hc2cfdft_10.c

快速fft变换

	       Rm[WS(rs, 4)] = KP500000000 * (T1x + T1U);
	       T1V = FNMS(KP250000000, T1U, T1x);
	       T21 = FNMS(KP559016994, T1W, T1V);
	       T1X = FMA(KP559016994, T1W, T1V);
	       Rm[0] = KP500000000 * (FNMS(KP951056516, T20, T1X));
	       Rp[WS(rs, 1)] = KP500000000 * (FMA(KP951056516, T20, T1X));
	       Rm[WS(rs, 2)] = KP500000000 * (FNMS(KP951056516, T22, T21));
	       Rp[WS(rs, 3)] = KP500000000 * (FMA(KP951056516, T22, T21));
	  }
     }
}

static const tw_instr twinstr[] = {
     {TW_FULL, 1, 10},
     {TW_NEXT, 1, 0}
};

static const hc2c_desc desc = { 10, "hc2cfdft_10", twinstr, &GENUS, {68, 38, 54, 0} };

void X(codelet_hc2cfdft_10) (planner *p) {
     X(khc2c_register) (p, hc2cfdft_10, &desc, HC2C_VIA_DFT);
}
#else				/* HAVE_FMA */

/* Generated by: ../../../genfft/gen_hc2cdft -compact -variables 4 -pipeline-latency 4 -n 10 -dit -name hc2cfdft_10 -include hc2cf.h */

/*
 * This function contains 122 FP additions, 68 FP multiplications,
 * (or, 92 additions, 38 multiplications, 30 fused multiply/add),
 * 62 stack variables, 5 constants, and 40 memory accesses
 */
#include "hc2cf.h"

static void hc2cfdft_10(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms)
{
     DK(KP293892626, +0.293892626146236564584352977319536384298826219);
     DK(KP475528258, +0.475528258147576786058219666689691071702849317);
     DK(KP125000000, +0.125000000000000000000000000000000000000000000);
     DK(KP500000000, +0.500000000000000000000000000000000000000000000);
     DK(KP279508497, +0.279508497187473712051146708591409529430077295);
     INT m;
     for (m = mb, W = W + ((mb - 1) * 18); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 18, MAKE_VOLATILE_STRIDE(rs)) {
	  E Tw, TL, TM, T1W, T1X, T27, T1Z, T20, T26, TX, T1a, T1b, T1d, T1e, T1f;
	  E T1q, T1t, T1u, T1x, T1A, T1B, T1g, T1h, T1i, Td, T25, T1k, T1F;
	  {
	       E T3, T1D, T19, T1z, T7, Tb, TR, T1v, Tm, T1o, TK, T1s, Tv, T1p, T12;
	       E T1y, TF, T1r, TW, T1w;
	       {
		    E T1, T2, T18, T14, T15, T16, T13, T17;
		    T1 = Ip[0];
		    T2 = Im[0];
		    T18 = T1 + T2;
		    T14 = Rm[0];
		    T15 = Rp[0];
		    T16 = T14 - T15;
		    T3 = T1 - T2;
		    T1D = T15 + T14;
		    T13 = W[0];
		    T17 = W[1];
		    T19 = FNMS(T17, T18, T13 * T16);
		    T1z = FMA(T17, T16, T13 * T18);
	       }
	       {
		    E T5, T6, TO, T9, Ta, TQ, TN, TP;
		    T5 = Ip[WS(rs, 2)];
		    T6 = Im[WS(rs, 2)];
		    TO = T5 - T6;
		    T9 = Rp[WS(rs, 2)];
		    Ta = Rm[WS(rs, 2)];
		    TQ = T9 + Ta;
		    T7 = T5 + T6;
		    Tb = T9 - Ta;
		    TN = W[6];
		    TP = W[7];
		    TR = FNMS(TP, TQ, TN * TO);
		    T1v = FMA(TP, TO, TN * TQ);
	       }
	       {
		    E Th, TJ, Tl, TH;
		    {
			 E Tf, Tg, Tj, Tk;
			 Tf = Ip[WS(rs, 1)];
			 Tg = Im[WS(rs, 1)];
			 Th = Tf - Tg;
			 TJ = Tf + Tg;
			 Tj = Rp[WS(rs, 1)];
			 Tk = Rm[WS(rs, 1)];
			 Tl = Tj + Tk;
			 TH = Tj - Tk;
		    }
		    {
			 E Te, Ti, TG, TI;
			 Te = W[2];
			 Ti = W[3];
			 Tm = FNMS(Ti, Tl, Te * Th);
			 T1o = FMA(Te, Tl, Ti * Th);
			 TG = W[4];
			 TI = W[5];
			 TK = FMA(TG, TH, TI * TJ);
			 T1s = FNMS(TI, TH, TG * TJ);
		    }
	       }
	       {
		    E Tq, TZ, Tu, T11;
		    {
			 E To, Tp, Ts, Tt;
			 To = Ip[WS(rs, 3)];
			 Tp = Im[WS(rs, 3)];
			 Tq = To + Tp;
			 TZ = To - Tp;
			 Ts = Rp[WS(rs, 3)];
			 Tt = Rm[WS(rs, 3)];
			 Tu = Ts - Tt;
			 T11 = Ts + Tt;
		    }
		    {
			 E Tn, Tr, TY, T10;
			 Tn = W[13];
			 Tr = W[12];
			 Tv = FMA(Tn, Tq, Tr * Tu);
			 T1p = FNMS(Tn, Tu, Tr * Tq);
			 TY = W[10];
			 T10 = W[11];
			 T12 = FNMS(T10, T11, TY * TZ);
			 T1y = FMA(T10, TZ, TY * T11);
		    }
	       }
	       {
		    E TA, TV, TE, TT;
		    {
			 E Ty, Tz, TC, TD;
			 Ty = Ip[WS(rs, 4)];
			 Tz = Im[WS(rs, 4)];
			 TA = Ty - Tz;
			 TV = Ty + Tz;
			 TC = Rp[WS(rs, 4)];
			 TD = Rm[WS(rs, 4)];
			 TE = TC + TD;
			 TT = TC - TD;
		    }
		    {
			 E Tx, TB, TS, TU;
			 Tx = W[14];
			 TB = W[15];
			 TF = FNMS(TB, TE, Tx * TA);
			 T1r = FMA(Tx, TE, TB * TA);
			 TS = W[16];
			 TU = W[17];
			 TW = FMA(TS, TT, TU * TV);
			 T1w = FNMS(TU, TT, TS * TV);
		    }
	       }
	       Tw = Tm - Tv;
	       TL = TF - TK;
	       TM = Tw + TL;
	       T1W = T1v + T1w;
	       T1X = T1y + T1z;
	       T27 = T1W + T1X;
	       T1Z = T1o + T1p;
	       T20 = T1s + T1r;
	       T26 = T1Z + T20;
	       TX = TR - TW;
	       T1a = T12 + T19;
	       T1b = TX + T1a;
	       T1d = T19 - T12;
	       T1e = TR + TW;
	       T1f = T1d - T1e;
	       T1q = T1o - T1p;
	       T1t = T1r - T1s;
	       T1u = T1q + T1t;
	       T1x = T1v - T1w;
	       T1A = T1y - T1z;
	       T1B = T1x + T1A;
	       T1g = Tm + Tv;
	       T1h = TK + TF;
	       T1i = T1g + T1h;
	       {
		    E Tc, T1E, T4, T8;
		    T4 = W[9];
		    T8 = W[8];
		    Tc = FMA(T4, T7, T8 * Tb);
		    T1E = FNMS(T4, Tb, T8 * T7);
		    Td = T3 - Tc;
		    T25 = T1D + T1E;
		    T1k = Tc + T3;
		    T1F = T1D - T1E;
	       }
	  }
	  {
	       E T1U, T1c, T1T, T22, T24, T1Y, T21, T23, T1V;
	       T1U = KP279508497 * (TM - T1b);
	       T1c = TM + T1b;
	       T1T = FNMS(KP125000000, T1c, KP500000000 * Td);
	       T1Y = T1W - T1X;
	       T21 = T1Z - T20;
	       T22 = FNMS(KP293892626, T21, KP475528258 * T1Y);
	       T24 = FMA(KP475528258, T21, KP293892626 * T1Y);
	       Ip[0] = KP500000000 * (Td + T1c);
	       T23 = T1U + T1T;
	       Ip[WS(rs, 4)] = T23 + T24;
	       Im[WS(rs, 3)] = T24 - T23;
	       T1V = T1T - T1U;
	       Ip[WS(rs, 2)] = T1V + T22;
	       Im[WS(rs, 1)] = T22 - T1V;
	  }
	  {
	       E T2a, T28, T29, T2e, T2g, T2c, T2d, T2f, T2b;
	       T2a = KP279508497 * (T26 - T27);
	       T28 = T26 + T27;
	       T29 = FNMS(KP125000000, T28, KP500000000 * T25);
	       T2c = TX - T1a;
	       T2d = Tw - TL;
	       T2e = FNMS(KP293892626, T2d, KP475528258 * T2c);
	       T2g = FMA(KP475528258, T2d, KP293892626 * T2c);
	       Rp[0] = KP500000000 * (T25 + T28);
	       T2f = T2a + T29;
	       Rp[WS(rs, 4)] = T2f - T2g;
	       Rm[WS(rs, 3)] = T2g + T2f;
	       T2b = T29 - T2a;
	       Rp[WS(rs, 2)] = T2b - T2e;
	       Rm[WS(rs, 1)] = T2e + T2b;
	  }
	  {
	       E T1M, T1j, T1L, T1Q, T1S, T1O, T1P, T1R, T1N;
	       T1M = KP279508497 * (T1i + T1f);
	       T1j = T1f - T1i;
	       T1L = FMA(KP500000000, T1k, KP125000000 * T1j);
	       T1O = T1A - T1x;
	       T1P = T1q - T1t;
	       T1Q = FNMS(KP475528258, T1P, KP293892626 * T1O);
	       T1S = FMA(KP293892626, T1P, KP475528258 * T1O);
	       Im[WS(rs, 4)] = KP500000000 * (T1j - T1k);
	       T1R = T1L - T1M;
	       Ip[WS(rs, 3)] = T1R + T1S;
	       Im[WS(rs, 2)] = T1S - T1R;
	       T1N = T1L + T1M;
	       Ip[WS(rs, 1)] = T1N + T1Q;
	       Im[0] = T1Q - T1N;
	  }
	  {
	       E T1C, T1G, T1H, T1n, T1J, T1l, T1m, T1K, T1I;
	       T1C = KP279508497 * (T1u - T1B);
	       T1G = T1u + T1B;
	       T1H = FNMS(KP125000000, T1G, KP500000000 * T1F);
	       T1l = T1g - T1h;
	       T1m = T1e + T1d;
	       T1n = FMA(KP475528258, T1l, KP293892626 * T1m);
	       T1J = FNMS(KP293892626, T1l, KP475528258 * T1m);
	       Rm[WS(rs, 4)] = KP500000000 * (T1F + T1G);
	       T1K = T1H - T1C;
	       Rp[WS(rs, 3)] = T1J + T1K;
	       Rm[WS(rs, 2)] = T1K - T1J;
	       T1I = T1C + T1H;
	       Rp[WS(rs, 1)] = T1n + T1I;
	       Rm[0] = T1I - T1n;
	  }
     }
}

static const tw_instr twinstr[] = {
     {TW_FULL, 1, 10},
     {TW_NEXT, 1, 0}
};

static const hc2c_desc desc = { 10, "hc2cfdft_10", twinstr, &GENUS, {92, 38, 30, 0} };

void X(codelet_hc2cfdft_10) (planner *p) {
     X(khc2c_register) (p, hc2cfdft_10, &desc, HC2C_VIA_DFT);
}
#endif				/* HAVE_FMA */