FLAMEGPU / FLAMEGPU2-circles-benchmark

Benchmarking of the Circles model in FLAME GPU 2

MIT License

1 stars 0 forks source link

Possible changes for benchmark #15

Open mondus opened 2 years ago

mondus commented 2 years ago

[ ] Drift guarding #14
[ ] Pop size doesn't give full utilisation on v100 (80 SMs, 2048 max threads per SM, but lower for message iteration kernels)
[x] FP64 use is massive there is a 3.14 someone rather than 3.14f
[x] Cube root is (maybe) in double. Is there a f version.
[ ] Broadcast access may be better as offset access. Really an issue for the main repo.

ptheywood commented 2 years ago

18 addresses:

the FP64 3.141 vs 3.141f issue (confirmed by profiling that the absence of the f literal does lead to FP64 pip utilisation)
cbrt was in fp64, but it should also be sqrtf anyway for correct behaviour. This impacts model behaviour rather than being pure perf.