Chris@82: /* Chris@82: * Copyright (c) 2003, 2007-14 Matteo Frigo Chris@82: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology Chris@82: * Chris@82: * This program is free software; you can redistribute it and/or modify Chris@82: * it under the terms of the GNU General Public License as published by Chris@82: * the Free Software Foundation; either version 2 of the License, or Chris@82: * (at your option) any later version. Chris@82: * Chris@82: * This program is distributed in the hope that it will be useful, Chris@82: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@82: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@82: * GNU General Public License for more details. Chris@82: * Chris@82: * You should have received a copy of the GNU General Public License Chris@82: * along with this program; if not, write to the Free Software Chris@82: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@82: * Chris@82: */ Chris@82: Chris@82: /* This file was automatically generated --- DO NOT EDIT */ Chris@82: /* Generated on Thu May 24 08:07:10 EDT 2018 */ Chris@82: Chris@82: #include "rdft/codelet-rdft.h" Chris@82: Chris@82: #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA) Chris@82: Chris@82: /* Generated by: ../../../genfft/gen_hc2cdft.native -fma -compact -variables 4 -pipeline-latency 4 -n 4 -dit -name hc2cfdft_4 -include rdft/scalar/hc2cf.h */ Chris@82: Chris@82: /* Chris@82: * This function contains 30 FP additions, 20 FP multiplications, Chris@82: * (or, 24 additions, 14 multiplications, 6 fused multiply/add), Chris@82: * 31 stack variables, 1 constants, and 16 memory accesses Chris@82: */ Chris@82: #include "rdft/scalar/hc2cf.h" Chris@82: Chris@82: static void hc2cfdft_4(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@82: { Chris@82: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@82: { Chris@82: INT m; Chris@82: 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@82: E Td, Tl, Tu, Tk, TC, Tf, Tj, T4, Tr, T9, To, T5, Tv, Tp, TA; Chris@82: E Tb, Tc; Chris@82: Tb = Ip[0]; Chris@82: Tc = Im[0]; Chris@82: Td = Tb - Tc; Chris@82: Tl = Tb + Tc; Chris@82: { Chris@82: E Tg, Th, Ti, T1, Tn; Chris@82: Tg = Rm[0]; Chris@82: Th = Rp[0]; Chris@82: Ti = Tg - Th; Chris@82: Tu = Th + Tg; Chris@82: Tk = W[1]; Chris@82: TC = Tk * Ti; Chris@82: Tf = W[0]; Chris@82: Tj = Tf * Ti; Chris@82: { Chris@82: E T2, T3, T7, T8; Chris@82: T2 = Ip[WS(rs, 1)]; Chris@82: T3 = Im[WS(rs, 1)]; Chris@82: T4 = T2 - T3; Chris@82: Tr = T2 + T3; Chris@82: T7 = Rp[WS(rs, 1)]; Chris@82: T8 = Rm[WS(rs, 1)]; Chris@82: T9 = T7 + T8; Chris@82: To = T7 - T8; Chris@82: } Chris@82: T1 = W[2]; Chris@82: T5 = T1 * T4; Chris@82: Tv = T1 * T9; Chris@82: Tn = W[4]; Chris@82: Tp = Tn * To; Chris@82: TA = Tn * Tr; Chris@82: } Chris@82: { Chris@82: E Tm, TD, Ta, Tw, Ts, TB, T6, Tq; Chris@82: Tm = FNMS(Tk, Tl, Tj); Chris@82: TD = FMA(Tf, Tl, TC); Chris@82: T6 = W[3]; Chris@82: Ta = FNMS(T6, T9, T5); Chris@82: Tw = FMA(T6, T4, Tv); Chris@82: Tq = W[5]; Chris@82: Ts = FMA(Tq, Tr, Tp); Chris@82: TB = FNMS(Tq, To, TA); Chris@82: { Chris@82: E Te, Tt, TF, TG; Chris@82: Te = Ta + Td; Chris@82: Tt = Tm - Ts; Chris@82: Ip[0] = KP500000000 * (Te + Tt); Chris@82: Im[WS(rs, 1)] = KP500000000 * (Tt - Te); Chris@82: TF = Tu + Tw; Chris@82: TG = TB + TD; Chris@82: Rm[WS(rs, 1)] = KP500000000 * (TF - TG); Chris@82: Rp[0] = KP500000000 * (TF + TG); Chris@82: } Chris@82: { Chris@82: E Tx, Ty, Tz, TE; Chris@82: Tx = Tu - Tw; Chris@82: Ty = Ts + Tm; Chris@82: Rm[0] = KP500000000 * (Tx - Ty); Chris@82: Rp[WS(rs, 1)] = KP500000000 * (Tx + Ty); Chris@82: Tz = Td - Ta; Chris@82: TE = TB - TD; Chris@82: Ip[WS(rs, 1)] = KP500000000 * (Tz + TE); Chris@82: Im[0] = KP500000000 * (TE - Tz); Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: Chris@82: static const tw_instr twinstr[] = { Chris@82: {TW_FULL, 1, 4}, Chris@82: {TW_NEXT, 1, 0} Chris@82: }; Chris@82: Chris@82: static const hc2c_desc desc = { 4, "hc2cfdft_4", twinstr, &GENUS, {24, 14, 6, 0} }; Chris@82: Chris@82: void X(codelet_hc2cfdft_4) (planner *p) { Chris@82: X(khc2c_register) (p, hc2cfdft_4, &desc, HC2C_VIA_DFT); Chris@82: } Chris@82: #else Chris@82: Chris@82: /* Generated by: ../../../genfft/gen_hc2cdft.native -compact -variables 4 -pipeline-latency 4 -n 4 -dit -name hc2cfdft_4 -include rdft/scalar/hc2cf.h */ Chris@82: Chris@82: /* Chris@82: * This function contains 30 FP additions, 20 FP multiplications, Chris@82: * (or, 24 additions, 14 multiplications, 6 fused multiply/add), Chris@82: * 18 stack variables, 1 constants, and 16 memory accesses Chris@82: */ Chris@82: #include "rdft/scalar/hc2cf.h" Chris@82: Chris@82: static void hc2cfdft_4(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@82: { Chris@82: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@82: { Chris@82: INT m; Chris@82: 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@82: E Tc, Tr, Tk, Tx, T9, Ts, Tp, Tw; Chris@82: { Chris@82: E Ta, Tb, Tj, Tf, Tg, Th, Te, Ti; Chris@82: Ta = Ip[0]; Chris@82: Tb = Im[0]; Chris@82: Tj = Ta + Tb; Chris@82: Tf = Rm[0]; Chris@82: Tg = Rp[0]; Chris@82: Th = Tf - Tg; Chris@82: Tc = Ta - Tb; Chris@82: Tr = Tg + Tf; Chris@82: Te = W[0]; Chris@82: Ti = W[1]; Chris@82: Tk = FNMS(Ti, Tj, Te * Th); Chris@82: Tx = FMA(Ti, Th, Te * Tj); Chris@82: } Chris@82: { Chris@82: E T4, To, T8, Tm; Chris@82: { Chris@82: E T2, T3, T6, T7; Chris@82: T2 = Ip[WS(rs, 1)]; Chris@82: T3 = Im[WS(rs, 1)]; Chris@82: T4 = T2 - T3; Chris@82: To = T2 + T3; Chris@82: T6 = Rp[WS(rs, 1)]; Chris@82: T7 = Rm[WS(rs, 1)]; Chris@82: T8 = T6 + T7; Chris@82: Tm = T6 - T7; Chris@82: } Chris@82: { Chris@82: E T1, T5, Tl, Tn; Chris@82: T1 = W[2]; Chris@82: T5 = W[3]; Chris@82: T9 = FNMS(T5, T8, T1 * T4); Chris@82: Ts = FMA(T1, T8, T5 * T4); Chris@82: Tl = W[4]; Chris@82: Tn = W[5]; Chris@82: Tp = FMA(Tl, Tm, Tn * To); Chris@82: Tw = FNMS(Tn, Tm, Tl * To); Chris@82: } Chris@82: } Chris@82: { Chris@82: E Td, Tq, Tz, TA; Chris@82: Td = T9 + Tc; Chris@82: Tq = Tk - Tp; Chris@82: Ip[0] = KP500000000 * (Td + Tq); Chris@82: Im[WS(rs, 1)] = KP500000000 * (Tq - Td); Chris@82: Tz = Tr + Ts; Chris@82: TA = Tw + Tx; Chris@82: Rm[WS(rs, 1)] = KP500000000 * (Tz - TA); Chris@82: Rp[0] = KP500000000 * (Tz + TA); Chris@82: } Chris@82: { Chris@82: E Tt, Tu, Tv, Ty; Chris@82: Tt = Tr - Ts; Chris@82: Tu = Tp + Tk; Chris@82: Rm[0] = KP500000000 * (Tt - Tu); Chris@82: Rp[WS(rs, 1)] = KP500000000 * (Tt + Tu); Chris@82: Tv = Tc - T9; Chris@82: Ty = Tw - Tx; Chris@82: Ip[WS(rs, 1)] = KP500000000 * (Tv + Ty); Chris@82: Im[0] = KP500000000 * (Ty - Tv); Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: Chris@82: static const tw_instr twinstr[] = { Chris@82: {TW_FULL, 1, 4}, Chris@82: {TW_NEXT, 1, 0} Chris@82: }; Chris@82: Chris@82: static const hc2c_desc desc = { 4, "hc2cfdft_4", twinstr, &GENUS, {24, 14, 6, 0} }; Chris@82: Chris@82: void X(codelet_hc2cfdft_4) (planner *p) { Chris@82: X(khc2c_register) (p, hc2cfdft_4, &desc, HC2C_VIA_DFT); Chris@82: } Chris@82: #endif