Closed brryan closed 1 year ago
OK I fixed the exp/log
inefficiency -- good point noticing that. I think that inefficiency goes back almost 15 years lol.
This passes the thincooling.py
test so I assume everything still works. Merging nowonce checks pass.
PR Summary
We were calling
LinearInterpLog
inside thewhile ()
of ado {} while ()
loop in the Monte Carlo sampling, and for some reason on device (power9 for me) this generates a segfault when thedNdlnu
ParArrayND<Real>
is passed to that inline function, either as a copy or a const reference. Don't know why that is, but the simple fix is to write out the interpolation routine in the actualpar_for
.This runs the
thincooling.pin
problem to completion though I didn't check the output for accuracy.PR Checklist
scripts/bash/format.sh
.