search

ComputationalRadiationPhysics / picongpu

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

Fix juwels booster profile to make use of all cores in a node. #5192

Closed steindev closed 3 weeks ago

steindev commented 4 weeks ago

For example, compiling on nodes took ages because only a single core was utilized. Fixed this in getDevice, getNode and the booster.tpl.

PrometheusPi commented 3 weeks ago

CI ran into a timeout. I retriggered it.

PrometheusPi commented 3 weeks ago

The CI failed again. @psychocoderHPC Might this be a real issue?

psychocoderHPC commented 3 weeks ago

The code must be formated correctly!

--- a/etc/picongpu/juwels-jsc/booster.tpl
+++ b/etc/picongpu/juwels-jsc/booster.tpl
@@ -61,7 +61,7 @@ export UCX_RC_TIMEOUT=3000000.00us # 3s instead of 1s
 .TBG_devicesPerNode=$(if [ $TBG_tasks -gt $TBG_numHostedDevicesPerNode ] ; then echo $TBG_numHostedDevicesPerNode; else echo $TBG_tasks; fi)

 # Cores per task. Theoretically we have 48 cores, we might leave one per task for the OS but then we would need to
-# hope that srun will do the pinning of cores to memory correctly in order to performantly read from memory. 
+# hope that srun will do the pinning of cores to memory correctly in order to performantly read from memory.
 .TBG_coresPerTask=12

There is a whitespace at the end of the line which must be removed.

steindev commented 3 weeks ago

@PrometheusPi ready to merge.