Closed inducer closed 2 years ago
@matthiasdiener This may be behind the NaNs that we were seeing, i.e. unrelated to the memory pool.
Yes, this PR does fix the NaNs we are seeing.
Force-merging since earlier tests passed and the latest changes only affected the (project-local) tests, which have passed.
@majosm Turns out #584 mishandled infinities, which are used frequently as neutral elements. @matthiasdiener This may be behind the NaNs that we were seeing, i.e. unrelated to the memory pool.