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 07:39:50 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 -n 5 -dit -name hf_5 -include hf.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 40 FP additions, 34 FP multiplications, cannam@95: * (or, 14 additions, 8 multiplications, 26 fused multiply/add), cannam@95: * 43 stack variables, 4 constants, and 20 memory accesses cannam@95: */ cannam@95: #include "hf.h" cannam@95: cannam@95: static void hf_5(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@95: DK(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@95: DK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@95: DK(KP618033988, +0.618033988749894848204586834365638117720309180); cannam@95: { cannam@95: INT m; cannam@95: 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)) { cannam@95: E T1, TJ, TK, TA, TR, Te, TC, Tk, TE, Tq; cannam@95: { cannam@95: E Tg, Tj, Tm, TB, Th, Tp, Tl, Ti, To, TD, Tn; cannam@95: T1 = cr[0]; cannam@95: TJ = ci[0]; cannam@95: { cannam@95: E T9, Tc, Ty, Ta, Tb, Tx, T7, Tf, Tz, Td; cannam@95: { cannam@95: E T3, T6, T8, Tw, T4, T2, T5; cannam@95: T3 = cr[WS(rs, 1)]; cannam@95: T6 = ci[WS(rs, 1)]; cannam@95: T2 = W[0]; cannam@95: T9 = cr[WS(rs, 4)]; cannam@95: Tc = ci[WS(rs, 4)]; cannam@95: T8 = W[6]; cannam@95: Tw = T2 * T6; cannam@95: T4 = T2 * T3; cannam@95: T5 = W[1]; cannam@95: Ty = T8 * Tc; cannam@95: Ta = T8 * T9; cannam@95: Tb = W[7]; cannam@95: Tx = FNMS(T5, T3, Tw); cannam@95: T7 = FMA(T5, T6, T4); cannam@95: } cannam@95: Tg = cr[WS(rs, 2)]; cannam@95: Tz = FNMS(Tb, T9, Ty); cannam@95: Td = FMA(Tb, Tc, Ta); cannam@95: Tj = ci[WS(rs, 2)]; cannam@95: Tf = W[2]; cannam@95: TK = Tx + Tz; cannam@95: TA = Tx - Tz; cannam@95: TR = Td - T7; cannam@95: Te = T7 + Td; cannam@95: Tm = cr[WS(rs, 3)]; cannam@95: TB = Tf * Tj; cannam@95: Th = Tf * Tg; cannam@95: Tp = ci[WS(rs, 3)]; cannam@95: Tl = W[4]; cannam@95: Ti = W[3]; cannam@95: To = W[5]; cannam@95: } cannam@95: TD = Tl * Tp; cannam@95: Tn = Tl * Tm; cannam@95: TC = FNMS(Ti, Tg, TB); cannam@95: Tk = FMA(Ti, Tj, Th); cannam@95: TE = FNMS(To, Tm, TD); cannam@95: Tq = FMA(To, Tp, Tn); cannam@95: } cannam@95: { cannam@95: E TG, TI, TO, TS, TU, Tu, TN, Tt, TL, TF; cannam@95: TL = TC + TE; cannam@95: TF = TC - TE; cannam@95: { cannam@95: E Tr, TQ, TM, Ts; cannam@95: Tr = Tk + Tq; cannam@95: TQ = Tk - Tq; cannam@95: TG = FMA(KP618033988, TF, TA); cannam@95: TI = FNMS(KP618033988, TA, TF); cannam@95: TO = TK - TL; cannam@95: TM = TK + TL; cannam@95: TS = FMA(KP618033988, TR, TQ); cannam@95: TU = FNMS(KP618033988, TQ, TR); cannam@95: Tu = Te - Tr; cannam@95: Ts = Te + Tr; cannam@95: ci[WS(rs, 4)] = TM + TJ; cannam@95: TN = FNMS(KP250000000, TM, TJ); cannam@95: cr[0] = T1 + Ts; cannam@95: Tt = FNMS(KP250000000, Ts, T1); cannam@95: } cannam@95: { cannam@95: E TT, TP, Tv, TH; cannam@95: TT = FMA(KP559016994, TO, TN); cannam@95: TP = FNMS(KP559016994, TO, TN); cannam@95: Tv = FMA(KP559016994, Tu, Tt); cannam@95: TH = FNMS(KP559016994, Tu, Tt); cannam@95: ci[WS(rs, 2)] = FMA(KP951056516, TS, TP); cannam@95: cr[WS(rs, 3)] = FMS(KP951056516, TS, TP); cannam@95: ci[WS(rs, 3)] = FMA(KP951056516, TU, TT); cannam@95: cr[WS(rs, 4)] = FMS(KP951056516, TU, TT); cannam@95: ci[WS(rs, 1)] = FMA(KP951056516, TI, TH); cannam@95: cr[WS(rs, 2)] = FNMS(KP951056516, TI, TH); cannam@95: cr[WS(rs, 1)] = FMA(KP951056516, TG, Tv); cannam@95: ci[0] = FNMS(KP951056516, TG, Tv); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_FULL, 1, 5}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2hc_desc desc = { 5, "hf_5", twinstr, &GENUS, {14, 8, 26, 0} }; cannam@95: cannam@95: void X(codelet_hf_5) (planner *p) { cannam@95: X(khc2hc_register) (p, hf_5, &desc); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -n 5 -dit -name hf_5 -include hf.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 40 FP additions, 28 FP multiplications, cannam@95: * (or, 26 additions, 14 multiplications, 14 fused multiply/add), cannam@95: * 29 stack variables, 4 constants, and 20 memory accesses cannam@95: */ cannam@95: #include "hf.h" cannam@95: cannam@95: static void hf_5(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@95: DK(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@95: DK(KP587785252, +0.587785252292473129168705954639072768597652438); cannam@95: DK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@95: { cannam@95: INT m; cannam@95: 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)) { cannam@95: E T1, TE, Tu, Tx, TC, TB, TF, TG, TH, Tc, Tn, To; cannam@95: T1 = cr[0]; cannam@95: TE = ci[0]; cannam@95: { cannam@95: E T6, Ts, Tm, Tw, Tb, Tt, Th, Tv; cannam@95: { cannam@95: E T3, T5, T2, T4; cannam@95: T3 = cr[WS(rs, 1)]; cannam@95: T5 = ci[WS(rs, 1)]; cannam@95: T2 = W[0]; cannam@95: T4 = W[1]; cannam@95: T6 = FMA(T2, T3, T4 * T5); cannam@95: Ts = FNMS(T4, T3, T2 * T5); cannam@95: } cannam@95: { cannam@95: E Tj, Tl, Ti, Tk; cannam@95: Tj = cr[WS(rs, 3)]; cannam@95: Tl = ci[WS(rs, 3)]; cannam@95: Ti = W[4]; cannam@95: Tk = W[5]; cannam@95: Tm = FMA(Ti, Tj, Tk * Tl); cannam@95: Tw = FNMS(Tk, Tj, Ti * Tl); cannam@95: } cannam@95: { cannam@95: E T8, Ta, T7, T9; cannam@95: T8 = cr[WS(rs, 4)]; cannam@95: Ta = ci[WS(rs, 4)]; cannam@95: T7 = W[6]; cannam@95: T9 = W[7]; cannam@95: Tb = FMA(T7, T8, T9 * Ta); cannam@95: Tt = FNMS(T9, T8, T7 * Ta); cannam@95: } cannam@95: { cannam@95: E Te, Tg, Td, Tf; cannam@95: Te = cr[WS(rs, 2)]; cannam@95: Tg = ci[WS(rs, 2)]; cannam@95: Td = W[2]; cannam@95: Tf = W[3]; cannam@95: Th = FMA(Td, Te, Tf * Tg); cannam@95: Tv = FNMS(Tf, Te, Td * Tg); cannam@95: } cannam@95: Tu = Ts - Tt; cannam@95: Tx = Tv - Tw; cannam@95: TC = Th - Tm; cannam@95: TB = Tb - T6; cannam@95: TF = Ts + Tt; cannam@95: TG = Tv + Tw; cannam@95: TH = TF + TG; cannam@95: Tc = T6 + Tb; cannam@95: Tn = Th + Tm; cannam@95: To = Tc + Tn; cannam@95: } cannam@95: cr[0] = T1 + To; cannam@95: { cannam@95: E Ty, TA, Tr, Tz, Tp, Tq; cannam@95: Ty = FMA(KP951056516, Tu, KP587785252 * Tx); cannam@95: TA = FNMS(KP587785252, Tu, KP951056516 * Tx); cannam@95: Tp = KP559016994 * (Tc - Tn); cannam@95: Tq = FNMS(KP250000000, To, T1); cannam@95: Tr = Tp + Tq; cannam@95: Tz = Tq - Tp; cannam@95: ci[0] = Tr - Ty; cannam@95: ci[WS(rs, 1)] = Tz + TA; cannam@95: cr[WS(rs, 1)] = Tr + Ty; cannam@95: cr[WS(rs, 2)] = Tz - TA; cannam@95: } cannam@95: ci[WS(rs, 4)] = TH + TE; cannam@95: { cannam@95: E TD, TL, TK, TM, TI, TJ; cannam@95: TD = FMA(KP587785252, TB, KP951056516 * TC); cannam@95: TL = FNMS(KP587785252, TC, KP951056516 * TB); cannam@95: TI = FNMS(KP250000000, TH, TE); cannam@95: TJ = KP559016994 * (TF - TG); cannam@95: TK = TI - TJ; cannam@95: TM = TJ + TI; cannam@95: cr[WS(rs, 3)] = TD - TK; cannam@95: ci[WS(rs, 3)] = TL + TM; cannam@95: ci[WS(rs, 2)] = TD + TK; cannam@95: cr[WS(rs, 4)] = TL - TM; cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_FULL, 1, 5}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2hc_desc desc = { 5, "hf_5", twinstr, &GENUS, {26, 14, 14, 0} }; cannam@95: cannam@95: void X(codelet_hf_5) (planner *p) { cannam@95: X(khc2hc_register) (p, hf_5, &desc); cannam@95: } cannam@95: #endif /* HAVE_FMA */