add thread_local for norm registry instance

Signed-off-by: Phuong Nguyen <[email protected]>
NVIDIA · Dec 11, 2024 · 26ba7be · 26ba7be
1 parent a149fb0
commit 26ba7be
Showing 1 changed file with 1 addition and 2 deletions.
diff --git a/transformer_engine/common/normalization/common.h b/transformer_engine/common/normalization/common.h
@@ -287,9 +287,8 @@ class CudnnNormalizationPlan : public NormalizationPlanBase {
 
 class NormalizationPlanRegistry {
  public:
-  // TODO thread-safe
   static NormalizationPlanRegistry& getInstance() {
-    static NormalizationPlanRegistry instance;
+    static thread_local NormalizationPlanRegistry instance;
     return instance;
   }