diff --git a/CMakeLists.txt b/CMakeLists.txt
index 05259984f3340a..04b32ba7c24fd1 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -744,7 +744,6 @@ if(NOT MSVC)
   string(APPEND CMAKE_CXX_FLAGS " -Wno-unknown-pragmas")
   string(APPEND CMAKE_CXX_FLAGS " -Wno-sign-compare")
   string(APPEND CMAKE_CXX_FLAGS " -Wno-unused-parameter")
-  string(APPEND CMAKE_CXX_FLAGS " -Wno-unused-variable")
   string(APPEND CMAKE_CXX_FLAGS " -Wno-unused-function")
   string(APPEND CMAKE_CXX_FLAGS " -Wno-unused-result")
   string(APPEND CMAKE_CXX_FLAGS " -Wno-unused-local-typedefs")
@@ -1068,6 +1067,9 @@ endif()
 
 # ---[ JNI
 if(BUILD_JNI)
+  if(NOT MSVC)
+    string(APPEND CMAKE_CXX_FLAGS " -Wno-unused-variable")
+  endif()
   set(BUILD_LIBTORCH_WITH_JNI 1)
   set(FBJNI_SKIP_TESTS 1)
   add_subdirectory(android/pytorch_android)
diff --git a/aten/src/ATen/mkl/Limits.h b/aten/src/ATen/mkl/Limits.h
index b0d3829e458ed1..2a0ed662a6d5cd 100644
--- a/aten/src/ATen/mkl/Limits.h
+++ b/aten/src/ATen/mkl/Limits.h
@@ -6,6 +6,6 @@ namespace at { namespace native {
 
   // Since size of MKL_LONG varies on different platforms (linux 64 bit, windows
   // 32 bit), we need to programmatically calculate the max.
-  static int64_t MKL_LONG_MAX = ((1LL << (sizeof(MKL_LONG) * 8 - 2)) - 1) * 2 + 1;
+  constexpr int64_t MKL_LONG_MAX = ((1LL << (sizeof(MKL_LONG) * 8 - 2)) - 1) * 2 + 1;
 
 }}  // namespace
diff --git a/aten/src/ATen/test/cuda_stream_test.cpp b/aten/src/ATen/test/cuda_stream_test.cpp
index 6bf83d2fcdffa6..5a117e772731cc 100644
--- a/aten/src/ATen/test/cuda_stream_test.cpp
+++ b/aten/src/ATen/test/cuda_stream_test.cpp
@@ -173,6 +173,7 @@ TEST(TestStream, StreamPoolTest) {
   if (!at::cuda::is_available()) return;
   std::vector<at::cuda::CUDAStream> streams{};
   for (const auto i : c10::irange(200)) {
+    (void)i;
     streams.emplace_back(at::cuda::getStreamFromPool());
   }
 
diff --git a/aten/src/ATen/test/cuda_tensor_interop_test.cpp b/aten/src/ATen/test/cuda_tensor_interop_test.cpp
index 89019d7950b9a4..2a2833b7b2748a 100644
--- a/aten/src/ATen/test/cuda_tensor_interop_test.cpp
+++ b/aten/src/ATen/test/cuda_tensor_interop_test.cpp
@@ -84,6 +84,8 @@ TEST(CUDAPytorchToCaffe2, Op) {
 
   auto* c2_tensor_a = BlobSetTensor(workspace.CreateBlob("a"), caffe2::Tensor(at_tensor_a));
   auto* c2_tensor_b = BlobSetTensor(workspace.CreateBlob("b"), caffe2::Tensor(at_tensor_b));
+  (void)c2_tensor_a;
+  (void)c2_tensor_b;
 
   // Test Alias
   {
diff --git a/aten/src/ATen/test/math_kernel_test.cpp b/aten/src/ATen/test/math_kernel_test.cpp
index 10bdd73b2d02f2..b6d7b5ae355a7b 100644
--- a/aten/src/ATen/test/math_kernel_test.cpp
+++ b/aten/src/ATen/test/math_kernel_test.cpp
@@ -54,8 +54,6 @@ TEST(MathKernelTest, NativeGroupNorm) {
 TEST(MathKernelTest, NativeLayerNorm) {
   const auto input = rand({20, 10, 10, 10});
   const auto input_shape = input.sizes();
-  // NOLINTNEXTLINE(clang-analyzer-deadcode.DeadStores)
-  const auto input_ndim = input.dim();
 
   double eps = 1e-05;
   for (bool undef_weight: {true, false}) {
diff --git a/benchmarks/cpp/nvfuser/CMakeLists.txt b/benchmarks/cpp/nvfuser/CMakeLists.txt
index 1024c4012a3cea..22b23e5e9ab7da 100644
--- a/benchmarks/cpp/nvfuser/CMakeLists.txt
+++ b/benchmarks/cpp/nvfuser/CMakeLists.txt
@@ -15,4 +15,8 @@ if(USE_CUDA)
     main.cpp)
 
   target_link_libraries(nvfuser_bench PRIVATE torch_library benchmark)
+  if(NOT MSVC)
+    target_compile_options(nvfuser_bench PRIVATE -Wno-unused-variable)
+  endif()
+
 endif()
diff --git a/c10/test/CMakeLists.txt b/c10/test/CMakeLists.txt
index acd9b4ded43846..a68d2b899bd79a 100644
--- a/c10/test/CMakeLists.txt
+++ b/c10/test/CMakeLists.txt
@@ -6,6 +6,9 @@ if(BUILD_TEST)
     get_filename_component(test_file_name ${test_src} NAME_WE)
     set(test_name "c10_${test_file_name}")
     add_executable(${test_name} "${test_src}")
+    if(NOT MSVC)
+      target_compile_options(${test_name} PRIVATE -Wno-unused-variable)
+    endif()
     target_link_libraries(${test_name} c10 gmock gtest gtest_main)
     add_test(NAME ${test_name} COMMAND $<TARGET_FILE:${test_name}>)
     if(INSTALL_TEST)
diff --git a/caffe2/CMakeLists.txt b/caffe2/CMakeLists.txt
index 5f2892350fe901..3743b5703b6209 100644
--- a/caffe2/CMakeLists.txt
+++ b/caffe2/CMakeLists.txt
@@ -1762,6 +1762,9 @@ if(BUILD_TEST)
     target_include_directories(${test_name} PRIVATE $<INSTALL_INTERFACE:include>)
     target_include_directories(${test_name} PRIVATE $<BUILD_INTERFACE:${CMAKE_BINARY_DIR}/include>)
     target_include_directories(${test_name} PRIVATE ${Caffe2_CPU_INCLUDE})
+    if(NOT MSVC)
+      target_compile_options(${test_name} PRIVATE -Wno-unused-variable)
+    endif()
     add_test(NAME ${test_name} COMMAND $<TARGET_FILE:${test_name}>)
     if(INSTALL_TEST)
       install(TARGETS ${test_name} DESTINATION test)
diff --git a/caffe2/core/blob_serialization.cc b/caffe2/core/blob_serialization.cc
index 70d1b59dc814ff..1b2c58bacd467f 100644
--- a/caffe2/core/blob_serialization.cc
+++ b/caffe2/core/blob_serialization.cc
@@ -399,8 +399,8 @@ void TensorSerializer::SerializeWithOptions(
   std::vector<std::future<void>> futures;
   if (tensor.numel() > chunk_size) {
     futures.reserve(FLAGS_caffe2_max_tensor_serializer_threads);
-    // NOLINTNEXTLINE(clang-analyzer-deadcode.DeadStores,clang-diagnostic-unused-variable)
     for (const auto i : c10::irange(FLAGS_caffe2_max_tensor_serializer_threads)) {
+      (void)i;
       futures.emplace_back(std::async(std::launch::async, task));
     }
   }
diff --git a/caffe2/operators/cast_op.h b/caffe2/operators/cast_op.h
index ab0323fc52d41e..478f2f30380c28 100644
--- a/caffe2/operators/cast_op.h
+++ b/caffe2/operators/cast_op.h
@@ -19,7 +19,6 @@ class CastOp : public Operator<Context> {
       : Operator<Context>(operator_def, ws) {
     const ArgumentHelper helper(operator_def);
     TensorProto_DataType to = cast::GetCastDataType(helper, "to");
-    TensorProto_DataType from = cast::GetCastDataType(helper, "from_type");
 
     SetBody(to);
   }
diff --git a/caffe2/operators/conv_op_cudnn.cc b/caffe2/operators/conv_op_cudnn.cc
index cfbebcfe601c4a..36a5f4c4bcaeb2 100644
--- a/caffe2/operators/conv_op_cudnn.cc
+++ b/caffe2/operators/conv_op_cudnn.cc
@@ -576,7 +576,9 @@ bool CudnnConvOp::DoRunWithType() {
     return true;
   }
 
+#if !CUDNN_VERSION_MIN(7, 0, 0)
   int group_offset_filter = filter.numel() / group_;
+#endif
 
   // Set up the cudnn algorithms & workspace if necessary
   bool input_changed = (X.sizes() != cudnn_input_dims_);
@@ -951,7 +953,9 @@ bool CudnnConvGradientOp::DoRunWithType() {
       "If you set group, the number of output channels should be divisible "
       "by group.");
 
+#if !CUDNN_VERSION_MIN(7, 0, 0)
   int group_offset_filter = filter.numel() / group_;
+#endif
   if (kernel_.size() == 1) {
     ConvPoolOpBase<CUDAContext>::ComputePads({H});
   } else if (kernel_.size() == 2) {
diff --git a/caffe2/operators/fused_rowwise_nbitfake_conversion_ops.cc b/caffe2/operators/fused_rowwise_nbitfake_conversion_ops.cc
index 6a483a8fb011e6..914b8d8abc8a7e 100644
--- a/caffe2/operators/fused_rowwise_nbitfake_conversion_ops.cc
+++ b/caffe2/operators/fused_rowwise_nbitfake_conversion_ops.cc
@@ -25,11 +25,10 @@ float compress_uniform_simplified_(
   float inverse_scale = 1.0f / scale;
 
   float norm = 0.0f;
-  // NOLINTNEXTLINE(clang-diagnostic-unused-variable)
-  constexpr int VLEN = 8;
   int i = 0;
 
 #ifdef __AVX__
+  constexpr int VLEN = 8;
   // vectorized loop
   __m256 norm_v = _mm256_setzero_ps();
   for (; i < N / VLEN * VLEN; i += VLEN) {
diff --git a/caffe2/operators/fused_rowwise_nbitfake_conversion_ops.h b/caffe2/operators/fused_rowwise_nbitfake_conversion_ops.h
index 53462073e9b247..105f9902b21599 100644
--- a/caffe2/operators/fused_rowwise_nbitfake_conversion_ops.h
+++ b/caffe2/operators/fused_rowwise_nbitfake_conversion_ops.h
@@ -72,7 +72,6 @@ class FloatToFusedNBitFakeRowwiseQuantizedOp final
       CAFFE_THROW("Unsupported data type");
     }
 
-    bool use_openmp = GREEDY;
 #ifdef _OPENMP
     vector<float> tmp_vec(input_columns * (GREEDY ? omp_get_max_threads() : 1));
 #else
diff --git a/caffe2/operators/gather_fused_8bit_rowwise_op.h b/caffe2/operators/gather_fused_8bit_rowwise_op.h
index b0d57436e09180..a111ff3eca4dbb 100644
--- a/caffe2/operators/gather_fused_8bit_rowwise_op.h
+++ b/caffe2/operators/gather_fused_8bit_rowwise_op.h
@@ -30,7 +30,6 @@ class GatherFused8BitRowwiseOp : public Operator<Context> {
     const std::vector<int64_t> shape = {indices.size(0), data.size(1) - 8};
     auto* output = Output(0, shape, at::dtype<float>());
 
-    int block_size = shape[1];
     auto block_bytesize = data.size_from_dim(1) * data.dtype().itemsize();
     int N = indices.numel();
 
diff --git a/caffe2/operators/generate_proposals_op_util_nms.h b/caffe2/operators/generate_proposals_op_util_nms.h
index 26e86c4407542f..ff68d49251c1e3 100644
--- a/caffe2/operators/generate_proposals_op_util_nms.h
+++ b/caffe2/operators/generate_proposals_op_util_nms.h
@@ -133,8 +133,7 @@ std::vector<int> soft_nms_cpu_upright(
 
     // Find proposal with max score among remaining proposals
     int max_pos;
-    // NOLINTNEXTLINE(clang-diagnostic-unused-variable)
-    auto max_score = GetSubArray(*out_scores, pending).maxCoeff(&max_pos);
+    GetSubArray(*out_scores, pending).maxCoeff(&max_pos);
     int i = pending[max_pos];
     keep.push_back(i);
 
@@ -635,8 +634,7 @@ std::vector<int> soft_nms_cpu_rotated(
 
     // Find proposal with max score among remaining proposals
     int max_pos;
-    // NOLINTNEXTLINE(clang-diagnostic-unused-variable)
-    auto max_score = GetSubArray(*out_scores, pending).maxCoeff(&max_pos);
+    GetSubArray(*out_scores, pending).maxCoeff(&max_pos);
     int i = pending[max_pos];
     keep.push_back(i);
 
diff --git a/caffe2/operators/h_softmax_op.cc b/caffe2/operators/h_softmax_op.cc
index 333fab96afef38..9913e81c297279 100644
--- a/caffe2/operators/h_softmax_op.cc
+++ b/caffe2/operators/h_softmax_op.cc
@@ -458,8 +458,6 @@ bool HuffmanTreeHierarchyOp<T, Context>::RunOnDevice() {
   std::vector<int> labelIndices;
   labelIndices.resize(num_classes_);
 
-  // NOLINTNEXTLINE(clang-diagnostic-unused-variable)
-  int current_node_index = 0;
   for (int i = 0; i < num_classes_; ++i) {
     Node node(i, labelCounts[i]);
     nodes.push(node);
diff --git a/caffe2/operators/layer_norm_op.h b/caffe2/operators/layer_norm_op.h
index e61eecdbd8c222..70a96d77ff21a2 100644
--- a/caffe2/operators/layer_norm_op.h
+++ b/caffe2/operators/layer_norm_op.h
@@ -132,8 +132,6 @@ class LayerNormGradientOp final : public Operator<Context> {
   template <typename T>
   bool DoRunWithType() {
     const auto& dY = Input(0);
-    // NOLINTNEXTLINE(clang-diagnostic-unused-variable)
-    const auto& Y = Input(1);
     const auto& mean = Input(2);
     const auto& sigma = Input(3);
     const auto& X = Input(4);
diff --git a/caffe2/operators/lengths_reducer_rowwise_8bit_ops.h b/caffe2/operators/lengths_reducer_rowwise_8bit_ops.h
index e4c0366833e755..67d792f4ca23ce 100644
--- a/caffe2/operators/lengths_reducer_rowwise_8bit_ops.h
+++ b/caffe2/operators/lengths_reducer_rowwise_8bit_ops.h
@@ -53,7 +53,6 @@ class SparseLengths8BitsRowwiseOp : public Operator<Context> {
         "the second dim of scale_bias has to be equal to 2");
     CAFFE_ENFORCE_EQ(1, indicesInput.dim(), "INDICES must be a vector");
     const IndexType* indices = indicesInput.template data<IndexType>();
-    int64_t dataToReduceSize = indicesInput.size(0);
 
     const int* lengths = lengthsInput.template data<int>();
     vector<int64_t> shape = dataInput.sizes().vec();
diff --git a/caffe2/operators/local_response_normalization_op_cudnn.cc b/caffe2/operators/local_response_normalization_op_cudnn.cc
index 5795ac6716c13c..dc6a29965df31b 100644
--- a/caffe2/operators/local_response_normalization_op_cudnn.cc
+++ b/caffe2/operators/local_response_normalization_op_cudnn.cc
@@ -193,8 +193,6 @@ bool CuDNNLRNGradientOp::DoRunWithType() {
 
 bool CuDNNLRNGradientOp::RunOnDevice() {
   // dispatch based on contents of tensor(s)
-  const auto& X = Input(0);
-  const auto& Y = Input(1);
   const auto& dY = Input(2);
   auto* dX = Output(0);
 
diff --git a/caffe2/operators/quantized/int8_add_op.h b/caffe2/operators/quantized/int8_add_op.h
index ea5cb8d5114cdc..67b48f0a87d834 100644
--- a/caffe2/operators/quantized/int8_add_op.h
+++ b/caffe2/operators/quantized/int8_add_op.h
@@ -55,8 +55,10 @@ class Int8AddOp final : public Operator<CPUContext> {
 
     initQNNPACK();
 
+#if !defined(FBCODE_CAFFE2) && defined(USE_INTERNAL_PTHREADPOOL_IMPL)
     pthreadpool_t threadpool =
         reinterpret_cast<pthreadpool_t>(ws_->GetThreadPool());
+#endif
 
     if (this->qnnpackOperator_ == nullptr) {
       const qnnp_status createStatus = qnnp_create_add_nc_q8(
diff --git a/caffe2/operators/quantized/int8_channel_shuffle_op.h b/caffe2/operators/quantized/int8_channel_shuffle_op.h
index 77a34c41c6be1d..52dabb31ed065a 100644
--- a/caffe2/operators/quantized/int8_channel_shuffle_op.h
+++ b/caffe2/operators/quantized/int8_channel_shuffle_op.h
@@ -47,7 +47,6 @@ class Int8ChannelShuffleOp final : public ConvPoolOpBase<CPUContext> {
     const auto C = X.t.dim32(3);
     const auto G = this->group_;
     CAFFE_ENFORCE(C % G == 0, "");
-    const auto B = X.t.numel() / C;
 
     initQNNPACK();
 
diff --git a/caffe2/operators/quantized/int8_conv_op.h b/caffe2/operators/quantized/int8_conv_op.h
index a309be9c11a6cf..055e85141e2619 100644
--- a/caffe2/operators/quantized/int8_conv_op.h
+++ b/caffe2/operators/quantized/int8_conv_op.h
@@ -60,8 +60,10 @@ class Int8ConvOp final : public ConvPoolOpBase<CPUContext> {
     runWithSharedBuffer<CPUContext>(ws_, [&](Tensor* buffer) {
       initQNNPACK();
 
+#if !defined(FBCODE_CAFFE2) && defined(USE_INTERNAL_PTHREADPOOL_IMPL)
       pthreadpool_t threadpool =
           reinterpret_cast<pthreadpool_t>(ws_->GetThreadPool());
+#endif
 
       if (this->qnnpackObject_ == nullptr) {
         CAFFE_ENFORCE(
diff --git a/caffe2/operators/quantized/int8_conv_transpose_op.h b/caffe2/operators/quantized/int8_conv_transpose_op.h
index 66f9ff9471064e..797acc2748aef3 100644
--- a/caffe2/operators/quantized/int8_conv_transpose_op.h
+++ b/caffe2/operators/quantized/int8_conv_transpose_op.h
@@ -39,17 +39,12 @@ class Int8ConvTransposeOp final : public ConvTransposeUnpoolBase<CPUContext> {
     const auto& W = Inputs()[1]->template Get<Int8TensorCPU>();
     const auto& B = Inputs()[2]->template Get<Int8TensorCPU>();
     auto* Y = Outputs()[0]->template GetMutable<Int8TensorCPU>();
-    const auto X_offset = -X.zero_point;
-    const auto W_offset = -W.zero_point;
     const int32_t Y_offset =
         this->template GetSingleArgument<int>("Y_zero_point", 0);
     double Y_scale = this->template GetSingleArgument<float>("Y_scale", 1);
     Y->scale = Y_scale;
     Y->zero_point = Y_offset;
 
-    const auto N = X.t.size(0);
-    const auto IH = X.t.size(1);
-    const auto IW = X.t.size(2);
     const auto IC = X.t.size(3);
 
     CHECK_EQ(IC, W.t.size(0));
@@ -64,8 +59,10 @@ class Int8ConvTransposeOp final : public ConvTransposeUnpoolBase<CPUContext> {
     runWithSharedBuffer<CPUContext>(ws_, [&](Tensor* buffer) {
       initQNNPACK();
 
+#if !defined(FBCODE_CAFFE2) && defined(USE_INTERNAL_PTHREADPOOL_IMPL)
       pthreadpool_t threadpool =
           reinterpret_cast<pthreadpool_t>(ws_->GetThreadPool());
+#endif
 
       if (this->qnnpackObject_ == nullptr) {
         const qnnp_status createStatus = qnnp_create_deconvolution2d_nhwc_q8(
diff --git a/caffe2/operators/quantized/int8_fc_op.h b/caffe2/operators/quantized/int8_fc_op.h
index 58e85caebebd52..3f0e65adc9f52a 100644
--- a/caffe2/operators/quantized/int8_fc_op.h
+++ b/caffe2/operators/quantized/int8_fc_op.h
@@ -47,8 +47,10 @@ class Int8FCOp final : public Operator<CPUContext> {
     runWithSharedBuffer<CPUContext>(ws_, [&](Tensor* buffer) {
       initQNNPACK();
 
+#if !defined(FBCODE_CAFFE2) && defined(USE_INTERNAL_PTHREADPOOL_IMPL)
       pthreadpool_t threadpool =
           reinterpret_cast<pthreadpool_t>(ws_->GetThreadPool());
+#endif
 
       if (this->qnnpackObject_ == nullptr) {
         const qnnp_status createStatus = qnnp_create_fully_connected_nc_q8(
diff --git a/caffe2/operators/quantized/int8_quantize_op.h b/caffe2/operators/quantized/int8_quantize_op.h
index 32bdd15ea1c334..7845107cf3e91c 100644
--- a/caffe2/operators/quantized/int8_quantize_op.h
+++ b/caffe2/operators/quantized/int8_quantize_op.h
@@ -19,10 +19,10 @@ void Int8Quantize(
     const int64_t N,
     const float Y_scale,
     const int32_t Y_offset) {
-  const float inv_scale = 1.0f / Y_scale;
   uint32_t i = 0;
 
 #ifdef INT8_NEON_SIMD
+  const float inv_scale = 1.0f / Y_scale;
   const float32x4_t vinv_scale = vdupq_n_f32(inv_scale);
   // magic float and magic int to take care of rounding
   // int magic_round(float f): interpret_int32(f + 12582912.0f) - 0x4B400000
diff --git a/caffe2/operators/quantized/int8_softmax_op.h b/caffe2/operators/quantized/int8_softmax_op.h
index e62b2237b1cd2c..deba3a42c7d960 100644
--- a/caffe2/operators/quantized/int8_softmax_op.h
+++ b/caffe2/operators/quantized/int8_softmax_op.h
@@ -38,7 +38,6 @@ class Int8SoftmaxOp final : public Operator<CPUContext> {
      * in-place, we may overwrite these parameters later, when we set
      * quantization parameters for output tensor.
      */
-    const uint8_t X_zero_point = X.zero_point;
     const float X_scale = X.scale;
 
     Y->scale = Y_scale;
diff --git a/caffe2/operators/text_file_reader.cc b/caffe2/operators/text_file_reader.cc
index fe2680bc2a28e0..96250cdd25812f 100644
--- a/caffe2/operators/text_file_reader.cc
+++ b/caffe2/operators/text_file_reader.cc
@@ -141,8 +141,6 @@ class TextFileReaderReadOp : public Operator<CPUContext> {
                   (field > 0 && token.startDelimId == 1),
               "Invalid number of columns at row ",
               instance->rowsRead + rowsRead + 1);
-          // NOLINTNEXTLINE(clang-diagnostic-unused-variable)
-          const auto& meta = instance->fieldMetas[field];
           char*& data = datas[field];
           convert(
               (TensorProto_DataType)instance->fieldTypes[field],
diff --git a/caffe2/operators/utility_ops.h b/caffe2/operators/utility_ops.h
index f0810813333865..e76649346dc31d 100644
--- a/caffe2/operators/utility_ops.h
+++ b/caffe2/operators/utility_ops.h
@@ -686,8 +686,7 @@ class ScatterAssignOp : public Operator<Context> {
     const auto dataType = TypeMetaToDataType(data.dtype());
     const auto slicesType = TypeMetaToDataType(slices.dtype());
     const auto indicesType = TypeMetaToDataType(indices.dtype());
-    // NOLINTNEXTLINE(clang-diagnostic-unused-variable)
-    auto* output = Output(0);
+    C10_UNUSED auto* output = Output(0);
 
     auto runner = GetRunner(dataType, slicesType, indicesType);
     (this->*runner)();
diff --git a/caffe2/operators/weighted_sample_op.cc b/caffe2/operators/weighted_sample_op.cc
index 8243fa1ce7ed95..cfc625f9f7c3aa 100644
--- a/caffe2/operators/weighted_sample_op.cc
+++ b/caffe2/operators/weighted_sample_op.cc
@@ -57,8 +57,7 @@ bool WeightedSampleOp<float, CPUContext>::RunOnDevice() {
       }
     }
   } else {
-    // NOLINTNEXTLINE(clang-diagnostic-unused-variable,clang-analyzer-deadcode.DeadStores)
-    auto* out_idx = Output(0, {0}, at::dtype<int>());
+    C10_UNUSED auto* out_idx = Output(0, {0}, at::dtype<int>());
     if (OutputSize() == 2) {
       auto* out_value = Output(1, {0}, at::dtype<float>());
       out_value->template mutable_data<float>();
diff --git a/caffe2/opt/bound_shape_inferencer.cc b/caffe2/opt/bound_shape_inferencer.cc
index 344242273db363..e359d17c57c1ce 100644
--- a/caffe2/opt/bound_shape_inferencer.cc
+++ b/caffe2/opt/bound_shape_inferencer.cc
@@ -436,13 +436,6 @@ void BoundShapeInferencer::InferSparseLengthsSum(const OperatorDef& op) {
        op.type() == "SparseLengthsWeightedSum4BitRowwiseSparse" ||
        op.type() == "SparseLengthsSum4BitRowwiseSparse");
 
-  // NOLINTNEXTLINE(clang-analyzer-deadcode.DeadStores,clang-diagnostic-unused-variable)
-  const bool isSparse =
-      (op.type() == "SparseLengthsSum4BitRowwiseSparse" ||
-       op.type() == "SparseLengthsWeightedSum4BitRowwiseSparse" ||
-       op.type() == "SparseLengthsSum8BitRowwiseSparse" ||
-       op.type() == "SparseLengthsWeightedSum8BitRowwiseSparse");
-
   if (weight) {
     CAFFE_ENFORCE_GE(
         op.input_size(),
diff --git a/caffe2/opt/optimize_ideep.cc b/caffe2/opt/optimize_ideep.cc
index e12d128b26d375..687126ee327d5d 100644
--- a/caffe2/opt/optimize_ideep.cc
+++ b/caffe2/opt/optimize_ideep.cc
@@ -533,8 +533,6 @@ bool fuseActivation(repr::NNModule* nn, caffe2::Workspace* ws) {
       continue;
     }
     auto relu_node = consumers.front();
-    // NOLINTNEXTLINE(clang-diagnostic-unused-variable,clang-analyzer-deadcode.DeadStores)
-    auto relu = repr::nn::get<repr::Relu>(relu_node);
 
     auto relu_outputs = repr::nn::getOutputs(relu_node);
     if (relu_outputs.size() != 1) {
@@ -893,10 +891,6 @@ void preConvertFiltersFormat(repr::NNModule* nn, caffe2::Workspace* ws) {
       initValue(strides, {1, 1});
       auto pads = convTranspose->getPads();
       initValue(pads, {0, 0, 0, 0});
-      // NOLINTNEXTLINE(clang-diagnostic-unused-variable,clang-analyzer-deadcode.DeadStores)
-      auto* op = getMutableOpDef(*convTranspose);
-      // NOLINTNEXTLINE(clang-diagnostic-unused-variable)
-      auto aalgorithm = ialgo::deconvolution_direct;
       auto dataType = filter->get_data_type();
       ideep::tensor::dims filter_dims_mkldnn{filter->get_dim(1),
                                              filter->get_dim(0),
diff --git a/caffe2/quantization/server/dynamic_histogram.cc b/caffe2/quantization/server/dynamic_histogram.cc
index 25648cbb7fcf7b..c7c7286bd6a1fc 100644
--- a/caffe2/quantization/server/dynamic_histogram.cc
+++ b/caffe2/quantization/server/dynamic_histogram.cc
@@ -64,12 +64,12 @@ void RemapHistograms(Histogram& src_hist, Histogram& dst_hist) {
     // NOLINTNEXTLINE(bugprone-narrowing-conversions,cppcoreguidelines-narrowing-conversions)
     float dst_bin_begin = dst_hist.Min() + dst_bin_width * dst_bin;
     float dst_bin_end = dst_bin_begin + dst_bin_width;
-    // NOLINTNEXTLINE(clang-diagnostic-unused-variable,clang-analyzer-deadcode.DeadStores)
     int dst_bin2 =
         // NOLINTNEXTLINE(bugprone-narrowing-conversions,cppcoreguidelines-narrowing-conversions)
         dst_bin_width == 0 ? 0 : (src_bin_end - dst_hist.Min()) / dst_bin_width;
     // 1 src_bin is mapped to at most 2 dst bin
     assert(dst_bin2 <= dst_bin + 2);
+    (void)dst_bin2;
 
     // dst_bin_cnt is the count from src_bin that should go to dst_bin
     // The remainder should go to dst_bin2
diff --git a/caffe2/quantization/server/fbgemm_pack_op.cc b/caffe2/quantization/server/fbgemm_pack_op.cc
index 22fa8a6fc790dc..99256fc630d318 100644
--- a/caffe2/quantization/server/fbgemm_pack_op.cc
+++ b/caffe2/quantization/server/fbgemm_pack_op.cc
@@ -698,9 +698,8 @@ TypeIdentifier Int8ConvDNNLowpPackedWeightBlobShapeFunctions::GetTypeMetaId() {
 
 TypeMeta Int8FCDNNLowpPackedWeightBlobShapeFunctions::GetExternalTensorType(
     const void* c) {
-  // NOLINTNEXTLINE(clang-diagnostic-unused-variable)
-  const Int8FCDNNLowPPackedWeightBlob* int8_tensor =
-      reinterpret_cast<const Int8FCDNNLowPPackedWeightBlob*>(c);
+  // const Int8FCDNNLowPPackedWeightBlob* int8_tensor =
+  //     reinterpret_cast<const Int8FCDNNLowPPackedWeightBlob*>(c);
   // We forced the output type to be uint8_t since we know it always is.
   // If it is going to be implemented elsewhere, we might need to change here.
   // return (int8_tensor->original_tensor).dtype();
@@ -709,9 +708,8 @@ TypeMeta Int8FCDNNLowpPackedWeightBlobShapeFunctions::GetExternalTensorType(
 
 TypeMeta Int8ConvDNNLowpPackedWeightBlobShapeFunctions::GetExternalTensorType(
     const void* c) {
-  // NOLINTNEXTLINE(clang-diagnostic-unused-variable)
-  const Int8ConvDNNLowPPackedWeightBlob* int8_tensor =
-      reinterpret_cast<const Int8ConvDNNLowPPackedWeightBlob*>(c);
+  // const Int8ConvDNNLowPPackedWeightBlob* int8_tensor =
+  //     reinterpret_cast<const Int8ConvDNNLowPPackedWeightBlob*>(c);
   // return (int8_tensor->original_tensor).dtype();
   return TypeMeta::Make<uint8_t>();
 }
diff --git a/caffe2/quantization/server/p99.cc b/caffe2/quantization/server/p99.cc
index 85f4c3567b15a9..22a7e2230b90c9 100644
--- a/caffe2/quantization/server/p99.cc
+++ b/caffe2/quantization/server/p99.cc
@@ -21,11 +21,7 @@ TensorQuantizationParams P99::ChooseQuantizationParams(
   float org_min = min;
   // NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
   float bin_width = (max - min) / nbins;
-  // NOLINTNEXTLINE(clang-diagnostic-unused-variable,bugprone-narrowing-conversions,cppcoreguidelines-narrowing-conversions,clang-analyzer-deadcode.DeadStores)
-  int zero_bin = round(-min / bin_width);
 
-  // NOLINTNEXTLINE(clang-diagnostic-unused-variable)
-  int best_width = 0;
   double total_sum = 0;
   for (int i = 0; i < nbins; ++i) {
     total_sum += bins_f[i];
diff --git a/caffe2/share/contrib/nnpack/conv_op.cc b/caffe2/share/contrib/nnpack/conv_op.cc
index bfc5ec16544150..7e429774c4c893 100644
--- a/caffe2/share/contrib/nnpack/conv_op.cc
+++ b/caffe2/share/contrib/nnpack/conv_op.cc
@@ -152,8 +152,7 @@ bool NNPACKConvOp::RunOnDeviceWithOrderNCHW() {
   auto& filter = Input(1);
   auto* Y = Output(0);
   CAFFE_ENFORCE(X.ndim() == 4, "Input dim should be 4");
-  // NOLINTNEXTLINE(clang-analyzer-deadcode.DeadStores,clang-diagnostic-unused-variable)
-  const int N = X.dim32(0), C = X.dim32(1), H = X.dim32(2), W = X.dim32(3);
+  const int C = X.dim32(1), H = X.dim32(2), W = X.dim32(3);
   CAFFE_ENFORCE(filter.ndim() == 4, "");
   const int M = filter.dim32(0);
   CAFFE_ENFORCE(C % this->group_ == 0, "");
@@ -181,12 +180,6 @@ bool NNPACKConvOp::RunOnDeviceWithOrderNCHW() {
     biasData = dummyBias_.data();
   }
 
-  // NOLINTNEXTLINE(clang-analyzer-deadcode.DeadStores,clang-diagnostic-unused-variable)
-  const size_t batch_size = X.dim32(0);
-  // NOLINTNEXTLINE(clang-analyzer-deadcode.DeadStores,clang-diagnostic-unused-variable)
-  const size_t input_channels = X.dim32(1);
-  // NOLINTNEXTLINE(clang-analyzer-deadcode.DeadStores,clang-diagnostic-unused-variable)
-  const size_t output_channels = Y->dim32(1);
   const nnp_size input_size = {.width = static_cast<size_t>(X.dim32(3)),
                                .height = static_cast<size_t>(X.dim32(2))};
   // filter is MCHW
diff --git a/test/benchmark_utils/test_benchmark_utils.py b/test/benchmark_utils/test_benchmark_utils.py
index 835fc89d033c17..58d8543db12f20 100644
--- a/test/benchmark_utils/test_benchmark_utils.py
+++ b/test/benchmark_utils/test_benchmark_utils.py
@@ -170,7 +170,7 @@ def test_timer(self):
     @unittest.skipIf(IS_SANDCASTLE, "C++ timing is OSS only.")
     def test_timer_tiny_fast_snippet(self):
         timer = benchmark_utils.Timer(
-            'auto x = 1;',
+            'auto x = 1;(void)x;',
             timer=timeit.default_timer,
             language=benchmark_utils.Language.CPP,
         )
diff --git a/test/cpp/api/CMakeLists.txt b/test/cpp/api/CMakeLists.txt
index f1fb4b6123a0f1..373895a0ce44d3 100644
--- a/test/cpp/api/CMakeLists.txt
+++ b/test/cpp/api/CMakeLists.txt
@@ -53,6 +53,9 @@ endif()
 add_executable(test_api ${TORCH_API_TEST_SOURCES})
 target_include_directories(test_api PRIVATE ${ATen_CPU_INCLUDE})
 target_link_libraries(test_api PRIVATE torch gtest)
+if(NOT MSVC)
+  target_compile_options(test_api PRIVATE -Wno-unused-variable)
+endif()
 
 if(USE_CUDA)
   target_link_libraries(test_api PRIVATE
diff --git a/test/cpp/jit/CMakeLists.txt b/test/cpp/jit/CMakeLists.txt
index d398e783194f6a..81ddaad8d1b108 100644
--- a/test/cpp/jit/CMakeLists.txt
+++ b/test/cpp/jit/CMakeLists.txt
@@ -97,6 +97,9 @@ endif(MSVC)
 
 target_link_libraries(test_jit PRIVATE ${JIT_TEST_DEPENDENCIES})
 target_include_directories(test_jit PRIVATE ${ATen_CPU_INCLUDE})
+if(NOT MSVC)
+  target_compile_options(test_jit PRIVATE -Wno-unused-variable)
+endif()
 
 if(LINUX)
   #Update to target_link_options when CMake version can be upgraded
diff --git a/test/cpp/jit/test_backend_compiler_lib.cpp b/test/cpp/jit/test_backend_compiler_lib.cpp
index 2aa6460cb8a33d..63aa4ecb6f7e0e 100644
--- a/test/cpp/jit/test_backend_compiler_lib.cpp
+++ b/test/cpp/jit/test_backend_compiler_lib.cpp
@@ -89,7 +89,6 @@ class BackendWithCompiler : public PyTorchBackendInterface {
     at::Tensor h = val1.toTensor();
 
     c10::List<at::Tensor> output_list;
-    double scalar_val = 1.0;
     for (const auto& token : handle.toList()) {
       IValue val = token;
       auto instruction = val.toTuple()->elements()[0].toStringRef();
diff --git a/test/cpp/tensorexpr/CMakeLists.txt b/test/cpp/tensorexpr/CMakeLists.txt
index 04dd9a133f3354..2a467e848d6a98 100644
--- a/test/cpp/tensorexpr/CMakeLists.txt
+++ b/test/cpp/tensorexpr/CMakeLists.txt
@@ -38,6 +38,9 @@ add_executable(test_tensorexpr
 target_link_libraries(test_tensorexpr PRIVATE torch gtest)
 target_include_directories(test_tensorexpr PRIVATE ${ATen_CPU_INCLUDE})
 target_compile_definitions(test_tensorexpr PRIVATE USE_GTEST)
+if(NOT MSVC)
+  target_compile_options(test_tensorexpr PRIVATE -Wno-unused-variable)
+endif()
 
 add_executable(tutorial_tensorexpr ${TENSOREXPR_TEST_ROOT}/tutorial.cpp)
 target_link_libraries(tutorial_tensorexpr PRIVATE torch)
diff --git a/torch/CMakeLists.txt b/torch/CMakeLists.txt
index 38baee4799b548..4d53eb080a072a 100644
--- a/torch/CMakeLists.txt
+++ b/torch/CMakeLists.txt
@@ -361,6 +361,10 @@ if(USE_PRECOMPILED_HEADERS)
       "$<$<COMPILE_LANGUAGE:CXX>:ATen/ATen.h>")
 endif()
 
+if(NOT MSVC)
+  target_compile_options(torch_python PRIVATE -Wno-unused-variable)
+endif()
+
 # Required workaround for generated sources
 # See https://samthursfield.wordpress.com/2015/11/21/cmake-dependencies-between-targets-and-files-and-custom-commands/#custom-commands-in-different-directories
 add_dependencies(torch_python generate-torch-sources)
diff --git a/torch/csrc/deploy/loader.cpp b/torch/csrc/deploy/loader.cpp
index cbf093d2bb8a0b..a95cf6a272cc51 100644
--- a/torch/csrc/deploy/loader.cpp
+++ b/torch/csrc/deploy/loader.cpp
@@ -1046,7 +1046,6 @@ struct __attribute__((visibility("hidden"))) CustomLibraryImpl
   }
 
   at::optional<TLSIndex> tls_lookup_symbol(Elf64_Xword r_info) {
-    const uint32_t r_type = ELF64_R_TYPE(r_info);
     const uint32_t r_sym = ELF64_R_SYM(r_info);
 
     if (r_sym == 0) {
diff --git a/torch/csrc/deploy/test_deploy.cpp b/torch/csrc/deploy/test_deploy.cpp
index 1eb2e74d467526..e0ad7bc3d550ea 100644
--- a/torch/csrc/deploy/test_deploy.cpp
+++ b/torch/csrc/deploy/test_deploy.cpp
@@ -107,8 +107,8 @@ TEST(TorchpyTest, MultiSerialSimpleModel) {
   size_t ninterp = 3;
   std::vector<at::Tensor> outputs;
 
-  // NOLINTNEXTLINE(clang-analyzer-deadcode.DeadStores)
   for (const auto i : c10::irange(ninterp)) {
+    (void)i;
     outputs.push_back(model({input.alias()}).toTensor());
   }
 
@@ -151,11 +151,12 @@ TEST(TorchpyTest, ThreadedSimpleModel) {
   std::vector<at::Tensor> outputs;
 
   std::vector<std::future<at::Tensor>> futures;
-  // NOLINTNEXTLINE(clang-analyzer-deadcode.DeadStores)
   for (const auto i : c10::irange(nthreads)) {
+    (void)i;
     futures.push_back(std::async(std::launch::async, [&model]() {
       auto input = torch::ones({10, 20});
-      for (const auto i : c10::irange(100)) {
+      for (const auto j : c10::irange(100)) {
+        (void)j;
         model({input.alias()}).toTensor();
       }
       auto result = model({input.alias()}).toTensor();
@@ -230,8 +231,8 @@ TEST(TorchpyTest, TaggingRace) {
   constexpr int64_t trials = 4;
   constexpr int64_t nthreads = 16;
   torch::deploy::InterpreterManager m(nthreads);
-  // NOLINTNEXTLINE(clang-analyzer-deadcode.DeadStores)
   for (const auto n : c10::irange(trials)) {
+    (void)n;
     at::Tensor t = torch::empty(2);
     std::atomic<int64_t> success(0);
     std::atomic<int64_t> failed(0);
@@ -283,6 +284,7 @@ TEST(TorchpyTest, FxModule) {
   std::vector<at::Tensor> outputs;
   auto input = torch::ones({5, 10});
   for (const auto i : c10::irange(nthreads)) {
+    (void)i;
     outputs.push_back(model({input.alias()}).toTensor());
   }
 
diff --git a/torch/csrc/distributed/c10d/ProcessGroupMPI.cpp b/torch/csrc/distributed/c10d/ProcessGroupMPI.cpp
index b75f4417e832aa..714f3a84deb61f 100644
--- a/torch/csrc/distributed/c10d/ProcessGroupMPI.cpp
+++ b/torch/csrc/distributed/c10d/ProcessGroupMPI.cpp
@@ -270,6 +270,7 @@ c10::intrusive_ptr<ProcessGroupMPI> ProcessGroupMPI::createProcessGroupMPI(
       bool groupComm_updated = false;
       MPI_Barrier(MPI_COMM_WORLD);
       for (const auto i : c10::irange(kMaxNumRetries)) {
+        (void)i;
         if (MPI_Comm_create(MPI_COMM_WORLD, ranksGroup, &groupComm)) {
           groupComm_updated = true;
           break;
diff --git a/torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp b/torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp
index 57eecc6bd81a15..2d31d25658dfef 100644
--- a/torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp
+++ b/torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp
@@ -939,6 +939,7 @@ std::vector<std::shared_ptr<NCCLComm>>& ProcessGroupNCCL::getNCCLComm(
   // created before encountering any communication calls. This is why we need
   // the following for loop.
   for (const auto i : c10::irange(ncclActiveGroupCounter_)) {
+    (void)i;
     C10D_NCCL_CHECK(ncclGroupEnd(), c10::nullopt);
   }
 
@@ -978,6 +979,7 @@ std::vector<std::shared_ptr<NCCLComm>>& ProcessGroupNCCL::getNCCLComm(
 
   // See [Group Start/End Note]
   for (const auto i : c10::irange(ncclActiveGroupCounter_)) {
+    (void)i;
     C10D_NCCL_CHECK(ncclGroupStart(), c10::nullopt);
   }
 
diff --git a/torch/csrc/distributed/c10d/reducer.cpp b/torch/csrc/distributed/c10d/reducer.cpp
index 1d54f365620561..4615174c8d79e8 100644
--- a/torch/csrc/distributed/c10d/reducer.cpp
+++ b/torch/csrc/distributed/c10d/reducer.cpp
@@ -1383,8 +1383,6 @@ void Reducer::finalize_bucket_dense(Bucket& bucket) {
   auto& replica = bucket.replicas[replica_index];
   for (const auto intra_bucket_index : c10::irange(replica.variables.size())) {
     auto& variable = replica.variables[intra_bucket_index];
-    const auto offset = replica.offsets[intra_bucket_index];
-    const auto length = replica.lengths[intra_bucket_index];
 
     bool global_unused = false;
     // See Note [Skip allreducing local_used_map_dev]
@@ -1634,6 +1632,7 @@ void Reducer::sync_bucket_indices(
     std::vector<size_t> bucket;
     bucket.reserve(bucket_size);
     for (const auto j : c10::irange(bucket_size)) {
+      (void)j;
       bucket.push_back(indices_accessor[indices_accessor_Index++]);
     }
     bucket_indices.emplace_back(std::move(bucket));
diff --git a/torch/csrc/jit/codegen/cuda/lower_thread_predicate.cpp b/torch/csrc/jit/codegen/cuda/lower_thread_predicate.cpp
index d8a6c7f79d9b0b..fe72baf085133f 100644
--- a/torch/csrc/jit/codegen/cuda/lower_thread_predicate.cpp
+++ b/torch/csrc/jit/codegen/cuda/lower_thread_predicate.cpp
@@ -280,7 +280,7 @@ void ThreadPredicateMap::insert(
 kir::Bool* ThreadPredicateMap::getPredicate(const TensorView* tv) const {
   // No thread predicate is needed when tv is an output of a
   // parallel broadcast expression.
-  if (auto bop = dynamic_cast<BroadcastOp*>(tv->definition())) {
+  if (dynamic_cast<BroadcastOp*>(tv->definition())) {
     if (getParallelBroadcastDomains(tv).any()) {
       return kir::IrBuilder(GpuLower::current()->kernel()).trueVal();
     }
diff --git a/torch/csrc/jit/frontend/ir_emitter.cpp b/torch/csrc/jit/frontend/ir_emitter.cpp
index 4c066c27c3b3dc..52ad5fd47d5bcb 100644
--- a/torch/csrc/jit/frontend/ir_emitter.cpp
+++ b/torch/csrc/jit/frontend/ir_emitter.cpp
@@ -5163,7 +5163,7 @@ std::unique_ptr<Function> CompilationUnit::define(
   if (shouldMangle) {
     // If `shouldMangle` is set, we should generate a unique name for this
     // function if there is already an existing one.
-    if (auto fn = find_function(name)) {
+    if (find_function(name)) {
       name = mangle(name);
     }
   }
diff --git a/torch/csrc/jit/frontend/sugared_value.cpp b/torch/csrc/jit/frontend/sugared_value.cpp
index a5f000769badc5..dbb4b55e33bb6e 100644
--- a/torch/csrc/jit/frontend/sugared_value.cpp
+++ b/torch/csrc/jit/frontend/sugared_value.cpp
@@ -153,7 +153,7 @@ std::shared_ptr<SugaredValue> SimpleValue::attr(
     }
   } else if (auto classType = value_->type()->cast<ClassType>()) {
     // This is a class, emit the proper attribute lookup
-    if (auto method = classType->findMethod(field)) {
+    if (classType->findMethod(field)) {
       return std::make_shared<MethodValue>(getValue(), field);
     }
     if (classType->hasAttribute(field)) {
@@ -169,7 +169,7 @@ std::shared_ptr<SugaredValue> SimpleValue::attr(
     }
   } else if (auto iface = value_->type()->cast<InterfaceType>()) {
     // accessing methods of interfaces
-    if (auto schema = iface->getMethod(field)) {
+    if (iface->getMethod(field)) {
       return std::make_shared<MethodValue>(getValue(), field);
     }
   } else if (auto enum_type = value_->type()->cast<EnumType>()) {
diff --git a/torch/csrc/jit/mobile/promoted_prim_ops.cpp b/torch/csrc/jit/mobile/promoted_prim_ops.cpp
index c96f09045c2d1c..5506e4debb3880 100644
--- a/torch/csrc/jit/mobile/promoted_prim_ops.cpp
+++ b/torch/csrc/jit/mobile/promoted_prim_ops.cpp
@@ -166,7 +166,7 @@ void numToTensorBool(Stack& stack) {
   push(stack, at::scalar_to_tensor(b));
 }
 
-static const std::array<mobile::prim_op_fn_register, 14> op_reg = {
+static const C10_UNUSED std::array<mobile::prim_op_fn_register, 14> op_reg = {
     mobile::prim_op_fn_register("prim::TupleIndex", tupleIndex),
     mobile::prim_op_fn_register("aten::Bool.Tensor", boolTensor),
     mobile::prim_op_fn_register("aten::format", aten_format),
diff --git a/torch/csrc/jit/passes/pass_manager.cpp b/torch/csrc/jit/passes/pass_manager.cpp
index cbd6daa34205db..ed8fe405978b58 100644
--- a/torch/csrc/jit/passes/pass_manager.cpp
+++ b/torch/csrc/jit/passes/pass_manager.cpp
@@ -3,6 +3,9 @@
 namespace torch {
 namespace jit {
 
+// Start UUID at 1
+static GraphPassNameType graphPassID = 1;
+
 std::vector<GraphPassEntry>& getCustomPostPasses() {
   static std::vector<GraphPassEntry> passes;
   return passes;
diff --git a/torch/csrc/jit/passes/pass_manager.h b/torch/csrc/jit/passes/pass_manager.h
index e9aeb09e8c0cce..8585c6ecdb3de5 100644
--- a/torch/csrc/jit/passes/pass_manager.h
+++ b/torch/csrc/jit/passes/pass_manager.h
@@ -26,8 +26,6 @@ using GraphPass = std::function<void(std::shared_ptr<Graph>&)>;
 // Since Passes are std::functions, we associate a UUID to each pass, this way
 // if we want to deregister a pass, we have something to reference it by.
 using GraphPassNameType = unsigned int;
-// Start UUID at 1
-static GraphPassNameType graphPassID = 1;
 
 // Graph pass entries have a name associated with them
 using GraphPassEntry = std::pair<GraphPass, GraphPassNameType>;
diff --git a/torch/csrc/jit/runtime/graph_executor.cpp b/torch/csrc/jit/runtime/graph_executor.cpp
index 39742c7815d3bc..21d2b714591b3c 100644
--- a/torch/csrc/jit/runtime/graph_executor.cpp
+++ b/torch/csrc/jit/runtime/graph_executor.cpp
@@ -158,8 +158,8 @@ struct CaptureList {
         case CAPTURE_LIST: {
           c10::List<at::Tensor> lst;
           auto size = *size_it++;
-          // NOLINTNEXTLINE(clang-analyzer-deadcode.DeadStores,clang-diagnostic-unused-variable)
           for (const auto i : c10::irange(size)) {
+            (void)i;
             lst.emplace_back(var_capture_it->unpack(saved_for));
             var_capture_it++;
           }
diff --git a/torch/csrc/jit/runtime/register_c10_ops.cpp b/torch/csrc/jit/runtime/register_c10_ops.cpp
index 4d541ec46bbbf1..0c8956f4bdf3e8 100644
--- a/torch/csrc/jit/runtime/register_c10_ops.cpp
+++ b/torch/csrc/jit/runtime/register_c10_ops.cpp
@@ -54,7 +54,7 @@ Registerer& registerer() {
 }
 
 // global instance to run its constructor on startup
-Registerer& dummy = registerer();
+C10_UNUSED Registerer& dummy = registerer();
 
 } // namespace
 
diff --git a/torch/csrc/jit/runtime/script_profile.cpp b/torch/csrc/jit/runtime/script_profile.cpp
index 3f43de788ad730..c393dcde1e016b 100644
--- a/torch/csrc/jit/runtime/script_profile.cpp
+++ b/torch/csrc/jit/runtime/script_profile.cpp
@@ -103,7 +103,7 @@ auto initBindings() {
   return nullptr;
 }
 
-const auto torchBindInitializer = initBindings();
+const auto C10_UNUSED torchBindInitializer = initBindings();
 
 } // namespace
 
diff --git a/torch/csrc/jit/tensorexpr/external_functions.cpp b/torch/csrc/jit/tensorexpr/external_functions.cpp
index 810fee1c8da873..a77b0a40186e06 100644
--- a/torch/csrc/jit/tensorexpr/external_functions.cpp
+++ b/torch/csrc/jit/tensorexpr/external_functions.cpp
@@ -25,8 +25,8 @@ std::vector<at::Tensor> constructTensors(
   for (const auto i : c10::irange(bufs_num)) {
     buf_data_vec.push_back(buf_data[i]);
     buf_dims_vec.emplace_back();
-    // NOLINTNEXTLINE(clang-diagnostic-unused-variable,clang-analyzer-deadcode.DeadStores)
     for (const auto dim : c10::irange(buf_ranks[i])) {
+      (void)dim;
       buf_dims_vec[i].push_back(buf_dims[buf_dims_idx++]);
     }
     buf_dtypes_vec.push_back(static_cast<c10::ScalarType>(buf_dtypes[i]));
diff --git a/torch/csrc/jit/tensorexpr/llvm_codegen.cpp b/torch/csrc/jit/tensorexpr/llvm_codegen.cpp
index 690edb997bfe76..9e3631deffaf69 100644
--- a/torch/csrc/jit/tensorexpr/llvm_codegen.cpp
+++ b/torch/csrc/jit/tensorexpr/llvm_codegen.cpp
@@ -1151,9 +1151,6 @@ void LLVMCodeGenImpl::visit(LoadPtr v) {
       throw std::runtime_error("invalid dtype in Load");
   }
 
-  // Detect whether the vector mask is all true
-  bool unmasked_load = true;
-
   // Handle the case where the load is contiguous and unmasked efficiently
   auto idx_ramp = to<Ramp>(v->flat_index());
   if (idx_ramp) {
@@ -1805,9 +1802,6 @@ void LLVMCodeGenImpl::visit(IntrinsicsPtr v) {
 }
 
 void LLVMCodeGenImpl::visit(ExternalCallPtr v) {
-  constexpr int max_buffers = 10;
-  constexpr int max_dimensions = 40;
-
   auto& func_registry = getNNCFunctionRegistry();
   if (!func_registry.count(v->func_name())) {
     throw unimplemented_lowering(v);
diff --git a/torch/csrc/onnx/onnx.h b/torch/csrc/onnx/onnx.h
index 4f6860e6026339..d01d8f5e804288 100644
--- a/torch/csrc/onnx/onnx.h
+++ b/torch/csrc/onnx/onnx.h
@@ -19,7 +19,7 @@ enum class TrainingMode {
 // We pin IR version instead of using onnx::IR_VERSION so that the
 // test_operators.py will be more stable. Only bump it when
 // necessary.
-static const size_t IR_VERSION = 7;
-static const char* PRODUCER_VERSION = "1.11";
+constexpr size_t IR_VERSION = 7;
+constexpr const char* PRODUCER_VERSION = "1.11";
 } // namespace onnx
 } // namespace torch
diff --git a/torch/csrc/utils/byte_order.cpp b/torch/csrc/utils/byte_order.cpp
index 37dd9ac9b42d10..d720ce689e152f 100644
--- a/torch/csrc/utils/byte_order.cpp
+++ b/torch/csrc/utils/byte_order.cpp
@@ -236,8 +236,8 @@ void THP_encodeInt16Buffer(uint8_t* dst, const int16_t* src, THPByteOrder order,
 {
   memcpy(dst, src, sizeof(int16_t) * len);
   if (order != THP_nativeByteOrder()) {
-    // NOLINTNEXTLINE(clang-analyzer-deadcode.DeadStores,clang-diagnostic-unused-variable)
     for(const auto i : c10::irange(len)) {
+      (void)i;
       swapBytes16(dst);
       dst += sizeof(int16_t);
     }
@@ -248,8 +248,8 @@ void THP_encodeInt32Buffer(uint8_t* dst, const int32_t* src, THPByteOrder order,
 {
   memcpy(dst, src, sizeof(int32_t) * len);
   if (order != THP_nativeByteOrder()) {
-    // NOLINTNEXTLINE(clang-analyzer-deadcode.DeadStores,clang-diagnostic-unused-variable)
     for(const auto i : c10::irange(len)) {
+      (void)i;
       swapBytes32(dst);
       dst += sizeof(int32_t);
     }
@@ -260,8 +260,8 @@ void THP_encodeInt64Buffer(uint8_t* dst, const int64_t* src, THPByteOrder order,
 {
   memcpy(dst, src, sizeof(int64_t) * len);
   if (order != THP_nativeByteOrder()) {
-    // NOLINTNEXTLINE(clang-analyzer-deadcode.DeadStores,clang-diagnostic-unused-variable)
     for(const auto i : c10::irange(len)) {
+      (void)i;
       swapBytes64(dst);
       dst += sizeof(int64_t);
     }
@@ -272,8 +272,8 @@ void THP_encodeFloatBuffer(uint8_t* dst, const float* src, THPByteOrder order, s
 {
   memcpy(dst, src, sizeof(float) * len);
   if (order != THP_nativeByteOrder()) {
-    // NOLINTNEXTLINE(clang-analyzer-deadcode.DeadStores,clang-diagnostic-unused-variable)
     for(const auto i : c10::irange(len)) {
+      (void)i;
       swapBytes32(dst);
       dst += sizeof(float);
     }
@@ -284,8 +284,8 @@ void THP_encodeDoubleBuffer(uint8_t* dst, const double* src, THPByteOrder order,
 {
   memcpy(dst, src, sizeof(double) * len);
   if (order != THP_nativeByteOrder()) {
-    // NOLINTNEXTLINE(clang-analyzer-deadcode.DeadStores,clang-diagnostic-unused-variable)
     for(const auto i : c10::irange(len)) {
+      (void)i;
       swapBytes64(dst);
       dst += sizeof(double);
     }
diff --git a/torch/csrc/utils/throughput_benchmark-inl.h b/torch/csrc/utils/throughput_benchmark-inl.h
index f49cb06d95ef3b..7a11d6f11a3b60 100644
--- a/torch/csrc/utils/throughput_benchmark-inl.h
+++ b/torch/csrc/utils/throughput_benchmark-inl.h
@@ -64,6 +64,7 @@ BenchmarkExecutionStats BenchmarkHelper<Input, Output, Model>::benchmark(
       // We use conditional variable as a barrier to make sure each thread
       // performs required warmeup iterations before we start measuring
       for (const auto j : c10::irange(config.num_warmup_iters)) {
+        (void)j;
         runOnce(std::move(thread_inputs[thread_id][input_iters[thread_id]]));
         ++input_iters[thread_id];
       }
diff --git a/torch/utils/benchmark/utils/timeit_template.cpp b/torch/utils/benchmark/utils/timeit_template.cpp
index 2a57236f2685af..6396d410dec4a5 100644
--- a/torch/utils/benchmark/utils/timeit_template.cpp
+++ b/torch/utils/benchmark/utils/timeit_template.cpp
@@ -31,6 +31,7 @@ double timeit(int n) {
     // Main loop
     auto start_time = std::chrono::high_resolution_clock::now();
     for(const auto loop_idx : c10::irange(n)) {
+        (void)loop_idx;
         // STMT_TEMPLATE_LOCATION
     }
     auto end_time = std::chrono::high_resolution_clock::now();
diff --git a/torch/utils/benchmark/utils/valgrind_wrapper/timer_callgrind_template.cpp b/torch/utils/benchmark/utils/valgrind_wrapper/timer_callgrind_template.cpp
index 3a11135638b7f9..1c6bd139a43e4e 100644
--- a/torch/utils/benchmark/utils/valgrind_wrapper/timer_callgrind_template.cpp
+++ b/torch/utils/benchmark/utils/valgrind_wrapper/timer_callgrind_template.cpp
@@ -44,14 +44,17 @@ int main(int argc, char* argv[]) {
 
     // Warmup
     for(const auto i : c10::irange(number_warmup)) {
+        (void)i;
         // STMT_TEMPLATE_LOCATION
     }
 
     // Main loop
     for(const auto repeat : c10::irange(repeats)) {
+        (void)repeat;
         CALLGRIND_TOGGLE_COLLECT;
 
         for(const auto i : c10::irange(number)) {
+        (void)i;
         // STMT_TEMPLATE_LOCATION
         }