have reaction loop print out failures even when on GPU
zingale opened this issue · comments
Currently, when running on the GPU, the different integrators do not print out the failures. But we now have standardized error codes, which we can add to the burn_t. Then we can store these in Castro to output once we leave the GPU kernel.
Some of this is discussed here:
In an ideal world, we'd setup a vector<burn_t> in Castro outside of the loop over the MF and then push back the burn_t of any zone that fails. Then outside of the GPU kernel, we'd loop over the vector and print out the failures.
This can really help with debugging.
Is there a way to do this easily with GPUs?
If you add the error code as a burn_t member, there's a couple ideas I have here for how to do some version of this.
the error codes are added to the burn_t in this PR:
AMReX-Astro/Microphysics#1254