cannam@127: /* cannam@127: * Copyright (c) 2003, 2007-14 Matteo Frigo cannam@127: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology cannam@127: * cannam@127: * This program is free software; you can redistribute it and/or modify cannam@127: * it under the terms of the GNU General Public License as published by cannam@127: * the Free Software Foundation; either version 2 of the License, or cannam@127: * (at your option) any later version. cannam@127: * cannam@127: * This program is distributed in the hope that it will be useful, cannam@127: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@127: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@127: * GNU General Public License for more details. cannam@127: * cannam@127: * You should have received a copy of the GNU General Public License cannam@127: * along with this program; if not, write to the Free Software cannam@127: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@127: * cannam@127: */ cannam@127: cannam@127: /* This file was automatically generated --- DO NOT EDIT */ cannam@127: /* Generated on Sat Jul 30 16:49:40 EDT 2016 */ cannam@127: cannam@127: #include "codelet-rdft.h" cannam@127: cannam@127: #ifdef HAVE_FMA cannam@127: cannam@127: /* 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 */ cannam@127: cannam@127: /* cannam@127: * This function contains 22 FP additions, 12 FP multiplications, cannam@127: * (or, 16 additions, 6 multiplications, 6 fused multiply/add), cannam@127: * 27 stack variables, 0 constants, and 16 memory accesses cannam@127: */ cannam@127: #include "hb.h" cannam@127: cannam@127: static void hb_4(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) cannam@127: { cannam@127: { cannam@127: INT m; cannam@127: 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@127: E T8, Th, Ta, T7, Ti, T9; cannam@127: { cannam@127: E Td, Tg, T3, T6, Tu, Tm, Tx, Tr; cannam@127: { cannam@127: E Tq, Tl, T4, T5, Tk, Tp; cannam@127: { cannam@127: E Tb, Tc, Te, Tf, T1, T2; cannam@127: Tb = ci[WS(rs, 3)]; cannam@127: Tc = cr[WS(rs, 2)]; cannam@127: Te = ci[WS(rs, 2)]; cannam@127: Tf = cr[WS(rs, 3)]; cannam@127: T1 = cr[0]; cannam@127: Tq = Tb + Tc; cannam@127: Td = Tb - Tc; cannam@127: T2 = ci[WS(rs, 1)]; cannam@127: Tl = Te + Tf; cannam@127: Tg = Te - Tf; cannam@127: T4 = cr[WS(rs, 1)]; cannam@127: T5 = ci[0]; cannam@127: T3 = T1 + T2; cannam@127: Tk = T1 - T2; cannam@127: } cannam@127: Tp = T4 - T5; cannam@127: T6 = T4 + T5; cannam@127: Tu = Tk + Tl; cannam@127: Tm = Tk - Tl; cannam@127: Tx = Tq - Tp; cannam@127: Tr = Tp + Tq; cannam@127: T8 = T3 - T6; cannam@127: } cannam@127: cr[0] = T3 + T6; cannam@127: { cannam@127: E Tj, To, Tw, Tv; cannam@127: Tj = W[0]; cannam@127: ci[0] = Td + Tg; cannam@127: To = W[1]; cannam@127: { cannam@127: E Tt, Ts, Tn, Ty; cannam@127: Tt = W[4]; cannam@127: Ts = Tj * Tr; cannam@127: Tn = Tj * Tm; cannam@127: Tw = W[5]; cannam@127: Ty = Tt * Tx; cannam@127: Tv = Tt * Tu; cannam@127: ci[WS(rs, 1)] = FMA(To, Tm, Ts); cannam@127: cr[WS(rs, 1)] = FNMS(To, Tr, Tn); cannam@127: ci[WS(rs, 3)] = FMA(Tw, Tu, Ty); cannam@127: } cannam@127: cr[WS(rs, 3)] = FNMS(Tw, Tx, Tv); cannam@127: Th = Td - Tg; cannam@127: Ta = W[3]; cannam@127: T7 = W[2]; cannam@127: } cannam@127: } cannam@127: Ti = Ta * T8; cannam@127: T9 = T7 * T8; cannam@127: ci[WS(rs, 2)] = FMA(T7, Th, Ti); cannam@127: cr[WS(rs, 2)] = FNMS(Ta, Th, T9); cannam@127: } cannam@127: } cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: {TW_FULL, 1, 4}, cannam@127: {TW_NEXT, 1, 0} cannam@127: }; cannam@127: cannam@127: static const hc2hc_desc desc = { 4, "hb_4", twinstr, &GENUS, {16, 6, 6, 0} }; cannam@127: cannam@127: void X(codelet_hb_4) (planner *p) { cannam@127: X(khc2hc_register) (p, hb_4, &desc); cannam@127: } cannam@127: #else /* HAVE_FMA */ cannam@127: cannam@127: /* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 4 -dif -name hb_4 -include hb.h */ cannam@127: cannam@127: /* cannam@127: * This function contains 22 FP additions, 12 FP multiplications, cannam@127: * (or, 16 additions, 6 multiplications, 6 fused multiply/add), cannam@127: * 13 stack variables, 0 constants, and 16 memory accesses cannam@127: */ cannam@127: #include "hb.h" cannam@127: cannam@127: static void hb_4(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) cannam@127: { cannam@127: { cannam@127: INT m; cannam@127: 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@127: E T3, Ti, T6, Tm, Tc, Tn, Tf, Tj; cannam@127: { cannam@127: E T1, T2, T4, T5; cannam@127: T1 = cr[0]; cannam@127: T2 = ci[WS(rs, 1)]; cannam@127: T3 = T1 + T2; cannam@127: Ti = T1 - T2; cannam@127: T4 = cr[WS(rs, 1)]; cannam@127: T5 = ci[0]; cannam@127: T6 = T4 + T5; cannam@127: Tm = T4 - T5; cannam@127: } cannam@127: { cannam@127: E Ta, Tb, Td, Te; cannam@127: Ta = ci[WS(rs, 3)]; cannam@127: Tb = cr[WS(rs, 2)]; cannam@127: Tc = Ta - Tb; cannam@127: Tn = Ta + Tb; cannam@127: Td = ci[WS(rs, 2)]; cannam@127: Te = cr[WS(rs, 3)]; cannam@127: Tf = Td - Te; cannam@127: Tj = Td + Te; cannam@127: } cannam@127: cr[0] = T3 + T6; cannam@127: ci[0] = Tc + Tf; cannam@127: { cannam@127: E T8, Tg, T7, T9; cannam@127: T8 = T3 - T6; cannam@127: Tg = Tc - Tf; cannam@127: T7 = W[2]; cannam@127: T9 = W[3]; cannam@127: cr[WS(rs, 2)] = FNMS(T9, Tg, T7 * T8); cannam@127: ci[WS(rs, 2)] = FMA(T9, T8, T7 * Tg); cannam@127: } cannam@127: { cannam@127: E Tk, To, Th, Tl; cannam@127: Tk = Ti - Tj; cannam@127: To = Tm + Tn; cannam@127: Th = W[0]; cannam@127: Tl = W[1]; cannam@127: cr[WS(rs, 1)] = FNMS(Tl, To, Th * Tk); cannam@127: ci[WS(rs, 1)] = FMA(Th, To, Tl * Tk); cannam@127: } cannam@127: { cannam@127: E Tq, Ts, Tp, Tr; cannam@127: Tq = Ti + Tj; cannam@127: Ts = Tn - Tm; cannam@127: Tp = W[4]; cannam@127: Tr = W[5]; cannam@127: cr[WS(rs, 3)] = FNMS(Tr, Ts, Tp * Tq); cannam@127: ci[WS(rs, 3)] = FMA(Tp, Ts, Tr * Tq); cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: {TW_FULL, 1, 4}, cannam@127: {TW_NEXT, 1, 0} cannam@127: }; cannam@127: cannam@127: static const hc2hc_desc desc = { 4, "hb_4", twinstr, &GENUS, {16, 6, 6, 0} }; cannam@127: cannam@127: void X(codelet_hb_4) (planner *p) { cannam@127: X(khc2hc_register) (p, hb_4, &desc); cannam@127: } cannam@127: #endif /* HAVE_FMA */