cannam@127: /* cannam@127: * Copyright (c) 2003, 2007-14 Matteo Frigo cannam@127: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology cannam@127: * cannam@127: * This program is free software; you can redistribute it and/or modify cannam@127: * it under the terms of the GNU General Public License as published by cannam@127: * the Free Software Foundation; either version 2 of the License, or cannam@127: * (at your option) any later version. cannam@127: * cannam@127: * This program is distributed in the hope that it will be useful, cannam@127: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@127: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@127: * GNU General Public License for more details. cannam@127: * cannam@127: * You should have received a copy of the GNU General Public License cannam@127: * along with this program; if not, write to the Free Software cannam@127: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@127: * cannam@127: */ cannam@127: cannam@127: /* This file was automatically generated --- DO NOT EDIT */ cannam@127: /* Generated on Sat Jul 30 16:37:11 EDT 2016 */ cannam@127: cannam@127: #include "codelet-dft.h" cannam@127: cannam@127: #ifdef HAVE_FMA cannam@127: cannam@127: /* 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@127: cannam@127: /* cannam@127: * This function contains 44 FP additions, 40 FP multiplications, cannam@127: * (or, 14 additions, 10 multiplications, 30 fused multiply/add), cannam@127: * 47 stack variables, 4 constants, and 20 memory accesses cannam@127: */ cannam@127: #include "t.h" cannam@127: cannam@127: static void t2_5(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@127: { cannam@127: DK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@127: DK(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@127: DK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@127: DK(KP618033988, +0.618033988749894848204586834365638117720309180); cannam@127: { cannam@127: INT m; cannam@127: 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@127: E Ta, T1, TO, Tp, TS, Ti, TL, TC, To, TE, Ts, TF, T2, T8, T5; cannam@127: E TT, Tt, TG; cannam@127: T2 = W[0]; cannam@127: Ta = W[3]; cannam@127: T8 = W[2]; cannam@127: T5 = W[1]; cannam@127: { cannam@127: E Tq, Tr, Te, T9; cannam@127: T1 = ri[0]; cannam@127: Te = T2 * Ta; cannam@127: T9 = T2 * T8; cannam@127: TO = ii[0]; cannam@127: { cannam@127: E T3, Tf, Tm, Tj, Tb, T4, T6, Tc, Tg; cannam@127: T3 = ri[WS(rs, 1)]; cannam@127: Tf = FMA(T5, T8, Te); cannam@127: Tm = FNMS(T5, T8, Te); cannam@127: Tj = FMA(T5, Ta, T9); cannam@127: Tb = FNMS(T5, Ta, T9); cannam@127: T4 = T2 * T3; cannam@127: T6 = ii[WS(rs, 1)]; cannam@127: Tc = ri[WS(rs, 4)]; cannam@127: Tg = ii[WS(rs, 4)]; cannam@127: { cannam@127: E Tk, Tl, Tn, TD; cannam@127: { cannam@127: E T7, Tz, Th, TB, Ty, Td, TA; cannam@127: Tk = ri[WS(rs, 2)]; cannam@127: T7 = FMA(T5, T6, T4); cannam@127: Ty = T2 * T6; cannam@127: Td = Tb * Tc; cannam@127: TA = Tb * Tg; cannam@127: Tl = Tj * Tk; cannam@127: Tz = FNMS(T5, T3, Ty); cannam@127: Th = FMA(Tf, Tg, Td); cannam@127: TB = FNMS(Tf, Tc, TA); cannam@127: Tn = ii[WS(rs, 2)]; cannam@127: Tp = ri[WS(rs, 3)]; cannam@127: TS = T7 - Th; cannam@127: Ti = T7 + Th; cannam@127: TL = Tz + TB; cannam@127: TC = Tz - TB; cannam@127: TD = Tj * Tn; cannam@127: Tq = T8 * Tp; cannam@127: Tr = ii[WS(rs, 3)]; cannam@127: } cannam@127: To = FMA(Tm, Tn, Tl); cannam@127: TE = FNMS(Tm, Tk, TD); cannam@127: } cannam@127: } cannam@127: Ts = FMA(Ta, Tr, Tq); cannam@127: TF = T8 * Tr; cannam@127: } cannam@127: TT = To - Ts; cannam@127: Tt = To + Ts; cannam@127: TG = FNMS(Ta, Tp, TF); cannam@127: { cannam@127: E TU, TW, TV, TR, Tw, Tu; cannam@127: TU = FMA(KP618033988, TT, TS); cannam@127: TW = FNMS(KP618033988, TS, TT); cannam@127: Tw = Ti - Tt; cannam@127: Tu = Ti + Tt; cannam@127: { cannam@127: E TM, TH, Tv, TI, TK; cannam@127: TM = TE + TG; cannam@127: TH = TE - TG; cannam@127: ri[0] = T1 + Tu; cannam@127: Tv = FNMS(KP250000000, Tu, T1); cannam@127: TI = FMA(KP618033988, TH, TC); cannam@127: TK = FNMS(KP618033988, TC, TH); cannam@127: { cannam@127: E TQ, TN, TJ, Tx, TP; cannam@127: TQ = TL - TM; cannam@127: TN = TL + TM; cannam@127: TJ = FNMS(KP559016994, Tw, Tv); cannam@127: Tx = FMA(KP559016994, Tw, Tv); cannam@127: ii[0] = TN + TO; cannam@127: TP = FNMS(KP250000000, TN, TO); cannam@127: ri[WS(rs, 1)] = FMA(KP951056516, TI, Tx); cannam@127: ri[WS(rs, 4)] = FNMS(KP951056516, TI, Tx); cannam@127: ri[WS(rs, 3)] = FMA(KP951056516, TK, TJ); cannam@127: ri[WS(rs, 2)] = FNMS(KP951056516, TK, TJ); cannam@127: TV = FNMS(KP559016994, TQ, TP); cannam@127: TR = FMA(KP559016994, TQ, TP); cannam@127: } cannam@127: } cannam@127: ii[WS(rs, 4)] = FMA(KP951056516, TU, TR); cannam@127: ii[WS(rs, 1)] = FNMS(KP951056516, TU, TR); cannam@127: ii[WS(rs, 3)] = FNMS(KP951056516, TW, TV); cannam@127: ii[WS(rs, 2)] = FMA(KP951056516, TW, TV); cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: {TW_CEXP, 0, 1}, cannam@127: {TW_CEXP, 0, 3}, cannam@127: {TW_NEXT, 1, 0} cannam@127: }; cannam@127: cannam@127: static const ct_desc desc = { 5, "t2_5", twinstr, &GENUS, {14, 10, 30, 0}, 0, 0, 0 }; cannam@127: cannam@127: void X(codelet_t2_5) (planner *p) { cannam@127: X(kdft_dit_register) (p, t2_5, &desc); cannam@127: } cannam@127: #else /* HAVE_FMA */ cannam@127: cannam@127: /* 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@127: cannam@127: /* cannam@127: * This function contains 44 FP additions, 32 FP multiplications, cannam@127: * (or, 30 additions, 18 multiplications, 14 fused multiply/add), cannam@127: * 37 stack variables, 4 constants, and 20 memory accesses cannam@127: */ cannam@127: #include "t.h" cannam@127: cannam@127: static void t2_5(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@127: { cannam@127: DK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@127: DK(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@127: DK(KP587785252, +0.587785252292473129168705954639072768597652438); cannam@127: DK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@127: { cannam@127: INT m; cannam@127: 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@127: E T2, T4, T7, T9, Tb, Tl, Tf, Tj; cannam@127: { cannam@127: E T8, Te, Ta, Td; cannam@127: T2 = W[0]; cannam@127: T4 = W[1]; cannam@127: T7 = W[2]; cannam@127: T9 = W[3]; cannam@127: T8 = T2 * T7; cannam@127: Te = T4 * T7; cannam@127: Ta = T4 * T9; cannam@127: Td = T2 * T9; cannam@127: Tb = T8 - Ta; cannam@127: Tl = Td - Te; cannam@127: Tf = Td + Te; cannam@127: Tj = T8 + Ta; cannam@127: } cannam@127: { cannam@127: E T1, TI, Ty, TB, TN, TM, TF, TG, TH, Ti, Tr, Ts; cannam@127: T1 = ri[0]; cannam@127: TI = ii[0]; cannam@127: { cannam@127: E T6, Tw, Tq, TA, Th, Tx, Tn, Tz; cannam@127: { cannam@127: E T3, T5, To, Tp; cannam@127: T3 = ri[WS(rs, 1)]; cannam@127: T5 = ii[WS(rs, 1)]; cannam@127: T6 = FMA(T2, T3, T4 * T5); cannam@127: Tw = FNMS(T4, T3, T2 * T5); cannam@127: To = ri[WS(rs, 3)]; cannam@127: Tp = ii[WS(rs, 3)]; cannam@127: Tq = FMA(T7, To, T9 * Tp); cannam@127: TA = FNMS(T9, To, T7 * Tp); cannam@127: } cannam@127: { cannam@127: E Tc, Tg, Tk, Tm; cannam@127: Tc = ri[WS(rs, 4)]; cannam@127: Tg = ii[WS(rs, 4)]; cannam@127: Th = FMA(Tb, Tc, Tf * Tg); cannam@127: Tx = FNMS(Tf, Tc, Tb * Tg); cannam@127: Tk = ri[WS(rs, 2)]; cannam@127: Tm = ii[WS(rs, 2)]; cannam@127: Tn = FMA(Tj, Tk, Tl * Tm); cannam@127: Tz = FNMS(Tl, Tk, Tj * Tm); cannam@127: } cannam@127: Ty = Tw - Tx; cannam@127: TB = Tz - TA; cannam@127: TN = Tn - Tq; cannam@127: TM = T6 - Th; cannam@127: TF = Tw + Tx; cannam@127: TG = Tz + TA; cannam@127: TH = TF + TG; cannam@127: Ti = T6 + Th; cannam@127: Tr = Tn + Tq; cannam@127: Ts = Ti + Tr; cannam@127: } cannam@127: ri[0] = T1 + Ts; cannam@127: ii[0] = TH + TI; cannam@127: { cannam@127: E TC, TE, Tv, TD, Tt, Tu; cannam@127: TC = FMA(KP951056516, Ty, KP587785252 * TB); cannam@127: TE = FNMS(KP587785252, Ty, KP951056516 * TB); cannam@127: Tt = KP559016994 * (Ti - Tr); cannam@127: Tu = FNMS(KP250000000, Ts, T1); cannam@127: Tv = Tt + Tu; cannam@127: TD = Tu - Tt; cannam@127: ri[WS(rs, 4)] = Tv - TC; cannam@127: ri[WS(rs, 3)] = TD + TE; cannam@127: ri[WS(rs, 1)] = Tv + TC; cannam@127: ri[WS(rs, 2)] = TD - TE; cannam@127: } cannam@127: { cannam@127: E TO, TP, TL, TQ, TJ, TK; cannam@127: TO = FMA(KP951056516, TM, KP587785252 * TN); cannam@127: TP = FNMS(KP587785252, TM, KP951056516 * TN); cannam@127: TJ = KP559016994 * (TF - TG); cannam@127: TK = FNMS(KP250000000, TH, TI); cannam@127: TL = TJ + TK; cannam@127: TQ = TK - TJ; cannam@127: ii[WS(rs, 1)] = TL - TO; cannam@127: ii[WS(rs, 3)] = TQ - TP; cannam@127: ii[WS(rs, 4)] = TO + TL; cannam@127: ii[WS(rs, 2)] = TP + TQ; cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: {TW_CEXP, 0, 1}, cannam@127: {TW_CEXP, 0, 3}, cannam@127: {TW_NEXT, 1, 0} cannam@127: }; cannam@127: cannam@127: static const ct_desc desc = { 5, "t2_5", twinstr, &GENUS, {30, 18, 14, 0}, 0, 0, 0 }; cannam@127: cannam@127: void X(codelet_t2_5) (planner *p) { cannam@127: X(kdft_dit_register) (p, t2_5, &desc); cannam@127: } cannam@127: #endif /* HAVE_FMA */