cannam@95: /*
cannam@95:  * Copyright (c) 2003, 2007-11 Matteo Frigo
cannam@95:  * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology
cannam@95:  *
cannam@95:  * This program is free software; you can redistribute it and/or modify
cannam@95:  * it under the terms of the GNU General Public License as published by
cannam@95:  * the Free Software Foundation; either version 2 of the License, or
cannam@95:  * (at your option) any later version.
cannam@95:  *
cannam@95:  * This program is distributed in the hope that it will be useful,
cannam@95:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
cannam@95:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
cannam@95:  * GNU General Public License for more details.
cannam@95:  *
cannam@95:  * You should have received a copy of the GNU General Public License
cannam@95:  * along with this program; if not, write to the Free Software
cannam@95:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
cannam@95:  *
cannam@95:  */
cannam@95: 
cannam@95: /* This file was automatically generated --- DO NOT EDIT */
cannam@95: /* Generated on Sun Nov 25 07:41:52 EST 2012 */
cannam@95: 
cannam@95: #include "codelet-rdft.h"
cannam@95: 
cannam@95: #ifdef HAVE_FMA
cannam@95: 
cannam@95: /* Generated by: ../../../genfft/gen_hc2c.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -sign 1 -n 2 -dif -name hc2cb_2 -include hc2cb.h */
cannam@95: 
cannam@95: /*
cannam@95:  * This function contains 6 FP additions, 4 FP multiplications,
cannam@95:  * (or, 4 additions, 2 multiplications, 2 fused multiply/add),
cannam@95:  * 11 stack variables, 0 constants, and 8 memory accesses
cannam@95:  */
cannam@95: #include "hc2cb.h"
cannam@95: 
cannam@95: static void hc2cb_2(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms)
cannam@95: {
cannam@95:      {
cannam@95: 	  INT m;
cannam@95: 	  for (m = mb, W = W + ((mb - 1) * 2); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 2, MAKE_VOLATILE_STRIDE(8, rs)) {
cannam@95: 	       E T5, T6, T9, T8, T7, Ta;
cannam@95: 	       {
cannam@95: 		    E T1, T2, T3, T4;
cannam@95: 		    T1 = Rp[0];
cannam@95: 		    T2 = Rm[0];
cannam@95: 		    T3 = Ip[0];
cannam@95: 		    T4 = Im[0];
cannam@95: 		    T5 = W[0];
cannam@95: 		    Rp[0] = T1 + T2;
cannam@95: 		    T6 = T1 - T2;
cannam@95: 		    Rm[0] = T3 - T4;
cannam@95: 		    T9 = T3 + T4;
cannam@95: 		    T8 = W[1];
cannam@95: 		    T7 = T5 * T6;
cannam@95: 	       }
cannam@95: 	       Ta = T8 * T6;
cannam@95: 	       Ip[0] = FNMS(T8, T9, T7);
cannam@95: 	       Im[0] = FMA(T5, T9, Ta);
cannam@95: 	  }
cannam@95:      }
cannam@95: }
cannam@95: 
cannam@95: static const tw_instr twinstr[] = {
cannam@95:      {TW_FULL, 1, 2},
cannam@95:      {TW_NEXT, 1, 0}
cannam@95: };
cannam@95: 
cannam@95: static const hc2c_desc desc = { 2, "hc2cb_2", twinstr, &GENUS, {4, 2, 2, 0} };
cannam@95: 
cannam@95: void X(codelet_hc2cb_2) (planner *p) {
cannam@95:      X(khc2c_register) (p, hc2cb_2, &desc, HC2C_VIA_RDFT);
cannam@95: }
cannam@95: #else				/* HAVE_FMA */
cannam@95: 
cannam@95: /* Generated by: ../../../genfft/gen_hc2c.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 2 -dif -name hc2cb_2 -include hc2cb.h */
cannam@95: 
cannam@95: /*
cannam@95:  * This function contains 6 FP additions, 4 FP multiplications,
cannam@95:  * (or, 4 additions, 2 multiplications, 2 fused multiply/add),
cannam@95:  * 9 stack variables, 0 constants, and 8 memory accesses
cannam@95:  */
cannam@95: #include "hc2cb.h"
cannam@95: 
cannam@95: static void hc2cb_2(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms)
cannam@95: {
cannam@95:      {
cannam@95: 	  INT m;
cannam@95: 	  for (m = mb, W = W + ((mb - 1) * 2); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 2, MAKE_VOLATILE_STRIDE(8, rs)) {
cannam@95: 	       E T1, T2, T6, T3, T4, T8, T5, T7;
cannam@95: 	       T1 = Rp[0];
cannam@95: 	       T2 = Rm[0];
cannam@95: 	       T6 = T1 - T2;
cannam@95: 	       T3 = Ip[0];
cannam@95: 	       T4 = Im[0];
cannam@95: 	       T8 = T3 + T4;
cannam@95: 	       Rp[0] = T1 + T2;
cannam@95: 	       Rm[0] = T3 - T4;
cannam@95: 	       T5 = W[0];
cannam@95: 	       T7 = W[1];
cannam@95: 	       Ip[0] = FNMS(T7, T8, T5 * T6);
cannam@95: 	       Im[0] = FMA(T7, T6, T5 * T8);
cannam@95: 	  }
cannam@95:      }
cannam@95: }
cannam@95: 
cannam@95: static const tw_instr twinstr[] = {
cannam@95:      {TW_FULL, 1, 2},
cannam@95:      {TW_NEXT, 1, 0}
cannam@95: };
cannam@95: 
cannam@95: static const hc2c_desc desc = { 2, "hc2cb_2", twinstr, &GENUS, {4, 2, 2, 0} };
cannam@95: 
cannam@95: void X(codelet_hc2cb_2) (planner *p) {
cannam@95:      X(khc2c_register) (p, hc2cb_2, &desc, HC2C_VIA_RDFT);
cannam@95: }
cannam@95: #endif				/* HAVE_FMA */