Benchmark against other frameworks with Python/numpy frontend

[learning-chip]

Hi, this is very cool work.

This repository https://github.com/dionhaefner/pyhpc-benchmarks compares several Python-HPC frameworks such as JAX, Numba, CuPy, PyTorch, etc. Would be useful to add DaCe to the comparison.

Related: dionhaefner/pyhpc-benchmarks#6

[tbennun]

Thanks! This looks very interesting and we would be happy to help. Is there a particular problem with implementing the benchmark?

[learning-chip]

Is there a particular problem with implementing the benchmark?

Thanks, @dionhaefner would be the expert to talk with😃

I am currently busy with other stuff and will try implementing some tests myself when time allows.

[dionhaefner]

I tried to apply @dace.program to the EOS benchmark, but got some errors that multiplications between floats are not implemented.

If you can get DaCe to support this, or supply me with a working implementation of the benchmark, I am happy to crunch the numbers on our hardware and add it to the repo.

[tbennun]

Interesting, thanks! I tried EOS with the latest master branch yesterday and it compiled successfully (I didn't run anything because of the __init__.py setup). I added import dace and decorated gsw_dHdT with @dace.program(auto_optimize=True).

This could be because of the released version being outdated. Since we'd like to release the latest version (which has improved Python support), it'd be great if we first make sure all three benchmarks run beforehand. What are the changes required to add an implementation to pyhpc-benchmarks?

[dionhaefner]

I've added instructions on how to add a new backend to pyhpc-benchmarks. Don't worry, it's really simple :)

Looking forward to your PR!

[tbennun]

That's great, thanks!