n2bv_20.c

快速fft变换

static const kdft_desc desc = { 20, "n2bv_20", {58, 4, 46, 0}, &GENUS, 0, 2, 0, 0 };
void X(codelet_n2bv_20) (planner *p) {
     X(kdft_register) (p, n2bv_20, &desc);
}

#else				/* HAVE_FMA */

/* Generated by: ../../../genfft/gen_notw_c -simd -compact -variables 4 -pipeline-latency 8 -sign 1 -n 20 -name n2bv_20 -with-ostride 2 -include n2b.h -store-multiple 2 */

/*
 * This function contains 104 FP additions, 24 FP multiplications,
 * (or, 92 additions, 12 multiplications, 12 fused multiply/add),
 * 57 stack variables, 4 constants, and 50 memory accesses
 */
#include "n2b.h"

static void n2bv_20(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs)
{
     DVK(KP587785252, +0.587785252292473129168705954639072768597652438);
     DVK(KP951056516, +0.951056516295153572116439333379382143405698634);
     DVK(KP250000000, +0.250000000000000000000000000000000000000000000);
     DVK(KP559016994, +0.559016994374947424102293417182819058860154590);
     INT i;
     const R *xi;
     R *xo;
     xi = ii;
     xo = io;
     for (i = v; i > 0; i = i - VL, xi = xi + (VL * ivs), xo = xo + (VL * ovs), MAKE_VOLATILE_STRIDE(is), MAKE_VOLATILE_STRIDE(os)) {
	  V T3, T1y, TH, T1i, Ts, TL, TM, Tz, T13, T16, T1j, T1u, T1v, T1w, T1r;
	  V T1s, T1t, T1a, T1d, T1k, Ti, Tk, TE, TI;
	  {
	       V T1, T2, T1g, TF, TG, T1h;
	       T1 = LD(&(xi[0]), ivs, &(xi[0]));
	       T2 = LD(&(xi[WS(is, 10)]), ivs, &(xi[0]));
	       T1g = VADD(T1, T2);
	       TF = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)]));
	       TG = LD(&(xi[WS(is, 15)]), ivs, &(xi[WS(is, 1)]));
	       T1h = VADD(TF, TG);
	       T3 = VSUB(T1, T2);
	       T1y = VADD(T1g, T1h);
	       TH = VSUB(TF, TG);
	       T1i = VSUB(T1g, T1h);
	  }
	  {
	       V T6, T11, Tv, T19, Ty, T1c, T9, T14, Td, T18, To, T12, Tr, T15, Tg;
	       V T1b;
	       {
		    V T4, T5, Tt, Tu;
		    T4 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0]));
		    T5 = LD(&(xi[WS(is, 14)]), ivs, &(xi[0]));
		    T6 = VSUB(T4, T5);
		    T11 = VADD(T4, T5);
		    Tt = LD(&(xi[WS(is, 13)]), ivs, &(xi[WS(is, 1)]));
		    Tu = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)]));
		    Tv = VSUB(Tt, Tu);
		    T19 = VADD(Tt, Tu);
	       }
	       {
		    V Tw, Tx, T7, T8;
		    Tw = LD(&(xi[WS(is, 17)]), ivs, &(xi[WS(is, 1)]));
		    Tx = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)]));
		    Ty = VSUB(Tw, Tx);
		    T1c = VADD(Tw, Tx);
		    T7 = LD(&(xi[WS(is, 16)]), ivs, &(xi[0]));
		    T8 = LD(&(xi[WS(is, 6)]), ivs, &(xi[0]));
		    T9 = VSUB(T7, T8);
		    T14 = VADD(T7, T8);
	       }
	       {
		    V Tb, Tc, Tm, Tn;
		    Tb = LD(&(xi[WS(is, 8)]), ivs, &(xi[0]));
		    Tc = LD(&(xi[WS(is, 18)]), ivs, &(xi[0]));
		    Td = VSUB(Tb, Tc);
		    T18 = VADD(Tb, Tc);
		    Tm = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)]));
		    Tn = LD(&(xi[WS(is, 19)]), ivs, &(xi[WS(is, 1)]));
		    To = VSUB(Tm, Tn);
		    T12 = VADD(Tm, Tn);
	       }
	       {
		    V Tp, Tq, Te, Tf;
		    Tp = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)]));
		    Tq = LD(&(xi[WS(is, 11)]), ivs, &(xi[WS(is, 1)]));
		    Tr = VSUB(Tp, Tq);
		    T15 = VADD(Tp, Tq);
		    Te = LD(&(xi[WS(is, 12)]), ivs, &(xi[0]));
		    Tf = LD(&(xi[WS(is, 2)]), ivs, &(xi[0]));
		    Tg = VSUB(Te, Tf);
		    T1b = VADD(Te, Tf);
	       }
	       Ts = VSUB(To, Tr);
	       TL = VSUB(T6, T9);
	       TM = VSUB(Td, Tg);
	       Tz = VSUB(Tv, Ty);
	       T13 = VSUB(T11, T12);
	       T16 = VSUB(T14, T15);
	       T1j = VADD(T13, T16);
	       T1u = VADD(T18, T19);
	       T1v = VADD(T1b, T1c);
	       T1w = VADD(T1u, T1v);
	       T1r = VADD(T11, T12);
	       T1s = VADD(T14, T15);
	       T1t = VADD(T1r, T1s);
	       T1a = VSUB(T18, T19);
	       T1d = VSUB(T1b, T1c);
	       T1k = VADD(T1a, T1d);
	       {
		    V Ta, Th, TC, TD;
		    Ta = VADD(T6, T9);
		    Th = VADD(Td, Tg);
		    Ti = VADD(Ta, Th);
		    Tk = VMUL(LDK(KP559016994), VSUB(Ta, Th));
		    TC = VADD(To, Tr);
		    TD = VADD(Tv, Ty);
		    TE = VMUL(LDK(KP559016994), VSUB(TC, TD));
		    TI = VADD(TC, TD);
	       }
	  }
	  {
	       V T1H, T1J, T1K, T1L, T1N, T1I, TZ, T10;
	       TZ = VADD(T3, Ti);
	       T10 = VBYI(VADD(TH, TI));
	       T1H = VSUB(TZ, T10);
	       STM2(&(xo[30]), T1H, ovs, &(xo[2]));
	       T1I = VADD(TZ, T10);
	       STM2(&(xo[10]), T1I, ovs, &(xo[2]));
	       {
		    V T1x, T1z, T1A, T1E, T1G, T1C, T1D, T1F, T1B, T1M;
		    T1x = VMUL(LDK(KP559016994), VSUB(T1t, T1w));
		    T1z = VADD(T1t, T1w);
		    T1A = VFNMS(LDK(KP250000000), T1z, T1y);
		    T1C = VSUB(T1r, T1s);
		    T1D = VSUB(T1u, T1v);
		    T1E = VBYI(VFMA(LDK(KP951056516), T1C, VMUL(LDK(KP587785252), T1D)));
		    T1G = VBYI(VFNMS(LDK(KP951056516), T1D, VMUL(LDK(KP587785252), T1C)));
		    T1J = VADD(T1y, T1z);
		    STM2(&(xo[0]), T1J, ovs, &(xo[0]));
		    T1F = VSUB(T1A, T1x);
		    T1K = VSUB(T1F, T1G);
		    STM2(&(xo[16]), T1K, ovs, &(xo[0]));
		    T1L = VADD(T1G, T1F);
		    STM2(&(xo[24]), T1L, ovs, &(xo[0]));
		    T1B = VADD(T1x, T1A);
		    T1M = VSUB(T1B, T1E);
		    STM2(&(xo[8]), T1M, ovs, &(xo[0]));
		    STN2(&(xo[8]), T1M, T1I, ovs);
		    T1N = VADD(T1E, T1B);
		    STM2(&(xo[32]), T1N, ovs, &(xo[0]));
	       }
	       {
		    V T1O, T1P, T1R, T1S;
		    {
			 V T1n, T1l, T1m, T1f, T1p, T17, T1e, T1q, T1Q, T1o;
			 T1n = VMUL(LDK(KP559016994), VSUB(T1j, T1k));
			 T1l = VADD(T1j, T1k);
			 T1m = VFNMS(LDK(KP250000000), T1l, T1i);
			 T17 = VSUB(T13, T16);
			 T1e = VSUB(T1a, T1d);
			 T1f = VBYI(VFNMS(LDK(KP951056516), T1e, VMUL(LDK(KP587785252), T17)));
			 T1p = VBYI(VFMA(LDK(KP951056516), T17, VMUL(LDK(KP587785252), T1e)));
			 T1O = VADD(T1i, T1l);
			 STM2(&(xo[20]), T1O, ovs, &(xo[0]));
			 T1q = VADD(T1n, T1m);
			 T1P = VADD(T1p, T1q);
			 STM2(&(xo[12]), T1P, ovs, &(xo[0]));
			 T1Q = VSUB(T1q, T1p);
			 STM2(&(xo[28]), T1Q, ovs, &(xo[0]));
			 STN2(&(xo[28]), T1Q, T1H, ovs);
			 T1o = VSUB(T1m, T1n);
			 T1R = VADD(T1f, T1o);
			 STM2(&(xo[4]), T1R, ovs, &(xo[0]));
			 T1S = VSUB(T1o, T1f);
			 STM2(&(xo[36]), T1S, ovs, &(xo[0]));
		    }
		    {
			 V TA, TN, TU, TS, TK, TV, Tl, TR, TJ, Tj;
			 TA = VFNMS(LDK(KP951056516), Tz, VMUL(LDK(KP587785252), Ts));
			 TN = VFNMS(LDK(KP951056516), TM, VMUL(LDK(KP587785252), TL));
			 TU = VFMA(LDK(KP951056516), TL, VMUL(LDK(KP587785252), TM));
			 TS = VFMA(LDK(KP951056516), Ts, VMUL(LDK(KP587785252), Tz));
			 TJ = VFNMS(LDK(KP250000000), TI, TH);
			 TK = VSUB(TE, TJ);
			 TV = VADD(TE, TJ);
			 Tj = VFNMS(LDK(KP250000000), Ti, T3);
			 Tl = VSUB(Tj, Tk);
			 TR = VADD(Tk, Tj);
			 {
			      V TB, TO, T1T, T1U;
			      TB = VSUB(Tl, TA);
			      TO = VBYI(VSUB(TK, TN));
			      T1T = VSUB(TB, TO);
			      STM2(&(xo[34]), T1T, ovs, &(xo[2]));
			      STN2(&(xo[32]), T1N, T1T, ovs);
			      T1U = VADD(TB, TO);
			      STM2(&(xo[6]), T1U, ovs, &(xo[2]));
			      STN2(&(xo[4]), T1R, T1U, ovs);
			 }
			 {
			      V TX, TY, T1V, T1W;
			      TX = VADD(TR, TS);
			      TY = VBYI(VSUB(TV, TU));
			      T1V = VSUB(TX, TY);
			      STM2(&(xo[22]), T1V, ovs, &(xo[2]));
			      STN2(&(xo[20]), T1O, T1V, ovs);
			      T1W = VADD(TX, TY);
			      STM2(&(xo[18]), T1W, ovs, &(xo[2]));
			      STN2(&(xo[16]), T1K, T1W, ovs);
			 }
			 {
			      V TP, TQ, T1X, T1Y;
			      TP = VADD(Tl, TA);
			      TQ = VBYI(VADD(TN, TK));
			      T1X = VSUB(TP, TQ);
			      STM2(&(xo[26]), T1X, ovs, &(xo[2]));
			      STN2(&(xo[24]), T1L, T1X, ovs);
			      T1Y = VADD(TP, TQ);
			      STM2(&(xo[14]), T1Y, ovs, &(xo[2]));
			      STN2(&(xo[12]), T1P, T1Y, ovs);
			 }
			 {
			      V TT, TW, T1Z, T20;
			      TT = VSUB(TR, TS);
			      TW = VBYI(VADD(TU, TV));
			      T1Z = VSUB(TT, TW);
			      STM2(&(xo[38]), T1Z, ovs, &(xo[2]));
			      STN2(&(xo[36]), T1S, T1Z, ovs);
			      T20 = VADD(TT, TW);
			      STM2(&(xo[2]), T20, ovs, &(xo[2]));
			      STN2(&(xo[0]), T1J, T20, ovs);
			 }
		    }
	       }
	  }
     }
}

static const kdft_desc desc = { 20, "n2bv_20", {92, 12, 12, 0}, &GENUS, 0, 2, 0, 0 };
void X(codelet_n2bv_20) (planner *p) {
     X(kdft_register) (p, n2bv_20, &desc);
}

#endif				/* HAVE_FMA */