[Mlir-commits] [mlir] [mlir] Added `Convergent` trait that matches LLVM's semantics (PR #152358)

[Nikolay Panchenko]

================
@@ -1054,7 +1058,8 @@ def NVVM_CpAsyncWaitGroupOp : NVVM_Op<"cp.async.wait.group">,
   let assemblyFormat = "$n attr-dict";
 }
 
-def NVVM_CpAsyncMBarrierArriveOp : NVVM_Op<"cp.async.mbarrier.arrive"> {
+def NVVM_CpAsyncMBarrierArriveOp : NVVM_Op<"cp.async.mbarrier.arrive",
+  [Convergent]> {
----------------
npanchen wrote:

> I do have low confidence for LLVM current annotation as a reliable source of documentation right now.

That's valid point. At the same time having different traits set on LLVM intrinsic and MLIR operation, that is lowered to that intrinsic, looks not only confusing, but stinky.
Ideally, for low-level dialects, such as NVVM, it will be great just to query that property from the intrinsic it's lowered to, but that does require to have LLVMContext.

Specifically about `convergent`, my understanding, if intrinsics is marked with it, but in reality it's not convergent, it's only bad for performance, but stability-wise it's still correct.
But yeah, it will be great to go through all LLVM intrinsics and make sure proper traits are used.

https://github.com/llvm/llvm-project/pull/152358