2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014,2015,2017,2018,2019, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
35 #include "gromacs/pbcutil/ishift.h"
36 #include "gromacs/simd/simd.h"
37 #include "gromacs/simd/simd_math.h"
38 #include "gromacs/simd/vector_operations.h"
39 #include "gromacs/utility/basedefinitions.h"
40 #ifdef CALC_COUL_EWALD
41 #include "gromacs/math/utilities.h"
46 #ifndef GMX_SIMD_J_UNROLL_SIZE
47 #error "Need to define GMX_SIMD_J_UNROLL_SIZE before including the 4xn kernel common header file"
51 #define UNROLLJ (GMX_SIMD_REAL_WIDTH/GMX_SIMD_J_UNROLL_SIZE)
53 static_assert(UNROLLI == c_nbnxnCpuIClusterSize, "UNROLLI should match the i-cluster size");
55 /* The stride of all the atom data arrays is max(UNROLLI,unrollj) */
56 #if GMX_SIMD_REAL_WIDTH >= UNROLLI
57 #define STRIDE (GMX_SIMD_REAL_WIDTH/GMX_SIMD_J_UNROLL_SIZE)
59 #define STRIDE (UNROLLI)
63 #if !defined GMX_NBNXN_SIMD_2XNN && !defined GMX_NBNXN_SIMD_4XN
64 #error "Must define an NBNxN kernel flavour before including NBNxN kernel utility functions"
67 // We use the FDV0 tables for width==4 (when we can load it in one go), or if we don't have any unaligned loads
68 #if GMX_SIMD_REAL_WIDTH == 4 || !GMX_SIMD_HAVE_GATHER_LOADU_BYSIMDINT_TRANSPOSE_REAL
74 /* Add energy register to possibly multiple terms in the energy array */
75 static inline void add_ener_grp(gmx::SimdReal e_S, real *v, const int *offset_jj)
80 /* We need to balance the number of store operations with
81 * the rapidly increases number of combinations of energy groups.
82 * We add to a temporary buffer for 1 i-group vs 2 j-groups.
84 for (jj = 0; jj < (UNROLLJ/2); jj++)
88 v_S = load<SimdReal>(v+offset_jj[jj]+jj*GMX_SIMD_REAL_WIDTH);
89 store(v+offset_jj[jj]+jj*GMX_SIMD_REAL_WIDTH, v_S + e_S);
94 #if GMX_SIMD_HAVE_INT32_LOGICAL
95 typedef gmx::SimdInt32 SimdBitMask;
97 typedef gmx::SimdReal SimdBitMask;
100 static inline void gmx_simdcall
101 gmx_load_simd_4xn_interactions(int excl,
102 SimdBitMask gmx_unused filter_S0,
103 SimdBitMask gmx_unused filter_S1,
104 SimdBitMask gmx_unused filter_S2,
105 SimdBitMask gmx_unused filter_S3,
106 const real gmx_unused *simd_interaction_array,
107 gmx::SimdBool *interact_S0,
108 gmx::SimdBool *interact_S1,
109 gmx::SimdBool *interact_S2,
110 gmx::SimdBool *interact_S3)
113 #if GMX_SIMD_HAVE_INT32_LOGICAL
114 /* Load integer interaction mask */
115 SimdInt32 mask_pr_S(excl);
116 *interact_S0 = cvtIB2B(testBits( mask_pr_S & filter_S0 ));
117 *interact_S1 = cvtIB2B(testBits( mask_pr_S & filter_S1 ));
118 *interact_S2 = cvtIB2B(testBits( mask_pr_S & filter_S2 ));
119 *interact_S3 = cvtIB2B(testBits( mask_pr_S & filter_S3 ));
120 #elif GMX_SIMD_HAVE_LOGICAL
132 SimdReal mask_pr_S(conv.r);
134 *interact_S0 = testBits( mask_pr_S & filter_S0 );
135 *interact_S1 = testBits( mask_pr_S & filter_S1 );
136 *interact_S2 = testBits( mask_pr_S & filter_S2 );
137 *interact_S3 = testBits( mask_pr_S & filter_S3 );
139 // Neither real or integer bitwise logical operations supported.
140 // Load masks from memory instead.
141 SimdReal zero = setZero();
142 *interact_S0 = ( zero < load<SimdReal>( simd_interaction_array + GMX_SIMD_REAL_WIDTH*((excl >> (0 * UNROLLJ)) & 0xF) ) );
143 *interact_S1 = ( zero < load<SimdReal>( simd_interaction_array + GMX_SIMD_REAL_WIDTH*((excl >> (1 * UNROLLJ)) & 0xF) ) );
144 *interact_S2 = ( zero < load<SimdReal>( simd_interaction_array + GMX_SIMD_REAL_WIDTH*((excl >> (2 * UNROLLJ)) & 0xF) ) );
145 *interact_S3 = ( zero < load<SimdReal>( simd_interaction_array + GMX_SIMD_REAL_WIDTH*((excl >> (3 * UNROLLJ)) & 0xF) ) );
149 /* All functionality defines are set here, except for:
150 * CALC_ENERGIES, ENERGY_GROUPS which are defined before.
151 * CHECK_EXCLS, which is set just before including the inner loop contents.
152 * The combination rule defines, LJ_COMB_GEOM or LJ_COMB_LB are currently
153 * set before calling the kernel function. We might want to move that
154 * to inside the n-loop and have a different combination rule for different
155 * ci's, as no combination rule gives a 50% performance hit for LJ.
158 /* We always calculate shift forces, because it's cheap anyhow */
159 #define CALC_SHIFTFORCES
161 /* Assumes all LJ parameters are identical */
162 /* #define FIX_LJ_C */