|
{
|
||
|
E T14, T12, T13, T1s, T1u, T1g, T1r, T1t, T15;
|
||
|
T14 = TY - T11;
|
||
|
T12 = TY + T11;
|
||
|
T13 = FNMS(KP250000000, T12, TV);
|
||
|
T1g = T1a - T1f;
|
||
|
T1r = T1l - T1q;
|
||
|
T1s = FMA(KP618033988, T1r, T1g);
|
||
|
T1u = FNMS(KP618033988, T1g, T1r);
|
||
|
ro[WS(os, 10)] = TV + T12;
|
||
|
T1t = FNMS(KP559016994, T14, T13);
|
||
|
ro[WS(os, 7)] = FNMS(KP951056516, T1u, T1t);
|
||
|
ro[WS(os, 13)] = FMA(KP951056516, T1u, T1t);
|
||
|
T15 = FMA(KP559016994, T14, T13);
|
||
|
ro[WS(os, 4)] = FNMS(KP951056516, T1s, T15);
|
||
|
ro[WS(os, 1)] = FMA(KP951056516, T1s, T15);
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
static const kdft_desc desc = { 15, "n1_15", { 72, 0, 84, 0 }, &GENUS, 0, 0, 0, 0 };
|
||
|
|
||
|
void X(codelet_n1_15) (planner *p) { X(kdft_register) (p, n1_15, &desc);
|
||
|
}
|
||
|
|
||
|
#else
|
||
|
|
||
|
/* Generated by: ../../../genfft/gen_notw.native -compact -variables 4 -pipeline-latency 4 -n 15 -name n1_15 -include dft/scalar/n.h */
|
||
|
|
||
|
/*
|
||
|
* This function contains 156 FP additions, 56 FP multiplications,
|
||
|
* (or, 128 additions, 28 multiplications, 28 fused multiply/add),
|
||
|
* 69 stack variables, 6 constants, and 60 memory accesses
|
||
|
*/
|
||
|
#include "dft/scalar/n.h"
|
||
|
|
||
|
static void n1_15(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs)
|
||
|
{
|
||
|
DK(KP587785252, +0.587785252292473129168705954639072768597652438);
|
||
|
DK(KP951056516, +0.951056516295153572116439333379382143405698634);
|
||
|
DK(KP250000000, +0.250000000000000000000000000000000000000000000);
|
||
|
DK(KP559016994, +0.559016994374947424102293417182819058860154590);
|
||
|
DK(KP500000000, +0.500000000000000000000000000000000000000000000);
|
||
|
DK(KP866025403, +0.866025403784438646763723170752936183471402627);
|
||
|
{
|
||
|
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(60, is), MAKE_VOLATILE_STRIDE(60, os)) {
|
||
|
E T5, T2l, Tx, TV, T1C, T20, Tl, Tq, Tr, TN, TS, TT, T2c, T2d, T2n;
|
||
|
E T1O, T1P, T22, T1l, T1q, T1w, TZ, T10, T11, Ta, Tf, Tg, TC, TH, TI;
|
||
|
E T2f, T2g, T2m, T1R, T1S, T21, T1a, T1f, T1v, TW, TX, TY;
|
||
|
{
|
||
|
E T1, T1z, T4, T1y, Tw, T1A, Tt, T1B;
|
||
|
T1 = ri[0];
|
||
|
T1z = ii[0];
|
||
|
{
|
||
|
E T2, T3, Tu, Tv;
|
||
|
T2 = ri[WS(is, 5)];
|
||
|
T3 = ri[WS(is, 10)];
|
||
|
T4 = T2 + T3;
|
||
|
T1y = KP866025403 * (T3 - T2);
|
||
|
Tu = ii[WS(is, 5)];
|
||
|
Tv = ii[WS(is, 10)];
|
||
|
Tw = KP866025403 * (Tu - Tv);
|
||
|
T1A = Tu + Tv;
|
||
|
}
|
||
|
T5 = T1 + T4;
|
||
|
T2l = T1z + T1A;
|
||
|
Tt = FNMS(KP500000000, T4, T1);
|
||
|
Tx = Tt - Tw;
|
||
|
TV = Tt + Tw;
|
||
|
T1B = FNMS(KP500000000, T1A, T1z);
|
||
|
T1C = T1y + T1B;
|
||
|
T20 = T1B - T1y;
|
||
|
}
|
||
|
{
|
||
|
E Th, Tk, TJ, T1h, T1i, T1j, TM, T1k, Tm, Tp, TO, T1m, T1n, T1o, TR;
|
||
|
E T1p;
|
||
|
{
|
||
|
E Ti, Tj, TK, TL;
|
||
|
Th = ri[WS(is, 6)];
|
||
|
Ti = ri[WS(is, 11)];
|
||
|
Tj = ri[WS(is, 1)];
|
||
|
Tk = Ti + Tj;
|
||
|
TJ = FNMS(KP500000000, Tk, Th);
|
||
|
T1h = KP866025403 * (Tj - Ti);
|
||
|
T1i = ii[WS(is, 6)];
|
||
|
TK = ii[WS(is, 11)];
|
||
|
TL = ii[WS(is, 1)];
|
||
|
T1j = TK + TL;
|
||
|
TM = KP866025403 * (TK - TL);
|
||
|
T1k = FNMS(KP500000000, T1j, T1i);
|
||
|
}
|
||
|
{
|
||
|
E Tn, To, TP, TQ;
|
||
|
Tm = ri[WS(is, 9)];
|
||
|
Tn = ri[WS(is, 14)];
|
||
|
To = ri[WS(is, 4)];
|
||
|
Tp = Tn + To;
|
||
|
TO = FNMS(KP500000000, Tp, Tm);
|
||
|
T1m = KP866025403 * (To - Tn);
|
||
|
T1n = ii[WS(is, 9)];
|
||
|
TP = ii[WS(is, 14)];
|
||
|
TQ = ii[WS(is, 4)];
|
||
|
T1o = TP + TQ;
|
||
|
TR = KP866025403 * (TP - TQ);
|
||
|
T1p = FNMS(KP500000000, T1o, T1n);
|
||
|
}
|
||
|
Tl = Th + Tk;
|
||
|
Tq = Tm + Tp;
|
||
|
Tr = Tl + Tq;
|
||
|
TN = TJ - TM;
|
||
|
TS = TO - TR;
|
||
|
TT = TN + TS;
|
||
|
T2c = T1i + T1j;
|
||
|
T2d = T1n + T1o;
|
||
|
T2n = T2c + T2d;
|
||
|
T1O = T1k - T1h;
|
||
|
T1P = T1p - T1m;
|
||
|
T22 = T1O + T1P;
|
||
|
T1l = T1h + T1k;
|
||
|
T1q = T1m + T1p;
|
||
|
T1w = T1l + T1q;
|
||
|
TZ = TJ + TM;
|
||
|
T10 = TO + TR;
|
||
|
T11 = TZ + T10;
|
||
|
}
|
||
|
{
|
||
|
E T6, T9, Ty, T16, T17, T18, TB, T19, Tb, Te, TD, T1b, T1c, T1d, TG;
|
||
|
E T1e;
|
||
|
{
|
||
|
E T7, T8, Tz, TA;
|
||
|
T6 = ri[WS(is, 3)];
|
||
|
T7 = ri[WS(is, 8)];
|
||
|
T8 = ri[WS(is, 13)];
|
||
|
T9 = T7 + T8;
|
||
|
Ty = FNMS(KP500000000, T9, T6);
|
||
|
T16 = KP866025403 * (T8 - T7);
|
||
|
T17 = ii[WS(is, 3)];
|
||
|
Tz = ii[WS(is, 8)];
|
||
|
TA = ii[WS(is, 13)];
|
||
|
T18 = Tz + TA;
|
||
|
TB = KP866025403 * (Tz - TA);
|
||
|
T19 = FNMS(KP500000000, T18, T17);
|
||
|
}
|
||
|
{
|
||
|
E Tc, Td, TE, TF;
|
||
|
Tb = ri[WS(is, 12)];
|
||
|
Tc = ri[WS(is, 2)];
|
||
|
Td = ri[WS(is, 7)];
|
||
|
Te = Tc + Td;
|
||
|
TD = FNMS(KP500000000, Te, Tb);
|
||
|
T1b = KP866025403 * (Td - Tc);
|
||
|
T1c = ii[WS(is, 12)];
|
||
|
TE = ii[WS(is, 2)];
|
||
|
TF = ii[WS(is, 7)];
|
||
|
T1d = TE + TF;
|
||
|
TG = KP866025403 * (TE - TF);
|
||
|
T1e = FNMS(KP500000000, T1d, T1c);
|
||
|
}
|
||
|
Ta = T6 + T9;
|
||
|
Tf = Tb + Te;
|
||
|
Tg = Ta + Tf;
|
||
|
TC = Ty - TB;
|
||
|
TH = TD - TG;
|
||
|
TI = TC + TH;
|
||
|
T2f = T17 + T18;
|
||
|
T2g = T1c + T1d;
|
||
|
T2m = T2f + T2g;
|
||
|
T1R = T19 - T16;
|
||
|
T1S = T1e - T1b;
|
||
|
T21 = T1R + T1S;
|
||
|
T1a = T16 + T19;
|
||
|
T1f = T1b + T1e;
|
||
|
T1v = T1a + T1f;
|
||
|
TW = Ty + TB;
|
||
|
TX = TD + TG;
|
||
|
TY = TW + TX;
|
||
|
}
|
||
|
{
|
||
|
E T2a, Ts, T29, T2i, T2k, T2e, T2h, T2j, T2b;
|
||
|
T2a = KP559016994 * (Tg - Tr);
|
||
|
Ts = Tg + Tr;
|
||
|
T29 = FNMS(KP250000000, Ts, T5);
|
||
|
T2e = T2c - T2d;
|
||
|
T2h = T2f - T2g;
|
||
|
T2i = FNMS(KP587785252, T2h, KP951056516 * T2e);
|
||
|
T2k = FMA(KP951056516, T2h, KP587785252 * T2e);
|
||
|
ro[0] = T5 + Ts;
|
||
|
T2j = T2a + T29;
|
||
|
ro[WS(os, 9)] = T2j - T2k;
|
||
|
ro[WS(os, 6)] = T2j + T2k;
|
||
|
T2b = T29 - T2a;
|
||
|
ro[WS(os, 12)] = T2b - T2i;
|
||
|
ro[WS(os, 3)] = T2b + T2i;
|
||
|
}
|
||
|
{
|
||
|
E T2q, T2o, T2p, T2u, T2w, T2s, T2t, T2v, T2r;
|
||
|
T2q = KP559016994 * (T2m - T2n);
|
||
|
T2o = T2m + T2n;
|
||
|
T2p = FNMS(KP250000000, T2o, T2l);
|
||
|
T2s = Tl - Tq;
|
||
|
T2t = Ta - Tf;
|
||
|
T2u = FNMS(KP587785252, T2t, KP951056516 * T2s);
|
||
|
T2w = FMA(KP951056516, T2t, KP587785252 * T2s);
|
||
|
io[0] = T2l + T2o;
|
||
|
T2v = T2q + T2p;
|
||
|
io[WS(os, 6)] = T2v - T2w;
|
||
|
io[WS(os, 9)] = T2w + T2v;
|
||
|
T2r = T2p - T2q;
|
||
|
io[WS(os, 3)] = T2r - T2u;
|
||
|
io[WS(os, 12)] = T2u + T2r;
|
||
|
}
|
||
|
{
|
||
|
E T1M, TU, T1L, T1U, T1W, T1Q, T1T, T1V, T1N;
|
||
|
T1M = KP559016994 * (TI - TT);
|
||
|
TU = TI + TT;
|
||
|
T1L = FNMS(KP250000000, TU, Tx);
|
||
|
T1Q = T1O - T1P;
|
||
|
T1T = T1R - T1S;
|
||
|
T1U = FNMS(KP587785252, T1T, KP951056516 * T1Q);
|
||
|
T1W = FMA(KP951056516, T1T, KP587785252 * T1Q);
|
||
|
ro[WS(os, 5)] = Tx + TU;
|
||
|
T1V = T1M + T1L;
|
||
|
ro[WS(os, 14)] = T1V - T1W;
|
||
|
ro[WS(os, 11)] = T1V + T1W;
|
||
|
T1N = T1L - T1M;
|
||
|
ro[WS(os, 2)] = T1N - T1U;
|
||
|
ro[WS(os, 8)] = T1N + T1U;
|
||
|
}
|
||
|
{
|
||
|
E T25, T23, T24, T1Z, T28, T1X, T1Y, T27, T26;
|
||
|
T25 = KP559016994 * (T21 - T22);
|
||
|
T23 = T21 + T22;
|
||
|
T24 = FNMS(KP250000000, T23, T20);
|
||
|
T1X = TN - TS;
|
||
|
T1Y = TC - TH;
|
||
|
T1Z = FNMS(KP587785252, T1Y, KP951056516 * T1X);
|
||
|
T28 = FMA(KP951056516, T1Y, KP587785252 * T1X);
|
||
|
io[WS(os, 5)] = T20 + T23;
|
||
|
T27 = T25 + T24;
|
||
|
io[WS(os, 11)] = T27 - T28;
|
||
|
io[WS(os, 14)] = T28 + T27;
|
||
|
T26 = T24 - T25;
|
||
|
io[WS(os, 2)] = T1Z + T26;
|
||
|
io[WS(os, 8)] = T26 - T1Z;
|
||
|
}
|
||
|
{
|
||
|
E T1x, T1D, T1E, T1I, T1J, T1G, T1H, T1K, T1F;
|
||
|
T1x = KP559016994 * (T1v - T1w);
|
||
|
T1D = T1v + T1w;
|
||
|
T1E = FNMS(KP250000000, T1D, T1C);
|
||
|
T1G = TW - TX;
|
||
|
T1H = TZ - T10;
|
||
|
T1I = FMA(KP951056516, T1G, KP587785252 * T1H);
|
||
|
T1J = FNMS(KP587785252, T1G, KP951056516 * T1H);
|
||
|
io[WS(os, 10)] = T1C + T1D;
|
||
|
T1K = T1E - T1x;
|
||
|
io[WS(os, 7)] = T1J + T1K;
|
||
|
io[WS(os, 13)] = T1K - T1J;
|
||
|
T1F = T1x + T1E;
|
||
|
io[WS(os, 1)] = T1F - T1I;
|
||
|
io[WS(os, 4)] = T1I + T1F;
|
||
|
}
|
||
|
{
|
||
|
E T13, T12, T14, T1s, T1u, T1g, T1r, T1t, T15;
|
||
|
T13 = KP559016994 * (TY - T11);
|
||
|
T12 = TY + T11;
|
||
|
T14 = FNMS(KP250000000, T12, TV);
|
||
|
T1g = T1a - T1f;
|
||
|
T1r = T1l - T1q;
|
||
|
T1s = FMA(KP951056516, T1g, KP587785252 * T1r);
|
||
|
T1u = FNMS(KP587785252, T1g, KP951056516 * T1r);
|
||
|
ro[WS(os, 10)] = TV + T12;
|
||
|
T1t = T14 - T13;
|
||
|
ro[WS(os, 7)] = T1t - T1u;
|
||
|
ro[WS(os, 13)] = T1t + T1u;
|
||
|
T15 = T13 + T14;
|
||
|
ro[WS(os, 4)] = T15 - T1s;
|
||
|
ro[WS(os, 1)] = T15 + T1s;
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
static const kdft_desc desc = { 15, "n1_15", { 128, 28, 28, 0 }, &GENUS, 0, 0, 0, 0 };
|
||
|
|
||
|
void X(codelet_n1_15) (planner *p) { X(kdft_register) (p, n1_15, &desc);
|
||
|
}
|
||
|
|
||
|
#endif
|