Chris@82: /* Chris@82: * Copyright (c) 2003, 2007-14 Matteo Frigo Chris@82: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology Chris@82: * Chris@82: * This program is free software; you can redistribute it and/or modify Chris@82: * it under the terms of the GNU General Public License as published by Chris@82: * the Free Software Foundation; either version 2 of the License, or Chris@82: * (at your option) any later version. Chris@82: * Chris@82: * This program is distributed in the hope that it will be useful, Chris@82: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@82: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@82: * GNU General Public License for more details. Chris@82: * Chris@82: * You should have received a copy of the GNU General Public License Chris@82: * along with this program; if not, write to the Free Software Chris@82: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@82: * Chris@82: */ Chris@82: Chris@82: /* This file was automatically generated --- DO NOT EDIT */ Chris@82: /* Generated on Thu May 24 08:07:31 EDT 2018 */ Chris@82: Chris@82: #include "rdft/codelet-rdft.h" Chris@82: Chris@82: #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA) Chris@82: Chris@82: /* Generated by: ../../../genfft/gen_hc2hc.native -fma -compact -variables 4 -pipeline-latency 4 -sign 1 -n 3 -dif -name hb_3 -include rdft/scalar/hb.h */ Chris@82: Chris@82: /* Chris@82: * This function contains 16 FP additions, 14 FP multiplications, Chris@82: * (or, 6 additions, 4 multiplications, 10 fused multiply/add), Chris@82: * 17 stack variables, 2 constants, and 12 memory accesses Chris@82: */ Chris@82: #include "rdft/scalar/hb.h" Chris@82: Chris@82: static void hb_3(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) Chris@82: { Chris@82: DK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@82: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@82: { Chris@82: INT m; Chris@82: 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)) { Chris@82: E T1, T4, T6, Tg, Td, Te, T9, Tf; Chris@82: { Chris@82: E T2, T3, T7, T8; Chris@82: T1 = cr[0]; Chris@82: T2 = cr[WS(rs, 1)]; Chris@82: T3 = ci[0]; Chris@82: T4 = T2 + T3; Chris@82: T6 = FNMS(KP500000000, T4, T1); Chris@82: Tg = T2 - T3; Chris@82: Td = ci[WS(rs, 2)]; Chris@82: T7 = ci[WS(rs, 1)]; Chris@82: T8 = cr[WS(rs, 2)]; Chris@82: Te = T7 - T8; Chris@82: T9 = T7 + T8; Chris@82: Tf = FNMS(KP500000000, Te, Td); Chris@82: } Chris@82: cr[0] = T1 + T4; Chris@82: ci[0] = Td + Te; Chris@82: { Chris@82: E Th, T5, Tb, Tc, Ti, Ta; Chris@82: Th = FMA(KP866025403, Tg, Tf); Chris@82: Ta = FNMS(KP866025403, T9, T6); Chris@82: T5 = W[0]; Chris@82: Tb = T5 * Ta; Chris@82: Tc = W[1]; Chris@82: Ti = Tc * Ta; Chris@82: cr[WS(rs, 1)] = FNMS(Tc, Th, Tb); Chris@82: ci[WS(rs, 1)] = FMA(T5, Th, Ti); Chris@82: } Chris@82: { Chris@82: E Tn, Tj, Tl, Tm, To, Tk; Chris@82: Tn = FNMS(KP866025403, Tg, Tf); Chris@82: Tk = FMA(KP866025403, T9, T6); Chris@82: Tj = W[2]; Chris@82: Tl = Tj * Tk; Chris@82: Tm = W[3]; Chris@82: To = Tm * Tk; Chris@82: cr[WS(rs, 2)] = FNMS(Tm, Tn, Tl); Chris@82: ci[WS(rs, 2)] = FMA(Tj, Tn, To); Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: Chris@82: static const tw_instr twinstr[] = { Chris@82: {TW_FULL, 1, 3}, Chris@82: {TW_NEXT, 1, 0} Chris@82: }; Chris@82: Chris@82: static const hc2hc_desc desc = { 3, "hb_3", twinstr, &GENUS, {6, 4, 10, 0} }; Chris@82: Chris@82: void X(codelet_hb_3) (planner *p) { Chris@82: X(khc2hc_register) (p, hb_3, &desc); Chris@82: } Chris@82: #else Chris@82: Chris@82: /* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 3 -dif -name hb_3 -include rdft/scalar/hb.h */ Chris@82: Chris@82: /* Chris@82: * This function contains 16 FP additions, 12 FP multiplications, Chris@82: * (or, 10 additions, 6 multiplications, 6 fused multiply/add), Chris@82: * 15 stack variables, 2 constants, and 12 memory accesses Chris@82: */ Chris@82: #include "rdft/scalar/hb.h" Chris@82: Chris@82: static void hb_3(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) Chris@82: { Chris@82: DK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@82: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@82: { Chris@82: INT m; Chris@82: 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)) { Chris@82: E T1, T4, Ta, Te, T5, T8, Tb, Tf; Chris@82: { Chris@82: E T2, T3, T6, T7; Chris@82: T1 = cr[0]; Chris@82: T2 = cr[WS(rs, 1)]; Chris@82: T3 = ci[0]; Chris@82: T4 = T2 + T3; Chris@82: Ta = FNMS(KP500000000, T4, T1); Chris@82: Te = KP866025403 * (T2 - T3); Chris@82: T5 = ci[WS(rs, 2)]; Chris@82: T6 = ci[WS(rs, 1)]; Chris@82: T7 = cr[WS(rs, 2)]; Chris@82: T8 = T6 - T7; Chris@82: Tb = KP866025403 * (T6 + T7); Chris@82: Tf = FNMS(KP500000000, T8, T5); Chris@82: } Chris@82: cr[0] = T1 + T4; Chris@82: ci[0] = T5 + T8; Chris@82: { Chris@82: E Tc, Tg, T9, Td; Chris@82: Tc = Ta - Tb; Chris@82: Tg = Te + Tf; Chris@82: T9 = W[0]; Chris@82: Td = W[1]; Chris@82: cr[WS(rs, 1)] = FNMS(Td, Tg, T9 * Tc); Chris@82: ci[WS(rs, 1)] = FMA(T9, Tg, Td * Tc); Chris@82: } Chris@82: { Chris@82: E Ti, Tk, Th, Tj; Chris@82: Ti = Ta + Tb; Chris@82: Tk = Tf - Te; Chris@82: Th = W[2]; Chris@82: Tj = W[3]; Chris@82: cr[WS(rs, 2)] = FNMS(Tj, Tk, Th * Ti); Chris@82: ci[WS(rs, 2)] = FMA(Th, Tk, Tj * Ti); Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: Chris@82: static const tw_instr twinstr[] = { Chris@82: {TW_FULL, 1, 3}, Chris@82: {TW_NEXT, 1, 0} Chris@82: }; Chris@82: Chris@82: static const hc2hc_desc desc = { 3, "hb_3", twinstr, &GENUS, {10, 6, 6, 0} }; Chris@82: Chris@82: void X(codelet_hb_3) (planner *p) { Chris@82: X(khc2hc_register) (p, hb_3, &desc); Chris@82: } Chris@82: #endif