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 07:36:09 EST 2012 */ cannam@95: cannam@95: #include "codelet-dft.h" cannam@95: cannam@95: #ifdef HAVE_FMA cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_twiddle.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -twiddle-log3 -precompute-twiddles -n 5 -name t2_5 -include t.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 44 FP additions, 40 FP multiplications, cannam@95: * (or, 14 additions, 10 multiplications, 30 fused multiply/add), cannam@95: * 47 stack variables, 4 constants, and 20 memory accesses cannam@95: */ cannam@95: #include "t.h" cannam@95: cannam@95: static void t2_5(R *ri, R *ii, 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 * 4); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 4, MAKE_VOLATILE_STRIDE(10, rs)) { cannam@95: E Ta, T1, TO, Tp, TS, Ti, TL, TC, To, TE, Ts, TF, T2, T8, T5; cannam@95: E TT, Tt, TG; cannam@95: T2 = W[0]; cannam@95: Ta = W[3]; cannam@95: T8 = W[2]; cannam@95: T5 = W[1]; cannam@95: { cannam@95: E Tq, Tr, Te, T9; cannam@95: T1 = ri[0]; cannam@95: Te = T2 * Ta; cannam@95: T9 = T2 * T8; cannam@95: TO = ii[0]; cannam@95: { cannam@95: E T3, Tf, Tm, Tj, Tb, T4, T6, Tc, Tg; cannam@95: T3 = ri[WS(rs, 1)]; cannam@95: Tf = FMA(T5, T8, Te); cannam@95: Tm = FNMS(T5, T8, Te); cannam@95: Tj = FMA(T5, Ta, T9); cannam@95: Tb = FNMS(T5, Ta, T9); cannam@95: T4 = T2 * T3; cannam@95: T6 = ii[WS(rs, 1)]; cannam@95: Tc = ri[WS(rs, 4)]; cannam@95: Tg = ii[WS(rs, 4)]; cannam@95: { cannam@95: E Tk, Tl, Tn, TD; cannam@95: { cannam@95: E T7, Tz, Th, TB, Ty, Td, TA; cannam@95: Tk = ri[WS(rs, 2)]; cannam@95: T7 = FMA(T5, T6, T4); cannam@95: Ty = T2 * T6; cannam@95: Td = Tb * Tc; cannam@95: TA = Tb * Tg; cannam@95: Tl = Tj * Tk; cannam@95: Tz = FNMS(T5, T3, Ty); cannam@95: Th = FMA(Tf, Tg, Td); cannam@95: TB = FNMS(Tf, Tc, TA); cannam@95: Tn = ii[WS(rs, 2)]; cannam@95: Tp = ri[WS(rs, 3)]; cannam@95: TS = T7 - Th; cannam@95: Ti = T7 + Th; cannam@95: TL = Tz + TB; cannam@95: TC = Tz - TB; cannam@95: TD = Tj * Tn; cannam@95: Tq = T8 * Tp; cannam@95: Tr = ii[WS(rs, 3)]; cannam@95: } cannam@95: To = FMA(Tm, Tn, Tl); cannam@95: TE = FNMS(Tm, Tk, TD); cannam@95: } cannam@95: } cannam@95: Ts = FMA(Ta, Tr, Tq); cannam@95: TF = T8 * Tr; cannam@95: } cannam@95: TT = To - Ts; cannam@95: Tt = To + Ts; cannam@95: TG = FNMS(Ta, Tp, TF); cannam@95: { cannam@95: E TU, TW, TV, TR, Tw, Tu; cannam@95: TU = FMA(KP618033988, TT, TS); cannam@95: TW = FNMS(KP618033988, TS, TT); cannam@95: Tw = Ti - Tt; cannam@95: Tu = Ti + Tt; cannam@95: { cannam@95: E TM, TH, Tv, TI, TK; cannam@95: TM = TE + TG; cannam@95: TH = TE - TG; cannam@95: ri[0] = T1 + Tu; cannam@95: Tv = FNMS(KP250000000, Tu, T1); cannam@95: TI = FMA(KP618033988, TH, TC); cannam@95: TK = FNMS(KP618033988, TC, TH); cannam@95: { cannam@95: E TQ, TN, TJ, Tx, TP; cannam@95: TQ = TL - TM; cannam@95: TN = TL + TM; cannam@95: TJ = FNMS(KP559016994, Tw, Tv); cannam@95: Tx = FMA(KP559016994, Tw, Tv); cannam@95: ii[0] = TN + TO; cannam@95: TP = FNMS(KP250000000, TN, TO); cannam@95: ri[WS(rs, 1)] = FMA(KP951056516, TI, Tx); cannam@95: ri[WS(rs, 4)] = FNMS(KP951056516, TI, Tx); cannam@95: ri[WS(rs, 3)] = FMA(KP951056516, TK, TJ); cannam@95: ri[WS(rs, 2)] = FNMS(KP951056516, TK, TJ); cannam@95: TV = FNMS(KP559016994, TQ, TP); cannam@95: TR = FMA(KP559016994, TQ, TP); cannam@95: } cannam@95: } cannam@95: ii[WS(rs, 4)] = FMA(KP951056516, TU, TR); cannam@95: ii[WS(rs, 1)] = FNMS(KP951056516, TU, TR); cannam@95: ii[WS(rs, 3)] = FNMS(KP951056516, TW, TV); cannam@95: ii[WS(rs, 2)] = FMA(KP951056516, TW, TV); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_CEXP, 0, 1}, cannam@95: {TW_CEXP, 0, 3}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const ct_desc desc = { 5, "t2_5", twinstr, &GENUS, {14, 10, 30, 0}, 0, 0, 0 }; cannam@95: cannam@95: void X(codelet_t2_5) (planner *p) { cannam@95: X(kdft_dit_register) (p, t2_5, &desc); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_twiddle.native -compact -variables 4 -pipeline-latency 4 -twiddle-log3 -precompute-twiddles -n 5 -name t2_5 -include t.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 44 FP additions, 32 FP multiplications, cannam@95: * (or, 30 additions, 18 multiplications, 14 fused multiply/add), cannam@95: * 37 stack variables, 4 constants, and 20 memory accesses cannam@95: */ cannam@95: #include "t.h" cannam@95: cannam@95: static void t2_5(R *ri, R *ii, 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 * 4); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 4, MAKE_VOLATILE_STRIDE(10, rs)) { cannam@95: E T2, T4, T7, T9, Tb, Tl, Tf, Tj; cannam@95: { cannam@95: E T8, Te, Ta, Td; cannam@95: T2 = W[0]; cannam@95: T4 = W[1]; cannam@95: T7 = W[2]; cannam@95: T9 = W[3]; cannam@95: T8 = T2 * T7; cannam@95: Te = T4 * T7; cannam@95: Ta = T4 * T9; cannam@95: Td = T2 * T9; cannam@95: Tb = T8 - Ta; cannam@95: Tl = Td - Te; cannam@95: Tf = Td + Te; cannam@95: Tj = T8 + Ta; cannam@95: } cannam@95: { cannam@95: E T1, TI, Ty, TB, TN, TM, TF, TG, TH, Ti, Tr, Ts; cannam@95: T1 = ri[0]; cannam@95: TI = ii[0]; cannam@95: { cannam@95: E T6, Tw, Tq, TA, Th, Tx, Tn, Tz; cannam@95: { cannam@95: E T3, T5, To, Tp; cannam@95: T3 = ri[WS(rs, 1)]; cannam@95: T5 = ii[WS(rs, 1)]; cannam@95: T6 = FMA(T2, T3, T4 * T5); cannam@95: Tw = FNMS(T4, T3, T2 * T5); cannam@95: To = ri[WS(rs, 3)]; cannam@95: Tp = ii[WS(rs, 3)]; cannam@95: Tq = FMA(T7, To, T9 * Tp); cannam@95: TA = FNMS(T9, To, T7 * Tp); cannam@95: } cannam@95: { cannam@95: E Tc, Tg, Tk, Tm; cannam@95: Tc = ri[WS(rs, 4)]; cannam@95: Tg = ii[WS(rs, 4)]; cannam@95: Th = FMA(Tb, Tc, Tf * Tg); cannam@95: Tx = FNMS(Tf, Tc, Tb * Tg); cannam@95: Tk = ri[WS(rs, 2)]; cannam@95: Tm = ii[WS(rs, 2)]; cannam@95: Tn = FMA(Tj, Tk, Tl * Tm); cannam@95: Tz = FNMS(Tl, Tk, Tj * Tm); cannam@95: } cannam@95: Ty = Tw - Tx; cannam@95: TB = Tz - TA; cannam@95: TN = Tn - Tq; cannam@95: TM = T6 - Th; cannam@95: TF = Tw + Tx; cannam@95: TG = Tz + TA; cannam@95: TH = TF + TG; cannam@95: Ti = T6 + Th; cannam@95: Tr = Tn + Tq; cannam@95: Ts = Ti + Tr; cannam@95: } cannam@95: ri[0] = T1 + Ts; cannam@95: ii[0] = TH + TI; cannam@95: { cannam@95: E TC, TE, Tv, TD, Tt, Tu; cannam@95: TC = FMA(KP951056516, Ty, KP587785252 * TB); cannam@95: TE = FNMS(KP587785252, Ty, KP951056516 * TB); cannam@95: Tt = KP559016994 * (Ti - Tr); cannam@95: Tu = FNMS(KP250000000, Ts, T1); cannam@95: Tv = Tt + Tu; cannam@95: TD = Tu - Tt; cannam@95: ri[WS(rs, 4)] = Tv - TC; cannam@95: ri[WS(rs, 3)] = TD + TE; cannam@95: ri[WS(rs, 1)] = Tv + TC; cannam@95: ri[WS(rs, 2)] = TD - TE; cannam@95: } cannam@95: { cannam@95: E TO, TP, TL, TQ, TJ, TK; cannam@95: TO = FMA(KP951056516, TM, KP587785252 * TN); cannam@95: TP = FNMS(KP587785252, TM, KP951056516 * TN); cannam@95: TJ = KP559016994 * (TF - TG); cannam@95: TK = FNMS(KP250000000, TH, TI); cannam@95: TL = TJ + TK; cannam@95: TQ = TK - TJ; cannam@95: ii[WS(rs, 1)] = TL - TO; cannam@95: ii[WS(rs, 3)] = TQ - TP; cannam@95: ii[WS(rs, 4)] = TO + TL; cannam@95: ii[WS(rs, 2)] = TP + TQ; cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_CEXP, 0, 1}, cannam@95: {TW_CEXP, 0, 3}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const ct_desc desc = { 5, "t2_5", twinstr, &GENUS, {30, 18, 14, 0}, 0, 0, 0 }; cannam@95: cannam@95: void X(codelet_t2_5) (planner *p) { cannam@95: X(kdft_dit_register) (p, t2_5, &desc); cannam@95: } cannam@95: #endif /* HAVE_FMA */