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 16:36:06 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 -n 5 -name t1_5 -include t.h */ cannam@127: cannam@127: /* cannam@127: * This function contains 40 FP additions, 34 FP multiplications, cannam@127: * (or, 14 additions, 8 multiplications, 26 fused multiply/add), cannam@127: * 43 stack variables, 4 constants, and 20 memory accesses cannam@127: */ cannam@127: #include "t.h" cannam@127: cannam@127: static void t1_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 * 8); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 8, MAKE_VOLATILE_STRIDE(10, rs)) { cannam@127: E T1, TM, TJ, TA, TQ, Te, TC, Tk, TE, Tq; cannam@127: { cannam@127: E Tg, Tj, Tm, TB, Th, Tp, Tl, Ti, To, TD, Tn; cannam@127: T1 = ri[0]; cannam@127: TM = ii[0]; cannam@127: { cannam@127: E T9, Tc, Ty, Ta, Tb, Tx, T7, Tf, Tz, Td; cannam@127: { cannam@127: E T3, T6, T8, Tw, T4, T2, T5; cannam@127: T3 = ri[WS(rs, 1)]; cannam@127: T6 = ii[WS(rs, 1)]; cannam@127: T2 = W[0]; cannam@127: T9 = ri[WS(rs, 4)]; cannam@127: Tc = ii[WS(rs, 4)]; cannam@127: T8 = W[6]; cannam@127: Tw = T2 * T6; cannam@127: T4 = T2 * T3; cannam@127: T5 = W[1]; cannam@127: Ty = T8 * Tc; cannam@127: Ta = T8 * T9; cannam@127: Tb = W[7]; cannam@127: Tx = FNMS(T5, T3, Tw); cannam@127: T7 = FMA(T5, T6, T4); cannam@127: } cannam@127: Tg = ri[WS(rs, 2)]; cannam@127: Tz = FNMS(Tb, T9, Ty); cannam@127: Td = FMA(Tb, Tc, Ta); cannam@127: Tj = ii[WS(rs, 2)]; cannam@127: Tf = W[2]; cannam@127: TJ = Tx + Tz; cannam@127: TA = Tx - Tz; cannam@127: TQ = T7 - Td; cannam@127: Te = T7 + Td; cannam@127: Tm = ri[WS(rs, 3)]; cannam@127: TB = Tf * Tj; cannam@127: Th = Tf * Tg; cannam@127: Tp = ii[WS(rs, 3)]; cannam@127: Tl = W[4]; cannam@127: Ti = W[3]; cannam@127: To = W[5]; cannam@127: } cannam@127: TD = Tl * Tp; cannam@127: Tn = Tl * Tm; cannam@127: TC = FNMS(Ti, Tg, TB); cannam@127: Tk = FMA(Ti, Tj, Th); cannam@127: TE = FNMS(To, Tm, TD); cannam@127: Tq = FMA(To, Tp, Tn); cannam@127: } cannam@127: { cannam@127: E TG, TI, TO, TS, TU, Tu, TN, Tt, TK, TF; cannam@127: TK = TC + TE; cannam@127: TF = TC - TE; cannam@127: { cannam@127: E Tr, TR, TL, Ts; cannam@127: Tr = Tk + Tq; cannam@127: TR = Tk - Tq; cannam@127: TG = FMA(KP618033988, TF, TA); cannam@127: TI = FNMS(KP618033988, TA, TF); cannam@127: TO = TJ - TK; cannam@127: TL = TJ + TK; cannam@127: TS = FMA(KP618033988, TR, TQ); cannam@127: TU = FNMS(KP618033988, TQ, TR); cannam@127: Tu = Te - Tr; cannam@127: Ts = Te + Tr; cannam@127: ii[0] = TL + TM; cannam@127: TN = FNMS(KP250000000, TL, TM); cannam@127: ri[0] = T1 + Ts; cannam@127: Tt = FNMS(KP250000000, Ts, T1); cannam@127: } cannam@127: { cannam@127: E TT, TP, TH, Tv; cannam@127: TT = FNMS(KP559016994, TO, TN); cannam@127: TP = FMA(KP559016994, TO, TN); cannam@127: TH = FNMS(KP559016994, Tu, Tt); cannam@127: Tv = FMA(KP559016994, Tu, Tt); cannam@127: ii[WS(rs, 4)] = FMA(KP951056516, TS, TP); cannam@127: ii[WS(rs, 1)] = FNMS(KP951056516, TS, TP); cannam@127: ii[WS(rs, 3)] = FNMS(KP951056516, TU, TT); cannam@127: ii[WS(rs, 2)] = FMA(KP951056516, TU, TT); cannam@127: ri[WS(rs, 1)] = FMA(KP951056516, TG, Tv); cannam@127: ri[WS(rs, 4)] = FNMS(KP951056516, TG, Tv); cannam@127: ri[WS(rs, 3)] = FMA(KP951056516, TI, TH); cannam@127: ri[WS(rs, 2)] = FNMS(KP951056516, TI, TH); cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: {TW_FULL, 0, 5}, cannam@127: {TW_NEXT, 1, 0} cannam@127: }; cannam@127: cannam@127: static const ct_desc desc = { 5, "t1_5", twinstr, &GENUS, {14, 8, 26, 0}, 0, 0, 0 }; cannam@127: cannam@127: void X(codelet_t1_5) (planner *p) { cannam@127: X(kdft_dit_register) (p, t1_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 -n 5 -name t1_5 -include t.h */ cannam@127: cannam@127: /* cannam@127: * This function contains 40 FP additions, 28 FP multiplications, cannam@127: * (or, 26 additions, 14 multiplications, 14 fused multiply/add), cannam@127: * 29 stack variables, 4 constants, and 20 memory accesses cannam@127: */ cannam@127: #include "t.h" cannam@127: cannam@127: static void t1_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 * 8); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 8, MAKE_VOLATILE_STRIDE(10, rs)) { cannam@127: E T1, TE, Tu, Tx, TJ, TI, TB, TC, TD, Tc, Tn, To; cannam@127: T1 = ri[0]; cannam@127: TE = ii[0]; cannam@127: { cannam@127: E T6, Ts, Tm, Tw, Tb, Tt, Th, Tv; cannam@127: { cannam@127: E T3, T5, T2, T4; cannam@127: T3 = ri[WS(rs, 1)]; cannam@127: T5 = ii[WS(rs, 1)]; cannam@127: T2 = W[0]; cannam@127: T4 = W[1]; cannam@127: T6 = FMA(T2, T3, T4 * T5); cannam@127: Ts = FNMS(T4, T3, T2 * T5); cannam@127: } cannam@127: { cannam@127: E Tj, Tl, Ti, Tk; cannam@127: Tj = ri[WS(rs, 3)]; cannam@127: Tl = ii[WS(rs, 3)]; cannam@127: Ti = W[4]; cannam@127: Tk = W[5]; cannam@127: Tm = FMA(Ti, Tj, Tk * Tl); cannam@127: Tw = FNMS(Tk, Tj, Ti * Tl); cannam@127: } cannam@127: { cannam@127: E T8, Ta, T7, T9; cannam@127: T8 = ri[WS(rs, 4)]; cannam@127: Ta = ii[WS(rs, 4)]; cannam@127: T7 = W[6]; cannam@127: T9 = W[7]; cannam@127: Tb = FMA(T7, T8, T9 * Ta); cannam@127: Tt = FNMS(T9, T8, T7 * Ta); cannam@127: } cannam@127: { cannam@127: E Te, Tg, Td, Tf; cannam@127: Te = ri[WS(rs, 2)]; cannam@127: Tg = ii[WS(rs, 2)]; cannam@127: Td = W[2]; cannam@127: Tf = W[3]; cannam@127: Th = FMA(Td, Te, Tf * Tg); cannam@127: Tv = FNMS(Tf, Te, Td * Tg); cannam@127: } cannam@127: Tu = Ts - Tt; cannam@127: Tx = Tv - Tw; cannam@127: TJ = Th - Tm; cannam@127: TI = T6 - Tb; cannam@127: TB = Ts + Tt; cannam@127: TC = Tv + Tw; cannam@127: TD = TB + TC; cannam@127: Tc = T6 + Tb; cannam@127: Tn = Th + Tm; cannam@127: To = Tc + Tn; cannam@127: } cannam@127: ri[0] = T1 + To; cannam@127: ii[0] = TD + TE; cannam@127: { cannam@127: E Ty, TA, Tr, Tz, Tp, Tq; cannam@127: Ty = FMA(KP951056516, Tu, KP587785252 * Tx); cannam@127: TA = FNMS(KP587785252, Tu, KP951056516 * Tx); cannam@127: Tp = KP559016994 * (Tc - Tn); cannam@127: Tq = FNMS(KP250000000, To, T1); cannam@127: Tr = Tp + Tq; cannam@127: Tz = Tq - Tp; cannam@127: ri[WS(rs, 4)] = Tr - Ty; cannam@127: ri[WS(rs, 3)] = Tz + TA; cannam@127: ri[WS(rs, 1)] = Tr + Ty; cannam@127: ri[WS(rs, 2)] = Tz - TA; cannam@127: } cannam@127: { cannam@127: E TK, TL, TH, TM, TF, TG; cannam@127: TK = FMA(KP951056516, TI, KP587785252 * TJ); cannam@127: TL = FNMS(KP587785252, TI, KP951056516 * TJ); cannam@127: TF = KP559016994 * (TB - TC); cannam@127: TG = FNMS(KP250000000, TD, TE); cannam@127: TH = TF + TG; cannam@127: TM = TG - TF; cannam@127: ii[WS(rs, 1)] = TH - TK; cannam@127: ii[WS(rs, 3)] = TM - TL; cannam@127: ii[WS(rs, 4)] = TK + TH; cannam@127: ii[WS(rs, 2)] = TL + TM; cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: {TW_FULL, 0, 5}, cannam@127: {TW_NEXT, 1, 0} cannam@127: }; cannam@127: cannam@127: static const ct_desc desc = { 5, "t1_5", twinstr, &GENUS, {26, 14, 14, 0}, 0, 0, 0 }; cannam@127: cannam@127: void X(codelet_t1_5) (planner *p) { cannam@127: X(kdft_dit_register) (p, t1_5, &desc); cannam@127: } cannam@127: #endif /* HAVE_FMA */