Closed bentsherman closed 7 years ago
I've run a few tests over the past few days, and I was able to fix a few memory errors. The only errors I'm seeing out of valgrind now are strange errors (not leaks) from LAPACKE and a few other nit-picky things from valgrind that I think we can disregard. If someone wants to explore these errors further, they are certainly welcome, but for now I think we can put this issue to rest.
I haven't done this in a while, so we should probably look for memory leaks. I've had issues in the past with
valgrind
because if I ran with--leak-check=full
,valgrind
would report literally millions of errors and quit. These errors seem to be coming from within BLAS/LAPACK. I think the issue comes from the fact that cblas and LAPACKE actually provide two interfaces: the "middle-level" interface, which is identical to the Fortran API, and the "high-level" interface, which abstracts some of the workspace variables. Maybevalgrind
is reporting uninitialized value errors or something like that. Or maybe BLAS and LAPACK are just hostile towardsvalgrind
. I don't know. But I was able to run without that option and still catch some memory errors. Anyway, another fun project for someone to do.