diff --git a/aten/src/ATen/native/cpu/FlashAttentionKernel.cpp b/aten/src/ATen/native/cpu/FlashAttentionKernel.cpp
index 7956d0c7f5f5..90e34fc70b81 100644
--- a/aten/src/ATen/native/cpu/FlashAttentionKernel.cpp
+++ b/aten/src/ATen/native/cpu/FlashAttentionKernel.cpp
@@ -473,8 +473,7 @@ void cpu_flash_attention(
         scalar_t* transpose_buffer_ptr = transpose_buffer.get();
         std::unique_ptr<scalar_t[]> v_copy_buffer = std::make_unique<scalar_t[]>(ekvSplitSize * packb_size);
         scalar_t* v_copy_buffer_ptr = v_copy_buffer.get();
-        for (const auto z : c10::irange(begin, end)) {
-          (void)z; // Suppress unused variable
+        for (C10_UNUSED auto z : c10::irange(begin, end)) {
           n = l * kvSplitSize;
           int64_t kvBlockSize = std::min(kvSplitSize, kvSize - n);
           int64_t ekvBlockSize = kvBlockSize % 2 == 0 ? kvBlockSize : kvBlockSize + 1;
@@ -567,8 +566,7 @@ void cpu_flash_attention(
             ? query_padding_ptr + ompIdx * qSplitSize * eheadSize
             : nullptr;
 
-    for (const auto z : c10::irange(begin, end)) {
-      (void)z; // Suppress unused variable
+    for (C10_UNUSED auto z : c10::irange(begin, end)) {
       int64_t m = k * qSplitSize;
       int64_t qBlockSize = std::min(qSplitSize, qSize - m);
       // Initialize max and sum
@@ -933,8 +931,7 @@ void cpu_flash_attention_backward(
 
     at::Tensor dsum = at::empty({qSplitSize}, query.options().dtype(accumulate_dtype));
     accum_t* dsum_data = dsum.data_ptr<accum_t>();
-    for (const auto z : c10::irange(begin, end)) {
-      (void)z; // Suppress unused variable
+    for (C10_UNUSED auto z : c10::irange(begin, end)) {
       // rowsum of grad_out * out
       for (int64_t m = 0; m < qSize; m += qSplitSize) {
         int64_t qBlockSize = std::min(qSplitSize, qSize - m);
diff --git a/aten/src/ATen/native/cuda/Sorting.cu b/aten/src/ATen/native/cuda/Sorting.cu
index 6272bbb9b75d..290be3926c6f 100644
--- a/aten/src/ATen/native/cuda/Sorting.cu
+++ b/aten/src/ATen/native/cuda/Sorting.cu
@@ -177,12 +177,11 @@ struct KthValueLauncher {
       cuda::detail::TensorInfo<scalar_t, index_t> values_info,
       int collapse_values_dim,
       cuda::detail::TensorInfo<int64_t, index_t> indices_info,
-      int collapse_indices_dim,
+      C10_UNUSED int collapse_indices_dim,
       cuda::detail::TensorInfo<const scalar_t, index_t> self_info,
       int collapse_self_dim,
       int64_t num_slices,
       int64_t slice_size) {
-    (void)collapse_indices_dim; // Suppress unused variable warning
     dim3 grid;
     if (!getGridFromTiles(num_slices, grid)) {
       AT_ERROR("slices are too many");
@@ -213,15 +212,13 @@ struct MedianLauncher {
   template <typename scalar_t, typename index_t, int all_dims>
   inline void launch(
       cuda::detail::TensorInfo<scalar_t, index_t> values_info,
-      int collapse_values_dim,
+      C10_UNUSED int collapse_values_dim,
       cuda::detail::TensorInfo<int64_t, index_t> indices_info,
-      int collapse_indices_dim,
+      C10_UNUSED int collapse_indices_dim,
       cuda::detail::TensorInfo<const scalar_t, index_t> self_info,
       int collapse_self_dim,
       int64_t num_slices,
       int64_t slice_size) {
-    (void)collapse_values_dim; // Suppress unused variable warning
-    (void)collapse_indices_dim; // Suppress unused variable warning
     dim3 grid;
     if (!getGridFromTiles(num_slices, grid)) {
       AT_ERROR("slices are too many");
diff --git a/aten/src/ATen/native/mps/operations/Normalization.mm b/aten/src/ATen/native/mps/operations/Normalization.mm
index 1d060f78ee23..ee32bc3d2274 100644
--- a/aten/src/ATen/native/mps/operations/Normalization.mm
+++ b/aten/src/ATen/native/mps/operations/Normalization.mm
@@ -904,8 +904,7 @@ std::tuple<Tensor, Tensor, Tensor> layer_norm_mps(const Tensor& input,
   for (const auto idx : c10::irange(axis)) {
     stat_shape.push_back(input_shape[idx]);
   }
-  for (const auto idx : c10::irange(axis, input.dim())) {
-    (void)idx; // Suppress unused variable
+  for (C10_UNUSED auto idx : c10::irange(axis, input.dim())) {
     stat_shape.push_back(1);
   }
   mean = mean.view(stat_shape);