portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-10 07:55:35 +00:00
Code Releases Activity

Reduce the loop over exterior for GPU to indirection table

Browse Source
This commit is contained in:
Peter Boyle 2022-06-01 14:29:25 -07:00
parent 34faa39f4f
commit e762c940c2
3 changed files with 19 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
Grid/qcd/action/fermion/WilsonCompressor.h
View File

@ -297,7 +297,7 @@ public:
void ZeroCountersi(void) { }
void Reporti(int calls) { }
std::vector<int> surface_list;
// Vector<int> surface_list;
WilsonStencil(GridBase *grid,
int npoints,
@ -307,10 +307,11 @@ public:
: CartesianStencil<vobj,cobj,Parameters> (grid,npoints,checkerboard,directions,distances,p)
{
ZeroCountersi();
surface_list.resize(0);
// surface_list.resize(0);
this->same_node.resize(npoints);
};
/*
void BuildSurfaceList(int Ls,int vol4){
// find same node for SHM
@ -331,7 +332,8 @@ public:
}
}
}
*/
template < class compressor>
void HaloExchangeOpt(const Lattice<vobj> &source,compressor &compress)
{

11
Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h
View File

@ -440,6 +440,17 @@ void WilsonKernels<Impl>::DhopDirKernel( StencilImpl &st, DoubledGaugeField &U,S
#define KERNEL_CALL(A) KERNEL_CALLNB(A); accelerator_barrier();
#define KERNEL_CALL_EXT(A) \
const uint64_t NN = Nsite*Ls; \
const uint64_t sz = st.surface_list.size(); \
auto ptr = &st.surface_list[0]; \
accelerator_forNB( ss, sz, Simd::Nsimd(), { \
int sF = ptr[ss]; \
int sU = ss/Ls; \
WilsonKernels<Impl>::A(st_v,U_v,buf,sF,sU,in_v,out_v); \
}); \
accelerator_barrier();
#define ASM_CALL(A) \
thread_for( ss, Nsite, { \
int sU = ss; \

4
Grid/stencil/Stencil.h
View File

@ -640,7 +640,9 @@ public:
}
}
if(local == 0) {
surface_list.push_back(site);
for(int s=0;s<Ls;s++){
surface_list.push_back(site*Ls+s);
}
}
}
}