Enable Wno-unused-private-field,Wunused-lambda-capture and fix CUDA warnings (#110856)

This PR enables Wno-unused-private-field,Wunused-lambda-capture  and some CUDA warnings were fixed.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/110856
Approved by: https://github.com/albanD, https://github.com/malfet

CMakeLists.txt

@@ -894,10 +894,8 @@ if(NOT MSVC)
   append_cxx_flag_if_supported("-Wno-error=inconsistent-missing-destructor-override" CMAKE_CXX_FLAGS)
   append_cxx_flag_if_supported("-Wconstant-conversion" CMAKE_CXX_FLAGS)
   append_cxx_flag_if_supported("-Wno-invalid-partial-specialization" CMAKE_CXX_FLAGS)
-  append_cxx_flag_if_supported("-Wno-unused-private-field" 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("-Wunused-lambda-capture" CMAKE_CXX_FLAGS)
   append_cxx_flag_if_supported("-Qunused-arguments" CMAKE_CXX_FLAGS)
 
   if(${USE_COLORIZE_OUTPUT})

aten/src/ATen/native/cuda/FractionalMaxPool2d.cu

@@ -142,15 +142,11 @@ TORCH_IMPL_FUNC(fractional_max_pool2d_out_cuda) (
   fractional_max_pool_check_shape</*ndim*/ 2>(input, randomSamples);
 
   int planeDim = 0;
-  int dimh = 1;
-  int dimw = 2;
 
   int ndims = input.ndimension();
 
   if (ndims == 4) {
     planeDim++;
-    dimh++;
-    dimw++;
   }
 
   /* sizes */

aten/src/ATen/native/cuda/Indexing.cu

@@ -367,7 +367,7 @@ computeLinearIndex(const Tensor & src, TensorList indices, bool check_range) {
   // this point. We also compute the number of dimensions before and after that
   // are not being index.
   Tensor linearIndex;
-  int64_t emptyBefore = 0, emptyAfter = 0, nElemBefore = 1, nElemAfter = 1, strideBefore =0;
+  int64_t nElemBefore = 1, nElemAfter = 1, strideBefore =0;
   for (const auto i: c10::irange(src.dim())) {
     if (indices[i].defined()) {
       // Cast index to the longType matching src's device
@@ -382,10 +382,8 @@ computeLinearIndex(const Tensor & src, TensorList indices, bool check_range) {
         }
       }
     } else if (linearIndex.defined()) {
-      emptyAfter++;
       nElemAfter *= src.size(i);
     } else {
-      emptyBefore++;
       nElemBefore *= src.size(i);
     }
   }

aten/src/ATen/native/cuda/ReplicationPadding.cu

@@ -242,13 +242,11 @@ void replication_pad2d_backward_out_cuda_template(
   const auto padR = paddingSize[1];
   const auto padT = paddingSize[2];
   const auto padB = paddingSize[3];
-  int planeDim = 0;
   int dimh = 1;
   int dimw = 2;
 
   int numInputDims = input.dim();
   if (numInputDims == 4) {
-    planeDim++;
     dimh++;
     dimw++;
   }
@@ -376,18 +374,8 @@ void replication_pad3d_backward_out_cuda_template(
   shapeAndGradOutputCheck3d(input, gradOutput, pleft, pright, ptop,
       pbottom, pfront, pback);
 
-  int planeDim = 0;
-  int dimd = 1;
-  int dimh = 2;
-  int dimw = 3;
 
   int numInputDims = input.dim();
-  if (numInputDims == 5) {
-    planeDim++;
-    dimd++;
-    dimh++;
-    dimw++;
-  }
 
   gradInput.resize_as_(input);
   if (gradInput.numel() == 0) {
@@ -498,12 +486,10 @@ TORCH_IMPL_FUNC(replication_pad1d_backward_out_cuda) (
       "output gradient tensor must fit into 32-bit index math");
 
   const int padL = paddingSize[0];
-  int planeDim = 0;
   int dimw = 1;
 
   int numInputDims = input.ndimension();
   if (numInputDims == 3) {
-    planeDim++;
     dimw++;
   }
   int iwidth = input.size(dimw);

caffe2/CMakeLists.txt

@@ -964,6 +964,7 @@ elseif(USE_CUDA)
   endif()
   set(CUDA_LINK_LIBRARIES_KEYWORD)
   torch_compile_options(torch_cuda)  # see cmake/public/utils.cmake
+  target_compile_options_if_supported(torch_cuda "-Wno-deprecated-copy")  # see cmake/public/utils.cmake
   target_compile_definitions(torch_cuda PRIVATE USE_CUDA)
 
   if(USE_CUSPARSELT)

test/cpp/lazy/test_ir.cpp

@@ -23,9 +23,7 @@ class TestLeafNode : public Node {
   }
 
   explicit TestLeafNode(size_t param)
-      : Node(ClassOpKind(), /* num_outputs */ 1),
-        hash_(Hash(param)),
-        param_(param) {}
+      : Node(ClassOpKind(), /* num_outputs */ 1), hash_(Hash(param)) {}
   ~TestLeafNode() override = default;
 
   const std::vector<Output>& operands() const override {
@@ -45,7 +43,6 @@ class TestLeafNode : public Node {
 
  private:
   hash_t hash_;
-  size_t param_;
 };
 
 TEST(IrTest, BasicTest) {

test/edge/CMakeLists.txt

@@ -4,6 +4,8 @@ set(TORCH_ROOT ${CMAKE_CURRENT_LIST_DIR}/../..)
 set(TEST_ROOT ${TORCH_ROOT}/test/edge)
 set(OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/out)
 file(GLOB_RECURSE all_python "${TORCH_ROOT}/torchgen/*.py")
+include(${TORCH_ROOT}/cmake/public/utils.cmake)
+append_cxx_flag_if_supported("-Wno-unused-private-field" CMAKE_CXX_FLAGS)
 
 # Generate unboxing kernels
 set(GEN_COMMAND

third_party/nvfuser/test/test_gpu3.cpp

@@ -3095,7 +3095,6 @@ TEST_F(NVFuserTest, FusionPredRemovalCheck_CUDA) {
     }
 
    private:
-    bool within_ite_ = false;
   } pred_checker;
 
   GpuLower gpulw(&fusion);