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 07:40:50 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 -twiddle-log3 -precompute-twiddles -n 4 -dit -name hc2cfdft2_4 -include hc2cf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 32 FP additions, 24 FP multiplications, Chris@10: * (or, 24 additions, 16 multiplications, 8 fused multiply/add), Chris@10: * 33 stack variables, 1 constants, and 16 memory accesses Chris@10: */ Chris@10: #include "hc2cf.h" Chris@10: Chris@10: static void hc2cfdft2_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) * 4); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 4, MAKE_VOLATILE_STRIDE(16, rs)) { Chris@10: E T1, T5, T2, T4; Chris@10: T1 = W[0]; Chris@10: T5 = W[3]; Chris@10: T2 = W[2]; Chris@10: T4 = W[1]; Chris@10: { Chris@10: E Tc, T6, Tp, Tj, Tw, Tt, T9, TE, To, TC, Ta, Tr, Tf, Tl, Tm; Chris@10: { Chris@10: E Th, Tb, T3, Ti; Chris@10: Th = Ip[0]; Chris@10: Tb = T1 * T5; Chris@10: T3 = T1 * T2; Chris@10: Ti = Im[0]; Chris@10: Tl = Rm[0]; Chris@10: Tc = FNMS(T4, T2, Tb); Chris@10: T6 = FMA(T4, T5, T3); Chris@10: Tp = Th + Ti; Chris@10: Tj = Th - Ti; Chris@10: Tm = Rp[0]; Chris@10: } Chris@10: { Chris@10: E T7, T8, Td, Tn, Te; Chris@10: T7 = Ip[WS(rs, 1)]; Chris@10: T8 = Im[WS(rs, 1)]; Chris@10: Td = Rp[WS(rs, 1)]; Chris@10: Tw = Tm + Tl; Chris@10: Tn = Tl - Tm; Chris@10: Tt = T7 + T8; Chris@10: T9 = T7 - T8; Chris@10: Te = Rm[WS(rs, 1)]; Chris@10: TE = T4 * Tn; Chris@10: To = T1 * Tn; Chris@10: TC = T2 * Tt; Chris@10: Ta = T6 * T9; Chris@10: Tr = Td - Te; Chris@10: Tf = Td + Te; Chris@10: } Chris@10: { Chris@10: E Tq, Tk, TB, Ty, Tu, TI, TG, TF; Chris@10: Tq = FNMS(T4, Tp, To); Chris@10: TF = FMA(T1, Tp, TE); Chris@10: { Chris@10: E Tg, Tx, TD, Ts; Chris@10: Tg = FNMS(Tc, Tf, Ta); Chris@10: Tx = T6 * Tf; Chris@10: TD = FNMS(T5, Tr, TC); Chris@10: Ts = T2 * Tr; Chris@10: Tk = Tg + Tj; Chris@10: TB = Tj - Tg; Chris@10: Ty = FMA(Tc, T9, Tx); Chris@10: Tu = FMA(T5, Tt, Ts); Chris@10: TI = TD + TF; Chris@10: TG = TD - TF; Chris@10: } Chris@10: { Chris@10: E Tz, TH, Tv, TA; Chris@10: Tz = Tw - Ty; Chris@10: TH = Tw + Ty; Chris@10: Tv = Tq - Tu; Chris@10: TA = Tu + Tq; Chris@10: Rp[0] = KP500000000 * (TH + TI); Chris@10: Rm[WS(rs, 1)] = KP500000000 * (TH - TI); Chris@10: Rm[0] = KP500000000 * (Tz - TA); Chris@10: Im[WS(rs, 1)] = KP500000000 * (Tv - Tk); Chris@10: Ip[0] = KP500000000 * (Tk + Tv); Chris@10: Im[0] = KP500000000 * (TG - TB); Chris@10: Rp[WS(rs, 1)] = KP500000000 * (Tz + TA); Chris@10: Ip[WS(rs, 1)] = KP500000000 * (TB + TG); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_CEXP, 1, 1}, Chris@10: {TW_CEXP, 1, 3}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 4, "hc2cfdft2_4", twinstr, &GENUS, {24, 16, 8, 0} }; Chris@10: Chris@10: void X(codelet_hc2cfdft2_4) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cfdft2_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 -twiddle-log3 -precompute-twiddles -n 4 -dit -name hc2cfdft2_4 -include hc2cf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 32 FP additions, 24 FP multiplications, Chris@10: * (or, 24 additions, 16 multiplications, 8 fused multiply/add), Chris@10: * 24 stack variables, 1 constants, and 16 memory accesses Chris@10: */ Chris@10: #include "hc2cf.h" Chris@10: Chris@10: static void hc2cfdft2_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) * 4); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 4, MAKE_VOLATILE_STRIDE(16, rs)) { Chris@10: E T1, T3, T2, T4, T5, T9; Chris@10: T1 = W[0]; Chris@10: T3 = W[1]; Chris@10: T2 = W[2]; Chris@10: T4 = W[3]; Chris@10: T5 = FMA(T1, T2, T3 * T4); Chris@10: T9 = FNMS(T3, T2, T1 * T4); Chris@10: { Chris@10: E Tg, Tr, Tm, Tx, Td, Tw, Tp, Ts; Chris@10: { Chris@10: E Te, Tf, Tl, Ti, Tj, Tk; Chris@10: Te = Ip[0]; Chris@10: Tf = Im[0]; Chris@10: Tl = Te + Tf; Chris@10: Ti = Rm[0]; Chris@10: Tj = Rp[0]; Chris@10: Tk = Ti - Tj; Chris@10: Tg = Te - Tf; Chris@10: Tr = Tj + Ti; Chris@10: Tm = FNMS(T3, Tl, T1 * Tk); Chris@10: Tx = FMA(T3, Tk, T1 * Tl); Chris@10: } Chris@10: { Chris@10: E T8, To, Tc, Tn; Chris@10: { Chris@10: E T6, T7, Ta, Tb; Chris@10: T6 = Ip[WS(rs, 1)]; Chris@10: T7 = Im[WS(rs, 1)]; Chris@10: T8 = T6 - T7; Chris@10: To = T6 + T7; Chris@10: Ta = Rp[WS(rs, 1)]; Chris@10: Tb = Rm[WS(rs, 1)]; Chris@10: Tc = Ta + Tb; Chris@10: Tn = Ta - Tb; Chris@10: } Chris@10: Td = FNMS(T9, Tc, T5 * T8); Chris@10: Tw = FNMS(T4, Tn, T2 * To); Chris@10: Tp = FMA(T2, Tn, T4 * To); Chris@10: Ts = FMA(T5, Tc, T9 * T8); Chris@10: } Chris@10: { Chris@10: E Th, Tq, Tz, TA; Chris@10: Th = Td + Tg; Chris@10: Tq = Tm - Tp; Chris@10: Ip[0] = KP500000000 * (Th + Tq); Chris@10: Im[WS(rs, 1)] = KP500000000 * (Tq - Th); 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 + Tm; Chris@10: Rm[0] = KP500000000 * (Tt - Tu); Chris@10: Rp[WS(rs, 1)] = KP500000000 * (Tt + Tu); Chris@10: Tv = Tg - Td; 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: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_CEXP, 1, 1}, Chris@10: {TW_CEXP, 1, 3}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 4, "hc2cfdft2_4", twinstr, &GENUS, {24, 16, 8, 0} }; Chris@10: Chris@10: void X(codelet_hc2cfdft2_4) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cfdft2_4, &desc, HC2C_VIA_DFT); Chris@10: } Chris@10: #endif /* HAVE_FMA */