[PATCH] D115808: [DAGCombiner] Avoid combining adjacent stores at -O0 to improve debug experience

[Sanjay Patel via Phabricator via llvm-commits]

spatel reopened this revision.
spatel added a comment.
This revision is now accepted and ready to land.

In D115808#3204444 <https://reviews.llvm.org/D115808#3204444>, @xgupta wrote:

> Sorry, I read it fast that I didn't understand the meaning, I think you agree with the change, for now, don't want to revert it, right?

Correct - based on previous patches/discussion, I see no reason to block this change. 
It's not pretty to keep adding these checks, but there doesn't seem to be a viable alternative. This mess goes away as we move towards more functionality in GISel?

So LGTM, but please wait a day or so to re-commit in case others have more suggestions.



================
Comment at: llvm/test/CodeGen/RISCV/optnone-store-no-combine.ll:4
+
+; This test case makes sure that the same store will not get optimize out by DAGCombiner
+; and verify that codegen has those stores in it. It will help to improve debug experience.
----------------
"This test verifies that a repeated store is not eliminated with optnone (improves debugging)."


================
Comment at: llvm/test/CodeGen/RISCV/optnone-store-no-combine.ll:7
+
+define dso_local void @foo(i32* %p) noinline nounwind optnone {
+; CHECK-LABEL: foo:
----------------
Remove dso_local, noinline, nounwind?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D115808/new/

https://reviews.llvm.org/D115808