portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-18 15:57:05 +01:00
Code Releases Activity

parallel_for elimination -> thread_loop

Browse Source
This commit is contained in:
paboyle
2018-01-28 01:01:14 +00:00
parent 9472b02771
commit 70e276e1ab
21 changed files with 269 additions and 300 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
lib/qcd/action/fermion/FermionOperatorImpl.h
View File

@ -268,13 +268,13 @@ public:
GaugeLinkField tmp(mat.Grid());
tmp = Zero();
parallel_for(int sss=0;sss<tmp.Grid()->oSites();sss++){
thread_loop( (int sss=0;sss<tmp.Grid()->oSites();sss++),{
int sU=sss;
for(int s=0;s<Ls;s++){
int sF = s+Ls*sU;
tmp[sU] = tmp[sU]+ traceIndex<SpinIndex>(outerProduct(Btilde[sF],Atilde[sF])); // ordering here
}
}
});
PokeIndex<LorentzIndex>(mat,tmp,mu);
}
@ -591,10 +591,10 @@ public:
Uconj = where(coor==neglink,-Uconj,Uconj);
}
parallel_for(auto ss=U.begin();ss<U.end();ss++){
thread_loop( (auto ss=U.begin();ss<U.end();ss++),{
Uds[ss](0)(mu) = U[ss]();
Uds[ss](1)(mu) = Uconj[ss]();
}
});
U = adj(Cshift(U ,mu,-1)); // correct except for spanning the boundary
Uconj = adj(Cshift(Uconj,mu,-1));
@ -605,18 +605,18 @@ public:
}
parallel_for(auto ss=U.begin();ss<U.end();ss++){
thread_loop((auto ss=U.begin();ss<U.end();ss++),{
Uds[ss](0)(mu+4) = Utmp[ss]();
}
});
Utmp = Uconj;
if ( Params.twists[mu] ) {
Utmp = where(coor==0,U,Utmp);
}
parallel_for(auto ss=U.begin();ss<U.end();ss++){
Uds[ss](1)(mu+4) = Utmp[ss]();
}
thread_loop((auto ss=U.begin();ss<U.end();ss++),{
Uds[ss](1)(mu+4) = Utmp[ss]();
});
}
}
@ -627,9 +627,9 @@ public:
GaugeLinkField link(mat.Grid());
// use lorentz for flavour as hack.
auto tmp = TraceIndex<SpinIndex>(outerProduct(Btilde, A));
parallel_for(auto ss = tmp.begin(); ss < tmp.end(); ss++) {
thread_loop((auto ss = tmp.begin(); ss < tmp.end(); ss++), {
link[ss]() = tmp[ss](0, 0) + conjugate(tmp[ss](1, 1));
}
});
PokeIndex<LorentzIndex>(mat, link, mu);
return;
}
@ -640,13 +640,13 @@ public:
GaugeLinkField tmp(mat.Grid());
tmp = Zero();
parallel_for(int ss = 0; ss < tmp.Grid()->oSites(); ss++) {
thread_loop((int ss = 0; ss < tmp.Grid()->oSites(); ss++) ,{
for (int s = 0; s < Ls; s++) {
int sF = s + Ls * ss;
auto ttmp = traceIndex<SpinIndex>(outerProduct(Btilde[sF], Atilde[sF]));
tmp[ss]() = tmp[ss]() + ttmp(0, 0) + conjugate(ttmp(1, 1));
}
}
});
PokeIndex<LorentzIndex>(mat, tmp, mu);
return;
}