Logo Search packages:      
Sourcecode: fftw3 version File versions  Download package

m1_16.c

/*
 * Copyright (c) 2003 Matteo Frigo
 * Copyright (c) 2003 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 Sat Jul  5 21:29:37 EDT 2003 */

#include "codelet-dft.h"

/* Generated by: /homee/stevenj/cvs/fftw3.0.1/genfft/gen_notw_noinline -compact -variables 4 -n 16 -name m1_16 -include n.h */

/*
 * This function contains 144 FP additions, 24 FP multiplications,
 * (or, 136 additions, 16 multiplications, 8 fused multiply/add),
 * 49 stack variables, and 64 memory accesses
 */
/*
 * Generator Id's : 
 * $Id: algsimp.ml,v 1.7 2003/03/15 20:29:42 stevenj Exp $
 * $Id: fft.ml,v 1.2 2003/03/15 20:29:42 stevenj Exp $
 * $Id: gen_notw_noinline.ml,v 1.1 2003/04/17 11:07:19 athena Exp $
 */

#include "n.h"

static void m1_16_0(const R *ri, const R *ii, R *ro, R *io, stride is, stride os)
{
     DK(KP382683432, +0.382683432365089771728459984030398866761344562);
     DK(KP923879532, +0.923879532511286756128183189396788286822416626);
     DK(KP707106781, +0.707106781186547524400844362104849039284835938);
     {
        E T7, T1R, T25, TC, TN, T1x, T1H, T1l, Tt, T22, T2h, T1b, T1g, T1E, T1Z;
        E T1D, Te, T1S, T26, TJ, TQ, T1m, T1n, TT, Tm, T1X, T2g, T10, T15, T1B;
        E T1U, T1A;
        {
             E T3, TL, Ty, T1k, T6, T1j, TB, TM;
             {
                E T1, T2, Tw, Tx;
                T1 = ri[0];
                T2 = ri[WS(is, 8)];
                T3 = T1 + T2;
                TL = T1 - T2;
                Tw = ii[0];
                Tx = ii[WS(is, 8)];
                Ty = Tw + Tx;
                T1k = Tw - Tx;
             }
             {
                E T4, T5, Tz, TA;
                T4 = ri[WS(is, 4)];
                T5 = ri[WS(is, 12)];
                T6 = T4 + T5;
                T1j = T4 - T5;
                Tz = ii[WS(is, 4)];
                TA = ii[WS(is, 12)];
                TB = Tz + TA;
                TM = Tz - TA;
             }
             T7 = T3 + T6;
             T1R = T3 - T6;
             T25 = Ty - TB;
             TC = Ty + TB;
             TN = TL - TM;
             T1x = TL + TM;
             T1H = T1k - T1j;
             T1l = T1j + T1k;
        }
        {
             E Tp, T17, T1f, T20, Ts, T1c, T1a, T21;
             {
                E Tn, To, T1d, T1e;
                Tn = ri[WS(is, 15)];
                To = ri[WS(is, 7)];
                Tp = Tn + To;
                T17 = Tn - To;
                T1d = ii[WS(is, 15)];
                T1e = ii[WS(is, 7)];
                T1f = T1d - T1e;
                T20 = T1d + T1e;
             }
             {
                E Tq, Tr, T18, T19;
                Tq = ri[WS(is, 3)];
                Tr = ri[WS(is, 11)];
                Ts = Tq + Tr;
                T1c = Tq - Tr;
                T18 = ii[WS(is, 3)];
                T19 = ii[WS(is, 11)];
                T1a = T18 - T19;
                T21 = T18 + T19;
             }
             Tt = Tp + Ts;
             T22 = T20 - T21;
             T2h = T20 + T21;
             T1b = T17 - T1a;
             T1g = T1c + T1f;
             T1E = T1f - T1c;
             T1Z = Tp - Ts;
             T1D = T17 + T1a;
        }
        {
             E Ta, TP, TF, TO, Td, TR, TI, TS;
             {
                E T8, T9, TD, TE;
                T8 = ri[WS(is, 2)];
                T9 = ri[WS(is, 10)];
                Ta = T8 + T9;
                TP = T8 - T9;
                TD = ii[WS(is, 2)];
                TE = ii[WS(is, 10)];
                TF = TD + TE;
                TO = TD - TE;
             }
             {
                E Tb, Tc, TG, TH;
                Tb = ri[WS(is, 14)];
                Tc = ri[WS(is, 6)];
                Td = Tb + Tc;
                TR = Tb - Tc;
                TG = ii[WS(is, 14)];
                TH = ii[WS(is, 6)];
                TI = TG + TH;
                TS = TG - TH;
             }
             Te = Ta + Td;
             T1S = TF - TI;
             T26 = Td - Ta;
             TJ = TF + TI;
             TQ = TO - TP;
             T1m = TR - TS;
             T1n = TP + TO;
             TT = TR + TS;
        }
        {
             E Ti, T11, TZ, T1V, Tl, TW, T14, T1W;
             {
                E Tg, Th, TX, TY;
                Tg = ri[WS(is, 1)];
                Th = ri[WS(is, 9)];
                Ti = Tg + Th;
                T11 = Tg - Th;
                TX = ii[WS(is, 1)];
                TY = ii[WS(is, 9)];
                TZ = TX - TY;
                T1V = TX + TY;
             }
             {
                E Tj, Tk, T12, T13;
                Tj = ri[WS(is, 5)];
                Tk = ri[WS(is, 13)];
                Tl = Tj + Tk;
                TW = Tj - Tk;
                T12 = ii[WS(is, 5)];
                T13 = ii[WS(is, 13)];
                T14 = T12 - T13;
                T1W = T12 + T13;
             }
             Tm = Ti + Tl;
             T1X = T1V - T1W;
             T2g = T1V + T1W;
             T10 = TW + TZ;
             T15 = T11 - T14;
             T1B = T11 + T14;
             T1U = Ti - Tl;
             T1A = TZ - TW;
        }
        {
             E Tf, Tu, T2j, T2k;
             Tf = T7 + Te;
             Tu = Tm + Tt;
             ro[WS(os, 8)] = Tf - Tu;
             ro[0] = Tf + Tu;
             T2j = TC + TJ;
             T2k = T2g + T2h;
             io[WS(os, 8)] = T2j - T2k;
             io[0] = T2j + T2k;
        }
        {
             E Tv, TK, T2f, T2i;
             Tv = Tt - Tm;
             TK = TC - TJ;
             io[WS(os, 4)] = Tv + TK;
             io[WS(os, 12)] = TK - Tv;
             T2f = T7 - Te;
             T2i = T2g - T2h;
             ro[WS(os, 12)] = T2f - T2i;
             ro[WS(os, 4)] = T2f + T2i;
        }
        {
             E T1T, T27, T24, T28, T1Y, T23;
             T1T = T1R + T1S;
             T27 = T25 - T26;
             T1Y = T1U + T1X;
             T23 = T1Z - T22;
             T24 = KP707106781 * (T1Y + T23);
             T28 = KP707106781 * (T23 - T1Y);
             ro[WS(os, 10)] = T1T - T24;
             io[WS(os, 6)] = T27 + T28;
             ro[WS(os, 2)] = T1T + T24;
             io[WS(os, 14)] = T27 - T28;
        }
        {
             E T29, T2d, T2c, T2e, T2a, T2b;
             T29 = T1R - T1S;
             T2d = T26 + T25;
             T2a = T1X - T1U;
             T2b = T1Z + T22;
             T2c = KP707106781 * (T2a - T2b);
             T2e = KP707106781 * (T2a + T2b);
             ro[WS(os, 14)] = T29 - T2c;
             io[WS(os, 2)] = T2d + T2e;
             ro[WS(os, 6)] = T29 + T2c;
             io[WS(os, 10)] = T2d - T2e;
        }
        {
             E TV, T1r, T1p, T1v, T1i, T1q, T1u, T1w, TU, T1o;
             TU = KP707106781 * (TQ - TT);
             TV = TN + TU;
             T1r = TN - TU;
             T1o = KP707106781 * (T1m - T1n);
             T1p = T1l - T1o;
             T1v = T1l + T1o;
             {
                E T16, T1h, T1s, T1t;
                T16 = FMA(KP923879532, T10, KP382683432 * T15);
                T1h = FNMS(KP923879532, T1g, KP382683432 * T1b);
                T1i = T16 + T1h;
                T1q = T1h - T16;
                T1s = FNMS(KP923879532, T15, KP382683432 * T10);
                T1t = FMA(KP382683432, T1g, KP923879532 * T1b);
                T1u = T1s - T1t;
                T1w = T1s + T1t;
             }
             ro[WS(os, 11)] = TV - T1i;
             io[WS(os, 11)] = T1v - T1w;
             ro[WS(os, 3)] = TV + T1i;
             io[WS(os, 3)] = T1v + T1w;
             io[WS(os, 15)] = T1p - T1q;
             ro[WS(os, 15)] = T1r - T1u;
             io[WS(os, 7)] = T1p + T1q;
             ro[WS(os, 7)] = T1r + T1u;
        }
        {
             E T1z, T1L, T1J, T1P, T1G, T1K, T1O, T1Q, T1y, T1I;
             T1y = KP707106781 * (T1n + T1m);
             T1z = T1x + T1y;
             T1L = T1x - T1y;
             T1I = KP707106781 * (TQ + TT);
             T1J = T1H - T1I;
             T1P = T1H + T1I;
             {
                E T1C, T1F, T1M, T1N;
                T1C = FMA(KP382683432, T1A, KP923879532 * T1B);
                T1F = FNMS(KP382683432, T1E, KP923879532 * T1D);
                T1G = T1C + T1F;
                T1K = T1F - T1C;
                T1M = FNMS(KP382683432, T1B, KP923879532 * T1A);
                T1N = FMA(KP923879532, T1E, KP382683432 * T1D);
                T1O = T1M - T1N;
                T1Q = T1M + T1N;
             }
             ro[WS(os, 9)] = T1z - T1G;
             io[WS(os, 9)] = T1P - T1Q;
             ro[WS(os, 1)] = T1z + T1G;
             io[WS(os, 1)] = T1P + T1Q;
             io[WS(os, 13)] = T1J - T1K;
             ro[WS(os, 13)] = T1L - T1O;
             io[WS(os, 5)] = T1J + T1K;
             ro[WS(os, 5)] = T1L + T1O;
        }
     }
}

static void m1_16(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, int v, int ivs, int ovs)
{
     int i;
     for (i = v; i > 0; i -= 1) {
        m1_16_0(ri, ii, ro, io, is, os);
        ri += ivs;
        ii += ivs;
        ro += ovs;
        io += ovs;
     }
}

static const kdft_desc desc = { 16, "m1_16", {136, 16, 8, 0}, &GENUS, 0, 0, 0, 0 };
void X(codelet_m1_16) (planner *p) {
     X(kdft_register) (p, m1_16, &desc);
}

Generated by  Doxygen 1.6.0   Back to index