[Mlir-commits] [mlir] [mlir][NVMM] Add `globaltimer_wait` op (PR #129065)

[Durgadoss R]

================
@@ -2614,6 +2613,31 @@ def NVVM_Breakpoint : NVVM_Op<"breakpoint"> {
   let assemblyFormat = "attr-dict";
 }
 
+def NVVM_GlobalTimerWaitOp : NVVM_PTXBuilder_Op<"globaltimer_wait">,  
+    Arguments<(ins I64:$time)> {
+  let summary = "globaltimer busy loop";
----------------
durga4github wrote:

I am wondering if we can leverage the nanosleep instruction for this?

https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#miscellaneous-instructions-nanosleep

https://github.com/llvm/llvm-project/pull/129065