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 07:39:49 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 3 -dit -name hf_3 -include hf.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 16 FP additions, 14 FP multiplications, cannam@95: * (or, 6 additions, 4 multiplications, 10 fused multiply/add), cannam@95: * 21 stack variables, 2 constants, and 12 memory accesses cannam@95: */ cannam@95: #include "hf.h" cannam@95: cannam@95: static void hf_3(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@95: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: { cannam@95: INT m; cannam@95: 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)) { cannam@95: E T1, Tl, T9, Tc, Tb, Th, T7, Ti, Ta, Tj, Td; cannam@95: T1 = cr[0]; cannam@95: Tl = ci[0]; cannam@95: { cannam@95: E T3, T6, T2, T5, Tg, T4, T8; cannam@95: T3 = cr[WS(rs, 1)]; cannam@95: T6 = ci[WS(rs, 1)]; cannam@95: T2 = W[0]; cannam@95: T5 = W[1]; cannam@95: T9 = cr[WS(rs, 2)]; cannam@95: Tc = ci[WS(rs, 2)]; cannam@95: Tg = T2 * T6; cannam@95: T4 = T2 * T3; cannam@95: T8 = W[2]; cannam@95: Tb = W[3]; cannam@95: Th = FNMS(T5, T3, Tg); cannam@95: T7 = FMA(T5, T6, T4); cannam@95: Ti = T8 * Tc; cannam@95: Ta = T8 * T9; cannam@95: } cannam@95: Tj = FNMS(Tb, T9, Ti); cannam@95: Td = FMA(Tb, Tc, Ta); cannam@95: { cannam@95: E Tk, Te, To, Tn, Tm, Tf; cannam@95: Tk = Th - Tj; cannam@95: Tm = Th + Tj; cannam@95: Te = T7 + Td; cannam@95: To = Td - T7; cannam@95: ci[WS(rs, 2)] = Tm + Tl; cannam@95: Tn = FNMS(KP500000000, Tm, Tl); cannam@95: cr[0] = T1 + Te; cannam@95: Tf = FNMS(KP500000000, Te, T1); cannam@95: ci[WS(rs, 1)] = FMA(KP866025403, To, Tn); cannam@95: cr[WS(rs, 2)] = FMS(KP866025403, To, Tn); cannam@95: cr[WS(rs, 1)] = FMA(KP866025403, Tk, Tf); cannam@95: ci[0] = FNMS(KP866025403, Tk, Tf); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_FULL, 1, 3}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2hc_desc desc = { 3, "hf_3", twinstr, &GENUS, {6, 4, 10, 0} }; cannam@95: cannam@95: void X(codelet_hf_3) (planner *p) { cannam@95: X(khc2hc_register) (p, hf_3, &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 3 -dit -name hf_3 -include hf.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 16 FP additions, 12 FP multiplications, cannam@95: * (or, 10 additions, 6 multiplications, 6 fused multiply/add), cannam@95: * 15 stack variables, 2 constants, and 12 memory accesses cannam@95: */ cannam@95: #include "hf.h" cannam@95: cannam@95: static void hf_3(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@95: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: { cannam@95: INT m; cannam@95: 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)) { cannam@95: E T1, Ti, T6, Te, Tb, Tf, Tc, Tj; cannam@95: T1 = cr[0]; cannam@95: Ti = ci[0]; cannam@95: { cannam@95: E T3, T5, T2, T4; cannam@95: T3 = cr[WS(rs, 1)]; cannam@95: T5 = ci[WS(rs, 1)]; cannam@95: T2 = W[0]; cannam@95: T4 = W[1]; cannam@95: T6 = FMA(T2, T3, T4 * T5); cannam@95: Te = FNMS(T4, T3, T2 * T5); cannam@95: } cannam@95: { cannam@95: E T8, Ta, T7, T9; cannam@95: T8 = cr[WS(rs, 2)]; cannam@95: Ta = ci[WS(rs, 2)]; cannam@95: T7 = W[2]; cannam@95: T9 = W[3]; cannam@95: Tb = FMA(T7, T8, T9 * Ta); cannam@95: Tf = FNMS(T9, T8, T7 * Ta); cannam@95: } cannam@95: Tc = T6 + Tb; cannam@95: Tj = Te + Tf; cannam@95: { cannam@95: E Td, Tg, Th, Tk; cannam@95: cr[0] = T1 + Tc; cannam@95: Td = FNMS(KP500000000, Tc, T1); cannam@95: Tg = KP866025403 * (Te - Tf); cannam@95: ci[0] = Td - Tg; cannam@95: cr[WS(rs, 1)] = Td + Tg; cannam@95: ci[WS(rs, 2)] = Tj + Ti; cannam@95: Th = KP866025403 * (Tb - T6); cannam@95: Tk = FNMS(KP500000000, Tj, Ti); cannam@95: cr[WS(rs, 2)] = Th - Tk; cannam@95: ci[WS(rs, 1)] = Th + Tk; cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_FULL, 1, 3}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2hc_desc desc = { 3, "hf_3", twinstr, &GENUS, {10, 6, 6, 0} }; cannam@95: cannam@95: void X(codelet_hf_3) (planner *p) { cannam@95: X(khc2hc_register) (p, hf_3, &desc); cannam@95: } cannam@95: #endif /* HAVE_FMA */