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
While B*Matrix(B') is always a symmetric matrix, when computed with OpenBLAS, this is not anymore true with MKL. Is any explanation for this? This is unexpected for me, being simply a dot product computed probably in two different ways.
In my tests I relied on the fact that B*Matrix(B') is symmetric. For MKL compliance I needed to change to something like (B*Matrix(B')+B*Matrix(B'))/2` to ensure exact symmetry (not a big trouble, but still unexpected somehow).
Note that issymmetric(B*B') is always true for both OpenBLAS and MKL,
The text was updated successfully, but these errors were encountered:
Following the discussions here, I started to make my code in PeriodicSystems compliant with MKL. During tests, I came accross to the following issue:
While
B*Matrix(B')
is always a symmetric matrix, when computed with OpenBLAS, this is not anymore true with MKL. Is any explanation for this? This is unexpected for me, being simply a dot product computed probably in two different ways.In my tests I relied on the fact that
B*Matrix(B')
is symmetric. For MKL compliance I needed to change to something like(B*Matrix(B')+
B*Matrix(B'))/2` to ensure exact symmetry (not a big trouble, but still unexpected somehow).Note that
issymmetric(B*B')
is alwaystrue
for both OpenBLAS and MKL,The text was updated successfully, but these errors were encountered: