diff --git a/third_party/intel/lib/TritonIntelGPUTransforms/RemoveLayoutConversions.cpp b/third_party/intel/lib/TritonIntelGPUTransforms/RemoveLayoutConversions.cpp
index 24ca2ef748..fe08490b95 100644
--- a/third_party/intel/lib/TritonIntelGPUTransforms/RemoveLayoutConversions.cpp
+++ b/third_party/intel/lib/TritonIntelGPUTransforms/RemoveLayoutConversions.cpp
@@ -417,7 +417,7 @@ SmallVector<Value> LayoutPropagation::propagateToUsers(Value value,
 
 void LayoutPropagation::propagateLayout() {
   SmallVector<Value> queue;
-  for (auto it : layouts) {
+  for (const auto &it : layouts) {
     queue.push_back(it.first);
   }
   while (!queue.empty()) {
diff --git a/third_party/intel/lib/TritonIntelGPUTransforms/RewriteTensorPointer.cpp b/third_party/intel/lib/TritonIntelGPUTransforms/RewriteTensorPointer.cpp
index 7bbd376763..e214f3ad05 100644
--- a/third_party/intel/lib/TritonIntelGPUTransforms/RewriteTensorPointer.cpp
+++ b/third_party/intel/lib/TritonIntelGPUTransforms/RewriteTensorPointer.cpp
@@ -645,7 +645,7 @@ class TritonIntelGPURewriteTensorPointerPass
 
       assert(rewritedInfo.count(newOperands[i]) &&
              "Expecting YieldOp operands in rewritedInfo");
-      auto info = rewritedInfo[newOperands[i]];
+      const auto &info = rewritedInfo[newOperands[i]];
       newOperands = generateNewOperands(newOperands, i, info.getOffsets());
       i += info.length() - 1;
       size += info.length() - 1;