[Mlir-commits] [mlir] [mlir][ROCDL] Add async variants of pre-gfx12 LDS load intrinsics (PR #181072)

[Krzysztof Drewniak]

================
@@ -1224,12 +1224,24 @@ llvm.func @rocdl.load.to.lds(%src : !llvm.ptr<7>, %dst: !llvm.ptr<3>) {
   llvm.return
 }
 
+llvm.func @rocdl.load.async.to.lds(%src : !llvm.ptr<7>, %dst: !llvm.ptr<3>) {
+  //CHECK: call void @llvm.amdgcn.load.async.to.lds.p7
----------------
krzysz00 wrote:

Exactly

You have to use asyncmark / wait.asyncmark instead to get correctness

https://github.com/llvm/llvm-project/pull/181072