Add torch.compile layernorm

Summary:
While this will probably also be covered in some microbenchmarks for
torch inductor, it seems like we should also have this in tritonbench to cover
triton-based changes in perf here.

Reviewed By: sijiac

Differential Revision: D55902435

fbshipit-source-id: 2622f807086d87a0036c65aa21524242e646a963

torchbenchmark/operators/layer_norm/__init__.py

@@ -23,6 +23,14 @@ def triton_layer_norm(self, *args):
     def torch_layer_norm(self, *args):
         return lambda: F.layer_norm(*args)
 
+    @register_benchmark()
+    def torch_compile_layer_norm(self, *args):
+        @torch.compile
+        def inner(*args):
+            return F.layer_norm(*args)
+
+        return lambda: inner(*args)
+
     def get_bwd_fn(self, fwd_fn: Callable) -> Callable:
         y = fwd_fn()
         dy = 0.1 * torch.randn_like(y)