From 26ba7be0fa00587f6eaea78d64e8ac73c08e8177 Mon Sep 17 00:00:00 2001
From: Phuong Nguyen <phuonguyen@nvidia.com>
Date: Wed, 11 Dec 2024 14:49:06 -0800
Subject: [PATCH] add thread_local for norm registry instance

Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>
---
 transformer_engine/common/normalization/common.h | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/transformer_engine/common/normalization/common.h b/transformer_engine/common/normalization/common.h
index 8a8df63ba4..d1d56d5cc9 100644
--- 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;
   }