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 07:35:47 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_twiddle.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 2 -name t1_2 -include t.h */
Chris@10: 
Chris@10: /*
Chris@10:  * This function contains 6 FP additions, 4 FP multiplications,
Chris@10:  * (or, 4 additions, 2 multiplications, 2 fused multiply/add),
Chris@10:  * 11 stack variables, 0 constants, and 8 memory accesses
Chris@10:  */
Chris@10: #include "t.h"
Chris@10: 
Chris@10: static void t1_2(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms)
Chris@10: {
Chris@10:      {
Chris@10: 	  INT m;
Chris@10: 	  for (m = mb, W = W + (mb * 2); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 2, MAKE_VOLATILE_STRIDE(4, rs)) {
Chris@10: 	       E T1, Ta, T3, T6, T2, T5;
Chris@10: 	       T1 = ri[0];
Chris@10: 	       Ta = ii[0];
Chris@10: 	       T3 = ri[WS(rs, 1)];
Chris@10: 	       T6 = ii[WS(rs, 1)];
Chris@10: 	       T2 = W[0];
Chris@10: 	       T5 = W[1];
Chris@10: 	       {
Chris@10: 		    E T8, T4, T9, T7;
Chris@10: 		    T8 = T2 * T6;
Chris@10: 		    T4 = T2 * T3;
Chris@10: 		    T9 = FNMS(T5, T3, T8);
Chris@10: 		    T7 = FMA(T5, T6, T4);
Chris@10: 		    ii[0] = T9 + Ta;
Chris@10: 		    ii[WS(rs, 1)] = Ta - T9;
Chris@10: 		    ri[0] = T1 + T7;
Chris@10: 		    ri[WS(rs, 1)] = T1 - T7;
Chris@10: 	       }
Chris@10: 	  }
Chris@10:      }
Chris@10: }
Chris@10: 
Chris@10: static const tw_instr twinstr[] = {
Chris@10:      {TW_FULL, 0, 2},
Chris@10:      {TW_NEXT, 1, 0}
Chris@10: };
Chris@10: 
Chris@10: static const ct_desc desc = { 2, "t1_2", twinstr, &GENUS, {4, 2, 2, 0}, 0, 0, 0 };
Chris@10: 
Chris@10: void X(codelet_t1_2) (planner *p) {
Chris@10:      X(kdft_dit_register) (p, t1_2, &desc);
Chris@10: }
Chris@10: #else				/* HAVE_FMA */
Chris@10: 
Chris@10: /* Generated by: ../../../genfft/gen_twiddle.native -compact -variables 4 -pipeline-latency 4 -n 2 -name t1_2 -include t.h */
Chris@10: 
Chris@10: /*
Chris@10:  * This function contains 6 FP additions, 4 FP multiplications,
Chris@10:  * (or, 4 additions, 2 multiplications, 2 fused multiply/add),
Chris@10:  * 9 stack variables, 0 constants, and 8 memory accesses
Chris@10:  */
Chris@10: #include "t.h"
Chris@10: 
Chris@10: static void t1_2(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms)
Chris@10: {
Chris@10:      {
Chris@10: 	  INT m;
Chris@10: 	  for (m = mb, W = W + (mb * 2); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 2, MAKE_VOLATILE_STRIDE(4, rs)) {
Chris@10: 	       E T1, T8, T6, T7;
Chris@10: 	       T1 = ri[0];
Chris@10: 	       T8 = ii[0];
Chris@10: 	       {
Chris@10: 		    E T3, T5, T2, T4;
Chris@10: 		    T3 = ri[WS(rs, 1)];
Chris@10: 		    T5 = ii[WS(rs, 1)];
Chris@10: 		    T2 = W[0];
Chris@10: 		    T4 = W[1];
Chris@10: 		    T6 = FMA(T2, T3, T4 * T5);
Chris@10: 		    T7 = FNMS(T4, T3, T2 * T5);
Chris@10: 	       }
Chris@10: 	       ri[WS(rs, 1)] = T1 - T6;
Chris@10: 	       ii[WS(rs, 1)] = T8 - T7;
Chris@10: 	       ri[0] = T1 + T6;
Chris@10: 	       ii[0] = T7 + T8;
Chris@10: 	  }
Chris@10:      }
Chris@10: }
Chris@10: 
Chris@10: static const tw_instr twinstr[] = {
Chris@10:      {TW_FULL, 0, 2},
Chris@10:      {TW_NEXT, 1, 0}
Chris@10: };
Chris@10: 
Chris@10: static const ct_desc desc = { 2, "t1_2", twinstr, &GENUS, {4, 2, 2, 0}, 0, 0, 0 };
Chris@10: 
Chris@10: void X(codelet_t1_2) (planner *p) {
Chris@10:      X(kdft_dit_register) (p, t1_2, &desc);
Chris@10: }
Chris@10: #endif				/* HAVE_FMA */