Add trim methods for INTEL

src/INTEL/npair_halffull_newtoff_trim_intel.h

@@ -0,0 +1,44 @@
+// clang-format off
+/* -*- c++ -*- ----------------------------------------------------------
+   LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
+   https://www.lammps.org/, Sandia National Laboratories
+   Steve Plimpton, sjplimp@sandia.gov
+
+   Copyright (2003) Sandia Corporation.  Under the terms of Contract
+   DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
+   certain rights in this software.  This software is distributed under
+   the GNU General Public License.
+
+   See the README file in the top-level LAMMPS directory.
+------------------------------------------------------------------------- */
+
+/* ----------------------------------------------------------------------
+   Contributing author: Stan Moore (SNL)
+------------------------------------------------------------------------- */
+
+// Only used for hybrid to generate list for non-intel style. Use
+// standard routines.
+
+#ifdef NPAIR_CLASS
+// clang-format off
+NPairStyle(halffull/newtoff/trim/intel,
+           NPairHalffullNewtoffTrim,
+           NP_HALF_FULL | NP_NEWTOFF | NP_NSQ | NP_BIN | NP_MULTI | NP_HALF |
+           NP_ORTHO | NP_TRI | NP_TRIM | NP_INTEL);
+
+NPairStyle(halffull/newtoff/skip/trim/intel,
+           NPairHalffullNewtoffTrim,
+           NP_HALF_FULL | NP_NEWTOFF | NP_NSQ | NP_BIN | NP_MULTI | NP_HALF |
+           NP_ORTHO | NP_TRI | NP_SKIP | NP_TRIM | NP_INTEL);
+
+NPairStyle(halffull/newtoff/ghost/trim/intel,
+           NPairHalffullNewtoffTrim,
+           NP_HALF_FULL | NP_NEWTOFF | NP_NSQ | NP_BIN | NP_MULTI | NP_HALF |
+           NP_ORTHO | NP_TRI | NP_GHOST | NP_TRIM | NP_INTEL);
+
+NPairStyle(halffull/newtoff/skip/ghost/trim/intel,
+           NPairHalffullNewtoffTrim,
+           NP_HALF_FULL | NP_NEWTOFF | NP_NSQ | NP_BIN | NP_MULTI | NP_HALF |
+           NP_ORTHO | NP_TRI | NP_SKIP | NP_GHOST | NP_TRIM | NP_INTEL);
+// clang-format on
+#endif

src/INTEL/npair_halffull_newton_trim_intel.cpp

@@ -0,0 +1,258 @@
+// clang-format off
+/* ----------------------------------------------------------------------
+   LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
+   https://www.lammps.org/, Sandia National Laboratories
+   Steve Plimpton, sjplimp@sandia.gov
+
+   Copyright (2003) Sandia Corporation.  Under the terms of Contract
+   DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
+   certain rights in this software.  This software is distributed under
+   the GNU General Public License.
+
+   See the README file in the top-level LAMMPS directory.
+------------------------------------------------------------------------- */
+
+/* ----------------------------------------------------------------------
+   Contributing author: Stan Moore (SNL)
+------------------------------------------------------------------------- */
+
+#include "npair_halffull_newton_trim_intel.h"
+
+#include "atom.h"
+#include "comm.h"
+#include "error.h"
+#include "modify.h"
+#include "my_page.h"
+#include "neigh_list.h"
+
+using namespace LAMMPS_NS;
+
+/* ---------------------------------------------------------------------- */
+
+NPairHalffullNewtonTrimIntel::NPairHalffullNewtonTrimIntel(LAMMPS *lmp) : NPair(lmp) {
+  _fix = static_cast<FixIntel *>(modify->get_fix_by_id("package_intel"));
+  if (!_fix) error->all(FLERR, "The 'package intel' command is required for /intel styles");
+}
+
+/* ----------------------------------------------------------------------
+   build half list from full list and trim to shorter cutoff
+   pair stored once if i,j are both owned and i < j
+   if j is ghost, only store if j coords are "above and to the right" of i
+   works if full list is a skip list
+------------------------------------------------------------------------- */
+
+template <class flt_t, class acc_t>
+void NPairHalffullNewtonTrimIntel::build_t(NeighList *list,
+                                       IntelBuffers<flt_t,acc_t> *buffers)
+{
+  const int inum_full = list->listfull->inum;
+  const int nlocal = atom->nlocal;
+  const int e_nall = nlocal + atom->nghost;
+  const ATOM_T * _noalias const x = buffers->get_x();
+  int * _noalias const ilist = list->ilist;
+  int * _noalias const numneigh = list->numneigh;
+  int ** _noalias const firstneigh = list->firstneigh;
+  const int * _noalias const ilist_full = list->listfull->ilist;
+  const int * _noalias const numneigh_full = list->listfull->numneigh;
+  const int ** _noalias const firstneigh_full = (const int ** const)list->listfull->firstneigh;  // NOLINT
+  
+  const double cutsq_custom = cutoff_custom * cutoff_custom;
+
+  #if defined(_OPENMP)
+  #pragma omp parallel
+  #endif
+  {
+    int tid, ifrom, ito;
+    IP_PRE_omp_range_id(ifrom, ito, tid, inum_full, comm->nthreads);
+
+    // each thread has its own page allocator
+    MyPage<int> &ipage = list->ipage[tid];
+    ipage.reset();
+
+    // loop over parent full list
+    for (int ii = ifrom; ii < ito; ii++) {
+      int n = 0;
+      int *neighptr = ipage.vget();
+
+      const int i = ilist_full[ii];
+      const flt_t xtmp = x[i].x;
+      const flt_t ytmp = x[i].y;
+      const flt_t ztmp = x[i].z;
+
+      // loop over full neighbor list
+
+      const int * _noalias const jlist = firstneigh_full[i];
+      const int jnum = numneigh_full[i];
+
+      #if defined(LMP_SIMD_COMPILER)
+      #pragma vector aligned
+      #pragma ivdep
+      #endif
+      for (int jj = 0; jj < jnum; jj++) {
+        const int joriginal = jlist[jj];
+        const int j = joriginal & NEIGHMASK;
+        int addme = 1;
+        if (j < nlocal) {
+          if (i > j) addme = 0;
+        } else {
+          if (x[j].z < ztmp) addme = 0;
+          if (x[j].z == ztmp) {
+            if (x[j].y < ytmp) addme = 0;
+            if (x[j].y == ytmp && x[j].x < xtmp) addme = 0;
+          }
+        }
+
+        // trim to shorter cutoff
+
+        const flt_t delx = xtmp - x[j].x;
+        const flt_t dely = ytmp - x[j].y;
+        const flt_t delz = ztmp - x[j].z;
+        const flt_t rsq = delx * delx + dely * dely + delz * delz;
+
+        if (rsq > cutsq_custom) addme = 0;
+
+        if (addme)
+          neighptr[n++] = joriginal;
+      }
+
+      ilist[ii] = i;
+      firstneigh[i] = neighptr;
+      numneigh[i] = n;
+
+      int pad_end = n;
+      IP_PRE_neighbor_pad(pad_end, 0);
+      #if defined(LMP_SIMD_COMPILER)
+      #pragma vector aligned
+      #pragma loop_count min=1, max=INTEL_COMPILE_WIDTH-1, \
+              avg=INTEL_COMPILE_WIDTH/2
+      #endif
+      for ( ; n < pad_end; n++)
+        neighptr[n] = e_nall;
+
+      ipage.vgot(n);
+      if (ipage.status())
+        error->one(FLERR,"Neighbor list overflow, boost neigh_modify one");
+    }
+  }
+  list->inum = inum_full;
+}
+
+/* ----------------------------------------------------------------------
+   build half list from full 3-body list and trim to shorter cutoff
+   half list is already stored as first part of 3-body list
+------------------------------------------------------------------------- */
+
+template <class flt_t, class acc_t>
+void NPairHalffullNewtonTrimIntel::build_t3(NeighList *list, int *numhalf,
+                                            IntelBuffers<flt_t,acc_t> *buffers)
+{
+  const int inum_full = list->listfull->inum;
+  const int e_nall = atom->nlocal + atom->nghost;
+  const ATOM_T * _noalias const x = buffers->get_x();
+  int * _noalias const ilist = list->ilist;
+  int * _noalias const numneigh = list->numneigh;
+  int ** _noalias const firstneigh = list->firstneigh;
+  const int * _noalias const ilist_full = list->listfull->ilist;
+  const int * _noalias const numneigh_full = numhalf;
+  const int ** _noalias const firstneigh_full = (const int ** const)list->listfull->firstneigh;  // NOLINT
+
+  const double cutsq_custom = cutoff_custom * cutoff_custom;
+
+  int packthreads = 1;
+  if (comm->nthreads > INTEL_HTHREADS) packthreads = comm->nthreads;
+
+  #if defined(_OPENMP)
+  #pragma omp parallel if (packthreads > 1)
+  #endif
+  {
+    int tid, ifrom, ito;
+    IP_PRE_omp_range_id(ifrom, ito, tid, inum_full, packthreads);
+
+    // each thread has its own page allocator
+    MyPage<int> &ipage = list->ipage[tid];
+    ipage.reset();
+
+    // loop over parent full list
+    for (int ii = ifrom; ii < ito; ii++) {
+      int n = 0;
+      int *neighptr = ipage.vget();
+
+      const int i = ilist_full[ii];
+      const flt_t xtmp = x[i].x;
+      const flt_t ytmp = x[i].y;
+      const flt_t ztmp = x[i].z;
+
+      // loop over full neighbor list
+
+      const int * _noalias const jlist = firstneigh_full[i];
+      const int jnum = numneigh_full[ii];
+
+      #if defined(LMP_SIMD_COMPILER)
+      #pragma vector aligned
+      #pragma ivdep
+      #endif
+      for (int jj = 0; jj < jnum; jj++) {
+        const int joriginal = jlist[jj];
+        const int j = joriginal & NEIGHMASK;
+        int addme = 1;
+
+        // trim to shorter cutoff
+
+        const flt_t delx = xtmp - x[j].x;
+        const flt_t dely = ytmp - x[j].y;
+        const flt_t delz = ztmp - x[j].z;
+        const flt_t rsq = delx * delx + dely * dely + delz * delz;
+
+        if (rsq > cutsq_custom) addme = 0;
+
+        if (addme)
+          neighptr[n++] = joriginal;
+      }
+
+      ilist[ii] = i;
+      firstneigh[i] = neighptr;
+      numneigh[i] = n;
+
+      int pad_end = n;
+      IP_PRE_neighbor_pad(pad_end, 0);
+      #if defined(LMP_SIMD_COMPILER)
+      #pragma vector aligned
+      #pragma loop_count min=1, max=INTEL_COMPILE_WIDTH-1, \
+              avg=INTEL_COMPILE_WIDTH/2
+      #endif
+      for ( ; n < pad_end; n++)
+        neighptr[n] = e_nall;
+
+      ipage.vgot(n);
+      if (ipage.status())
+        error->one(FLERR,"Neighbor list overflow, boost neigh_modify one");
+    }
+  }
+  list->inum = inum_full;
+}
+
+/* ---------------------------------------------------------------------- */
+
+void NPairHalffullNewtonTrimIntel::build(NeighList *list)
+{
+  if (_fix->three_body_neighbor() == 0) {
+    if (_fix->precision() == FixIntel::PREC_MODE_MIXED)
+      build_t(list, _fix->get_mixed_buffers());
+    else if (_fix->precision() == FixIntel::PREC_MODE_DOUBLE)
+      build_t(list, _fix->get_double_buffers());
+    else
+      build_t(list, _fix->get_single_buffers());
+  } else {
+    int *nhalf, *cnum;
+    if (_fix->precision() == FixIntel::PREC_MODE_MIXED) {
+      _fix->get_mixed_buffers()->get_list_data3(list->listfull, nhalf, cnum);
+      build_t3<float>(list, nhalf, _fix->get_mixed_buffers());
+    } else if (_fix->precision() == FixIntel::PREC_MODE_DOUBLE) {
+      _fix->get_double_buffers()->get_list_data3(list->listfull, nhalf, cnum);
+      build_t3<double>(list, nhalf, _fix->get_double_buffers());
+    } else {
+      _fix->get_single_buffers()->get_list_data3(list->listfull, nhalf, cnum);
+      build_t3<float>(list, nhalf, _fix->get_single_buffers());
+    }
+  }
+}

src/INTEL/npair_halffull_newton_trim_intel.h

@@ -0,0 +1,61 @@
+// clang-format off
+/* -*- c++ -*- ----------------------------------------------------------
+   LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
+   https://www.lammps.org/, Sandia National Laboratories
+   Steve Plimpton, sjplimp@sandia.gov
+
+   Copyright (2003) Sandia Corporation.  Under the terms of Contract
+   DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
+   certain rights in this software.  This software is distributed under
+   the GNU General Public License.
+
+   See the README file in the top-level LAMMPS directory.
+------------------------------------------------------------------------- */
+
+/* ----------------------------------------------------------------------
+   Contributing author: Stan Moore (SNL)
+------------------------------------------------------------------------- */
+
+#ifdef NPAIR_CLASS
+// clang-format off
+NPairStyle(halffull/newton/trim/intel,
+           NPairHalffullNewtonTrimIntel,
+           NP_HALF_FULL | NP_NEWTON | NP_HALF | NP_NSQ | NP_BIN | NP_MULTI |
+           NP_ORTHO | NP_TRI| NP_TRIM | NP_INTEL);
+
+NPairStyle(halffull/newton/skip/trim/intel,
+           NPairHalffullNewtonTrimIntel,
+           NP_HALF_FULL | NP_NEWTON | NP_HALF | NP_NSQ | NP_BIN | NP_MULTI |
+           NP_ORTHO | NP_TRI | NP_SKIP | NP_TRIM | NP_INTEL);
+// clang-format on
+#else
+
+#ifndef LMP_NPAIR_HALFFULL_NEWTON_TRIM_INTEL_H
+#define LMP_NPAIR_HALFFULL_NEWTON_TRIM_INTEL_H
+
+#include "fix_intel.h"
+#include "npair.h"
+
+#if defined(_OPENMP)
+#include <omp.h>
+#endif
+
+namespace LAMMPS_NS {
+
+class NPairHalffullNewtonTrimIntel : public NPair {
+ public:
+  NPairHalffullNewtonTrimIntel(class LAMMPS *);
+  void build(class NeighList *) override;
+
+ protected:
+  FixIntel *_fix;
+
+  template <class flt_t, class acc_t> void build_t(NeighList *, IntelBuffers<flt_t, acc_t> *);
+
+  template <class flt_t, class acc_t> void build_t3(NeighList *, int *, IntelBuffers<flt_t, acc_t> *);
+};
+
+}    // namespace LAMMPS_NS
+
+#endif
+#endif

src/INTEL/npair_trim_intel.cpp

@@ -0,0 +1,138 @@
+// clang-format off
+/* ----------------------------------------------------------------------
+   LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
+   https://www.lammps.org/, Sandia National Laboratories
+   Steve Plimpton, sjplimp@sandia.gov
+
+   Copyright (2003) Sandia Corporation.  Under the terms of Contract
+   DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
+   certain rights in this software.  This software is distributed under
+   the GNU General Public License.
+
+   See the README file in the top-level LAMMPS directory.
+------------------------------------------------------------------------- */
+
+/* ----------------------------------------------------------------------
+   Contributing author: Stan Moore (SNL)
+------------------------------------------------------------------------- */
+
+#include "npair_trim_intel.h"
+
+#include "atom.h"
+#include "comm.h"
+#include "error.h"
+#include "modify.h"
+#include "my_page.h"
+#include "neigh_list.h"
+
+using namespace LAMMPS_NS;
+
+/* ---------------------------------------------------------------------- */
+
+NPairTrimIntel::NPairTrimIntel(LAMMPS *lmp) : NPair(lmp) {
+  _fix = static_cast<FixIntel *>(modify->get_fix_by_id("package_intel"));
+  if (!_fix) error->all(FLERR, "The 'package intel' command is required for /intel styles");
+}
+
+/* ----------------------------------------------------------------------
+   trim from copy list to shorter cutoff
+------------------------------------------------------------------------- */
+
+template <class flt_t, class acc_t>
+void NPairTrimIntel::build_t(NeighList *list,
+                                       IntelBuffers<flt_t,acc_t> *buffers)
+{
+  const int inum_copy = list->listcopy->inum;
+  const int nlocal = atom->nlocal;
+  const int e_nall = nlocal + atom->nghost;
+  const ATOM_T * _noalias const x = buffers->get_x();
+  int * _noalias const ilist = list->ilist;
+  int * _noalias const numneigh = list->numneigh;
+  int ** _noalias const firstneigh = list->firstneigh;
+  const int * _noalias const ilist_copy = list->listcopy->ilist;
+  const int * _noalias const numneigh_copy = list->listcopy->numneigh;
+  const int ** _noalias const firstneigh_copy = (const int ** const)list->listcopy->firstneigh;  // NOLINT
+  
+  const double cutsq_custom = cutoff_custom * cutoff_custom;
+
+  #if defined(_OPENMP)
+  #pragma omp parallel
+  #endif
+  {
+    int tid, ifrom, ito;
+    IP_PRE_omp_range_id(ifrom, ito, tid, inum_copy, comm->nthreads);
+
+    // each thread has its own page allocator
+    MyPage<int> &ipage = list->ipage[tid];
+    ipage.reset();
+
+    // loop over parent copy list
+    for (int ii = ifrom; ii < ito; ii++) {
+      int n = 0;
+      int *neighptr = ipage.vget();
+
+      const int i = ilist_copy[ii];
+      const flt_t xtmp = x[i].x;
+      const flt_t ytmp = x[i].y;
+      const flt_t ztmp = x[i].z;
+
+      // loop over copy neighbor list
+
+      const int * _noalias const jlist = firstneigh_copy[i];
+      const int jnum = numneigh_copy[i];
+
+      #if defined(LMP_SIMD_COMPILER)
+      #pragma vector aligned
+      #pragma ivdep
+      #endif
+      for (int jj = 0; jj < jnum; jj++) {
+        const int joriginal = jlist[jj];
+        const int j = joriginal & NEIGHMASK;
+        int addme = 1;
+
+        // trim to shorter cutoff
+
+        const flt_t delx = xtmp - x[j].x;
+        const flt_t dely = ytmp - x[j].y;
+        const flt_t delz = ztmp - x[j].z;
+        const flt_t rsq = delx * delx + dely * dely + delz * delz;
+
+        if (rsq > cutsq_custom) addme = 0;
+
+        if (addme)
+          neighptr[n++] = joriginal;
+      }
+
+      ilist[ii] = i;
+      firstneigh[i] = neighptr;
+      numneigh[i] = n;
+
+      int pad_end = n;
+      IP_PRE_neighbor_pad(pad_end, 0);
+      #if defined(LMP_SIMD_COMPILER)
+      #pragma vector aligned
+      #pragma loop_count min=1, max=INTEL_COMPILE_WIDTH-1, \
+              avg=INTEL_COMPILE_WIDTH/2
+      #endif
+      for ( ; n < pad_end; n++)
+        neighptr[n] = e_nall;
+
+      ipage.vgot(n);
+      if (ipage.status())
+        error->one(FLERR,"Neighbor list overflow, boost neigh_modify one");
+    }
+  }
+  list->inum = inum_copy;
+}
+
+/* ---------------------------------------------------------------------- */
+
+void NPairTrimIntel::build(NeighList *list)
+{
+  if (_fix->precision() == FixIntel::PREC_MODE_MIXED)
+    build_t(list, _fix->get_mixed_buffers());
+  else if (_fix->precision() == FixIntel::PREC_MODE_DOUBLE)
+    build_t(list, _fix->get_double_buffers());
+  else
+    build_t(list, _fix->get_single_buffers());
+}

src/INTEL/npair_trim_intel.h

@@ -0,0 +1,53 @@
+// clang-format off
+/* -*- c++ -*- ----------------------------------------------------------
+   LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
+   https://www.lammps.org/, Sandia National Laboratories
+   Steve Plimpton, sjplimp@sandia.gov
+
+   Copyright (2003) Sandia Corporation.  Under the terms of Contract
+   DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
+   certain rights in this software.  This software is distributed under
+   the GNU General Public License.
+
+   See the README file in the top-level LAMMPS directory.
+------------------------------------------------------------------------- */
+
+/* ----------------------------------------------------------------------
+   Contributing author: Stan Moore (SNL)
+------------------------------------------------------------------------- */
+
+#ifdef NPAIR_CLASS
+// clang-format off
+NPairStyle(trim/intel,
+           NPairTrimIntel,
+           NP_COPY | NP_TRIM | NP_INTEL);
+// clang-format on
+#else
+
+#ifndef LMP_NPAIR_TRIM_INTEL_H
+#define LMP_NPAIR_TRIM_INTEL_H
+
+#include "fix_intel.h"
+#include "npair.h"
+
+#if defined(_OPENMP)
+#include <omp.h>
+#endif
+
+namespace LAMMPS_NS {
+
+class NPairTrimIntel : public NPair {
+ public:
+  NPairTrimIntel(class LAMMPS *);
+  void build(class NeighList *) override;
+
+ protected:
+  FixIntel *_fix;
+
+  template <class flt_t, class acc_t> void build_t(NeighList *, IntelBuffers<flt_t, acc_t> *);
+};
+
+}    // namespace LAMMPS_NS
+
+#endif
+#endif