src/gromacs/nbnxm/kernels_simd_4xm/kernel_common.h

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017 by the GROMACS development team.
   5  * Copyright (c) 2018,2019,2020, by the GROMACS development team, led by
   6  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   7  * and including many others, as listed in the AUTHORS file in the
   8  * top-level source directory and at http://www.gromacs.org.
   9  *
  10  * GROMACS is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU Lesser General Public License
  12  * as published by the Free Software Foundation; either version 2.1
  13  * of the License, or (at your option) any later version.
  14  *
  15  * GROMACS is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * Lesser General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU Lesser General Public
  21  * License along with GROMACS; if not, see
  22  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  23  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  24  *
  25  * If you want to redistribute modifications to GROMACS, please
  26  * consider that scientific software is very special. Version
  27  * control is crucial - bugs must be traceable. We will be happy to
  28  * consider code for inclusion in the official distribution, but
  29  * derived work must not be called official GROMACS. Details are found
  30  * in the README & COPYING files - if they are missing, get the
  31  * official version at http://www.gromacs.org.
  32  *
  33  * To help us fund GROMACS development, we humbly ask that you cite
  34  * the research papers on the package. Check out http://www.gromacs.org.
  35  */
  36 #include "gromacs/pbcutil/ishift.h"
  37 #include "gromacs/simd/simd.h"
  38 #include "gromacs/simd/simd_math.h"
  39 #include "gromacs/simd/vector_operations.h"
  40 #include "gromacs/utility/basedefinitions.h"
  41 #ifdef CALC_COUL_EWALD
  42 #    include "gromacs/math/utilities.h"
  43 #endif
  44
  45 #include "config.h"
  46
  47 #ifndef GMX_SIMD_J_UNROLL_SIZE
  48 #    error "Need to define GMX_SIMD_J_UNROLL_SIZE before including the 4xn kernel common header file"
  49 #endif
  50
  51 #define UNROLLI 4
  52 #define UNROLLJ (GMX_SIMD_REAL_WIDTH / GMX_SIMD_J_UNROLL_SIZE)
  53
  54 static_assert(UNROLLI == c_nbnxnCpuIClusterSize, "UNROLLI should match the i-cluster size");
  55
  56 /* The stride of all the atom data arrays is max(UNROLLI,unrollj) */
  57 #if GMX_SIMD_REAL_WIDTH >= UNROLLI
  58 #    define STRIDE (GMX_SIMD_REAL_WIDTH / GMX_SIMD_J_UNROLL_SIZE)
  59 #else
  60 #    define STRIDE (UNROLLI)
  61 #endif
  62
  63
  64 #if !defined GMX_NBNXN_SIMD_2XNN && !defined GMX_NBNXN_SIMD_4XN
  65 #    error "Must define an NBNxN kernel flavour before including NBNxN kernel utility functions"
  66 #endif
  67
  68 // 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
  69 #if GMX_SIMD_REAL_WIDTH == 4 || !GMX_SIMD_HAVE_GATHER_LOADU_BYSIMDINT_TRANSPOSE_REAL
  70 #    define TAB_FDV0
  71 #endif
  72
  73
  74 #ifdef UNROLLJ
  75 /* Add energy register to possibly multiple terms in the energy array */
  76 static inline void add_ener_grp(gmx::SimdReal e_S, real* v, const int* offset_jj)
  77 {
  78     using namespace gmx;
  79     int jj;
  80
  81     /* We need to balance the number of store operations with
  82      * the rapidly increases number of combinations of energy groups.
  83      * We add to a temporary buffer for 1 i-group vs 2 j-groups.
  84      */
  85     for (jj = 0; jj < (UNROLLJ / 2); jj++)
  86     {
  87         SimdReal v_S;
  88
  89         v_S = load<SimdReal>(v + offset_jj[jj] + jj * GMX_SIMD_REAL_WIDTH);
  90         store(v + offset_jj[jj] + jj * GMX_SIMD_REAL_WIDTH, v_S + e_S);
  91     }
  92 }
  93 #endif
  94
  95 #if GMX_SIMD_HAVE_INT32_LOGICAL
  96 typedef gmx::SimdInt32 SimdBitMask;
  97 #else
  98 typedef gmx::SimdReal SimdBitMask;
  99 #endif
 100
 101 static inline void gmx_simdcall 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)
 111 {
 112     using namespace gmx;
 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
 121     union {
 122 #    if GMX_DOUBLE
 123         std::int64_t i;
 124 #    else
 125         std::int32_t i;
 126 #    endif
 127         real         r;
 128     } conv;
 129
 130     conv.i = excl;
 131     SimdReal mask_pr_S(conv.r);
 132
 133     *interact_S0 = testBits(mask_pr_S & filter_S0);
 134     *interact_S1 = testBits(mask_pr_S & filter_S1);
 135     *interact_S2 = testBits(mask_pr_S & filter_S2);
 136     *interact_S3 = testBits(mask_pr_S & filter_S3);
 137 #else
 138     // Neither real or integer bitwise logical operations supported.
 139     // Load masks from memory instead.
 140     SimdReal zero = setZero();
 141     *interact_S0  = (zero < load<SimdReal>(simd_interaction_array
 142                                           + GMX_SIMD_REAL_WIDTH * ((excl >> (0 * UNROLLJ)) & 0xF)));
 143     *interact_S1  = (zero < load<SimdReal>(simd_interaction_array
 144                                           + GMX_SIMD_REAL_WIDTH * ((excl >> (1 * UNROLLJ)) & 0xF)));
 145     *interact_S2  = (zero < load<SimdReal>(simd_interaction_array
 146                                           + GMX_SIMD_REAL_WIDTH * ((excl >> (2 * UNROLLJ)) & 0xF)));
 147     *interact_S3  = (zero < load<SimdReal>(simd_interaction_array
 148                                           + GMX_SIMD_REAL_WIDTH * ((excl >> (3 * UNROLLJ)) & 0xF)));
 149 #endif
 150 }
 151
 152 /* All functionality defines are set here, except for:
 153  * CALC_ENERGIES, ENERGY_GROUPS which are defined before.
 154  * CHECK_EXCLS, which is set just before including the inner loop contents.
 155  * The combination rule defines, LJ_COMB_GEOM or LJ_COMB_LB are currently
 156  * set before calling the kernel function. We might want to move that
 157  * to inside the n-loop and have a different combination rule for different
 158  * ci's, as no combination rule gives a 50% performance hit for LJ.
 159  */
 160
 161 /* We always calculate shift forces, because it's cheap anyhow */
 162 #define CALC_SHIFTFORCES
 163
 164 /* Assumes all LJ parameters are identical */
 165 /* #define FIX_LJ_C */