adding a new episode for cupy #2
Conversation
qianglise
requested review from
ashwinvis,
code4yonglei,
ffrancesco94 and
jadgt
There was a problem hiding this comment.
I really liked it. The cupy vs numpy/scipy gotcha section is gold. I only pointed out a couple of minor typos and a doubt I had about one thing, but as soon as the broken examples are fixed (or the reason why they're broken is known) for me it's mergeable!
EDIT: regarding the stream thing, I think it's ok like this. It could be part of a bonus episode maybe. But as things look now, we either assume that people already know about the GPU execution model or they don't care and just want to gpu-ify their code, so they don't want to bother with that level of granularity.
ashwinvis
left a comment
ashwinvis
left a comment
There was a problem hiding this comment.
Few more comments, and another general one which applies to many places: try to look for the term cupy. and use backticks so that we recognize it is code and not part of the text.
| (see [CUDA Array Interface](https://numba.readthedocs.io/en/stable/cuda/cuda_array_interface.html) | ||
| for details). It means one can pass CuPy arrays to kernels JITed with Numba. | ||
|
|
||
| ################### FIXME: crashes when launching |
There was a problem hiding this comment.
it still crashes for me when I did the test on Tetralith.
I have to make a test on a different cluster
fixed in f22c022 |
|
@ashwinvis @ffrancesco94 all the comments are implemented except one which is the crashing of the numba. I need more time on that. For now I would say I will merge the PR, and later remove the FIXME warning after testing on a different machine. |
|
Thanks @qianglise, I will merge as it is. |
adding cupy episode related to ENCCS/hpda-python#111