git-svn-id: svn://svn.icms.temple.edu/lammps-ro/trunk@6406 f3b2605a-c512-4ea7-a41b-209d697bcdaa

2011-06-14 13:33:48 +00:00
parent 479275be1d
commit 9418bacff7
17 changed files with 3065 additions and 0 deletions
--- a/lib/gpu/re_squared.h
+++ b/lib/gpu/re_squared.h
@ -0,0 +1,91 @@
+/***************************************************************************
+                                 re_squared.h
+                             -------------------
+                               W. Michael Brown
+
+  Host code for RE-Squared potential acceleration
+
+ __________________________________________________________________________
+    This file is part of the LAMMPS Accelerator Library (LAMMPS_AL)
+ __________________________________________________________________________
+
+    begin                : Fri May 06 2011
+    email                : brownw@ornl.gov
+ ***************************************************************************/
+
+#ifndef RE_SQUARED_H
+#define RE_SQUARED_H
+
+#include "base_ellipsoid.h"
+#include "mpi.h"
+
+namespace LAMMPS_AL {
+
+template <class numtyp, class acctyp>
+class RESquared : public BaseEllipsoid<numtyp, acctyp> {
+ public:
+  RESquared();
+  ~RESquared(); 
+
+  /// Clear any previous data and set up for a new LAMMPS run
+  /** \param gpu_nbor true if neighboring performed on device
+    * \param max_nbors initial number of rows in the neighbor matrix
+    * \param cell_size cutoff + skin
+    * \param gpu_split fraction of particles handled by device 
+    * \return false if there is not sufficient memory or device init prob
+    * 
+    * Returns:
+    * -  0 if successfull
+    * - -1 if fix gpu not found
+    * - -3 if there is an out of memory error
+    * - -4 if the GPU library was not compiled for GPU
+    * - -5 Double precision is not supported on card **/
+  int init(const int ntypes, double **host_shape, double **host_well,
+           double **host_cutsq, double **host_sigma,  double **host_epsilon,
+           int **h_form, double **host_lj1, double **host_lj2, 
+           double **host_lj3, double **host_lj4, double **host_offset,
+           const double *host_special_lj, const int nlocal, const int nall,
+           const int max_nbors, const int maxspecial, const double cell_size,
+           const double gpu_split, FILE *screen);
+
+  /// Clear all host and device data
+  /** \note This is called at the beginning of the init() routine **/
+  void clear();
+ 
+  /// Returns memory usage on device per atom
+  int bytes_per_atom(const int max_nbors) const;
+
+  /// Total host memory used by library for pair style
+  double host_memory_usage() const;
+
+  /// Device Error Flag - Set if a bad matrix inversion occurs
+  UCL_D_Vec<int> dev_error;
+  
+  // --------------------------- TYPE DATA -------------------------- 
+
+  /// lj1.x = lj1, lj1.y = lj2, lj1.z = cutsq, lj1.w = form
+  UCL_D_Vec<numtyp4> lj1;
+  /// lj3.x = lj3, lj3.y = lj4, lj3.z = offset
+  UCL_D_Vec<numtyp4> lj3;
+  /// sigma_epsilon.x = sigma, sigma_epsilon.y = epsilon
+  UCL_D_Vec<numtyp2> sigma_epsilon;
+  /// special lj 0-4
+  UCL_D_Vec<numtyp> special_lj;
+  
+  /// If atom type constants fit in shared memory, use fast kernels
+  bool _shared_types;
+  int _lj_types;
+   
+  // --------------------------- ATOM DATA -------------------------- 
+
+  /// Aspherical Const Data for Atoms
+  UCL_D_Vec<numtyp4> shape, well;
+
+ private:
+  bool _allocated;
+  void loop(const bool _eflag, const bool _vflag);
+};
+
+}
+
+#endif