cannam@167: /* cannam@167: * Copyright (c) 2003, 2007-14 Matteo Frigo cannam@167: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology cannam@167: * cannam@167: * This program is free software; you can redistribute it and/or modify cannam@167: * it under the terms of the GNU General Public License as published by cannam@167: * the Free Software Foundation; either version 2 of the License, or cannam@167: * (at your option) any later version. cannam@167: * cannam@167: * This program is distributed in the hope that it will be useful, cannam@167: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@167: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@167: * GNU General Public License for more details. cannam@167: * cannam@167: * You should have received a copy of the GNU General Public License cannam@167: * along with this program; if not, write to the Free Software cannam@167: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@167: * cannam@167: */ cannam@167: cannam@167: /* This file was automatically generated --- DO NOT EDIT */ cannam@167: /* Generated on Thu May 24 08:04:29 EDT 2018 */ cannam@167: cannam@167: #include "dft/codelet-dft.h" cannam@167: cannam@167: #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA) cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_twidsq.native -fma -compact -variables 4 -pipeline-latency 4 -reload-twiddle -dif -n 2 -name q1_2 -include dft/scalar/q.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 12 FP additions, 8 FP multiplications, cannam@167: * (or, 8 additions, 4 multiplications, 4 fused multiply/add), cannam@167: * 17 stack variables, 0 constants, and 16 memory accesses cannam@167: */ cannam@167: #include "dft/scalar/q.h" cannam@167: cannam@167: static void q1_2(R *rio, R *iio, const R *W, stride rs, stride vs, INT mb, INT me, INT ms) cannam@167: { cannam@167: { cannam@167: INT m; cannam@167: for (m = mb, W = W + (mb * 2); m < me; m = m + 1, rio = rio + ms, iio = iio + ms, W = W + 2, MAKE_VOLATILE_STRIDE(4, rs), MAKE_VOLATILE_STRIDE(0, vs)) { cannam@167: E T1, T2, T4, T7, T8, T9, Tb, Tc, Te, Th, Ti, Tj; cannam@167: T1 = rio[0]; cannam@167: T2 = rio[WS(rs, 1)]; cannam@167: T4 = T1 - T2; cannam@167: T7 = iio[0]; cannam@167: T8 = iio[WS(rs, 1)]; cannam@167: T9 = T7 - T8; cannam@167: Tb = rio[WS(vs, 1)]; cannam@167: Tc = rio[WS(vs, 1) + WS(rs, 1)]; cannam@167: Te = Tb - Tc; cannam@167: Th = iio[WS(vs, 1)]; cannam@167: Ti = iio[WS(vs, 1) + WS(rs, 1)]; cannam@167: Tj = Th - Ti; cannam@167: rio[0] = T1 + T2; cannam@167: iio[0] = T7 + T8; cannam@167: rio[WS(rs, 1)] = Tb + Tc; cannam@167: iio[WS(rs, 1)] = Th + Ti; cannam@167: { cannam@167: E Tf, Tk, Td, Tg; cannam@167: Td = W[0]; cannam@167: Tf = Td * Te; cannam@167: Tk = Td * Tj; cannam@167: Tg = W[1]; cannam@167: rio[WS(vs, 1) + WS(rs, 1)] = FMA(Tg, Tj, Tf); cannam@167: iio[WS(vs, 1) + WS(rs, 1)] = FNMS(Tg, Te, Tk); cannam@167: } cannam@167: { cannam@167: E T5, Ta, T3, T6; cannam@167: T3 = W[0]; cannam@167: T5 = T3 * T4; cannam@167: Ta = T3 * T9; cannam@167: T6 = W[1]; cannam@167: rio[WS(vs, 1)] = FMA(T6, T9, T5); cannam@167: iio[WS(vs, 1)] = FNMS(T6, T4, Ta); cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: {TW_FULL, 0, 2}, cannam@167: {TW_NEXT, 1, 0} cannam@167: }; cannam@167: cannam@167: static const ct_desc desc = { 2, "q1_2", twinstr, &GENUS, {8, 4, 4, 0}, 0, 0, 0 }; cannam@167: cannam@167: void X(codelet_q1_2) (planner *p) { cannam@167: X(kdft_difsq_register) (p, q1_2, &desc); cannam@167: } cannam@167: #else cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_twidsq.native -compact -variables 4 -pipeline-latency 4 -reload-twiddle -dif -n 2 -name q1_2 -include dft/scalar/q.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 12 FP additions, 8 FP multiplications, cannam@167: * (or, 8 additions, 4 multiplications, 4 fused multiply/add), cannam@167: * 17 stack variables, 0 constants, and 16 memory accesses cannam@167: */ cannam@167: #include "dft/scalar/q.h" cannam@167: cannam@167: static void q1_2(R *rio, R *iio, const R *W, stride rs, stride vs, INT mb, INT me, INT ms) cannam@167: { cannam@167: { cannam@167: INT m; cannam@167: for (m = mb, W = W + (mb * 2); m < me; m = m + 1, rio = rio + ms, iio = iio + ms, W = W + 2, MAKE_VOLATILE_STRIDE(4, rs), MAKE_VOLATILE_STRIDE(0, vs)) { cannam@167: E T1, T2, T4, T6, T7, T8, T9, Ta, Tc, Te, Tf, Tg; cannam@167: T1 = rio[0]; cannam@167: T2 = rio[WS(rs, 1)]; cannam@167: T4 = T1 - T2; cannam@167: T6 = iio[0]; cannam@167: T7 = iio[WS(rs, 1)]; cannam@167: T8 = T6 - T7; cannam@167: T9 = rio[WS(vs, 1)]; cannam@167: Ta = rio[WS(vs, 1) + WS(rs, 1)]; cannam@167: Tc = T9 - Ta; cannam@167: Te = iio[WS(vs, 1)]; cannam@167: Tf = iio[WS(vs, 1) + WS(rs, 1)]; cannam@167: Tg = Te - Tf; cannam@167: rio[0] = T1 + T2; cannam@167: iio[0] = T6 + T7; cannam@167: rio[WS(rs, 1)] = T9 + Ta; cannam@167: iio[WS(rs, 1)] = Te + Tf; cannam@167: { cannam@167: E Tb, Td, T3, T5; cannam@167: Tb = W[0]; cannam@167: Td = W[1]; cannam@167: rio[WS(vs, 1) + WS(rs, 1)] = FMA(Tb, Tc, Td * Tg); cannam@167: iio[WS(vs, 1) + WS(rs, 1)] = FNMS(Td, Tc, Tb * Tg); cannam@167: T3 = W[0]; cannam@167: T5 = W[1]; cannam@167: rio[WS(vs, 1)] = FMA(T3, T4, T5 * T8); cannam@167: iio[WS(vs, 1)] = FNMS(T5, T4, T3 * T8); cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: {TW_FULL, 0, 2}, cannam@167: {TW_NEXT, 1, 0} cannam@167: }; cannam@167: cannam@167: static const ct_desc desc = { 2, "q1_2", twinstr, &GENUS, {8, 4, 4, 0}, 0, 0, 0 }; cannam@167: cannam@167: void X(codelet_q1_2) (planner *p) { cannam@167: X(kdft_difsq_register) (p, q1_2, &desc); cannam@167: } cannam@167: #endif