ocl: revised creating sub-devices - Githubissues

cp2k / dbcsr

DBCSR: Distributed Block Compressed Sparse Row matrix library

https://cp2k.github.io/dbcsr/

GNU General Public License v2.0

135 stars 47 forks source link

ocl: revised creating sub-devices #831

Closed hfp closed 4 months ago

hfp commented 4 months ago

Handle ACC_OPENCL_DEVSPLIT=1 like ZE_FLAT_DEVICE_HIERARCHY=COMPOSITE.
Partition devices into number of local ranks by default (Intel MPI).
Adjusted device-split' default according to MPI rank-count.
Auto-split devices and take rank-count into account.
Correctly handle ACC_OPENCL_DEVSPLIT<0.