Fix RuntimeError: value cannot be converted to type int64_t without overflow (pytorch#147492)

The exact call is coming from here:

https://github.com/pytorch/pytorch/blob/78a94c911435bf9b1bb45888033a29081e406ec2/torch/_inductor/memory.py#L161

I have no idea why this error is being thrown and what mode/modes might be failing for this

Pull Request resolved: pytorch#147492
Approved by: https://github.com/eellison

Author: drisspg

torch/_inductor/utils.py

@@ -969,6 +969,10 @@ def evaluate(expr: Union[bool, torch.SymInt, sympy.Expr]) -> bool:
 
 @functools.lru_cache(8)
 def get_dtype_size(dtype: torch.dtype) -> int:
+    # TODO: Investigate why uint64 tensor creation causes overflow error:
+    # Workaround for RuntimeError in memory size calculation, but underlying cause unclear
+    if dtype == torch.uint64:
+        return 8
     return torch.empty((), dtype=dtype).element_size()