hc2cbdftv_16.c

快速fft变换

static const hc2c_desc desc = { 16, "hc2cbdftv_16", twinstr, &GENUS, {53, 30, 50, 0} };

void X(codelet_hc2cbdftv_16) (planner *p) {
     X(khc2c_register) (p, hc2cbdftv_16, &desc, HC2C_VIA_DFT);
}
#else				/* HAVE_FMA */

/* Generated by: ../../../genfft/gen_hc2cdft_c -simd -compact -variables 4 -pipeline-latency 8 -trivial-stores -variables 32 -no-generate-bytw -n 16 -dif -sign 1 -name hc2cbdftv_16 -include hc2cbv.h */

/*
 * This function contains 103 FP additions, 42 FP multiplications,
 * (or, 99 additions, 38 multiplications, 4 fused multiply/add),
 * 83 stack variables, 3 constants, and 32 memory accesses
 */
#include "hc2cbv.h"

static void hc2cbdftv_16(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms)
{
     DVK(KP382683432, +0.382683432365089771728459984030398866761344562);
     DVK(KP923879532, +0.923879532511286756128183189396788286822416626);
     DVK(KP707106781, +0.707106781186547524400844362104849039284835938);
     INT m;
     for (m = mb, W = W + ((mb - 1) * ((TWVL / VL) * 30)); m < me; m = m + VL, Rp = Rp + (VL * ms), Ip = Ip + (VL * ms), Rm = Rm - (VL * ms), Im = Im - (VL * ms), W = W + (TWVL * 30), MAKE_VOLATILE_STRIDE(rs)) {
	  V Tf, T16, TZ, T1C, TI, T1a, TV, T1D, T1F, T1G, Ty, T19, TC, T17, TS;
	  V T10;
	  {
	       V T2, TD, T4, TF, Tc, Tb, Td, T6, T8, T9, T3, TE, Ta, T7, T5;
	       V Te, TX, TY, TG, TH, TT, TU, Tj, TM, Tw, TQ, Tn, TN, Ts, TP;
	       V Tg, Ti, Th, Tt, Tv, Tu, Tk, Tm, Tl, Tr, Tq, Tp, To, Tx, TA;
	       V TB, TO, TR;
	       T2 = LD(&(Rp[0]), ms, &(Rp[0]));
	       TD = LD(&(Rp[WS(rs, 4)]), ms, &(Rp[0]));
	       T3 = LD(&(Rm[WS(rs, 7)]), -ms, &(Rm[WS(rs, 1)]));
	       T4 = VCONJ(T3);
	       TE = LD(&(Rm[WS(rs, 3)]), -ms, &(Rm[WS(rs, 1)]));
	       TF = VCONJ(TE);
	       Tc = LD(&(Rp[WS(rs, 6)]), ms, &(Rp[0]));
	       Ta = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)]));
	       Tb = VCONJ(Ta);
	       Td = VSUB(Tb, Tc);
	       T6 = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0]));
	       T7 = LD(&(Rm[WS(rs, 5)]), -ms, &(Rm[WS(rs, 1)]));
	       T8 = VCONJ(T7);
	       T9 = VSUB(T6, T8);
	       T5 = VSUB(T2, T4);
	       Te = VMUL(LDK(KP707106781), VADD(T9, Td));
	       Tf = VADD(T5, Te);
	       T16 = VSUB(T5, Te);
	       TX = VADD(T2, T4);
	       TY = VADD(TD, TF);
	       TZ = VSUB(TX, TY);
	       T1C = VADD(TX, TY);
	       TG = VSUB(TD, TF);
	       TH = VMUL(LDK(KP707106781), VSUB(T9, Td));
	       TI = VADD(TG, TH);
	       T1a = VSUB(TH, TG);
	       TT = VADD(T6, T8);
	       TU = VADD(Tb, Tc);
	       TV = VSUB(TT, TU);
	       T1D = VADD(TT, TU);
	       Tg = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)]));
	       Th = LD(&(Rm[WS(rs, 6)]), -ms, &(Rm[0]));
	       Ti = VCONJ(Th);
	       Tj = VSUB(Tg, Ti);
	       TM = VADD(Tg, Ti);
	       Tt = LD(&(Rp[WS(rs, 3)]), ms, &(Rp[WS(rs, 1)]));
	       Tu = LD(&(Rm[WS(rs, 4)]), -ms, &(Rm[0]));
	       Tv = VCONJ(Tu);
	       Tw = VSUB(Tt, Tv);
	       TQ = VADD(Tt, Tv);
	       Tk = LD(&(Rp[WS(rs, 5)]), ms, &(Rp[WS(rs, 1)]));
	       Tl = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0]));
	       Tm = VCONJ(Tl);
	       Tn = VSUB(Tk, Tm);
	       TN = VADD(Tk, Tm);
	       Tr = LD(&(Rp[WS(rs, 7)]), ms, &(Rp[WS(rs, 1)]));
	       Tp = LD(&(Rm[0]), -ms, &(Rm[0]));
	       Tq = VCONJ(Tp);
	       Ts = VSUB(Tq, Tr);
	       TP = VADD(Tq, Tr);
	       T1F = VADD(TM, TN);
	       T1G = VADD(TP, TQ);
	       To = VFNMS(LDK(KP382683432), Tn, VMUL(LDK(KP923879532), Tj));
	       Tx = VFMA(LDK(KP923879532), Ts, VMUL(LDK(KP382683432), Tw));
	       Ty = VADD(To, Tx);
	       T19 = VSUB(To, Tx);
	       TA = VFMA(LDK(KP382683432), Tj, VMUL(LDK(KP923879532), Tn));
	       TB = VFNMS(LDK(KP382683432), Ts, VMUL(LDK(KP923879532), Tw));
	       TC = VADD(TA, TB);
	       T17 = VSUB(TA, TB);
	       TO = VSUB(TM, TN);
	       TR = VSUB(TP, TQ);
	       TS = VMUL(LDK(KP707106781), VSUB(TO, TR));
	       T10 = VMUL(LDK(KP707106781), VADD(TO, TR));
	  }
	  {
	       V T21, T1W, T1u, T20, T1I, T1O, TK, T1S, T12, T1e, T1k, T1A, T1o, T1w, T1c;
	       V T1M, T1U, T1V, T1T, T1s, T1t, T1r, T1Z, T1E, T1H, T1B, T1N, Tz, TJ, T1;
	       V T1R, TW, T11, TL, T1d, T1i, T1j, T1h, T1z, T1m, T1n, T1l, T1v, T18, T1b;
	       V T15, T1L, T13, T1g, T1X, T23, T14, T1f, T1Y, T22, T1p, T1y, T1J, T1Q, T1q;
	       V T1x, T1K, T1P;
	       T1U = VADD(T1C, T1D);
	       T1V = VADD(T1F, T1G);
	       T21 = VADD(T1U, T1V);
	       T1T = LDW(&(W[TWVL * 14]));
	       T1W = VZMUL(T1T, VSUB(T1U, T1V));
	       T1s = VADD(Tf, Ty);
	       T1t = VBYI(VADD(TI, TC));
	       T1r = LDW(&(W[TWVL * 28]));
	       T1u = VZMULI(T1r, VSUB(T1s, T1t));
	       T1Z = LDW(&(W[0]));
	       T20 = VZMULI(T1Z, VADD(T1s, T1t));
	       T1E = VSUB(T1C, T1D);
	       T1H = VBYI(VSUB(T1F, T1G));
	       T1B = LDW(&(W[TWVL * 22]));
	       T1I = VZMUL(T1B, VSUB(T1E, T1H));
	       T1N = LDW(&(W[TWVL * 6]));
	       T1O = VZMUL(T1N, VADD(T1E, T1H));
	       Tz = VSUB(Tf, Ty);
	       TJ = VBYI(VSUB(TC, TI));
	       T1 = LDW(&(W[TWVL * 12]));
	       TK = VZMULI(T1, VADD(Tz, TJ));
	       T1R = LDW(&(W[TWVL * 16]));
	       T1S = VZMULI(T1R, VSUB(Tz, TJ));
	       TW = VBYI(VSUB(TS, TV));
	       T11 = VSUB(TZ, T10);
	       TL = LDW(&(W[TWVL * 10]));
	       T12 = VZMUL(TL, VADD(TW, T11));
	       T1d = LDW(&(W[TWVL * 18]));
	       T1e = VZMUL(T1d, VSUB(T11, TW));
	       T1i = VBYI(VADD(T1a, T19));
	       T1j = VADD(T16, T17);
	       T1h = LDW(&(W[TWVL * 4]));
	       T1k = VZMULI(T1h, VADD(T1i, T1j));
	       T1z = LDW(&(W[TWVL * 24]));
	       T1A = VZMULI(T1z, VSUB(T1j, T1i));
	       T1m = VBYI(VADD(TV, TS));
	       T1n = VADD(TZ, T10);
	       T1l = LDW(&(W[TWVL * 2]));
	       T1o = VZMUL(T1l, VADD(T1m, T1n));
	       T1v = LDW(&(W[TWVL * 26]));
	       T1w = VZMUL(T1v, VSUB(T1n, T1m));
	       T18 = VSUB(T16, T17);
	       T1b = VBYI(VSUB(T19, T1a));
	       T15 = LDW(&(W[TWVL * 20]));
	       T1c = VZMULI(T15, VSUB(T18, T1b));
	       T1L = LDW(&(W[TWVL * 8]));
	       T1M = VZMULI(T1L, VADD(T1b, T18));
	       T13 = VADD(TK, T12);
	       ST(&(Rp[WS(rs, 3)]), T13, ms, &(Rp[WS(rs, 1)]));
	       T1g = VCONJ(VSUB(T1e, T1c));
	       ST(&(Rm[WS(rs, 5)]), T1g, -ms, &(Rm[WS(rs, 1)]));
	       T1X = VADD(T1S, T1W);
	       ST(&(Rp[WS(rs, 4)]), T1X, ms, &(Rp[0]));
	       T23 = VCONJ(VSUB(T21, T20));
	       ST(&(Rm[0]), T23, -ms, &(Rm[0]));
	       T14 = VCONJ(VSUB(T12, TK));
	       ST(&(Rm[WS(rs, 3)]), T14, -ms, &(Rm[WS(rs, 1)]));
	       T1f = VADD(T1c, T1e);
	       ST(&(Rp[WS(rs, 5)]), T1f, ms, &(Rp[WS(rs, 1)]));
	       T1Y = VCONJ(VSUB(T1W, T1S));
	       ST(&(Rm[WS(rs, 4)]), T1Y, -ms, &(Rm[0]));
	       T22 = VADD(T20, T21);
	       ST(&(Rp[0]), T22, ms, &(Rp[0]));
	       T1p = VADD(T1k, T1o);
	       ST(&(Rp[WS(rs, 1)]), T1p, ms, &(Rp[WS(rs, 1)]));
	       T1y = VCONJ(VSUB(T1w, T1u));
	       ST(&(Rm[WS(rs, 7)]), T1y, -ms, &(Rm[WS(rs, 1)]));
	       T1J = VADD(T1A, T1I);
	       ST(&(Rp[WS(rs, 6)]), T1J, ms, &(Rp[0]));
	       T1Q = VCONJ(VSUB(T1O, T1M));
	       ST(&(Rm[WS(rs, 2)]), T1Q, -ms, &(Rm[0]));
	       T1q = VCONJ(VSUB(T1o, T1k));
	       ST(&(Rm[WS(rs, 1)]), T1q, -ms, &(Rm[WS(rs, 1)]));
	       T1x = VADD(T1u, T1w);
	       ST(&(Rp[WS(rs, 7)]), T1x, ms, &(Rp[WS(rs, 1)]));
	       T1K = VCONJ(VSUB(T1I, T1A));
	       ST(&(Rm[WS(rs, 6)]), T1K, -ms, &(Rm[0]));
	       T1P = VADD(T1M, T1O);
	       ST(&(Rp[WS(rs, 2)]), T1P, ms, &(Rp[0]));
	  }
     }
}

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

static const hc2c_desc desc = { 16, "hc2cbdftv_16", twinstr, &GENUS, {99, 38, 4, 0} };

void X(codelet_hc2cbdftv_16) (planner *p) {
     X(khc2c_register) (p, hc2cbdftv_16, &desc, HC2C_VIA_DFT);
}
#endif				/* HAVE_FMA */