Chris@42: /*
Chris@42:  * Copyright (c) 2003, 2007-14 Matteo Frigo
Chris@42:  * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
Chris@42:  *
Chris@42:  * This program is free software; you can redistribute it and/or modify
Chris@42:  * it under the terms of the GNU General Public License as published by
Chris@42:  * the Free Software Foundation; either version 2 of the License, or
Chris@42:  * (at your option) any later version.
Chris@42:  *
Chris@42:  * This program is distributed in the hope that it will be useful,
Chris@42:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
Chris@42:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
Chris@42:  * GNU General Public License for more details.
Chris@42:  *
Chris@42:  * You should have received a copy of the GNU General Public License
Chris@42:  * along with this program; if not, write to the Free Software
Chris@42:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
Chris@42:  *
Chris@42:  */
Chris@42: 
Chris@42: /* This file was automatically generated --- DO NOT EDIT */
Chris@42: /* Generated on Sat Jul 30 16:50:34 EDT 2016 */
Chris@42: 
Chris@42: #include "codelet-rdft.h"
Chris@42: 
Chris@42: #ifdef HAVE_FMA
Chris@42: 
Chris@42: /* Generated by: ../../../genfft/gen_r2cb.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -sign 1 -n 4 -name r2cbIII_4 -dft-III -include r2cbIII.h */
Chris@42: 
Chris@42: /*
Chris@42:  * This function contains 6 FP additions, 4 FP multiplications,
Chris@42:  * (or, 6 additions, 4 multiplications, 0 fused multiply/add),
Chris@42:  * 9 stack variables, 2 constants, and 8 memory accesses
Chris@42:  */
Chris@42: #include "r2cbIII.h"
Chris@42: 
Chris@42: static void r2cbIII_4(R *R0, R *R1, R *Cr, R *Ci, stride rs, stride csr, stride csi, INT v, INT ivs, INT ovs)
Chris@42: {
Chris@42:      DK(KP1_414213562, +1.414213562373095048801688724209698078569671875);
Chris@42:      DK(KP2_000000000, +2.000000000000000000000000000000000000000000000);
Chris@42:      {
Chris@42: 	  INT i;
Chris@42: 	  for (i = v; i > 0; i = i - 1, R0 = R0 + ovs, R1 = R1 + ovs, Cr = Cr + ivs, Ci = Ci + ivs, MAKE_VOLATILE_STRIDE(16, rs), MAKE_VOLATILE_STRIDE(16, csr), MAKE_VOLATILE_STRIDE(16, csi)) {
Chris@42: 	       E T1, T2, T4, T5, T3, T6;
Chris@42: 	       T1 = Cr[0];
Chris@42: 	       T2 = Cr[WS(csr, 1)];
Chris@42: 	       T4 = Ci[0];
Chris@42: 	       T5 = Ci[WS(csi, 1)];
Chris@42: 	       R0[0] = KP2_000000000 * (T1 + T2);
Chris@42: 	       T3 = T1 - T2;
Chris@42: 	       R0[WS(rs, 1)] = KP2_000000000 * (T5 - T4);
Chris@42: 	       T6 = T4 + T5;
Chris@42: 	       R1[WS(rs, 1)] = -(KP1_414213562 * (T3 + T6));
Chris@42: 	       R1[0] = KP1_414213562 * (T3 - T6);
Chris@42: 	  }
Chris@42:      }
Chris@42: }
Chris@42: 
Chris@42: static const kr2c_desc desc = { 4, "r2cbIII_4", {6, 4, 0, 0}, &GENUS };
Chris@42: 
Chris@42: void X(codelet_r2cbIII_4) (planner *p) {
Chris@42:      X(kr2c_register) (p, r2cbIII_4, &desc);
Chris@42: }
Chris@42: 
Chris@42: #else				/* HAVE_FMA */
Chris@42: 
Chris@42: /* Generated by: ../../../genfft/gen_r2cb.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 4 -name r2cbIII_4 -dft-III -include r2cbIII.h */
Chris@42: 
Chris@42: /*
Chris@42:  * This function contains 6 FP additions, 4 FP multiplications,
Chris@42:  * (or, 6 additions, 4 multiplications, 0 fused multiply/add),
Chris@42:  * 9 stack variables, 2 constants, and 8 memory accesses
Chris@42:  */
Chris@42: #include "r2cbIII.h"
Chris@42: 
Chris@42: static void r2cbIII_4(R *R0, R *R1, R *Cr, R *Ci, stride rs, stride csr, stride csi, INT v, INT ivs, INT ovs)
Chris@42: {
Chris@42:      DK(KP1_414213562, +1.414213562373095048801688724209698078569671875);
Chris@42:      DK(KP2_000000000, +2.000000000000000000000000000000000000000000000);
Chris@42:      {
Chris@42: 	  INT i;
Chris@42: 	  for (i = v; i > 0; i = i - 1, R0 = R0 + ovs, R1 = R1 + ovs, Cr = Cr + ivs, Ci = Ci + ivs, MAKE_VOLATILE_STRIDE(16, rs), MAKE_VOLATILE_STRIDE(16, csr), MAKE_VOLATILE_STRIDE(16, csi)) {
Chris@42: 	       E T1, T2, T3, T4, T5, T6;
Chris@42: 	       T1 = Cr[0];
Chris@42: 	       T2 = Cr[WS(csr, 1)];
Chris@42: 	       T3 = T1 - T2;
Chris@42: 	       T4 = Ci[0];
Chris@42: 	       T5 = Ci[WS(csi, 1)];
Chris@42: 	       T6 = T4 + T5;
Chris@42: 	       R0[0] = KP2_000000000 * (T1 + T2);
Chris@42: 	       R0[WS(rs, 1)] = KP2_000000000 * (T5 - T4);
Chris@42: 	       R1[0] = KP1_414213562 * (T3 - T6);
Chris@42: 	       R1[WS(rs, 1)] = -(KP1_414213562 * (T3 + T6));
Chris@42: 	  }
Chris@42:      }
Chris@42: }
Chris@42: 
Chris@42: static const kr2c_desc desc = { 4, "r2cbIII_4", {6, 4, 0, 0}, &GENUS };
Chris@42: 
Chris@42: void X(codelet_r2cbIII_4) (planner *p) {
Chris@42:      X(kr2c_register) (p, r2cbIII_4, &desc);
Chris@42: }
Chris@42: 
Chris@42: #endif				/* HAVE_FMA */