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 07:39:49 EST 2012 */ Chris@10: Chris@10: #include "codelet-rdft.h" Chris@10: Chris@10: #ifdef HAVE_FMA Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_hc2hc.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 3 -dit -name hf_3 -include hf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 16 FP additions, 14 FP multiplications, Chris@10: * (or, 6 additions, 4 multiplications, 10 fused multiply/add), Chris@10: * 21 stack variables, 2 constants, and 12 memory accesses Chris@10: */ Chris@10: #include "hf.h" Chris@10: Chris@10: static void hf_3(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * 4); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 4, MAKE_VOLATILE_STRIDE(6, rs)) { Chris@10: E T1, Tl, T9, Tc, Tb, Th, T7, Ti, Ta, Tj, Td; Chris@10: T1 = cr[0]; Chris@10: Tl = ci[0]; Chris@10: { Chris@10: E T3, T6, T2, T5, Tg, T4, T8; Chris@10: T3 = cr[WS(rs, 1)]; Chris@10: T6 = ci[WS(rs, 1)]; Chris@10: T2 = W[0]; Chris@10: T5 = W[1]; Chris@10: T9 = cr[WS(rs, 2)]; Chris@10: Tc = ci[WS(rs, 2)]; Chris@10: Tg = T2 * T6; Chris@10: T4 = T2 * T3; Chris@10: T8 = W[2]; Chris@10: Tb = W[3]; Chris@10: Th = FNMS(T5, T3, Tg); Chris@10: T7 = FMA(T5, T6, T4); Chris@10: Ti = T8 * Tc; Chris@10: Ta = T8 * T9; Chris@10: } Chris@10: Tj = FNMS(Tb, T9, Ti); Chris@10: Td = FMA(Tb, Tc, Ta); Chris@10: { Chris@10: E Tk, Te, To, Tn, Tm, Tf; Chris@10: Tk = Th - Tj; Chris@10: Tm = Th + Tj; Chris@10: Te = T7 + Td; Chris@10: To = Td - T7; Chris@10: ci[WS(rs, 2)] = Tm + Tl; Chris@10: Tn = FNMS(KP500000000, Tm, Tl); Chris@10: cr[0] = T1 + Te; Chris@10: Tf = FNMS(KP500000000, Te, T1); Chris@10: ci[WS(rs, 1)] = FMA(KP866025403, To, Tn); Chris@10: cr[WS(rs, 2)] = FMS(KP866025403, To, Tn); Chris@10: cr[WS(rs, 1)] = FMA(KP866025403, Tk, Tf); Chris@10: ci[0] = FNMS(KP866025403, Tk, Tf); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 3}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2hc_desc desc = { 3, "hf_3", twinstr, &GENUS, {6, 4, 10, 0} }; Chris@10: Chris@10: void X(codelet_hf_3) (planner *p) { Chris@10: X(khc2hc_register) (p, hf_3, &desc); Chris@10: } Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -n 3 -dit -name hf_3 -include hf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 16 FP additions, 12 FP multiplications, Chris@10: * (or, 10 additions, 6 multiplications, 6 fused multiply/add), Chris@10: * 15 stack variables, 2 constants, and 12 memory accesses Chris@10: */ Chris@10: #include "hf.h" Chris@10: Chris@10: static void hf_3(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * 4); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 4, MAKE_VOLATILE_STRIDE(6, rs)) { Chris@10: E T1, Ti, T6, Te, Tb, Tf, Tc, Tj; Chris@10: T1 = cr[0]; Chris@10: Ti = ci[0]; Chris@10: { Chris@10: E T3, T5, T2, T4; Chris@10: T3 = cr[WS(rs, 1)]; Chris@10: T5 = ci[WS(rs, 1)]; Chris@10: T2 = W[0]; Chris@10: T4 = W[1]; Chris@10: T6 = FMA(T2, T3, T4 * T5); Chris@10: Te = FNMS(T4, T3, T2 * T5); Chris@10: } Chris@10: { Chris@10: E T8, Ta, T7, T9; Chris@10: T8 = cr[WS(rs, 2)]; Chris@10: Ta = ci[WS(rs, 2)]; Chris@10: T7 = W[2]; Chris@10: T9 = W[3]; Chris@10: Tb = FMA(T7, T8, T9 * Ta); Chris@10: Tf = FNMS(T9, T8, T7 * Ta); Chris@10: } Chris@10: Tc = T6 + Tb; Chris@10: Tj = Te + Tf; Chris@10: { Chris@10: E Td, Tg, Th, Tk; Chris@10: cr[0] = T1 + Tc; Chris@10: Td = FNMS(KP500000000, Tc, T1); Chris@10: Tg = KP866025403 * (Te - Tf); Chris@10: ci[0] = Td - Tg; Chris@10: cr[WS(rs, 1)] = Td + Tg; Chris@10: ci[WS(rs, 2)] = Tj + Ti; Chris@10: Th = KP866025403 * (Tb - T6); Chris@10: Tk = FNMS(KP500000000, Tj, Ti); Chris@10: cr[WS(rs, 2)] = Th - Tk; Chris@10: ci[WS(rs, 1)] = Th + Tk; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 3}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2hc_desc desc = { 3, "hf_3", twinstr, &GENUS, {10, 6, 6, 0} }; Chris@10: Chris@10: void X(codelet_hf_3) (planner *p) { Chris@10: X(khc2hc_register) (p, hf_3, &desc); Chris@10: } Chris@10: #endif /* HAVE_FMA */