Carl Pearson
Carl Pearson
-- edit --- I made an error, please see https://github.com/kokkos/kokkos/pull/8169#issuecomment-2963782804 below -- original comment -- We could just disable perf tests...Our unit tests build and run okay, it has something...
A large number of our unit tests also trigger this error under CUDA 12.9, so it's not just a matter of disabling the perf tests.
Sorry for my mistaken comment, we actually have this same issue compiling many of our unit tests under CUDA 12.9
@Yurlungur thanks! You will need to sign off on each commit, and also run clang-format 16 (or you can wait for the format check to fail and look in the...
I can't re-trigger the SNL CI checks, possibly rebasing this would fix it?
I'm happy to defer to the framework team either way, however this might only be worth doing if 3.28 provides some fixes or quality-of-life improvements that are useful to the...
In my experience more than half the battle of benchmarking a pure function is convincing the compiler not to optimize your code away. Not to say it's impossible, but it...
Thoughts on something like this? I was experimenting with it. `do_not_optimize` is compiler-specific, but it might not be too hard to get it right for a few platforms we care...
eh, there seems to be some additional optimization between PTX and SASS that deletes the unused values despite the `asm` statements. I don't know how to defeat that.
> @cwpearson Can you also run the benchmark with `SYCL` on `blake`? Updated in description.