Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.
And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.
Question: maximum ndim discrepancy between CuPy and NumPy?
See original GitHub issueI noticed in CuPy we set MAX_NDIM = 25:
https://github.com/cupy/cupy/blob/46e5ff0bdf401ad88f404cd7cb7d26dd62b061c0/cupy/core/_carray.pxd#L10
whereas in NumPy this is set to 32 more than a decade ago:
https://github.com/numpy/numpy/blob/9563a3a63fd3dfb3b687fbabad134b5ded72bd46/numpy/core/include/numpy/ndarraytypes.h#L40
Is there a particular reason to pick 25? I was experimenting some iterator stuff and realized I need to know the largest possible ndim at build time. I can work around things so this is not critical at all, just simply curious.
Issue Analytics
- State:
- Created 3 years ago
- Comments:7 (6 by maintainers)
Top Results From Across the Web
Top Related Medium Post
Top Related StackOverflow Question
Troubleshoot Live Code
Top Related Reddit Thread
Top Related Hackernoon Post
Top Related Tweet
Top Related Dev.to Post
Top Related Hashnode Post
Yeah, if I remember correctly. That just meant there was a well-used tensor library that chose the number without any problem for years.
I talked with @okuta regarding this. It seems there’s no strong reason why 25 is picked. I think changing it to 32 for NumPy compatibility is better (but must verify that there’s no huge performance degradation).
If I recall correctly there was a MAX_NDIM validation in the past versions of CuPy but seems it’s lost at some point. It should be added too.
update: @beam2d told me that it was picked after cutorch: https://github.com/torch/cutorch/blob/5e9d86cb982a6048d3077aeb0e0cee19847b4c08/lib/THC/THCTensorInfo.cuh#L10