Commit d3d4b420 authored by Michael Brown's avatar Michael Brown
Browse files

USER-INTEL: Changes to support LAMMPS_BIGBIG in USER-INTEL + fixed check for... 

USER-INTEL: Changes to support LAMMPS_BIGBIG in USER-INTEL + fixed check for per-atom virials + fixed check for exclusion with offload.
parent a7c9560d

src/USER-INTEL/fix_intel.cpp

+4 −8
Original line number Diff line number Diff line
@@ -34,10 +34,6 @@ 
#include <cstdio>

#include <cmath>



#ifdef LAMMPS_BIGBIG

#error "The USER-INTEL package is not compatible with -DLAMMPS_BIGBIG"

#endif



#ifdef _LMP_INTEL_OFFLOAD

#ifndef INTEL_OFFLOAD_NOAFFINITY

#include <unistd.h>
@@ -378,13 +374,13 @@ void FixIntel::setup(int vflag) 
  if (neighbor->style != Neighbor::BIN)

    error->all(FLERR,

            "Currently, neighbor style BIN must be used with Intel package.");

  if (neighbor->exclude_setting() != 0)

    error->all(FLERR,

            "Currently, cannot use neigh_modify exclude with Intel package.");

  if (vflag_atom)

  if (vflag > 3)

   error->all(FLERR,

               "Cannot currently get per-atom virials with Intel package.");

  #ifdef _LMP_INTEL_OFFLOAD

  if (neighbor->exclude_setting() != 0)

    error->all(FLERR,

     "Currently, cannot use neigh_modify exclude with Intel package offload.");

  post_force(vflag);

  #endif

}

src/USER-INTEL/fix_intel.h

+2 −2
Original line number Diff line number Diff line
@@ -495,9 +495,9 @@ E: Currently, neighbor style BIN must be used with Intel package. 
This is the only neighbor style that has been implemented for the Intel

package.



E: Currently, cannot use neigh_modify exclude with Intel package.

E: Currently, cannot use neigh_modify exclude with Intel package offload.



This is a current restriction of the Intel package.

This is a current restriction of the Intel package when built for offload.



W: Unknown Intel Compiler Version

src/USER-INTEL/intel_buffers.cpp

+7 −6
Original line number Diff line number Diff line
@@ -186,8 +186,8 @@ void IntelBuffers<flt_t, acc_t>::free_nmax() 
{

  #ifdef _LMP_INTEL_OFFLOAD

  if (_off_map_nmax > 0) {

    const int * tag = _off_map_tag;

    const int * special = _off_map_special;

    const tagint * tag = _off_map_tag;

    const tagint * special = _off_map_special;

    const int * nspecial = _off_map_nspecial;

    #pragma offload_transfer target(mic:_cop) \

      nocopy(tag:alloc_if(0) free_if(1)) \
@@ -209,7 +209,8 @@ void IntelBuffers<flt_t, acc_t>::_grow_nmax(const int offload_end) 
  _host_nmax = size;



  if (!offload_end) return;

  int *special, *nspecial;

  tagint *special;

  int *nspecial;

  int tag_length, special_length, nspecial_length;

  if (lmp->atom->molecular) {

    special = lmp->atom->special[0];
@@ -226,7 +227,7 @@ void IntelBuffers<flt_t, acc_t>::_grow_nmax(const int offload_end) 
    tag_length = size;

  else

    tag_length = 1;

  int *tag = lmp->atom->tag;

  tagint *tag = lmp->atom->tag;

  #pragma offload_transfer target(mic:_cop) \

    nocopy(tag:length(tag_length) alloc_if(1) free_if(0)) \

    nocopy(special:length(special_length) alloc_if(1) free_if(0)) \
@@ -523,7 +524,7 @@ void IntelBuffers<flt_t, acc_t>::free_ncache() 
    flt_t *ncachez = _ncachez;

    int *ncachej = _ncachej;

    int *ncachejtype = _ncachejtype;

    int *ncachetag = _ncachetag;

    tagint *ncachetag = _ncachetag;



    #ifdef _LMP_INTEL_OFFLOAD

    if (_off_ncache) {
@@ -603,7 +604,7 @@ void IntelBuffers<flt_t, acc_t>::grow_ncache(const int off_flag, 
      tsize = 16;

      lmp->memory->create(_ncachetag, tsize, "_ncachetag");

    }

    int *ncachetag = _ncachetag;

    tagint *ncachetag = _ncachetag;

    #pragma offload_transfer target(mic:_cop)                   \

      nocopy(ncachetag:length(tsize) alloc_if(1) free_if(0))

    _off_ncache = 1;

src/USER-INTEL/intel_buffers.h

+5 −4
Original line number Diff line number Diff line
@@ -132,7 +132,7 @@ class IntelBuffers { 
  inline flt_t * get_ncachez() { return _ncachez; }

  inline int * get_ncachej() { return _ncachej; }

  inline int * get_ncachejtype() { return _ncachejtype; }

  inline int * get_ncachetag() { return _ncachetag; }

  inline tagint * get_ncachetag() { return _ncachetag; }



  inline int get_max_nbors() {

    int mn = lmp->neighbor->oneatom * sizeof(int) /
@@ -336,7 +336,8 @@ class IntelBuffers { 


  int _ncache_stride, _ncache_alloc;

  flt_t *_ncachex, *_ncachey, *_ncachez;

  int *_ncachej, *_ncachejtype, *_ncachetag;

  int *_ncachej, *_ncachejtype;

  tagint *_ncachetag;



  int _need_tag, _host_nmax;
@@ -352,8 +353,8 @@ class IntelBuffers { 
  quat_t *_host_quat;

  vec3_acc_t *_off_f;

  int _off_map_nmax, _cop, _off_ccache, _off_ncache;

  int *_off_map_ilist;

  int *_off_map_special, *_off_map_nspecial, *_off_map_tag;

  int *_off_map_ilist, *_off_map_nspecial;

  tagint *_off_map_tag, *_off_map_special;

  int **_off_map_firstneigh, *_off_map_numneigh;

  bool _off_list_alloc;

  #endif

src/USER-INTEL/npair_full_bin_ghost_intel.cpp

+3 −3
Original line number Diff line number Diff line
@@ -194,7 +194,7 @@ void NPairFullBinGhostIntel::fbi(const int offload, NeighList * list, 
  flt_t * _noalias const ncachez = buffers->get_ncachez();

  int * _noalias const ncachej = buffers->get_ncachej();

  int * _noalias const ncachejtype = buffers->get_ncachejtype();

  int * _noalias const ncachetag = buffers->get_ncachetag();

  tagint * _noalias const ncachetag = buffers->get_ncachetag();

  const int ncache_stride = buffers->ncache_stride();



  const int mbinx = this->mbinx;
@@ -304,7 +304,7 @@ void NPairFullBinGhostIntel::fbi(const int offload, NeighList * list, 
      flt_t * _noalias const tz = ncachez + toffs;

      int * _noalias const tj = ncachej + toffs;

      int * _noalias const tjtype = ncachejtype + toffs;

      int * _noalias const ttag = ncachetag + toffs;

      tagint * _noalias const ttag = ncachetag + toffs;



      // loop over all atoms in other bins in stencil, store every pair

      int ncount, oldbin = -9999999;
@@ -392,7 +392,7 @@ void NPairFullBinGhostIntel::fbi(const int offload, NeighList * list, 
          const flt_t dely = ytmp - ty[u];

          const flt_t delz = ztmp - tz[u];

          const int jtype = tjtype[u];

          const int jtag = ttag[u];

          const tagint jtag = ttag[u];

          const flt_t rsq = delx * delx + dely * dely + delz * delz;

          if (rsq > cutsq[ioffset + jtype]) addme = 0;

CRA Git | Maintained and supported by SUSTech CRA and CCSE