Page Menu
Home
c4science
Search
Configure Global Search
Log In
Files
F66224307
lal_born.cu
No One
Temporary
Actions
Download File
Edit File
Delete File
View Transforms
Subscribe
Mute Notifications
Award Token
Subscribers
None
File Metadata
Details
File Info
Storage
Attached
Created
Sun, Jun 9, 03:12
Size
6 KB
Mime Type
text/x-c
Expires
Tue, Jun 11, 03:12 (2 d)
Engine
blob
Format
Raw Data
Handle
18188253
Attached To
rLAMMPS lammps
lal_born.cu
View Options
// **************************************************************************
// born.cu
// -------------------
// Trung Dac Nguyen (ORNL)
//
// Device code for acceleration of the born pair style
//
// __________________________________________________________________________
// This file is part of the LAMMPS Accelerator Library (LAMMPS_AL)
// __________________________________________________________________________
//
// begin :
// email : nguyentd@ornl.gov
// ***************************************************************************/
#ifdef NV_KERNEL
#include "lal_aux_fun1.h"
#ifndef _DOUBLE_DOUBLE
texture<float4> pos_tex;
#else
texture<int4,1> pos_tex;
#endif
#else
#define pos_tex x_
#endif
__kernel void k_born(const __global numtyp4 *restrict x_,
const __global numtyp4 *restrict coeff1,
const __global numtyp4 *restrict coeff2,
const __global numtyp2 *restrict cutsq_sigma,
const int lj_types,
const __global numtyp *restrict sp_lj_in,
const __global int *dev_nbor,
const __global int *dev_packed,
__global acctyp4 *restrict ans,
__global acctyp *restrict engv,
const int eflag, const int vflag, const int inum,
const int nbor_pitch, const int t_per_atom) {
int tid, ii, offset;
atom_info(t_per_atom,ii,tid,offset);
__local numtyp sp_lj[4];
sp_lj[0]=sp_lj_in[0];
sp_lj[1]=sp_lj_in[1];
sp_lj[2]=sp_lj_in[2];
sp_lj[3]=sp_lj_in[3];
acctyp energy=(acctyp)0;
acctyp4 f;
f.x=(acctyp)0; f.y=(acctyp)0; f.z=(acctyp)0;
acctyp virial[6];
for (int i=0; i<6; i++)
virial[i]=(acctyp)0;
if (ii<inum) {
int nbor, nbor_end;
int i, numj;
__local int n_stride;
nbor_info(dev_nbor,dev_packed,nbor_pitch,t_per_atom,ii,offset,i,numj,
n_stride,nbor_end,nbor);
numtyp4 ix; fetch4(ix,i,pos_tex); //x_[i];
int itype=ix.w;
numtyp factor_lj;
for ( ; nbor<nbor_end; nbor+=n_stride) {
int j=dev_packed[nbor];
factor_lj = sp_lj[sbmask(j)];
j &= NEIGHMASK;
numtyp4 jx; fetch4(jx,j,pos_tex); //x_[j];
int jtype=jx.w;
// Compute r12
numtyp delx = ix.x-jx.x;
numtyp dely = ix.y-jx.y;
numtyp delz = ix.z-jx.z;
numtyp r2inv = delx*delx+dely*dely+delz*delz;
int mtype=itype*lj_types+jtype;
if (r2inv<cutsq_sigma[mtype].x) {
numtyp r=ucl_sqrt(r2inv);
numtyp rexp = ucl_exp((cutsq_sigma[mtype].y-r)*coeff1[mtype].x);
r2inv=ucl_recip(r2inv);
numtyp r6inv = r2inv*r2inv*r2inv;
numtyp force = r2inv*(coeff1[mtype].y*r*rexp
- coeff1[mtype].z*r6inv + coeff1[mtype].w*r2inv*r6inv);
force*=factor_lj;
f.x+=delx*force;
f.y+=dely*force;
f.z+=delz*force;
if (eflag>0) {
numtyp e=coeff2[mtype].x*rexp - coeff2[mtype].y*r6inv
+ coeff2[mtype].z*r2inv*r6inv;
energy+=factor_lj*(e-coeff2[mtype].w);
}
if (vflag>0) {
virial[0] += delx*delx*force;
virial[1] += dely*dely*force;
virial[2] += delz*delz*force;
virial[3] += delx*dely*force;
virial[4] += delx*delz*force;
virial[5] += dely*delz*force;
}
}
} // for nbor
store_answers(f,energy,virial,ii,inum,tid,t_per_atom,offset,eflag,vflag,
ans,engv);
} // if ii
}
__kernel void k_born_fast(const __global numtyp4 *restrict x_,
const __global numtyp4 *restrict coeff1_in,
const __global numtyp4 *restrict coeff2_in,
const __global numtyp2 *restrict cutsq_sigma,
const __global numtyp *restrict sp_lj_in,
const __global int *dev_nbor,
const __global int *dev_packed,
__global acctyp4 *restrict ans,
__global acctyp *restrict engv,
const int eflag, const int vflag, const int inum,
const int nbor_pitch, const int t_per_atom) {
int tid, ii, offset;
atom_info(t_per_atom,ii,tid,offset);
__local numtyp4 coeff1[MAX_SHARED_TYPES*MAX_SHARED_TYPES];
__local numtyp4 coeff2[MAX_SHARED_TYPES*MAX_SHARED_TYPES];
__local numtyp sp_lj[4];
if (tid<4)
sp_lj[tid]=sp_lj_in[tid];
if (tid<MAX_SHARED_TYPES*MAX_SHARED_TYPES) {
coeff1[tid]=coeff1_in[tid];
if (eflag>0)
coeff2[tid]=coeff2_in[tid];
}
acctyp energy=(acctyp)0;
acctyp4 f;
f.x=(acctyp)0; f.y=(acctyp)0; f.z=(acctyp)0;
acctyp virial[6];
for (int i=0; i<6; i++)
virial[i]=(acctyp)0;
__syncthreads();
if (ii<inum) {
int nbor, nbor_end;
int i, numj;
__local int n_stride;
nbor_info(dev_nbor,dev_packed,nbor_pitch,t_per_atom,ii,offset,i,numj,
n_stride,nbor_end,nbor);
numtyp4 ix; fetch4(ix,i,pos_tex); //x_[i];
int iw=ix.w;
int itype=fast_mul((int)MAX_SHARED_TYPES,iw);
numtyp factor_lj;
for ( ; nbor<nbor_end; nbor+=n_stride) {
int j=dev_packed[nbor];
factor_lj = sp_lj[sbmask(j)];
j &= NEIGHMASK;
numtyp4 jx; fetch4(jx,j,pos_tex); //x_[j];
int mtype=itype+jx.w;
// Compute r12
numtyp delx = ix.x-jx.x;
numtyp dely = ix.y-jx.y;
numtyp delz = ix.z-jx.z;
numtyp r2inv = delx*delx+dely*dely+delz*delz;
if (r2inv<cutsq_sigma[mtype].x) {
numtyp r=ucl_sqrt(r2inv);
numtyp rexp = ucl_exp((cutsq_sigma[mtype].y-r)*coeff1[mtype].x);
r2inv=ucl_recip(r2inv);
numtyp r6inv = r2inv*r2inv*r2inv;
numtyp force = r2inv*(coeff1[mtype].y*r*rexp
- coeff1[mtype].z*r6inv + coeff1[mtype].w*r2inv*r6inv);
force*=factor_lj;
f.x+=delx*force;
f.y+=dely*force;
f.z+=delz*force;
if (eflag>0) {
numtyp e=coeff2[mtype].x*rexp - coeff2[mtype].y*r6inv
+ coeff2[mtype].z*r2inv*r6inv;
energy+=factor_lj*(e-coeff2[mtype].w);
}
if (vflag>0) {
virial[0] += delx*delx*force;
virial[1] += dely*dely*force;
virial[2] += delz*delz*force;
virial[3] += delx*dely*force;
virial[4] += delx*delz*force;
virial[5] += dely*delz*force;
}
}
} // for nbor
store_answers(f,energy,virial,ii,inum,tid,t_per_atom,offset,eflag,vflag,
ans,engv);
} // if ii
}
Event Timeline
Log In to Comment