[PATCH] D29048: [RegisterCoalescer] Do not call LiveIntervals::getInstructionIndex with a DBG_VALUE

[Matthias Braun via Phabricator via llvm-commits]

MatzeB added a comment.

In https://reviews.llvm.org/D29048#654084, @bcahoon wrote:

> In https://reviews.llvm.org/D29048#653998, @MatzeB wrote:
>
> > At a first glance this looks like it fixes symptoms but not the root cause. It would probably be good if we could get a public reproducer.
>
>
> I agree with the symptom vs. root cause comment. I also am not sure if I was doing something else that might be problematic.  Especially after I was unable to generate a public test case that reproduced the problem.  I tried a bunch of different things with ARM since that target has a composable SubRegs without any luck.


Sometimes you can take an internal/private benchmark and reduce it with bugpoint to reduce it something simple that doesn't resemble the original code anymore but still fails.

As for other targets: hexagon and amdgpu are the only public targets with subregister liveness enabled which I suspect is necessary to trigger this.

I can take a look at this later whether I can spot something obvious, but a reproducer would obviously help...


https://reviews.llvm.org/D29048