rapidsai / ptxcompiler Goto Github PK

If you pass a tuple of something other than strings as the options to compile(), then:

puts null pointers into the list of options to pass to nvPTXCompilerCompile, and causes a segfault.

Numba shouldn't be a hard dependency of ptxcompiler - it has other uses besides just Numba. However, importing ptxcompiler unconditionally imports Numba though patch.py. This import should be conditional, and check for a supported version (i.e. 0.54 and above).

I'd like to request CUDA 11.6 conda packages when possible.

The missing ptxcompiler packages for CUDA 11.6 are the only blocker for me to use CUDA 11.6 with rapids-compose. I currently have a workaround of removing the ptxcompiler dependency from the cuDF conda environment.

Hi! I'd like to request that this package be updated to install with CUDA 11.7. Currently, this dependency requirement blocks cuDF from installing on my system. Would it be possible to increase this package's bound for __cuda?

Dears,

I've tried to install rapids using mamba in place of conda and got this error:

bash$ mamba create -n rapids-mamba -c rapidsai -c nvidia -c conda-forge rapids=22.04 python=3.9 cudatoolkit=11.2 dask-sql
Encountered problems while solving.
Problem: package rapids-22.04.00-cuda11_py39_ge08d166_149 requires ptxcompiler, but none of the providers can be installed

If I try to install ptxcompiler only using mamba I got this issue:

bash$ mamba install -c rapidsai -c nvidia -c conda-forge ptxcompiler
Problem: package ptxcompiler-0.2.0-py39h114ffae_0 has constraint __cuda >=11.0,<=11.5 conflicting with __cuda-11.7-0

I have installed cuda-11-2 only, so I'm surprised but my nvidia-smi show conda 11.7 in the header...

There is no issue if I use conda, so it may be a mamba issue.

But regarding defect 13 #13 maybe it is a version mismatch...

So I prefer to get your point before cross posting

Thanks in advance

A new release, 0.3.0 was tagged. Would be good to build new packages for it.