jianmu
/
lammps
forked from lijiext/lammps
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Reduce unnecessary data exchanges

This commit is contained in:
Vsevak 2020-01-19 02:05:16 +03:00
parent 868df1f640
commit f69a17dd3a
2 changed files with 20 additions and 24 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

37
lib/gpu/lal_lj_tip4p_long.cpp
View File

@ -198,8 +198,7 @@ void LJTIP4PLongT::loop(const bool _eflag, const bool _vflag) {
&nall, &ainum,
&nbor_pitch, &this->_threads_per_atom,
&hneight, &m, &TypeO, &TypeH, &alpha,
&this->atom->q, &tag, &map_array,
&atom_sametag);
&this->atom->q);
GX=static_cast<int>(ceil(static_cast<double>(this->ans->inum())/
(BX/this->_threads_per_atom)));
@ -217,8 +216,7 @@ void LJTIP4PLongT::loop(const bool _eflag, const bool _vflag) {
&ainum, &nbor_pitch, &this->_threads_per_atom,
&hneight, &m, &TypeO, &TypeH, &alpha,
&this->atom->q, &cutsq, &_qqrd2e, &_g_ewald,
&cut_coulsq, &cut_coulsqplus, &tag, &map_array,
&atom_sametag, &this->ansO);
&cut_coulsq, &cut_coulsqplus, &this->ansO);
GX=static_cast<int>(ceil(static_cast<double>(this->ans->inum())/BX));
this->k_pair_distrib.set_size(GX,BX);
this->k_pair_distrib.run(&this->atom->x, &this->ans->force, &this->ans->engv,
@ -234,25 +232,26 @@ void LJTIP4PLongT::copy_relations_data(int n, tagint *tag, int *map_array,
int nall = n;
const int hn_sz = n*4; // matrix size = col size * col number
hneight.resize_ib(hn_sz);
if (ago == 0)
hneight.zero();
m.resize_ib(n);
m.zero();
if (ago == 0) {
hneight.zero();
UCL_H_Vec<int> host_tag_write(nall,*(this->ucl_device),UCL_WRITE_ONLY);
this->tag.resize_ib(nall);
for(int i=0; i<nall; ++i) host_tag_write[i] = tag[i];
ucl_copy(this->tag, host_tag_write, nall, false);
UCL_H_Vec<int> host_tag_write(nall,*(this->ucl_device),UCL_WRITE_ONLY);
this->tag.resize_ib(nall);
for(int i=0; i<nall; ++i) host_tag_write[i] = tag[i];
ucl_copy(this->tag, host_tag_write, nall, false);
host_tag_write.resize_ib(max_same);
this->atom_sametag.resize_ib(max_same);
for(int i=0; i<max_same; ++i) host_tag_write[i] = sametag[i];
ucl_copy(this->atom_sametag, host_tag_write, max_same, false);
host_tag_write.resize_ib(max_same);
this->atom_sametag.resize_ib(max_same);
for(int i=0; i<max_same; ++i) host_tag_write[i] = sametag[i];
ucl_copy(this->atom_sametag, host_tag_write, max_same, false);
host_tag_write.resize_ib(map_size);
this->map_array.resize_ib(map_size);
for(int i=0; i<map_size; ++i) host_tag_write[i] = map_array[i];
ucl_copy(this->map_array, host_tag_write, map_size, false);
host_tag_write.resize_ib(map_size);
this->map_array.resize_ib(map_size);
for(int i=0; i<map_size; ++i) host_tag_write[i] = map_array[i];
ucl_copy(this->map_array, host_tag_write, map_size, false);
}
}

7
lib/gpu/lal_lj_tip4p_long.cu
View File

@ -226,9 +226,7 @@ __kernel void k_lj_tip4p_newsite(const __global numtyp4 *restrict x_,
__global int *restrict hneigh,
__global numtyp4 *restrict m,
const int typeO, const int typeH,
const numtyp alpha, const __global numtyp *restrict q_,
const __global int *restrict tag, const __global int *restrict map,
const __global int *restrict sametag) {
const numtyp alpha, const __global numtyp *restrict q_) {
int tid, ii, offset;
atom_info(t_per_atom,ii,tid,offset);
int i = BLOCK_ID_X*(BLOCK_SIZE_X)+tid;
@ -268,8 +266,7 @@ __kernel void k_lj_tip4p_long(const __global numtyp4 *restrict x_,
const __global numtyp *restrict cutsq,
const numtyp qqrd2e, const numtyp g_ewald,
const numtyp cut_coulsq, const numtyp cut_coulsqplus,
const __global int *restrict tag, const __global int *restrict map,
const __global int *restrict sametag, __global acctyp4 *restrict ansO) {
__global acctyp4 *restrict ansO) {
int tid, ii, offset;
atom_info(t_per_atom,ii,tid,offset);