TY  - CONF
AU  - Baumann, Thomas
AU  - Speck, Robert
TI  - Porting mpi4py-fft to GPU
M1  - FZJ-2024-04850
PY  - 2024
AB  - The mpi4py-fft library enables distributed fast Fourier transforms on CPUs with an easy to use interface and scales very well. We attempt to port this to GPUs, which significantly outperform the CPU counterpart at a given node count. While the porting is straightforward for the most part, the best communication strategy is still an open question for us.The algorithm relies on MPI alltoallw. Even with CUDA-aware MPI, this exhibits very poor performance on the Juelich computers. By replacing it with a custom communication strategy, throughput can be increased at a slight loss of generality. We would like to discuss optimising the strategy, or even if the performance of alltoallw can be increased by some measure.
T2  - 16th JLESC Workshop
CY  - 16 Apr 2024 - 18 Apr 2024, Kobe (Japan)
Y2  - 16 Apr 2024 - 18 Apr 2024
M2  - Kobe, Japan
LB  - PUB:(DE-HGF)6
DO  - DOI:10.34734/FZJ-2024-04850
UR  - https://juser.fz-juelich.de/record/1028864
ER  -