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 07:41:57 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_hc2c.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -sign 1 -twiddle-log3 -precompute-twiddles -n 4 -dif -name hc2cb2_4 -include hc2cb.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 24 FP additions, 16 FP multiplications, Chris@10: * (or, 16 additions, 8 multiplications, 8 fused multiply/add), Chris@10: * 30 stack variables, 0 constants, and 16 memory accesses Chris@10: */ Chris@10: #include "hc2cb.h" Chris@10: Chris@10: static void hc2cb2_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) * 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)) { Chris@10: E Tg, Tc, Te, To, Tn; Chris@10: { Chris@10: E T7, Tb, T8, Ta; Chris@10: T7 = W[0]; Chris@10: Tb = W[3]; Chris@10: T8 = W[2]; Chris@10: Ta = W[1]; Chris@10: { Chris@10: E Tu, Tj, T3, Tm, Tx, Tr, T6, Tt; Chris@10: { Chris@10: E T4, Tp, Tq, T5; Chris@10: { Chris@10: E T1, T2, Tk, Tl; Chris@10: { Chris@10: E Th, Tf, T9, Ti; Chris@10: Th = Ip[0]; Chris@10: Tf = T7 * Tb; Chris@10: T9 = T7 * T8; Chris@10: Ti = Im[WS(rs, 1)]; Chris@10: T1 = Rp[0]; Chris@10: Tg = FNMS(Ta, T8, Tf); Chris@10: Tc = FMA(Ta, Tb, T9); Chris@10: Tu = Th + Ti; Chris@10: Tj = Th - Ti; Chris@10: T2 = Rm[WS(rs, 1)]; Chris@10: } Chris@10: Tk = Ip[WS(rs, 1)]; Chris@10: Tl = Im[0]; Chris@10: T4 = Rp[WS(rs, 1)]; Chris@10: T3 = T1 + T2; Chris@10: Tp = T1 - T2; Chris@10: Tm = Tk - Tl; Chris@10: Tq = Tk + Tl; Chris@10: T5 = Rm[0]; Chris@10: } Chris@10: Tx = Tp + Tq; Chris@10: Tr = Tp - Tq; Chris@10: T6 = T4 + T5; Chris@10: Tt = T4 - T5; Chris@10: } Chris@10: { Chris@10: E Tz, Tv, Td, Ts, Tw, TA, Ty; Chris@10: Rm[0] = Tj + Tm; Chris@10: Ts = T7 * Tr; Chris@10: Tz = Tu - Tt; Chris@10: Tv = Tt + Tu; Chris@10: Rp[0] = T3 + T6; Chris@10: Td = T3 - T6; Chris@10: Ip[0] = FNMS(Ta, Tv, Ts); Chris@10: Tw = T7 * Tv; Chris@10: TA = T8 * Tz; Chris@10: Ty = T8 * Tx; Chris@10: Te = Tc * Td; Chris@10: Im[0] = FMA(Ta, Tr, Tw); Chris@10: Im[WS(rs, 1)] = FMA(Tb, Tx, TA); Chris@10: Ip[WS(rs, 1)] = FNMS(Tb, Tz, Ty); Chris@10: To = Tg * Td; Chris@10: Tn = Tj - Tm; Chris@10: } Chris@10: } Chris@10: } Chris@10: Rm[WS(rs, 1)] = FMA(Tc, Tn, To); Chris@10: Rp[WS(rs, 1)] = FNMS(Tg, Tn, Te); Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_CEXP, 1, 1}, Chris@10: {TW_CEXP, 1, 3}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 4, "hc2cb2_4", twinstr, &GENUS, {16, 8, 8, 0} }; Chris@10: Chris@10: void X(codelet_hc2cb2_4) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cb2_4, &desc, HC2C_VIA_RDFT); Chris@10: } Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_hc2c.native -compact -variables 4 -pipeline-latency 4 -sign 1 -twiddle-log3 -precompute-twiddles -n 4 -dif -name hc2cb2_4 -include hc2cb.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 24 FP additions, 16 FP multiplications, Chris@10: * (or, 16 additions, 8 multiplications, 8 fused multiply/add), Chris@10: * 21 stack variables, 0 constants, and 16 memory accesses Chris@10: */ Chris@10: #include "hc2cb.h" Chris@10: Chris@10: static void hc2cb2_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) * 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)) { Chris@10: E T7, T9, T8, Ta, Tb, Td; Chris@10: T7 = W[0]; Chris@10: T9 = W[1]; Chris@10: T8 = W[2]; Chris@10: Ta = W[3]; Chris@10: Tb = FMA(T7, T8, T9 * Ta); Chris@10: Td = FNMS(T9, T8, T7 * Ta); Chris@10: { Chris@10: E T3, Tl, Tg, Tp, T6, To, Tj, Tm, Tc, Tk; Chris@10: { Chris@10: E T1, T2, Te, Tf; Chris@10: T1 = Rp[0]; Chris@10: T2 = Rm[WS(rs, 1)]; Chris@10: T3 = T1 + T2; Chris@10: Tl = T1 - T2; Chris@10: Te = Ip[0]; Chris@10: Tf = Im[WS(rs, 1)]; Chris@10: Tg = Te - Tf; Chris@10: Tp = Te + Tf; Chris@10: } Chris@10: { Chris@10: E T4, T5, Th, Ti; Chris@10: T4 = Rp[WS(rs, 1)]; Chris@10: T5 = Rm[0]; Chris@10: T6 = T4 + T5; Chris@10: To = T4 - T5; Chris@10: Th = Ip[WS(rs, 1)]; Chris@10: Ti = Im[0]; Chris@10: Tj = Th - Ti; Chris@10: Tm = Th + Ti; Chris@10: } Chris@10: Rp[0] = T3 + T6; Chris@10: Rm[0] = Tg + Tj; Chris@10: Tc = T3 - T6; Chris@10: Tk = Tg - Tj; Chris@10: Rp[WS(rs, 1)] = FNMS(Td, Tk, Tb * Tc); Chris@10: Rm[WS(rs, 1)] = FMA(Td, Tc, Tb * Tk); Chris@10: { Chris@10: E Tn, Tq, Tr, Ts; Chris@10: Tn = Tl - Tm; Chris@10: Tq = To + Tp; Chris@10: Ip[0] = FNMS(T9, Tq, T7 * Tn); Chris@10: Im[0] = FMA(T7, Tq, T9 * Tn); Chris@10: Tr = Tl + Tm; Chris@10: Ts = Tp - To; Chris@10: Ip[WS(rs, 1)] = FNMS(Ta, Ts, T8 * Tr); Chris@10: Im[WS(rs, 1)] = FMA(T8, Ts, Ta * Tr); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_CEXP, 1, 1}, Chris@10: {TW_CEXP, 1, 3}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 4, "hc2cb2_4", twinstr, &GENUS, {16, 8, 8, 0} }; Chris@10: Chris@10: void X(codelet_hc2cb2_4) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cb2_4, &desc, HC2C_VIA_RDFT); Chris@10: } Chris@10: #endif /* HAVE_FMA */