ICODE-ADC/lammps
4
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Feb2021 GPU Package Update - GPU Package Files

Browse Source
This commit is contained in:
Michael Brown
2021-02-15 08:20:50 -08:00
parent 16004e8f45
commit e7e2d2323b
345 changed files with 13424 additions and 7708 deletions
Download Patch File Download Diff File Expand all files Collapse all files

44
lib/gpu/lal_colloid.cu
View File

@ -42,22 +42,25 @@ __kernel void k_colloid(const __global numtyp4 *restrict x_,
atom_info(t_per_atom,ii,tid,offset);
__local numtyp sp_lj[4];
int n_stride;
local_allocate_store_pair();
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;
acctyp energy, virial[6];
if (EVFLAG) {
energy=(acctyp)0;
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);
@ -146,7 +149,7 @@ __kernel void k_colloid(const __global numtyp4 *restrict x_,
f.y+=dely*force;
f.z+=delz*force;
if (eflag>0) {
if (EVFLAG && eflag) {
numtyp e=(numtyp)0.0;
if (form[mtype]==0) {
e=r6inv*(lj3[mtype].x*r6inv-lj3[mtype].y);
@ -160,7 +163,7 @@ __kernel void k_colloid(const __global numtyp4 *restrict x_,
}
energy+=factor_lj*(e-lj3[mtype].z);
}
if (vflag>0) {
if (EVFLAG && vflag) {
virial[0] += delx*delx*force;
virial[1] += dely*dely*force;
virial[2] += delz*delz*force;
@ -171,9 +174,9 @@ __kernel void k_colloid(const __global numtyp4 *restrict x_,
}
} // for nbor
store_answers(f,energy,virial,ii,inum,tid,t_per_atom,offset,eflag,vflag,
ans,engv);
} // if ii
store_answers(f,energy,virial,ii,inum,tid,t_per_atom,offset,eflag,vflag,
ans,engv);
}
__kernel void k_colloid_fast(const __global numtyp4 *restrict x_,
@ -198,6 +201,9 @@ __kernel void k_colloid_fast(const __global numtyp4 *restrict x_,
__local numtyp4 colloid2[MAX_SHARED_TYPES*MAX_SHARED_TYPES];
__local int form[MAX_SHARED_TYPES*MAX_SHARED_TYPES];
__local numtyp sp_lj[4];
int n_stride;
local_allocate_store_pair();
if (tid<4)
sp_lj[tid]=sp_lj_in[tid];
if (tid<MAX_SHARED_TYPES*MAX_SHARED_TYPES) {
@ -205,23 +211,23 @@ __kernel void k_colloid_fast(const __global numtyp4 *restrict x_,
colloid1[tid]=colloid1_in[tid];
colloid2[tid]=colloid2_in[tid];
form[tid]=form_in[tid];
if (eflag>0)
if (EVFLAG && eflag)
lj3[tid]=lj3_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;
acctyp energy, virial[6];
if (EVFLAG) {
energy=(acctyp)0;
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);
@ -310,7 +316,7 @@ __kernel void k_colloid_fast(const __global numtyp4 *restrict x_,
f.y+=dely*force;
f.z+=delz*force;
if (eflag>0) {
if (EVFLAG && eflag) {
numtyp e=(numtyp)0.0;
if (form[mtype]==0) {
e=r6inv*(lj3[mtype].x*r6inv-lj3[mtype].y);
@ -325,7 +331,7 @@ __kernel void k_colloid_fast(const __global numtyp4 *restrict x_,
}
energy+=factor_lj*(e-lj3[mtype].z);
}
if (vflag>0) {
if (EVFLAG && vflag) {
virial[0] += delx*delx*force;
virial[1] += dely*dely*force;
virial[2] += delz*delz*force;
@ -336,8 +342,8 @@ __kernel void k_colloid_fast(const __global numtyp4 *restrict x_,
}
} // for nbor
store_answers(f,energy,virial,ii,inum,tid,t_per_atom,offset,eflag,vflag,
ans,engv);
} // if ii
store_answers(f,energy,virial,ii,inum,tid,t_per_atom,offset,eflag,vflag,
ans,engv);
}