Update GPU configuration for graupel

[iomaganaris]

Set the gpu_maxnreg to a lower value. This allows increased occupancy of the GPU (tested on GH200). Even if there is excessive register spilling that's not a problem for graupel because I expect that the register spilling will only happen when the mask is enabled. When the mask is disabled (most of the time) we get the benefit of higher occupancy

[edopao]

Can you please measure (approximately) the speedup provided by this change on top of main? I would like to know if its benefits are visible without self-copy removal, to decide whether this change should go separately or together with self-copy removal.

[github-actions]

Mandatory Tests

Please make sure you run these tests via comment before you merge!

• cscs-ci run default
• cscs-ci run distributed

Optional Tests

To run benchmarks you can use:

• cscs-ci run benchmark-bencher

To run tests and benchmarks with the DaCe backend you can use:

• cscs-ci run dace

To run test levels ignored by the default test suite (mostly simple datatest for static fields computations) you can use:

• cscs-ci run extra

For more detailed information please look at CI in the EXCLAIM universe.

[iomaganaris]

R02B06 main

For 100 iterations it took 0.6509003639221191 seconds!
For 100 iterations it took 0.6452944278717041 seconds!
For 100 iterations it took 0.6502292156219482 seconds!
For 100 iterations it took 0.6481423377990723 seconds!

R02B08 main

For 100 iterations it took 8.613034963607788 seconds!
For 100 iterations it took 8.910228967666626 seconds!
For 100 iterations it took 8.931719303131104 seconds!
For 100 iterations it took 8.839605808258057 seconds!

vs
R02B06 graupel_gpu_opt

For 100 iterations it took 0.6054491996765137 seconds!
For 100 iterations it took 0.6052908897399902 seconds!
For 100 iterations it took 0.5998942852020264 seconds!
For 100 iterations it took 0.5863747596740723 seconds!

R02B08 graupel_gpu_opt

For 100 iterations it took 8.290613174438477 seconds!
For 100 iterations it took 8.158649444580078 seconds!
For 100 iterations it took 8.326894760131836 seconds!
For 100 iterations it took 8.067954063415527 seconds!

That's an ~8% increase in performance on average for R02B06 and ~7% for R02B08

[iomaganaris]

cscs-ci run default

[iomaganaris]

cscs-ci run distributed

[edopao]

Thank you for the time measurement, very good results!