portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-14 01:35:36 +00:00
Code Releases Activity

Remove SVM use

Browse Source
This commit is contained in:
Peter Boyle 2024-07-25 18:05:40 +00:00
parent 04cf902791
commit 6f1328160c
1 changed files with 1 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
Grid/lattice/Lattice_reduction.h
View File

@ -264,24 +264,8 @@ inline ComplexD rankInnerProduct(const Lattice<vobj> &left,const Lattice<vobj> &
const uint64_t sites = grid->oSites(); const uint64_t sites = grid->oSites();
// Might make all code paths go this way. // Might make all code paths go this way.
#if 0
typedef decltype(innerProductD(vobj(),vobj())) inner_t;
Vector<inner_t> inner_tmp(sites);
auto inner_tmp_v = &inner_tmp[0];
{
autoView( left_v , left, AcceleratorRead);
autoView( right_v,right, AcceleratorRead);
// This code could read coalesce
// GPU - SIMT lane compliance...
accelerator_for( ss, sites, nsimd,{
auto x_l = left_v(ss);
auto y_l = right_v(ss);
coalescedWrite(inner_tmp_v[ss],innerProductD(x_l,y_l));
});
}
#else
typedef decltype(innerProduct(vobj(),vobj())) inner_t; typedef decltype(innerProduct(vobj(),vobj())) inner_t;
Vector<inner_t> inner_tmp(sites); deviceVector<inner_t> inner_tmp(sites);
auto inner_tmp_v = &inner_tmp[0]; auto inner_tmp_v = &inner_tmp[0];
{ {
@ -295,7 +279,6 @@ inline ComplexD rankInnerProduct(const Lattice<vobj> &left,const Lattice<vobj> &
coalescedWrite(inner_tmp_v[ss],innerProduct(x_l,y_l)); coalescedWrite(inner_tmp_v[ss],innerProduct(x_l,y_l));
}); });
} }
#endif
// This is in single precision and fails some tests // This is in single precision and fails some tests
auto anrm = sumD(inner_tmp_v,sites); auto anrm = sumD(inner_tmp_v,sites);
nrm = anrm; nrm = anrm;