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 16:48:40 EDT 2016 */ cannam@127: cannam@127: #include "codelet-rdft.h" cannam@127: cannam@127: #ifdef HAVE_FMA cannam@127: cannam@127: /* 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 */ cannam@127: cannam@127: /* cannam@127: * This function contains 30 FP additions, 20 FP multiplications, cannam@127: * (or, 24 additions, 14 multiplications, 6 fused multiply/add), cannam@127: * 32 stack variables, 1 constants, and 16 memory accesses cannam@127: */ cannam@127: #include "hc2cf.h" cannam@127: cannam@127: static void hc2cfdft_4(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@127: { cannam@127: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@127: { cannam@127: INT m; cannam@127: 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)) { cannam@127: E Td, Tu, Tr, T4, Tm, To, T9, T5, TA, Tp, Tv, TD, T6, Tq; cannam@127: { cannam@127: E Tk, Tl, Tf, TC, Tj, T7, T8, T1, Tn, Tb, Tc; cannam@127: Tb = Ip[0]; cannam@127: Tc = Im[0]; cannam@127: { cannam@127: E Ti, Tg, Th, T2, T3; cannam@127: Tg = Rm[0]; cannam@127: Th = Rp[0]; cannam@127: Tk = W[1]; cannam@127: Tl = Tb + Tc; cannam@127: Td = Tb - Tc; cannam@127: Tu = Th + Tg; cannam@127: Ti = Tg - Th; cannam@127: Tf = W[0]; cannam@127: T2 = Ip[WS(rs, 1)]; cannam@127: T3 = Im[WS(rs, 1)]; cannam@127: TC = Tk * Ti; cannam@127: Tj = Tf * Ti; cannam@127: T7 = Rp[WS(rs, 1)]; cannam@127: Tr = T2 + T3; cannam@127: T4 = T2 - T3; cannam@127: T8 = Rm[WS(rs, 1)]; cannam@127: T1 = W[2]; cannam@127: Tn = W[4]; cannam@127: } cannam@127: Tm = FNMS(Tk, Tl, Tj); cannam@127: To = T7 - T8; cannam@127: T9 = T7 + T8; cannam@127: T5 = T1 * T4; cannam@127: TA = Tn * Tr; cannam@127: Tp = Tn * To; cannam@127: Tv = T1 * T9; cannam@127: TD = FMA(Tf, Tl, TC); cannam@127: T6 = W[3]; cannam@127: Tq = W[5]; cannam@127: } cannam@127: { cannam@127: E Tw, Ta, TB, Ts; cannam@127: Tw = FMA(T6, T4, Tv); cannam@127: Ta = FNMS(T6, T9, T5); cannam@127: TB = FNMS(Tq, To, TA); cannam@127: Ts = FMA(Tq, Tr, Tp); cannam@127: { cannam@127: E TF, Tx, Te, Tz; cannam@127: TF = Tu + Tw; cannam@127: Tx = Tu - Tw; cannam@127: Te = Ta + Td; cannam@127: Tz = Td - Ta; cannam@127: { cannam@127: E TG, TE, Tt, Ty; cannam@127: TG = TB + TD; cannam@127: TE = TB - TD; cannam@127: Tt = Tm - Ts; cannam@127: Ty = Ts + Tm; cannam@127: Im[0] = KP500000000 * (TE - Tz); cannam@127: Ip[WS(rs, 1)] = KP500000000 * (Tz + TE); cannam@127: Rp[0] = KP500000000 * (TF + TG); cannam@127: Rm[WS(rs, 1)] = KP500000000 * (TF - TG); cannam@127: Rp[WS(rs, 1)] = KP500000000 * (Tx + Ty); cannam@127: Rm[0] = KP500000000 * (Tx - Ty); cannam@127: Im[WS(rs, 1)] = KP500000000 * (Tt - Te); cannam@127: Ip[0] = KP500000000 * (Te + Tt); cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: {TW_FULL, 1, 4}, cannam@127: {TW_NEXT, 1, 0} cannam@127: }; cannam@127: cannam@127: static const hc2c_desc desc = { 4, "hc2cfdft_4", twinstr, &GENUS, {24, 14, 6, 0} }; cannam@127: cannam@127: void X(codelet_hc2cfdft_4) (planner *p) { cannam@127: X(khc2c_register) (p, hc2cfdft_4, &desc, HC2C_VIA_DFT); cannam@127: } cannam@127: #else /* HAVE_FMA */ cannam@127: cannam@127: /* Generated by: ../../../genfft/gen_hc2cdft.native -compact -variables 4 -pipeline-latency 4 -n 4 -dit -name hc2cfdft_4 -include hc2cf.h */ cannam@127: cannam@127: /* cannam@127: * This function contains 30 FP additions, 20 FP multiplications, cannam@127: * (or, 24 additions, 14 multiplications, 6 fused multiply/add), cannam@127: * 18 stack variables, 1 constants, and 16 memory accesses cannam@127: */ cannam@127: #include "hc2cf.h" cannam@127: cannam@127: static void hc2cfdft_4(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@127: { cannam@127: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@127: { cannam@127: INT m; cannam@127: 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)) { cannam@127: E Tc, Tr, Tk, Tx, T9, Ts, Tp, Tw; cannam@127: { cannam@127: E Ta, Tb, Tj, Tf, Tg, Th, Te, Ti; cannam@127: Ta = Ip[0]; cannam@127: Tb = Im[0]; cannam@127: Tj = Ta + Tb; cannam@127: Tf = Rm[0]; cannam@127: Tg = Rp[0]; cannam@127: Th = Tf - Tg; cannam@127: Tc = Ta - Tb; cannam@127: Tr = Tg + Tf; cannam@127: Te = W[0]; cannam@127: Ti = W[1]; cannam@127: Tk = FNMS(Ti, Tj, Te * Th); cannam@127: Tx = FMA(Ti, Th, Te * Tj); cannam@127: } cannam@127: { cannam@127: E T4, To, T8, Tm; cannam@127: { cannam@127: E T2, T3, T6, T7; cannam@127: T2 = Ip[WS(rs, 1)]; cannam@127: T3 = Im[WS(rs, 1)]; cannam@127: T4 = T2 - T3; cannam@127: To = T2 + T3; cannam@127: T6 = Rp[WS(rs, 1)]; cannam@127: T7 = Rm[WS(rs, 1)]; cannam@127: T8 = T6 + T7; cannam@127: Tm = T6 - T7; cannam@127: } cannam@127: { cannam@127: E T1, T5, Tl, Tn; cannam@127: T1 = W[2]; cannam@127: T5 = W[3]; cannam@127: T9 = FNMS(T5, T8, T1 * T4); cannam@127: Ts = FMA(T1, T8, T5 * T4); cannam@127: Tl = W[4]; cannam@127: Tn = W[5]; cannam@127: Tp = FMA(Tl, Tm, Tn * To); cannam@127: Tw = FNMS(Tn, Tm, Tl * To); cannam@127: } cannam@127: } cannam@127: { cannam@127: E Td, Tq, Tz, TA; cannam@127: Td = T9 + Tc; cannam@127: Tq = Tk - Tp; cannam@127: Ip[0] = KP500000000 * (Td + Tq); cannam@127: Im[WS(rs, 1)] = KP500000000 * (Tq - Td); cannam@127: Tz = Tr + Ts; cannam@127: TA = Tw + Tx; cannam@127: Rm[WS(rs, 1)] = KP500000000 * (Tz - TA); cannam@127: Rp[0] = KP500000000 * (Tz + TA); cannam@127: } cannam@127: { cannam@127: E Tt, Tu, Tv, Ty; cannam@127: Tt = Tr - Ts; cannam@127: Tu = Tp + Tk; cannam@127: Rm[0] = KP500000000 * (Tt - Tu); cannam@127: Rp[WS(rs, 1)] = KP500000000 * (Tt + Tu); cannam@127: Tv = Tc - T9; cannam@127: Ty = Tw - Tx; cannam@127: Ip[WS(rs, 1)] = KP500000000 * (Tv + Ty); cannam@127: Im[0] = KP500000000 * (Ty - Tv); cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: {TW_FULL, 1, 4}, cannam@127: {TW_NEXT, 1, 0} cannam@127: }; cannam@127: cannam@127: static const hc2c_desc desc = { 4, "hc2cfdft_4", twinstr, &GENUS, {24, 14, 6, 0} }; cannam@127: cannam@127: void X(codelet_hc2cfdft_4) (planner *p) { cannam@127: X(khc2c_register) (p, hc2cfdft_4, &desc, HC2C_VIA_DFT); cannam@127: } cannam@127: #endif /* HAVE_FMA */