Chris@10: /* Chris@10: * Copyright (c) 2003, 2007-11 Matteo Frigo Chris@10: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology Chris@10: * Chris@10: * This program is free software; you can redistribute it and/or modify Chris@10: * it under the terms of the GNU General Public License as published by Chris@10: * the Free Software Foundation; either version 2 of the License, or Chris@10: * (at your option) any later version. Chris@10: * Chris@10: * This program is distributed in the hope that it will be useful, Chris@10: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@10: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@10: * GNU General Public License for more details. Chris@10: * Chris@10: * You should have received a copy of the GNU General Public License Chris@10: * along with this program; if not, write to the Free Software Chris@10: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@10: * Chris@10: */ Chris@10: Chris@10: /* This file was automatically generated --- DO NOT EDIT */ Chris@10: /* Generated on Sun Nov 25 07:39:50 EST 2012 */ Chris@10: Chris@10: #include "codelet-rdft.h" Chris@10: Chris@10: #ifdef HAVE_FMA Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_hc2hc.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 5 -dit -name hf_5 -include hf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 40 FP additions, 34 FP multiplications, Chris@10: * (or, 14 additions, 8 multiplications, 26 fused multiply/add), Chris@10: * 43 stack variables, 4 constants, and 20 memory accesses Chris@10: */ Chris@10: #include "hf.h" Chris@10: Chris@10: static void hf_5(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP951056516, +0.951056516295153572116439333379382143405698634); Chris@10: DK(KP559016994, +0.559016994374947424102293417182819058860154590); Chris@10: DK(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@10: DK(KP618033988, +0.618033988749894848204586834365638117720309180); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * 8); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 8, MAKE_VOLATILE_STRIDE(10, rs)) { Chris@10: E T1, TJ, TK, TA, TR, Te, TC, Tk, TE, Tq; Chris@10: { Chris@10: E Tg, Tj, Tm, TB, Th, Tp, Tl, Ti, To, TD, Tn; Chris@10: T1 = cr[0]; Chris@10: TJ = ci[0]; Chris@10: { Chris@10: E T9, Tc, Ty, Ta, Tb, Tx, T7, Tf, Tz, Td; Chris@10: { Chris@10: E T3, T6, T8, Tw, T4, T2, T5; Chris@10: T3 = cr[WS(rs, 1)]; Chris@10: T6 = ci[WS(rs, 1)]; Chris@10: T2 = W[0]; Chris@10: T9 = cr[WS(rs, 4)]; Chris@10: Tc = ci[WS(rs, 4)]; Chris@10: T8 = W[6]; Chris@10: Tw = T2 * T6; Chris@10: T4 = T2 * T3; Chris@10: T5 = W[1]; Chris@10: Ty = T8 * Tc; Chris@10: Ta = T8 * T9; Chris@10: Tb = W[7]; Chris@10: Tx = FNMS(T5, T3, Tw); Chris@10: T7 = FMA(T5, T6, T4); Chris@10: } Chris@10: Tg = cr[WS(rs, 2)]; Chris@10: Tz = FNMS(Tb, T9, Ty); Chris@10: Td = FMA(Tb, Tc, Ta); Chris@10: Tj = ci[WS(rs, 2)]; Chris@10: Tf = W[2]; Chris@10: TK = Tx + Tz; Chris@10: TA = Tx - Tz; Chris@10: TR = Td - T7; Chris@10: Te = T7 + Td; Chris@10: Tm = cr[WS(rs, 3)]; Chris@10: TB = Tf * Tj; Chris@10: Th = Tf * Tg; Chris@10: Tp = ci[WS(rs, 3)]; Chris@10: Tl = W[4]; Chris@10: Ti = W[3]; Chris@10: To = W[5]; Chris@10: } Chris@10: TD = Tl * Tp; Chris@10: Tn = Tl * Tm; Chris@10: TC = FNMS(Ti, Tg, TB); Chris@10: Tk = FMA(Ti, Tj, Th); Chris@10: TE = FNMS(To, Tm, TD); Chris@10: Tq = FMA(To, Tp, Tn); Chris@10: } Chris@10: { Chris@10: E TG, TI, TO, TS, TU, Tu, TN, Tt, TL, TF; Chris@10: TL = TC + TE; Chris@10: TF = TC - TE; Chris@10: { Chris@10: E Tr, TQ, TM, Ts; Chris@10: Tr = Tk + Tq; Chris@10: TQ = Tk - Tq; Chris@10: TG = FMA(KP618033988, TF, TA); Chris@10: TI = FNMS(KP618033988, TA, TF); Chris@10: TO = TK - TL; Chris@10: TM = TK + TL; Chris@10: TS = FMA(KP618033988, TR, TQ); Chris@10: TU = FNMS(KP618033988, TQ, TR); Chris@10: Tu = Te - Tr; Chris@10: Ts = Te + Tr; Chris@10: ci[WS(rs, 4)] = TM + TJ; Chris@10: TN = FNMS(KP250000000, TM, TJ); Chris@10: cr[0] = T1 + Ts; Chris@10: Tt = FNMS(KP250000000, Ts, T1); Chris@10: } Chris@10: { Chris@10: E TT, TP, Tv, TH; Chris@10: TT = FMA(KP559016994, TO, TN); Chris@10: TP = FNMS(KP559016994, TO, TN); Chris@10: Tv = FMA(KP559016994, Tu, Tt); Chris@10: TH = FNMS(KP559016994, Tu, Tt); Chris@10: ci[WS(rs, 2)] = FMA(KP951056516, TS, TP); Chris@10: cr[WS(rs, 3)] = FMS(KP951056516, TS, TP); Chris@10: ci[WS(rs, 3)] = FMA(KP951056516, TU, TT); Chris@10: cr[WS(rs, 4)] = FMS(KP951056516, TU, TT); Chris@10: ci[WS(rs, 1)] = FMA(KP951056516, TI, TH); Chris@10: cr[WS(rs, 2)] = FNMS(KP951056516, TI, TH); Chris@10: cr[WS(rs, 1)] = FMA(KP951056516, TG, Tv); Chris@10: ci[0] = FNMS(KP951056516, TG, Tv); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 5}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2hc_desc desc = { 5, "hf_5", twinstr, &GENUS, {14, 8, 26, 0} }; Chris@10: Chris@10: void X(codelet_hf_5) (planner *p) { Chris@10: X(khc2hc_register) (p, hf_5, &desc); Chris@10: } Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -n 5 -dit -name hf_5 -include hf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 40 FP additions, 28 FP multiplications, Chris@10: * (or, 26 additions, 14 multiplications, 14 fused multiply/add), Chris@10: * 29 stack variables, 4 constants, and 20 memory accesses Chris@10: */ Chris@10: #include "hf.h" Chris@10: Chris@10: static void hf_5(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@10: DK(KP559016994, +0.559016994374947424102293417182819058860154590); Chris@10: DK(KP587785252, +0.587785252292473129168705954639072768597652438); Chris@10: DK(KP951056516, +0.951056516295153572116439333379382143405698634); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * 8); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 8, MAKE_VOLATILE_STRIDE(10, rs)) { Chris@10: E T1, TE, Tu, Tx, TC, TB, TF, TG, TH, Tc, Tn, To; Chris@10: T1 = cr[0]; Chris@10: TE = ci[0]; Chris@10: { Chris@10: E T6, Ts, Tm, Tw, Tb, Tt, Th, Tv; Chris@10: { Chris@10: E T3, T5, T2, T4; Chris@10: T3 = cr[WS(rs, 1)]; Chris@10: T5 = ci[WS(rs, 1)]; Chris@10: T2 = W[0]; Chris@10: T4 = W[1]; Chris@10: T6 = FMA(T2, T3, T4 * T5); Chris@10: Ts = FNMS(T4, T3, T2 * T5); Chris@10: } Chris@10: { Chris@10: E Tj, Tl, Ti, Tk; Chris@10: Tj = cr[WS(rs, 3)]; Chris@10: Tl = ci[WS(rs, 3)]; Chris@10: Ti = W[4]; Chris@10: Tk = W[5]; Chris@10: Tm = FMA(Ti, Tj, Tk * Tl); Chris@10: Tw = FNMS(Tk, Tj, Ti * Tl); Chris@10: } Chris@10: { Chris@10: E T8, Ta, T7, T9; Chris@10: T8 = cr[WS(rs, 4)]; Chris@10: Ta = ci[WS(rs, 4)]; Chris@10: T7 = W[6]; Chris@10: T9 = W[7]; Chris@10: Tb = FMA(T7, T8, T9 * Ta); Chris@10: Tt = FNMS(T9, T8, T7 * Ta); Chris@10: } Chris@10: { Chris@10: E Te, Tg, Td, Tf; Chris@10: Te = cr[WS(rs, 2)]; Chris@10: Tg = ci[WS(rs, 2)]; Chris@10: Td = W[2]; Chris@10: Tf = W[3]; Chris@10: Th = FMA(Td, Te, Tf * Tg); Chris@10: Tv = FNMS(Tf, Te, Td * Tg); Chris@10: } Chris@10: Tu = Ts - Tt; Chris@10: Tx = Tv - Tw; Chris@10: TC = Th - Tm; Chris@10: TB = Tb - T6; Chris@10: TF = Ts + Tt; Chris@10: TG = Tv + Tw; Chris@10: TH = TF + TG; Chris@10: Tc = T6 + Tb; Chris@10: Tn = Th + Tm; Chris@10: To = Tc + Tn; Chris@10: } Chris@10: cr[0] = T1 + To; Chris@10: { Chris@10: E Ty, TA, Tr, Tz, Tp, Tq; Chris@10: Ty = FMA(KP951056516, Tu, KP587785252 * Tx); Chris@10: TA = FNMS(KP587785252, Tu, KP951056516 * Tx); Chris@10: Tp = KP559016994 * (Tc - Tn); Chris@10: Tq = FNMS(KP250000000, To, T1); Chris@10: Tr = Tp + Tq; Chris@10: Tz = Tq - Tp; Chris@10: ci[0] = Tr - Ty; Chris@10: ci[WS(rs, 1)] = Tz + TA; Chris@10: cr[WS(rs, 1)] = Tr + Ty; Chris@10: cr[WS(rs, 2)] = Tz - TA; Chris@10: } Chris@10: ci[WS(rs, 4)] = TH + TE; Chris@10: { Chris@10: E TD, TL, TK, TM, TI, TJ; Chris@10: TD = FMA(KP587785252, TB, KP951056516 * TC); Chris@10: TL = FNMS(KP587785252, TC, KP951056516 * TB); Chris@10: TI = FNMS(KP250000000, TH, TE); Chris@10: TJ = KP559016994 * (TF - TG); Chris@10: TK = TI - TJ; Chris@10: TM = TJ + TI; Chris@10: cr[WS(rs, 3)] = TD - TK; Chris@10: ci[WS(rs, 3)] = TL + TM; Chris@10: ci[WS(rs, 2)] = TD + TK; Chris@10: cr[WS(rs, 4)] = TL - TM; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 5}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2hc_desc desc = { 5, "hf_5", twinstr, &GENUS, {26, 14, 14, 0} }; Chris@10: Chris@10: void X(codelet_hf_5) (planner *p) { Chris@10: X(khc2hc_register) (p, hf_5, &desc); Chris@10: } Chris@10: #endif /* HAVE_FMA */