[PATCH] D65280: Add a pass to lower is.constant and objectsize intrinsics

Mon Aug 5 17:53:58 PDT 2019

chandlerc added a comment.

In D65280#1614752 <https://reviews.llvm.org/D65280#1614752>, @joerg wrote:

> Looking a bit more into the details. Chandler, you've originally suggested going with the LowerAtomic route and that actually does create code that fails the SDAG lowering if the pass is skipped, e.g. on ARM.

Sorry, I had forgotten about that discussion. Thank you for digging it up and reminding me to be consistent with my past self.

If this pass is going into the `llc` pipeline like the lower atomic pass, you're right it totally makes sense to fail hard in SDAG lowering. Sorry for the confusion again.

> The second part is currently overlapping with the CodeGenPrepare pass. I can cleanup the implementation somewhat by reusing the same functionality as that pass is using OR I could factor out a minimal branch for doing the constant folding optimization from CodeGenPrepare as a general branch that is included in the pass chain for `-O0`, e.g. instead of the more general CodeGenPrepare pass. The main difference is that the non-optimized build would not get any recursive folding from PHI-simplification, but I think that's fine for the original use case. It would also not get the block merging, but again, that seems to be fine for the constraints.

I'm not really sure I understand the problem?

This pass needs to run (at least) in the same general area as the atomics lowering for the same reasons (it would be part of SDAG but it can't be due to CFG mutation).  It might optionally be run in other places of course.

I think you have all the code you need in this patch already. I'm just suggesting a different strategy for applying it to the IR that I think will be simpler (and not require any changes to instsimplify). Maybe my suggestion isn't clear? If so I can try to clarify....

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D65280/new/

https://reviews.llvm.org/D65280