Chris@82: /*
Chris@82:  * Copyright (c) 2003, 2007-14 Matteo Frigo
Chris@82:  * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
Chris@82:  *
Chris@82:  * This program is free software; you can redistribute it and/or modify
Chris@82:  * it under the terms of the GNU General Public License as published by
Chris@82:  * the Free Software Foundation; either version 2 of the License, or
Chris@82:  * (at your option) any later version.
Chris@82:  *
Chris@82:  * This program is distributed in the hope that it will be useful,
Chris@82:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
Chris@82:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
Chris@82:  * GNU General Public License for more details.
Chris@82:  *
Chris@82:  * You should have received a copy of the GNU General Public License
Chris@82:  * along with this program; if not, write to the Free Software
Chris@82:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
Chris@82:  *
Chris@82:  */
Chris@82: 
Chris@82: /* This file was automatically generated --- DO NOT EDIT */
Chris@82: /* Generated on Thu May 24 08:07:31 EDT 2018 */
Chris@82: 
Chris@82: #include "rdft/codelet-rdft.h"
Chris@82: 
Chris@82: #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA)
Chris@82: 
Chris@82: /* Generated by: ../../../genfft/gen_hc2hc.native -fma -compact -variables 4 -pipeline-latency 4 -sign 1 -n 2 -dif -name hb_2 -include rdft/scalar/hb.h */
Chris@82: 
Chris@82: /*
Chris@82:  * This function contains 6 FP additions, 4 FP multiplications,
Chris@82:  * (or, 4 additions, 2 multiplications, 2 fused multiply/add),
Chris@82:  * 11 stack variables, 0 constants, and 8 memory accesses
Chris@82:  */
Chris@82: #include "rdft/scalar/hb.h"
Chris@82: 
Chris@82: static void hb_2(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms)
Chris@82: {
Chris@82:      {
Chris@82: 	  INT m;
Chris@82: 	  for (m = mb, W = W + ((mb - 1) * 2); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 2, MAKE_VOLATILE_STRIDE(4, rs)) {
Chris@82: 	       E T1, T2, T6, T3, T4, T9;
Chris@82: 	       T1 = cr[0];
Chris@82: 	       T2 = ci[0];
Chris@82: 	       T6 = T1 - T2;
Chris@82: 	       T3 = ci[WS(rs, 1)];
Chris@82: 	       T4 = cr[WS(rs, 1)];
Chris@82: 	       T9 = T3 + T4;
Chris@82: 	       cr[0] = T1 + T2;
Chris@82: 	       ci[0] = T3 - T4;
Chris@82: 	       {
Chris@82: 		    E T5, T7, T8, Ta;
Chris@82: 		    T5 = W[0];
Chris@82: 		    T7 = T5 * T6;
Chris@82: 		    T8 = W[1];
Chris@82: 		    Ta = T8 * T6;
Chris@82: 		    cr[WS(rs, 1)] = FNMS(T8, T9, T7);
Chris@82: 		    ci[WS(rs, 1)] = FMA(T5, T9, Ta);
Chris@82: 	       }
Chris@82: 	  }
Chris@82:      }
Chris@82: }
Chris@82: 
Chris@82: static const tw_instr twinstr[] = {
Chris@82:      {TW_FULL, 1, 2},
Chris@82:      {TW_NEXT, 1, 0}
Chris@82: };
Chris@82: 
Chris@82: static const hc2hc_desc desc = { 2, "hb_2", twinstr, &GENUS, {4, 2, 2, 0} };
Chris@82: 
Chris@82: void X(codelet_hb_2) (planner *p) {
Chris@82:      X(khc2hc_register) (p, hb_2, &desc);
Chris@82: }
Chris@82: #else
Chris@82: 
Chris@82: /* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 2 -dif -name hb_2 -include rdft/scalar/hb.h */
Chris@82: 
Chris@82: /*
Chris@82:  * This function contains 6 FP additions, 4 FP multiplications,
Chris@82:  * (or, 4 additions, 2 multiplications, 2 fused multiply/add),
Chris@82:  * 9 stack variables, 0 constants, and 8 memory accesses
Chris@82:  */
Chris@82: #include "rdft/scalar/hb.h"
Chris@82: 
Chris@82: static void hb_2(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms)
Chris@82: {
Chris@82:      {
Chris@82: 	  INT m;
Chris@82: 	  for (m = mb, W = W + ((mb - 1) * 2); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 2, MAKE_VOLATILE_STRIDE(4, rs)) {
Chris@82: 	       E T1, T2, T6, T3, T4, T8, T5, T7;
Chris@82: 	       T1 = cr[0];
Chris@82: 	       T2 = ci[0];
Chris@82: 	       T6 = T1 - T2;
Chris@82: 	       T3 = ci[WS(rs, 1)];
Chris@82: 	       T4 = cr[WS(rs, 1)];
Chris@82: 	       T8 = T3 + T4;
Chris@82: 	       cr[0] = T1 + T2;
Chris@82: 	       ci[0] = T3 - T4;
Chris@82: 	       T5 = W[0];
Chris@82: 	       T7 = W[1];
Chris@82: 	       cr[WS(rs, 1)] = FNMS(T7, T8, T5 * T6);
Chris@82: 	       ci[WS(rs, 1)] = FMA(T7, T6, T5 * T8);
Chris@82: 	  }
Chris@82:      }
Chris@82: }
Chris@82: 
Chris@82: static const tw_instr twinstr[] = {
Chris@82:      {TW_FULL, 1, 2},
Chris@82:      {TW_NEXT, 1, 0}
Chris@82: };
Chris@82: 
Chris@82: static const hc2hc_desc desc = { 2, "hb_2", twinstr, &GENUS, {4, 2, 2, 0} };
Chris@82: 
Chris@82: void X(codelet_hb_2) (planner *p) {
Chris@82:      X(khc2hc_register) (p, hb_2, &desc);
Chris@82: }
Chris@82: #endif