You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In ITensor v3 the QDense(IndexSet,QN) is slower than the v2 version because QN addition now needs to check that all of the indices in the IndexSet have the same QNs. That means that constructor is quite slow when constructing QDense with many blocks from the divergence (so it affects constructors like randomITensor(QN(0),i,j,k) when the indices have many blocks, and gets worse with the order of the tensor). We should be able to optimize that constructor by first checking that the indices all have the same QNs, and then call a specialized QN addition and comparison function that assumes the QNs are all the same.
The text was updated successfully, but these errors were encountered:
In ITensor v3 the
QDense(IndexSet,QN)
is slower than the v2 version because QN addition now needs to check that all of the indices in the IndexSet have the same QNs. That means that constructor is quite slow when constructing QDense with many blocks from the divergence (so it affects constructors likerandomITensor(QN(0),i,j,k)
when the indices have many blocks, and gets worse with the order of the tensor). We should be able to optimize that constructor by first checking that the indices all have the same QNs, and then call a specialized QN addition and comparison function that assumes the QNs are all the same.The text was updated successfully, but these errors were encountered: