use is_symm_mem_tensor if the API is available

Author: shunting314

helion/_dist_utils.py

@@ -75,20 +75,9 @@ def is_master_rank() -> bool:
 
 
 def is_symm_mem_tensor(t: Tensor, process_group_name: str | None = None) -> bool:
-    if not isinstance(t, Tensor) or not dist.is_initialized():
-        return False
-
-    assert process_group_name is not None
-    try:
-        hdl = symm_mem.rendezvous(
-            t,
-            group=process_group_name,  # pyrefly: ignore[bad-argument-type]
-        )
-        return hdl is not None
-    except RuntimeError:
-        # PyTorch right now throws a RuntimeError if the tensor passed
-        # to rendezvious is not from symm-mem
+    if not isinstance(t, Tensor):
         return False
+    return symm_mem.is_symm_mem_tensor(t)
 
 
 def get_signal_pad_ptrs_dev(t: Tensor, process_group_name: str | None = None) -> int: