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 07:37:21 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_notw_c.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -n 2 -name n2fv_2 -with-ostride 2 -include n2f.h -store-multiple 2 */
Chris@10: 
Chris@10: /*
Chris@10:  * This function contains 2 FP additions, 0 FP multiplications,
Chris@10:  * (or, 2 additions, 0 multiplications, 0 fused multiply/add),
Chris@10:  * 7 stack variables, 0 constants, and 5 memory accesses
Chris@10:  */
Chris@10: #include "n2f.h"
Chris@10: 
Chris@10: static void n2fv_2(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs)
Chris@10: {
Chris@10:      {
Chris@10: 	  INT i;
Chris@10: 	  const R *xi;
Chris@10: 	  R *xo;
Chris@10: 	  xi = ri;
Chris@10: 	  xo = ro;
Chris@10: 	  for (i = v; i > 0; i = i - VL, xi = xi + (VL * ivs), xo = xo + (VL * ovs), MAKE_VOLATILE_STRIDE(4, is), MAKE_VOLATILE_STRIDE(4, os)) {
Chris@10: 	       V T1, T2, T3, T4;
Chris@10: 	       T1 = LD(&(xi[0]), ivs, &(xi[0]));
Chris@10: 	       T2 = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)]));
Chris@10: 	       T3 = VADD(T1, T2);
Chris@10: 	       STM2(&(xo[0]), T3, ovs, &(xo[0]));
Chris@10: 	       T4 = VSUB(T1, T2);
Chris@10: 	       STM2(&(xo[2]), T4, ovs, &(xo[2]));
Chris@10: 	       STN2(&(xo[0]), T3, T4, ovs);
Chris@10: 	  }
Chris@10:      }
Chris@10:      VLEAVE();
Chris@10: }
Chris@10: 
Chris@10: static const kdft_desc desc = { 2, XSIMD_STRING("n2fv_2"), {2, 0, 0, 0}, &GENUS, 0, 2, 0, 0 };
Chris@10: 
Chris@10: void XSIMD(codelet_n2fv_2) (planner *p) {
Chris@10:      X(kdft_register) (p, n2fv_2, &desc);
Chris@10: }
Chris@10: 
Chris@10: #else				/* HAVE_FMA */
Chris@10: 
Chris@10: /* Generated by: ../../../genfft/gen_notw_c.native -simd -compact -variables 4 -pipeline-latency 8 -n 2 -name n2fv_2 -with-ostride 2 -include n2f.h -store-multiple 2 */
Chris@10: 
Chris@10: /*
Chris@10:  * This function contains 2 FP additions, 0 FP multiplications,
Chris@10:  * (or, 2 additions, 0 multiplications, 0 fused multiply/add),
Chris@10:  * 7 stack variables, 0 constants, and 5 memory accesses
Chris@10:  */
Chris@10: #include "n2f.h"
Chris@10: 
Chris@10: static void n2fv_2(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs)
Chris@10: {
Chris@10:      {
Chris@10: 	  INT i;
Chris@10: 	  const R *xi;
Chris@10: 	  R *xo;
Chris@10: 	  xi = ri;
Chris@10: 	  xo = ro;
Chris@10: 	  for (i = v; i > 0; i = i - VL, xi = xi + (VL * ivs), xo = xo + (VL * ovs), MAKE_VOLATILE_STRIDE(4, is), MAKE_VOLATILE_STRIDE(4, os)) {
Chris@10: 	       V T1, T2, T3, T4;
Chris@10: 	       T1 = LD(&(xi[0]), ivs, &(xi[0]));
Chris@10: 	       T2 = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)]));
Chris@10: 	       T3 = VSUB(T1, T2);
Chris@10: 	       STM2(&(xo[2]), T3, ovs, &(xo[2]));
Chris@10: 	       T4 = VADD(T1, T2);
Chris@10: 	       STM2(&(xo[0]), T4, ovs, &(xo[0]));
Chris@10: 	       STN2(&(xo[0]), T4, T3, ovs);
Chris@10: 	  }
Chris@10:      }
Chris@10:      VLEAVE();
Chris@10: }
Chris@10: 
Chris@10: static const kdft_desc desc = { 2, XSIMD_STRING("n2fv_2"), {2, 0, 0, 0}, &GENUS, 0, 2, 0, 0 };
Chris@10: 
Chris@10: void XSIMD(codelet_n2fv_2) (planner *p) {
Chris@10:      X(kdft_register) (p, n2fv_2, &desc);
Chris@10: }
Chris@10: 
Chris@10: #endif				/* HAVE_FMA */