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 07:42:03 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 -sign 1 -n 4 -dif -name hc2cbdft_4 -include hc2cb.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 30 FP additions, 12 FP multiplications, cannam@95: * (or, 24 additions, 6 multiplications, 6 fused multiply/add), cannam@95: * 35 stack variables, 0 constants, and 16 memory accesses cannam@95: */ cannam@95: #include "hc2cb.h" cannam@95: cannam@95: static void hc2cbdft_4(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: { cannam@95: INT m; cannam@95: 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@95: E Ty, TB, Tw, TE, TA, TF, Tz, TG, TC; cannam@95: { cannam@95: E T4, Tg, T3, Tm, Tc, T5, Th, Ti; cannam@95: { cannam@95: E T1, T2, Ta, Tb; cannam@95: T1 = Rp[0]; cannam@95: T2 = Rm[WS(rs, 1)]; cannam@95: Ta = Ip[0]; cannam@95: Tb = Im[WS(rs, 1)]; cannam@95: T4 = Rp[WS(rs, 1)]; cannam@95: Tg = T1 - T2; cannam@95: T3 = T1 + T2; cannam@95: Tm = Ta - Tb; cannam@95: Tc = Ta + Tb; cannam@95: T5 = Rm[0]; cannam@95: Th = Ip[WS(rs, 1)]; cannam@95: Ti = Im[0]; cannam@95: } cannam@95: { cannam@95: E T8, Td, T7, Ts, To, Tv, Tk, Te, Tf; cannam@95: T8 = W[0]; cannam@95: { cannam@95: E T9, T6, Tn, Tj; cannam@95: T9 = T4 - T5; cannam@95: T6 = T4 + T5; cannam@95: Tn = Th - Ti; cannam@95: Tj = Th + Ti; cannam@95: Ty = Tc - T9; cannam@95: Td = T9 + Tc; cannam@95: T7 = T3 + T6; cannam@95: Ts = T3 - T6; cannam@95: To = Tm + Tn; cannam@95: Tv = Tm - Tn; cannam@95: TB = Tg + Tj; cannam@95: Tk = Tg - Tj; cannam@95: Te = T8 * Td; cannam@95: } cannam@95: Tf = W[1]; cannam@95: { cannam@95: E Tr, Tu, Tt, TD, Tx, Tp, Tl, Tq; cannam@95: Tr = W[2]; cannam@95: Tp = T8 * Tk; cannam@95: Tu = W[3]; cannam@95: Tl = FMA(Tf, Tk, Te); cannam@95: Tt = Tr * Ts; cannam@95: Tq = FNMS(Tf, Td, Tp); cannam@95: TD = Tu * Ts; cannam@95: Rm[0] = T7 + Tl; cannam@95: Rp[0] = T7 - Tl; cannam@95: Im[0] = Tq - To; cannam@95: Ip[0] = To + Tq; cannam@95: Tx = W[4]; cannam@95: Tw = FNMS(Tu, Tv, Tt); cannam@95: TE = FMA(Tr, Tv, TD); cannam@95: TA = W[5]; cannam@95: TF = Tx * TB; cannam@95: Tz = Tx * Ty; cannam@95: } cannam@95: } cannam@95: } cannam@95: TG = FNMS(TA, Ty, TF); cannam@95: TC = FMA(TA, TB, Tz); cannam@95: Im[WS(rs, 1)] = TG - TE; cannam@95: Ip[WS(rs, 1)] = TE + TG; cannam@95: Rm[WS(rs, 1)] = Tw + TC; cannam@95: Rp[WS(rs, 1)] = Tw - TC; cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_FULL, 1, 4}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 4, "hc2cbdft_4", twinstr, &GENUS, {24, 6, 6, 0} }; cannam@95: cannam@95: void X(codelet_hc2cbdft_4) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cbdft_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 -sign 1 -n 4 -dif -name hc2cbdft_4 -include hc2cb.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 30 FP additions, 12 FP multiplications, cannam@95: * (or, 24 additions, 6 multiplications, 6 fused multiply/add), cannam@95: * 19 stack variables, 0 constants, and 16 memory accesses cannam@95: */ cannam@95: #include "hc2cb.h" cannam@95: cannam@95: static void hc2cbdft_4(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: { cannam@95: INT m; cannam@95: 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@95: E T3, Tl, T6, Tm, Td, Tj, Tx, Tv, Ts, Tq; cannam@95: { cannam@95: E Tf, Tc, T9, Ti; cannam@95: { cannam@95: E T1, T2, Ta, Tb; cannam@95: T1 = Rp[0]; cannam@95: T2 = Rm[WS(rs, 1)]; cannam@95: T3 = T1 + T2; cannam@95: Tf = T1 - T2; cannam@95: Ta = Ip[0]; cannam@95: Tb = Im[WS(rs, 1)]; cannam@95: Tc = Ta + Tb; cannam@95: Tl = Ta - Tb; cannam@95: } cannam@95: { cannam@95: E T4, T5, Tg, Th; cannam@95: T4 = Rp[WS(rs, 1)]; cannam@95: T5 = Rm[0]; cannam@95: T6 = T4 + T5; cannam@95: T9 = T4 - T5; cannam@95: Tg = Ip[WS(rs, 1)]; cannam@95: Th = Im[0]; cannam@95: Ti = Tg + Th; cannam@95: Tm = Tg - Th; cannam@95: } cannam@95: Td = T9 + Tc; cannam@95: Tj = Tf - Ti; cannam@95: Tx = Tf + Ti; cannam@95: Tv = Tc - T9; cannam@95: Ts = Tl - Tm; cannam@95: Tq = T3 - T6; cannam@95: } cannam@95: { cannam@95: E T7, Tn, Tk, To, T8, Te; cannam@95: T7 = T3 + T6; cannam@95: Tn = Tl + Tm; cannam@95: T8 = W[0]; cannam@95: Te = W[1]; cannam@95: Tk = FMA(T8, Td, Te * Tj); cannam@95: To = FNMS(Te, Td, T8 * Tj); cannam@95: Rp[0] = T7 - Tk; cannam@95: Ip[0] = Tn + To; cannam@95: Rm[0] = T7 + Tk; cannam@95: Im[0] = To - Tn; cannam@95: } cannam@95: { cannam@95: E Tt, Tz, Ty, TA; cannam@95: { cannam@95: E Tp, Tr, Tu, Tw; cannam@95: Tp = W[2]; cannam@95: Tr = W[3]; cannam@95: Tt = FNMS(Tr, Ts, Tp * Tq); cannam@95: Tz = FMA(Tr, Tq, Tp * Ts); cannam@95: Tu = W[4]; cannam@95: Tw = W[5]; cannam@95: Ty = FMA(Tu, Tv, Tw * Tx); cannam@95: TA = FNMS(Tw, Tv, Tu * Tx); cannam@95: } cannam@95: Rp[WS(rs, 1)] = Tt - Ty; cannam@95: Ip[WS(rs, 1)] = Tz + TA; cannam@95: Rm[WS(rs, 1)] = Tt + Ty; cannam@95: Im[WS(rs, 1)] = TA - Tz; cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_FULL, 1, 4}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 4, "hc2cbdft_4", twinstr, &GENUS, {24, 6, 6, 0} }; cannam@95: cannam@95: void X(codelet_hc2cbdft_4) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cbdft_4, &desc, HC2C_VIA_DFT); cannam@95: } cannam@95: #endif /* HAVE_FMA */