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 07:41:12 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 -sign 1 -n 4 -dif -name hb_4 -include hb.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 22 FP additions, 12 FP multiplications, Chris@10: * (or, 16 additions, 6 multiplications, 6 fused multiply/add), Chris@10: * 27 stack variables, 0 constants, and 16 memory accesses Chris@10: */ Chris@10: #include "hb.h" Chris@10: Chris@10: static void hb_4(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * 6); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 6, MAKE_VOLATILE_STRIDE(8, rs)) { Chris@10: E T8, Th, Ta, T7, Ti, T9; Chris@10: { Chris@10: E Td, Tg, T3, T6, Tu, Tm, Tx, Tr; Chris@10: { Chris@10: E Tq, Tl, T4, T5, Tk, Tp; Chris@10: { Chris@10: E Tb, Tc, Te, Tf, T1, T2; Chris@10: Tb = ci[WS(rs, 3)]; Chris@10: Tc = cr[WS(rs, 2)]; Chris@10: Te = ci[WS(rs, 2)]; Chris@10: Tf = cr[WS(rs, 3)]; Chris@10: T1 = cr[0]; Chris@10: Tq = Tb + Tc; Chris@10: Td = Tb - Tc; Chris@10: T2 = ci[WS(rs, 1)]; Chris@10: Tl = Te + Tf; Chris@10: Tg = Te - Tf; Chris@10: T4 = cr[WS(rs, 1)]; Chris@10: T5 = ci[0]; Chris@10: T3 = T1 + T2; Chris@10: Tk = T1 - T2; Chris@10: } Chris@10: Tp = T4 - T5; Chris@10: T6 = T4 + T5; Chris@10: Tu = Tk + Tl; Chris@10: Tm = Tk - Tl; Chris@10: Tx = Tq - Tp; Chris@10: Tr = Tp + Tq; Chris@10: T8 = T3 - T6; Chris@10: } Chris@10: cr[0] = T3 + T6; Chris@10: { Chris@10: E Tj, To, Tw, Tv; Chris@10: Tj = W[0]; Chris@10: ci[0] = Td + Tg; Chris@10: To = W[1]; Chris@10: { Chris@10: E Tt, Ts, Tn, Ty; Chris@10: Tt = W[4]; Chris@10: Ts = Tj * Tr; Chris@10: Tn = Tj * Tm; Chris@10: Tw = W[5]; Chris@10: Ty = Tt * Tx; Chris@10: Tv = Tt * Tu; Chris@10: ci[WS(rs, 1)] = FMA(To, Tm, Ts); Chris@10: cr[WS(rs, 1)] = FNMS(To, Tr, Tn); Chris@10: ci[WS(rs, 3)] = FMA(Tw, Tu, Ty); Chris@10: } Chris@10: cr[WS(rs, 3)] = FNMS(Tw, Tx, Tv); Chris@10: Th = Td - Tg; Chris@10: Ta = W[3]; Chris@10: T7 = W[2]; Chris@10: } Chris@10: } Chris@10: Ti = Ta * T8; Chris@10: T9 = T7 * T8; Chris@10: ci[WS(rs, 2)] = FMA(T7, Th, Ti); Chris@10: cr[WS(rs, 2)] = FNMS(Ta, Th, T9); Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 4}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2hc_desc desc = { 4, "hb_4", twinstr, &GENUS, {16, 6, 6, 0} }; Chris@10: Chris@10: void X(codelet_hb_4) (planner *p) { Chris@10: X(khc2hc_register) (p, hb_4, &desc); Chris@10: } Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 4 -dif -name hb_4 -include hb.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 22 FP additions, 12 FP multiplications, Chris@10: * (or, 16 additions, 6 multiplications, 6 fused multiply/add), Chris@10: * 13 stack variables, 0 constants, and 16 memory accesses Chris@10: */ Chris@10: #include "hb.h" Chris@10: Chris@10: static void hb_4(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * 6); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 6, MAKE_VOLATILE_STRIDE(8, rs)) { Chris@10: E T3, Ti, T6, Tm, Tc, Tn, Tf, Tj; Chris@10: { Chris@10: E T1, T2, T4, T5; Chris@10: T1 = cr[0]; Chris@10: T2 = ci[WS(rs, 1)]; Chris@10: T3 = T1 + T2; Chris@10: Ti = T1 - T2; Chris@10: T4 = cr[WS(rs, 1)]; Chris@10: T5 = ci[0]; Chris@10: T6 = T4 + T5; Chris@10: Tm = T4 - T5; Chris@10: } Chris@10: { Chris@10: E Ta, Tb, Td, Te; Chris@10: Ta = ci[WS(rs, 3)]; Chris@10: Tb = cr[WS(rs, 2)]; Chris@10: Tc = Ta - Tb; Chris@10: Tn = Ta + Tb; Chris@10: Td = ci[WS(rs, 2)]; Chris@10: Te = cr[WS(rs, 3)]; Chris@10: Tf = Td - Te; Chris@10: Tj = Td + Te; Chris@10: } Chris@10: cr[0] = T3 + T6; Chris@10: ci[0] = Tc + Tf; Chris@10: { Chris@10: E T8, Tg, T7, T9; Chris@10: T8 = T3 - T6; Chris@10: Tg = Tc - Tf; Chris@10: T7 = W[2]; Chris@10: T9 = W[3]; Chris@10: cr[WS(rs, 2)] = FNMS(T9, Tg, T7 * T8); Chris@10: ci[WS(rs, 2)] = FMA(T9, T8, T7 * Tg); Chris@10: } Chris@10: { Chris@10: E Tk, To, Th, Tl; Chris@10: Tk = Ti - Tj; Chris@10: To = Tm + Tn; Chris@10: Th = W[0]; Chris@10: Tl = W[1]; Chris@10: cr[WS(rs, 1)] = FNMS(Tl, To, Th * Tk); Chris@10: ci[WS(rs, 1)] = FMA(Th, To, Tl * Tk); Chris@10: } Chris@10: { Chris@10: E Tq, Ts, Tp, Tr; Chris@10: Tq = Ti + Tj; Chris@10: Ts = Tn - Tm; Chris@10: Tp = W[4]; Chris@10: Tr = W[5]; Chris@10: cr[WS(rs, 3)] = FNMS(Tr, Ts, Tp * Tq); Chris@10: ci[WS(rs, 3)] = FMA(Tp, Ts, Tr * Tq); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 4}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2hc_desc desc = { 4, "hb_4", twinstr, &GENUS, {16, 6, 6, 0} }; Chris@10: Chris@10: void X(codelet_hb_4) (planner *p) { Chris@10: X(khc2hc_register) (p, hb_4, &desc); Chris@10: } Chris@10: #endif /* HAVE_FMA */