Fix clang compiler warnings (#2898)

* Fix clang compiler warnings

* Remove unintended file.

* style

* Not part of PR

* Another extra closure ref

* More warnings from merges

* Lambda arg was used

src/ngraph/builder/quantization/quantized_linear_convolution.cpp

@@ -104,7 +104,7 @@ namespace ngraph
                                               output_zero_point->get_element_type(),
                                               axes,
                                               op::Quantize::RoundMode::ROUND_NEAREST_TOWARD_EVEN);
-                return q_convolution;
+                return move(q_convolution);
             }
 
             shared_ptr<Node> QuantizedLinearConvolutionBias(const shared_ptr<Node>& input,

src/ngraph/op/result.cpp

@@ -49,7 +49,7 @@ shared_ptr<Node> op::Result::copy_with_new_args(const NodeVector& new_args) cons
     {
         res->set_needs_default_layout(m_needs_default_layout);
     }
-    return res;
+    return std::move(res);
 }
 
 void op::Result::generate_adjoints(autodiff::Adjoints& adjoints, const NodeVector& deltas)

src/ngraph/op/reverse_sequence.cpp

@@ -97,7 +97,7 @@ shared_ptr<Node> op::ReverseSequence::copy_with_new_args(const NodeVector& new_a
     check_new_args_count(this, new_args);
     auto res =
         make_shared<ReverseSequence>(new_args.at(0), new_args.at(1), m_batch_axis, m_seq_axis);
-    return res;
+    return move(res);
 }
 
 void op::ReverseSequence::generate_adjoints(autodiff::Adjoints& adjoints, const NodeVector& deltas)

src/ngraph/pass/batch_fusion.cpp

@@ -160,7 +160,7 @@ std::shared_ptr<Node> fuse_group_convolution(const std::shared_ptr<Node>& n)
                                                            sconv->get_data_dilation_strides(),
                                                            n->get_arguments().size());
 
-    return new_conv;
+    return move(new_conv);
 }
 
 bool ngraph::pass::BatchFusion::run_on_function(std::shared_ptr<Function> func)

src/ngraph/pass/core_fusion.cpp

@@ -410,7 +410,7 @@ static std::shared_ptr<Node> reduce_broadcast(std::shared_ptr<Node> broadcast)
         std::make_shared<op::Broadcast>(matched_broadcast_w1->get_argument(0),
                                         shape_w1,
                                         matched_broadcast_w1->get_broadcast_axes());
-    return new_broadcast_w1;
+    return move(new_broadcast_w1);
 }
 
 static size_t shape_to_index(Shape shape)

src/ngraph/runtime/cpu/builder/get_output_element.cpp

@@ -38,8 +38,8 @@ namespace ngraph
                 size_t n = goe->get_n();
                 auto arg_buffer_index = external_function->get_buffer_index(args[n].get_name());
                 auto out_buffer_index = external_function->get_buffer_index(out[0].get_name());
-                auto functor = [&, n, arg_buffer_index, out_buffer_index](
-                    CPURuntimeContext* ctx, CPUExecutionContext* ectx) {
+                auto functor = [&, arg_buffer_index, out_buffer_index](CPURuntimeContext* ctx,
+                                                                       CPUExecutionContext* ectx) {
                     if (ctx->buffer_data[arg_buffer_index] != ctx->buffer_data[out_buffer_index])
                     {
                         throw ngraph_error("GOE's input and out must be equal");

src/ngraph/runtime/cpu/cpu_op_annotations.cpp

@@ -30,7 +30,7 @@ namespace ngraph
                     []() -> std::shared_ptr<ngraph::op::util::OpAnnotations> {
                     auto op_annotations =
                         std::make_shared<ngraph::runtime::cpu::CPUOpAnnotations>();
-                    return op_annotations;
+                    return std::move(op_annotations);
                 };
                 return func;
             }

src/ngraph/runtime/cpu/mkldnn_emitter.hpp

@@ -73,7 +73,7 @@ namespace ngraph
 
             // TODO (nbpatel) Templatize the return type when we have double scales
             template <typename OP>
-            static std::vector<float> extract_scale_value(const ngraph::Node* node, int index)
+            inline std::vector<float> extract_scale_value(const ngraph::Node* node, int index)
             {
                 auto qc = static_cast<const OP*>(node);
                 std::vector<float> scale_val = {1.0f};

src/ngraph/runtime/cpu/pass/cpu_fusion.cpp

@@ -1680,7 +1680,7 @@ void ngraph::runtime::cpu::pass::CPUFusion::construct_groupconv_batchnorm_global
     auto bn =
         std::make_shared<ngraph::op::BatchNormInference>(eps, gamma, beta, conv_label, mean, var);
 
-    auto callback = [input, filters, conv_label, mean, var, gamma, beta, eps](pattern::Matcher& m) {
+    auto callback = [input, filters, conv_label, mean, var, gamma, beta](pattern::Matcher& m) {
 
         NGRAPH_DEBUG << "In callback for groupconv BatchNorm folding against node = "
                      << m.get_match_root()->get_name();
@@ -1838,7 +1838,7 @@ void ngraph::runtime::cpu::pass::CPUFusion::construct_deconvolution_affine_foldi
     double eps = 0.001;
     auto bn = std::make_shared<op::BatchNormInference>(eps, gamma, beta, conv_label, mean, var);
 
-    auto callback = [data_label, filters, out_delta, conv_label, mean, var, gamma, beta, eps](
+    auto callback = [data_label, filters, out_delta, conv_label, mean, var, gamma, beta](
         pattern::Matcher& m) {
         NGRAPH_DEBUG << "In callback for deconv affine folding against node = "
                      << m.get_match_root()->get_name();

src/ngraph/runtime/cpu/pass/cpu_mat_fusion.cpp

@@ -502,7 +502,7 @@ std::shared_ptr<Node> fuse_group_convolution(const std::shared_ptr<Node>& n)
                                                            sconv->get_data_dilation_strides(),
                                                            n->get_arguments().size());
 
-    return new_conv;
+    return move(new_conv);
 }
 
 std::shared_ptr<Node> fuse_batch_mat_mul_transpose(const std::shared_ptr<Node>& n)

src/ngraph/runtime/reference/topk.hpp

@@ -31,7 +31,7 @@ namespace ngraph
             // Had to split out these two functions. They used to be lambda expressions but
             // MSVC had difficulty compiling. This way is more explicit.
             template <typename T, typename U>
-            static bool compare_max(const std::tuple<T, U>& a, const std::tuple<T, U>& b)
+            inline bool compare_max(const std::tuple<T, U>& a, const std::tuple<T, U>& b)
             {
 // this is intentional to be able to compare floats directly
 // without using relative or absolute tolerance
@@ -46,7 +46,7 @@ namespace ngraph
                 return a > b;
             }
             template <typename T, typename U>
-            static bool compare_min(const std::tuple<T, U>& a, const std::tuple<T, U>& b)
+            inline bool compare_min(const std::tuple<T, U>& a, const std::tuple<T, U>& b)
             {
                 return a < b;
             }

src/ngraph/serializer.cpp

@@ -228,7 +228,7 @@ static json write_partial_shape(const PartialShape& s)
         {
             vals[i] = write_dimension(s[i]);
         }
-        return vals;
+        return move(vals);
     }
 }
 

src/ngraph/type/element_type.cpp

@@ -110,7 +110,7 @@ std::vector<const element::Type*> element::Type::get_known_types()
 element::Type::Type(
     size_t bitwidth, bool is_real, bool is_signed, bool is_quantized, const std::string& cname)
 {
-    for (const pair<element::Type_t, TypeInfo>& t : get_type_info_map())
+    for (auto& t : get_type_info_map())
     {
         const TypeInfo& info = t.second;
         if (bitwidth == info.m_bitwidth && is_real == info.m_is_real &&

src/tools/nbench/nbench.cpp

@@ -85,7 +85,7 @@ multimap<size_t, string> aggregate_timing_details(const vector<PerfShape>& perf_
     }
 
     multimap<size_t, string> rc;
-    for (const pair<string, size_t>& t : timing)
+    for (auto& t : timing)
     {
         rc.insert({t.second, t.first + to_string(count[t.first])});
     }
@@ -103,7 +103,7 @@ multimap<size_t, string> aggregate_timing(const vector<PerfShape>& perf_data)
     }
 
     multimap<size_t, string> rc;
-    for (const pair<string, size_t>& t : timing)
+    for (auto& t : timing)
     {
         rc.insert({t.second, t.first});
     }
@@ -115,7 +115,7 @@ void print_times(const multimap<size_t, string>& timing)
     // set the column widths
     int name_width = 0;
     int time_width = 0;
-    for (const pair<size_t, string>& p : timing)
+    for (auto& p : timing)
     {
         name_width = max(name_width, static_cast<int>(p.second.size()));
         time_width = max(time_width, static_cast<int>(locale_string(p.first).size()));
@@ -411,7 +411,7 @@ OPTIONS
                     cout << "    " << type << "\n";
                 }
                 cout << "--\n";
-                for (const pair<string, size_t>& op_info : op_list)
+                for (auto& op_info : op_list)
                 {
                     cout << op_info.first << ": " << op_info.second << " ops" << endl;
                 }

test/control_dependencies.cpp

@@ -50,7 +50,7 @@ public:
     virtual std::shared_ptr<Node> copy_with_new_args(const NodeVector& new_args) const override
     {
         auto clone = make_shared<ControlDependencyOp>(new_args, std::set<std::shared_ptr<Node>>{});
-        return clone;
+        return move(clone);
     }
 
     ControlDependencyOp(const NodeVector& args, const std::set<std::shared_ptr<Node>>& deps)

test/cpu_fusion.cpp

@@ -3206,7 +3206,7 @@ static std::shared_ptr<Function>
         auto dot = std::make_shared<op::Dot>(data_param_reshape, W_reshape);
         auto bias_broadcast = make_shared<op::Broadcast>(bias, dot->get_shape(), AxisSet{0});
         auto add_bias = std::make_shared<op::Add>(dot, bias_broadcast);
-        return add_bias;
+        return move(add_bias);
 
     };