cannam@95: /* cannam@95: * Copyright (c) 2003, 2007-11 Matteo Frigo cannam@95: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology cannam@95: * cannam@95: * This program is free software; you can redistribute it and/or modify cannam@95: * it under the terms of the GNU General Public License as published by cannam@95: * the Free Software Foundation; either version 2 of the License, or cannam@95: * (at your option) any later version. cannam@95: * cannam@95: * This program is distributed in the hope that it will be useful, cannam@95: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@95: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@95: * GNU General Public License for more details. cannam@95: * cannam@95: * You should have received a copy of the GNU General Public License cannam@95: * along with this program; if not, write to the Free Software cannam@95: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@95: * cannam@95: */ cannam@95: cannam@95: /* This file was automatically generated --- DO NOT EDIT */ cannam@95: /* Generated on Sun Nov 25 07:39:50 EST 2012 */ cannam@95: cannam@95: #include "codelet-rdft.h" cannam@95: cannam@95: #ifdef HAVE_FMA cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_hc2hc.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 4 -dit -name hf_4 -include hf.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 22 FP additions, 12 FP multiplications, cannam@95: * (or, 16 additions, 6 multiplications, 6 fused multiply/add), cannam@95: * 31 stack variables, 0 constants, and 16 memory accesses cannam@95: */ cannam@95: #include "hf.h" cannam@95: cannam@95: static void hf_4(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: { cannam@95: INT m; cannam@95: 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)) { cannam@95: E To, Te, Tm, T8, Ty, Tw, Tq, Tk; cannam@95: { cannam@95: E T1, Tv, Tu, T7, Tg, Tj, Tf, Ti, Tp, Th; cannam@95: T1 = cr[0]; cannam@95: Tv = ci[0]; cannam@95: { cannam@95: E T3, T6, T2, T5; cannam@95: T3 = cr[WS(rs, 2)]; cannam@95: T6 = ci[WS(rs, 2)]; cannam@95: T2 = W[2]; cannam@95: T5 = W[3]; cannam@95: { cannam@95: E Ta, Td, Tc, Tn, Tb, Tt, T4, T9; cannam@95: Ta = cr[WS(rs, 1)]; cannam@95: Td = ci[WS(rs, 1)]; cannam@95: Tt = T2 * T6; cannam@95: T4 = T2 * T3; cannam@95: T9 = W[0]; cannam@95: Tc = W[1]; cannam@95: Tu = FNMS(T5, T3, Tt); cannam@95: T7 = FMA(T5, T6, T4); cannam@95: Tn = T9 * Td; cannam@95: Tb = T9 * Ta; cannam@95: Tg = cr[WS(rs, 3)]; cannam@95: Tj = ci[WS(rs, 3)]; cannam@95: To = FNMS(Tc, Ta, Tn); cannam@95: Te = FMA(Tc, Td, Tb); cannam@95: Tf = W[4]; cannam@95: Ti = W[5]; cannam@95: } cannam@95: } cannam@95: Tm = T1 - T7; cannam@95: T8 = T1 + T7; cannam@95: Tp = Tf * Tj; cannam@95: Th = Tf * Tg; cannam@95: Ty = Tv - Tu; cannam@95: Tw = Tu + Tv; cannam@95: Tq = FNMS(Ti, Tg, Tp); cannam@95: Tk = FMA(Ti, Tj, Th); cannam@95: } cannam@95: { cannam@95: E Tr, Ts, Tl, Tx; cannam@95: Tr = To - Tq; cannam@95: Ts = To + Tq; cannam@95: Tl = Te + Tk; cannam@95: Tx = Tk - Te; cannam@95: ci[WS(rs, 3)] = Ts + Tw; cannam@95: cr[WS(rs, 2)] = Ts - Tw; cannam@95: cr[WS(rs, 1)] = Tm + Tr; cannam@95: ci[0] = Tm - Tr; cannam@95: ci[WS(rs, 2)] = Tx + Ty; cannam@95: cr[WS(rs, 3)] = Tx - Ty; cannam@95: cr[0] = T8 + Tl; cannam@95: ci[WS(rs, 1)] = T8 - Tl; cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_FULL, 1, 4}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2hc_desc desc = { 4, "hf_4", twinstr, &GENUS, {16, 6, 6, 0} }; cannam@95: cannam@95: void X(codelet_hf_4) (planner *p) { cannam@95: X(khc2hc_register) (p, hf_4, &desc); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -n 4 -dit -name hf_4 -include hf.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 22 FP additions, 12 FP multiplications, cannam@95: * (or, 16 additions, 6 multiplications, 6 fused multiply/add), cannam@95: * 13 stack variables, 0 constants, and 16 memory accesses cannam@95: */ cannam@95: #include "hf.h" cannam@95: cannam@95: static void hf_4(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: { cannam@95: INT m; cannam@95: 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)) { cannam@95: E T1, Tp, T6, To, Tc, Tk, Th, Tl; cannam@95: T1 = cr[0]; cannam@95: Tp = ci[0]; cannam@95: { cannam@95: E T3, T5, T2, T4; cannam@95: T3 = cr[WS(rs, 2)]; cannam@95: T5 = ci[WS(rs, 2)]; cannam@95: T2 = W[2]; cannam@95: T4 = W[3]; cannam@95: T6 = FMA(T2, T3, T4 * T5); cannam@95: To = FNMS(T4, T3, T2 * T5); cannam@95: } cannam@95: { cannam@95: E T9, Tb, T8, Ta; cannam@95: T9 = cr[WS(rs, 1)]; cannam@95: Tb = ci[WS(rs, 1)]; cannam@95: T8 = W[0]; cannam@95: Ta = W[1]; cannam@95: Tc = FMA(T8, T9, Ta * Tb); cannam@95: Tk = FNMS(Ta, T9, T8 * Tb); cannam@95: } cannam@95: { cannam@95: E Te, Tg, Td, Tf; cannam@95: Te = cr[WS(rs, 3)]; cannam@95: Tg = ci[WS(rs, 3)]; cannam@95: Td = W[4]; cannam@95: Tf = W[5]; cannam@95: Th = FMA(Td, Te, Tf * Tg); cannam@95: Tl = FNMS(Tf, Te, Td * Tg); cannam@95: } cannam@95: { cannam@95: E T7, Ti, Tj, Tm; cannam@95: T7 = T1 + T6; cannam@95: Ti = Tc + Th; cannam@95: ci[WS(rs, 1)] = T7 - Ti; cannam@95: cr[0] = T7 + Ti; cannam@95: Tj = T1 - T6; cannam@95: Tm = Tk - Tl; cannam@95: ci[0] = Tj - Tm; cannam@95: cr[WS(rs, 1)] = Tj + Tm; cannam@95: } cannam@95: { cannam@95: E Tn, Tq, Tr, Ts; cannam@95: Tn = Tk + Tl; cannam@95: Tq = To + Tp; cannam@95: cr[WS(rs, 2)] = Tn - Tq; cannam@95: ci[WS(rs, 3)] = Tn + Tq; cannam@95: Tr = Th - Tc; cannam@95: Ts = Tp - To; cannam@95: cr[WS(rs, 3)] = Tr - Ts; cannam@95: ci[WS(rs, 2)] = Tr + Ts; cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_FULL, 1, 4}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2hc_desc desc = { 4, "hf_4", twinstr, &GENUS, {16, 6, 6, 0} }; cannam@95: cannam@95: void X(codelet_hf_4) (planner *p) { cannam@95: X(khc2hc_register) (p, hf_4, &desc); cannam@95: } cannam@95: #endif /* HAVE_FMA */