cannam@95: /* cannam@95: * Copyright (c) 2003, 2007-11 Matteo Frigo cannam@95: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology cannam@95: * cannam@95: * This program is free software; you can redistribute it and/or modify cannam@95: * it under the terms of the GNU General Public License as published by cannam@95: * the Free Software Foundation; either version 2 of the License, or cannam@95: * (at your option) any later version. cannam@95: * cannam@95: * This program is distributed in the hope that it will be useful, cannam@95: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@95: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@95: * GNU General Public License for more details. cannam@95: * cannam@95: * You should have received a copy of the GNU General Public License cannam@95: * along with this program; if not, write to the Free Software cannam@95: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@95: * cannam@95: */ cannam@95: cannam@95: /* This file was automatically generated --- DO NOT EDIT */ cannam@95: /* Generated on Sun Nov 25 07:40:50 EST 2012 */ cannam@95: cannam@95: #include "codelet-rdft.h" cannam@95: cannam@95: #ifdef HAVE_FMA cannam@95: cannam@95: /* 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 */ cannam@95: cannam@95: /* cannam@95: * This function contains 32 FP additions, 24 FP multiplications, cannam@95: * (or, 24 additions, 16 multiplications, 8 fused multiply/add), cannam@95: * 33 stack variables, 1 constants, and 16 memory accesses cannam@95: */ cannam@95: #include "hc2cf.h" cannam@95: cannam@95: static void hc2cfdft2_4(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: { cannam@95: INT m; cannam@95: 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)) { cannam@95: E T1, T5, T2, T4; cannam@95: T1 = W[0]; cannam@95: T5 = W[3]; cannam@95: T2 = W[2]; cannam@95: T4 = W[1]; cannam@95: { cannam@95: E Tc, T6, Tp, Tj, Tw, Tt, T9, TE, To, TC, Ta, Tr, Tf, Tl, Tm; cannam@95: { cannam@95: E Th, Tb, T3, Ti; cannam@95: Th = Ip[0]; cannam@95: Tb = T1 * T5; cannam@95: T3 = T1 * T2; cannam@95: Ti = Im[0]; cannam@95: Tl = Rm[0]; cannam@95: Tc = FNMS(T4, T2, Tb); cannam@95: T6 = FMA(T4, T5, T3); cannam@95: Tp = Th + Ti; cannam@95: Tj = Th - Ti; cannam@95: Tm = Rp[0]; cannam@95: } cannam@95: { cannam@95: E T7, T8, Td, Tn, Te; cannam@95: T7 = Ip[WS(rs, 1)]; cannam@95: T8 = Im[WS(rs, 1)]; cannam@95: Td = Rp[WS(rs, 1)]; cannam@95: Tw = Tm + Tl; cannam@95: Tn = Tl - Tm; cannam@95: Tt = T7 + T8; cannam@95: T9 = T7 - T8; cannam@95: Te = Rm[WS(rs, 1)]; cannam@95: TE = T4 * Tn; cannam@95: To = T1 * Tn; cannam@95: TC = T2 * Tt; cannam@95: Ta = T6 * T9; cannam@95: Tr = Td - Te; cannam@95: Tf = Td + Te; cannam@95: } cannam@95: { cannam@95: E Tq, Tk, TB, Ty, Tu, TI, TG, TF; cannam@95: Tq = FNMS(T4, Tp, To); cannam@95: TF = FMA(T1, Tp, TE); cannam@95: { cannam@95: E Tg, Tx, TD, Ts; cannam@95: Tg = FNMS(Tc, Tf, Ta); cannam@95: Tx = T6 * Tf; cannam@95: TD = FNMS(T5, Tr, TC); cannam@95: Ts = T2 * Tr; cannam@95: Tk = Tg + Tj; cannam@95: TB = Tj - Tg; cannam@95: Ty = FMA(Tc, T9, Tx); cannam@95: Tu = FMA(T5, Tt, Ts); cannam@95: TI = TD + TF; cannam@95: TG = TD - TF; cannam@95: } cannam@95: { cannam@95: E Tz, TH, Tv, TA; cannam@95: Tz = Tw - Ty; cannam@95: TH = Tw + Ty; cannam@95: Tv = Tq - Tu; cannam@95: TA = Tu + Tq; cannam@95: Rp[0] = KP500000000 * (TH + TI); cannam@95: Rm[WS(rs, 1)] = KP500000000 * (TH - TI); cannam@95: Rm[0] = KP500000000 * (Tz - TA); cannam@95: Im[WS(rs, 1)] = KP500000000 * (Tv - Tk); cannam@95: Ip[0] = KP500000000 * (Tk + Tv); cannam@95: Im[0] = KP500000000 * (TG - TB); cannam@95: Rp[WS(rs, 1)] = KP500000000 * (Tz + TA); cannam@95: Ip[WS(rs, 1)] = KP500000000 * (TB + TG); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_CEXP, 1, 1}, cannam@95: {TW_CEXP, 1, 3}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 4, "hc2cfdft2_4", twinstr, &GENUS, {24, 16, 8, 0} }; cannam@95: cannam@95: void X(codelet_hc2cfdft2_4) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cfdft2_4, &desc, HC2C_VIA_DFT); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* 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 */ cannam@95: cannam@95: /* cannam@95: * This function contains 32 FP additions, 24 FP multiplications, cannam@95: * (or, 24 additions, 16 multiplications, 8 fused multiply/add), cannam@95: * 24 stack variables, 1 constants, and 16 memory accesses cannam@95: */ cannam@95: #include "hc2cf.h" cannam@95: cannam@95: static void hc2cfdft2_4(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: { cannam@95: INT m; cannam@95: 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)) { cannam@95: E T1, T3, T2, T4, T5, T9; cannam@95: T1 = W[0]; cannam@95: T3 = W[1]; cannam@95: T2 = W[2]; cannam@95: T4 = W[3]; cannam@95: T5 = FMA(T1, T2, T3 * T4); cannam@95: T9 = FNMS(T3, T2, T1 * T4); cannam@95: { cannam@95: E Tg, Tr, Tm, Tx, Td, Tw, Tp, Ts; cannam@95: { cannam@95: E Te, Tf, Tl, Ti, Tj, Tk; cannam@95: Te = Ip[0]; cannam@95: Tf = Im[0]; cannam@95: Tl = Te + Tf; cannam@95: Ti = Rm[0]; cannam@95: Tj = Rp[0]; cannam@95: Tk = Ti - Tj; cannam@95: Tg = Te - Tf; cannam@95: Tr = Tj + Ti; cannam@95: Tm = FNMS(T3, Tl, T1 * Tk); cannam@95: Tx = FMA(T3, Tk, T1 * Tl); cannam@95: } cannam@95: { cannam@95: E T8, To, Tc, Tn; cannam@95: { cannam@95: E T6, T7, Ta, Tb; cannam@95: T6 = Ip[WS(rs, 1)]; cannam@95: T7 = Im[WS(rs, 1)]; cannam@95: T8 = T6 - T7; cannam@95: To = T6 + T7; cannam@95: Ta = Rp[WS(rs, 1)]; cannam@95: Tb = Rm[WS(rs, 1)]; cannam@95: Tc = Ta + Tb; cannam@95: Tn = Ta - Tb; cannam@95: } cannam@95: Td = FNMS(T9, Tc, T5 * T8); cannam@95: Tw = FNMS(T4, Tn, T2 * To); cannam@95: Tp = FMA(T2, Tn, T4 * To); cannam@95: Ts = FMA(T5, Tc, T9 * T8); cannam@95: } cannam@95: { cannam@95: E Th, Tq, Tz, TA; cannam@95: Th = Td + Tg; cannam@95: Tq = Tm - Tp; cannam@95: Ip[0] = KP500000000 * (Th + Tq); cannam@95: Im[WS(rs, 1)] = KP500000000 * (Tq - Th); cannam@95: Tz = Tr + Ts; cannam@95: TA = Tw + Tx; cannam@95: Rm[WS(rs, 1)] = KP500000000 * (Tz - TA); cannam@95: Rp[0] = KP500000000 * (Tz + TA); cannam@95: } cannam@95: { cannam@95: E Tt, Tu, Tv, Ty; cannam@95: Tt = Tr - Ts; cannam@95: Tu = Tp + Tm; cannam@95: Rm[0] = KP500000000 * (Tt - Tu); cannam@95: Rp[WS(rs, 1)] = KP500000000 * (Tt + Tu); cannam@95: Tv = Tg - Td; cannam@95: Ty = Tw - Tx; cannam@95: Ip[WS(rs, 1)] = KP500000000 * (Tv + Ty); cannam@95: Im[0] = KP500000000 * (Ty - Tv); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_CEXP, 1, 1}, cannam@95: {TW_CEXP, 1, 3}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 4, "hc2cfdft2_4", twinstr, &GENUS, {24, 16, 8, 0} }; cannam@95: cannam@95: void X(codelet_hc2cfdft2_4) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cfdft2_4, &desc, HC2C_VIA_DFT); cannam@95: } cannam@95: #endif /* HAVE_FMA */