Revert "Enable -Wunused on torch targets (#150077)"

This reverts commit 688adc9941.

Reverted https://github.com/pytorch/pytorch/pull/150077 on behalf of https://github.com/wdvr due to failing internally with use of undeclared identifier ([comment](https://github.com/pytorch/pytorch/pull/150077#issuecomment-2846499828))

aten/src/ATen/detail/CUDAHooksInterface.h

@@ -86,7 +86,7 @@ struct TORCH_API CUDAHooksInterface : AcceleratorHooksInterface {
     TORCH_CHECK(false, "Cannot get device of pointer on CUDA without ATen_cuda library. ", CUDA_HELP);
   }
 
-  bool isPinnedPtr(const void* /*data*/)  const override {
+  bool isPinnedPtr(const void* data) const override {
     return false;
   }
 

aten/src/ATen/detail/HIPHooksInterface.h

@@ -6,6 +6,8 @@
 
 #include <ATen/detail/AcceleratorHooksInterface.h>
 
+#include <memory>
+
 // NB: Class must live in `at` due to limitations of Registry.h.
 namespace at {
 
@@ -35,7 +37,7 @@ struct TORCH_API HIPHooksInterface : AcceleratorHooksInterface {
     return -1;
   }
 
-  bool isPinnedPtr(const void* /*data*/ ) const override {
+  bool isPinnedPtr(const void* data) const override {
     return false;
   }
 
@@ -47,7 +49,7 @@ struct TORCH_API HIPHooksInterface : AcceleratorHooksInterface {
     return 0;
   }
 
-  bool hasPrimaryContext(DeviceIndex /*device_index*/ ) const override {
+  bool hasPrimaryContext(DeviceIndex device_index) const override {
     TORCH_CHECK(false, "Cannot check primary context without ATen_hip library.");
   }
 };

aten/src/ATen/detail/IPUHooksInterface.h

@@ -15,7 +15,7 @@ struct TORCH_API IPUHooksInterface : AcceleratorHooksInterface {
     TORCH_CHECK(false, "Cannot initialize IPU without ATen_ipu library.");
   }
 
-  bool hasPrimaryContext(DeviceIndex /*device_index*/) const override {
+  bool hasPrimaryContext(DeviceIndex device_index) const override {
     TORCH_CHECK(false, "Cannot initialize IPU without ATen_ipu library.");
     return false;
   }
@@ -26,7 +26,7 @@ struct TORCH_API IPUHooksInterface : AcceleratorHooksInterface {
   }
 
   Generator getNewGenerator(
-      DeviceIndex /*device_index*/ = -1) const override {
+      DeviceIndex device_index [[maybe_unused]] = -1) const override {
     TORCH_CHECK(false, "Cannot initialize IPU without ATen_ipu library.");
   }
 };

aten/src/ATen/detail/MAIAHooksInterface.h

@@ -17,7 +17,7 @@ struct TORCH_API MAIAHooksInterface : AcceleratorHooksInterface {
     TORCH_CHECK(false, "Cannot initialize MAIA without ATen_maia library.");
   }
 
-  bool hasPrimaryContext(DeviceIndex /*device_index*/) const override {
+  bool hasPrimaryContext(DeviceIndex device_index) const override {
     TORCH_CHECK(false, "Cannot initialize MAIA without ATen_maia library.");
     return false;
   }

aten/src/ATen/detail/MTIAHooksInterface.h

@@ -12,6 +12,7 @@
 #include <ATen/detail/AcceleratorHooksInterface.h>
 
 #include <string>
+C10_DIAGNOSTIC_PUSH_AND_IGNORED_IF_DEFINED("-Wunused-parameter")
 namespace at {
 class Context;
 }
@@ -45,7 +46,7 @@ struct TORCH_API MTIAHooksInterface : AcceleratorHooksInterface {
     return 0;
   }
 
-  virtual void deviceSynchronize(c10::DeviceIndex /*device_index*/) const {
+  virtual void deviceSynchronize(c10::DeviceIndex device_index) const {
     FAIL_MTIAHOOKS_FUNC(__func__);
   }
 
@@ -53,11 +54,11 @@ struct TORCH_API MTIAHooksInterface : AcceleratorHooksInterface {
     FAIL_MTIAHOOKS_FUNC(__func__);
   }
 
-  bool hasPrimaryContext(DeviceIndex /*device_index*/) const override {
+  bool hasPrimaryContext(DeviceIndex device_index) const override {
     return false;
   }
 
-  void setCurrentDevice(DeviceIndex /*device*/) const override {
+  void setCurrentDevice(DeviceIndex device) const override {
     FAIL_MTIAHOOKS_FUNC(__func__);
   }
 
@@ -66,36 +67,36 @@ struct TORCH_API MTIAHooksInterface : AcceleratorHooksInterface {
     return -1;
   }
 
-  DeviceIndex exchangeDevice(DeviceIndex /*device*/) const override {
+  DeviceIndex exchangeDevice(DeviceIndex device) const override {
     FAIL_MTIAHOOKS_FUNC(__func__);
     return -1;
   }
 
-  DeviceIndex maybeExchangeDevice(DeviceIndex /*device*/) const override {
+  DeviceIndex maybeExchangeDevice(DeviceIndex device) const override {
     FAIL_MTIAHOOKS_FUNC(__func__);
     return -1;
   }
 
-  virtual c10::Stream getCurrentStream(DeviceIndex /*device*/) const {
+  virtual c10::Stream getCurrentStream(DeviceIndex device) const {
     FAIL_MTIAHOOKS_FUNC(__func__);
     return c10::Stream::unpack3(-1, 0, c10::DeviceType::MTIA);
   }
 
-  virtual int64_t getCurrentRawStream(DeviceIndex /*device*/) const {
+  virtual int64_t getCurrentRawStream(DeviceIndex device) const {
     FAIL_MTIAHOOKS_FUNC(__func__);
     return -1;
   }
 
-  virtual c10::Stream getDefaultStream(DeviceIndex /*device*/) const {
+  virtual c10::Stream getDefaultStream(DeviceIndex device) const {
     FAIL_MTIAHOOKS_FUNC(__func__);
     return c10::Stream::unpack3(-1, 0, c10::DeviceType::MTIA);
   }
 
-  virtual void setCurrentStream(const c10::Stream& /*stream*/ ) const {
+  virtual void setCurrentStream(const c10::Stream& stream) const {
     FAIL_MTIAHOOKS_FUNC(__func__);
   }
 
-  bool isPinnedPtr(const void* /*data*/) const override {
+  bool isPinnedPtr(const void* data) const override {
     return false;
   }
 
@@ -104,12 +105,12 @@ struct TORCH_API MTIAHooksInterface : AcceleratorHooksInterface {
     return nullptr;
   }
 
-  virtual PyObject* memoryStats(DeviceIndex /*device*/) const {
+  virtual PyObject* memoryStats(DeviceIndex device) const {
     FAIL_MTIAHOOKS_FUNC(__func__);
     return nullptr;
   }
 
-  virtual PyObject* getDeviceCapability(DeviceIndex /*device*/) const {
+  virtual PyObject* getDeviceCapability(DeviceIndex device) const {
     FAIL_MTIAHOOKS_FUNC(__func__);
     return nullptr;
   }
@@ -120,9 +121,9 @@ struct TORCH_API MTIAHooksInterface : AcceleratorHooksInterface {
 
 
   virtual void recordMemoryHistory(
-    const std::optional<std::string>& /*enabled*/,
-    const std::string& /*stacks*/,
-    size_t /*max_entries*/) const {
+    const std::optional<std::string>& enabled,
+    const std::string& stacks,
+    size_t max_entries) const {
     FAIL_MTIAHOOKS_FUNC(__func__);
   }
 
@@ -136,7 +137,7 @@ struct TORCH_API MTIAHooksInterface : AcceleratorHooksInterface {
     return 0;
   }
 
-  virtual void resetPeakMemoryStats(DeviceIndex /*device*/) const {
+  virtual void resetPeakMemoryStats(DeviceIndex device) const {
     FAIL_MTIAHOOKS_FUNC(__func__);
   }
 
@@ -157,3 +158,4 @@ TORCH_API const MTIAHooksInterface& getMTIAHooks();
 TORCH_API bool isMTIAHooksBuilt();
 } // namespace detail
 } // namespace at
+C10_DIAGNOSTIC_POP()

aten/src/ATen/native/TensorShape.cpp

@@ -4201,7 +4201,8 @@ static inline void handle_unflatten_exception(
     const std::runtime_error& e,
     const Tensor& self,
     int64_t dim,
-    SymIntArrayRef sizes) {
+    SymIntArrayRef sizes,
+    std::optional<DimnameList> names) {
   if (!strstr(e.what(), "is invalid for input of size")) {
     TORCH_CHECK(false, "unflatten got an unexpected error:\n", e.what());
   }
@@ -4255,7 +4256,7 @@ static Tensor unflatten_impl(
     // at::infer_size would throw std::runtime_error for invalid size,
     // catch the runtime_error and display the error message in a more
     // user-friendly way for both tensors and named tensors
-    handle_unflatten_exception(e, self, dim, sizes);
+    handle_unflatten_exception(e, self, dim, sizes, names);
   }
 
   SymDimVector shape(self.sym_sizes().begin(), self.sym_sizes().end());

aten/src/ATen/native/cuda/MemoryAccess.cuh

@@ -48,7 +48,7 @@ struct static_unroll {
 template<template<int i> typename func, int end>
 struct static_unroll<func, end, end> {
   template<typename... Args>
-  static inline C10_HOST_DEVICE void with_args(Args... /*args*/) {}
+  static inline C10_HOST_DEVICE void with_args(Args... args) {}
 };
 
 // helper structs to be used with static_unroll to load arguments
@@ -516,7 +516,7 @@ inline C10_HOST_DEVICE int can_vectorize_up_to(char *pointer) {
 template<int i>
 struct can_vectorize_up_to_helper {
   template <typename array_t, typename traits>
-  static C10_HOST_DEVICE void apply(int &result, array_t pointers, traits /*_*/) {
+  static C10_HOST_DEVICE void apply(int &result, array_t pointers, traits _) {
     using arg_t = typename traits::template arg<i>::type;
     // `pointers` hold the data_ptr for tensors [output, input0, input1, ...], so we
     // need a +1 offset to get the input

aten/src/ATen/native/quantized/cpu/qnnpack/CMakeLists.txt

@@ -331,9 +331,6 @@ if(NOT TARGET clog)
     "${CONFU_DEPENDENCIES_BINARY_DIR}/clog")
   # We build static version of clog but a dynamic library may indirectly depend on it
   set_property(TARGET clog PROPERTY POSITION_INDEPENDENT_CODE ON)
-  if(CMAKE_CXX_COMPILER_ID MATCHES "Clang" OR CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
-    target_compile_options(clog PRIVATE "-Wno-unused-result")
-  endif()
 endif()
 target_link_libraries(pytorch_qnnpack PUBLIC clog)
 

c10/core/GeneratorImpl.cpp

@@ -32,7 +32,7 @@ c10::intrusive_ptr<GeneratorImpl> GeneratorImpl::clone() const {
 }
 
 void GeneratorImpl::graphsafe_set_state(
-    const c10::intrusive_ptr<c10::GeneratorImpl>& /*state*/) {
+    const c10::intrusive_ptr<c10::GeneratorImpl>& state) {
   TORCH_CHECK_NOT_IMPLEMENTED(
       false, "graphsafe_set_state is not supported in this Generator");
 }

c10/core/TensorImpl.cpp

@@ -102,7 +102,7 @@ TensorImpl::TensorImpl(
 
 // NOLINTNEXTLINE(cppcoreguidelines-pro-type-member-init)
 TensorImpl::TensorImpl(
-    ImplType /*type*/,
+    ImplType type,
     Storage&& storage,
     DispatchKeySet key_set,
     const caffe2::TypeMeta data_type)

c10/core/impl/FakeGuardImpl.h

@@ -59,15 +59,15 @@ struct FakeGuardImpl final : public DeviceGuardImplInterface {
 
   // Event-related functions
   void record(
-      void** /*event*/,
-      const Stream& /*stream*/,
-      const DeviceIndex /*device_index*/,
-      const EventFlag /*flag*/) const override {}
-  void block(void* /*event*/, const Stream& /*stream*/) const override {}
-  bool queryEvent(void* /*event*/) const override {
+      void** event,
+      const Stream& stream,
+      const DeviceIndex device_index,
+      const EventFlag flag) const override {}
+  void block(void* event, const Stream& stream) const override {}
+  bool queryEvent(void* event) const override {
     return true;
   }
-  void destroyEvent(void* /*event*/, const DeviceIndex /*device_index*/)
+  void destroyEvent(void* event, const DeviceIndex device_index)
       const noexcept override {}
 
   // Convenience methods for testing

c10/core/impl/PyInterpreter.cpp

@@ -1,7 +1,7 @@
 #include <c10/core/SymIntArrayRef.h>
 #include <c10/core/TensorImpl.h>
 #include <c10/core/impl/PyInterpreter.h>
-C10_DIAGNOSTIC_PUSH_AND_IGNORED_IF_DEFINED("-Wunused-parameter")
+
 namespace c10::impl {
 
 struct NoopPyInterpreterVTable final : public PyInterpreterVTable {
@@ -145,4 +145,3 @@ void PyInterpreter::disarm() noexcept {
 }
 
 } // namespace c10::impl
-C10_DIAGNOSTIC_POP()

c10/core/impl/alloc_cpu.cpp

@@ -70,7 +70,7 @@ inline bool is_thp_alloc(size_t nbytes) {
 }
 
 #elif !defined(__ANDROID__) && !defined(_MSC_VER)
-constexpr size_t c10_compute_alignment(size_t /*nbytes*/) {
+constexpr size_t c10_compute_alignment([[maybe_unused]] size_t nbytes) {
   return gAlignment;
 }
 

c10/cuda/CUDACachingAllocator.h

@@ -227,9 +227,7 @@ class CUDAAllocator : public Allocator {
       c10::DeviceIndex device,
       MempoolId_t mempool_id) = 0;
   virtual void releasePool(c10::DeviceIndex device, MempoolId_t mempool_id) = 0;
-  virtual int getPoolUseCount(
-      c10::DeviceIndex /*device*/,
-      MempoolId_t /*mempool_id*/) {
+  virtual int getPoolUseCount(c10::DeviceIndex device, MempoolId_t mempool_id) {
     TORCH_CHECK(
         false,
         name(),
@@ -237,8 +235,8 @@ class CUDAAllocator : public Allocator {
         "If you need it, please file an issue describing your use case.");
   }
   virtual void ensureExistsAndIncrefPool(
-      c10::DeviceIndex /*device*/,
-      MempoolId_t /*mempool_id*/) {
+      c10::DeviceIndex device,
+      MempoolId_t mempool_id) {
     TORCH_CHECK(
         false,
         name(),
@@ -258,9 +256,9 @@ class CUDAAllocator : public Allocator {
 
   // returns true if the allocated blocks are equal to expected live allocations
   virtual bool checkPoolLiveAllocations(
-      c10::DeviceIndex /*device*/,
-      MempoolId_t /*mempool_id*/,
-      const std::unordered_set<void*>& /*expected_live_allocations*/) {
+      c10::DeviceIndex device,
+      MempoolId_t mempool_id,
+      const std::unordered_set<void*>& expected_live_allocations) {
     TORCH_CHECK(
         false,
         name(),
@@ -283,7 +281,7 @@ class CUDAAllocator : public Allocator {
       RecordContext when,
       bool clearHistory) = 0;
   virtual void recordAnnotation(
-      const std::vector<std::pair<std::string, std::string>>& /*md*/) {}
+      const std::vector<std::pair<std::string, std::string>>& md) {}
   virtual void attachOutOfMemoryObserver(OutOfMemoryObserver observer) = 0;
 
   // Attached AllocatorTraceTracker callbacks will be called while the

c10/cuda/CUDADeviceAssertionHost.cpp

@@ -213,11 +213,11 @@ bool CUDAKernelLaunchRegistry::check_env_for_dsa_enabled() const {
 }
 
 uint32_t CUDAKernelLaunchRegistry::insert(
-    const char* /*launch_filename*/,
-    const char* /*launch_function*/,
-    const uint32_t /*launch_linenum*/,
-    const char* /*kernel_name*/,
-    const int32_t /*stream_id*/) {
+    const char* launch_filename,
+    const char* launch_function,
+    const uint32_t launch_linenum,
+    const char* kernel_name,
+    const int32_t stream_id) {
 #ifdef TORCH_USE_CUDA_DSA
   if (!enabled_at_runtime) {
     return 0;

c10/cuda/CUDAException.cpp

@@ -10,9 +10,9 @@ namespace c10::cuda {
 
 void c10_cuda_check_implementation(
     const int32_t err,
-    const char* /*filename*/,
-    const char* /*function_name*/,
-    const int /*line_number*/,
+    const char* filename,
+    const char* function_name,
+    const int line_number,
     const bool include_device_assertions) {
   const auto cuda_error = static_cast<cudaError_t>(err);
   const auto cuda_kernel_failure = include_device_assertions
@@ -24,6 +24,7 @@ void c10_cuda_check_implementation(
   }
 
   [[maybe_unused]] auto error_unused = cudaGetLastError();
+  (void)error_unused;
 
   std::string check_message;
 #ifndef STRIP_ERROR_MESSAGES

c10/util/Exception.cpp

@@ -213,7 +213,7 @@ Warning::Warning(
 Warning::Warning(
     warning_variant_t type,
     SourceLocation source_location,
-    detail::CompileTimeEmptyString /*msg*/,
+    detail::CompileTimeEmptyString msg,
     const bool verbatim)
     : Warning(type, source_location, "", verbatim) {}
 

c10/util/Logging.cpp

@@ -139,7 +139,7 @@ APIUsageLoggerType* GetAPIUsageLogger() {
 APIUsageMetadataLoggerType* GetAPIUsageMetadataLogger() {
   static APIUsageMetadataLoggerType func =
       [](const std::string&,
-         const std::map<std::string, std::string>& /*metadata_map*/) {};
+         const std::map<std::string, std::string>& metadata_map) {};
   return &func;
 }
 
@@ -386,7 +386,7 @@ void initLogging() {
   detail::setLogLevelFlagFromEnv();
 }
 
-bool InitCaffeLogging(int* argc, char** /*argv*/) {
+bool InitCaffeLogging(int* argc, char** argv) {
   // When doing InitCaffeLogging, we will assume that caffe's flag parser has
   // already finished.
   if (*argc == 0)

c10/util/NetworkFlow.cpp

@@ -216,7 +216,9 @@ struct DinicFlowGraph {
     return seen;
   }
 
-  std::pair<std::vector<size_t>, std::vector<size_t>> partition(size_t t) {
+  std::pair<std::vector<size_t>, std::vector<size_t>> partition(
+      size_t s,
+      size_t t) {
     // Note: the partitioning returns "reachable" / "unreachable",
     //   but specifically, for "unreachable", it returns "all vertices
     //   that are reachable from t in the reverse residual graph"
@@ -256,7 +258,7 @@ struct DinicFlowGraph {
       };
     }
 
-    auto [reachable_idxs, unreachable_idxs] = partition(t_int);
+    auto [reachable_idxs, unreachable_idxs] = partition(s_int, t_int);
     std::vector<std::string> reachable, unreachable;
 
     auto idxs_to_names = [&](std::vector<size_t>& src,

c10/util/numa.cpp

@@ -113,9 +113,9 @@ bool IsNUMAEnabled() {
   return false;
 }
 
-void NUMABind(int /*numa_node_id*/) {}
+void NUMABind(int numa_node_id) {}
 
-int GetNUMANode(const void* /*ptr*/) {
+int GetNUMANode(const void* ptr) {
   return -1;
 }
 
@@ -123,7 +123,7 @@ int GetNumNUMANodes() {
   return -1;
 }
 
-void NUMAMove(void* /*ptr*/, size_t /*size*/, int /*numa_node_id*/) {}
+void NUMAMove(void* ptr, size_t size, int numa_node_id) {}
 
 int GetCurrentNUMANode() {
   return -1;

cmake/public/utils.cmake

@@ -383,7 +383,6 @@ function(torch_compile_options libname)
       -Wall
       -Wextra
       -Wdeprecated
-      -Wunused
       -Wno-unused-parameter
       -Wno-missing-field-initializers
       -Wno-array-bounds
@@ -391,11 +390,13 @@ function(torch_compile_options libname)
       -Wno-strict-overflow
       -Wno-strict-aliasing
       )
+    list(APPEND private_compile_options -Wunused-function)
+    list(APPEND private_compile_options -Wunused-variable)
     if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
-      list(APPEND private_compile_options -Wredundant-move)
+      list(APPEND private_compile_options -Wunused-but-set-variable -Wredundant-move)
     endif()
     if(CMAKE_CXX_COMPILER_ID MATCHES "Clang")
-      list(APPEND private_compile_options -Wextra-semi -Wno-error=extra-semi -Wmove)
+      list(APPEND private_compile_options -Wunused-private-field -Wextra-semi -Wno-error=extra-semi -Wmove)
     else()
       list(APPEND private_compile_options
         # Considered to be flaky.  See the discussion at
@@ -408,9 +409,9 @@ function(torch_compile_options libname)
         -Werror
         -Werror=inconsistent-missing-override
         -Werror=inconsistent-missing-destructor-override
+        -Werror=unused-function
+        -Werror=unused-variable
         -Werror=pedantic
-        -Werror=unused
-        -Wno-error=unused-parameter
       )
       if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
         list(APPEND private_compile_options -Werror=unused-but-set-variable)