always with shared.dyn

microsoft · Aug 9, 2024 · 8b5f083 · 8b5f083
1 parent bcffc49
commit 8b5f083
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/bitblas/base/roller/policy/tensorcore.py b/bitblas/base/roller/policy/tensorcore.py
@@ -334,7 +334,7 @@ def _score(node, thread):  # small is better
             codegen_dict.shared_scope = "shared.dyn"
 
         # Or assume we always use shared memory
-        # codegen_dict.shared_scope = "shared.dyn"
+        codegen_dict.shared_scope = "shared.dyn"
 
         codegen_dict.complete_config(node)
         codegen_dict.vectorize = self._plan_vectorize(self.prim_func_node, td, block_size)