src/gromacs/nbnxm/kernels_simd_2xmm/kernel_common.h

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2018,2019, 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.
   8  *
   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.
  13  *
  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.
  18  *
  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.
  23  *
  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.
  31  *
  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.
  34  */
  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 #ifdef CALC_COUL_EWALD
  40 #include "gromacs/math/utilities.h"
  41 #endif
  42
  43 #include "config.h"
  44
  45 #include <cstdint>
  46
  47 #if !GMX_SIMD_HAVE_HSIMD_UTIL_REAL
  48 #error "Half-simd utility operations are required for the 2xNN kernels"
  49 #endif
  50
  51 #ifndef GMX_SIMD_J_UNROLL_SIZE
  52 #error "Need to define GMX_SIMD_J_UNROLL_SIZE before including the 2xnn kernel common header file"
  53 #endif
  54
  55 #define UNROLLI    4
  56 #define UNROLLJ    (GMX_SIMD_REAL_WIDTH/GMX_SIMD_J_UNROLL_SIZE)
  57
  58 static_assert(UNROLLI == c_nbnxnCpuIClusterSize, "UNROLLI should match the i-cluster size");
  59
  60 /* The stride of all the atom data arrays is equal to half the SIMD width */
  61 #define STRIDE     UNROLLJ
  62
  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"
  65 #endif
  66
  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
  69 #define TAB_FDV0
  70 #endif
  71
  72 #if defined UNROLLJ
  73 /* As add_ener_grp, but for two groups of UNROLLJ/2 stored in
  74  * a single SIMD register.
  75  */
  76 static inline void
  77 add_ener_grp_halves(gmx::SimdReal e_S, real *v0, real *v1, const int *offset_jj)
  78 {
  79     for (int jj = 0; jj < (UNROLLJ/2); jj++)
  80     {
  81         incrDualHsimd(v0+offset_jj[jj]+jj*GMX_SIMD_REAL_WIDTH/2,
  82                       v1+offset_jj[jj]+jj*GMX_SIMD_REAL_WIDTH/2, e_S);
  83     }
  84 }
  85 #endif
  86
  87 #if GMX_SIMD_HAVE_INT32_LOGICAL
  88 typedef gmx::SimdInt32    SimdBitMask;
  89 #else
  90 typedef gmx::SimdReal     SimdBitMask;
  91 #endif
  92
  93
  94 static inline void gmx_simdcall
  95 gmx_load_simd_2xnn_interactions(int                  excl,
  96                                 SimdBitMask          filter_S0,
  97                                 SimdBitMask          filter_S2,
  98                                 gmx::SimdBool       *interact_S0,
  99                                 gmx::SimdBool       *interact_S2)
 100 {
 101     using namespace gmx;
 102 #if GMX_SIMD_HAVE_INT32_LOGICAL
 103     SimdInt32 mask_pr_S(excl);
 104     *interact_S0  = cvtIB2B( testBits( mask_pr_S & filter_S0 ) );
 105     *interact_S2  = cvtIB2B( testBits( mask_pr_S & filter_S2 ) );
 106 #elif GMX_SIMD_HAVE_LOGICAL
 107     union
 108     {
 109 #if GMX_DOUBLE
 110         std::int64_t i;
 111 #else
 112         std::int32_t i;
 113 #endif
 114         real         r;
 115     } conv;
 116
 117     conv.i = excl;
 118     SimdReal      mask_pr_S(conv.r);
 119
 120     *interact_S0  = testBits( mask_pr_S & filter_S0 );
 121     *interact_S2  = testBits( mask_pr_S & filter_S2 );
 122 #endif
 123 }
 124
 125 /* All functionality defines are set here, except for:
 126  * CALC_ENERGIES, ENERGY_GROUPS which are defined before.
 127  * CHECK_EXCLS, which is set just before including the inner loop contents.
 128  * The combination rule defines, LJ_COMB_GEOM or LJ_COMB_LB are currently
 129  * set before calling the kernel function. We might want to move that
 130  * to inside the n-loop and have a different combination rule for different
 131  * ci's, as no combination rule gives a 50% performance hit for LJ.
 132  */
 133
 134 /* We always calculate shift forces, because it's cheap anyhow */
 135 #define CALC_SHIFTFORCES
 136
 137 /* Assumes all LJ parameters are identical */
 138 /* #define FIX_LJ_C */