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.
Recommended way to develop a cupy wrapper for an original cuda kernel
See original GitHub issueI’m a NN developer who want to transport a pytorch Module (https://github.com/jonas-koehler/s2cnn) to chainer.
The original Module uses a homemade kernel, wrapped by pynvrtc and cupy.cuda.function.Module.get_function().
In searching information for the transportation, I found a PR which refer not to use pynvrtc in cupy. https://github.com/cupy/cupy/pull/33#issuecomment-301306224
My question is which is the better to use pynvrtc or cupy.cuda.compiler to create an original kernel used for chainer.
Issue Analytics
- State:
- Created 5 years ago
- Comments:6 (3 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
Currently these low-level APIs are undocumented as it is intended for private use in CuPy. However it is a simple API that you only need to pass the CUDA code. You can find some examples in: https://github.com/cupy/cupy/blob/v5.0.0a1/cupy/core/core.pyx#L4339-L4356 If you have multiple GPUs with different compute capability, make sure to run compilation for each device.
Thank you very much! That was really I hoped!