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 16:36:05 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 -n 4 -name t1_4 -include t.h */ cannam@127: cannam@127: /* cannam@127: * This function contains 22 FP additions, 12 FP multiplications, cannam@127: * (or, 16 additions, 6 multiplications, 6 fused multiply/add), cannam@127: * 31 stack variables, 0 constants, and 16 memory accesses cannam@127: */ cannam@127: #include "t.h" cannam@127: cannam@127: static void t1_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 * 6); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 6, MAKE_VOLATILE_STRIDE(8, rs)) { cannam@127: E To, Te, Tm, T8, Tw, Tx, Tq, Tk; cannam@127: { cannam@127: E T1, Tv, Tu, T7, Tg, Tj, Tf, Ti, Tp, Th; cannam@127: T1 = ri[0]; cannam@127: Tv = ii[0]; cannam@127: { cannam@127: E T3, T6, T2, T5; cannam@127: T3 = ri[WS(rs, 2)]; cannam@127: T6 = ii[WS(rs, 2)]; cannam@127: T2 = W[2]; cannam@127: T5 = W[3]; cannam@127: { cannam@127: E Ta, Td, Tc, Tn, Tb, Tt, T4, T9; cannam@127: Ta = ri[WS(rs, 1)]; cannam@127: Td = ii[WS(rs, 1)]; cannam@127: Tt = T2 * T6; cannam@127: T4 = T2 * T3; cannam@127: T9 = W[0]; cannam@127: Tc = W[1]; cannam@127: Tu = FNMS(T5, T3, Tt); cannam@127: T7 = FMA(T5, T6, T4); cannam@127: Tn = T9 * Td; cannam@127: Tb = T9 * Ta; cannam@127: Tg = ri[WS(rs, 3)]; cannam@127: Tj = ii[WS(rs, 3)]; cannam@127: To = FNMS(Tc, Ta, Tn); cannam@127: Te = FMA(Tc, Td, Tb); cannam@127: Tf = W[4]; cannam@127: Ti = W[5]; cannam@127: } cannam@127: } cannam@127: Tm = T1 - T7; cannam@127: T8 = T1 + T7; cannam@127: Tw = Tu + Tv; cannam@127: Tx = Tv - Tu; cannam@127: Tp = Tf * Tj; cannam@127: Th = Tf * Tg; cannam@127: Tq = FNMS(Ti, Tg, Tp); cannam@127: Tk = FMA(Ti, Tj, Th); cannam@127: } cannam@127: { cannam@127: E Ts, Tr, Tl, Ty; cannam@127: Ts = To + Tq; cannam@127: Tr = To - Tq; cannam@127: Tl = Te + Tk; cannam@127: Ty = Te - Tk; cannam@127: ri[WS(rs, 1)] = Tm + Tr; cannam@127: ri[WS(rs, 3)] = Tm - Tr; cannam@127: ii[WS(rs, 2)] = Tw - Ts; cannam@127: ii[0] = Ts + Tw; cannam@127: ii[WS(rs, 3)] = Ty + Tx; cannam@127: ii[WS(rs, 1)] = Tx - Ty; cannam@127: ri[0] = T8 + Tl; cannam@127: ri[WS(rs, 2)] = T8 - Tl; cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: {TW_FULL, 0, 4}, cannam@127: {TW_NEXT, 1, 0} cannam@127: }; cannam@127: cannam@127: static const ct_desc desc = { 4, "t1_4", twinstr, &GENUS, {16, 6, 6, 0}, 0, 0, 0 }; cannam@127: cannam@127: void X(codelet_t1_4) (planner *p) { cannam@127: X(kdft_dit_register) (p, t1_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 -n 4 -name t1_4 -include t.h */ cannam@127: cannam@127: /* cannam@127: * This function contains 22 FP additions, 12 FP multiplications, cannam@127: * (or, 16 additions, 6 multiplications, 6 fused multiply/add), cannam@127: * 13 stack variables, 0 constants, and 16 memory accesses cannam@127: */ cannam@127: #include "t.h" cannam@127: cannam@127: static void t1_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 * 6); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 6, MAKE_VOLATILE_STRIDE(8, rs)) { cannam@127: E T1, Tp, T6, To, Tc, Tk, Th, Tl; cannam@127: T1 = ri[0]; cannam@127: Tp = ii[0]; cannam@127: { cannam@127: E T3, T5, T2, T4; cannam@127: T3 = ri[WS(rs, 2)]; cannam@127: T5 = ii[WS(rs, 2)]; cannam@127: T2 = W[2]; cannam@127: T4 = W[3]; cannam@127: T6 = FMA(T2, T3, T4 * T5); cannam@127: To = FNMS(T4, T3, T2 * T5); cannam@127: } cannam@127: { cannam@127: E T9, Tb, T8, Ta; cannam@127: T9 = ri[WS(rs, 1)]; cannam@127: Tb = ii[WS(rs, 1)]; cannam@127: T8 = W[0]; cannam@127: Ta = W[1]; cannam@127: Tc = FMA(T8, T9, Ta * Tb); cannam@127: Tk = FNMS(Ta, T9, T8 * Tb); cannam@127: } cannam@127: { cannam@127: E Te, Tg, Td, Tf; cannam@127: Te = ri[WS(rs, 3)]; cannam@127: Tg = ii[WS(rs, 3)]; cannam@127: Td = W[4]; cannam@127: Tf = W[5]; cannam@127: Th = FMA(Td, Te, Tf * Tg); cannam@127: Tl = FNMS(Tf, Te, Td * Tg); cannam@127: } cannam@127: { cannam@127: E T7, Ti, Tn, Tq; cannam@127: T7 = T1 + T6; cannam@127: Ti = Tc + Th; cannam@127: ri[WS(rs, 2)] = T7 - Ti; cannam@127: ri[0] = T7 + 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 - T6; 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 = Tc - 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: static const tw_instr twinstr[] = { cannam@127: {TW_FULL, 0, 4}, cannam@127: {TW_NEXT, 1, 0} cannam@127: }; cannam@127: cannam@127: static const ct_desc desc = { 4, "t1_4", twinstr, &GENUS, {16, 6, 6, 0}, 0, 0, 0 }; cannam@127: cannam@127: void X(codelet_t1_4) (planner *p) { cannam@127: X(kdft_dit_register) (p, t1_4, &desc); cannam@127: } cannam@127: #endif /* HAVE_FMA */