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

Speed up Cshift more with coalesced

Browse Source
This commit is contained in:
Peter Boyle 2020-09-01 15:28:15 -04:00
parent c273fb051c
commit 5791021dcd
2 changed files with 8 additions and 23 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
Grid/cshift/Cshift.h
View File

@ -52,23 +52,8 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
NAMESPACE_BEGIN(Grid);
template<typename Op, typename T1>
auto Cshift(const LatticeUnaryExpression<Op,T1> &expr,int dim,int shift)
-> Lattice<decltype(expr.op.func(eval(0, expr.arg1)))>
{
return Cshift(closure(expr),dim,shift);
}
template <class Op, class T1, class T2>
auto Cshift(const LatticeBinaryExpression<Op,T1,T2> &expr,int dim,int shift)
-> Lattice<decltype(expr.op.func(eval(0, expr.arg1),eval(0, expr.arg2)))>
{
return Cshift(closure(expr),dim,shift);
}
template <class Op, class T1, class T2, class T3>
auto Cshift(const LatticeTrinaryExpression<Op,T1,T2,T3> &expr,int dim,int shift)
-> Lattice<decltype(expr.op.func(eval(0, expr.arg1),
eval(0, expr.arg2),
eval(0, expr.arg3)))>
template<class Expression,typename std::enable_if<is_lattice_expr<Expression>::value,void>::type * = nullptr>
auto Cshift(const Expression &expr,int dim,int shift) -> decltype(closure(expr))
{
return Cshift(closure(expr),dim,shift);
}

12
Grid/cshift/Cshift_common.h
View File

@ -76,8 +76,8 @@ Gather_plane_simple (const Lattice<vobj> &rhs,commVector<vobj> &buffer,int dimen
autoView(rhs_v , rhs, AcceleratorRead);
auto buffer_p = & buffer[0];
auto table = &Cshift_table[0];
accelerator_for(i,ent,1,{
buffer_p[table[i].first]=rhs_v[table[i].second];
accelerator_for(i,ent,vobj::Nsimd(),{
coalescedWrite(buffer_p[table[i].first],coalescedRead(rhs_v[table[i].second]));
});
}
}
@ -185,8 +185,8 @@ template<class vobj> void Scatter_plane_simple (Lattice<vobj> &rhs,commVector<vo
autoView( rhs_v, rhs, AcceleratorWrite);
auto buffer_p = & buffer[0];
auto table = &Cshift_table[0];
accelerator_for(i,ent,1,{
rhs_v[table[i].first]=buffer_p[table[i].second];
accelerator_for(i,ent,vobj::Nsimd(),{
coalescedWrite(rhs_v[table[i].first],coalescedRead(buffer_p[table[i].second]));
});
}
}
@ -282,8 +282,8 @@ template<class vobj> void Copy_plane(Lattice<vobj>& lhs,const Lattice<vobj> &rhs
autoView(rhs_v , rhs, AcceleratorRead);
autoView(lhs_v , lhs, AcceleratorWrite);
auto table = &Cshift_table[0];
accelerator_for(i,ent,1,{
lhs_v[table[i].first]=rhs_v[table[i].second];
accelerator_for(i,ent,vobj::Nsimd(),{
coalescedWrite(lhs_v[table[i].first],coalescedRead(rhs_v[table[i].second]));
});
}
}