Chris@10: /*
Chris@10:  * Copyright (c) 2003, 2007-11 Matteo Frigo
Chris@10:  * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology
Chris@10:  *
Chris@10:  * This program is free software; you can redistribute it and/or modify
Chris@10:  * it under the terms of the GNU General Public License as published by
Chris@10:  * the Free Software Foundation; either version 2 of the License, or
Chris@10:  * (at your option) any later version.
Chris@10:  *
Chris@10:  * This program is distributed in the hope that it will be useful,
Chris@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
Chris@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
Chris@10:  * GNU General Public License for more details.
Chris@10:  *
Chris@10:  * You should have received a copy of the GNU General Public License
Chris@10:  * along with this program; if not, write to the Free Software
Chris@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
Chris@10:  *
Chris@10:  */
Chris@10: 
Chris@10: /* This file was automatically generated --- DO NOT EDIT */
Chris@10: /* Generated on Sun Nov 25 07:36:17 EST 2012 */
Chris@10: 
Chris@10: #include "codelet-dft.h"
Chris@10: 
Chris@10: #ifdef HAVE_FMA
Chris@10: 
Chris@10: /* Generated by: ../../../genfft/gen_twidsq.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -reload-twiddle -dif -n 4 -name q1_4 -include q.h */
Chris@10: 
Chris@10: /*
Chris@10:  * This function contains 88 FP additions, 48 FP multiplications,
Chris@10:  * (or, 64 additions, 24 multiplications, 24 fused multiply/add),
Chris@10:  * 76 stack variables, 0 constants, and 64 memory accesses
Chris@10:  */
Chris@10: #include "q.h"
Chris@10: 
Chris@10: static void q1_4(R *rio, R *iio, const R *W, stride rs, stride vs, INT mb, INT me, INT ms)
Chris@10: {
Chris@10:      {
Chris@10: 	  INT m;
Chris@10: 	  for (m = mb, W = W + (mb * 6); m < me; m = m + 1, rio = rio + ms, iio = iio + ms, W = W + 6, MAKE_VOLATILE_STRIDE(8, rs), MAKE_VOLATILE_STRIDE(0, vs)) {
Chris@10: 	       E T1X, T1S, T1L, T1Y, T1R;
Chris@10: 	       {
Chris@10: 		    E T3, Tf, Tv, Ti, Tw, Tx, T6, Tm, Tc, Ts, T1T, T1H, T29, T1W, T2a;
Chris@10: 		    E T2b, T1K, T20, T1Q, T26, TN, TB, T13, TQ, T14, T15, TE, TU, TK, T10;
Chris@10: 		    E T1l, T19, T1a, T1h, T1B, T1o, T1C, T1b, T1D, T1e, T1c;
Chris@10: 		    {
Chris@10: 			 E T1I, T1P, T1J, T1M;
Chris@10: 			 {
Chris@10: 			      E Tb, T4, T5, T8;
Chris@10: 			      {
Chris@10: 				   E T1, T2, T9, Ta, Tg, Th;
Chris@10: 				   T1 = rio[0];
Chris@10: 				   T2 = rio[WS(rs, 2)];
Chris@10: 				   T9 = iio[0];
Chris@10: 				   Ta = iio[WS(rs, 2)];
Chris@10: 				   Tg = iio[WS(rs, 1)];
Chris@10: 				   T3 = T1 + T2;
Chris@10: 				   Tf = T1 - T2;
Chris@10: 				   Th = iio[WS(rs, 3)];
Chris@10: 				   Tv = T9 + Ta;
Chris@10: 				   Tb = T9 - Ta;
Chris@10: 				   T4 = rio[WS(rs, 1)];
Chris@10: 				   Ti = Tg - Th;
Chris@10: 				   Tw = Tg + Th;
Chris@10: 				   T5 = rio[WS(rs, 3)];
Chris@10: 			      }
Chris@10: 			      Tx = Tv - Tw;
Chris@10: 			      T8 = T4 - T5;
Chris@10: 			      T6 = T4 + T5;
Chris@10: 			      {
Chris@10: 				   E T1N, T1O, T1F, T1G, T1U, T1V;
Chris@10: 				   T1F = rio[WS(vs, 3)];
Chris@10: 				   T1G = rio[WS(vs, 3) + WS(rs, 2)];
Chris@10: 				   Tm = Tb - T8;
Chris@10: 				   Tc = T8 + Tb;
Chris@10: 				   Ts = T3 - T6;
Chris@10: 				   T1T = T1F - T1G;
Chris@10: 				   T1H = T1F + T1G;
Chris@10: 				   T1N = iio[WS(vs, 3)];
Chris@10: 				   T1O = iio[WS(vs, 3) + WS(rs, 2)];
Chris@10: 				   T1U = iio[WS(vs, 3) + WS(rs, 1)];
Chris@10: 				   T1V = iio[WS(vs, 3) + WS(rs, 3)];
Chris@10: 				   T1I = rio[WS(vs, 3) + WS(rs, 1)];
Chris@10: 				   T1P = T1N - T1O;
Chris@10: 				   T29 = T1N + T1O;
Chris@10: 				   T1W = T1U - T1V;
Chris@10: 				   T2a = T1U + T1V;
Chris@10: 				   T1J = rio[WS(vs, 3) + WS(rs, 3)];
Chris@10: 			      }
Chris@10: 			 }
Chris@10: 			 T2b = T29 - T2a;
Chris@10: 			 T1M = T1I - T1J;
Chris@10: 			 T1K = T1I + T1J;
Chris@10: 			 {
Chris@10: 			      E TC, TJ, TD, TG;
Chris@10: 			      {
Chris@10: 				   E TH, TI, Tz, TA, TO, TP;
Chris@10: 				   Tz = rio[WS(vs, 1)];
Chris@10: 				   TA = rio[WS(vs, 1) + WS(rs, 2)];
Chris@10: 				   T20 = T1P - T1M;
Chris@10: 				   T1Q = T1M + T1P;
Chris@10: 				   T26 = T1H - T1K;
Chris@10: 				   TN = Tz - TA;
Chris@10: 				   TB = Tz + TA;
Chris@10: 				   TH = iio[WS(vs, 1)];
Chris@10: 				   TI = iio[WS(vs, 1) + WS(rs, 2)];
Chris@10: 				   TO = iio[WS(vs, 1) + WS(rs, 1)];
Chris@10: 				   TP = iio[WS(vs, 1) + WS(rs, 3)];
Chris@10: 				   TC = rio[WS(vs, 1) + WS(rs, 1)];
Chris@10: 				   TJ = TH - TI;
Chris@10: 				   T13 = TH + TI;
Chris@10: 				   TQ = TO - TP;
Chris@10: 				   T14 = TO + TP;
Chris@10: 				   TD = rio[WS(vs, 1) + WS(rs, 3)];
Chris@10: 			      }
Chris@10: 			      T15 = T13 - T14;
Chris@10: 			      TG = TC - TD;
Chris@10: 			      TE = TC + TD;
Chris@10: 			      {
Chris@10: 				   E T1f, T1g, T17, T18, T1m, T1n;
Chris@10: 				   T17 = rio[WS(vs, 2)];
Chris@10: 				   T18 = rio[WS(vs, 2) + WS(rs, 2)];
Chris@10: 				   TU = TJ - TG;
Chris@10: 				   TK = TG + TJ;
Chris@10: 				   T10 = TB - TE;
Chris@10: 				   T1l = T17 - T18;
Chris@10: 				   T19 = T17 + T18;
Chris@10: 				   T1f = iio[WS(vs, 2)];
Chris@10: 				   T1g = iio[WS(vs, 2) + WS(rs, 2)];
Chris@10: 				   T1m = iio[WS(vs, 2) + WS(rs, 1)];
Chris@10: 				   T1n = iio[WS(vs, 2) + WS(rs, 3)];
Chris@10: 				   T1a = rio[WS(vs, 2) + WS(rs, 1)];
Chris@10: 				   T1h = T1f - T1g;
Chris@10: 				   T1B = T1f + T1g;
Chris@10: 				   T1o = T1m - T1n;
Chris@10: 				   T1C = T1m + T1n;
Chris@10: 				   T1b = rio[WS(vs, 2) + WS(rs, 3)];
Chris@10: 			      }
Chris@10: 			 }
Chris@10: 		    }
Chris@10: 		    T1D = T1B - T1C;
Chris@10: 		    T1e = T1a - T1b;
Chris@10: 		    T1c = T1a + T1b;
Chris@10: 		    {
Chris@10: 			 E T1s, T1i, T1y, T28, T27, Tr, Tu;
Chris@10: 			 rio[0] = T3 + T6;
Chris@10: 			 iio[0] = Tv + Tw;
Chris@10: 			 T1s = T1h - T1e;
Chris@10: 			 T1i = T1e + T1h;
Chris@10: 			 T1y = T19 - T1c;
Chris@10: 			 rio[WS(rs, 1)] = TB + TE;
Chris@10: 			 iio[WS(rs, 1)] = T13 + T14;
Chris@10: 			 rio[WS(rs, 2)] = T19 + T1c;
Chris@10: 			 iio[WS(rs, 2)] = T1B + T1C;
Chris@10: 			 iio[WS(rs, 3)] = T29 + T2a;
Chris@10: 			 rio[WS(rs, 3)] = T1H + T1K;
Chris@10: 			 Tr = W[2];
Chris@10: 			 Tu = W[3];
Chris@10: 			 {
Chris@10: 			      E T25, Ty, Tt, T2c;
Chris@10: 			      T25 = W[2];
Chris@10: 			      T28 = W[3];
Chris@10: 			      Ty = Tr * Tx;
Chris@10: 			      Tt = Tr * Ts;
Chris@10: 			      T2c = T25 * T2b;
Chris@10: 			      T27 = T25 * T26;
Chris@10: 			      iio[WS(vs, 2)] = FNMS(Tu, Ts, Ty);
Chris@10: 			      rio[WS(vs, 2)] = FMA(Tu, Tx, Tt);
Chris@10: 			      iio[WS(vs, 2) + WS(rs, 3)] = FNMS(T28, T26, T2c);
Chris@10: 			 }
Chris@10: 			 rio[WS(vs, 2) + WS(rs, 3)] = FMA(T28, T2b, T27);
Chris@10: 			 {
Chris@10: 			      E Tp, T1v, T23, T22, T1Z, TR, TM, TF;
Chris@10: 			      {
Chris@10: 				   E T1A, T1z, TZ, T12;
Chris@10: 				   TZ = W[2];
Chris@10: 				   T12 = W[3];
Chris@10: 				   {
Chris@10: 					E T1x, T16, T11, T1E;
Chris@10: 					T1x = W[2];
Chris@10: 					T1A = W[3];
Chris@10: 					T16 = TZ * T15;
Chris@10: 					T11 = TZ * T10;
Chris@10: 					T1E = T1x * T1D;
Chris@10: 					T1z = T1x * T1y;
Chris@10: 					iio[WS(vs, 2) + WS(rs, 1)] = FNMS(T12, T10, T16);
Chris@10: 					rio[WS(vs, 2) + WS(rs, 1)] = FMA(T12, T15, T11);
Chris@10: 					iio[WS(vs, 2) + WS(rs, 2)] = FNMS(T1A, T1y, T1E);
Chris@10: 				   }
Chris@10: 				   rio[WS(vs, 2) + WS(rs, 2)] = FMA(T1A, T1D, T1z);
Chris@10: 				   {
Chris@10: 					E Tj, Te, T7, T1p, T1k, T1j;
Chris@10: 					Tp = Tf + Ti;
Chris@10: 					Tj = Tf - Ti;
Chris@10: 					Te = W[5];
Chris@10: 					T7 = W[4];
Chris@10: 					{
Chris@10: 					     E T1d, T1q, Tk, Td;
Chris@10: 					     T1p = T1l - T1o;
Chris@10: 					     T1v = T1l + T1o;
Chris@10: 					     T1k = W[5];
Chris@10: 					     Tk = Te * Tc;
Chris@10: 					     Td = T7 * Tc;
Chris@10: 					     T1d = W[4];
Chris@10: 					     T1q = T1k * T1i;
Chris@10: 					     rio[WS(vs, 3)] = FMA(T7, Tj, Tk);
Chris@10: 					     iio[WS(vs, 3)] = FNMS(Te, Tj, Td);
Chris@10: 					     T1j = T1d * T1i;
Chris@10: 					     rio[WS(vs, 3) + WS(rs, 2)] = FMA(T1d, T1p, T1q);
Chris@10: 					}
Chris@10: 					T23 = T1T + T1W;
Chris@10: 					T1X = T1T - T1W;
Chris@10: 					T22 = W[1];
Chris@10: 					iio[WS(vs, 3) + WS(rs, 2)] = FNMS(T1k, T1p, T1j);
Chris@10: 					T1Z = W[0];
Chris@10: 				   }
Chris@10: 			      }
Chris@10: 			      {
Chris@10: 				   E TX, TW, TT, TY, TV, T24, T21;
Chris@10: 				   TX = TN + TQ;
Chris@10: 				   TR = TN - TQ;
Chris@10: 				   T24 = T22 * T20;
Chris@10: 				   TW = W[1];
Chris@10: 				   T21 = T1Z * T20;
Chris@10: 				   TT = W[0];
Chris@10: 				   rio[WS(vs, 1) + WS(rs, 3)] = FMA(T1Z, T23, T24);
Chris@10: 				   TY = TW * TU;
Chris@10: 				   iio[WS(vs, 1) + WS(rs, 3)] = FNMS(T22, T23, T21);
Chris@10: 				   TV = TT * TU;
Chris@10: 				   rio[WS(vs, 1) + WS(rs, 1)] = FMA(TT, TX, TY);
Chris@10: 				   TM = W[5];
Chris@10: 				   iio[WS(vs, 1) + WS(rs, 1)] = FNMS(TW, TX, TV);
Chris@10: 				   TF = W[4];
Chris@10: 			      }
Chris@10: 			      {
Chris@10: 				   E To, Tl, Tq, Tn, TS, TL;
Chris@10: 				   TS = TM * TK;
Chris@10: 				   To = W[1];
Chris@10: 				   TL = TF * TK;
Chris@10: 				   Tl = W[0];
Chris@10: 				   rio[WS(vs, 3) + WS(rs, 1)] = FMA(TF, TR, TS);
Chris@10: 				   Tq = To * Tm;
Chris@10: 				   iio[WS(vs, 3) + WS(rs, 1)] = FNMS(TM, TR, TL);
Chris@10: 				   Tn = Tl * Tm;
Chris@10: 				   {
Chris@10: 					E T1u, T1r, T1w, T1t;
Chris@10: 					rio[WS(vs, 1)] = FMA(Tl, Tp, Tq);
Chris@10: 					T1u = W[1];
Chris@10: 					iio[WS(vs, 1)] = FNMS(To, Tp, Tn);
Chris@10: 					T1r = W[0];
Chris@10: 					T1w = T1u * T1s;
Chris@10: 					T1S = W[5];
Chris@10: 					T1t = T1r * T1s;
Chris@10: 					T1L = W[4];
Chris@10: 					rio[WS(vs, 1) + WS(rs, 2)] = FMA(T1r, T1v, T1w);
Chris@10: 					T1Y = T1S * T1Q;
Chris@10: 					iio[WS(vs, 1) + WS(rs, 2)] = FNMS(T1u, T1v, T1t);
Chris@10: 					T1R = T1L * T1Q;
Chris@10: 				   }
Chris@10: 			      }
Chris@10: 			 }
Chris@10: 		    }
Chris@10: 	       }
Chris@10: 	       rio[WS(vs, 3) + WS(rs, 3)] = FMA(T1L, T1X, T1Y);
Chris@10: 	       iio[WS(vs, 3) + WS(rs, 3)] = FNMS(T1S, T1X, T1R);
Chris@10: 	  }
Chris@10:      }
Chris@10: }
Chris@10: 
Chris@10: static const tw_instr twinstr[] = {
Chris@10:      {TW_FULL, 0, 4},
Chris@10:      {TW_NEXT, 1, 0}
Chris@10: };
Chris@10: 
Chris@10: static const ct_desc desc = { 4, "q1_4", twinstr, &GENUS, {64, 24, 24, 0}, 0, 0, 0 };
Chris@10: 
Chris@10: void X(codelet_q1_4) (planner *p) {
Chris@10:      X(kdft_difsq_register) (p, q1_4, &desc);
Chris@10: }
Chris@10: #else				/* HAVE_FMA */
Chris@10: 
Chris@10: /* Generated by: ../../../genfft/gen_twidsq.native -compact -variables 4 -pipeline-latency 4 -reload-twiddle -dif -n 4 -name q1_4 -include q.h */
Chris@10: 
Chris@10: /*
Chris@10:  * This function contains 88 FP additions, 48 FP multiplications,
Chris@10:  * (or, 64 additions, 24 multiplications, 24 fused multiply/add),
Chris@10:  * 37 stack variables, 0 constants, and 64 memory accesses
Chris@10:  */
Chris@10: #include "q.h"
Chris@10: 
Chris@10: static void q1_4(R *rio, R *iio, const R *W, stride rs, stride vs, INT mb, INT me, INT ms)
Chris@10: {
Chris@10:      {
Chris@10: 	  INT m;
Chris@10: 	  for (m = mb, W = W + (mb * 6); m < me; m = m + 1, rio = rio + ms, iio = iio + ms, W = W + 6, MAKE_VOLATILE_STRIDE(8, rs), MAKE_VOLATILE_STRIDE(0, vs)) {
Chris@10: 	       E T3, Te, Tb, Tq, T6, T8, Th, Tr, Tv, TG, TD, TS, Ty, TA, TJ;
Chris@10: 	       E TT, TX, T18, T15, T1k, T10, T12, T1b, T1l, T1p, T1A, T1x, T1M, T1s, T1u;
Chris@10: 	       E T1D, T1N;
Chris@10: 	       {
Chris@10: 		    E T1, T2, T9, Ta;
Chris@10: 		    T1 = rio[0];
Chris@10: 		    T2 = rio[WS(rs, 2)];
Chris@10: 		    T3 = T1 + T2;
Chris@10: 		    Te = T1 - T2;
Chris@10: 		    T9 = iio[0];
Chris@10: 		    Ta = iio[WS(rs, 2)];
Chris@10: 		    Tb = T9 - Ta;
Chris@10: 		    Tq = T9 + Ta;
Chris@10: 	       }
Chris@10: 	       {
Chris@10: 		    E T4, T5, Tf, Tg;
Chris@10: 		    T4 = rio[WS(rs, 1)];
Chris@10: 		    T5 = rio[WS(rs, 3)];
Chris@10: 		    T6 = T4 + T5;
Chris@10: 		    T8 = T4 - T5;
Chris@10: 		    Tf = iio[WS(rs, 1)];
Chris@10: 		    Tg = iio[WS(rs, 3)];
Chris@10: 		    Th = Tf - Tg;
Chris@10: 		    Tr = Tf + Tg;
Chris@10: 	       }
Chris@10: 	       {
Chris@10: 		    E Tt, Tu, TB, TC;
Chris@10: 		    Tt = rio[WS(vs, 1)];
Chris@10: 		    Tu = rio[WS(vs, 1) + WS(rs, 2)];
Chris@10: 		    Tv = Tt + Tu;
Chris@10: 		    TG = Tt - Tu;
Chris@10: 		    TB = iio[WS(vs, 1)];
Chris@10: 		    TC = iio[WS(vs, 1) + WS(rs, 2)];
Chris@10: 		    TD = TB - TC;
Chris@10: 		    TS = TB + TC;
Chris@10: 	       }
Chris@10: 	       {
Chris@10: 		    E Tw, Tx, TH, TI;
Chris@10: 		    Tw = rio[WS(vs, 1) + WS(rs, 1)];
Chris@10: 		    Tx = rio[WS(vs, 1) + WS(rs, 3)];
Chris@10: 		    Ty = Tw + Tx;
Chris@10: 		    TA = Tw - Tx;
Chris@10: 		    TH = iio[WS(vs, 1) + WS(rs, 1)];
Chris@10: 		    TI = iio[WS(vs, 1) + WS(rs, 3)];
Chris@10: 		    TJ = TH - TI;
Chris@10: 		    TT = TH + TI;
Chris@10: 	       }
Chris@10: 	       {
Chris@10: 		    E TV, TW, T13, T14;
Chris@10: 		    TV = rio[WS(vs, 2)];
Chris@10: 		    TW = rio[WS(vs, 2) + WS(rs, 2)];
Chris@10: 		    TX = TV + TW;
Chris@10: 		    T18 = TV - TW;
Chris@10: 		    T13 = iio[WS(vs, 2)];
Chris@10: 		    T14 = iio[WS(vs, 2) + WS(rs, 2)];
Chris@10: 		    T15 = T13 - T14;
Chris@10: 		    T1k = T13 + T14;
Chris@10: 	       }
Chris@10: 	       {
Chris@10: 		    E TY, TZ, T19, T1a;
Chris@10: 		    TY = rio[WS(vs, 2) + WS(rs, 1)];
Chris@10: 		    TZ = rio[WS(vs, 2) + WS(rs, 3)];
Chris@10: 		    T10 = TY + TZ;
Chris@10: 		    T12 = TY - TZ;
Chris@10: 		    T19 = iio[WS(vs, 2) + WS(rs, 1)];
Chris@10: 		    T1a = iio[WS(vs, 2) + WS(rs, 3)];
Chris@10: 		    T1b = T19 - T1a;
Chris@10: 		    T1l = T19 + T1a;
Chris@10: 	       }
Chris@10: 	       {
Chris@10: 		    E T1n, T1o, T1v, T1w;
Chris@10: 		    T1n = rio[WS(vs, 3)];
Chris@10: 		    T1o = rio[WS(vs, 3) + WS(rs, 2)];
Chris@10: 		    T1p = T1n + T1o;
Chris@10: 		    T1A = T1n - T1o;
Chris@10: 		    T1v = iio[WS(vs, 3)];
Chris@10: 		    T1w = iio[WS(vs, 3) + WS(rs, 2)];
Chris@10: 		    T1x = T1v - T1w;
Chris@10: 		    T1M = T1v + T1w;
Chris@10: 	       }
Chris@10: 	       {
Chris@10: 		    E T1q, T1r, T1B, T1C;
Chris@10: 		    T1q = rio[WS(vs, 3) + WS(rs, 1)];
Chris@10: 		    T1r = rio[WS(vs, 3) + WS(rs, 3)];
Chris@10: 		    T1s = T1q + T1r;
Chris@10: 		    T1u = T1q - T1r;
Chris@10: 		    T1B = iio[WS(vs, 3) + WS(rs, 1)];
Chris@10: 		    T1C = iio[WS(vs, 3) + WS(rs, 3)];
Chris@10: 		    T1D = T1B - T1C;
Chris@10: 		    T1N = T1B + T1C;
Chris@10: 	       }
Chris@10: 	       rio[0] = T3 + T6;
Chris@10: 	       iio[0] = Tq + Tr;
Chris@10: 	       rio[WS(rs, 1)] = Tv + Ty;
Chris@10: 	       iio[WS(rs, 1)] = TS + TT;
Chris@10: 	       rio[WS(rs, 2)] = TX + T10;
Chris@10: 	       iio[WS(rs, 2)] = T1k + T1l;
Chris@10: 	       iio[WS(rs, 3)] = T1M + T1N;
Chris@10: 	       rio[WS(rs, 3)] = T1p + T1s;
Chris@10: 	       {
Chris@10: 		    E Tc, Ti, T7, Td;
Chris@10: 		    Tc = T8 + Tb;
Chris@10: 		    Ti = Te - Th;
Chris@10: 		    T7 = W[4];
Chris@10: 		    Td = W[5];
Chris@10: 		    iio[WS(vs, 3)] = FNMS(Td, Ti, T7 * Tc);
Chris@10: 		    rio[WS(vs, 3)] = FMA(Td, Tc, T7 * Ti);
Chris@10: 	       }
Chris@10: 	       {
Chris@10: 		    E T1K, T1O, T1J, T1L;
Chris@10: 		    T1K = T1p - T1s;
Chris@10: 		    T1O = T1M - T1N;
Chris@10: 		    T1J = W[2];
Chris@10: 		    T1L = W[3];
Chris@10: 		    rio[WS(vs, 2) + WS(rs, 3)] = FMA(T1J, T1K, T1L * T1O);
Chris@10: 		    iio[WS(vs, 2) + WS(rs, 3)] = FNMS(T1L, T1K, T1J * T1O);
Chris@10: 	       }
Chris@10: 	       {
Chris@10: 		    E Tk, Tm, Tj, Tl;
Chris@10: 		    Tk = Tb - T8;
Chris@10: 		    Tm = Te + Th;
Chris@10: 		    Tj = W[0];
Chris@10: 		    Tl = W[1];
Chris@10: 		    iio[WS(vs, 1)] = FNMS(Tl, Tm, Tj * Tk);
Chris@10: 		    rio[WS(vs, 1)] = FMA(Tl, Tk, Tj * Tm);
Chris@10: 	       }
Chris@10: 	       {
Chris@10: 		    E To, Ts, Tn, Tp;
Chris@10: 		    To = T3 - T6;
Chris@10: 		    Ts = Tq - Tr;
Chris@10: 		    Tn = W[2];
Chris@10: 		    Tp = W[3];
Chris@10: 		    rio[WS(vs, 2)] = FMA(Tn, To, Tp * Ts);
Chris@10: 		    iio[WS(vs, 2)] = FNMS(Tp, To, Tn * Ts);
Chris@10: 	       }
Chris@10: 	       {
Chris@10: 		    E T16, T1c, T11, T17;
Chris@10: 		    T16 = T12 + T15;
Chris@10: 		    T1c = T18 - T1b;
Chris@10: 		    T11 = W[4];
Chris@10: 		    T17 = W[5];
Chris@10: 		    iio[WS(vs, 3) + WS(rs, 2)] = FNMS(T17, T1c, T11 * T16);
Chris@10: 		    rio[WS(vs, 3) + WS(rs, 2)] = FMA(T17, T16, T11 * T1c);
Chris@10: 	       }
Chris@10: 	       {
Chris@10: 		    E T1G, T1I, T1F, T1H;
Chris@10: 		    T1G = T1x - T1u;
Chris@10: 		    T1I = T1A + T1D;
Chris@10: 		    T1F = W[0];
Chris@10: 		    T1H = W[1];
Chris@10: 		    iio[WS(vs, 1) + WS(rs, 3)] = FNMS(T1H, T1I, T1F * T1G);
Chris@10: 		    rio[WS(vs, 1) + WS(rs, 3)] = FMA(T1H, T1G, T1F * T1I);
Chris@10: 	       }
Chris@10: 	       {
Chris@10: 		    E TQ, TU, TP, TR;
Chris@10: 		    TQ = Tv - Ty;
Chris@10: 		    TU = TS - TT;
Chris@10: 		    TP = W[2];
Chris@10: 		    TR = W[3];
Chris@10: 		    rio[WS(vs, 2) + WS(rs, 1)] = FMA(TP, TQ, TR * TU);
Chris@10: 		    iio[WS(vs, 2) + WS(rs, 1)] = FNMS(TR, TQ, TP * TU);
Chris@10: 	       }
Chris@10: 	       {
Chris@10: 		    E T1e, T1g, T1d, T1f;
Chris@10: 		    T1e = T15 - T12;
Chris@10: 		    T1g = T18 + T1b;
Chris@10: 		    T1d = W[0];
Chris@10: 		    T1f = W[1];
Chris@10: 		    iio[WS(vs, 1) + WS(rs, 2)] = FNMS(T1f, T1g, T1d * T1e);
Chris@10: 		    rio[WS(vs, 1) + WS(rs, 2)] = FMA(T1f, T1e, T1d * T1g);
Chris@10: 	       }
Chris@10: 	       {
Chris@10: 		    E T1i, T1m, T1h, T1j;
Chris@10: 		    T1i = TX - T10;
Chris@10: 		    T1m = T1k - T1l;
Chris@10: 		    T1h = W[2];
Chris@10: 		    T1j = W[3];
Chris@10: 		    rio[WS(vs, 2) + WS(rs, 2)] = FMA(T1h, T1i, T1j * T1m);
Chris@10: 		    iio[WS(vs, 2) + WS(rs, 2)] = FNMS(T1j, T1i, T1h * T1m);
Chris@10: 	       }
Chris@10: 	       {
Chris@10: 		    E T1y, T1E, T1t, T1z;
Chris@10: 		    T1y = T1u + T1x;
Chris@10: 		    T1E = T1A - T1D;
Chris@10: 		    T1t = W[4];
Chris@10: 		    T1z = W[5];
Chris@10: 		    iio[WS(vs, 3) + WS(rs, 3)] = FNMS(T1z, T1E, T1t * T1y);
Chris@10: 		    rio[WS(vs, 3) + WS(rs, 3)] = FMA(T1z, T1y, T1t * T1E);
Chris@10: 	       }
Chris@10: 	       {
Chris@10: 		    E TM, TO, TL, TN;
Chris@10: 		    TM = TD - TA;
Chris@10: 		    TO = TG + TJ;
Chris@10: 		    TL = W[0];
Chris@10: 		    TN = W[1];
Chris@10: 		    iio[WS(vs, 1) + WS(rs, 1)] = FNMS(TN, TO, TL * TM);
Chris@10: 		    rio[WS(vs, 1) + WS(rs, 1)] = FMA(TN, TM, TL * TO);
Chris@10: 	       }
Chris@10: 	       {
Chris@10: 		    E TE, TK, Tz, TF;
Chris@10: 		    TE = TA + TD;
Chris@10: 		    TK = TG - TJ;
Chris@10: 		    Tz = W[4];
Chris@10: 		    TF = W[5];
Chris@10: 		    iio[WS(vs, 3) + WS(rs, 1)] = FNMS(TF, TK, Tz * TE);
Chris@10: 		    rio[WS(vs, 3) + WS(rs, 1)] = FMA(TF, TE, Tz * TK);
Chris@10: 	       }
Chris@10: 	  }
Chris@10:      }
Chris@10: }
Chris@10: 
Chris@10: static const tw_instr twinstr[] = {
Chris@10:      {TW_FULL, 0, 4},
Chris@10:      {TW_NEXT, 1, 0}
Chris@10: };
Chris@10: 
Chris@10: static const ct_desc desc = { 4, "q1_4", twinstr, &GENUS, {64, 24, 24, 0}, 0, 0, 0 };
Chris@10: 
Chris@10: void X(codelet_q1_4) (planner *p) {
Chris@10:      X(kdft_difsq_register) (p, q1_4, &desc);
Chris@10: }
Chris@10: #endif				/* HAVE_FMA */