Logo Search packages:      
Sourcecode: blender version File versions

n1_12.c

/*
 * 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 Sat Jul  1 13:54:27 EDT 2006 */

#include "codelet-dft.h"

#ifdef HAVE_FMA

/* Generated by: ../../../genfft/gen_notw -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 12 -name n1_12 -include n.h */

/*
 * This function contains 96 FP additions, 24 FP multiplications,
 * (or, 72 additions, 0 multiplications, 24 fused multiply/add),
 * 63 stack variables, and 48 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_notw.ml,v 1.30 2006-02-12 23:34:12 athena Exp $
 */

#include "n.h"

static void n1_12(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs)
{
     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(is), MAKE_VOLATILE_STRIDE(os)) {
        E TT, TW, TF, T1q, TY, TQ, TX, T1n;
        {
             E TA, TS, TR, T5, Ts, Tz, TD, TV, TU, Ta, Tx, TC, T1d, Th, TJ;
             E TG, Tg, T1u, T1c, T1f, TM, TN, Tk, T1i;
             {
                E T6, Tt, Tu, Tv, T9;
                {
                   E T1, To, Tp, Tq, T4, T2, T3, T7, T8, Tr;
                   T1 = ri[0];
                   T2 = ri[WS(is, 4)];
                   T3 = ri[WS(is, 8)];
                   To = ii[0];
                   Tp = ii[WS(is, 4)];
                   Tq = ii[WS(is, 8)];
                   T4 = T2 + T3;
                   TA = T3 - T2;
                   T6 = ri[WS(is, 6)];
                   TS = Tp - Tq;
                   Tr = Tp + Tq;
                   TR = FNMS(KP500000000, T4, T1);
                   T5 = T1 + T4;
                   T7 = ri[WS(is, 10)];
                   Ts = To + Tr;
                   Tz = FNMS(KP500000000, Tr, To);
                   T8 = ri[WS(is, 2)];
                   Tt = ii[WS(is, 6)];
                   Tu = ii[WS(is, 10)];
                   Tv = ii[WS(is, 2)];
                   T9 = T7 + T8;
                   TD = T8 - T7;
                }
                {
                   E Tc, T1a, TH, TI, Tf, Td, Te, Tw, Ti, Tj, T1b;
                   Tc = ri[WS(is, 3)];
                   TV = Tu - Tv;
                   Tw = Tu + Tv;
                   TU = FNMS(KP500000000, T9, T6);
                   Ta = T6 + T9;
                   Td = ri[WS(is, 7)];
                   Tx = Tt + Tw;
                   TC = FNMS(KP500000000, Tw, Tt);
                   Te = ri[WS(is, 11)];
                   T1a = ii[WS(is, 3)];
                   TH = ii[WS(is, 7)];
                   TI = ii[WS(is, 11)];
                   Tf = Td + Te;
                   T1d = Te - Td;
                   Th = ri[WS(is, 9)];
                   TJ = TH - TI;
                   T1b = TH + TI;
                   TG = FNMS(KP500000000, Tf, Tc);
                   Tg = Tc + Tf;
                   Ti = ri[WS(is, 1)];
                   T1u = T1a + T1b;
                   T1c = FNMS(KP500000000, T1b, T1a);
                   Tj = ri[WS(is, 5)];
                   T1f = ii[WS(is, 9)];
                   TM = ii[WS(is, 1)];
                   TN = ii[WS(is, 5)];
                   Tk = Ti + Tj;
                   T1i = Tj - Ti;
                }
             }
             {
                E T1t, TO, TL, T1h, T1w, Tb, T1g, Tl;
                T1t = T5 - Ta;
                Tb = T5 + Ta;
                TO = TM - TN;
                T1g = TM + TN;
                TL = FNMS(KP500000000, Tk, Th);
                Tl = Th + Tk;
                {
                   E T1x, Ty, T1v, Tn, Tm, T1y;
                   T1x = Ts + Tx;
                   Ty = Ts - Tx;
                   T1v = T1f + T1g;
                   T1h = FNMS(KP500000000, T1g, T1f);
                   Tn = Tg - Tl;
                   Tm = Tg + Tl;
                   T1y = T1u + T1v;
                   T1w = T1u - T1v;
                   ro[0] = Tb + Tm;
                   ro[WS(os, 6)] = Tb - Tm;
                   io[WS(os, 3)] = Tn + Ty;
                   io[0] = T1x + T1y;
                   io[WS(os, 6)] = T1x - T1y;
                   io[WS(os, 9)] = Ty - Tn;
                }
                {
                   E TB, TE, T1o, T11, T1p, TK, TP, T15, T1k, T18, T14, T16, T1l, T1m;
                   {
                        E T1e, T1j, TZ, T10, T12, T13;
                        TB = FNMS(KP866025403, TA, Tz);
                        TZ = FMA(KP866025403, TA, Tz);
                        T10 = FMA(KP866025403, TD, TC);
                        TE = FNMS(KP866025403, TD, TC);
                        T1o = FNMS(KP866025403, T1d, T1c);
                        T1e = FMA(KP866025403, T1d, T1c);
                        ro[WS(os, 9)] = T1t + T1w;
                        ro[WS(os, 3)] = T1t - T1w;
                        T1l = TZ + T10;
                        T11 = TZ - T10;
                        T1j = FMA(KP866025403, T1i, T1h);
                        T1p = FNMS(KP866025403, T1i, T1h);
                        TK = FNMS(KP866025403, TJ, TG);
                        T12 = FMA(KP866025403, TJ, TG);
                        T13 = FMA(KP866025403, TO, TL);
                        TP = FNMS(KP866025403, TO, TL);
                        TT = FNMS(KP866025403, TS, TR);
                        T15 = FMA(KP866025403, TS, TR);
                        T1m = T1e + T1j;
                        T1k = T1e - T1j;
                        T18 = T12 + T13;
                        T14 = T12 - T13;
                        T16 = FMA(KP866025403, TV, TU);
                        TW = FNMS(KP866025403, TV, TU);
                   }
                   io[WS(os, 10)] = T1l - T1m;
                   io[WS(os, 4)] = T1l + T1m;
                   io[WS(os, 7)] = T11 + T14;
                   io[WS(os, 1)] = T11 - T14;
                   {
                        E T17, T19, T1r, T1s;
                        T17 = T15 + T16;
                        T19 = T15 - T16;
                        ro[WS(os, 7)] = T19 - T1k;
                        ro[WS(os, 1)] = T19 + T1k;
                        ro[WS(os, 4)] = T17 + T18;
                        ro[WS(os, 10)] = T17 - T18;
                        T1r = TB + TE;
                        TF = TB - TE;
                        T1s = T1o + T1p;
                        T1q = T1o - T1p;
                        TY = TK + TP;
                        TQ = TK - TP;
                        io[WS(os, 2)] = T1r - T1s;
                        io[WS(os, 8)] = T1r + T1s;
                   }
                }
             }
        }
        io[WS(os, 11)] = TF + TQ;
        io[WS(os, 5)] = TF - TQ;
        TX = TT + TW;
        T1n = TT - TW;
        ro[WS(os, 11)] = T1n - T1q;
        ro[WS(os, 5)] = T1n + T1q;
        ro[WS(os, 8)] = TX + TY;
        ro[WS(os, 2)] = TX - TY;
     }
}

static const kdft_desc desc = { 12, "n1_12", {72, 0, 24, 0}, &GENUS, 0, 0, 0, 0 };
void X(codelet_n1_12) (planner *p) {
     X(kdft_register) (p, n1_12, &desc);
}

#else                   /* HAVE_FMA */

/* Generated by: ../../../genfft/gen_notw -compact -variables 4 -pipeline-latency 4 -n 12 -name n1_12 -include n.h */

/*
 * This function contains 96 FP additions, 16 FP multiplications,
 * (or, 88 additions, 8 multiplications, 8 fused multiply/add),
 * 43 stack variables, and 48 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_notw.ml,v 1.30 2006-02-12 23:34:12 athena Exp $
 */

#include "n.h"

static void n1_12(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs)
{
     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(is), MAKE_VOLATILE_STRIDE(os)) {
        E T5, TR, TA, Ts, TS, Tz, Ta, TU, TD, Tx, TV, TC, Tg, T1a, TG;
        E TJ, T1u, T1d, Tl, T1f, TL, TO, T1v, T1i;
        {
             E T1, T2, T3, T4;
             T1 = ri[0];
             T2 = ri[WS(is, 4)];
             T3 = ri[WS(is, 8)];
             T4 = T2 + T3;
             T5 = T1 + T4;
             TR = FNMS(KP500000000, T4, T1);
             TA = KP866025403 * (T3 - T2);
        }
        {
             E To, Tp, Tq, Tr;
             To = ii[0];
             Tp = ii[WS(is, 4)];
             Tq = ii[WS(is, 8)];
             Tr = Tp + Tq;
             Ts = To + Tr;
             TS = KP866025403 * (Tp - Tq);
             Tz = FNMS(KP500000000, Tr, To);
        }
        {
             E T6, T7, T8, T9;
             T6 = ri[WS(is, 6)];
             T7 = ri[WS(is, 10)];
             T8 = ri[WS(is, 2)];
             T9 = T7 + T8;
             Ta = T6 + T9;
             TU = FNMS(KP500000000, T9, T6);
             TD = KP866025403 * (T8 - T7);
        }
        {
             E Tt, Tu, Tv, Tw;
             Tt = ii[WS(is, 6)];
             Tu = ii[WS(is, 10)];
             Tv = ii[WS(is, 2)];
             Tw = Tu + Tv;
             Tx = Tt + Tw;
             TV = KP866025403 * (Tu - Tv);
             TC = FNMS(KP500000000, Tw, Tt);
        }
        {
             E Tc, Td, Te, Tf;
             Tc = ri[WS(is, 3)];
             Td = ri[WS(is, 7)];
             Te = ri[WS(is, 11)];
             Tf = Td + Te;
             Tg = Tc + Tf;
             T1a = KP866025403 * (Te - Td);
             TG = FNMS(KP500000000, Tf, Tc);
        }
        {
             E T1b, TH, TI, T1c;
             T1b = ii[WS(is, 3)];
             TH = ii[WS(is, 7)];
             TI = ii[WS(is, 11)];
             T1c = TH + TI;
             TJ = KP866025403 * (TH - TI);
             T1u = T1b + T1c;
             T1d = FNMS(KP500000000, T1c, T1b);
        }
        {
             E Th, Ti, Tj, Tk;
             Th = ri[WS(is, 9)];
             Ti = ri[WS(is, 1)];
             Tj = ri[WS(is, 5)];
             Tk = Ti + Tj;
             Tl = Th + Tk;
             T1f = KP866025403 * (Tj - Ti);
             TL = FNMS(KP500000000, Tk, Th);
        }
        {
             E T1g, TM, TN, T1h;
             T1g = ii[WS(is, 9)];
             TM = ii[WS(is, 1)];
             TN = ii[WS(is, 5)];
             T1h = TM + TN;
             TO = KP866025403 * (TM - TN);
             T1v = T1g + T1h;
             T1i = FNMS(KP500000000, T1h, T1g);
        }
        {
             E Tb, Tm, T1t, T1w;
             Tb = T5 + Ta;
             Tm = Tg + Tl;
             ro[WS(os, 6)] = Tb - Tm;
             ro[0] = Tb + Tm;
             {
                E T1x, T1y, Tn, Ty;
                T1x = Ts + Tx;
                T1y = T1u + T1v;
                io[WS(os, 6)] = T1x - T1y;
                io[0] = T1x + T1y;
                Tn = Tg - Tl;
                Ty = Ts - Tx;
                io[WS(os, 3)] = Tn + Ty;
                io[WS(os, 9)] = Ty - Tn;
             }
             T1t = T5 - Ta;
             T1w = T1u - T1v;
             ro[WS(os, 3)] = T1t - T1w;
             ro[WS(os, 9)] = T1t + T1w;
             {
                E T11, T1l, T1k, T1m, T14, T18, T17, T19;
                {
                   E TZ, T10, T1e, T1j;
                   TZ = TA + Tz;
                   T10 = TD + TC;
                   T11 = TZ - T10;
                   T1l = TZ + T10;
                   T1e = T1a + T1d;
                   T1j = T1f + T1i;
                   T1k = T1e - T1j;
                   T1m = T1e + T1j;
                }
                {
                   E T12, T13, T15, T16;
                   T12 = TG + TJ;
                   T13 = TL + TO;
                   T14 = T12 - T13;
                   T18 = T12 + T13;
                   T15 = TR + TS;
                   T16 = TU + TV;
                   T17 = T15 + T16;
                   T19 = T15 - T16;
                }
                io[WS(os, 1)] = T11 - T14;
                ro[WS(os, 1)] = T19 + T1k;
                io[WS(os, 7)] = T11 + T14;
                ro[WS(os, 7)] = T19 - T1k;
                ro[WS(os, 10)] = T17 - T18;
                io[WS(os, 10)] = T1l - T1m;
                ro[WS(os, 4)] = T17 + T18;
                io[WS(os, 4)] = T1l + T1m;
             }
             {
                E TF, T1r, T1q, T1s, TQ, TY, TX, T1n;
                {
                   E TB, TE, T1o, T1p;
                   TB = Tz - TA;
                   TE = TC - TD;
                   TF = TB - TE;
                   T1r = TB + TE;
                   T1o = T1d - T1a;
                   T1p = T1i - T1f;
                   T1q = T1o - T1p;
                   T1s = T1o + T1p;
                }
                {
                   E TK, TP, TT, TW;
                   TK = TG - TJ;
                   TP = TL - TO;
                   TQ = TK - TP;
                   TY = TK + TP;
                   TT = TR - TS;
                   TW = TU - TV;
                   TX = TT + TW;
                   T1n = TT - TW;
                }
                io[WS(os, 5)] = TF - TQ;
                ro[WS(os, 5)] = T1n + T1q;
                io[WS(os, 11)] = TF + TQ;
                ro[WS(os, 11)] = T1n - T1q;
                ro[WS(os, 2)] = TX - TY;
                io[WS(os, 2)] = T1r - T1s;
                ro[WS(os, 8)] = TX + TY;
                io[WS(os, 8)] = T1r + T1s;
             }
        }
     }
}

static const kdft_desc desc = { 12, "n1_12", {88, 8, 8, 0}, &GENUS, 0, 0, 0, 0 };
void X(codelet_n1_12) (planner *p) {
     X(kdft_register) (p, n1_12, &desc);
}

#endif                        /* HAVE_FMA */

Generated by  Doxygen 1.6.0   Back to index