hf2_16.c

快速fft变换

	  ci[WS(rs, 2)] = FMA(KP923879532, T35, T32);
	  cr[WS(rs, 5)] = FNMS(KP923879532, T35, T32);
     }
}

static const tw_instr twinstr[] = {
     {TW_CEXP, 1, 1},
     {TW_CEXP, 1, 3},
     {TW_CEXP, 1, 9},
     {TW_CEXP, 1, 15},
     {TW_NEXT, 1, 0}
};

static const hc2hc_desc desc = { 16, "hf2_16", twinstr, &GENUS, {104, 42, 92, 0} };

void X(codelet_hf2_16) (planner *p) {
     X(khc2hc_register) (p, hf2_16, &desc);
}
#else				/* HAVE_FMA */

/* Generated by: ../../../genfft/gen_hc2hc -compact -variables 4 -pipeline-latency 4 -twiddle-log3 -precompute-twiddles -n 16 -dit -name hf2_16 -include hf.h */

/*
 * This function contains 196 FP additions, 108 FP multiplications,
 * (or, 156 additions, 68 multiplications, 40 fused multiply/add),
 * 82 stack variables, 3 constants, and 64 memory accesses
 */
#include "hf.h"

static void hf2_16(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms)
{
     DK(KP382683432, +0.382683432365089771728459984030398866761344562);
     DK(KP923879532, +0.923879532511286756128183189396788286822416626);
     DK(KP707106781, +0.707106781186547524400844362104849039284835938);
     INT m;
     for (m = mb, W = W + ((mb - 1) * 8); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 8, MAKE_VOLATILE_STRIDE(rs)) {
	  E T2, T5, Tg, Ti, Tk, To, TE, TC, T6, T3, T8, TW, TJ, Tt, TU;
	  E Tc, Tx, TH, TN, TO, TP, TR, T1f, T1k, T1b, T1i, T1y, T1H, T1u, T1F;
	  {
	       E T7, Tv, Ta, Ts, T4, Tw, Tb, Tr;
	       {
		    E Th, Tn, Tj, Tm;
		    T2 = W[0];
		    T5 = W[1];
		    Tg = W[2];
		    Ti = W[3];
		    Th = T2 * Tg;
		    Tn = T5 * Tg;
		    Tj = T5 * Ti;
		    Tm = T2 * Ti;
		    Tk = Th - Tj;
		    To = Tm + Tn;
		    TE = Tm - Tn;
		    TC = Th + Tj;
		    T6 = W[5];
		    T7 = T5 * T6;
		    Tv = Tg * T6;
		    Ta = T2 * T6;
		    Ts = Ti * T6;
		    T3 = W[4];
		    T4 = T2 * T3;
		    Tw = Ti * T3;
		    Tb = T5 * T3;
		    Tr = Tg * T3;
	       }
	       T8 = T4 + T7;
	       TW = Tv - Tw;
	       TJ = Ta + Tb;
	       Tt = Tr - Ts;
	       TU = Tr + Ts;
	       Tc = Ta - Tb;
	       Tx = Tv + Tw;
	       TH = T4 - T7;
	       TN = W[6];
	       TO = W[7];
	       TP = FMA(T2, TN, T5 * TO);
	       TR = FNMS(T5, TN, T2 * TO);
	       {
		    E T1d, T1e, T19, T1a;
		    T1d = Tk * T6;
		    T1e = To * T3;
		    T1f = T1d - T1e;
		    T1k = T1d + T1e;
		    T19 = Tk * T3;
		    T1a = To * T6;
		    T1b = T19 + T1a;
		    T1i = T19 - T1a;
	       }
	       {
		    E T1w, T1x, T1s, T1t;
		    T1w = TC * T6;
		    T1x = TE * T3;
		    T1y = T1w - T1x;
		    T1H = T1w + T1x;
		    T1s = TC * T3;
		    T1t = TE * T6;
		    T1u = T1s + T1t;
		    T1F = T1s - T1t;
	       }
	  }
	  {
	       E Tf, T3s, T1N, T3e, TA, T3r, T1Q, T3b, TM, T2N, T1W, T2w, TZ, T2M, T21;
	       E T2x, T1B, T1K, T2V, T2W, T2X, T2Y, T2j, T2E, T2o, T2D, T18, T1n, T2Q, T2R;
	       E T2S, T2T, T28, T2B, T2d, T2A;
	       {
		    E T1, T3d, Te, T3c, T9, Td;
		    T1 = cr[0];
		    T3d = ci[0];
		    T9 = cr[WS(rs, 8)];
		    Td = ci[WS(rs, 8)];
		    Te = FMA(T8, T9, Tc * Td);
		    T3c = FNMS(Tc, T9, T8 * Td);
		    Tf = T1 + Te;
		    T3s = T3d - T3c;
		    T1N = T1 - Te;
		    T3e = T3c + T3d;
	       }
	       {
		    E Tq, T1O, Tz, T1P;
		    {
			 E Tl, Tp, Tu, Ty;
			 Tl = cr[WS(rs, 4)];
			 Tp = ci[WS(rs, 4)];
			 Tq = FMA(Tk, Tl, To * Tp);
			 T1O = FNMS(To, Tl, Tk * Tp);
			 Tu = cr[WS(rs, 12)];
			 Ty = ci[WS(rs, 12)];
			 Tz = FMA(Tt, Tu, Tx * Ty);
			 T1P = FNMS(Tx, Tu, Tt * Ty);
		    }
		    TA = Tq + Tz;
		    T3r = Tq - Tz;
		    T1Q = T1O - T1P;
		    T3b = T1O + T1P;
	       }
	       {
		    E TG, T1T, TL, T1U, T1S, T1V;
		    {
			 E TD, TF, TI, TK;
			 TD = cr[WS(rs, 2)];
			 TF = ci[WS(rs, 2)];
			 TG = FMA(TC, TD, TE * TF);
			 T1T = FNMS(TE, TD, TC * TF);
			 TI = cr[WS(rs, 10)];
			 TK = ci[WS(rs, 10)];
			 TL = FMA(TH, TI, TJ * TK);
			 T1U = FNMS(TJ, TI, TH * TK);
		    }
		    TM = TG + TL;
		    T2N = T1T + T1U;
		    T1S = TG - TL;
		    T1V = T1T - T1U;
		    T1W = T1S - T1V;
		    T2w = T1S + T1V;
	       }
	       {
		    E TT, T1Y, TY, T1Z, T1X, T20;
		    {
			 E TQ, TS, TV, TX;
			 TQ = cr[WS(rs, 14)];
			 TS = ci[WS(rs, 14)];
			 TT = FMA(TP, TQ, TR * TS);
			 T1Y = FNMS(TR, TQ, TP * TS);
			 TV = cr[WS(rs, 6)];
			 TX = ci[WS(rs, 6)];
			 TY = FMA(TU, TV, TW * TX);
			 T1Z = FNMS(TW, TV, TU * TX);
		    }
		    TZ = TT + TY;
		    T2M = T1Y + T1Z;
		    T1X = TT - TY;
		    T20 = T1Y - T1Z;
		    T21 = T1X + T20;
		    T2x = T1X - T20;
	       }
	       {
		    E T1r, T2f, T1J, T2m, T1A, T2g, T1E, T2l;
		    {
			 E T1p, T1q, T1G, T1I;
			 T1p = cr[WS(rs, 15)];
			 T1q = ci[WS(rs, 15)];
			 T1r = FMA(TN, T1p, TO * T1q);
			 T2f = FNMS(TO, T1p, TN * T1q);
			 T1G = cr[WS(rs, 11)];
			 T1I = ci[WS(rs, 11)];
			 T1J = FMA(T1F, T1G, T1H * T1I);
			 T2m = FNMS(T1H, T1G, T1F * T1I);
		    }
		    {
			 E T1v, T1z, T1C, T1D;
			 T1v = cr[WS(rs, 7)];
			 T1z = ci[WS(rs, 7)];
			 T1A = FMA(T1u, T1v, T1y * T1z);
			 T2g = FNMS(T1y, T1v, T1u * T1z);
			 T1C = cr[WS(rs, 3)];
			 T1D = ci[WS(rs, 3)];
			 T1E = FMA(Tg, T1C, Ti * T1D);
			 T2l = FNMS(Ti, T1C, Tg * T1D);
		    }
		    T1B = T1r + T1A;
		    T1K = T1E + T1J;
		    T2V = T1B - T1K;
		    T2W = T2f + T2g;
		    T2X = T2l + T2m;
		    T2Y = T2W - T2X;
		    {
			 E T2h, T2i, T2k, T2n;
			 T2h = T2f - T2g;
			 T2i = T1E - T1J;
			 T2j = T2h + T2i;
			 T2E = T2h - T2i;
			 T2k = T1r - T1A;
			 T2n = T2l - T2m;
			 T2o = T2k - T2n;
			 T2D = T2k + T2n;
		    }
	       }
	       {
		    E T14, T29, T1m, T26, T17, T2a, T1h, T25;
		    {
			 E T12, T13, T1j, T1l;
			 T12 = cr[WS(rs, 1)];
			 T13 = ci[WS(rs, 1)];
			 T14 = FMA(T2, T12, T5 * T13);
			 T29 = FNMS(T5, T12, T2 * T13);
			 T1j = cr[WS(rs, 13)];
			 T1l = ci[WS(rs, 13)];
			 T1m = FMA(T1i, T1j, T1k * T1l);
			 T26 = FNMS(T1k, T1j, T1i * T1l);
		    }
		    {
			 E T15, T16, T1c, T1g;
			 T15 = cr[WS(rs, 9)];
			 T16 = ci[WS(rs, 9)];
			 T17 = FMA(T3, T15, T6 * T16);
			 T2a = FNMS(T6, T15, T3 * T16);
			 T1c = cr[WS(rs, 5)];
			 T1g = ci[WS(rs, 5)];
			 T1h = FMA(T1b, T1c, T1f * T1g);
			 T25 = FNMS(T1f, T1c, T1b * T1g);
		    }
		    T18 = T14 + T17;
		    T1n = T1h + T1m;
		    T2Q = T18 - T1n;
		    T2R = T29 + T2a;
		    T2S = T25 + T26;
		    T2T = T2R - T2S;
		    {
			 E T24, T27, T2b, T2c;
			 T24 = T14 - T17;
			 T27 = T25 - T26;
			 T28 = T24 - T27;
			 T2B = T24 + T27;
			 T2b = T29 - T2a;
			 T2c = T1h - T1m;
			 T2d = T2b + T2c;
			 T2A = T2b - T2c;
		    }
	       }
	       {
		    E T23, T2r, T3u, T3w, T2q, T3v, T2u, T3p;
		    {
			 E T1R, T22, T3q, T3t;
			 T1R = T1N - T1Q;
			 T22 = KP707106781 * (T1W + T21);
			 T23 = T1R + T22;
			 T2r = T1R - T22;
			 T3q = KP707106781 * (T2w - T2x);
			 T3t = T3r + T3s;
			 T3u = T3q + T3t;
			 T3w = T3t - T3q;
		    }
		    {
			 E T2e, T2p, T2s, T2t;
			 T2e = FNMS(KP382683432, T2d, KP923879532 * T28);
			 T2p = FMA(KP382683432, T2j, KP923879532 * T2o);
			 T2q = T2e + T2p;
			 T3v = T2p - T2e;
			 T2s = FMA(KP923879532, T2d, KP382683432 * T28);
			 T2t = FNMS(KP923879532, T2j, KP382683432 * T2o);
			 T2u = T2s + T2t;
			 T3p = T2t - T2s;
		    }
		    cr[WS(rs, 7)] = T23 - T2q;
		    cr[WS(rs, 11)] = T3v - T3w;
		    ci[WS(rs, 12)] = T3v + T3w;
		    ci[0] = T23 + T2q;
		    ci[WS(rs, 4)] = T2r - T2u;
		    cr[WS(rs, 15)] = T3p - T3u;
		    ci[WS(rs, 8)] = T3p + T3u;
		    cr[WS(rs, 3)] = T2r + T2u;
	       }
	       {
		    E T11, T35, T3g, T3i, T1M, T3h, T38, T39;
		    {
			 E TB, T10, T3a, T3f;
			 TB = Tf + TA;
			 T10 = TM + TZ;
			 T11 = TB + T10;
			 T35 = TB - T10;
			 T3a = T2N + T2M;
			 T3f = T3b + T3e;
			 T3g = T3a + T3f;
			 T3i = T3f - T3a;
		    }
		    {
			 E T1o, T1L, T36, T37;
			 T1o = T18 + T1n;
			 T1L = T1B + T1K;
			 T1M = T1o + T1L;
			 T3h = T1L - T1o;
			 T36 = T2W + T2X;
			 T37 = T2R + T2S;
			 T38 = T36 - T37;
			 T39 = T37 + T36;
		    }
		    ci[WS(rs, 7)] = T11 - T1M;
		    cr[WS(rs, 12)] = T3h - T3i;
		    ci[WS(rs, 11)] = T3h + T3i;
		    cr[0] = T11 + T1M;
		    cr[WS(rs, 4)] = T35 - T38;
		    cr[WS(rs, 8)] = T39 - T3g;
		    ci[WS(rs, 15)] = T39 + T3g;
		    ci[WS(rs, 3)] = T35 + T38;
	       }
	       {
		    E T2z, T2H, T3A, T3C, T2G, T3B, T2K, T3x;
		    {
			 E T2v, T2y, T3y, T3z;
			 T2v = T1N + T1Q;
			 T2y = KP707106781 * (T2w + T2x);
			 T2z = T2v + T2y;
			 T2H = T2v - T2y;
			 T3y = KP707106781 * (T21 - T1W);
			 T3z = T3s - T3r;
			 T3A = T3y + T3z;
			 T3C = T3z - T3y;
		    }
		    {
			 E T2C, T2F, T2I, T2J;
			 T2C = FMA(KP382683432, T2A, KP923879532 * T2B);
			 T2F = FNMS(KP382683432, T2E, KP923879532 * T2D);
			 T2G = T2C + T2F;
			 T3B = T2F - T2C;
			 T2I = FNMS(KP923879532, T2A, KP382683432 * T2B);
			 T2J = FMA(KP923879532, T2E, KP382683432 * T2D);
			 T2K = T2I + T2J;
			 T3x = T2J - T2I;
		    }
		    ci[WS(rs, 6)] = T2z - T2G;
		    cr[WS(rs, 13)] = T3B - T3C;
		    ci[WS(rs, 10)] = T3B + T3C;
		    cr[WS(rs, 1)] = T2z + T2G;
		    cr[WS(rs, 5)] = T2H - T2K;
		    cr[WS(rs, 9)] = T3x - T3A;
		    ci[WS(rs, 14)] = T3x + T3A;
		    ci[WS(rs, 2)] = T2H + T2K;
	       }
	       {
		    E T2P, T31, T3m, T3o, T30, T3j, T34, T3n;
		    {
			 E T2L, T2O, T3k, T3l;
			 T2L = Tf - TA;
			 T2O = T2M - T2N;
			 T2P = T2L - T2O;
			 T31 = T2L + T2O;
			 T3k = TM - TZ;
			 T3l = T3e - T3b;
			 T3m = T3k + T3l;
			 T3o = T3l - T3k;
		    }
		    {
			 E T2U, T2Z, T32, T33;
			 T2U = T2Q + T2T;
			 T2Z = T2V - T2Y;
			 T30 = KP707106781 * (T2U + T2Z);
			 T3j = KP707106781 * (T2Z - T2U);
			 T32 = T2Q - T2T;
			 T33 = T2V + T2Y;
			 T34 = KP707106781 * (T32 + T33);
			 T3n = KP707106781 * (T33 - T32);
		    }
		    ci[WS(rs, 5)] = T2P - T30;
		    cr[WS(rs, 10)] = T3n - T3o;
		    ci[WS(rs, 13)] = T3n + T3o;
		    cr[WS(rs, 2)] = T2P + T30;
		    cr[WS(rs, 6)] = T31 - T34;
		    cr[WS(rs, 14)] = T3j - T3m;
		    ci[WS(rs, 9)] = T3j + T3m;
		    ci[WS(rs, 1)] = T31 + T34;
	       }
	  }
     }
}

static const tw_instr twinstr[] = {
     {TW_CEXP, 1, 1},
     {TW_CEXP, 1, 3},
     {TW_CEXP, 1, 9},
     {TW_CEXP, 1, 15},
     {TW_NEXT, 1, 0}
};

static const hc2hc_desc desc = { 16, "hf2_16", twinstr, &GENUS, {156, 68, 40, 0} };

void X(codelet_hf2_16) (planner *p) {
     X(khc2hc_register) (p, hf2_16, &desc);
}
#endif				/* HAVE_FMA */