cannam@127: /* cannam@127: * Copyright (c) 2003, 2007-14 Matteo Frigo cannam@127: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology cannam@127: * cannam@127: * This program is free software; you can redistribute it and/or modify cannam@127: * it under the terms of the GNU General Public License as published by cannam@127: * the Free Software Foundation; either version 2 of the License, or cannam@127: * (at your option) any later version. cannam@127: * cannam@127: * This program is distributed in the hope that it will be useful, cannam@127: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@127: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@127: * GNU General Public License for more details. cannam@127: * cannam@127: * You should have received a copy of the GNU General Public License cannam@127: * along with this program; if not, write to the Free Software cannam@127: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@127: * cannam@127: */ cannam@127: cannam@127: /* This file was automatically generated --- DO NOT EDIT */ cannam@127: /* Generated on Sat Jul 30 16:36:42 EDT 2016 */ cannam@127: cannam@127: #include "codelet-dft.h" cannam@127: cannam@127: #ifdef HAVE_FMA cannam@127: cannam@127: /* Generated by: ../../../genfft/gen_twiddle.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -twiddle-log3 -precompute-twiddles -n 4 -name t2_4 -include t.h */ cannam@127: cannam@127: /* cannam@127: * This function contains 24 FP additions, 16 FP multiplications, cannam@127: * (or, 16 additions, 8 multiplications, 8 fused multiply/add), cannam@127: * 33 stack variables, 0 constants, and 16 memory accesses cannam@127: */ cannam@127: #include "t.h" cannam@127: cannam@127: static void t2_4(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@127: { cannam@127: { cannam@127: INT m; cannam@127: for (m = mb, W = W + (mb * 4); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 4, MAKE_VOLATILE_STRIDE(8, rs)) { cannam@127: E Ti, Tq, To, Te, Ty, Tz, Tm, Ts; cannam@127: { cannam@127: E T2, T6, T3, T5; cannam@127: T2 = W[0]; cannam@127: T6 = W[3]; cannam@127: T3 = W[2]; cannam@127: T5 = W[1]; cannam@127: { cannam@127: E T1, Tx, Td, Tw, Tj, Tl, Ta, T4, Tk, Tr; cannam@127: T1 = ri[0]; cannam@127: Ta = T2 * T6; cannam@127: T4 = T2 * T3; cannam@127: Tx = ii[0]; cannam@127: { cannam@127: E T8, Tb, T7, Tc; cannam@127: T8 = ri[WS(rs, 2)]; cannam@127: Tb = FNMS(T5, T3, Ta); cannam@127: T7 = FMA(T5, T6, T4); cannam@127: Tc = ii[WS(rs, 2)]; cannam@127: { cannam@127: E Tf, Th, T9, Tv, Tg, Tp; cannam@127: Tf = ri[WS(rs, 1)]; cannam@127: Th = ii[WS(rs, 1)]; cannam@127: T9 = T7 * T8; cannam@127: Tv = T7 * Tc; cannam@127: Tg = T2 * Tf; cannam@127: Tp = T2 * Th; cannam@127: Td = FMA(Tb, Tc, T9); cannam@127: Tw = FNMS(Tb, T8, Tv); cannam@127: Ti = FMA(T5, Th, Tg); cannam@127: Tq = FNMS(T5, Tf, Tp); cannam@127: } cannam@127: Tj = ri[WS(rs, 3)]; cannam@127: Tl = ii[WS(rs, 3)]; cannam@127: } cannam@127: To = T1 - Td; cannam@127: Te = T1 + Td; cannam@127: Ty = Tw + Tx; cannam@127: Tz = Tx - Tw; cannam@127: Tk = T3 * Tj; cannam@127: Tr = T3 * Tl; cannam@127: Tm = FMA(T6, Tl, Tk); cannam@127: Ts = FNMS(T6, Tj, Tr); cannam@127: } cannam@127: } cannam@127: { cannam@127: E Tn, TA, Tu, Tt; cannam@127: Tn = Ti + Tm; cannam@127: TA = Ti - Tm; cannam@127: Tu = Tq + Ts; cannam@127: Tt = Tq - Ts; cannam@127: ii[WS(rs, 3)] = TA + Tz; cannam@127: ii[WS(rs, 1)] = Tz - TA; cannam@127: ri[0] = Te + Tn; cannam@127: ri[WS(rs, 2)] = Te - Tn; cannam@127: ri[WS(rs, 1)] = To + Tt; cannam@127: ri[WS(rs, 3)] = To - Tt; cannam@127: ii[WS(rs, 2)] = Ty - Tu; cannam@127: ii[0] = Tu + Ty; cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: {TW_CEXP, 0, 1}, cannam@127: {TW_CEXP, 0, 3}, cannam@127: {TW_NEXT, 1, 0} cannam@127: }; cannam@127: cannam@127: static const ct_desc desc = { 4, "t2_4", twinstr, &GENUS, {16, 8, 8, 0}, 0, 0, 0 }; cannam@127: cannam@127: void X(codelet_t2_4) (planner *p) { cannam@127: X(kdft_dit_register) (p, t2_4, &desc); cannam@127: } cannam@127: #else /* HAVE_FMA */ cannam@127: cannam@127: /* Generated by: ../../../genfft/gen_twiddle.native -compact -variables 4 -pipeline-latency 4 -twiddle-log3 -precompute-twiddles -n 4 -name t2_4 -include t.h */ cannam@127: cannam@127: /* cannam@127: * This function contains 24 FP additions, 16 FP multiplications, cannam@127: * (or, 16 additions, 8 multiplications, 8 fused multiply/add), cannam@127: * 21 stack variables, 0 constants, and 16 memory accesses cannam@127: */ cannam@127: #include "t.h" cannam@127: cannam@127: static void t2_4(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@127: { cannam@127: { cannam@127: INT m; cannam@127: for (m = mb, W = W + (mb * 4); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 4, MAKE_VOLATILE_STRIDE(8, rs)) { cannam@127: E T2, T4, T3, T5, T6, T8; cannam@127: T2 = W[0]; cannam@127: T4 = W[1]; cannam@127: T3 = W[2]; cannam@127: T5 = W[3]; cannam@127: T6 = FMA(T2, T3, T4 * T5); cannam@127: T8 = FNMS(T4, T3, T2 * T5); cannam@127: { cannam@127: E T1, Tp, Ta, To, Te, Tk, Th, Tl, T7, T9; cannam@127: T1 = ri[0]; cannam@127: Tp = ii[0]; cannam@127: T7 = ri[WS(rs, 2)]; cannam@127: T9 = ii[WS(rs, 2)]; cannam@127: Ta = FMA(T6, T7, T8 * T9); cannam@127: To = FNMS(T8, T7, T6 * T9); cannam@127: { cannam@127: E Tc, Td, Tf, Tg; cannam@127: Tc = ri[WS(rs, 1)]; cannam@127: Td = ii[WS(rs, 1)]; cannam@127: Te = FMA(T2, Tc, T4 * Td); cannam@127: Tk = FNMS(T4, Tc, T2 * Td); cannam@127: Tf = ri[WS(rs, 3)]; cannam@127: Tg = ii[WS(rs, 3)]; cannam@127: Th = FMA(T3, Tf, T5 * Tg); cannam@127: Tl = FNMS(T5, Tf, T3 * Tg); cannam@127: } cannam@127: { cannam@127: E Tb, Ti, Tn, Tq; cannam@127: Tb = T1 + Ta; cannam@127: Ti = Te + Th; cannam@127: ri[WS(rs, 2)] = Tb - Ti; cannam@127: ri[0] = Tb + Ti; cannam@127: Tn = Tk + Tl; cannam@127: Tq = To + Tp; cannam@127: ii[0] = Tn + Tq; cannam@127: ii[WS(rs, 2)] = Tq - Tn; cannam@127: } cannam@127: { cannam@127: E Tj, Tm, Tr, Ts; cannam@127: Tj = T1 - Ta; cannam@127: Tm = Tk - Tl; cannam@127: ri[WS(rs, 3)] = Tj - Tm; cannam@127: ri[WS(rs, 1)] = Tj + Tm; cannam@127: Tr = Tp - To; cannam@127: Ts = Te - Th; cannam@127: ii[WS(rs, 1)] = Tr - Ts; cannam@127: ii[WS(rs, 3)] = Ts + Tr; cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: {TW_CEXP, 0, 1}, cannam@127: {TW_CEXP, 0, 3}, cannam@127: {TW_NEXT, 1, 0} cannam@127: }; cannam@127: cannam@127: static const ct_desc desc = { 4, "t2_4", twinstr, &GENUS, {16, 8, 8, 0}, 0, 0, 0 }; cannam@127: cannam@127: void X(codelet_t2_4) (planner *p) { cannam@127: X(kdft_dit_register) (p, t2_4, &desc); cannam@127: } cannam@127: #endif /* HAVE_FMA */