Page Menu
Home
c4science
Search
Configure Global Search
Log In
Files
F87252218
lal_eam.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
Fri, Oct 11, 13:38
Size
14 KB
Mime Type
text/x-c
Expires
Sun, Oct 13, 13:38 (2 d)
Engine
blob
Format
Raw Data
Handle
21563655
Attached To
rLAMMPS lammps
lal_eam.cu
View Options
// **************************************************************************
// lal_eam.cu
// -------------------
// Trung Dac Nguyen, W. Michael Brown (ORNL)
//
// Device code for acceleration of the eam pair style
//
// __________________________________________________________________________
// This file is part of the LAMMPS Accelerator Library (LAMMPS_AL)
// __________________________________________________________________________
//
// begin :
// email : brownw@ornl.gov nguyentd@ornl.gov
// ***************************************************************************/
#ifdef NV_KERNEL
#include "lal_aux_fun1.h"
texture<float4> pos_tex;
texture<float> fp_tex;
#ifndef _DOUBLE_DOUBLE
ucl_inline float4 fetch_pos(const int& i, const float4 *pos)
{ return tex1Dfetch(pos_tex, i); }
ucl_inline float fetch_q(const int& i, const float *fp)
{ return tex1Dfetch(fp_tex, i); }
#endif
#endif
#define MIN(A,B) ((A) < (B) ? (A) : (B))
#define MAX(A,B) ((A) > (B) ? (A) : (B))
#define store_answers_eam(f, energy, virial, ii, inum, tid, t_per_atom, \
offset, elag, vflag, ans, engv) \
if (t_per_atom>1) { \
__local acctyp red_acc[6][BLOCK_PAIR]; \
red_acc[0][tid]=f.x; \
red_acc[1][tid]=f.y; \
red_acc[2][tid]=f.z; \
red_acc[3][tid]=energy; \
for (unsigned int s=t_per_atom/2; s>0; s>>=1) { \
if (offset < s) { \
for (int r=0; r<4; r++) \
red_acc[r][tid] += red_acc[r][tid+s]; \
} \
} \
f.x=red_acc[0][tid]; \
f.y=red_acc[1][tid]; \
f.z=red_acc[2][tid]; \
energy=red_acc[3][tid]; \
if (vflag>0) { \
for (int r=0; r<6; r++) \
red_acc[r][tid]=virial[r]; \
for (unsigned int s=t_per_atom/2; s>0; s>>=1) { \
if (offset < s) { \
for (int r=0; r<6; r++) \
red_acc[r][tid] += red_acc[r][tid+s]; \
} \
} \
for (int r=0; r<6; r++) \
virial[r]=red_acc[r][tid]; \
} \
} \
if (offset==0) { \
engv+=ii; \
if (eflag>0) { \
*engv+=energy; \
engv+=inum; \
} \
if (vflag>0) { \
for (int i=0; i<6; i++) { \
*engv=virial[i]; \
engv+=inum; \
} \
} \
ans[ii]=f; \
}
__kernel void kernel_energy(__global numtyp4 *x_,
__global numtyp2 *type2rhor_z2r, __global numtyp *type2frho,
__global numtyp4 *rhor_spline2, __global numtyp4 *frho_spline1,
__global numtyp4 *frho_spline2,
__global int *dev_nbor, __global int *dev_packed,
__global numtyp *fp_,
__global acctyp *engv, const int eflag,
const int inum,
const int nbor_pitch,
const int ntypes, const numtyp cutforcesq,
const numtyp rdr, const numtyp rdrho,
const int nrho, const int nr,
const int t_per_atom) {
int tid, ii, offset;
atom_info(t_per_atom,ii,tid,offset);
acctyp rho = (acctyp)0;
acctyp energy = (acctyp)0;
if (ii<inum) {
__global int *nbor, *list_end;
int i, numj, n_stride;
nbor_info(dev_nbor,dev_packed,nbor_pitch,t_per_atom,ii,offset,i,numj,
n_stride,list_end,nbor);
numtyp4 ix=fetch_pos(i,x_); //x_[i];
int itype=ix.w;
for ( ; nbor<list_end; nbor+=n_stride) {
int j=*nbor;
j &= NEIGHMASK;
numtyp4 jx=fetch_pos(j,x_); //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 rsq = delx*delx+dely*dely+delz*delz;
if (rsq<cutforcesq) {
numtyp p = ucl_sqrt(rsq)*rdr + (numtyp)1.0;
int m=p;
m = MIN(m,nr-1);
p -= m;
p = MIN(p,(numtyp)1.0);
int mtype = jtype*ntypes+itype;
int index = type2rhor_z2r[mtype].x*(nr+1)+m;
numtyp4 coeff = rhor_spline2[index];
rho += ((coeff.x*p + coeff.y)*p + coeff.z)*p + coeff.w;
}
} // for nbor
// reduce to get the density at atom ii
if (t_per_atom>1) {
__local acctyp red_acc[BLOCK_PAIR];
red_acc[tid]=rho;
for (unsigned int s=t_per_atom/2; s>0; s>>=1) {
if (offset < s)
red_acc[tid] += red_acc[tid+s];
}
rho=red_acc[tid];
}
// calculate the embedded force for ii
if (offset==0) {
numtyp p = rho*rdrho + (numtyp)1.0;
int m=p;
m = MAX(1,MIN(m,nrho-1));
p -= m;
p = MIN(p,(numtyp)1.0);
int index = type2frho[itype]*(nr+1)+m;
numtyp4 coeff = frho_spline1[index];
numtyp fp = (coeff.x*p + coeff.y)*p + coeff.z;
fp_[ii]=fp;
engv+=ii;
if (eflag>0) {
coeff = frho_spline2[index];
energy = ((coeff.x*p + coeff.y)*p + coeff.z)*p + coeff.w;
*engv=(acctyp)2.0*energy;
}
}
} // if ii
}
__kernel void kernel_pair(__global numtyp4 *x_, __global numtyp *fp_,
__global numtyp2 *type2rhor_z2r,
__global numtyp4 *rhor_spline1, __global numtyp *z2r_spline,
__global int *dev_nbor, __global int *dev_packed,
__global acctyp4 *ans, __global acctyp *engv,
const int eflag, const int vflag,
const int inum, const int nbor_pitch,
const int ntypes, const numtyp cutforcesq,
const numtyp rdr, const int nr,
const int t_per_atom) {
int tid, ii, offset;
atom_info(t_per_atom,ii,tid,offset);
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) {
__global int *nbor, *list_end;
int i, numj, n_stride;
nbor_info(dev_nbor,dev_packed,nbor_pitch,t_per_atom,ii,offset,i,numj,
n_stride,list_end,nbor);
numtyp4 ix=fetch_pos(i,x_); //x_[i];
numtyp ifp=fetch_q(i,fp_); //fp_[i];
int itype=ix.w;
for ( ; nbor<list_end; nbor+=n_stride) {
int j=*nbor;
j &= NEIGHMASK;
numtyp4 jx=fetch_pos(j,x_); //x_[j];
numtyp jfp=fetch_q(j,fp_); //fp_[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 rsq = delx*delx+dely*dely+delz*delz;
if (rsq<cutforcesq) {
numtyp r = ucl_sqrt(rsq);
numtyp p = r*rdr + (numtyp)1.0;
int m=p;
m = MIN(m,nr-1);
p -= m;
p = MIN(p,(numtyp)1.0);
int mtype,index;
numtyp coeff0,coeff1,coeff2,coeff3,coeff4,coeff5,coeff6;
numtyp4 coeff;
mtype = itype*ntypes+jtype;
index = type2rhor_z2r[mtype].x*(nr+1)+m;
coeff = rhor_spline1[index];
numtyp rhoip = (coeff.x*p + coeff.y)*p + coeff.z;
mtype = jtype*ntypes+itype;
index = type2rhor_z2r[mtype].x*(nr+1)+m;
coeff = rhor_spline1[index];
numtyp rhojp = (coeff.x*p + coeff.y)*p + coeff.z;
mtype = itype*ntypes+jtype;
index = type2rhor_z2r[mtype].y*(nr+1)*8+m*8;
coeff0 = z2r_spline[index+0];
coeff1 = z2r_spline[index+1];
coeff2 = z2r_spline[index+2];
coeff3 = z2r_spline[index+3];
coeff4 = z2r_spline[index+4];
coeff5 = z2r_spline[index+5];
coeff6 = z2r_spline[index+6];
numtyp z2p = (coeff0*p + coeff1)*p + coeff2;
numtyp z2 = ((coeff3*p + coeff4)*p + coeff5)*p + coeff6;
numtyp recip = (numtyp)1.0/r;
numtyp phi = z2*recip;
numtyp phip = z2p*recip - phi*recip;
numtyp psip = ifp*rhojp + jfp*rhoip + phip;
numtyp force = -psip*recip;
f.x+=delx*force;
f.y+=dely*force;
f.z+=delz*force;
if (eflag>0) {
energy += phi;
}
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_eam(f,energy,virial,ii,inum,tid,t_per_atom,offset,eflag,vflag,
ans,engv);
} // if ii
}
__kernel void kernel_pair_fast(__global numtyp4 *x_, __global numtyp *fp_,
__global numtyp2 *type2rhor_z2r,
__global numtyp4 *rhor_spline1, __global numtyp *z2r_spline,
__global int *dev_nbor, __global int *dev_packed,
__global acctyp4 *ans, __global acctyp *engv,
const int eflag, const int vflag, const int inum,
const int nbor_pitch,
const numtyp cutforcesq,
const numtyp rdr, const int nr,
const int t_per_atom) {
int tid, ii, offset;
atom_info(t_per_atom,ii,tid,offset);
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) {
__global int *nbor, *list_end;
int i, numj, n_stride;
nbor_info(dev_nbor,dev_packed,nbor_pitch,t_per_atom,ii,offset,i,numj,
n_stride,list_end,nbor);
numtyp4 ix=fetch_pos(i,x_); //x_[i];
numtyp ifp=fetch_q(i,fp_); //fp_[i];
int iw=ix.w;
int itype=fast_mul((int)MAX_SHARED_TYPES,iw);
for ( ; nbor<list_end; nbor+=n_stride) {
int j=*nbor;
j &= NEIGHMASK;
numtyp4 jx=fetch_pos(j,x_); //x_[j];
numtyp jfp=fetch_q(j,fp_); //fp_[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 rsq = delx*delx+dely*dely+delz*delz;
if (rsq<cutforcesq) {
numtyp r = ucl_sqrt(rsq);
numtyp p = r*rdr + (numtyp)1.0;
int m=p;
m = MIN(m,nr-1);
p -= m;
p = MIN(p,(numtyp)1.0);
numtyp coeff0,coeff1,coeff2,coeff3,coeff4,coeff5,coeff6;
numtyp4 coeff;
int mtype,index;
mtype = itype+jx.w;
index = type2rhor_z2r[mtype].x*(nr+1)+m;
coeff = rhor_spline1[index];
numtyp rhoip = (coeff.x*p + coeff.y)*p + coeff.z;
mtype = jtype+ix.w;
index = type2rhor_z2r[mtype].x*(nr+1)+m;
coeff = rhor_spline1[index];
numtyp rhojp = (coeff.x*p + coeff.y)*p + coeff.z;
mtype = itype+jx.w;
index = type2rhor_z2r[mtype].y*(nr+1)*8+m*8;
coeff0 = z2r_spline[index+0];
coeff1 = z2r_spline[index+1];
coeff2 = z2r_spline[index+2];
coeff3 = z2r_spline[index+3];
coeff4 = z2r_spline[index+4];
coeff5 = z2r_spline[index+5];
coeff6 = z2r_spline[index+6];
numtyp z2p = (coeff0*p + coeff1)*p + coeff2;
numtyp z2 = ((coeff3*p + coeff4)*p + coeff5)*p + coeff6;
numtyp recip = (numtyp)1.0/r;
numtyp phi = z2*recip;
numtyp phip = z2p*recip - phi*recip;
numtyp psip = ifp*rhojp + jfp*rhoip + phip;
numtyp force = -psip*recip;
f.x+=delx*force;
f.y+=dely*force;
f.z+=delz*force;
if (eflag>0) {
energy += phi;
}
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_eam(f,energy,virial,ii,inum,tid,t_per_atom,offset,eflag,vflag,
ans,engv);
} // if ii
}
Event Timeline
Log In to Comment