furnace/extern/fftw/dft/scalar/codelets/t2_8.c

391 lines
9.9 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 10:44:32 EDT 2021 */
#include "dft/codelet-dft.h"
#if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA)
/* Generated by: ../../../genfft/gen_twiddle.native -fma -compact -variables 4 -pipeline-latency 4 -twiddle-log3 -precompute-twiddles -n 8 -name t2_8 -include dft/scalar/t.h */
/*
* This function contains 74 FP additions, 50 FP multiplications,
* (or, 44 additions, 20 multiplications, 30 fused multiply/add),
* 48 stack variables, 1 constants, and 32 memory accesses
*/
#include "dft/scalar/t.h"
static void t2_8(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms)
{
DK(KP707106781, +0.707106781186547524400844362104849039284835938);
{
INT m;
for (m = mb, W = W + (mb * 6); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 6, MAKE_VOLATILE_STRIDE(16, rs)) {
E T2, T3, Tl, Tn, T5, T6, Tf, T7, Ts, Tb, To, Ti, TC, TG;
{
E T4, Tm, Tr, Ta, TB, TF;
T2 = W[0];
T3 = W[2];
T4 = T2 * T3;
Tl = W[4];
Tm = T2 * Tl;
Tn = W[5];
Tr = T2 * Tn;
T5 = W[1];
T6 = W[3];
Ta = T2 * T6;
Tf = FMA(T5, T6, T4);
T7 = FNMS(T5, T6, T4);
Ts = FNMS(T5, Tl, Tr);
Tb = FMA(T5, T3, Ta);
To = FMA(T5, Tn, Tm);
TB = Tf * Tl;
TF = Tf * Tn;
Ti = FNMS(T5, T3, Ta);
TC = FMA(Ti, Tn, TB);
TG = FNMS(Ti, Tl, TF);
}
{
E T1, T1s, Td, T1r, Tu, TY, Tk, TW, TN, TR, T18, T1a, T1c, T1d, TA;
E TI, T11, T13, T15, T16;
T1 = ri[0];
T1s = ii[0];
{
E T8, T9, Tc, T1q;
T8 = ri[WS(rs, 4)];
T9 = T7 * T8;
Tc = ii[WS(rs, 4)];
T1q = T7 * Tc;
Td = FMA(Tb, Tc, T9);
T1r = FNMS(Tb, T8, T1q);
}
{
E Tp, Tq, Tt, TX;
Tp = ri[WS(rs, 6)];
Tq = To * Tp;
Tt = ii[WS(rs, 6)];
TX = To * Tt;
Tu = FMA(Ts, Tt, Tq);
TY = FNMS(Ts, Tp, TX);
}
{
E Tg, Th, Tj, TV;
Tg = ri[WS(rs, 2)];
Th = Tf * Tg;
Tj = ii[WS(rs, 2)];
TV = Tf * Tj;
Tk = FMA(Ti, Tj, Th);
TW = FNMS(Ti, Tg, TV);
}
{
E TK, TL, TM, T19, TO, TP, TQ, T1b;
TK = ri[WS(rs, 7)];
TL = Tl * TK;
TM = ii[WS(rs, 7)];
T19 = Tl * TM;
TO = ri[WS(rs, 3)];
TP = T3 * TO;
TQ = ii[WS(rs, 3)];
T1b = T3 * TQ;
TN = FMA(Tn, TM, TL);
TR = FMA(T6, TQ, TP);
T18 = TN - TR;
T1a = FNMS(Tn, TK, T19);
T1c = FNMS(T6, TO, T1b);
T1d = T1a - T1c;
}
{
E Tx, Ty, Tz, T12, TD, TE, TH, T14;
Tx = ri[WS(rs, 1)];
Ty = T2 * Tx;
Tz = ii[WS(rs, 1)];
T12 = T2 * Tz;
TD = ri[WS(rs, 5)];
TE = TC * TD;
TH = ii[WS(rs, 5)];
T14 = TC * TH;
TA = FMA(T5, Tz, Ty);
TI = FMA(TG, TH, TE);
T11 = TA - TI;
T13 = FNMS(T5, Tx, T12);
T15 = FNMS(TG, TD, T14);
T16 = T13 - T15;
}
{
E T10, T1g, T1z, T1B, T1f, T1C, T1j, T1A;
{
E TU, TZ, T1x, T1y;
TU = T1 - Td;
TZ = TW - TY;
T10 = TU + TZ;
T1g = TU - TZ;
T1x = T1s - T1r;
T1y = Tk - Tu;
T1z = T1x - T1y;
T1B = T1y + T1x;
}
{
E T17, T1e, T1h, T1i;
T17 = T11 + T16;
T1e = T18 - T1d;
T1f = T17 + T1e;
T1C = T1e - T17;
T1h = T16 - T11;
T1i = T18 + T1d;
T1j = T1h - T1i;
T1A = T1h + T1i;
}
ri[WS(rs, 5)] = FNMS(KP707106781, T1f, T10);
ii[WS(rs, 5)] = FNMS(KP707106781, T1A, T1z);
ri[WS(rs, 1)] = FMA(KP707106781, T1f, T10);
ii[WS(rs, 1)] = FMA(KP707106781, T1A, T1z);
ri[WS(rs, 7)] = FNMS(KP707106781, T1j, T1g);
ii[WS(rs, 7)] = FNMS(KP707106781, T1C, T1B);
ri[WS(rs, 3)] = FMA(KP707106781, T1j, T1g);
ii[WS(rs, 3)] = FMA(KP707106781, T1C, T1B);
}
{
E Tw, T1k, T1u, T1w, TT, T1v, T1n, T1o;
{
E Te, Tv, T1p, T1t;
Te = T1 + Td;
Tv = Tk + Tu;
Tw = Te + Tv;
T1k = Te - Tv;
T1p = TW + TY;
T1t = T1r + T1s;
T1u = T1p + T1t;
T1w = T1t - T1p;
}
{
E TJ, TS, T1l, T1m;
TJ = TA + TI;
TS = TN + TR;
TT = TJ + TS;
T1v = TS - TJ;
T1l = T13 + T15;
T1m = T1a + T1c;
T1n = T1l - T1m;
T1o = T1l + T1m;
}
ri[WS(rs, 4)] = Tw - TT;
ii[WS(rs, 4)] = T1u - T1o;
ri[0] = Tw + TT;
ii[0] = T1o + T1u;
ri[WS(rs, 6)] = T1k - T1n;
ii[WS(rs, 6)] = T1w - T1v;
ri[WS(rs, 2)] = T1k + T1n;
ii[WS(rs, 2)] = T1v + T1w;
}
}
}
}
}
static const tw_instr twinstr[] = {
{ TW_CEXP, 0, 1 },
{ TW_CEXP, 0, 3 },
{ TW_CEXP, 0, 7 },
{ TW_NEXT, 1, 0 }
};
static const ct_desc desc = { 8, "t2_8", twinstr, &GENUS, { 44, 20, 30, 0 }, 0, 0, 0 };
void X(codelet_t2_8) (planner *p) {
X(kdft_dit_register) (p, t2_8, &desc);
}
#else
/* Generated by: ../../../genfft/gen_twiddle.native -compact -variables 4 -pipeline-latency 4 -twiddle-log3 -precompute-twiddles -n 8 -name t2_8 -include dft/scalar/t.h */
/*
* This function contains 74 FP additions, 44 FP multiplications,
* (or, 56 additions, 26 multiplications, 18 fused multiply/add),
* 42 stack variables, 1 constants, and 32 memory accesses
*/
#include "dft/scalar/t.h"
static void t2_8(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms)
{
DK(KP707106781, +0.707106781186547524400844362104849039284835938);
{
INT m;
for (m = mb, W = W + (mb * 6); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 6, MAKE_VOLATILE_STRIDE(16, rs)) {
E T2, T5, T3, T6, T8, Tc, Tg, Ti, Tl, Tm, Tn, Tz, Tp, Tx;
{
E T4, Tb, T7, Ta;
T2 = W[0];
T5 = W[1];
T3 = W[2];
T6 = W[3];
T4 = T2 * T3;
Tb = T5 * T3;
T7 = T5 * T6;
Ta = T2 * T6;
T8 = T4 - T7;
Tc = Ta + Tb;
Tg = T4 + T7;
Ti = Ta - Tb;
Tl = W[4];
Tm = W[5];
Tn = FMA(T2, Tl, T5 * Tm);
Tz = FNMS(Ti, Tl, Tg * Tm);
Tp = FNMS(T5, Tl, T2 * Tm);
Tx = FMA(Tg, Tl, Ti * Tm);
}
{
E Tf, T1i, TL, T1d, TJ, T17, TV, TY, Ts, T1j, TO, T1a, TC, T16, TQ;
E TT;
{
E T1, T1c, Te, T1b, T9, Td;
T1 = ri[0];
T1c = ii[0];
T9 = ri[WS(rs, 4)];
Td = ii[WS(rs, 4)];
Te = FMA(T8, T9, Tc * Td);
T1b = FNMS(Tc, T9, T8 * Td);
Tf = T1 + Te;
T1i = T1c - T1b;
TL = T1 - Te;
T1d = T1b + T1c;
}
{
E TF, TW, TI, TX;
{
E TD, TE, TG, TH;
TD = ri[WS(rs, 7)];
TE = ii[WS(rs, 7)];
TF = FMA(Tl, TD, Tm * TE);
TW = FNMS(Tm, TD, Tl * TE);
TG = ri[WS(rs, 3)];
TH = ii[WS(rs, 3)];
TI = FMA(T3, TG, T6 * TH);
TX = FNMS(T6, TG, T3 * TH);
}
TJ = TF + TI;
T17 = TW + TX;
TV = TF - TI;
TY = TW - TX;
}
{
E Tk, TM, Tr, TN;
{
E Th, Tj, To, Tq;
Th = ri[WS(rs, 2)];
Tj = ii[WS(rs, 2)];
Tk = FMA(Tg, Th, Ti * Tj);
TM = FNMS(Ti, Th, Tg * Tj);
To = ri[WS(rs, 6)];
Tq = ii[WS(rs, 6)];
Tr = FMA(Tn, To, Tp * Tq);
TN = FNMS(Tp, To, Tn * Tq);
}
Ts = Tk + Tr;
T1j = Tk - Tr;
TO = TM - TN;
T1a = TM + TN;
}
{
E Tw, TR, TB, TS;
{
E Tu, Tv, Ty, TA;
Tu = ri[WS(rs, 1)];
Tv = ii[WS(rs, 1)];
Tw = FMA(T2, Tu, T5 * Tv);
TR = FNMS(T5, Tu, T2 * Tv);
Ty = ri[WS(rs, 5)];
TA = ii[WS(rs, 5)];
TB = FMA(Tx, Ty, Tz * TA);
TS = FNMS(Tz, Ty, Tx * TA);
}
TC = Tw + TB;
T16 = TR + TS;
TQ = Tw - TB;
TT = TR - TS;
}
{
E Tt, TK, T1f, T1g;
Tt = Tf + Ts;
TK = TC + TJ;
ri[WS(rs, 4)] = Tt - TK;
ri[0] = Tt + TK;
{
E T19, T1e, T15, T18;
T19 = T16 + T17;
T1e = T1a + T1d;
ii[0] = T19 + T1e;
ii[WS(rs, 4)] = T1e - T19;
T15 = Tf - Ts;
T18 = T16 - T17;
ri[WS(rs, 6)] = T15 - T18;
ri[WS(rs, 2)] = T15 + T18;
}
T1f = TJ - TC;
T1g = T1d - T1a;
ii[WS(rs, 2)] = T1f + T1g;
ii[WS(rs, 6)] = T1g - T1f;
{
E T11, T1k, T14, T1h, T12, T13;
T11 = TL - TO;
T1k = T1i - T1j;
T12 = TT - TQ;
T13 = TV + TY;
T14 = KP707106781 * (T12 - T13);
T1h = KP707106781 * (T12 + T13);
ri[WS(rs, 7)] = T11 - T14;
ii[WS(rs, 5)] = T1k - T1h;
ri[WS(rs, 3)] = T11 + T14;
ii[WS(rs, 1)] = T1h + T1k;
}
{
E TP, T1m, T10, T1l, TU, TZ;
TP = TL + TO;
T1m = T1j + T1i;
TU = TQ + TT;
TZ = TV - TY;
T10 = KP707106781 * (TU + TZ);
T1l = KP707106781 * (TZ - TU);
ri[WS(rs, 5)] = TP - T10;
ii[WS(rs, 7)] = T1m - T1l;
ri[WS(rs, 1)] = TP + T10;
ii[WS(rs, 3)] = T1l + T1m;
}
}
}
}
}
}
static const tw_instr twinstr[] = {
{ TW_CEXP, 0, 1 },
{ TW_CEXP, 0, 3 },
{ TW_CEXP, 0, 7 },
{ TW_NEXT, 1, 0 }
};
static const ct_desc desc = { 8, "t2_8", twinstr, &GENUS, { 56, 26, 18, 0 }, 0, 0, 0 };
void X(codelet_t2_8) (planner *p) {
X(kdft_dit_register) (p, t2_8, &desc);
}
#endif