[GPU] GPU serialization fix (#14164)

* kernel_impl_params should be serialized before primitive_impl

* reduced the range of min max values ​​for this test case

* updated to use default get_mem when min max are not specified

src/plugins/intel_gpu/include/intel_gpu/graph/serialization/layout_serializer.hpp

@@ -32,7 +32,7 @@ public:
         size_t num_dimensions;
         buffer >> num_dimensions;
         for (size_t i = 0; i < num_dimensions; i++) {
-            ov::Dimension::value_type min_val, max_val;
+            ov::Interval::value_type min_val, max_val;
             buffer >> min_val >> max_val;
             partial_shape.push_back(ov::Dimension(min_val, max_val));
         }

src/plugins/intel_gpu/src/graph/primitive_inst.cpp

@@ -1059,14 +1059,17 @@ static primitive_id find_dep_by_mem(const cldnn::primitive_inst* p_inst, memory&
 }
 
 // Cache blob format:
-//     [ primitive_impl ]
 //     [ kernel_impl_params ]
+//     [ primitive_impl ]
 //     [ member variables of primitive_inst ]
 //     [ output memory information ]
 //     [ memory dependency information ]
 //     [ execution dependency information ]
 //     [ intermediate memory information ]
 void primitive_inst::save(cldnn::BinaryOutputBuffer& ob) const {
+    _impl_params->save(ob);
+    ob.setKernlImplParams(_impl_params.get());
+
     if (_impl != nullptr) {
         ob << true;
         kernel_arguments_data args = _impl->get_arguments(*this);
@@ -1078,9 +1081,6 @@ void primitive_inst::save(cldnn::BinaryOutputBuffer& ob) const {
         ob << false;
     }
 
-    _impl_params->save(ob);
-    ob.setKernlImplParams(_impl_params.get());
-
     ob << _node_output_layout;
     ob << has_mutable_input();
     ob << mem_allocated();
@@ -1169,6 +1169,11 @@ int32_t primitive_inst::get_index_in_deps(memory::cptr arg) const {
 }
 
 void primitive_inst::load(cldnn::BinaryInputBuffer& ib) {
+    _impl_params.release();
+    _impl_params = make_unique<kernel_impl_params>();
+    _impl_params->load(ib);
+    ib.setKernlImplParams(_impl_params.get());
+
     bool has_impl;
     ib >> has_impl;
     if (has_impl) {
@@ -1176,11 +1181,6 @@ void primitive_inst::load(cldnn::BinaryInputBuffer& ib) {
         ib >> _impl;
     }
 
-    _impl_params.release();
-    _impl_params = make_unique<kernel_impl_params>();
-    _impl_params->load(ib);
-    ib.setKernlImplParams(_impl_params.get());
-
     ib >> _node_output_layout;
     ib >> _has_mutable_input;
     ib >> _mem_allocated;

src/plugins/intel_gpu/tests/fusions/convolution_fusion_test.cpp

@@ -80,8 +80,13 @@ struct conv_eltw_test_params {
 
 class ConvFusingTest : public BaseFusingTest<convolution_test_params> {
 public:
-    void execute(convolution_test_params& p) {
-        auto input_prim = get_mem(get_input_layout(p));
+    void execute(convolution_test_params& p, int min=0, int max=0) {
+        cldnn::memory::ptr input_prim;
+        if (min == max) {
+            input_prim = get_mem(get_input_layout(p));
+        } else {
+            input_prim = get_mem(get_input_layout(p), min, max);
+        }
         network network_not_fused(this->engine, this->topology_non_fused, bo_not_fused);
         network network_fused(this->engine, this->topology_fused, bo_fused);
         network_fused.set_input_data("input", input_prim);
@@ -1744,7 +1749,7 @@ TEST_P(conv_int8_scale_shift_swish, basic) {
 
     // high tolerance because many eltwise operations
     tolerance = default_tolerance(p.default_type) * 10;
-    execute(p);
+    execute(p, -20, 20);
 }
 
 INSTANTIATE_TEST_SUITE_P(fusings_gpu, conv_int8_scale_shift_swish, ::testing::ValuesIn(std::vector<convolution_test_params>{