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 07:41:12 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 -sign 1 -n 3 -dif -name hb_3 -include hb.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: * 27 stack variables, 2 constants, and 12 memory accesses cannam@95: */ cannam@95: #include "hb.h" cannam@95: cannam@95: static void hb_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 Tk, Tj, Tn, Tl, Tm, To; cannam@95: { cannam@95: E T1, Td, T7, T8, T4, Tg, T2, T3; cannam@95: T1 = cr[0]; cannam@95: T2 = cr[WS(rs, 1)]; cannam@95: T3 = ci[0]; cannam@95: Td = ci[WS(rs, 2)]; cannam@95: T7 = ci[WS(rs, 1)]; cannam@95: T8 = cr[WS(rs, 2)]; cannam@95: T4 = T2 + T3; cannam@95: Tg = T2 - T3; cannam@95: { cannam@95: E T5, Tc, Tf, Ta, T9, Te, T6, Th, Ti, Tb; cannam@95: T5 = W[0]; cannam@95: T9 = T7 + T8; cannam@95: Te = T7 - T8; cannam@95: cr[0] = T1 + T4; cannam@95: T6 = FNMS(KP500000000, T4, T1); cannam@95: Tc = W[1]; cannam@95: ci[0] = Td + Te; cannam@95: Tf = FNMS(KP500000000, Te, Td); cannam@95: Tk = FMA(KP866025403, T9, T6); cannam@95: Ta = FNMS(KP866025403, T9, T6); cannam@95: Tj = W[2]; cannam@95: Tn = FNMS(KP866025403, Tg, Tf); cannam@95: Th = FMA(KP866025403, Tg, Tf); cannam@95: Ti = Tc * Ta; cannam@95: Tb = T5 * Ta; cannam@95: Tl = Tj * Tk; cannam@95: Tm = W[3]; cannam@95: ci[WS(rs, 1)] = FMA(T5, Th, Ti); cannam@95: cr[WS(rs, 1)] = FNMS(Tc, Th, Tb); cannam@95: } cannam@95: } cannam@95: cr[WS(rs, 2)] = FNMS(Tm, Tn, Tl); cannam@95: To = Tm * Tk; cannam@95: ci[WS(rs, 2)] = FMA(Tj, Tn, To); 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, "hb_3", twinstr, &GENUS, {6, 4, 10, 0} }; cannam@95: cannam@95: void X(codelet_hb_3) (planner *p) { cannam@95: X(khc2hc_register) (p, hb_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 -sign 1 -n 3 -dif -name hb_3 -include hb.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 "hb.h" cannam@95: cannam@95: static void hb_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, T4, Ta, Te, T5, T8, Tb, Tf; cannam@95: { cannam@95: E T2, T3, T6, T7; cannam@95: T1 = cr[0]; cannam@95: T2 = cr[WS(rs, 1)]; cannam@95: T3 = ci[0]; cannam@95: T4 = T2 + T3; cannam@95: Ta = FNMS(KP500000000, T4, T1); cannam@95: Te = KP866025403 * (T2 - T3); cannam@95: T5 = ci[WS(rs, 2)]; cannam@95: T6 = ci[WS(rs, 1)]; cannam@95: T7 = cr[WS(rs, 2)]; cannam@95: T8 = T6 - T7; cannam@95: Tb = KP866025403 * (T6 + T7); cannam@95: Tf = FNMS(KP500000000, T8, T5); cannam@95: } cannam@95: cr[0] = T1 + T4; cannam@95: ci[0] = T5 + T8; cannam@95: { cannam@95: E Tc, Tg, T9, Td; cannam@95: Tc = Ta - Tb; cannam@95: Tg = Te + Tf; cannam@95: T9 = W[0]; cannam@95: Td = W[1]; cannam@95: cr[WS(rs, 1)] = FNMS(Td, Tg, T9 * Tc); cannam@95: ci[WS(rs, 1)] = FMA(T9, Tg, Td * Tc); cannam@95: } cannam@95: { cannam@95: E Ti, Tk, Th, Tj; cannam@95: Ti = Ta + Tb; cannam@95: Tk = Tf - Te; cannam@95: Th = W[2]; cannam@95: Tj = W[3]; cannam@95: cr[WS(rs, 2)] = FNMS(Tj, Tk, Th * Ti); cannam@95: ci[WS(rs, 2)] = FMA(Th, Tk, Tj * Ti); 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, "hb_3", twinstr, &GENUS, {10, 6, 6, 0} }; cannam@95: cannam@95: void X(codelet_hb_3) (planner *p) { cannam@95: X(khc2hc_register) (p, hb_3, &desc); cannam@95: } cannam@95: #endif /* HAVE_FMA */