fix cmake cxx flags (PaddlePaddle#9467)

cmake/flags.cmake

@@ -119,6 +119,7 @@ if (NOT WIN32)
 set(COMMON_FLAGS
     -fPIC
     -fno-omit-frame-pointer
+    -Werror
     -Wall
     -Wextra
     -Wnon-virtual-dtor
@@ -139,10 +140,14 @@ set(COMMON_FLAGS
     -Wno-error=maybe-uninitialized # Warning in boost gcc 7.2
 )
 
-if (NOT EMSCRIPTEN)
-  # disable -Werror for Emscripten
-  set(COMMON_FLAGS "${COMMON_FLAGS} -Werror")
-endif(NOT EMSCRIPTEN)
+if((CMAKE_CXX_COMPILER_ID STREQUAL "Clang") OR EMSCRIPTEN)
+  # disable -Werror
+  list(REMOVE_ITEM COMMON_FLAGS "-Werror")
+endif()
+
+if((CMAKE_CXX_COMPILER_ID STREQUAL "GNU") AND (CMAKE_CXX_COMPILER_VERSION VERSION_GREATER_EQUAL 9.3))
+  list(APPEND COMMON_FLAGS "-Wno-error=deprecated-copy") # Warning in Eigen
+endif()
 
 set(GPU_COMMON_FLAGS
     -fPIC

lite/backends/nnadapter/nnadapter/src/driver/cambricon_mlu/optimizer/fix_non_max_suppression.cc

@@ -221,7 +221,6 @@ bool NMSFixer::HandleMatchedResults(core::Model* model,
   auto concat0_operand0 = concat0_operation->input_operands[0];
   auto concat0_operand1 = concat0_operation->input_operands[1];
   auto concat0_operand2 = concat0_operation->input_operands[2];
-  auto concat0_operand3 = concat0_operation->input_operands[3];
   auto concat0_operand4 = concat0_operation->input_operands[4];
   concat0_operation->input_operands = {
       concat0_operand0, concat0_operand1, concat0_operand2, concat0_operand4};
@@ -230,14 +229,12 @@ bool NMSFixer::HandleMatchedResults(core::Model* model,
   auto concat1_operand0 = concat1_operation->input_operands[0];
   auto concat1_operand1 = concat1_operation->input_operands[1];
   auto concat1_operand2 = concat1_operation->input_operands[2];
-  auto concat1_operand3 = concat1_operation->input_operands[3];
   auto concat1_operand4 = concat1_operation->input_operands[4];
   concat1_operation->input_operands = {
       concat1_operand0, concat1_operand1, concat1_operand2, concat1_operand4};
 
   auto concat2_operation = nodes.at("concat2")->operation;
   auto concat2_operand0 = concat2_operation->input_operands[0];
-  auto concat2_operand1 = concat2_operation->input_operands[1];
   auto concat2_operand2 = concat2_operation->input_operands[2];
   auto nms_operation = nodes.at("nms")->operation;
   auto nms_index_operand = nms_operation->output_operands[2];

lite/backends/nnadapter/nnadapter/src/optimizer/constant_fold_operations.cc

@@ -50,7 +50,7 @@ class ConstantFoldOperationsFuser : public PatternMatcher {
 
 void ConstantFoldOperationsFuser::BuildPattern() {
   // Create patterns
-  auto operation = CreatePattern("operation")->IsOperation();
+  CreatePattern("operation")->IsOperation();
 }
 
 bool ConstantFoldOperationsFuser::HandleMatchedResults(

lite/backends/nnadapter/nnadapter/src/optimizer/fuse_conv2d_activation_into_conv2d.cc

@@ -47,7 +47,7 @@ void Conv2DActivationFuser::BuildPattern() {
   // Operand patterns
   auto conv2d_input_pattern =
       CreatePattern("conv2d_input")->IsOperationInputOperand(conv2d_type_, 0);
-  int conv2d_fuse_code_index;
+  int conv2d_fuse_code_index = -1;
   if (conv2d_type_ == NNADAPTER_CONV_2D) {
     conv2d_fuse_code_index = 8;
   } else if (conv2d_type_ == NNADAPTER_CONV_2D_TRANSPOSE) {

lite/backends/nnadapter/nnadapter/src/optimizer/fuse_conv2d_add_into_conv2d.cc

@@ -44,7 +44,7 @@ void Conv2DAddFuser::BuildPattern() {
   // Operand patterns
   auto conv2d_input_pattern =
       CreatePattern("conv2d_input")->IsOperationInputOperand(conv2d_type_, 0);
-  int conv2d_fuse_code_index;
+  int conv2d_fuse_code_index = -1;
   if (conv2d_type_ == NNADAPTER_CONV_2D) {
     conv2d_fuse_code_index = 8;
   } else if (conv2d_type_ == NNADAPTER_CONV_2D_TRANSPOSE) {
@@ -91,8 +91,6 @@ bool Conv2DAddFuser::HandleMatchedResults(
   // Get the operands and operations from the matched subgraph nodes.
   auto conv2d_operation = nodes.at("conv2d")->operation;
   auto conv2d_fuse_code_operand = nodes.at("conv2d_fuse_code")->operand;
-  auto conv2d_input_operand = conv2d_operation->input_operands[0];
-  auto conv2d_output_operand = conv2d_operation->output_operands[0];
   auto conv2d_filter_operand = conv2d_operation->input_operands[1];
   auto conv2d_bias_operand = conv2d_operation->input_operands[2];
   auto conv2d_group =
@@ -103,7 +101,6 @@ bool Conv2DAddFuser::HandleMatchedResults(
     conv2d_output_channel_size =
         conv2d_filter_operand->type.dimensions.data[1] * conv2d_group;
   }
-  auto add_operation = nodes.at("add")->operation;
   auto add_input_operand = nodes.at("add_input")->operand;
   auto add_fuse_code_operand = nodes.at("add_fuse_code")->operand;
   auto add_output_operand = nodes.at("add_output")->operand;

lite/backends/nnadapter/nnadapter/src/optimizer/fuse_matmul_dequant_add_into_fully_connected_dequant.cc

@@ -119,7 +119,6 @@ bool MatMulDequantAddFuser::HandleMatchedResults(
   auto add_output_operand = add_operation->output_operands[0];
   auto dequant_operation = nodes.at("dequant")->operation;
   auto dequant_input_operand = dequant_operation->input_operands[0];
-  auto dequant_output_operand = dequant_operation->output_operands[0];
   auto matmul_transpose_y =
       *reinterpret_cast<bool*>(matmul_transpose_y_operand->buffer);
   auto matmul_num_units =

lite/backends/nnadapter/nnadapter/src/optimizer/remove_useless_operations.cc

@@ -36,7 +36,7 @@ class RemoveUselessCastFuser : public PatternMatcher {
 
 void RemoveUselessCastFuser::BuildPattern() {
   // Create patterns
-  auto cast = CreatePattern("cast", NNADAPTER_CAST);
+  CreatePattern("cast", NNADAPTER_CAST);
 }
 
 bool RemoveUselessCastFuser::HandleMatchedResults(
@@ -149,7 +149,6 @@ void RemoveUselessMulFuser::BuildPattern() {
 bool RemoveUselessMulFuser::HandleMatchedResults(
     core::Model* model, const std::map<std::string, Node*>& nodes) {
   auto mul_x = nodes.at("mul_x")->operand;
-  auto mul = nodes.at("mul")->operation;
   auto mul_out = nodes.at("mul_out")->operand;
   if (IsModelInputOperand(mul_x) && IsModelOutputOperand(mul_out)) return false;
 

lite/backends/nnadapter/nnadapter/src/runtime/device.cc

@@ -74,6 +74,7 @@ class Context {
 };
 
 Context::Context(void* device, const char* properties) : device_(device) {
+  NNADAPTER_CHECK(device_);
   // Extract the runtime parameters from the context properties
   NNADAPTER_LOG(INFO) << "properties: " << std::string(properties);
   std::string key_value;
@@ -89,7 +90,9 @@ Context::Context(void* device, const char* properties) : device_(device) {
 
 class Program {
  public:
-  explicit Program(Context* context) : context_(context) {}
+  explicit Program(Context* context) : context_(context) {
+    NNADAPTER_CHECK(context_);
+  }
   ~Program() { Clear(); }
 
   int Validate(const core::Model* model, bool* supported_operations);

lite/backends/x86/math/conv_direct_fp32.cc

@@ -180,10 +180,10 @@ void conv_direct::generate_code(int ic,
 
   // compute output whole line
   // three parts: [left, middle, right]
-  auto cal_out_whole_line = [=, &temp](int oc_group, int ic_group) {
+  auto cal_out_whole_line = [=](int oc_group, int ic_group) {
     int ow_bulk_i = ow / ow_bulk;
 
-    auto cal_bulk = [=, &temp](
+    auto cal_bulk = [=](
         int oc_group, int ic_group, int l_pad, int r_pad, int bulk) {
       load(oc_group, bulk);
 
@@ -255,7 +255,7 @@ void conv_direct::generate_code(int ic,
   };
 
   //
-  auto cal_with_ic_fixed = [=, &temp](int ic_group) {
+  auto cal_with_ic_fixed = [=](int ic_group) {
     // ic_group is fixed
     // according to oc !
     Xbyak::Label label_oc_remain;

lite/core/optimizer/mir/fusion/quant_dequant_op_fuser.cc

@@ -322,13 +322,12 @@ void DequantOpFuser::InsertNewNode(SSAGraph* graph,
 #endif
   op_desc.SetInputScale(weight_name, weight_scale);
 
-  // change the weight from the float type to int8 type.
+// change the weight from the float type to int8 type.
+#ifdef LITE_WITH_FPGA
   Tensor temp_tensor;
   temp_tensor.CopyDataFrom(*quantized_weight_t);
   float* temp_data = temp_tensor.mutable_data<float>();
   size_t weight_num = quantized_weight_t->data_size();
-
-#ifdef LITE_WITH_FPGA
   float* quantized_weight_data = quantized_weight_t->mutable_data<float>();
   for (size_t i = 0; i < weight_num; i++) {
     quantized_weight_data[i] = temp_data[i] * whole_weight_scale;
@@ -442,11 +441,10 @@ void ChannelWiseDequantOpFuser::InsertNewNode(SSAGraph* graph,
   auto quantized_weight_var_name = quantized_op_weight->arg()->name;
   auto quantized_weight_t =
       scope->FindVar(quantized_weight_var_name)->GetMutable<lite::Tensor>();
+#ifdef LITE_WITH_FPGA
   Tensor temp_tensor;
   temp_tensor.CopyDataFrom(*quantized_weight_t);
   float* temp_data = temp_tensor.mutable_data<float>();
-
-#ifdef LITE_WITH_FPGA
   float* quantized_weight_data = quantized_weight_t->mutable_data<float>();
   int channel = channel_scale_tensor->data_size();
   int weight_chw = quantized_weight_t->data_size() / channel;
@@ -456,7 +454,6 @@ void ChannelWiseDequantOpFuser::InsertNewNode(SSAGraph* graph,
   }
   quantized_weight_t->set_persistable(true);
   quantized_weight_t->set_precision(PRECISION(kFloat));
-
 #else
   CastPersistableTensorInPlace(quantized_weight_t, weight_bit_length);
 #endif

lite/kernels/opencl/conv_image_compute.cc

@@ -1841,6 +1841,8 @@ void ConvImageCompute::Conv2d1x1opt() {
   }
 }
 
+void ConvImageCompute::Conv2d3x3() { LOG(FATAL) << "Not implement."; }
+
 void ConvImageCompute::Conv2dnxnopt() {
   int arg_idx = 0;
   status_ = kernel_.setArg(arg_idx++, c_blk_);

lite/kernels/xpu/topk_v2_compute.cc

@@ -25,9 +25,6 @@ namespace xpu {
 void TopkV2Compute::Run() {
   auto& param = this->Param<operators::TopkParam>();
   auto& ctx = this->ctx_->As<XPUContext>();
-  const float* x_data = param.X->data<float>();
-  float* out_val = param.Out->mutable_data<float>();
-  auto out_ind = param.Indices->mutable_data<int64_t>();
 
   DDim x_dims = param.X->dims();
   int axis = param.axis;

lite/operators/__xpu__multi_encoder_op.cc

@@ -143,9 +143,10 @@ bool XPUMultiEncoderOp::AttachImpl(const cpp::OpDesc& op_desc,
   param_.norm_before = op_desc.GetAttr<bool>("norm_before");
   param_.adaptive_seqlen = op_desc.GetAttr<bool>("adaptive_seqlen");
   param_.per_channel = op_desc.GetAttr<bool>("per_channel");
-  if (op_desc.HasAttr("enable_int8") && op_desc.GetAttr<bool>("enable_int8") ||
-      op_desc.HasAttr("enable_int16") &&
-          op_desc.GetAttr<bool>("enable_int16")) {
+  if ((op_desc.HasAttr("enable_int8") &&
+       op_desc.GetAttr<bool>("enable_int8")) ||
+      (op_desc.HasAttr("enable_int16") &&
+       op_desc.GetAttr<bool>("enable_int16"))) {
     param_.input_max = op_desc.GetAttr<std::vector<float>>("FCInputMax");
   }
   param_.weight_max.clear();

lite/tests/kernels/generate_proposals_v2_compute_test.cc

@@ -258,10 +258,6 @@ class GenerateProposalsV2ComputeTester : public arena::TestCase {
     const T* im_info_data = im_info.data<T>();
     T offset = pixel_offset ? static_cast<T>(1) : 0;
     T zero(0);
-    T im_w =
-        is_scale ? round(im_info_data[1] / im_info_data[2]) : im_info_data[1];
-    T im_h =
-        is_scale ? round(im_info_data[0] / im_info_data[2]) : im_info_data[0];
     for (int64_t i = 0; i < boxes->numel(); ++i) {
       if (i % 4 == 0) {
         boxes_data[i] =
@@ -301,13 +297,6 @@ class GenerateProposalsV2ComputeTester : public arena::TestCase {
         T x_ctr = boxes_data[4 * i] + ws / 2;
         T y_ctr = boxes_data[4 * i + 1] + hs / 2;
 
-        if (is_scale) {
-          T ws =
-              (boxes_data[4 * i + 2] - boxes_data[4 * i]) / im_info_data[2] + 1;
-          T hs = (boxes_data[4 * i + 3] - boxes_data[4 * i + 1]) /
-                     im_info_data[2] +
-                 1;
-        }
         if (ws >= min_size && hs >= min_size && x_ctr <= im_info_data[1] &&
             y_ctr <= im_info_data[0]) {
           keep_data[keep_len++] = i;

lite/tests/kernels/roll_compute_test.cc

@@ -97,7 +97,6 @@ class RollComputeTester : public arena::TestCase {
 
   void RunBaseline(Scope* scope) override {
     auto* x = scope->FindTensor(x_);
-    const auto* x_data = x->data<float>();
     std::vector<int64_t> shifts;
     if (!shifts_tensor_.empty()) {
       auto* shift = scope->FindTensor(shifts_tensor_);
@@ -121,7 +120,6 @@ class RollComputeTester : public arena::TestCase {
     out->CopyDataFrom(*x);
     auto* out_data = out->mutable_data<float>();
     for (size_t i = 0; i < nums; i++) {
-      int64_t input_size = input_dim.size();
       ShiftAlongDim(out_data, input_dim, axis_[i], shifts_[i]);
     }
   }

lite/tests/math/sgemm_compute_test.cc

@@ -122,7 +122,6 @@ bool test_sgemm(bool tra,
   auto da = ta.mutable_data<float>();
   auto db = tb.mutable_data<float>();
   auto dc = tc.mutable_data<float>();
-  auto dc_sve = tc_sve.mutable_data<float>();
   auto dc_basic = tc_basic.mutable_data<float>();
   auto dc_backup = tc_backup.mutable_data<float>();
   auto dbias = tbias.mutable_data<float>();
@@ -212,6 +211,7 @@ bool test_sgemm(bool tra,
   }
 #ifdef LITE_WITH_ARM8_SVE2
   // sve
+  auto dc_sve = tc_sve.mutable_data<float>();
   Timer t1;
   for (int i = 0; i < FLAGS_repeats; ++i) {
     if (i == FLAGS_repeats - 1) {