diff --git a/.clang-format b/.clang-format
index 0b94540e7a25..2e5161504103 100644
--- a/.clang-format
+++ b/.clang-format
@@ -106,6 +106,8 @@ StatementMacros:
   - C10_DEFINE_int32
   - C10_DEFINE_int64
   - C10_DEFINE_string
+  - C10_DEFINE_REGISTRY_WITHOUT_WARNING
+  - C10_REGISTER_CREATOR
   - DEFINE_BINARY
   - PyObject_HEAD
   - PyObject_VAR_HEAD
diff --git a/CMakeLists.txt b/CMakeLists.txt
index c3ba29a50092..b43137523cfe 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -1057,7 +1057,6 @@ if(NOT MSVC)
   append_cxx_flag_if_supported("-Wconstant-conversion" CMAKE_CXX_FLAGS)
   append_cxx_flag_if_supported("-Wno-aligned-allocation-unavailable"
                                CMAKE_CXX_FLAGS)
-  append_cxx_flag_if_supported("-Wno-missing-braces" CMAKE_CXX_FLAGS)
   append_cxx_flag_if_supported("-Qunused-arguments" CMAKE_CXX_FLAGS)
 
   if(${USE_COLORIZE_OUTPUT})
diff --git a/aten/src/ATen/native/QuantizedLinear.cpp b/aten/src/ATen/native/QuantizedLinear.cpp
index e2f3f06f64c8..6d6c6909fe32 100644
--- a/aten/src/ATen/native/QuantizedLinear.cpp
+++ b/aten/src/ATen/native/QuantizedLinear.cpp
@@ -1,5 +1,4 @@
 #define TORCH_ASSERT_ONLY_METHOD_OPERATORS
-#include <vector>
 
 #include <ATen/core/Tensor.h>
 #include <ATen/Parallel.h>
@@ -116,7 +115,7 @@ Tensor fbgemm_linear_int8_weight_fp32_activation(
   const Tensor bias_contig = bias.contiguous();
 
   // Allocate output Tensor and a buffer for fbgemmPacked to use
-  std::vector<int64_t> output_size = input.sizes().vec();
+  auto output_size = input.sizes().vec();
   output_size.back() = N;
   Tensor output = at::empty(output_size, input.options().dtype(at::kFloat), LEGACY_CONTIGUOUS_MEMORY_FORMAT);
   Tensor buffer = at::empty(output_size, input.options().dtype(at::kInt), LEGACY_CONTIGUOUS_MEMORY_FORMAT);
diff --git a/aten/src/ATen/native/RNN.cpp b/aten/src/ATen/native/RNN.cpp
index eee703ce4733..bd503854899c 100644
--- a/aten/src/ATen/native/RNN.cpp
+++ b/aten/src/ATen/native/RNN.cpp
@@ -62,8 +62,6 @@
 #include <utility>
 #endif
 
-int register_linear_params();
-
 namespace at::native {
 
 namespace {
diff --git a/aten/src/ATen/native/quantized/cpu/LinearUnpackImpl.cpp b/aten/src/ATen/native/quantized/cpu/LinearUnpackImpl.cpp
index df74b10d70f9..c9c09cf2464f 100644
--- a/aten/src/ATen/native/quantized/cpu/LinearUnpackImpl.cpp
+++ b/aten/src/ATen/native/quantized/cpu/LinearUnpackImpl.cpp
@@ -19,8 +19,6 @@
 #include <ATen/ops/from_blob.h>
 #endif
 
-int register_linear_params();
-
 #ifdef USE_FBGEMM
 std::tuple<at::Tensor, std::optional<at::Tensor>> PackedLinearWeight::unpack() {
   auto packB = w.get();
diff --git a/aten/src/ATen/native/quantized/cpu/fbgemm_utils.cpp b/aten/src/ATen/native/quantized/cpu/fbgemm_utils.cpp
index fc77d44e18c1..33edbfac1fc4 100644
--- a/aten/src/ATen/native/quantized/cpu/fbgemm_utils.cpp
+++ b/aten/src/ATen/native/quantized/cpu/fbgemm_utils.cpp
@@ -28,7 +28,6 @@
 #include <utility>
 #endif
 
-int register_embedding_params();
 
 #ifdef USE_FBGEMM
 
@@ -381,9 +380,7 @@ namespace {
   }
 }
 
-template <int kSpatialDim = 2>
-TORCH_API int
-register_conv_params() {
+template <int kSpatialDim> int register_conv_params() {
   static auto register_conv_params =
     torch::selective_class_<ConvPackedParamsBase<kSpatialDim>>(
         "quantized", TORCH_SELECTIVE_CLASS(_hack_int_to_class_name(kSpatialDim)))
@@ -420,9 +417,7 @@ TORCH_API int register_conv_params<2>();
 template
 TORCH_API int register_conv_params<3>();
 
-TORCH_API int register_linear_params();
-
-TORCH_API int register_linear_params() {
+int register_linear_params() {
   using SerializationType = std::tuple<at::Tensor, std::optional<at::Tensor>>;
   static auto register_linear_params =
       torch::selective_class_<LinearPackedParamsBase>(
diff --git a/aten/src/ATen/native/quantized/cpu/fbgemm_utils.h b/aten/src/ATen/native/quantized/cpu/fbgemm_utils.h
index 05d63c8476ac..62a21c439b78 100644
--- a/aten/src/ATen/native/quantized/cpu/fbgemm_utils.h
+++ b/aten/src/ATen/native/quantized/cpu/fbgemm_utils.h
@@ -6,7 +6,7 @@
 #include <c10/core/QScheme.h>
 #include <c10/util/irange.h>
 
-#ifdef USE_FBGEMM
+#if defined(USE_FBGEMM) && __has_include(<fbgemm/Fbgemm.h>)
 #include <fbgemm/Fbgemm.h>
 C10_DIAGNOSTIC_PUSH_AND_IGNORED_IF_DEFINED("-Winconsistent-missing-destructor-override")
 #include <fbgemm/FbgemmFP16.h>
@@ -407,3 +407,8 @@ struct TORCH_API PackedEmbeddingBagWeight : public EmbeddingPackedParamsBase {
       bool include_last_offset,
       bool is_embedding_op) override;
 };
+
+TORCH_API int register_linear_params();
+int register_embedding_params();
+
+template <int kSpatialDim = 2> TORCH_API int register_conv_params();
diff --git a/aten/src/ATen/native/quantized/cpu/qlinear.cpp b/aten/src/ATen/native/quantized/cpu/qlinear.cpp
index 27139bcb40cb..05be6ebbf8d4 100644
--- a/aten/src/ATen/native/quantized/cpu/qlinear.cpp
+++ b/aten/src/ATen/native/quantized/cpu/qlinear.cpp
@@ -31,8 +31,6 @@
 #include <algorithm>
 #include <string>
 
-int register_linear_params();
-
 #ifdef USE_FBGEMM
 template <bool ReluFused>
 at::Tensor& PackedLinearWeight::apply_impl(
diff --git a/aten/src/ATen/native/quantized/cpu/qlinear_dynamic.cpp b/aten/src/ATen/native/quantized/cpu/qlinear_dynamic.cpp
index 091e309cd95d..27e1b24d2362 100644
--- a/aten/src/ATen/native/quantized/cpu/qlinear_dynamic.cpp
+++ b/aten/src/ATen/native/quantized/cpu/qlinear_dynamic.cpp
@@ -29,8 +29,6 @@
 #include <string>
 #include <type_traits>
 
-int register_linear_params();
-
 #ifdef USE_FBGEMM
 template <bool ReluFused>
 at::Tensor PackedLinearWeight::apply_dynamic_impl(
diff --git a/aten/src/ATen/native/quantized/cpu/qlinear_prepack.cpp b/aten/src/ATen/native/quantized/cpu/qlinear_prepack.cpp
index 3e4ce69a8994..d40552037057 100644
--- a/aten/src/ATen/native/quantized/cpu/qlinear_prepack.cpp
+++ b/aten/src/ATen/native/quantized/cpu/qlinear_prepack.cpp
@@ -31,8 +31,6 @@
 #include <utility>
 #include <vector>
 
-int register_linear_params();
-
 #ifdef USE_FBGEMM
 namespace {
 // Calculate the column offsets.
diff --git a/aten/src/ATen/native/quantized/cudnn/Conv.cpp b/aten/src/ATen/native/quantized/cudnn/Conv.cpp
index d37f674f5c77..dff887118ee5 100644
--- a/aten/src/ATen/native/quantized/cudnn/Conv.cpp
+++ b/aten/src/ATen/native/quantized/cudnn/Conv.cpp
@@ -9,6 +9,7 @@
 #include <ATen/cuda/Exceptions.h>
 #include <ATen/cudnn/Handle.h>
 #include <ATen/native/cudnn/ConvShared.h>
+#include <ATen/native/quantized/cpu/fbgemm_utils.h>
 #include <ATen/native/quantized/cudnn/utils.h>
 #include <ATen/native/quantized/ConvUtils.h>
 #include <ATen/native/quantized/PackedParams.h>
@@ -22,12 +23,6 @@
 #include <unordered_map>
 #include <vector>
 
-template <int kSpatialDim = 2>
-int register_conv_params();
-
-extern template int register_conv_params<2>();
-extern template int register_conv_params<3>();
-
 // TODO: there is a table from input dtype and weight dtype to operator qdtype,
 // we can derive the operator dtype based on input dtype
 cudnn_frontend::ConvDesc_v8 getConvDescriptor(cudnnDataType_t dataType, c10::IntArrayRef padding, c10::IntArrayRef stride, c10::IntArrayRef dilation) {
diff --git a/aten/src/ATen/native/quantized/cudnn/ConvPrepack.cpp b/aten/src/ATen/native/quantized/cudnn/ConvPrepack.cpp
index 9103bdd0d414..da46dccab316 100644
--- a/aten/src/ATen/native/quantized/cudnn/ConvPrepack.cpp
+++ b/aten/src/ATen/native/quantized/cudnn/ConvPrepack.cpp
@@ -5,6 +5,7 @@
 
 #include <ATen/ATen.h>
 #include <torch/library.h>
+#include <ATen/native/quantized/cpu/fbgemm_utils.h>
 #include <ATen/native/quantized/cpu/QuantUtils.h>
 #include <ATen/native/quantized/cudnn/utils.h>
 #include <ATen/native/quantized/PackedParams.h>
@@ -15,12 +16,6 @@
 
 #include <utility>
 
-template <int kSpatialDim = 2>
-int register_conv_params();
-
-extern template int register_conv_params<2>();
-extern template int register_conv_params<3>();
-
 template <int kSpatialDim>
 c10::intrusive_ptr<ConvPackedParamsBase<kSpatialDim>> PackedConvWeightCudnn<
     kSpatialDim>::
diff --git a/aten/src/ATen/native/quantized/library.cpp b/aten/src/ATen/native/quantized/library.cpp
index 72dcda2b74de..7d2ed9f42e04 100644
--- a/aten/src/ATen/native/quantized/library.cpp
+++ b/aten/src/ATen/native/quantized/library.cpp
@@ -1,13 +1,5 @@
 #include <torch/library.h>
-
-int register_linear_params();
-
-template <int kSpatialDim = 2>
-int register_conv_params();
-
-extern template int register_conv_params<2>();
-extern template int register_conv_params<3>();
-int register_embedding_params();
+#include <aten/src/ATen/native/quantized/cpu/fbgemm_utils.h>
 
 TORCH_LIBRARY(quantized, m) {
   m.set_python_module("caffe2.torch.fb.model_transform.splitting.split_dispatcher");
diff --git a/aten/src/ATen/native/quantized/qconv_unpack.cpp b/aten/src/ATen/native/quantized/qconv_unpack.cpp
index f33bd6cf96fe..f613097fdba1 100644
--- a/aten/src/ATen/native/quantized/qconv_unpack.cpp
+++ b/aten/src/ATen/native/quantized/qconv_unpack.cpp
@@ -28,13 +28,6 @@ and /cudnn/ConvUnpackImpl.cpp, for cudnn.
 #include <ATen/ops/from_blob.h>
 #endif
 
-template <int kSpatialDim = 2>
-int register_conv_params();
-
-extern template int register_conv_params<2>();
-extern template int register_conv_params<3>();
-
-
 
 namespace at::native {
 namespace {
diff --git a/aten/src/ATen/native/quantized/qlinear_unpack.cpp b/aten/src/ATen/native/quantized/qlinear_unpack.cpp
index a485094df3ca..2999e183b9ef 100644
--- a/aten/src/ATen/native/quantized/qlinear_unpack.cpp
+++ b/aten/src/ATen/native/quantized/qlinear_unpack.cpp
@@ -13,9 +13,6 @@ and /cudnn/linear_unpack_impl.cpp, for cudnn.
 #include <torch/custom_class.h>
 #include <torch/library.h>
 
-int register_linear_params();
-
-
 namespace at::native {
 namespace {
 
diff --git a/torch/csrc/distributed/c10d/GlooDeviceFactory.cpp b/torch/csrc/distributed/c10d/GlooDeviceFactory.cpp
index 47a9a02ae810..478b73e09938 100644
--- a/torch/csrc/distributed/c10d/GlooDeviceFactory.cpp
+++ b/torch/csrc/distributed/c10d/GlooDeviceFactory.cpp
@@ -96,7 +96,7 @@ static std::shared_ptr<::gloo::transport::Device> makeTCPTLSDevice(
       attr, pkey, cert, caFile, caPath);
 }
 
-C10_REGISTER_CREATOR(GlooDeviceRegistry, TCP_TLS, makeTCPTLSDevice);
+C10_REGISTER_CREATOR(GlooDeviceRegistry, TCP_TLS, makeTCPTLSDevice)
 #endif
 
 #if GLOO_HAVE_TRANSPORT_UV
@@ -120,9 +120,9 @@ static std::shared_ptr<::gloo::transport::Device> makeUVDevice(
 // Registry priority is per key identifier. We register UV to `APPLE` for
 // the flexibility of other application to override by priority. Register
 // UV to `UV` for env "GLOO_DEVICE_TRANSPORT" override.
-C10_REGISTER_CREATOR(GlooDeviceRegistry, APPLE, makeUVDevice);
-C10_REGISTER_CREATOR(GlooDeviceRegistry, WIN32, makeUVDevice);
-C10_REGISTER_CREATOR(GlooDeviceRegistry, UV, makeUVDevice);
+C10_REGISTER_CREATOR(GlooDeviceRegistry, APPLE, makeUVDevice)
+C10_REGISTER_CREATOR(GlooDeviceRegistry, WIN32, makeUVDevice)
+C10_REGISTER_CREATOR(GlooDeviceRegistry, UV, makeUVDevice)
 #endif
 
 namespace {
diff --git a/torch/csrc/distributed/rpc/tensorpipe_agent.cpp b/torch/csrc/distributed/rpc/tensorpipe_agent.cpp
index a08aaa4576f6..9801a0327ddf 100644
--- a/torch/csrc/distributed/rpc/tensorpipe_agent.cpp
+++ b/torch/csrc/distributed/rpc/tensorpipe_agent.cpp
@@ -153,11 +153,11 @@ void makeStreamsWaitOnOthers(
 
 C10_DEFINE_REGISTRY_WITHOUT_WARNING(
     TensorPipeTransportRegistry,
-    TransportRegistration);
+    TransportRegistration)
 
 C10_DEFINE_REGISTRY_WITHOUT_WARNING(
     TensorPipeChannelRegistry,
-    ChannelRegistration);
+    ChannelRegistration)
 
 const std::string& TensorPipeAgent::guessAddress() {
   static const std::string uvAddress = []() {
@@ -284,7 +284,7 @@ std::unique_ptr<ChannelRegistration> makeMultiplexedUvChannel() {
 C10_REGISTER_CREATOR(
     TensorPipeChannelRegistry,
     mpt_uv,
-    makeMultiplexedUvChannel);
+    makeMultiplexedUvChannel)
 
 } // namespace
 
diff --git a/torch/csrc/distributed/rpc/tensorpipe_cuda.cpp b/torch/csrc/distributed/rpc/tensorpipe_cuda.cpp
index 97341a41899c..4c326b6a0e27 100644
--- a/torch/csrc/distributed/rpc/tensorpipe_cuda.cpp
+++ b/torch/csrc/distributed/rpc/tensorpipe_cuda.cpp
@@ -24,7 +24,7 @@ std::unique_ptr<ChannelRegistration> makeCudaIpcChannel() {
 }
 
 // The cuda_ipc channels use cudaMemcpy to transmit CUDA tensor across processes
-C10_REGISTER_CREATOR(TensorPipeChannelRegistry, cuda_ipc, makeCudaIpcChannel);
+C10_REGISTER_CREATOR(TensorPipeChannelRegistry, cuda_ipc, makeCudaIpcChannel)
 
 #endif
 
@@ -44,7 +44,7 @@ std::unique_ptr<ChannelRegistration> makeCudaGdrChannel() {
 // in order to ensure readiness and to agree on the device indices and thus the
 // queue pair to use. It automatically pairs each GPU to the "closest" NIC if
 // there are multiple of them (closest = longest prefix match in PCI tree).
-C10_REGISTER_CREATOR(TensorPipeChannelRegistry, cuda_gdr, makeCudaGdrChannel);
+C10_REGISTER_CREATOR(TensorPipeChannelRegistry, cuda_gdr, makeCudaGdrChannel)
 
 #endif
 
@@ -55,7 +55,7 @@ std::unique_ptr<ChannelRegistration> makeCudaXthChannel() {
 }
 
 // The cuda_xth channel supports same-process GPU-to-GPU comm
-C10_REGISTER_CREATOR(TensorPipeChannelRegistry, cuda_xth, makeCudaXthChannel);
+C10_REGISTER_CREATOR(TensorPipeChannelRegistry, cuda_xth, makeCudaXthChannel)
 
 std::unique_ptr<ChannelRegistration> makeCudaBasicChannel() {
   auto context = tensorpipe::channel::cuda_basic::create(
@@ -68,7 +68,7 @@ std::unique_ptr<ChannelRegistration> makeCudaBasicChannel() {
 C10_REGISTER_CREATOR(
     TensorPipeChannelRegistry,
     cuda_basic,
-    makeCudaBasicChannel);
+    makeCudaBasicChannel)
 
 class TensorpipeCudaConverter : public TensorpipeDeviceTypeConverter {
  public:
diff --git a/torch/csrc/jit/passes/onnx/constant_map.h b/torch/csrc/jit/passes/onnx/constant_map.h
index 60d4470c1b12..7b447fa74231 100644
--- a/torch/csrc/jit/passes/onnx/constant_map.h
+++ b/torch/csrc/jit/passes/onnx/constant_map.h
@@ -2,15 +2,10 @@
 
 #include <c10/macros/Macros.h>
 
-C10_DIAGNOSTIC_PUSH_AND_IGNORED_IF_DEFINED("-Wsuggest-override")
-C10_DIAGNOSTIC_PUSH_AND_IGNORED_IF_DEFINED("-Wnewline-eof")
 #include <onnx/shape_inference/implementation.h>
-C10_DIAGNOSTIC_POP()
-C10_DIAGNOSTIC_POP()
 
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/serialization/export.h>
-#include <mutex>
 #include <unordered_map>
 
 namespace torch::jit {
diff --git a/torch/csrc/jit/serialization/export.cpp b/torch/csrc/jit/serialization/export.cpp
index 84c0a46a7746..d95c45a7e6b9 100644
--- a/torch/csrc/jit/serialization/export.cpp
+++ b/torch/csrc/jit/serialization/export.cpp
@@ -19,18 +19,14 @@
 #include <torch/csrc/onnx/back_compat.h>
 #include <torch/csrc/onnx/onnx.h>
 #include <torch/version.h>
-#include <optional>
 
-C10_DIAGNOSTIC_PUSH_AND_IGNORED_IF_DEFINED("-Wnewline-eof")
 #include <onnx/checker.h>
-C10_DIAGNOSTIC_POP()
 #include <onnx/onnx_pb.h>
 #include <onnx/proto_utils.h>
-C10_DIAGNOSTIC_PUSH_AND_IGNORED_IF_DEFINED("-Wsuggest-override")
 #include <onnx/shape_inference/implementation.h>
-C10_DIAGNOSTIC_POP()
 
 #include <memory>
+#include <optional>
 #include <regex>
 #include <set>
 #include <sstream>