Boolean masking backwards doesn't work even with dynamic output shape…

… ops, break accordingly (pytorch#114126)

Signed-off-by: Edward Z. Yang <ezyang@meta.com>

Pull Request resolved: pytorch#114126
Approved by: https://github.com/albanD

torch/_dynamo/variables/tensor.py

@@ -589,13 +589,12 @@ def has_bool_key(v):
                     return False
 
             if (
-                not config.capture_dynamic_output_shape_ops
-                and has_bool_key(key)
+                has_bool_key(key)
                 and isinstance(value, TensorVariable)
                 and value.requires_grad
             ):
                 unimplemented(
-                    "boolean masking setitem backwards requires dynamic shapes"
+                    "boolean masking setitem backwards, see https://github.com/pytorch/pytorch/issues/114123"
                 )
             tx.output.create_proxy(
                 "call_function",