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 08:04:12 EDT 2018 */ Chris@82: Chris@82: #include "dft/codelet-dft.h" Chris@82: Chris@82: #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA) Chris@82: Chris@82: /* Generated by: ../../../genfft/gen_twiddle.native -fma -compact -variables 4 -pipeline-latency 4 -n 3 -name t1_3 -include dft/scalar/t.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: * 15 stack variables, 2 constants, and 12 memory accesses Chris@82: */ Chris@82: #include "dft/scalar/t.h" Chris@82: Chris@82: static void t1_3(R *ri, R *ii, 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 * 4); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 4, MAKE_VOLATILE_STRIDE(6, rs)) { Chris@82: E T1, Tm, T7, Th, Td, Tj; Chris@82: T1 = ri[0]; Chris@82: Tm = ii[0]; Chris@82: { Chris@82: E T3, T6, T4, Tg, T2, T5; Chris@82: T3 = ri[WS(rs, 1)]; Chris@82: T6 = ii[WS(rs, 1)]; Chris@82: T2 = W[0]; Chris@82: T4 = T2 * T3; Chris@82: Tg = T2 * T6; Chris@82: T5 = W[1]; Chris@82: T7 = FMA(T5, T6, T4); Chris@82: Th = FNMS(T5, T3, Tg); Chris@82: } Chris@82: { Chris@82: E T9, Tc, Ta, Ti, T8, Tb; Chris@82: T9 = ri[WS(rs, 2)]; Chris@82: Tc = ii[WS(rs, 2)]; Chris@82: T8 = W[2]; Chris@82: Ta = T8 * T9; Chris@82: Ti = T8 * Tc; Chris@82: Tb = W[3]; Chris@82: Td = FMA(Tb, Tc, Ta); Chris@82: Tj = FNMS(Tb, T9, Ti); Chris@82: } Chris@82: { Chris@82: E Tk, Te, Tf, To, Tl, Tn; Chris@82: Tk = Th - Tj; Chris@82: Te = T7 + Td; Chris@82: Tf = FNMS(KP500000000, Te, T1); Chris@82: ri[0] = T1 + Te; Chris@82: ri[WS(rs, 1)] = FMA(KP866025403, Tk, Tf); Chris@82: ri[WS(rs, 2)] = FNMS(KP866025403, Tk, Tf); Chris@82: To = Td - T7; Chris@82: Tl = Th + Tj; Chris@82: Tn = FNMS(KP500000000, Tl, Tm); Chris@82: ii[0] = Tl + Tm; Chris@82: ii[WS(rs, 2)] = FNMS(KP866025403, To, Tn); Chris@82: ii[WS(rs, 1)] = FMA(KP866025403, To, Tn); Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: Chris@82: static const tw_instr twinstr[] = { Chris@82: {TW_FULL, 0, 3}, Chris@82: {TW_NEXT, 1, 0} Chris@82: }; Chris@82: Chris@82: static const ct_desc desc = { 3, "t1_3", twinstr, &GENUS, {6, 4, 10, 0}, 0, 0, 0 }; Chris@82: Chris@82: void X(codelet_t1_3) (planner *p) { Chris@82: X(kdft_dit_register) (p, t1_3, &desc); Chris@82: } Chris@82: #else Chris@82: Chris@82: /* Generated by: ../../../genfft/gen_twiddle.native -compact -variables 4 -pipeline-latency 4 -n 3 -name t1_3 -include dft/scalar/t.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 "dft/scalar/t.h" Chris@82: Chris@82: static void t1_3(R *ri, R *ii, 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 * 4); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 4, MAKE_VOLATILE_STRIDE(6, rs)) { Chris@82: E T1, Ti, T6, Te, Tb, Tf, Tc, Th; Chris@82: T1 = ri[0]; Chris@82: Ti = ii[0]; Chris@82: { Chris@82: E T3, T5, T2, T4; Chris@82: T3 = ri[WS(rs, 1)]; Chris@82: T5 = ii[WS(rs, 1)]; Chris@82: T2 = W[0]; Chris@82: T4 = W[1]; Chris@82: T6 = FMA(T2, T3, T4 * T5); Chris@82: Te = FNMS(T4, T3, T2 * T5); Chris@82: } Chris@82: { Chris@82: E T8, Ta, T7, T9; Chris@82: T8 = ri[WS(rs, 2)]; Chris@82: Ta = ii[WS(rs, 2)]; Chris@82: T7 = W[2]; Chris@82: T9 = W[3]; Chris@82: Tb = FMA(T7, T8, T9 * Ta); Chris@82: Tf = FNMS(T9, T8, T7 * Ta); Chris@82: } Chris@82: Tc = T6 + Tb; Chris@82: Th = Te + Tf; Chris@82: ri[0] = T1 + Tc; Chris@82: ii[0] = Th + Ti; Chris@82: { Chris@82: E Td, Tg, Tj, Tk; Chris@82: Td = FNMS(KP500000000, Tc, T1); Chris@82: Tg = KP866025403 * (Te - Tf); Chris@82: ri[WS(rs, 2)] = Td - Tg; Chris@82: ri[WS(rs, 1)] = Td + Tg; Chris@82: Tj = KP866025403 * (Tb - T6); Chris@82: Tk = FNMS(KP500000000, Th, Ti); Chris@82: ii[WS(rs, 1)] = Tj + Tk; Chris@82: ii[WS(rs, 2)] = Tk - Tj; Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: Chris@82: static const tw_instr twinstr[] = { Chris@82: {TW_FULL, 0, 3}, Chris@82: {TW_NEXT, 1, 0} Chris@82: }; Chris@82: Chris@82: static const ct_desc desc = { 3, "t1_3", twinstr, &GENUS, {10, 6, 6, 0}, 0, 0, 0 }; Chris@82: Chris@82: void X(codelet_t1_3) (planner *p) { Chris@82: X(kdft_dit_register) (p, t1_3, &desc); Chris@82: } Chris@82: #endif