Closed zingale closed 7 months ago
run this together with this Microphysics PR: https://github.com/AMReX-Astro/Microphysics/pull/1441
these hardcode the parameters for the subchandra problem, which is one of the problems that fails with later versions of ROCm
The way this works is that we pass --constexpr
to the python scripts that write the parameters headers. This changes them to declare them as constexpr
instead of (extern) managed. The values are no hardcoded in, and there are no parmparse queries for them.
PR summary
This is an experiment to see if keeping the runtime parameters as constexpr instead of managed helps with later versions of ROCm. Note: the values are compiled in, so they will need to be hacked to override the defaults for a particular problem.
PR motivation
PR checklist
CHANGES
file has been updated, if appropriate