[Mlir-commits] [mlir] [mlir][ROCDL] Wrap asyncmark and wait.asyncmark intrinsics (PR #181054)

[Krzysztof Drewniak]

================
@@ -590,6 +590,61 @@ def ROCDL_WaitTensorcntOp: ROCDL_ConcreteNonMemIntrOp<"s.wait.tensorcnt", [], 0,
   let assemblyFormat = "$count attr-dict";
 }
 
+def ROCDL_AsyncmarkOp : ROCDL_ConcreteNonMemIntrOp<"asyncmark", [], 0>,
+    Arguments<(ins)> {
+  let summary = "Mark the end of a group of asynchronous operations";
+  let description = [{
+      This operation, in conjunction with `rocdl.wait.asyncmark`, forms the
+      compiler-provided framework for tracking explicitly asynchronous
+      memory operations, such as copies to LDS that use async intrinsics
+      and gfx1250's tensor loads.
+
+      Details of its behavior can be found in
+      [the LLVM documentation on async tracking](/llvm/docs/AMDGPUAsyncOperations.rst).
+
+      See `rocdl.wait.asyncmark`'s documentation for a usage example.
+
+      Available on gfx9 and later.
----------------
krzysz00 wrote:

There's a pointer to an example in the immediately following op

I'm not sure if an example that is just straight up the name of the operation adds anything, so I didn't put one in.

https://github.com/llvm/llvm-project/pull/181054