search

ComputationalRadiationPhysics / picongpu

Performance-Portable Particle-in-Cell Simulations for the Exascale Era :sparkles:
https://picongpu.readthedocs.io
Other
692 stars 218 forks source link

openPMD-api writes wrong blocks in meshes - indexing bug #4641

Closed PrometheusPi closed 1 year ago

PrometheusPi commented 1 year ago

As discussed in https://github.com/openPMD/openPMD-api/issues/1485 PIConGPU seems to write incorrect blocks in openPMD causing shift in data and undefined regions.

Please see the linked issue for details.

PrometheusPi commented 1 year ago

I have revised ticket 21796 at HZDR. Most likely the "empty blocks" were caused by the same problem. EDIT: While I see the same bug at iteration 0, later iteration are correct. Both problems seem to be independent.

PrometheusPi commented 1 year ago

The misalignment of written blocks only occurs at the 0th iteration. All other iterations cover the entire cell volume. Tested with:

for i in `seq 0 1000 10000`; do export FN=`printf simOutput_%06d.bp $i`; bpls -D $FN | grep fields/e_all_chargeDensity -A 33; done > test.txt

test.txt

@psychocoderHPC might this be caused by -m --windowMovePoint 0.0 ?

psychocoderHPC commented 1 year ago

Could you please provide you file tbg/submit.start I am wondering about your setup because I can not reproduce the isse.

psychocoderHPC commented 1 year ago

The error is not reproducible on our development system but on hemera. The issue is coming from an unset variable and therefore is random. Fix is on the way.

bug introduced with #4585

PrometheusPi commented 1 year ago

This was fixed with #4644 - I will close the issue