mirror of
https://github.com/paboyle/Grid.git
synced 2025-12-21 21:24:30 +00:00
Make view specify where and drive data motion - first cut.
This is a compile tiime option --enable-unified=yes/no
This commit is contained in:
@@ -41,8 +41,8 @@ NAMESPACE_BEGIN(Grid);
|
||||
template<class vobj>
|
||||
inline Lattice<vobj> transpose(const Lattice<vobj> &lhs){
|
||||
Lattice<vobj> ret(lhs.Grid());
|
||||
auto ret_v = ret.View();
|
||||
auto lhs_v = lhs.View();
|
||||
auto ret_v = ret.View(AcceleratorWrite);
|
||||
auto lhs_v = lhs.View(AcceleratorRead);
|
||||
accelerator_for(ss,lhs_v.size(),vobj::Nsimd(),{
|
||||
coalescedWrite(ret_v[ss], transpose(lhs_v(ss)));
|
||||
});
|
||||
@@ -56,8 +56,8 @@ template<int Index,class vobj>
|
||||
inline auto TransposeIndex(const Lattice<vobj> &lhs) -> Lattice<decltype(transposeIndex<Index>(vobj()))>
|
||||
{
|
||||
Lattice<decltype(transposeIndex<Index>(vobj()))> ret(lhs.Grid());
|
||||
auto ret_v = ret.View();
|
||||
auto lhs_v = lhs.View();
|
||||
auto ret_v = ret.View(AcceleratorWrite);
|
||||
auto lhs_v = lhs.View(AcceleratorRead);
|
||||
accelerator_for(ss,lhs_v.size(),vobj::Nsimd(),{
|
||||
coalescedWrite(ret_v[ss] , transposeIndex<Index>(lhs_v(ss)));
|
||||
});
|
||||
|
||||
Reference in New Issue
Block a user