cannam@95: /* cannam@95: * Copyright (c) 2003, 2007-11 Matteo Frigo cannam@95: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology cannam@95: * cannam@95: * This program is free software; you can redistribute it and/or modify cannam@95: * it under the terms of the GNU General Public License as published by cannam@95: * the Free Software Foundation; either version 2 of the License, or cannam@95: * (at your option) any later version. cannam@95: * cannam@95: * This program is distributed in the hope that it will be useful, cannam@95: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@95: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@95: * GNU General Public License for more details. cannam@95: * cannam@95: * You should have received a copy of the GNU General Public License cannam@95: * along with this program; if not, write to the Free Software cannam@95: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@95: * cannam@95: */ cannam@95: cannam@95: /* This file was automatically generated --- DO NOT EDIT */ cannam@95: /* Generated on Sun Nov 25 07:40:02 EST 2012 */ cannam@95: cannam@95: #include "codelet-rdft.h" cannam@95: cannam@95: #ifdef HAVE_FMA cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_hc2hc.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -twiddle-log3 -precompute-twiddles -n 4 -dit -name hf2_4 -include hf.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 24 FP additions, 16 FP multiplications, cannam@95: * (or, 16 additions, 8 multiplications, 8 fused multiply/add), cannam@95: * 33 stack variables, 0 constants, and 16 memory accesses cannam@95: */ cannam@95: #include "hf.h" cannam@95: cannam@95: static void hf2_4(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: { cannam@95: INT m; cannam@95: 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(8, rs)) { cannam@95: E Ti, Tq, To, Te, TA, Ty, Tm, Ts; cannam@95: { cannam@95: E T2, T6, T3, T5; cannam@95: T2 = W[0]; cannam@95: T6 = W[3]; cannam@95: T3 = W[2]; cannam@95: T5 = W[1]; cannam@95: { cannam@95: E T1, Tx, Td, Tw, Tj, Tl, Ta, T4, Tk, Tr; cannam@95: T1 = cr[0]; cannam@95: Ta = T2 * T6; cannam@95: T4 = T2 * T3; cannam@95: Tx = ci[0]; cannam@95: { cannam@95: E T8, Tb, T7, Tc; cannam@95: T8 = cr[WS(rs, 2)]; cannam@95: Tb = FNMS(T5, T3, Ta); cannam@95: T7 = FMA(T5, T6, T4); cannam@95: Tc = ci[WS(rs, 2)]; cannam@95: { cannam@95: E Tf, Th, T9, Tv, Tg, Tp; cannam@95: Tf = cr[WS(rs, 1)]; cannam@95: Th = ci[WS(rs, 1)]; cannam@95: T9 = T7 * T8; cannam@95: Tv = T7 * Tc; cannam@95: Tg = T2 * Tf; cannam@95: Tp = T2 * Th; cannam@95: Td = FMA(Tb, Tc, T9); cannam@95: Tw = FNMS(Tb, T8, Tv); cannam@95: Ti = FMA(T5, Th, Tg); cannam@95: Tq = FNMS(T5, Tf, Tp); cannam@95: } cannam@95: Tj = cr[WS(rs, 3)]; cannam@95: Tl = ci[WS(rs, 3)]; cannam@95: } cannam@95: To = T1 - Td; cannam@95: Te = T1 + Td; cannam@95: Tk = T3 * Tj; cannam@95: Tr = T3 * Tl; cannam@95: TA = Tx - Tw; cannam@95: Ty = Tw + Tx; cannam@95: Tm = FMA(T6, Tl, Tk); cannam@95: Ts = FNMS(T6, Tj, Tr); cannam@95: } cannam@95: } cannam@95: { cannam@95: E Tn, Tz, Tt, Tu; cannam@95: Tn = Ti + Tm; cannam@95: Tz = Tm - Ti; cannam@95: Tt = Tq - Ts; cannam@95: Tu = Tq + Ts; cannam@95: ci[WS(rs, 2)] = Tz + TA; cannam@95: cr[WS(rs, 3)] = Tz - TA; cannam@95: cr[0] = Te + Tn; cannam@95: ci[WS(rs, 1)] = Te - Tn; cannam@95: ci[WS(rs, 3)] = Tu + Ty; cannam@95: cr[WS(rs, 2)] = Tu - Ty; cannam@95: cr[WS(rs, 1)] = To + Tt; cannam@95: ci[0] = To - Tt; cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_CEXP, 1, 1}, cannam@95: {TW_CEXP, 1, 3}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2hc_desc desc = { 4, "hf2_4", twinstr, &GENUS, {16, 8, 8, 0} }; cannam@95: cannam@95: void X(codelet_hf2_4) (planner *p) { cannam@95: X(khc2hc_register) (p, hf2_4, &desc); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -twiddle-log3 -precompute-twiddles -n 4 -dit -name hf2_4 -include hf.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 24 FP additions, 16 FP multiplications, cannam@95: * (or, 16 additions, 8 multiplications, 8 fused multiply/add), cannam@95: * 21 stack variables, 0 constants, and 16 memory accesses cannam@95: */ cannam@95: #include "hf.h" cannam@95: cannam@95: static void hf2_4(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: { cannam@95: INT m; cannam@95: 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(8, rs)) { cannam@95: E T2, T4, T3, T5, T6, T8; cannam@95: T2 = W[0]; cannam@95: T4 = W[1]; cannam@95: T3 = W[2]; cannam@95: T5 = W[3]; cannam@95: T6 = FMA(T2, T3, T4 * T5); cannam@95: T8 = FNMS(T4, T3, T2 * T5); cannam@95: { cannam@95: E T1, Tp, Ta, To, Te, Tk, Th, Tl, T7, T9; cannam@95: T1 = cr[0]; cannam@95: Tp = ci[0]; cannam@95: T7 = cr[WS(rs, 2)]; cannam@95: T9 = ci[WS(rs, 2)]; cannam@95: Ta = FMA(T6, T7, T8 * T9); cannam@95: To = FNMS(T8, T7, T6 * T9); cannam@95: { cannam@95: E Tc, Td, Tf, Tg; cannam@95: Tc = cr[WS(rs, 1)]; cannam@95: Td = ci[WS(rs, 1)]; cannam@95: Te = FMA(T2, Tc, T4 * Td); cannam@95: Tk = FNMS(T4, Tc, T2 * Td); cannam@95: Tf = cr[WS(rs, 3)]; cannam@95: Tg = ci[WS(rs, 3)]; cannam@95: Th = FMA(T3, Tf, T5 * Tg); cannam@95: Tl = FNMS(T5, Tf, T3 * Tg); cannam@95: } cannam@95: { cannam@95: E Tb, Ti, Tj, Tm; cannam@95: Tb = T1 + Ta; cannam@95: Ti = Te + Th; cannam@95: ci[WS(rs, 1)] = Tb - Ti; cannam@95: cr[0] = Tb + Ti; cannam@95: Tj = T1 - Ta; cannam@95: Tm = Tk - Tl; cannam@95: ci[0] = Tj - Tm; cannam@95: cr[WS(rs, 1)] = Tj + Tm; cannam@95: } cannam@95: { cannam@95: E Tn, Tq, Tr, Ts; cannam@95: Tn = Tk + Tl; cannam@95: Tq = To + Tp; cannam@95: cr[WS(rs, 2)] = Tn - Tq; cannam@95: ci[WS(rs, 3)] = Tn + Tq; cannam@95: Tr = Th - Te; cannam@95: Ts = Tp - To; cannam@95: cr[WS(rs, 3)] = Tr - Ts; cannam@95: ci[WS(rs, 2)] = Tr + Ts; cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_CEXP, 1, 1}, cannam@95: {TW_CEXP, 1, 3}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2hc_desc desc = { 4, "hf2_4", twinstr, &GENUS, {16, 8, 8, 0} }; cannam@95: cannam@95: void X(codelet_hf2_4) (planner *p) { cannam@95: X(khc2hc_register) (p, hf2_4, &desc); cannam@95: } cannam@95: #endif /* HAVE_FMA */