issues
search
cp2k
/
dbcsr
DBCSR: Distributed Block Compressed Sparse Row matrix library
https://cp2k.github.io/dbcsr/
GNU General Public License v2.0
134
stars
45
forks
source link
ocl: discover device properties once and simplified setting up WG-size
#775
Closed
hfp
closed
3 months ago
hfp
commented
3 months ago
Store cl_device_id in c_dbcsr_acc_opencl_device_t and rely on it (instead of querying it repeatedly).
Make maximum size of work-group (WG) and preferred multiple available as part of device structure.
Discover subgroups. Avoid some compile-time decision (OpenCL 3.0).
Simplify handling WG-size (and SG-size).
Code cleanup (_DEBUG).