q1fv_5.c

快速fft变换

static const ct_desc desc = { 5, "q1fv_5", twinstr, &GENUS, {55, 50, 45, 0}, 0, 0, 0 };

void X(codelet_q1fv_5) (planner *p) {
     X(kdft_difsq_register) (p, q1fv_5, &desc);
}
#else				/* HAVE_FMA */

/* Generated by: ../../../genfft/gen_twidsq_c -simd -compact -variables 4 -pipeline-latency 8 -n 5 -dif -name q1fv_5 -include q1f.h */

/*
 * This function contains 100 FP additions, 70 FP multiplications,
 * (or, 85 additions, 55 multiplications, 15 fused multiply/add),
 * 44 stack variables, 4 constants, and 50 memory accesses
 */
#include "q1f.h"

static void q1fv_5(R *ri, R *ii, const R *W, stride rs, stride vs, INT mb, INT me, INT ms)
{
     DVK(KP250000000, +0.250000000000000000000000000000000000000000000);
     DVK(KP587785252, +0.587785252292473129168705954639072768597652438);
     DVK(KP951056516, +0.951056516295153572116439333379382143405698634);
     DVK(KP559016994, +0.559016994374947424102293417182819058860154590);
     INT m;
     R *x;
     x = ri;
     for (m = mb, W = W + (mb * ((TWVL / VL) * 8)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 8), MAKE_VOLATILE_STRIDE(rs), MAKE_VOLATILE_STRIDE(vs)) {
	  V T8, T7, Th, Te, T9, Ta, T1q, T1p, T1z, T1w, T1r, T1s, Ts, Tr, TB;
	  V Ty, Tt, Tu, TM, TL, TV, TS, TN, TO, T16, T15, T1f, T1c, T17, T18;
	  {
	       V T6, Td, T3, Tc;
	       T8 = LD(&(x[0]), ms, &(x[0]));
	       {
		    V T4, T5, T1, T2;
		    T4 = LD(&(x[WS(rs, 2)]), ms, &(x[0]));
		    T5 = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)]));
		    T6 = VADD(T4, T5);
		    Td = VSUB(T4, T5);
		    T1 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)]));
		    T2 = LD(&(x[WS(rs, 4)]), ms, &(x[0]));
		    T3 = VADD(T1, T2);
		    Tc = VSUB(T1, T2);
	       }
	       T7 = VMUL(LDK(KP559016994), VSUB(T3, T6));
	       Th = VBYI(VFNMS(LDK(KP587785252), Tc, VMUL(LDK(KP951056516), Td)));
	       Te = VBYI(VFMA(LDK(KP951056516), Tc, VMUL(LDK(KP587785252), Td)));
	       T9 = VADD(T3, T6);
	       Ta = VFNMS(LDK(KP250000000), T9, T8);
	  }
	  {
	       V T1o, T1v, T1l, T1u;
	       T1q = LD(&(x[WS(vs, 4)]), ms, &(x[WS(vs, 4)]));
	       {
		    V T1m, T1n, T1j, T1k;
		    T1m = LD(&(x[WS(vs, 4) + WS(rs, 2)]), ms, &(x[WS(vs, 4)]));
		    T1n = LD(&(x[WS(vs, 4) + WS(rs, 3)]), ms, &(x[WS(vs, 4) + WS(rs, 1)]));
		    T1o = VADD(T1m, T1n);
		    T1v = VSUB(T1m, T1n);
		    T1j = LD(&(x[WS(vs, 4) + WS(rs, 1)]), ms, &(x[WS(vs, 4) + WS(rs, 1)]));
		    T1k = LD(&(x[WS(vs, 4) + WS(rs, 4)]), ms, &(x[WS(vs, 4)]));
		    T1l = VADD(T1j, T1k);
		    T1u = VSUB(T1j, T1k);
	       }
	       T1p = VMUL(LDK(KP559016994), VSUB(T1l, T1o));
	       T1z = VBYI(VFNMS(LDK(KP587785252), T1u, VMUL(LDK(KP951056516), T1v)));
	       T1w = VBYI(VFMA(LDK(KP951056516), T1u, VMUL(LDK(KP587785252), T1v)));
	       T1r = VADD(T1l, T1o);
	       T1s = VFNMS(LDK(KP250000000), T1r, T1q);
	  }
	  {
	       V Tq, Tx, Tn, Tw;
	       Ts = LD(&(x[WS(vs, 1)]), ms, &(x[WS(vs, 1)]));
	       {
		    V To, Tp, Tl, Tm;
		    To = LD(&(x[WS(vs, 1) + WS(rs, 2)]), ms, &(x[WS(vs, 1)]));
		    Tp = LD(&(x[WS(vs, 1) + WS(rs, 3)]), ms, &(x[WS(vs, 1) + WS(rs, 1)]));
		    Tq = VADD(To, Tp);
		    Tx = VSUB(To, Tp);
		    Tl = LD(&(x[WS(vs, 1) + WS(rs, 1)]), ms, &(x[WS(vs, 1) + WS(rs, 1)]));
		    Tm = LD(&(x[WS(vs, 1) + WS(rs, 4)]), ms, &(x[WS(vs, 1)]));
		    Tn = VADD(Tl, Tm);
		    Tw = VSUB(Tl, Tm);
	       }
	       Tr = VMUL(LDK(KP559016994), VSUB(Tn, Tq));
	       TB = VBYI(VFNMS(LDK(KP587785252), Tw, VMUL(LDK(KP951056516), Tx)));
	       Ty = VBYI(VFMA(LDK(KP951056516), Tw, VMUL(LDK(KP587785252), Tx)));
	       Tt = VADD(Tn, Tq);
	       Tu = VFNMS(LDK(KP250000000), Tt, Ts);
	  }
	  {
	       V TK, TR, TH, TQ;
	       TM = LD(&(x[WS(vs, 2)]), ms, &(x[WS(vs, 2)]));
	       {
		    V TI, TJ, TF, TG;
		    TI = LD(&(x[WS(vs, 2) + WS(rs, 2)]), ms, &(x[WS(vs, 2)]));
		    TJ = LD(&(x[WS(vs, 2) + WS(rs, 3)]), ms, &(x[WS(vs, 2) + WS(rs, 1)]));
		    TK = VADD(TI, TJ);
		    TR = VSUB(TI, TJ);
		    TF = LD(&(x[WS(vs, 2) + WS(rs, 1)]), ms, &(x[WS(vs, 2) + WS(rs, 1)]));
		    TG = LD(&(x[WS(vs, 2) + WS(rs, 4)]), ms, &(x[WS(vs, 2)]));
		    TH = VADD(TF, TG);
		    TQ = VSUB(TF, TG);
	       }
	       TL = VMUL(LDK(KP559016994), VSUB(TH, TK));
	       TV = VBYI(VFNMS(LDK(KP587785252), TQ, VMUL(LDK(KP951056516), TR)));
	       TS = VBYI(VFMA(LDK(KP951056516), TQ, VMUL(LDK(KP587785252), TR)));
	       TN = VADD(TH, TK);
	       TO = VFNMS(LDK(KP250000000), TN, TM);
	  }
	  {
	       V T14, T1b, T11, T1a;
	       T16 = LD(&(x[WS(vs, 3)]), ms, &(x[WS(vs, 3)]));
	       {
		    V T12, T13, TZ, T10;
		    T12 = LD(&(x[WS(vs, 3) + WS(rs, 2)]), ms, &(x[WS(vs, 3)]));
		    T13 = LD(&(x[WS(vs, 3) + WS(rs, 3)]), ms, &(x[WS(vs, 3) + WS(rs, 1)]));
		    T14 = VADD(T12, T13);
		    T1b = VSUB(T12, T13);
		    TZ = LD(&(x[WS(vs, 3) + WS(rs, 1)]), ms, &(x[WS(vs, 3) + WS(rs, 1)]));
		    T10 = LD(&(x[WS(vs, 3) + WS(rs, 4)]), ms, &(x[WS(vs, 3)]));
		    T11 = VADD(TZ, T10);
		    T1a = VSUB(TZ, T10);
	       }
	       T15 = VMUL(LDK(KP559016994), VSUB(T11, T14));
	       T1f = VBYI(VFNMS(LDK(KP587785252), T1a, VMUL(LDK(KP951056516), T1b)));
	       T1c = VBYI(VFMA(LDK(KP951056516), T1a, VMUL(LDK(KP587785252), T1b)));
	       T17 = VADD(T11, T14);
	       T18 = VFNMS(LDK(KP250000000), T17, T16);
	  }
	  ST(&(x[0]), VADD(T8, T9), ms, &(x[0]));
	  ST(&(x[WS(rs, 4)]), VADD(T1q, T1r), ms, &(x[0]));
	  ST(&(x[WS(rs, 2)]), VADD(TM, TN), ms, &(x[0]));
	  ST(&(x[WS(rs, 3)]), VADD(T16, T17), ms, &(x[WS(rs, 1)]));
	  ST(&(x[WS(rs, 1)]), VADD(Ts, Tt), ms, &(x[WS(rs, 1)]));
	  {
	       V Tj, Tk, Ti, T1B, T1C, T1A;
	       Ti = VSUB(Ta, T7);
	       Tj = BYTWJ(&(W[TWVL * 2]), VADD(Th, Ti));
	       Tk = BYTWJ(&(W[TWVL * 4]), VSUB(Ti, Th));
	       ST(&(x[WS(vs, 2)]), Tj, ms, &(x[WS(vs, 2)]));
	       ST(&(x[WS(vs, 3)]), Tk, ms, &(x[WS(vs, 3)]));
	       T1A = VSUB(T1s, T1p);
	       T1B = BYTWJ(&(W[TWVL * 2]), VADD(T1z, T1A));
	       T1C = BYTWJ(&(W[TWVL * 4]), VSUB(T1A, T1z));
	       ST(&(x[WS(vs, 2) + WS(rs, 4)]), T1B, ms, &(x[WS(vs, 2)]));
	       ST(&(x[WS(vs, 3) + WS(rs, 4)]), T1C, ms, &(x[WS(vs, 3)]));
	  }
	  {
	       V T1h, T1i, T1g, TD, TE, TC;
	       T1g = VSUB(T18, T15);
	       T1h = BYTWJ(&(W[TWVL * 2]), VADD(T1f, T1g));
	       T1i = BYTWJ(&(W[TWVL * 4]), VSUB(T1g, T1f));
	       ST(&(x[WS(vs, 2) + WS(rs, 3)]), T1h, ms, &(x[WS(vs, 2) + WS(rs, 1)]));
	       ST(&(x[WS(vs, 3) + WS(rs, 3)]), T1i, ms, &(x[WS(vs, 3) + WS(rs, 1)]));
	       TC = VSUB(Tu, Tr);
	       TD = BYTWJ(&(W[TWVL * 2]), VADD(TB, TC));
	       TE = BYTWJ(&(W[TWVL * 4]), VSUB(TC, TB));
	       ST(&(x[WS(vs, 2) + WS(rs, 1)]), TD, ms, &(x[WS(vs, 2) + WS(rs, 1)]));
	       ST(&(x[WS(vs, 3) + WS(rs, 1)]), TE, ms, &(x[WS(vs, 3) + WS(rs, 1)]));
	  }
	  {
	       V TX, TY, TW, TT, TU, TP;
	       TW = VSUB(TO, TL);
	       TX = BYTWJ(&(W[TWVL * 2]), VADD(TV, TW));
	       TY = BYTWJ(&(W[TWVL * 4]), VSUB(TW, TV));
	       ST(&(x[WS(vs, 2) + WS(rs, 2)]), TX, ms, &(x[WS(vs, 2)]));
	       ST(&(x[WS(vs, 3) + WS(rs, 2)]), TY, ms, &(x[WS(vs, 3)]));
	       TP = VADD(TL, TO);
	       TT = BYTWJ(&(W[0]), VSUB(TP, TS));
	       TU = BYTWJ(&(W[TWVL * 6]), VADD(TS, TP));
	       ST(&(x[WS(vs, 1) + WS(rs, 2)]), TT, ms, &(x[WS(vs, 1)]));
	       ST(&(x[WS(vs, 4) + WS(rs, 2)]), TU, ms, &(x[WS(vs, 4)]));
	  }
	  {
	       V Tf, Tg, Tb, Tz, TA, Tv;
	       Tb = VADD(T7, Ta);
	       Tf = BYTWJ(&(W[0]), VSUB(Tb, Te));
	       Tg = BYTWJ(&(W[TWVL * 6]), VADD(Te, Tb));
	       ST(&(x[WS(vs, 1)]), Tf, ms, &(x[WS(vs, 1)]));
	       ST(&(x[WS(vs, 4)]), Tg, ms, &(x[WS(vs, 4)]));
	       Tv = VADD(Tr, Tu);
	       Tz = BYTWJ(&(W[0]), VSUB(Tv, Ty));
	       TA = BYTWJ(&(W[TWVL * 6]), VADD(Ty, Tv));
	       ST(&(x[WS(vs, 1) + WS(rs, 1)]), Tz, ms, &(x[WS(vs, 1) + WS(rs, 1)]));
	       ST(&(x[WS(vs, 4) + WS(rs, 1)]), TA, ms, &(x[WS(vs, 4) + WS(rs, 1)]));
	  }
	  {
	       V T1d, T1e, T19, T1x, T1y, T1t;
	       T19 = VADD(T15, T18);
	       T1d = BYTWJ(&(W[0]), VSUB(T19, T1c));
	       T1e = BYTWJ(&(W[TWVL * 6]), VADD(T1c, T19));
	       ST(&(x[WS(vs, 1) + WS(rs, 3)]), T1d, ms, &(x[WS(vs, 1) + WS(rs, 1)]));
	       ST(&(x[WS(vs, 4) + WS(rs, 3)]), T1e, ms, &(x[WS(vs, 4) + WS(rs, 1)]));
	       T1t = VADD(T1p, T1s);
	       T1x = BYTWJ(&(W[0]), VSUB(T1t, T1w));
	       T1y = BYTWJ(&(W[TWVL * 6]), VADD(T1w, T1t));
	       ST(&(x[WS(vs, 1) + WS(rs, 4)]), T1x, ms, &(x[WS(vs, 1)]));
	       ST(&(x[WS(vs, 4) + WS(rs, 4)]), T1y, ms, &(x[WS(vs, 4)]));
	  }
     }
}

static const tw_instr twinstr[] = {
     VTW(0, 1),
     VTW(0, 2),
     VTW(0, 3),
     VTW(0, 4),
     {TW_NEXT, VL, 0}
};

static const ct_desc desc = { 5, "q1fv_5", twinstr, &GENUS, {85, 55, 15, 0}, 0, 0, 0 };

void X(codelet_q1fv_5) (planner *p) {
     X(kdft_difsq_register) (p, q1fv_5, &desc);
}
#endif				/* HAVE_FMA */