[PATCH] D29105: Fix regalloc assignment of overlapping registers

Stanislav Mekhanoshin via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Tue Jan 31 16:52:15 PST 2017

rampitec added a comment.

In https://reviews.llvm.org/D29105#662438, @MatzeB wrote:

> In https://reviews.llvm.org/D29105#661368, @alex-t wrote:
> > Given that AMDGPU backend is the only one tracking lane liveness, I'd like this change to be accepted now.
> > LGTM
> Hexagon is using subreg liveness as well and there is at least 1 more out-of-tree target using it.
> This patch feels incomplete, it should just be a question of time until you run into the `report_fatal_error()` condition, but it's better than the status quo, so go ahead and push it there is not better solution.

I agree, it is incomplete but better than what we have now. At least we will see this fatal error. It took me 2 months to catch this error in a huge program without any diagnostics from the compiler. Next time we will see it right away.



More information about the llvm-commits mailing list