diff --git a/onnxruntime/core/mlas/lib/kleidiai/mlasi_kleidiai.h b/onnxruntime/core/mlas/lib/kleidiai/mlasi_kleidiai.h
index 2e9c4574fd057..cd68a9d61a680 100644
--- a/onnxruntime/core/mlas/lib/kleidiai/mlasi_kleidiai.h
+++ b/onnxruntime/core/mlas/lib/kleidiai/mlasi_kleidiai.h
@@ -6,7 +6,7 @@
 
 #pragma once
 
-#include "mlasi.h"
+#include "../mlasi.h"
 
 // Fix to ensure compatibility with MSVC build
 #if defined(_MSC_VER)
diff --git a/onnxruntime/test/onnx/microbenchmark/tptest.cc b/onnxruntime/test/onnx/microbenchmark/tptest.cc
index e51b12454294f..1c377f1a5fa96 100644
--- a/onnxruntime/test/onnx/microbenchmark/tptest.cc
+++ b/onnxruntime/test/onnx/microbenchmark/tptest.cc
@@ -37,7 +37,7 @@ BENCHMARK(BM_CreateThreadPool)
 void SimpleForLoop(ptrdiff_t first, ptrdiff_t last) {
   size_t sum = 0;
   for (; first != last; ++first) {
-    ++sum;
+    benchmark::DoNotOptimize(++sum);
   }
 }
 #ifdef _WIN32
@@ -102,7 +102,8 @@ static void BM_ThreadPoolSimpleParallelFor(benchmark::State& state) {
   for (auto _ : state) {
     for (int j = 0; j < 100; j++) {
       ThreadPool::TrySimpleParallelFor(tp.get(), len, [&](size_t) {
-        for (volatile size_t x = 0; x < body; x++) {
+        for (size_t x = 0; x < body; x++) {
+          benchmark::DoNotOptimize(x);
         }
       });
     }