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 07:40:14 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_r2cf.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 8 -name r2cfII_8 -dft-II -include r2cfII.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 22 FP additions, 16 FP multiplications, Chris@10: * (or, 6 additions, 0 multiplications, 16 fused multiply/add), Chris@10: * 22 stack variables, 3 constants, and 16 memory accesses Chris@10: */ Chris@10: #include "r2cfII.h" Chris@10: Chris@10: static void r2cfII_8(R *R0, R *R1, R *Cr, R *Ci, stride rs, stride csr, stride csi, INT v, INT ivs, INT ovs) Chris@10: { Chris@10: DK(KP923879532, +0.923879532511286756128183189396788286822416626); Chris@10: DK(KP707106781, +0.707106781186547524400844362104849039284835938); Chris@10: DK(KP414213562, +0.414213562373095048801688724209698078569671875); Chris@10: { Chris@10: INT i; Chris@10: for (i = v; i > 0; i = i - 1, R0 = R0 + ivs, R1 = R1 + ivs, Cr = Cr + ovs, Ci = Ci + ovs, MAKE_VOLATILE_STRIDE(32, rs), MAKE_VOLATILE_STRIDE(32, csr), MAKE_VOLATILE_STRIDE(32, csi)) { Chris@10: E Te, T8, Td, T5, Tj, Tl, Tf, Tb; Chris@10: { Chris@10: E T1, Th, T9, Ti, T4, Ta; Chris@10: T1 = R0[0]; Chris@10: Th = R0[WS(rs, 2)]; Chris@10: { Chris@10: E T2, T3, T6, T7; Chris@10: T2 = R0[WS(rs, 1)]; Chris@10: T3 = R0[WS(rs, 3)]; Chris@10: T6 = R1[0]; Chris@10: T7 = R1[WS(rs, 2)]; Chris@10: T9 = R1[WS(rs, 3)]; Chris@10: Ti = T2 + T3; Chris@10: T4 = T2 - T3; Chris@10: Te = FMA(KP414213562, T6, T7); Chris@10: T8 = FNMS(KP414213562, T7, T6); Chris@10: Ta = R1[WS(rs, 1)]; Chris@10: } Chris@10: Td = FNMS(KP707106781, T4, T1); Chris@10: T5 = FMA(KP707106781, T4, T1); Chris@10: Tj = FMA(KP707106781, Ti, Th); Chris@10: Tl = FNMS(KP707106781, Ti, Th); Chris@10: Tf = FMA(KP414213562, T9, Ta); Chris@10: Tb = FMS(KP414213562, Ta, T9); Chris@10: } Chris@10: { Chris@10: E Tk, Tg, Tc, Tm; Chris@10: Tk = Te + Tf; Chris@10: Tg = Te - Tf; Chris@10: Tc = T8 + Tb; Chris@10: Tm = Tb - T8; Chris@10: Cr[WS(csr, 1)] = FMA(KP923879532, Tg, Td); Chris@10: Cr[WS(csr, 2)] = FNMS(KP923879532, Tg, Td); Chris@10: Ci[WS(csi, 3)] = FNMS(KP923879532, Tk, Tj); Chris@10: Ci[0] = -(FMA(KP923879532, Tk, Tj)); Chris@10: Ci[WS(csi, 1)] = FMA(KP923879532, Tm, Tl); Chris@10: Ci[WS(csi, 2)] = FMS(KP923879532, Tm, Tl); Chris@10: Cr[0] = FMA(KP923879532, Tc, T5); Chris@10: Cr[WS(csr, 3)] = FNMS(KP923879532, Tc, T5); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const kr2c_desc desc = { 8, "r2cfII_8", {6, 0, 16, 0}, &GENUS }; Chris@10: Chris@10: void X(codelet_r2cfII_8) (planner *p) { Chris@10: X(kr2c_register) (p, r2cfII_8, &desc); Chris@10: } Chris@10: Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_r2cf.native -compact -variables 4 -pipeline-latency 4 -n 8 -name r2cfII_8 -dft-II -include r2cfII.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 22 FP additions, 10 FP multiplications, Chris@10: * (or, 18 additions, 6 multiplications, 4 fused multiply/add), Chris@10: * 18 stack variables, 3 constants, and 16 memory accesses Chris@10: */ Chris@10: #include "r2cfII.h" Chris@10: Chris@10: static void r2cfII_8(R *R0, R *R1, R *Cr, R *Ci, stride rs, stride csr, stride csi, INT v, INT ivs, INT ovs) Chris@10: { Chris@10: DK(KP382683432, +0.382683432365089771728459984030398866761344562); Chris@10: DK(KP923879532, +0.923879532511286756128183189396788286822416626); Chris@10: DK(KP707106781, +0.707106781186547524400844362104849039284835938); Chris@10: { Chris@10: INT i; Chris@10: for (i = v; i > 0; i = i - 1, R0 = R0 + ivs, R1 = R1 + ivs, Cr = Cr + ovs, Ci = Ci + ovs, MAKE_VOLATILE_STRIDE(32, rs), MAKE_VOLATILE_STRIDE(32, csr), MAKE_VOLATILE_STRIDE(32, csi)) { Chris@10: E T1, Tj, T4, Ti, T8, Te, Tb, Tf, T2, T3; Chris@10: T1 = R0[0]; Chris@10: Tj = R0[WS(rs, 2)]; Chris@10: T2 = R0[WS(rs, 1)]; Chris@10: T3 = R0[WS(rs, 3)]; Chris@10: T4 = KP707106781 * (T2 - T3); Chris@10: Ti = KP707106781 * (T2 + T3); Chris@10: { Chris@10: E T6, T7, T9, Ta; Chris@10: T6 = R1[0]; Chris@10: T7 = R1[WS(rs, 2)]; Chris@10: T8 = FNMS(KP382683432, T7, KP923879532 * T6); Chris@10: Te = FMA(KP382683432, T6, KP923879532 * T7); Chris@10: T9 = R1[WS(rs, 1)]; Chris@10: Ta = R1[WS(rs, 3)]; Chris@10: Tb = FNMS(KP923879532, Ta, KP382683432 * T9); Chris@10: Tf = FMA(KP923879532, T9, KP382683432 * Ta); Chris@10: } Chris@10: { Chris@10: E T5, Tc, Th, Tk; Chris@10: T5 = T1 + T4; Chris@10: Tc = T8 + Tb; Chris@10: Cr[WS(csr, 3)] = T5 - Tc; Chris@10: Cr[0] = T5 + Tc; Chris@10: Th = Te + Tf; Chris@10: Tk = Ti + Tj; Chris@10: Ci[0] = -(Th + Tk); Chris@10: Ci[WS(csi, 3)] = Tk - Th; Chris@10: } Chris@10: { Chris@10: E Td, Tg, Tl, Tm; Chris@10: Td = T1 - T4; Chris@10: Tg = Te - Tf; Chris@10: Cr[WS(csr, 2)] = Td - Tg; Chris@10: Cr[WS(csr, 1)] = Td + Tg; Chris@10: Tl = Tb - T8; Chris@10: Tm = Tj - Ti; Chris@10: Ci[WS(csi, 2)] = Tl - Tm; Chris@10: Ci[WS(csi, 1)] = Tl + Tm; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const kr2c_desc desc = { 8, "r2cfII_8", {18, 6, 4, 0}, &GENUS }; Chris@10: Chris@10: void X(codelet_r2cfII_8) (planner *p) { Chris@10: X(kr2c_register) (p, r2cfII_8, &desc); Chris@10: } Chris@10: Chris@10: #endif /* HAVE_FMA */