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 07:41:57 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_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 */ cannam@95: cannam@95: /* cannam@95: * This function contains 24 FP additions, 16 FP multiplications, cannam@95: * (or, 16 additions, 8 multiplications, 8 fused multiply/add), cannam@95: * 30 stack variables, 0 constants, and 16 memory accesses cannam@95: */ cannam@95: #include "hc2cb.h" cannam@95: cannam@95: static void hc2cb2_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) * 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 Tg, Tc, Te, To, Tn; cannam@95: { cannam@95: E T7, Tb, T8, Ta; cannam@95: T7 = W[0]; cannam@95: Tb = W[3]; cannam@95: T8 = W[2]; cannam@95: Ta = W[1]; cannam@95: { cannam@95: E Tu, Tj, T3, Tm, Tx, Tr, T6, Tt; cannam@95: { cannam@95: E T4, Tp, Tq, T5; cannam@95: { cannam@95: E T1, T2, Tk, Tl; cannam@95: { cannam@95: E Th, Tf, T9, Ti; cannam@95: Th = Ip[0]; cannam@95: Tf = T7 * Tb; cannam@95: T9 = T7 * T8; cannam@95: Ti = Im[WS(rs, 1)]; cannam@95: T1 = Rp[0]; cannam@95: Tg = FNMS(Ta, T8, Tf); cannam@95: Tc = FMA(Ta, Tb, T9); cannam@95: Tu = Th + Ti; cannam@95: Tj = Th - Ti; cannam@95: T2 = Rm[WS(rs, 1)]; cannam@95: } cannam@95: Tk = Ip[WS(rs, 1)]; cannam@95: Tl = Im[0]; cannam@95: T4 = Rp[WS(rs, 1)]; cannam@95: T3 = T1 + T2; cannam@95: Tp = T1 - T2; cannam@95: Tm = Tk - Tl; cannam@95: Tq = Tk + Tl; cannam@95: T5 = Rm[0]; cannam@95: } cannam@95: Tx = Tp + Tq; cannam@95: Tr = Tp - Tq; cannam@95: T6 = T4 + T5; cannam@95: Tt = T4 - T5; cannam@95: } cannam@95: { cannam@95: E Tz, Tv, Td, Ts, Tw, TA, Ty; cannam@95: Rm[0] = Tj + Tm; cannam@95: Ts = T7 * Tr; cannam@95: Tz = Tu - Tt; cannam@95: Tv = Tt + Tu; cannam@95: Rp[0] = T3 + T6; cannam@95: Td = T3 - T6; cannam@95: Ip[0] = FNMS(Ta, Tv, Ts); cannam@95: Tw = T7 * Tv; cannam@95: TA = T8 * Tz; cannam@95: Ty = T8 * Tx; cannam@95: Te = Tc * Td; cannam@95: Im[0] = FMA(Ta, Tr, Tw); cannam@95: Im[WS(rs, 1)] = FMA(Tb, Tx, TA); cannam@95: Ip[WS(rs, 1)] = FNMS(Tb, Tz, Ty); cannam@95: To = Tg * Td; cannam@95: Tn = Tj - Tm; cannam@95: } cannam@95: } cannam@95: } cannam@95: Rm[WS(rs, 1)] = FMA(Tc, Tn, To); cannam@95: Rp[WS(rs, 1)] = FNMS(Tg, Tn, Te); 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, "hc2cb2_4", twinstr, &GENUS, {16, 8, 8, 0} }; cannam@95: cannam@95: void X(codelet_hc2cb2_4) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cb2_4, &desc, HC2C_VIA_RDFT); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* 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 */ cannam@95: cannam@95: /* cannam@95: * This function contains 24 FP additions, 16 FP multiplications, cannam@95: * (or, 16 additions, 8 multiplications, 8 fused multiply/add), cannam@95: * 21 stack variables, 0 constants, and 16 memory accesses cannam@95: */ cannam@95: #include "hc2cb.h" cannam@95: cannam@95: static void hc2cb2_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) * 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 T7, T9, T8, Ta, Tb, Td; cannam@95: T7 = W[0]; cannam@95: T9 = W[1]; cannam@95: T8 = W[2]; cannam@95: Ta = W[3]; cannam@95: Tb = FMA(T7, T8, T9 * Ta); cannam@95: Td = FNMS(T9, T8, T7 * Ta); cannam@95: { cannam@95: E T3, Tl, Tg, Tp, T6, To, Tj, Tm, Tc, Tk; cannam@95: { cannam@95: E T1, T2, Te, Tf; cannam@95: T1 = Rp[0]; cannam@95: T2 = Rm[WS(rs, 1)]; cannam@95: T3 = T1 + T2; cannam@95: Tl = T1 - T2; cannam@95: Te = Ip[0]; cannam@95: Tf = Im[WS(rs, 1)]; cannam@95: Tg = Te - Tf; cannam@95: Tp = Te + Tf; cannam@95: } cannam@95: { cannam@95: E T4, T5, Th, Ti; cannam@95: T4 = Rp[WS(rs, 1)]; cannam@95: T5 = Rm[0]; cannam@95: T6 = T4 + T5; cannam@95: To = T4 - T5; cannam@95: Th = Ip[WS(rs, 1)]; cannam@95: Ti = Im[0]; cannam@95: Tj = Th - Ti; cannam@95: Tm = Th + Ti; cannam@95: } cannam@95: Rp[0] = T3 + T6; cannam@95: Rm[0] = Tg + Tj; cannam@95: Tc = T3 - T6; cannam@95: Tk = Tg - Tj; cannam@95: Rp[WS(rs, 1)] = FNMS(Td, Tk, Tb * Tc); cannam@95: Rm[WS(rs, 1)] = FMA(Td, Tc, Tb * Tk); cannam@95: { cannam@95: E Tn, Tq, Tr, Ts; cannam@95: Tn = Tl - Tm; cannam@95: Tq = To + Tp; cannam@95: Ip[0] = FNMS(T9, Tq, T7 * Tn); cannam@95: Im[0] = FMA(T7, Tq, T9 * Tn); cannam@95: Tr = Tl + Tm; cannam@95: Ts = Tp - To; cannam@95: Ip[WS(rs, 1)] = FNMS(Ta, Ts, T8 * Tr); cannam@95: Im[WS(rs, 1)] = FMA(T8, Ts, Ta * Tr); 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, "hc2cb2_4", twinstr, &GENUS, {16, 8, 8, 0} }; cannam@95: cannam@95: void X(codelet_hc2cb2_4) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cb2_4, &desc, HC2C_VIA_RDFT); cannam@95: } cannam@95: #endif /* HAVE_FMA */