t1sv_16.c

快速傅立叶变换库函数

     VTW(13),
     VTW(14),
     VTW(15),
     {TW_NEXT, (2 * VL), 0}
};

static const ct_desc desc = { 16, "t1sv_16", twinstr, &GENUS, {104, 30, 70, 0}, 0, 0, 0 };

void X(codelet_t1sv_16) (planner *p) {
     X(kdft_dit_register) (p, t1sv_16, &desc);
}
#else				/* HAVE_FMA */

/* Generated by: ../../../genfft/gen_twiddle -simd -compact -variables 4 -pipeline-latency 8 -n 16 -name t1sv_16 -include ts.h */

/*
 * This function contains 174 FP additions, 84 FP multiplications,
 * (or, 136 additions, 46 multiplications, 38 fused multiply/add),
 * 52 stack variables, and 64 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_twiddle.ml,v 1.24 2006-02-12 23:34:12 athena Exp $
 */

#include "ts.h"

static const R *t1sv_16(R *ri, R *ii, const R *W, stride ios, INT m, INT dist)
{
     DVK(KP382683432, +0.382683432365089771728459984030398866761344562);
     DVK(KP923879532, +0.923879532511286756128183189396788286822416626);
     DVK(KP707106781, +0.707106781186547524400844362104849039284835938);
     INT i;
     for (i = m; i > 0; i = i - (2 * VL), ri = ri + ((2 * VL) * dist), ii = ii + ((2 * VL) * dist), W = W + ((2 * VL) * 30), MAKE_VOLATILE_STRIDE(ios)) {
	  V T7, T37, T1t, T2U, Ti, T38, T1w, T2R, Tu, T2s, T1C, T2c, TF, T2t, T1H;
	  V T2d, T1f, T1q, T2B, T2C, T2D, T2E, T1Z, T2j, T24, T2k, TS, T13, T2w, T2x;
	  V T2y, T2z, T1O, T2g, T1T, T2h;
	  {
	       V T1, T2T, T6, T2S;
	       T1 = LD(&(ri[0]), dist, &(ri[0]));
	       T2T = LD(&(ii[0]), dist, &(ii[0]));
	       {
		    V T3, T5, T2, T4;
		    T3 = LD(&(ri[WS(ios, 8)]), dist, &(ri[0]));
		    T5 = LD(&(ii[WS(ios, 8)]), dist, &(ii[0]));
		    T2 = LDW(&(W[TWVL * 14]));
		    T4 = LDW(&(W[TWVL * 15]));
		    T6 = VFMA(T2, T3, VMUL(T4, T5));
		    T2S = VFNMS(T4, T3, VMUL(T2, T5));
	       }
	       T7 = VADD(T1, T6);
	       T37 = VSUB(T2T, T2S);
	       T1t = VSUB(T1, T6);
	       T2U = VADD(T2S, T2T);
	  }
	  {
	       V Tc, T1u, Th, T1v;
	       {
		    V T9, Tb, T8, Ta;
		    T9 = LD(&(ri[WS(ios, 4)]), dist, &(ri[0]));
		    Tb = LD(&(ii[WS(ios, 4)]), dist, &(ii[0]));
		    T8 = LDW(&(W[TWVL * 6]));
		    Ta = LDW(&(W[TWVL * 7]));
		    Tc = VFMA(T8, T9, VMUL(Ta, Tb));
		    T1u = VFNMS(Ta, T9, VMUL(T8, Tb));
	       }
	       {
		    V Te, Tg, Td, Tf;
		    Te = LD(&(ri[WS(ios, 12)]), dist, &(ri[0]));
		    Tg = LD(&(ii[WS(ios, 12)]), dist, &(ii[0]));
		    Td = LDW(&(W[TWVL * 22]));
		    Tf = LDW(&(W[TWVL * 23]));
		    Th = VFMA(Td, Te, VMUL(Tf, Tg));
		    T1v = VFNMS(Tf, Te, VMUL(Td, Tg));
	       }
	       Ti = VADD(Tc, Th);
	       T38 = VSUB(Tc, Th);
	       T1w = VSUB(T1u, T1v);
	       T2R = VADD(T1u, T1v);
	  }
	  {
	       V To, T1y, Tt, T1z, T1A, T1B;
	       {
		    V Tl, Tn, Tk, Tm;
		    Tl = LD(&(ri[WS(ios, 2)]), dist, &(ri[0]));
		    Tn = LD(&(ii[WS(ios, 2)]), dist, &(ii[0]));
		    Tk = LDW(&(W[TWVL * 2]));
		    Tm = LDW(&(W[TWVL * 3]));
		    To = VFMA(Tk, Tl, VMUL(Tm, Tn));
		    T1y = VFNMS(Tm, Tl, VMUL(Tk, Tn));
	       }
	       {
		    V Tq, Ts, Tp, Tr;
		    Tq = LD(&(ri[WS(ios, 10)]), dist, &(ri[0]));
		    Ts = LD(&(ii[WS(ios, 10)]), dist, &(ii[0]));
		    Tp = LDW(&(W[TWVL * 18]));
		    Tr = LDW(&(W[TWVL * 19]));
		    Tt = VFMA(Tp, Tq, VMUL(Tr, Ts));
		    T1z = VFNMS(Tr, Tq, VMUL(Tp, Ts));
	       }
	       Tu = VADD(To, Tt);
	       T2s = VADD(T1y, T1z);
	       T1A = VSUB(T1y, T1z);
	       T1B = VSUB(To, Tt);
	       T1C = VSUB(T1A, T1B);
	       T2c = VADD(T1B, T1A);
	  }
	  {
	       V Tz, T1E, TE, T1F, T1D, T1G;
	       {
		    V Tw, Ty, Tv, Tx;
		    Tw = LD(&(ri[WS(ios, 14)]), dist, &(ri[0]));
		    Ty = LD(&(ii[WS(ios, 14)]), dist, &(ii[0]));
		    Tv = LDW(&(W[TWVL * 26]));
		    Tx = LDW(&(W[TWVL * 27]));
		    Tz = VFMA(Tv, Tw, VMUL(Tx, Ty));
		    T1E = VFNMS(Tx, Tw, VMUL(Tv, Ty));
	       }
	       {
		    V TB, TD, TA, TC;
		    TB = LD(&(ri[WS(ios, 6)]), dist, &(ri[0]));
		    TD = LD(&(ii[WS(ios, 6)]), dist, &(ii[0]));
		    TA = LDW(&(W[TWVL * 10]));
		    TC = LDW(&(W[TWVL * 11]));
		    TE = VFMA(TA, TB, VMUL(TC, TD));
		    T1F = VFNMS(TC, TB, VMUL(TA, TD));
	       }
	       TF = VADD(Tz, TE);
	       T2t = VADD(T1E, T1F);
	       T1D = VSUB(Tz, TE);
	       T1G = VSUB(T1E, T1F);
	       T1H = VADD(T1D, T1G);
	       T2d = VSUB(T1D, T1G);
	  }
	  {
	       V T19, T20, T1p, T1X, T1e, T21, T1k, T1W;
	       {
		    V T16, T18, T15, T17;
		    T16 = LD(&(ri[WS(ios, 15)]), dist, &(ri[WS(ios, 1)]));
		    T18 = LD(&(ii[WS(ios, 15)]), dist, &(ii[WS(ios, 1)]));
		    T15 = LDW(&(W[TWVL * 28]));
		    T17 = LDW(&(W[TWVL * 29]));
		    T19 = VFMA(T15, T16, VMUL(T17, T18));
		    T20 = VFNMS(T17, T16, VMUL(T15, T18));
	       }
	       {
		    V T1m, T1o, T1l, T1n;
		    T1m = LD(&(ri[WS(ios, 11)]), dist, &(ri[WS(ios, 1)]));
		    T1o = LD(&(ii[WS(ios, 11)]), dist, &(ii[WS(ios, 1)]));
		    T1l = LDW(&(W[TWVL * 20]));
		    T1n = LDW(&(W[TWVL * 21]));
		    T1p = VFMA(T1l, T1m, VMUL(T1n, T1o));
		    T1X = VFNMS(T1n, T1m, VMUL(T1l, T1o));
	       }
	       {
		    V T1b, T1d, T1a, T1c;
		    T1b = LD(&(ri[WS(ios, 7)]), dist, &(ri[WS(ios, 1)]));
		    T1d = LD(&(ii[WS(ios, 7)]), dist, &(ii[WS(ios, 1)]));
		    T1a = LDW(&(W[TWVL * 12]));
		    T1c = LDW(&(W[TWVL * 13]));
		    T1e = VFMA(T1a, T1b, VMUL(T1c, T1d));
		    T21 = VFNMS(T1c, T1b, VMUL(T1a, T1d));
	       }
	       {
		    V T1h, T1j, T1g, T1i;
		    T1h = LD(&(ri[WS(ios, 3)]), dist, &(ri[WS(ios, 1)]));
		    T1j = LD(&(ii[WS(ios, 3)]), dist, &(ii[WS(ios, 1)]));
		    T1g = LDW(&(W[TWVL * 4]));
		    T1i = LDW(&(W[TWVL * 5]));
		    T1k = VFMA(T1g, T1h, VMUL(T1i, T1j));
		    T1W = VFNMS(T1i, T1h, VMUL(T1g, T1j));
	       }
	       T1f = VADD(T19, T1e);
	       T1q = VADD(T1k, T1p);
	       T2B = VSUB(T1f, T1q);
	       T2C = VADD(T20, T21);
	       T2D = VADD(T1W, T1X);
	       T2E = VSUB(T2C, T2D);
	       {
		    V T1V, T1Y, T22, T23;
		    T1V = VSUB(T19, T1e);
		    T1Y = VSUB(T1W, T1X);
		    T1Z = VSUB(T1V, T1Y);
		    T2j = VADD(T1V, T1Y);
		    T22 = VSUB(T20, T21);
		    T23 = VSUB(T1k, T1p);
		    T24 = VADD(T22, T23);
		    T2k = VSUB(T22, T23);
	       }
	  }
	  {
	       V TM, T1K, T12, T1R, TR, T1L, TX, T1Q;
	       {
		    V TJ, TL, TI, TK;
		    TJ = LD(&(ri[WS(ios, 1)]), dist, &(ri[WS(ios, 1)]));
		    TL = LD(&(ii[WS(ios, 1)]), dist, &(ii[WS(ios, 1)]));
		    TI = LDW(&(W[0]));
		    TK = LDW(&(W[TWVL * 1]));
		    TM = VFMA(TI, TJ, VMUL(TK, TL));
		    T1K = VFNMS(TK, TJ, VMUL(TI, TL));
	       }
	       {
		    V TZ, T11, TY, T10;
		    TZ = LD(&(ri[WS(ios, 13)]), dist, &(ri[WS(ios, 1)]));
		    T11 = LD(&(ii[WS(ios, 13)]), dist, &(ii[WS(ios, 1)]));
		    TY = LDW(&(W[TWVL * 24]));
		    T10 = LDW(&(W[TWVL * 25]));
		    T12 = VFMA(TY, TZ, VMUL(T10, T11));
		    T1R = VFNMS(T10, TZ, VMUL(TY, T11));
	       }
	       {
		    V TO, TQ, TN, TP;
		    TO = LD(&(ri[WS(ios, 9)]), dist, &(ri[WS(ios, 1)]));
		    TQ = LD(&(ii[WS(ios, 9)]), dist, &(ii[WS(ios, 1)]));
		    TN = LDW(&(W[TWVL * 16]));
		    TP = LDW(&(W[TWVL * 17]));
		    TR = VFMA(TN, TO, VMUL(TP, TQ));
		    T1L = VFNMS(TP, TO, VMUL(TN, TQ));
	       }
	       {
		    V TU, TW, TT, TV;
		    TU = LD(&(ri[WS(ios, 5)]), dist, &(ri[WS(ios, 1)]));
		    TW = LD(&(ii[WS(ios, 5)]), dist, &(ii[WS(ios, 1)]));
		    TT = LDW(&(W[TWVL * 8]));
		    TV = LDW(&(W[TWVL * 9]));
		    TX = VFMA(TT, TU, VMUL(TV, TW));
		    T1Q = VFNMS(TV, TU, VMUL(TT, TW));
	       }
	       TS = VADD(TM, TR);
	       T13 = VADD(TX, T12);
	       T2w = VSUB(TS, T13);
	       T2x = VADD(T1K, T1L);
	       T2y = VADD(T1Q, T1R);
	       T2z = VSUB(T2x, T2y);
	       {
		    V T1M, T1N, T1P, T1S;
		    T1M = VSUB(T1K, T1L);
		    T1N = VSUB(TX, T12);
		    T1O = VADD(T1M, T1N);
		    T2g = VSUB(T1M, T1N);
		    T1P = VSUB(TM, TR);
		    T1S = VSUB(T1Q, T1R);
		    T1T = VSUB(T1P, T1S);
		    T2h = VADD(T1P, T1S);
	       }
	  }
	  {
	       V T1J, T27, T3g, T3i, T26, T3h, T2a, T3d;
	       {
		    V T1x, T1I, T3e, T3f;
		    T1x = VSUB(T1t, T1w);
		    T1I = VMUL(LDK(KP707106781), VSUB(T1C, T1H));
		    T1J = VADD(T1x, T1I);
		    T27 = VSUB(T1x, T1I);
		    T3e = VMUL(LDK(KP707106781), VSUB(T2d, T2c));
		    T3f = VADD(T38, T37);
		    T3g = VADD(T3e, T3f);
		    T3i = VSUB(T3f, T3e);
	       }
	       {
		    V T1U, T25, T28, T29;
		    T1U = VFMA(LDK(KP923879532), T1O, VMUL(LDK(KP382683432), T1T));
		    T25 = VFNMS(LDK(KP923879532), T24, VMUL(LDK(KP382683432), T1Z));
		    T26 = VADD(T1U, T25);
		    T3h = VSUB(T25, T1U);
		    T28 = VFNMS(LDK(KP923879532), T1T, VMUL(LDK(KP382683432), T1O));
		    T29 = VFMA(LDK(KP382683432), T24, VMUL(LDK(KP923879532), T1Z));
		    T2a = VSUB(T28, T29);
		    T3d = VADD(T28, T29);
	       }
	       ST(&(ri[WS(ios, 11)]), VSUB(T1J, T26), dist, &(ri[WS(ios, 1)]));
	       ST(&(ii[WS(ios, 11)]), VSUB(T3g, T3d), dist, &(ii[WS(ios, 1)]));
	       ST(&(ri[WS(ios, 3)]), VADD(T1J, T26), dist, &(ri[WS(ios, 1)]));
	       ST(&(ii[WS(ios, 3)]), VADD(T3d, T3g), dist, &(ii[WS(ios, 1)]));
	       ST(&(ri[WS(ios, 15)]), VSUB(T27, T2a), dist, &(ri[WS(ios, 1)]));
	       ST(&(ii[WS(ios, 15)]), VSUB(T3i, T3h), dist, &(ii[WS(ios, 1)]));
	       ST(&(ri[WS(ios, 7)]), VADD(T27, T2a), dist, &(ri[WS(ios, 1)]));
	       ST(&(ii[WS(ios, 7)]), VADD(T3h, T3i), dist, &(ii[WS(ios, 1)]));
	  }
	  {
	       V T2v, T2H, T32, T34, T2G, T33, T2K, T2Z;
	       {
		    V T2r, T2u, T30, T31;
		    T2r = VSUB(T7, Ti);
		    T2u = VSUB(T2s, T2t);
		    T2v = VADD(T2r, T2u);
		    T2H = VSUB(T2r, T2u);
		    T30 = VSUB(TF, Tu);
		    T31 = VSUB(T2U, T2R);
		    T32 = VADD(T30, T31);
		    T34 = VSUB(T31, T30);
	       }
	       {
		    V T2A, T2F, T2I, T2J;
		    T2A = VADD(T2w, T2z);
		    T2F = VSUB(T2B, T2E);
		    T2G = VMUL(LDK(KP707106781), VADD(T2A, T2F));
		    T33 = VMUL(LDK(KP707106781), VSUB(T2F, T2A));
		    T2I = VSUB(T2z, T2w);
		    T2J = VADD(T2B, T2E);
		    T2K = VMUL(LDK(KP707106781), VSUB(T2I, T2J));
		    T2Z = VMUL(LDK(KP707106781), VADD(T2I, T2J));
	       }
	       ST(&(ri[WS(ios, 10)]), VSUB(T2v, T2G), dist, &(ri[0]));
	       ST(&(ii[WS(ios, 10)]), VSUB(T32, T2Z), dist, &(ii[0]));
	       ST(&(ri[WS(ios, 2)]), VADD(T2v, T2G), dist, &(ri[0]));
	       ST(&(ii[WS(ios, 2)]), VADD(T2Z, T32), dist, &(ii[0]));
	       ST(&(ri[WS(ios, 14)]), VSUB(T2H, T2K), dist, &(ri[0]));
	       ST(&(ii[WS(ios, 14)]), VSUB(T34, T33), dist, &(ii[0]));
	       ST(&(ri[WS(ios, 6)]), VADD(T2H, T2K), dist, &(ri[0]));
	       ST(&(ii[WS(ios, 6)]), VADD(T33, T34), dist, &(ii[0]));
	  }
	  {
	       V T2f, T2n, T3a, T3c, T2m, T3b, T2q, T35;
	       {
		    V T2b, T2e, T36, T39;
		    T2b = VADD(T1t, T1w);
		    T2e = VMUL(LDK(KP707106781), VADD(T2c, T2d));
		    T2f = VADD(T2b, T2e);
		    T2n = VSUB(T2b, T2e);
		    T36 = VMUL(LDK(KP707106781), VADD(T1C, T1H));
		    T39 = VSUB(T37, T38);
		    T3a = VADD(T36, T39);
		    T3c = VSUB(T39, T36);
	       }
	       {
		    V T2i, T2l, T2o, T2p;
		    T2i = VFMA(LDK(KP382683432), T2g, VMUL(LDK(KP923879532), T2h));
		    T2l = VFNMS(LDK(KP382683432), T2k, VMUL(LDK(KP923879532), T2j));
		    T2m = VADD(T2i, T2l);
		    T3b = VSUB(T2l, T2i);
		    T2o = VFNMS(LDK(KP382683432), T2h, VMUL(LDK(KP923879532), T2g));
		    T2p = VFMA(LDK(KP923879532), T2k, VMUL(LDK(KP382683432), T2j));
		    T2q = VSUB(T2o, T2p);
		    T35 = VADD(T2o, T2p);
	       }
	       ST(&(ri[WS(ios, 9)]), VSUB(T2f, T2m), dist, &(ri[WS(ios, 1)]));
	       ST(&(ii[WS(ios, 9)]), VSUB(T3a, T35), dist, &(ii[WS(ios, 1)]));
	       ST(&(ri[WS(ios, 1)]), VADD(T2f, T2m), dist, &(ri[WS(ios, 1)]));
	       ST(&(ii[WS(ios, 1)]), VADD(T35, T3a), dist, &(ii[WS(ios, 1)]));
	       ST(&(ri[WS(ios, 13)]), VSUB(T2n, T2q), dist, &(ri[WS(ios, 1)]));
	       ST(&(ii[WS(ios, 13)]), VSUB(T3c, T3b), dist, &(ii[WS(ios, 1)]));
	       ST(&(ri[WS(ios, 5)]), VADD(T2n, T2q), dist, &(ri[WS(ios, 1)]));
	       ST(&(ii[WS(ios, 5)]), VADD(T3b, T3c), dist, &(ii[WS(ios, 1)]));
	  }
	  {
	       V TH, T2L, T2W, T2Y, T1s, T2X, T2O, T2P;
	       {
		    V Tj, TG, T2Q, T2V;
		    Tj = VADD(T7, Ti);
		    TG = VADD(Tu, TF);
		    TH = VADD(Tj, TG);
		    T2L = VSUB(Tj, TG);
		    T2Q = VADD(T2s, T2t);
		    T2V = VADD(T2R, T2U);
		    T2W = VADD(T2Q, T2V);
		    T2Y = VSUB(T2V, T2Q);
	       }
	       {
		    V T14, T1r, T2M, T2N;
		    T14 = VADD(TS, T13);
		    T1r = VADD(T1f, T1q);
		    T1s = VADD(T14, T1r);
		    T2X = VSUB(T1r, T14);
		    T2M = VADD(T2x, T2y);
		    T2N = VADD(T2C, T2D);
		    T2O = VSUB(T2M, T2N);
		    T2P = VADD(T2M, T2N);
	       }
	       ST(&(ri[WS(ios, 8)]), VSUB(TH, T1s), dist, &(ri[0]));
	       ST(&(ii[WS(ios, 8)]), VSUB(T2W, T2P), dist, &(ii[0]));
	       ST(&(ri[0]), VADD(TH, T1s), dist, &(ri[0]));
	       ST(&(ii[0]), VADD(T2P, T2W), dist, &(ii[0]));
	       ST(&(ri[WS(ios, 12)]), VSUB(T2L, T2O), dist, &(ri[0]));
	       ST(&(ii[WS(ios, 12)]), VSUB(T2Y, T2X), dist, &(ii[0]));
	       ST(&(ri[WS(ios, 4)]), VADD(T2L, T2O), dist, &(ri[0]));
	       ST(&(ii[WS(ios, 4)]), VADD(T2X, T2Y), dist, &(ii[0]));
	  }
     }
     return W;
}

static const tw_instr twinstr[] = {
     VTW(1),
     VTW(2),
     VTW(3),
     VTW(4),
     VTW(5),
     VTW(6),
     VTW(7),
     VTW(8),
     VTW(9),
     VTW(10),
     VTW(11),
     VTW(12),
     VTW(13),
     VTW(14),
     VTW(15),
     {TW_NEXT, (2 * VL), 0}
};

static const ct_desc desc = { 16, "t1sv_16", twinstr, &GENUS, {136, 46, 38, 0}, 0, 0, 0 };

void X(codelet_t1sv_16) (planner *p) {
     X(kdft_dit_register) (p, t1sv_16, &desc);
}
#endif				/* HAVE_FMA */