[ET-VK] Fix op benchmarks code generation (#6528)

## Context As title. There was a small error in the operator benchmark code generation which impacted operators which return a collection of tensors (e.g. layer norm). The issue was that the generated code would store the output `ValueRefs` in variables that shadowed class variables of the benchmark. Differential Revision: [D65071024](https://our.internmc.facebook.com/intern/diff/D65071024/) ghstack-source-id: 250448840 Pull Request resolved: #6527 Co-authored-by: Stephen Jia <[email protected]>
pytorch · Oct 28, 2024 · 85d3ff6 · 85d3ff6
1 parent 3366edf
commit 85d3ff6
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/backends/vulkan/test/op_tests/utils/gen_computegraph.py b/backends/vulkan/test/op_tests/utils/gen_computegraph.py
@@ -448,7 +448,7 @@ def set_output(self, ref: ValueRefList, include_declarations: bool = True) -> st
         if isinstance(ref, list):
             ret_str = ""
             for r in ref[:-1]:
-                ret_str += self.set_output(r)
+                ret_str += self.set_output(r, include_declarations)
             return ret_str
         elif ref.src_cpp_type == TENSOR_VECTOR:
             assert ref.is_out