Update on "[cutlass backend] Reduce log level for cutlass runtime error"

henrylhtsang · henrylhtsang · commit 2896e50e1b1e · 2025-05-16T14:45:25.000-07:00
Want to make sure we always call self.cleanup_run_fn() even if we crash. I think this is the reason why sometimes we get ``` in _dlclose TypeError: 'NoneType' object is not callable ``` Differential Revision: [D74629230](https://our.internmc.facebook.com/intern/diff/D74629230/) cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov [ghstack-poisoned]
diff --git a/torch/_inductor/autotune_process.py b/torch/_inductor/autotune_process.py
@@ -733,11 +733,11 @@ def make_run_fn(
         except RuntimeError as e:
             err_msg = str(e)
 
-            def dummy_function():
+            def raise_runtime_error():
                 raise RuntimeError(err_msg)
 
             self.cleanup_run_fn()
-            return dummy_function
+            return raise_runtime_error
 
         return ret
 
diff --git a/torch/_inductor/select_algorithm.py b/torch/_inductor/select_algorithm.py
@@ -2269,14 +2269,15 @@ def benchmark_choices(
 
                 if not isinstance(choice, CUDATemplateCaller):
                     log.error(
-                        "CUDA compilation error during autotuning: \n%s. \nIgnoring this choice.",
+                        "CUDA runtime error during autotuning: \n%s. \nIgnoring this choice.",
                         e,
                     )
                 msg = str(e)
                 if "invalid argument" in msg:
                     msg += "\n\nThis may mean this GPU is too small for max_autotune mode.\n\n"
-                elif "illegal memory access" in msg:
-                    msg += "\n\nEither error in template or triton bug.\n"
+                else:
+                    if "illegal memory access" in msg:
+                        msg += "\n\nEither error in template or triton bug.\n"
 
                 if isinstance(choice, CUDATemplateCaller):
                     log.debug(
