Andreas Klöckner
Andreas Klöckner
Thanks for reporting these. Contributions to address these would be very welcome.
I'm not wildly opposed, but I feel that it's also easy to just hack a `1/0` into that code path when needed. And for automated determinism tests, there are probably...
The SSL certificate on https://gmplib.org/ is broken, breaking all the wheel builds. It looks like they're aware of it.
@isuruf Was there a big reason why we used `gmp` instead of `imath-32` in the wheel? I'm getting ready to drop gmp here because I'm tired of waiting for them...
So... it turns out that this is what has kept https://github.com/inducer/loopy/issues/576 from biting as hard as it deserved to. https://gist.github.com/inducer/9eb97cb5a6064baee63d0ef64233ee3b together with this patch to loopy illustrates the issue: ```diff...
> I considered bundling both changes into one PR but they are really orthogonal and this did not feel right. Sounds good. Thanks for making an effort to keep PRs...
Still reproducible on pocl 3.1 with LLVM 15.
That completes without error. (LLVM 15, 9d57e12d693b2f1fcdb6521101d2aebf920a6ddb)
I'd call it pretty usable. We're using it as a "daily driver" in a reasonably large research center. Maybe @matthiasdiener has some recent data sitting around?
12569d59cb548c15094d11b2a1cc1bd36ebe4895 removed the flattening in `IdentityMapper`.