Closed rosejn closed 1 year ago
I would, except I need a version with gl enabled and it seems for some reason that isn’t the case with the release. I had to compile it myself, but it would be great if you made a version on pypi. (We’re copying data from torch tensors output from models directly into GL textures.)
Thanks!
On Thu, Jul 6, 2023 at 12:56 PM Andreas Klöckner @.***> wrote:
- A few releases were missing from Github. This has been fixed. Thanks for pointing this out.
- Please don't use the zip downloads from Github. Instead, please always download releases from https://pypi.org/project/pycuda/. The tarballs are not equivalent, and the Github ones are broken in a way that I can't fix.
— Reply to this email directly, view it on GitHub https://github.com/inducer/pycuda/issues/418#issuecomment-1624168190, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAAI53V55XV45ISGMVJUXZ3XO4C6TANCNFSM6AAAAAA2A2MV74 . You are receiving this because you authored the thread.Message ID: @.***>
Sorry, I'm not totally sure what you mean by "GL enabled" here. The code in the Github tarball and the code on the package index should be the same, so they should not differ in their GL-enabled-ness.
oh, I mean compiling it with the --cuda-enable-gl flag. Without that it won't let me copy data from a torch tensor into a GL texture buffer. Would it be possible to add that to the standard compilation used when installed via pypi?
Thanks for the suggestion. Unfortunately, I think that would cause all sorts of compatibility issues. That said, if you can figure out how to get flags through pip
to setup.py
, I'd be happy to take that PR.
I just downloaded the latest release zip (pycuda-2022.2.zip) from github, and it seems to be an old version that doesn't include the latest updates to support cuda 12 (no COMPUTE_* beyond 21 I think). Maybe there's a bug in the release script or something. I can build the latest by cloning the repo and that succeeds though. Thanks for the library!