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 07:40:07 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 -twiddle-log3 -precompute-twiddles -n 5 -dit -name hf2_5 -include hf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 44 FP additions, 40 FP multiplications, Chris@10: * (or, 14 additions, 10 multiplications, 30 fused multiply/add), Chris@10: * 47 stack variables, 4 constants, and 20 memory accesses Chris@10: */ Chris@10: #include "hf.h" Chris@10: Chris@10: static void hf2_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) * 4); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 4, MAKE_VOLATILE_STRIDE(10, rs)) { Chris@10: E Ta, T1, TL, Tp, TT, Ti, TM, TC, To, TE, Ts, TF, T2, T8, T5; Chris@10: E TS, Tt, TG; Chris@10: T2 = W[0]; Chris@10: Ta = W[3]; Chris@10: T8 = W[2]; Chris@10: T5 = W[1]; Chris@10: { Chris@10: E Tq, Tr, Te, T9; Chris@10: T1 = cr[0]; Chris@10: Te = T2 * Ta; Chris@10: T9 = T2 * T8; Chris@10: TL = ci[0]; Chris@10: { Chris@10: E T3, Tf, Tm, Tj, Tb, T4, T6, Tc, Tg; Chris@10: T3 = cr[WS(rs, 1)]; Chris@10: Tf = FMA(T5, T8, Te); Chris@10: Tm = FNMS(T5, T8, Te); Chris@10: Tj = FMA(T5, Ta, T9); Chris@10: Tb = FNMS(T5, Ta, T9); Chris@10: T4 = T2 * T3; Chris@10: T6 = ci[WS(rs, 1)]; Chris@10: Tc = cr[WS(rs, 4)]; Chris@10: Tg = ci[WS(rs, 4)]; Chris@10: { Chris@10: E Tk, Tl, Tn, TD; Chris@10: { Chris@10: E T7, Tz, Th, TB, Ty, Td, TA; Chris@10: Tk = cr[WS(rs, 2)]; Chris@10: T7 = FMA(T5, T6, T4); Chris@10: Ty = T2 * T6; Chris@10: Td = Tb * Tc; Chris@10: TA = Tb * Tg; Chris@10: Tl = Tj * Tk; Chris@10: Tz = FNMS(T5, T3, Ty); Chris@10: Th = FMA(Tf, Tg, Td); Chris@10: TB = FNMS(Tf, Tc, TA); Chris@10: Tn = ci[WS(rs, 2)]; Chris@10: Tp = cr[WS(rs, 3)]; Chris@10: TT = Th - T7; Chris@10: Ti = T7 + Th; Chris@10: TM = Tz + TB; Chris@10: TC = Tz - TB; Chris@10: TD = Tj * Tn; Chris@10: Tq = T8 * Tp; Chris@10: Tr = ci[WS(rs, 3)]; Chris@10: } Chris@10: To = FMA(Tm, Tn, Tl); Chris@10: TE = FNMS(Tm, Tk, TD); Chris@10: } Chris@10: } Chris@10: Ts = FMA(Ta, Tr, Tq); Chris@10: TF = T8 * Tr; Chris@10: } Chris@10: TS = To - Ts; Chris@10: Tt = To + Ts; Chris@10: TG = FNMS(Ta, Tp, TF); Chris@10: { Chris@10: E TU, TW, TV, TR, Tw, Tu; Chris@10: TU = FMA(KP618033988, TT, TS); Chris@10: TW = FNMS(KP618033988, TS, TT); Chris@10: Tw = Ti - Tt; Chris@10: Tu = Ti + Tt; Chris@10: { Chris@10: E TN, TH, Tv, TI, TK; Chris@10: TN = TE + TG; Chris@10: TH = TE - TG; Chris@10: cr[0] = T1 + Tu; Chris@10: Tv = FNMS(KP250000000, Tu, T1); Chris@10: TI = FMA(KP618033988, TH, TC); Chris@10: TK = FNMS(KP618033988, TC, TH); Chris@10: { Chris@10: E TQ, TO, Tx, TJ, TP; Chris@10: TQ = TM - TN; Chris@10: TO = TM + TN; Chris@10: Tx = FMA(KP559016994, Tw, Tv); Chris@10: TJ = FNMS(KP559016994, Tw, Tv); Chris@10: ci[WS(rs, 4)] = TO + TL; Chris@10: TP = FNMS(KP250000000, TO, TL); Chris@10: ci[WS(rs, 1)] = FMA(KP951056516, TK, TJ); Chris@10: cr[WS(rs, 2)] = FNMS(KP951056516, TK, TJ); Chris@10: cr[WS(rs, 1)] = FMA(KP951056516, TI, Tx); Chris@10: ci[0] = FNMS(KP951056516, TI, Tx); Chris@10: TV = FMA(KP559016994, TQ, TP); Chris@10: TR = FNMS(KP559016994, TQ, TP); Chris@10: } Chris@10: } Chris@10: ci[WS(rs, 2)] = FMA(KP951056516, TU, TR); Chris@10: cr[WS(rs, 3)] = FMS(KP951056516, TU, TR); Chris@10: ci[WS(rs, 3)] = FMA(KP951056516, TW, TV); Chris@10: cr[WS(rs, 4)] = FMS(KP951056516, TW, TV); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_CEXP, 1, 1}, Chris@10: {TW_CEXP, 1, 3}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2hc_desc desc = { 5, "hf2_5", twinstr, &GENUS, {14, 10, 30, 0} }; Chris@10: Chris@10: void X(codelet_hf2_5) (planner *p) { Chris@10: X(khc2hc_register) (p, hf2_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 -twiddle-log3 -precompute-twiddles -n 5 -dit -name hf2_5 -include hf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 44 FP additions, 32 FP multiplications, Chris@10: * (or, 30 additions, 18 multiplications, 14 fused multiply/add), Chris@10: * 37 stack variables, 4 constants, and 20 memory accesses Chris@10: */ Chris@10: #include "hf.h" Chris@10: Chris@10: static void hf2_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) * 4); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 4, MAKE_VOLATILE_STRIDE(10, rs)) { Chris@10: E T2, T4, T7, T9, Tb, Tl, Tf, Tj; Chris@10: { Chris@10: E T8, Te, Ta, Td; Chris@10: T2 = W[0]; Chris@10: T4 = W[1]; Chris@10: T7 = W[2]; Chris@10: T9 = W[3]; Chris@10: T8 = T2 * T7; Chris@10: Te = T4 * T7; Chris@10: Ta = T4 * T9; Chris@10: Td = T2 * T9; Chris@10: Tb = T8 - Ta; Chris@10: Tl = Td - Te; Chris@10: Tf = Td + Te; Chris@10: Tj = T8 + Ta; Chris@10: } Chris@10: { Chris@10: E T1, TI, Ty, TB, TG, TF, TJ, TK, TL, Ti, Tr, Ts; Chris@10: T1 = cr[0]; Chris@10: TI = ci[0]; Chris@10: { Chris@10: E T6, Tw, Tq, TA, Th, Tx, Tn, Tz; Chris@10: { Chris@10: E T3, T5, To, Tp; Chris@10: T3 = cr[WS(rs, 1)]; Chris@10: T5 = ci[WS(rs, 1)]; Chris@10: T6 = FMA(T2, T3, T4 * T5); Chris@10: Tw = FNMS(T4, T3, T2 * T5); Chris@10: To = cr[WS(rs, 3)]; Chris@10: Tp = ci[WS(rs, 3)]; Chris@10: Tq = FMA(T7, To, T9 * Tp); Chris@10: TA = FNMS(T9, To, T7 * Tp); Chris@10: } Chris@10: { Chris@10: E Tc, Tg, Tk, Tm; Chris@10: Tc = cr[WS(rs, 4)]; Chris@10: Tg = ci[WS(rs, 4)]; Chris@10: Th = FMA(Tb, Tc, Tf * Tg); Chris@10: Tx = FNMS(Tf, Tc, Tb * Tg); Chris@10: Tk = cr[WS(rs, 2)]; Chris@10: Tm = ci[WS(rs, 2)]; Chris@10: Tn = FMA(Tj, Tk, Tl * Tm); Chris@10: Tz = FNMS(Tl, Tk, Tj * Tm); Chris@10: } Chris@10: Ty = Tw - Tx; Chris@10: TB = Tz - TA; Chris@10: TG = Tn - Tq; Chris@10: TF = Th - T6; Chris@10: TJ = Tw + Tx; Chris@10: TK = Tz + TA; Chris@10: TL = TJ + TK; Chris@10: Ti = T6 + Th; Chris@10: Tr = Tn + Tq; Chris@10: Ts = Ti + Tr; Chris@10: } Chris@10: cr[0] = T1 + Ts; Chris@10: { Chris@10: E TC, TE, Tv, TD, Tt, Tu; Chris@10: TC = FMA(KP951056516, Ty, KP587785252 * TB); Chris@10: TE = FNMS(KP587785252, Ty, KP951056516 * TB); Chris@10: Tt = KP559016994 * (Ti - Tr); Chris@10: Tu = FNMS(KP250000000, Ts, T1); Chris@10: Tv = Tt + Tu; Chris@10: TD = Tu - Tt; Chris@10: ci[0] = Tv - TC; Chris@10: ci[WS(rs, 1)] = TD + TE; Chris@10: cr[WS(rs, 1)] = Tv + TC; Chris@10: cr[WS(rs, 2)] = TD - TE; Chris@10: } Chris@10: ci[WS(rs, 4)] = TL + TI; Chris@10: { Chris@10: E TH, TP, TO, TQ, TM, TN; Chris@10: TH = FMA(KP587785252, TF, KP951056516 * TG); Chris@10: TP = FNMS(KP587785252, TG, KP951056516 * TF); Chris@10: TM = FNMS(KP250000000, TL, TI); Chris@10: TN = KP559016994 * (TJ - TK); Chris@10: TO = TM - TN; Chris@10: TQ = TN + TM; Chris@10: cr[WS(rs, 3)] = TH - TO; Chris@10: ci[WS(rs, 3)] = TP + TQ; Chris@10: ci[WS(rs, 2)] = TH + TO; Chris@10: cr[WS(rs, 4)] = TP - TQ; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_CEXP, 1, 1}, Chris@10: {TW_CEXP, 1, 3}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2hc_desc desc = { 5, "hf2_5", twinstr, &GENUS, {30, 18, 14, 0} }; Chris@10: Chris@10: void X(codelet_hf2_5) (planner *p) { Chris@10: X(khc2hc_register) (p, hf2_5, &desc); Chris@10: } Chris@10: #endif /* HAVE_FMA */