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 07:42:07 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 -sign 1 -n 4 -dif -name hc2cbdft2_4 -include hc2cb.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 30 FP additions, 12 FP multiplications, Chris@10: * (or, 24 additions, 6 multiplications, 6 fused multiply/add), Chris@10: * 35 stack variables, 0 constants, and 16 memory accesses Chris@10: */ Chris@10: #include "hc2cb.h" Chris@10: Chris@10: static void hc2cbdft2_4(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { 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 Ty, TB, Tw, TE, TA, TF, Tz, TG, TC; Chris@10: { Chris@10: E T4, Tg, T3, Tm, Tc, T5, Th, Ti; Chris@10: { Chris@10: E T1, T2, Ta, Tb; Chris@10: T1 = Rp[0]; Chris@10: T2 = Rm[WS(rs, 1)]; Chris@10: Ta = Ip[0]; Chris@10: Tb = Im[WS(rs, 1)]; Chris@10: T4 = Rp[WS(rs, 1)]; Chris@10: Tg = T1 - T2; Chris@10: T3 = T1 + T2; Chris@10: Tm = Ta - Tb; Chris@10: Tc = Ta + Tb; Chris@10: T5 = Rm[0]; Chris@10: Th = Ip[WS(rs, 1)]; Chris@10: Ti = Im[0]; Chris@10: } Chris@10: { Chris@10: E T8, Td, T7, Ts, To, Tv, Tk, Te, Tf; Chris@10: T8 = W[0]; Chris@10: { Chris@10: E T9, T6, Tn, Tj; Chris@10: T9 = T4 - T5; Chris@10: T6 = T4 + T5; Chris@10: Tn = Th - Ti; Chris@10: Tj = Th + Ti; Chris@10: Ty = Tc - T9; Chris@10: Td = T9 + Tc; Chris@10: T7 = T3 + T6; Chris@10: Ts = T3 - T6; Chris@10: To = Tm + Tn; Chris@10: Tv = Tm - Tn; Chris@10: TB = Tg + Tj; Chris@10: Tk = Tg - Tj; Chris@10: Te = T8 * Td; Chris@10: } Chris@10: Tf = W[1]; Chris@10: { Chris@10: E Tr, Tu, Tt, TD, Tx, Tp, Tl, Tq; Chris@10: Tr = W[2]; Chris@10: Tp = T8 * Tk; Chris@10: Tu = W[3]; Chris@10: Tl = FMA(Tf, Tk, Te); Chris@10: Tt = Tr * Ts; Chris@10: Tq = FNMS(Tf, Td, Tp); Chris@10: TD = Tu * Ts; Chris@10: Rm[0] = T7 + Tl; Chris@10: Rp[0] = T7 - Tl; Chris@10: Im[0] = Tq - To; Chris@10: Ip[0] = To + Tq; Chris@10: Tx = W[4]; Chris@10: Tw = FNMS(Tu, Tv, Tt); Chris@10: TE = FMA(Tr, Tv, TD); Chris@10: TA = W[5]; Chris@10: TF = Tx * TB; Chris@10: Tz = Tx * Ty; Chris@10: } Chris@10: } Chris@10: } Chris@10: TG = FNMS(TA, Ty, TF); Chris@10: TC = FMA(TA, TB, Tz); Chris@10: Im[WS(rs, 1)] = TG - TE; Chris@10: Ip[WS(rs, 1)] = TE + TG; Chris@10: Rm[WS(rs, 1)] = Tw + TC; Chris@10: Rp[WS(rs, 1)] = Tw - TC; 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, "hc2cbdft2_4", twinstr, &GENUS, {24, 6, 6, 0} }; Chris@10: Chris@10: void X(codelet_hc2cbdft2_4) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cbdft2_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 -sign 1 -n 4 -dif -name hc2cbdft2_4 -include hc2cb.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 30 FP additions, 12 FP multiplications, Chris@10: * (or, 24 additions, 6 multiplications, 6 fused multiply/add), Chris@10: * 19 stack variables, 0 constants, and 16 memory accesses Chris@10: */ Chris@10: #include "hc2cb.h" Chris@10: Chris@10: static void hc2cbdft2_4(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { 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 T3, Tl, T6, Tm, Td, Tj, Tx, Tv, Ts, Tq; Chris@10: { Chris@10: E Tf, Tc, T9, Ti; Chris@10: { Chris@10: E T1, T2, Ta, Tb; Chris@10: T1 = Rp[0]; Chris@10: T2 = Rm[WS(rs, 1)]; Chris@10: T3 = T1 + T2; Chris@10: Tf = T1 - T2; Chris@10: Ta = Ip[0]; Chris@10: Tb = Im[WS(rs, 1)]; Chris@10: Tc = Ta + Tb; Chris@10: Tl = Ta - Tb; Chris@10: } Chris@10: { Chris@10: E T4, T5, Tg, Th; Chris@10: T4 = Rp[WS(rs, 1)]; Chris@10: T5 = Rm[0]; Chris@10: T6 = T4 + T5; Chris@10: T9 = T4 - T5; Chris@10: Tg = Ip[WS(rs, 1)]; Chris@10: Th = Im[0]; Chris@10: Ti = Tg + Th; Chris@10: Tm = Tg - Th; Chris@10: } Chris@10: Td = T9 + Tc; Chris@10: Tj = Tf - Ti; Chris@10: Tx = Tf + Ti; Chris@10: Tv = Tc - T9; Chris@10: Ts = Tl - Tm; Chris@10: Tq = T3 - T6; Chris@10: } Chris@10: { Chris@10: E T7, Tn, Tk, To, T8, Te; Chris@10: T7 = T3 + T6; Chris@10: Tn = Tl + Tm; Chris@10: T8 = W[0]; Chris@10: Te = W[1]; Chris@10: Tk = FMA(T8, Td, Te * Tj); Chris@10: To = FNMS(Te, Td, T8 * Tj); Chris@10: Rp[0] = T7 - Tk; Chris@10: Ip[0] = Tn + To; Chris@10: Rm[0] = T7 + Tk; Chris@10: Im[0] = To - Tn; Chris@10: } Chris@10: { Chris@10: E Tt, Tz, Ty, TA; Chris@10: { Chris@10: E Tp, Tr, Tu, Tw; Chris@10: Tp = W[2]; Chris@10: Tr = W[3]; Chris@10: Tt = FNMS(Tr, Ts, Tp * Tq); Chris@10: Tz = FMA(Tr, Tq, Tp * Ts); Chris@10: Tu = W[4]; Chris@10: Tw = W[5]; Chris@10: Ty = FMA(Tu, Tv, Tw * Tx); Chris@10: TA = FNMS(Tw, Tv, Tu * Tx); Chris@10: } Chris@10: Rp[WS(rs, 1)] = Tt - Ty; Chris@10: Ip[WS(rs, 1)] = Tz + TA; Chris@10: Rm[WS(rs, 1)] = Tt + Ty; Chris@10: Im[WS(rs, 1)] = TA - Tz; 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, "hc2cbdft2_4", twinstr, &GENUS, {24, 6, 6, 0} }; Chris@10: Chris@10: void X(codelet_hc2cbdft2_4) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cbdft2_4, &desc, HC2C_VIA_DFT); Chris@10: } Chris@10: #endif /* HAVE_FMA */