Remove stack dump for compiler failure which can happen often (#7207)

Co-authored-by: Ubuntu <ubuntu@ip-172-31-28-162.us-west-2.compute.internal>
pytorch · Jun 8, 2024 · 5338a3b · 5338a3b
1 parent 1ac8968
commit 5338a3b
Show file tree

Hide file tree

Showing 2 changed files with 8 additions and 1 deletion.
diff --git a/torch_xla/csrc/runtime/debug_macros.h b/torch_xla/csrc/runtime/debug_macros.h
@@ -21,4 +21,10 @@ T ConsumeValue(xla::StatusOr<T>&& status) {
   return std::move(status).value();
 }
 
+template <typename T>
+T ConsumeValueNoStackDump(xla::StatusOr<T>&& status) {
+  TF_CHECK_OK(status.status());
+  return std::move(status).value();
+}
+
 #endif  // XLA_CLIENT_DEBUG_MACROS_H_
diff --git a/torch_xla/csrc/runtime/pjrt_computation_client.cc b/torch_xla/csrc/runtime/pjrt_computation_client.cc
@@ -627,7 +627,8 @@ std::vector<ComputationClient::ComputationPtr> PjRtComputationClient::Compile(
       TF_VLOG(3) << "memory usage is not availiable";
     }
 
-    const auto& hlo_modules = ConsumeValue(executable->GetHloModules());
+    const auto& hlo_modules =
+        ConsumeValueNoStackDump(executable->GetHloModules());
     xla::HloComputation* hlo_computation = hlo_modules[0]->entry_computation();
     std::shared_ptr<PjRtComputation> pjrt_computation =
         std::make_shared<PjRtComputation>(