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 07:40:44 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_hc2cdft.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 4 -dit -name hc2cfdft_4 -include hc2cf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 30 FP additions, 20 FP multiplications, Chris@10: * (or, 24 additions, 14 multiplications, 6 fused multiply/add), Chris@10: * 32 stack variables, 1 constants, and 16 memory accesses Chris@10: */ Chris@10: #include "hc2cf.h" Chris@10: Chris@10: static void hc2cfdft_4(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * 6); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 6, MAKE_VOLATILE_STRIDE(16, rs)) { Chris@10: E Td, Tu, Tr, T4, Tm, To, T9, T5, TA, Tp, Tv, TD, T6, Tq; Chris@10: { Chris@10: E Tk, Tl, Tf, TC, Tj, T7, T8, T1, Tn, Tb, Tc; Chris@10: Tb = Ip[0]; Chris@10: Tc = Im[0]; Chris@10: { Chris@10: E Ti, Tg, Th, T2, T3; Chris@10: Tg = Rm[0]; Chris@10: Th = Rp[0]; Chris@10: Tk = W[1]; Chris@10: Tl = Tb + Tc; Chris@10: Td = Tb - Tc; Chris@10: Tu = Th + Tg; Chris@10: Ti = Tg - Th; Chris@10: Tf = W[0]; Chris@10: T2 = Ip[WS(rs, 1)]; Chris@10: T3 = Im[WS(rs, 1)]; Chris@10: TC = Tk * Ti; Chris@10: Tj = Tf * Ti; Chris@10: T7 = Rp[WS(rs, 1)]; Chris@10: Tr = T2 + T3; Chris@10: T4 = T2 - T3; Chris@10: T8 = Rm[WS(rs, 1)]; Chris@10: T1 = W[2]; Chris@10: Tn = W[4]; Chris@10: } Chris@10: Tm = FNMS(Tk, Tl, Tj); Chris@10: To = T7 - T8; Chris@10: T9 = T7 + T8; Chris@10: T5 = T1 * T4; Chris@10: TA = Tn * Tr; Chris@10: Tp = Tn * To; Chris@10: Tv = T1 * T9; Chris@10: TD = FMA(Tf, Tl, TC); Chris@10: T6 = W[3]; Chris@10: Tq = W[5]; Chris@10: } Chris@10: { Chris@10: E Tw, Ta, TB, Ts; Chris@10: Tw = FMA(T6, T4, Tv); Chris@10: Ta = FNMS(T6, T9, T5); Chris@10: TB = FNMS(Tq, To, TA); Chris@10: Ts = FMA(Tq, Tr, Tp); Chris@10: { Chris@10: E TF, Tx, Te, Tz; Chris@10: TF = Tu + Tw; Chris@10: Tx = Tu - Tw; Chris@10: Te = Ta + Td; Chris@10: Tz = Td - Ta; Chris@10: { Chris@10: E TG, TE, Tt, Ty; Chris@10: TG = TB + TD; Chris@10: TE = TB - TD; Chris@10: Tt = Tm - Ts; Chris@10: Ty = Ts + Tm; Chris@10: Im[0] = KP500000000 * (TE - Tz); Chris@10: Ip[WS(rs, 1)] = KP500000000 * (Tz + TE); Chris@10: Rp[0] = KP500000000 * (TF + TG); Chris@10: Rm[WS(rs, 1)] = KP500000000 * (TF - TG); Chris@10: Rp[WS(rs, 1)] = KP500000000 * (Tx + Ty); Chris@10: Rm[0] = KP500000000 * (Tx - Ty); Chris@10: Im[WS(rs, 1)] = KP500000000 * (Tt - Te); Chris@10: Ip[0] = KP500000000 * (Te + Tt); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 4}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 4, "hc2cfdft_4", twinstr, &GENUS, {24, 14, 6, 0} }; Chris@10: Chris@10: void X(codelet_hc2cfdft_4) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cfdft_4, &desc, HC2C_VIA_DFT); Chris@10: } Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_hc2cdft.native -compact -variables 4 -pipeline-latency 4 -n 4 -dit -name hc2cfdft_4 -include hc2cf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 30 FP additions, 20 FP multiplications, Chris@10: * (or, 24 additions, 14 multiplications, 6 fused multiply/add), Chris@10: * 18 stack variables, 1 constants, and 16 memory accesses Chris@10: */ Chris@10: #include "hc2cf.h" Chris@10: Chris@10: static void hc2cfdft_4(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * 6); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 6, MAKE_VOLATILE_STRIDE(16, rs)) { Chris@10: E Tc, Tr, Tk, Tx, T9, Ts, Tp, Tw; Chris@10: { Chris@10: E Ta, Tb, Tj, Tf, Tg, Th, Te, Ti; Chris@10: Ta = Ip[0]; Chris@10: Tb = Im[0]; Chris@10: Tj = Ta + Tb; Chris@10: Tf = Rm[0]; Chris@10: Tg = Rp[0]; Chris@10: Th = Tf - Tg; Chris@10: Tc = Ta - Tb; Chris@10: Tr = Tg + Tf; Chris@10: Te = W[0]; Chris@10: Ti = W[1]; Chris@10: Tk = FNMS(Ti, Tj, Te * Th); Chris@10: Tx = FMA(Ti, Th, Te * Tj); Chris@10: } Chris@10: { Chris@10: E T4, To, T8, Tm; Chris@10: { Chris@10: E T2, T3, T6, T7; Chris@10: T2 = Ip[WS(rs, 1)]; Chris@10: T3 = Im[WS(rs, 1)]; Chris@10: T4 = T2 - T3; Chris@10: To = T2 + T3; Chris@10: T6 = Rp[WS(rs, 1)]; Chris@10: T7 = Rm[WS(rs, 1)]; Chris@10: T8 = T6 + T7; Chris@10: Tm = T6 - T7; Chris@10: } Chris@10: { Chris@10: E T1, T5, Tl, Tn; Chris@10: T1 = W[2]; Chris@10: T5 = W[3]; Chris@10: T9 = FNMS(T5, T8, T1 * T4); Chris@10: Ts = FMA(T1, T8, T5 * T4); Chris@10: Tl = W[4]; Chris@10: Tn = W[5]; Chris@10: Tp = FMA(Tl, Tm, Tn * To); Chris@10: Tw = FNMS(Tn, Tm, Tl * To); Chris@10: } Chris@10: } Chris@10: { Chris@10: E Td, Tq, Tz, TA; Chris@10: Td = T9 + Tc; Chris@10: Tq = Tk - Tp; Chris@10: Ip[0] = KP500000000 * (Td + Tq); Chris@10: Im[WS(rs, 1)] = KP500000000 * (Tq - Td); Chris@10: Tz = Tr + Ts; Chris@10: TA = Tw + Tx; Chris@10: Rm[WS(rs, 1)] = KP500000000 * (Tz - TA); Chris@10: Rp[0] = KP500000000 * (Tz + TA); Chris@10: } Chris@10: { Chris@10: E Tt, Tu, Tv, Ty; Chris@10: Tt = Tr - Ts; Chris@10: Tu = Tp + Tk; Chris@10: Rm[0] = KP500000000 * (Tt - Tu); Chris@10: Rp[WS(rs, 1)] = KP500000000 * (Tt + Tu); Chris@10: Tv = Tc - T9; Chris@10: Ty = Tw - Tx; Chris@10: Ip[WS(rs, 1)] = KP500000000 * (Tv + Ty); Chris@10: Im[0] = KP500000000 * (Ty - Tv); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 4}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 4, "hc2cfdft_4", twinstr, &GENUS, {24, 14, 6, 0} }; Chris@10: Chris@10: void X(codelet_hc2cfdft_4) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cfdft_4, &desc, HC2C_VIA_DFT); Chris@10: } Chris@10: #endif /* HAVE_FMA */