[GPU] Fix a bug of remove redundant reorder pass (#7995)

+ Avoid dangling node exception

inference-engine/thirdparty/clDNN/src/graph_optimizer/remove_redundant_reorders.cpp

@@ -340,28 +340,28 @@ void remove_redundant_reorders::run(program& p) {
     }
 
     // Remove reorder for Convolution bfyx -> fs_b_yx_fsv32 (+ onednn: bfyx -> b_fs_yx_fsv32)
-    auto try_fuse_reorder_bfyx_to_fsv32 = [&](reorder_node* node) {
+    auto try_fuse_reorder_bfyx_to_fsv32 = [&](reorder_node* node) -> bool {
         if (node->get_users().size() != 1)
-            return;
+            return false;
 
         auto& usr = node->get_users().front();
         auto& dep = node->get_dependency(0);
         if (!(usr->is_type<convolution>()) ||
             usr->get_output_layout().data_type != dep.get_output_layout().data_type ||
             dep.get_output_layout().format != format::bfyx)
-            return;
+            return false;
         if (usr->as<convolution>().get_preferred_impl_type() != impl_types::onednn &&
             usr->get_output_layout().format != format::fs_b_yx_fsv32)
-            return;
+            return false;
         if (usr->as<convolution>().get_preferred_impl_type() == impl_types::onednn &&
             usr->get_output_layout().format != format::b_fs_yx_fsv32)
-            return;
+            return false;
 
         if (dep.is_type<input_layout>())
-            return;
+            return false;
 
         if (usr->as<convolution>().get_primitive()->groups != 1)
-            return;
+            return false;
 
         dep.merge_output_padding(node->get_output_layout().data_padding);
         p.replace_all_usages(*node, dep);
@@ -369,36 +369,37 @@ void remove_redundant_reorders::run(program& p) {
         p.add_optimized_primitive_info(node->id());
         p.remove_all_connections(*node);
         p.remove_if_dangling(*node);
+        return true;
     };
 
     // Remove reorder for Convolution b_fs_yx_fsv16 -> bfyx
-    auto try_fuse_reorder_fsv16_to_bfyx = [&](reorder_node* node) {
+    auto try_fuse_reorder_fsv16_to_bfyx = [&](reorder_node* node) -> bool {
         if (!node->get_fused_activations_funcs().empty() ||
             !node->get_fused_primitives().empty())
-            return;
+            return false;
 
         auto& input = node->input();
 
         if (!(input.is_type<convolution>()) ||
             !(input.get_output_layout().format == format::b_fs_yx_fsv16) ||
             !(node->get_output_layout().format == format::bfyx))
-            return;
+            return false;
 
         if (input.as<convolution>().get_primitive()->groups != 1)
-            return;
+            return false;
 
         // Avoid onednn convolution selects ref kernel for fsv16 -> bfyx
         if (input.as<convolution>().get_preferred_impl_type() == impl_types::onednn)
-            return;
+            return false;
 
         if (input.get_users().size() != 1)
-            return;
+            return false;
 
         auto& input_dep = input.get_dependency(0);
         if (input_dep.get_output_layout().format != format::b_fs_yx_fsv16 ||
             input_dep.get_output_layout().data_type == data_types::u8 ||
             input_dep.get_output_layout().data_type == data_types::i8)
-            return;
+            return false;
 
         for (auto& user : node->get_users()) {
             // if concat is reorder's user and concat's axis is 0(Batch) or 1(Feature), conv's output would have padding.
@@ -407,7 +408,7 @@ void remove_redundant_reorders::run(program& p) {
                 auto& concat_node = user->as<concatenation>();
                 auto concat_axis = concat_node.get_primitive()->axis;
                 if (concat_axis == 0 || concat_axis == 1)
-                    return;
+                    return false;
             }
         }
 
@@ -431,6 +432,7 @@ void remove_redundant_reorders::run(program& p) {
             p.add_optimized_primitive_info(node->id());
             p.extract_and_remove(*node);
         }
+        return true;
     };
 
     if (enable_reorder_fusing) {
@@ -446,9 +448,11 @@ void remove_redundant_reorders::run(program& p) {
             auto& r_node = node->as<reorder>();
 
             // Remove reorder for Convolution bfyx -> fs_b_yx_fsv32
-            try_fuse_reorder_bfyx_to_fsv32(&r_node);
+            if (try_fuse_reorder_bfyx_to_fsv32(&r_node))
+                continue;
             // Remove reorder for Convolution b_fs_yx_fsv16 -> bfyx
-            try_fuse_reorder_fsv16_to_bfyx(&r_node);
+            if (try_fuse_reorder_fsv16_to_bfyx(&r_node))
+                continue;
         }
     }