/*
* Copyright (c) 2003, 2006 Matteo Frigo
* Copyright (c) 2003, 2006 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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*
*/
/* This file was automatically generated --- DO NOT EDIT */
/* Generated on Sun Jul 2 15:50:20 EDT 2006 */
#include "codelet-rdft.h"
#ifdef HAVE_FMA
/* Generated by: ../../../genfft/gen_hc2hc -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 8 -dit -name hf_8 -include hf.h */
/*
* This function contains 66 FP additions, 36 FP multiplications,
* (or, 44 additions, 14 multiplications, 22 fused multiply/add),
* 61 stack variables, and 32 memory accesses
*/
/*
* Generator Id's :
* $Id: algsimp.ml,v 1.9 2006-02-12 23:34:12 athena Exp $
* $Id: fft.ml,v 1.4 2006-01-05 03:04:27 stevenj Exp $
* $Id: gen_hc2hc.ml,v 1.16 2006-02-12 23:34:12 athena Exp $
*/
#include "hf.h"
static const R *hf_8(R *rio, R *iio, const R *W, stride ios, INT m, INT dist)
{
DK(KP707106781, +0.707106781186547524400844362104849039284835938);
INT i;
for (i = m - 2; i > 0; i = i - 2, rio = rio + dist, iio = iio - dist, W = W + 14, MAKE_VOLATILE_STRIDE(ios)) {
E T1g, T1f, T1e, Tm, T1q, T1o, T1p, TN, T1h, T1i;
{
E T1, T1m, TS, Tk, TQ, Te, T1l, T7, To, Tr, T17, TM, T12, Tu, TW;
E Tp, Tx, Tt, Tq, Tw;
{
E Tg, Tj, Tf, Ti;
T1 = rio[0];
T1m = iio[-WS(ios, 7)];
Tg = rio[WS(ios, 6)];
Tj = iio[-WS(ios, 1)];
Tf = W[10];
Ti = W[11];
{
E T3, T6, T2, T5;
{
E Ta, Td, Tc, TP, Tb, TR, Th, T9;
Ta = rio[WS(ios, 2)];
Td = iio[-WS(ios, 5)];
TR = Tf * Tj;
Th = Tf * Tg;
T9 = W[2];
Tc = W[3];
TS = FNMS(Ti, Tg, TR);
Tk = FMA(Ti, Tj, Th);
TP = T9 * Td;
Tb = T9 * Ta;
T3 = rio[WS(ios, 4)];
T6 = iio[-WS(ios, 3)];
TQ = FNMS(Tc, Ta, TP);
Te = FMA(Tc, Td, Tb);
T2 = W[6];
T5 = W[7];
}
{
E TB, TE, TH, T13, TC, TK, TG, TD, TJ, T1k, T4, TA, Tn;
TB = rio[WS(ios, 7)];
TE = iio[0];
T1k = T2 * T6;
T4 = T2 * T3;
TA = W[12];
TH = rio[WS(ios, 3)];
T1l = FNMS(T5, T3, T1k);
T7 = FMA(T5, T6, T4);
T13 = TA * TE;
TC = TA * TB;
TK = iio[-WS(ios, 4)];
TG = W[4];
TD = W[13];
TJ = W[5];
{
E T14, TF, T16, TL, T15, TI;
To = rio[WS(ios, 1)];
T15 = TG * TK;
TI = TG * TH;
T14 = FNMS(TD, TB, T13);
TF = FMA(TD, TE, TC);
T16 = FNMS(TJ, TH, T15);
TL = FMA(TJ, TK, TI);
Tr = iio[-WS(ios, 6)];
Tn = W[0];
T17 = T14 - T16;
T1g = T14 + T16;
TM = TF + TL;
T12 = TF - TL;
}
Tu = rio[WS(ios, 5)];
TW = Tn * Tr;
Tp = Tn * To;
Tx = iio[-WS(ios, 2)];
Tt = W[8];
Tq = W[1];
Tw = W[9];
}
}
}
{
E T8, T1j, T1n, Tz, T1a, TU, Tl, T1b, T1c, T1v, T1t, T1w, T19, T1u, T1d;
{
E T1r, T10, TV, T1s, T11, T18;
{
E TO, TX, Ts, TZ, Ty, TT, TY, Tv;
T8 = T1 + T7;
TO = T1 - T7;
TY = Tt * Tx;
Tv = Tt * Tu;
TX = FNMS(Tq, To, TW);
Ts = FMA(Tq, Tr, Tp);
TZ = FNMS(Tw, Tu, TY);
Ty = FMA(Tw, Tx, Tv);
TT = TQ - TS;
T1j = TQ + TS;
T1n = T1l + T1m;
T1r = T1m - T1l;
T10 = TX - TZ;
T1f = TX + TZ;
Tz = Ts + Ty;
TV = Ts - Ty;
T1a = TO - TT;
TU = TO + TT;
T1s = Te - Tk;
Tl = Te + Tk;
}
T1b = T10 - TV;
T11 = TV + T10;
T18 = T12 - T17;
T1c = T12 + T17;
T1v = T1s + T1r;
T1t = T1r - T1s;
T1w = T18 - T11;
T19 = T11 + T18;
}
iio[-WS(ios, 3)] = FMA(KP707106781, T1w, T1v);
rio[WS(ios, 7)] = FMS(KP707106781, T1w, T1v);
rio[WS(ios, 1)] = FMA(KP707106781, T19, TU);
iio[-WS(ios, 5)] = FNMS(KP707106781, T19, TU);
T1u = T1b + T1c;
T1d = T1b - T1c;
iio[-WS(ios, 1)] = FMA(KP707106781, T1u, T1t);
rio[WS(ios, 5)] = FMS(KP707106781, T1u, T1t);
rio[WS(ios, 3)] = FMA(KP707106781, T1d, T1a);
iio[-WS(ios, 7)] = FNMS(KP707106781, T1d, T1a);
T1e = T8 - Tl;
Tm = T8 + Tl;
T1q = T1n - T1j;
T1o = T1j + T1n;
T1p = TM - Tz;
TN = Tz + TM;
}
}
iio[-WS(ios, 2)] = T1p + T1q;
rio[WS(ios, 6)] = T1p - T1q;
rio[0] = Tm + TN;
iio[-WS(ios, 4)] = Tm - TN;
T1h = T1f - T1g;
T1i = T1f + T1g;
iio[0] = T1i + T1o;
rio[WS(ios, 4)] = T1i - T1o;
rio[WS(ios, 2)] = T1e + T1h;
iio[-WS(ios, 6)] = T1e - T1h;
}
return W;
}
static const tw_instr twinstr[] = {
{TW_FULL, 0, 8},
{TW_NEXT, 1, 0}
};
static const hc2hc_desc desc = { 8, "hf_8", twinstr, &GENUS, {44, 14, 22, 0}, 0, 0, 0 };
void X(codelet_hf_8) (planner *p) {
X(khc2hc_register) (p, hf_8, &desc);
}
#else /* HAVE_FMA */
/* Generated by: ../../../genfft/gen_hc2hc -compact -variables 4 -pipeline-latency 4 -n 8 -dit -name hf_8 -include hf.h */
/*
* This function contains 66 FP additions, 32 FP multiplications,
* (or, 52 additions, 18 multiplications, 14 fused multiply/add),
* 28 stack variables, and 32 memory accesses
*/
/*
* Generator Id's :
* $Id: algsimp.ml,v 1.9 2006-02-12 23:34:12 athena Exp $
* $Id: fft.ml,v 1.4 2006-01-05 03:04:27 stevenj Exp $
* $Id: gen_hc2hc.ml,v 1.16 2006-02-12 23:34:12 athena Exp $
*/
#include "hf.h"
static const R *hf_8(R *rio, R *iio, const R *W, stride ios, INT m, INT dist)
{
DK(KP707106781, +0.707106781186547524400844362104849039284835938);
INT i;
for (i = m - 2; i > 0; i = i - 2, rio = rio + dist, iio = iio - dist, W = W + 14, MAKE_VOLATILE_STRIDE(ios)) {
E T7, T1e, TH, T19, TF, T13, TR, TU, Ti, T1f, TK, T16, Tu, T12, TM;
E TP;
{
E T1, T18, T6, T17;
T1 = rio[0];
T18 = iio[-WS(ios, 7)];
{
E T3, T5, T2, T4;
T3 = rio[WS(ios, 4)];
T5 = iio[-WS(ios, 3)];
T2 = W[6];
T4 = W[7];
T6 = FMA(T2, T3, T4 * T5);
T17 = FNMS(T4, T3, T2 * T5);
}
T7 = T1 + T6;
T1e = T18 - T17;
TH = T1 - T6;
T19 = T17 + T18;
}
{
E Tz, TS, TE, TT;
{
E Tw, Ty, Tv, Tx;
Tw = rio[WS(ios, 7)];
Ty = iio[0];
Tv = W[12];
Tx = W[13];
Tz = FMA(Tv, Tw, Tx * Ty);
TS = FNMS(Tx, Tw, Tv * Ty);
}
{
E TB, TD, TA, TC;
TB = rio[WS(ios, 3)];
TD = iio[-WS(ios, 4)];
TA = W[4];
TC = W[5];
TE = FMA(TA, TB, TC * TD);
TT = FNMS(TC, TB, TA * TD);
}
TF = Tz + TE;
T13 = TS + TT;
TR = Tz - TE;
TU = TS - TT;
}
{
E Tc, TI, Th, TJ;
{
E T9, Tb, T8, Ta;
T9 = rio[WS(ios, 2)];
Tb = iio[-WS(ios, 5)];
T8 = W[2];
Ta = W[3];
Tc = FMA(T8, T9, Ta * Tb);
TI = FNMS(Ta, T9, T8 * Tb);
}
{
E Te, Tg, Td, Tf;
Te = rio[WS(ios, 6)];
Tg = iio[-WS(ios, 1)];
Td = W[10];
Tf = W[11];
Th = FMA(Td, Te, Tf * Tg);
TJ = FNMS(Tf, Te, Td * Tg);
}
Ti = Tc + Th;
T1f = Tc - Th;
TK = TI - TJ;
T16 = TI + TJ;
}
{
E To, TN, Tt, TO;
{
E Tl, Tn, Tk, Tm;
Tl = rio[WS(ios, 1)];
Tn = iio[-WS(ios, 6)];
Tk = W[0];
Tm = W[1];
To = FMA(Tk, Tl, Tm * Tn);
TN = FNMS(Tm, Tl, Tk * Tn);
}
{
E Tq, Ts, Tp, Tr;
Tq = rio[WS(ios, 5)];
Ts = iio[-WS(ios, 2)];
Tp = W[8];
Tr = W[9];
Tt = FMA(Tp, Tq, Tr * Ts);
TO = FNMS(Tr, Tq, Tp * Ts);
}
Tu = To + Tt;
T12 = TN + TO;
TM = To - Tt;
TP = TN - TO;
}
{
E Tj, TG, T1b, T1c;
Tj = T7 + Ti;
TG = Tu + TF;
iio[-WS(ios, 4)] = Tj - TG;
rio[0] = Tj + TG;
{
E T15, T1a, T11, T14;
T15 = T12 + T13;
T1a = T16 + T19;
rio[WS(ios, 4)] = T15 - T1a;
iio[0] = T15 + T1a;
T11 = T7 - Ti;
T14 = T12 - T13;
iio[-WS(ios, 6)] = T11 - T14;
rio[WS(ios, 2)] = T11 + T14;
}
T1b = TF - Tu;
T1c = T19 - T16;
rio[WS(ios, 6)] = T1b - T1c;
iio[-WS(ios, 2)] = T1b + T1c;
{
E TX, T1g, T10, T1d, TY, TZ;
TX = TH - TK;
T1g = T1e - T1f;
TY = TP - TM;
TZ = TR + TU;
T10 = KP707106781 * (TY - TZ);
T1d = KP707106781 * (TY + TZ);
iio[-WS(ios, 7)] = TX - T10;
iio[-WS(ios, 1)] = T1d + T1g;
rio[WS(ios, 3)] = TX + T10;
rio[WS(ios, 5)] = T1d - T1g;
}
{
E TL, T1i, TW, T1h, TQ, TV;
TL = TH + TK;
T1i = T1f + T1e;
TQ = TM + TP;
TV = TR - TU;
TW = KP707106781 * (TQ + TV);
T1h = KP707106781 * (TV - TQ);
iio[-WS(ios, 5)] = TL - TW;
iio[-WS(ios, 3)] = T1h + T1i;
rio[WS(ios, 1)] = TL + TW;
rio[WS(ios, 7)] = T1h - T1i;
}
}
}
return W;
}
static const tw_instr twinstr[] = {
{TW_FULL, 0, 8},
{TW_NEXT, 1, 0}
};
static const hc2hc_desc desc = { 8, "hf_8", twinstr, &GENUS, {52, 18, 14, 0}, 0, 0, 0 };
void X(codelet_hf_8) (planner *p) {
X(khc2hc_register) (p, hf_8, &desc);
}
#endif /* HAVE_FMA */
syntax highlighted by Code2HTML, v. 0.9.1