Andreas Klöckner
Andreas Klöckner
So the fact that the backtrace contains `clReleaseContext` points to the notion that the Nvidia runtime has some bug that makes it not like decreasing the context refcount (perhaps: doing...
Thank you for offering up your efforts. Do you remember the steps that you took, and would you be willing to package those up in a CI job? You could...
Agreed, I've been thinking about how to best support such a thing. I'd be happy to discuss ideas for a patch.
(or take a patch, for that matter.)
Not really. What @rutsky describes above is still the "state of the art". I would welcome patches to make this easier.
Could you explain how `c_vec` differs semantically from `vec`? Off the cuff, rather than introduce a new tag, I think I would prefer realizing the existing `vec` tag the way...
OIC, thanks for those links. In that case, I'm not sure how `cvec` (as an array axis tag) is different from a normal (strided) array axis. And I don't really...
Thanks @tj-sun for your help! I'm still of the mind to address the problems you encountered (and allow a unified "vocabulary" across targets) rather than introduce a target-specific widget, at...
We're seeing something pretty similar on Linux: https://gitlab.tiker.net/inducer/pycuda/-/jobs/38205 I don't have a lot of time to debug this right now (tenure track...), but I'd appreciate patches. I'll disable that test...
Let me know if the skipping helped for you. It didn't for me: https://gitlab.tiker.net/inducer/pycuda/pipelines/9559