Page Menu
Home
c4science
Search
Configure Global Search
Log In
Files
F91973122
coul_long.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
Sat, Nov 16, 07:05
Size
10 KB
Mime Type
text/x-c
Expires
Mon, Nov 18, 07:05 (1 d, 23 h)
Engine
blob
Format
Raw Data
Handle
22356873
Attached To
rLAMMPS lammps
coul_long.cu
View Options
// **************************************************************************
// coul_long.cu
// -------------------
// Axel Kohlmeyer (Temple)
//
// Device code for acceleration of the coul/long pair style
//
// __________________________________________________________________________
// This file is part of the LAMMPS Accelerator Library (LAMMPS_AL)
// __________________________________________________________________________
//
// begin : July 2011
// email : a.kohlmeyer@temple.edu
// ***************************************************************************/
#ifdef NV_KERNEL
#include "nv_kernel_def.h"
texture<float4> pos_tex;
texture<float> q_tex;
#ifdef _DOUBLE_DOUBLE
__inline double4 fetch_pos(const int& i, const double4 *pos)
{
return pos[i];
}
__inline double fetch_q(const int& i, const double *q)
{
return q[i];
}
#else
__inline float4 fetch_pos(const int& i, const float4 *pos)
{
return tex1Dfetch(pos_tex, i);
}
__inline float fetch_q(const int& i, const float *q)
{
return tex1Dfetch(q_tex, i);
}
#endif
#else
#pragma OPENCL EXTENSION cl_khr_fp64: enable
#define GLOBAL_ID_X get_global_id(0)
#define THREAD_ID_X get_local_id(0)
#define BLOCK_ID_X get_group_id(0)
#define BLOCK_SIZE_X get_local_size(0)
#define __syncthreads() barrier(CLK_LOCAL_MEM_FENCE)
#define __inline inline
#define fetch_pos(i,y) x_[i]
#define fetch_q(i,y) q_[i]
#define BLOCK_PAIR 64
#define MAX_SHARED_TYPES 8
#endif
#ifdef _DOUBLE_DOUBLE
#define numtyp double
#define numtyp2 double2
#define numtyp4 double4
#define acctyp double
#define acctyp4 double4
#endif
#ifdef _SINGLE_DOUBLE
#define numtyp float
#define numtyp2 float2
#define numtyp4 float4
#define acctyp double
#define acctyp4 double4
#endif
#ifndef numtyp
#define numtyp float
#define numtyp2 float2
#define numtyp4 float4
#define acctyp float
#define acctyp4 float4
#endif
#define EWALD_F (numtyp)1.12837917
#define EWALD_P (numtyp)0.3275911
#define A1 (numtyp)0.254829592
#define A2 (numtyp)-0.284496736
#define A3 (numtyp)1.421413741
#define A4 (numtyp)-1.453152027
#define A5 (numtyp)1.061405429
#define SBBITS 30
#define NEIGHMASK 0x3FFFFFFF
__inline int sbmask(int j) { return j >> SBBITS & 3; }
__kernel void kernel_pair(__global numtyp4 *x_, __global numtyp4 *lj1,
__global numtyp4* lj3, const int lj_types,
__global numtyp *sp_cl_in, __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, __global numtyp *q_,
const numtyp cut_coulsq, const numtyp qqrd2e,
const numtyp g_ewald, const int t_per_atom) {
int tid=THREAD_ID_X;
int ii=mul24((int)BLOCK_ID_X,(int)(BLOCK_SIZE_X)/t_per_atom);
ii+=tid/t_per_atom;
int offset=tid%t_per_atom;
__local numtyp sp_cl[4];
sp_cl[0]=sp_cl_in[0];
sp_cl[1]=sp_cl_in[1];
sp_cl[2]=sp_cl_in[2];
sp_cl[3]=sp_cl_in[3];
acctyp e_coul=(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=dev_nbor+ii;
int i=*nbor;
nbor+=nbor_pitch;
int numj=*nbor;
nbor+=nbor_pitch;
int n_stride;
__global int *list_end;
if (dev_nbor==dev_packed) {
list_end=nbor+mul24(numj,nbor_pitch);
nbor+=mul24(offset,nbor_pitch);
n_stride=mul24(t_per_atom,nbor_pitch);
} else {
nbor=dev_packed+*nbor;
list_end=nbor+numj;
n_stride=t_per_atom;
nbor+=offset;
}
numtyp4 ix=fetch_pos(i,x_); //x_[i];
numtyp qtmp=fetch_q(i,q_);
for ( ; nbor<list_end; nbor+=n_stride) {
int j=*nbor;
numtyp factor_coul;
factor_coul = (numtyp)1.0-sp_cl[sbmask(j)];
j &= NEIGHMASK;
numtyp4 jx=fetch_pos(j,x_); //x_[j];
// 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 < cut_coulsq) {
numtyp r2inv=(numtyp)1.0/rsq;
numtyp force, prefactor, _erfc;
numtyp r = sqrt(rsq);
numtyp grij = g_ewald * r;
numtyp expm2 = exp(-grij*grij);
numtyp t = (numtyp)1.0 / ((numtyp)1.0 + EWALD_P*grij);
_erfc = t * (A1+t*(A2+t*(A3+t*(A4+t*A5)))) * expm2;
prefactor = qqrd2e * qtmp*fetch_q(j,q_)/r;
force = prefactor * (_erfc + EWALD_F*grij*expm2-factor_coul) * r2inv;
f.x+=delx*force;
f.y+=dely*force;
f.z+=delz*force;
if (eflag>0) {
e_coul += prefactor*(_erfc-factor_coul);
}
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
} // if ii
// Reduce answers
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]=e_coul;
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];
e_coul=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];
}
}
// Store answers
if (ii<inum && offset==0) {
__global acctyp *ap1=engv+ii;
if (eflag>0) {
*ap1=(acctyp)0;
ap1+=inum;
*ap1=e_coul;
ap1+=inum;
}
if (vflag>0) {
for (int i=0; i<6; i++) {
*ap1=virial[i];
ap1+=inum;
}
}
ans[ii]=f;
} // if ii
}
__kernel void kernel_pair_fast(__global numtyp4 *x_, __global numtyp4 *lj1_in,
__global numtyp4* lj3_in,
__global numtyp* sp_cl_in,
__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, __global numtyp *q_,
const numtyp cut_coulsq, const numtyp qqrd2e,
const numtyp g_ewald, const int t_per_atom) {
int tid=THREAD_ID_X;
int ii=mul24((int)BLOCK_ID_X,(int)(BLOCK_SIZE_X)/t_per_atom);
ii+=tid/t_per_atom;
int offset=tid%t_per_atom;
__local numtyp sp_cl[4];
if (tid<4)
sp_cl[tid]=sp_cl_in[tid];
acctyp e_coul=(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) {
__global int *nbor=dev_nbor+ii;
int i=*nbor;
nbor+=nbor_pitch;
int numj=*nbor;
nbor+=nbor_pitch;
int n_stride;
__global int *list_end;
if (dev_nbor==dev_packed) {
list_end=nbor+mul24(numj,nbor_pitch);
nbor+=mul24(offset,nbor_pitch);
n_stride=mul24(t_per_atom,nbor_pitch);
} else {
nbor=dev_packed+*nbor;
list_end=nbor+numj;
n_stride=t_per_atom;
nbor+=offset;
}
numtyp4 ix=fetch_pos(i,x_); //x_[i];
numtyp qtmp=fetch_q(i,q_);
for ( ; nbor<list_end; nbor+=n_stride) {
int j=*nbor;
numtyp factor_coul;
factor_coul = (numtyp)1.0-sp_cl[sbmask(j)];
j &= NEIGHMASK;
numtyp4 jx=fetch_pos(j,x_); //x_[j];
// 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 < cut_coulsq) {
numtyp r2inv=(numtyp)1.0/rsq;
numtyp force, prefactor, _erfc;
numtyp r = sqrt(rsq);
numtyp grij = g_ewald * r;
numtyp expm2 = exp(-grij*grij);
numtyp t = (numtyp)1.0 / ((numtyp)1.0 + EWALD_P*grij);
_erfc = t * (A1+t*(A2+t*(A3+t*(A4+t*A5)))) * expm2;
prefactor = qqrd2e * qtmp*fetch_q(j,q_)/r;
force = prefactor * (_erfc + EWALD_F*grij*expm2-factor_coul) * r2inv;
f.x+=delx*force;
f.y+=dely*force;
f.z+=delz*force;
if (eflag>0) {
e_coul += prefactor*(_erfc-factor_coul);
}
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
} // if ii
// Reduce answers
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]=e_coul;
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];
e_coul=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];
}
}
// Store answers
if (ii<inum && offset==0) {
__global acctyp *ap1=engv+ii;
if (eflag>0) {
*ap1=(acctyp)0;
ap1+=inum;
*ap1=e_coul;
ap1+=inum;
}
if (vflag>0) {
for (int i=0; i<6; i++) {
*ap1=virial[i];
ap1+=inum;
}
}
ans[ii]=f;
} // if ii*/
}
Event Timeline
Log In to Comment