q1bv_8.c

快速fft变换

			      T2F = BYTW(&(W[TWVL * 6]), VSUB(T2D, T2E));
			      {
				   V T14, Tb, Tg, T28, T3U, T3Z;
				   T28 = BYTW(&(W[TWVL * 6]), VSUB(T26, T27));
				   ST(&(x[WS(vs, 4) + WS(rs, 5)]), T3c, ms, &(x[WS(vs, 4) + WS(rs, 1)]));
				   ST(&(x[WS(vs, 4) + WS(rs, 7)]), T4g, ms, &(x[WS(vs, 4) + WS(rs, 1)]));
				   T14 = BYTW(&(W[TWVL * 6]), VSUB(T12, T13));
				   Tj = VFMA(LDK(KP707106781), Ta, T3);
				   Tb = VFNMS(LDK(KP707106781), Ta, T3);
				   ST(&(x[WS(vs, 4) + WS(rs, 6)]), T3J, ms, &(x[WS(vs, 4)]));
				   Tk = VFMA(LDK(KP707106781), Tf, Te);
				   Tg = VFNMS(LDK(KP707106781), Tf, Te);
				   ST(&(x[WS(vs, 4) + WS(rs, 4)]), T2F, ms, &(x[WS(vs, 4)]));
				   ST(&(x[WS(vs, 4) + WS(rs, 3)]), T28, ms, &(x[WS(vs, 4) + WS(rs, 1)]));
				   T3U = VFNMS(LDK(KP707106781), T3T, T3M);
				   T42 = VFMA(LDK(KP707106781), T3T, T3M);
				   T43 = VFMA(LDK(KP707106781), T3Y, T3X);
				   T3Z = VFNMS(LDK(KP707106781), T3Y, T3X);
				   ST(&(x[WS(vs, 4) + WS(rs, 1)]), T14, ms, &(x[WS(vs, 4) + WS(rs, 1)]));
				   T2r = VFMA(LDK(KP707106781), T2i, T2b);
				   T2j = VFNMS(LDK(KP707106781), T2i, T2b);
				   T2o = VFNMS(LDK(KP707106781), T2n, T2m);
				   T2s = VFMA(LDK(KP707106781), T2n, T2m);
				   Ti = BYTW(&(W[TWVL * 8]), VFMAI(Tg, Tb));
				   Th = BYTW(&(W[TWVL * 4]), VFNMSI(Tg, Tb));
				   T1U = VFMA(LDK(KP707106781), T1L, T1E);
				   T1M = VFNMS(LDK(KP707106781), T1L, T1E);
				   T1R = VFNMS(LDK(KP707106781), T1Q, T1P);
				   T1V = VFMA(LDK(KP707106781), T1Q, T1P);
				   T41 = BYTW(&(W[TWVL * 8]), VFMAI(T3Z, T3U));
				   T40 = BYTW(&(W[TWVL * 4]), VFNMSI(T3Z, T3U));
			      }
			 }
			 {
			      V TQ, TR, T1n, T1o, T3v, T3w;
			      {
				   V TI, TN, T1f, T1k, T3n, T3s;
				   {
					V T1T, T1S, T2q, T2p;
					TQ = VFMA(LDK(KP707106781), TH, TA);
					TI = VFNMS(LDK(KP707106781), TH, TA);
					T2q = BYTW(&(W[TWVL * 8]), VFMAI(T2o, T2j));
					T2p = BYTW(&(W[TWVL * 4]), VFNMSI(T2o, T2j));
					ST(&(x[WS(vs, 5)]), Ti, ms, &(x[WS(vs, 5)]));
					ST(&(x[WS(vs, 3)]), Th, ms, &(x[WS(vs, 3)]));
					T1T = BYTW(&(W[TWVL * 8]), VFMAI(T1R, T1M));
					T1S = BYTW(&(W[TWVL * 4]), VFNMSI(T1R, T1M));
					ST(&(x[WS(vs, 5) + WS(rs, 7)]), T41, ms, &(x[WS(vs, 5) + WS(rs, 1)]));
					ST(&(x[WS(vs, 3) + WS(rs, 7)]), T40, ms, &(x[WS(vs, 3) + WS(rs, 1)]));
					ST(&(x[WS(vs, 5) + WS(rs, 4)]), T2q, ms, &(x[WS(vs, 5)]));
					ST(&(x[WS(vs, 3) + WS(rs, 4)]), T2p, ms, &(x[WS(vs, 3)]));
					TN = VFNMS(LDK(KP707106781), TM, TL);
					TR = VFMA(LDK(KP707106781), TM, TL);
					T1n = VFMA(LDK(KP707106781), T1e, T17);
					T1f = VFNMS(LDK(KP707106781), T1e, T17);
					ST(&(x[WS(vs, 5) + WS(rs, 3)]), T1T, ms, &(x[WS(vs, 5) + WS(rs, 1)]));
					ST(&(x[WS(vs, 3) + WS(rs, 3)]), T1S, ms, &(x[WS(vs, 3) + WS(rs, 1)]));
					T1k = VFNMS(LDK(KP707106781), T1j, T1i);
					T1o = VFMA(LDK(KP707106781), T1j, T1i);
					T3v = VFMA(LDK(KP707106781), T3m, T3f);
					T3n = VFNMS(LDK(KP707106781), T3m, T3f);
					T3s = VFNMS(LDK(KP707106781), T3r, T3q);
					T3w = VFMA(LDK(KP707106781), T3r, T3q);
				   }
				   {
					V T2Q, TP, TO, T2V, T2X, T2W;
					T2Y = VFMA(LDK(KP707106781), T2P, T2I);
					T2Q = VFNMS(LDK(KP707106781), T2P, T2I);
					TP = BYTW(&(W[TWVL * 8]), VFMAI(TN, TI));
					TO = BYTW(&(W[TWVL * 4]), VFNMSI(TN, TI));
					T2V = VFNMS(LDK(KP707106781), T2U, T2T);
					T2Z = VFMA(LDK(KP707106781), T2U, T2T);
					{
					     V T1m, T1l, T3u, T3t;
					     T1m = BYTW(&(W[TWVL * 8]), VFMAI(T1k, T1f));
					     T1l = BYTW(&(W[TWVL * 4]), VFNMSI(T1k, T1f));
					     T3u = BYTW(&(W[TWVL * 8]), VFMAI(T3s, T3n));
					     T3t = BYTW(&(W[TWVL * 4]), VFNMSI(T3s, T3n));
					     ST(&(x[WS(vs, 5) + WS(rs, 1)]), TP, ms, &(x[WS(vs, 5) + WS(rs, 1)]));
					     ST(&(x[WS(vs, 3) + WS(rs, 1)]), TO, ms, &(x[WS(vs, 3) + WS(rs, 1)]));
					     T2X = BYTW(&(W[TWVL * 8]), VFMAI(T2V, T2Q));
					     T2W = BYTW(&(W[TWVL * 4]), VFNMSI(T2V, T2Q));
					     ST(&(x[WS(vs, 5) + WS(rs, 2)]), T1m, ms, &(x[WS(vs, 5)]));
					     ST(&(x[WS(vs, 3) + WS(rs, 2)]), T1l, ms, &(x[WS(vs, 3)]));
					     ST(&(x[WS(vs, 5) + WS(rs, 6)]), T3u, ms, &(x[WS(vs, 5)]));
					     ST(&(x[WS(vs, 3) + WS(rs, 6)]), T3t, ms, &(x[WS(vs, 3)]));
					}
					ST(&(x[WS(vs, 5) + WS(rs, 5)]), T2X, ms, &(x[WS(vs, 5) + WS(rs, 1)]));
					ST(&(x[WS(vs, 3) + WS(rs, 5)]), T2W, ms, &(x[WS(vs, 3) + WS(rs, 1)]));
				   }
			      }
			      {
				   V T3y, T3x, T1q, T1p;
				   T1q = BYTW(&(W[TWVL * 12]), VFNMSI(T1o, T1n));
				   T1p = BYTW(&(W[0]), VFMAI(T1o, T1n));
				   {
					V Tm, Tl, T2u, T2t;
					Tm = BYTW(&(W[TWVL * 12]), VFNMSI(Tk, Tj));
					Tl = BYTW(&(W[0]), VFMAI(Tk, Tj));
					T2u = BYTW(&(W[TWVL * 12]), VFNMSI(T2s, T2r));
					T2t = BYTW(&(W[0]), VFMAI(T2s, T2r));
					ST(&(x[WS(vs, 7) + WS(rs, 2)]), T1q, ms, &(x[WS(vs, 7)]));
					ST(&(x[WS(vs, 1) + WS(rs, 2)]), T1p, ms, &(x[WS(vs, 1)]));
					T3y = BYTW(&(W[TWVL * 12]), VFNMSI(T3w, T3v));
					T3x = BYTW(&(W[0]), VFMAI(T3w, T3v));
					ST(&(x[WS(vs, 7)]), Tm, ms, &(x[WS(vs, 7)]));
					ST(&(x[WS(vs, 1)]), Tl, ms, &(x[WS(vs, 1)]));
					ST(&(x[WS(vs, 7) + WS(rs, 4)]), T2u, ms, &(x[WS(vs, 7)]));
					ST(&(x[WS(vs, 1) + WS(rs, 4)]), T2t, ms, &(x[WS(vs, 1)]));
				   }
				   ST(&(x[WS(vs, 7) + WS(rs, 6)]), T3y, ms, &(x[WS(vs, 7)]));
				   ST(&(x[WS(vs, 1) + WS(rs, 6)]), T3x, ms, &(x[WS(vs, 1)]));
				   TT = BYTW(&(W[TWVL * 12]), VFNMSI(TR, TQ));
				   TS = BYTW(&(W[0]), VFMAI(TR, TQ));
			      }
			 }
		    }
	       }
	  }
	  {
	       V T1X, T1W, T31, T30;
	       T1X = BYTW(&(W[TWVL * 12]), VFNMSI(T1V, T1U));
	       T1W = BYTW(&(W[0]), VFMAI(T1V, T1U));
	       T31 = BYTW(&(W[TWVL * 12]), VFNMSI(T2Z, T2Y));
	       T30 = BYTW(&(W[0]), VFMAI(T2Z, T2Y));
	       ST(&(x[WS(vs, 7) + WS(rs, 1)]), TT, ms, &(x[WS(vs, 7) + WS(rs, 1)]));
	       ST(&(x[WS(vs, 1) + WS(rs, 1)]), TS, ms, &(x[WS(vs, 1) + WS(rs, 1)]));
	       T45 = BYTW(&(W[TWVL * 12]), VFNMSI(T43, T42));
	       T44 = BYTW(&(W[0]), VFMAI(T43, T42));
	       ST(&(x[WS(vs, 7) + WS(rs, 3)]), T1X, ms, &(x[WS(vs, 7) + WS(rs, 1)]));
	       ST(&(x[WS(vs, 1) + WS(rs, 3)]), T1W, ms, &(x[WS(vs, 1) + WS(rs, 1)]));
	       ST(&(x[WS(vs, 7) + WS(rs, 5)]), T31, ms, &(x[WS(vs, 7) + WS(rs, 1)]));
	       ST(&(x[WS(vs, 1) + WS(rs, 5)]), T30, ms, &(x[WS(vs, 1) + WS(rs, 1)]));
	  }
	  ST(&(x[WS(vs, 7) + WS(rs, 7)]), T45, ms, &(x[WS(vs, 7) + WS(rs, 1)]));
	  ST(&(x[WS(vs, 1) + WS(rs, 7)]), T44, ms, &(x[WS(vs, 1) + WS(rs, 1)]));
     }
}

static const tw_instr twinstr[] = {
     VTW(0, 1),
     VTW(0, 2),
     VTW(0, 3),
     VTW(0, 4),
     VTW(0, 5),
     VTW(0, 6),
     VTW(0, 7),
     {TW_NEXT, VL, 0}
};

static const ct_desc desc = { 8, "q1bv_8", twinstr, &GENUS, {184, 112, 80, 0}, 0, 0, 0 };

void X(codelet_q1bv_8) (planner *p) {
     X(kdft_difsq_register) (p, q1bv_8, &desc);
}
#else				/* HAVE_FMA */

/* Generated by: ../../../genfft/gen_twidsq_c -simd -compact -variables 4 -pipeline-latency 8 -n 8 -dif -name q1bv_8 -include q1b.h -sign 1 */

/*
 * This function contains 264 FP additions, 128 FP multiplications,
 * (or, 264 additions, 128 multiplications, 0 fused multiply/add),
 * 77 stack variables, 1 constants, and 128 memory accesses
 */
#include "q1b.h"

static void q1bv_8(R *ri, R *ii, const R *W, stride rs, stride vs, INT mb, INT me, INT ms)
{
     DVK(KP707106781, +0.707106781186547524400844362104849039284835938);
     INT m;
     R *x;
     x = ii;
     for (m = mb, W = W + (mb * ((TWVL / VL) * 14)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 14), MAKE_VOLATILE_STRIDE(rs), MAKE_VOLATILE_STRIDE(vs)) {
	  V Ta, Tv, Te, Tp, T1L, T26, T1P, T20, T2i, T2D, T2m, T2x, T3T, T4e, T3X;
	  V T48, TH, T12, TL, TW, T1e, T1z, T1i, T1t, T2P, T3a, T2T, T34, T3m, T3H;
	  V T3q, T3B, T7, Tw, Tf, Ts, T1I, T27, T1Q, T23, T2f, T2E, T2n, T2A, T3Q;
	  V T4f, T3Y, T4b, TE, T13, TM, TZ, T1b, T1A, T1j, T1w, T2M, T3b, T2U, T37;
	  V T3j, T3I, T3r, T3E, T28, T14;
	  {
	       V T8, T9, To, Tc, Td, Tn;
	       T8 = LD(&(x[WS(rs, 2)]), ms, &(x[0]));
	       T9 = LD(&(x[WS(rs, 6)]), ms, &(x[0]));
	       To = VADD(T8, T9);
	       Tc = LD(&(x[0]), ms, &(x[0]));
	       Td = LD(&(x[WS(rs, 4)]), ms, &(x[0]));
	       Tn = VADD(Tc, Td);
	       Ta = VSUB(T8, T9);
	       Tv = VADD(Tn, To);
	       Te = VSUB(Tc, Td);
	       Tp = VSUB(Tn, To);
	  }
	  {
	       V T1J, T1K, T1Z, T1N, T1O, T1Y;
	       T1J = LD(&(x[WS(vs, 3) + WS(rs, 2)]), ms, &(x[WS(vs, 3)]));
	       T1K = LD(&(x[WS(vs, 3) + WS(rs, 6)]), ms, &(x[WS(vs, 3)]));
	       T1Z = VADD(T1J, T1K);
	       T1N = LD(&(x[WS(vs, 3)]), ms, &(x[WS(vs, 3)]));
	       T1O = LD(&(x[WS(vs, 3) + WS(rs, 4)]), ms, &(x[WS(vs, 3)]));
	       T1Y = VADD(T1N, T1O);
	       T1L = VSUB(T1J, T1K);
	       T26 = VADD(T1Y, T1Z);
	       T1P = VSUB(T1N, T1O);
	       T20 = VSUB(T1Y, T1Z);
	  }
	  {
	       V T2g, T2h, T2w, T2k, T2l, T2v;
	       T2g = LD(&(x[WS(vs, 4) + WS(rs, 2)]), ms, &(x[WS(vs, 4)]));
	       T2h = LD(&(x[WS(vs, 4) + WS(rs, 6)]), ms, &(x[WS(vs, 4)]));
	       T2w = VADD(T2g, T2h);
	       T2k = LD(&(x[WS(vs, 4)]), ms, &(x[WS(vs, 4)]));
	       T2l = LD(&(x[WS(vs, 4) + WS(rs, 4)]), ms, &(x[WS(vs, 4)]));
	       T2v = VADD(T2k, T2l);
	       T2i = VSUB(T2g, T2h);
	       T2D = VADD(T2v, T2w);
	       T2m = VSUB(T2k, T2l);
	       T2x = VSUB(T2v, T2w);
	  }
	  {
	       V T3R, T3S, T47, T3V, T3W, T46;
	       T3R = LD(&(x[WS(vs, 7) + WS(rs, 2)]), ms, &(x[WS(vs, 7)]));
	       T3S = LD(&(x[WS(vs, 7) + WS(rs, 6)]), ms, &(x[WS(vs, 7)]));
	       T47 = VADD(T3R, T3S);
	       T3V = LD(&(x[WS(vs, 7)]), ms, &(x[WS(vs, 7)]));
	       T3W = LD(&(x[WS(vs, 7) + WS(rs, 4)]), ms, &(x[WS(vs, 7)]));
	       T46 = VADD(T3V, T3W);
	       T3T = VSUB(T3R, T3S);
	       T4e = VADD(T46, T47);
	       T3X = VSUB(T3V, T3W);
	       T48 = VSUB(T46, T47);
	  }
	  {
	       V TF, TG, TV, TJ, TK, TU;
	       TF = LD(&(x[WS(vs, 1) + WS(rs, 2)]), ms, &(x[WS(vs, 1)]));
	       TG = LD(&(x[WS(vs, 1) + WS(rs, 6)]), ms, &(x[WS(vs, 1)]));
	       TV = VADD(TF, TG);
	       TJ = LD(&(x[WS(vs, 1)]), ms, &(x[WS(vs, 1)]));
	       TK = LD(&(x[WS(vs, 1) + WS(rs, 4)]), ms, &(x[WS(vs, 1)]));
	       TU = VADD(TJ, TK);
	       TH = VSUB(TF, TG);
	       T12 = VADD(TU, TV);
	       TL = VSUB(TJ, TK);
	       TW = VSUB(TU, TV);
	  }
	  {
	       V T1c, T1d, T1s, T1g, T1h, T1r;
	       T1c = LD(&(x[WS(vs, 2) + WS(rs, 2)]), ms, &(x[WS(vs, 2)]));
	       T1d = LD(&(x[WS(vs, 2) + WS(rs, 6)]), ms, &(x[WS(vs, 2)]));
	       T1s = VADD(T1c, T1d);
	       T1g = LD(&(x[WS(vs, 2)]), ms, &(x[WS(vs, 2)]));
	       T1h = LD(&(x[WS(vs, 2) + WS(rs, 4)]), ms, &(x[WS(vs, 2)]));
	       T1r = VADD(T1g, T1h);
	       T1e = VSUB(T1c, T1d);
	       T1z = VADD(T1r, T1s);
	       T1i = VSUB(T1g, T1h);
	       T1t = VSUB(T1r, T1s);
	  }
	  {
	       V T2N, T2O, T33, T2R, T2S, T32;
	       T2N = LD(&(x[WS(vs, 5) + WS(rs, 2)]), ms, &(x[WS(vs, 5)]));
	       T2O = LD(&(x[WS(vs, 5) + WS(rs, 6)]), ms, &(x[WS(vs, 5)]));
	       T33 = VADD(T2N, T2O);
	       T2R = LD(&(x[WS(vs, 5)]), ms, &(x[WS(vs, 5)]));
	       T2S = LD(&(x[WS(vs, 5) + WS(rs, 4)]), ms, &(x[WS(vs, 5)]));
	       T32 = VADD(T2R, T2S);
	       T2P = VSUB(T2N, T2O);
	       T3a = VADD(T32, T33);
	       T2T = VSUB(T2R, T2S);
	       T34 = VSUB(T32, T33);
	  }
	  {
	       V T3k, T3l, T3A, T3o, T3p, T3z;
	       T3k = LD(&(x[WS(vs, 6) + WS(rs, 2)]), ms, &(x[WS(vs, 6)]));
	       T3l = LD(&(x[WS(vs, 6) + WS(rs, 6)]), ms, &(x[WS(vs, 6)]));
	       T3A = VADD(T3k, T3l);
	       T3o = LD(&(x[WS(vs, 6)]), ms, &(x[WS(vs, 6)]));
	       T3p = LD(&(x[WS(vs, 6) + WS(rs, 4)]), ms, &(x[WS(vs, 6)]));
	       T3z = VADD(T3o, T3p);
	       T3m = VSUB(T3k, T3l);
	       T3H = VADD(T3z, T3A);
	       T3q = VSUB(T3o, T3p);
	       T3B = VSUB(T3z, T3A);
	  }
	  {
	       V T3, Tq, T6, Tr;
	       {
		    V T1, T2, T4, T5;
		    T1 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)]));
		    T2 = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)]));
		    T3 = VSUB(T1, T2);
		    Tq = VADD(T1, T2);
		    T4 = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)]));
		    T5 = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)]));
		    T6 = VSUB(T4, T5);
		    Tr = VADD(T4, T5);
	       }
	       T7 = VMUL(LDK(KP707106781), VSUB(T3, T6));
	       Tw = VADD(Tq, Tr);
	       Tf = VMUL(LDK(KP707106781), VADD(T3, T6));
	       Ts = VBYI(VSUB(Tq, Tr));
	  }
	  {
	       V T1E, T21, T1H, T22;
	       {
		    V T1C, T1D, T1F, T1G;
		    T1C = LD(&(x[WS(vs, 3) + WS(rs, 1)]), ms, &(x[WS(vs, 3) + WS(rs, 1)]));
		    T1D = LD(&(x[WS(vs, 3) + WS(rs, 5)]), ms, &(x[WS(vs, 3) + WS(rs, 1)]));
		    T1E = VSUB(T1C, T1D);
		    T21 = VADD(T1C, T1D);
		    T1F = LD(&(x[WS(vs, 3) + WS(rs, 7)]), ms, &(x[WS(vs, 3) + WS(rs, 1)]));
		    T1G = LD(&(x[WS(vs, 3) + WS(rs, 3)]), ms, &(x[WS(vs, 3) + WS(rs, 1)]));
		    T1H = VSUB(T1F, T1G);
		    T22 = VADD(T1F, T1G);
	       }
	       T1I = VMUL(LDK(KP707106781), VSUB(T1E, T1H));
	       T27 = VADD(T21, T22);
	       T1Q = VMUL(LDK(KP707106781), VADD(T1E, T1H));
	       T23 = VBYI(VSUB(T21, T22));
	  }
	  {
	       V T2b, T2y, T2e, T2z;
	       {
		    V T29, T2a, T2c, T2d;
		    T29 = LD(&(x[WS(vs, 4) + WS(rs, 1)]), ms, &(x[WS(vs, 4) + WS(rs, 1)]));
		    T2a = LD(&(x[WS(vs, 4) + WS(rs, 5)]), ms, &(x[WS(vs, 4) + WS(rs, 1)]));
		    T2b = VSUB(T29, T2a);
		    T2y = VADD(T29, T2a);
		    T2c = LD(&(x[WS(vs, 4) + WS(rs, 7)]), ms, &(x[WS(vs, 4) + WS(rs, 1)]));
		    T2d = LD(&(x[WS(vs, 4) + WS(rs, 3)]), ms, &(x[WS(vs, 4) + WS(rs, 1)]));
		    T2e = VSUB(T2c, T2d);
		    T2z = VADD(T2c, T2d);