AMReX-Astro / Microphysics

common astrophysical microphysics routines with interfaces for the different AMReX codes
https://amrex-astro.github.io/Microphysics
Other
35 stars 34 forks source link

time and temp_start not properly output after burn failure on Frontier GPU #1630

Closed brady-ryan closed 2 months ago

brady-ryan commented 2 months ago

When running subch_planar with GPUs on Frontier, the time and temp_start of a failed burn state does not seem to be properly stored. Here is an example output:

[ERROR] integration failed in net, istate = -6, time = 0, dt = 7.54401e-07, dens_start = 1.87438e+06, temp_start = 0

yut23 commented 2 months ago

I'm not sure about the time, but state_save.T_in is indeed only set on CPU: https://github.com/AMReX-Astro/Microphysics/blob/ecbc595ec086227d1120b97efc51d92aa6e14dc4/integration/integrator_setup_sdc.H#L115-L120 https://github.com/AMReX-Astro/Microphysics/blob/ecbc595ec086227d1120b97efc51d92aa6e14dc4/integration/integrator_setup_strang.H#L105-L110

zingale commented 2 months ago

this is fixed by #1635 I don't think there's an issue with time -- if I run burn_cell_sdc on GPUs and force it to fail by giving it a hard burn, then I get the expected intermediary time.