[GPU] Reshape keeps data padding of padded input (#23880)

### Details:
 - *Fix to keep reshape data padding of padded input*

### Tickets:
 - *137553*

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

@@ -399,7 +399,7 @@ void primitive_inst::update_shape() {
     _impl_params->memory_deps = memory_deps;
 
     auto update_output_layout = [&](layout& layout, size_t idx) {
-        if (!_node->is_type<reshape>()) {
+        if (!_node->is_type<reshape>() || (!_node->get_input_layout(0).has_dynamic_pad() && !_node->can_be_optimized())) {
             auto data_padding = padding::max(_impl_params->get_output_layout(idx).data_padding, layout.data_padding);
             layout.data_padding = padding::max(_node->get_primitive()->get_output_padding(idx), data_padding);
         }

src/plugins/intel_gpu/tests/unit/test_cases/reshape_gpu_test.cpp

@@ -1568,3 +1568,123 @@ TEST(reshape_gpu_f32, shrink_chain_full_cached) {
 TEST(reshape_gpu_f32, shrink_chain_out_cached) {
     test_shrink_chain_out<float>(true);
 }
+
+TEST(reshape_gpu_f32, followed_by_convolution_dynamic) {
+    auto& engine = get_test_engine();
+
+    ov::Shape in0_shape = { 1, 1, 4, 5 };
+
+    auto in0_dyn_layout = layout{ov::PartialShape::dynamic(in0_shape.size()), data_types::f32, format::bfyx};
+    auto weights = engine.allocate_memory({ data_types::f32, format::bfyx, { 1, 1, 3, 2 } });
+
+    set_values(weights, {
+        1.0f, 2.0f, 1.0f,
+        2.0f, 1.0f, 2.0f
+    });
+
+    topology topology(
+        input_layout("input", in0_dyn_layout),
+        shape_of("shape_of_input", input_info("input"), data_types::i32),
+        reshape("reshape", input_info("input"), input_info("shape_of_input"), false, ov::PartialShape::dynamic(4),
+                cldnn::reshape::reshape_mode::base, padding({0, 0, 1, 1}, {0, 0, 1, 1})),
+        data("weights", weights),
+        convolution("conv", input_info("reshape"), "weights", "", 1, { 2, 1 }, {1, 1}, {0, 0}, {0, 0}, false));
+
+    ExecutionConfig config = get_test_default_config(engine);
+    config.set_property(ov::intel_gpu::allow_new_shape_infer(true));
+    network network(engine, topology, config);
+
+    // first execute
+    {
+        auto input0 = engine.allocate_memory({ in0_shape, data_types::f32, format::bfyx });
+        set_values(input0, {
+            1.0f, 2.0f, 3.0f, 4.0f, 5.0f,
+            2.0f, 2.0f, 3.0f, 4.0f, 6.0f,
+            3.0f, 3.0f, 3.0f, 5.0f, 1.0f,
+            1.0f, 1.0f, 1.0f, 1.0f, 1.0f
+        });
+        network.set_input_data("input", input0);
+
+        auto inst = network.get_primitive("conv");
+        auto impl = inst->get_impl();
+        ASSERT_TRUE(impl != nullptr);
+        ASSERT_TRUE(impl->is_dynamic());
+
+        auto outputs = network.execute();
+        ASSERT_EQ(outputs.size(), size_t(1));
+        ASSERT_EQ(outputs.begin()->first, "conv");
+
+        auto output_memory = outputs.at("conv").get_memory();
+        auto output_layout = output_memory->get_layout();
+        cldnn::mem_lock<float> output_ptr(output_memory, get_test_stream());
+
+        int y_size = output_layout.spatial(1);
+        int x_size = output_layout.spatial(0);
+        int f_size = output_layout.feature();
+        int b_size = output_layout.batch();
+        ASSERT_EQ(output_layout.format, format::bfyx);
+        ASSERT_EQ(y_size, 2);
+        ASSERT_EQ(x_size, 3);
+        ASSERT_EQ(f_size, 1);
+        ASSERT_EQ(b_size, 1);
+
+        VVF<float> output_vec = {
+            { 20.0f, 27.0f, 38.0f },
+            { 17.0f, 19.0f, 19.0f }
+        };
+        for (int y = 0; y < y_size; ++y) {
+            for (int x = 0; x < x_size; ++x) {
+                ASSERT_EQ(output_vec[y][x], output_ptr[y * x_size + x]);
+            }
+        }
+    }
+
+    // second execute
+    {
+        in0_shape = { 1, 1, 6, 4 };
+        auto input0 = engine.allocate_memory({ in0_shape, data_types::f32, format::bfyx });
+        set_values(input0, {
+            1.0f, 2.0f, 3.0f, 4.0f,
+            2.0f, 2.0f, 3.0f, 4.0f,
+            3.0f, 3.0f, 3.0f, 5.0f,
+            1.0f, 1.0f, 1.0f, 1.0f,
+            5.0f, 4.0f, 3.0f, 2.0f,
+            4.0f, 4.0f, 3.0f, 3.0f,
+        });
+        network.set_input_data("input", input0);
+
+        auto inst = network.get_primitive("conv");
+        auto impl = inst->get_impl();
+        ASSERT_TRUE(impl != nullptr);
+        ASSERT_TRUE(impl->is_dynamic());
+
+        auto outputs = network.execute();
+        ASSERT_EQ(outputs.size(), size_t(1));
+        ASSERT_EQ(outputs.begin()->first, "conv");
+
+        auto output_memory = outputs.at("conv").get_memory();
+        auto output_layout = output_memory->get_layout();
+        cldnn::mem_lock<float> output_ptr(output_memory, get_test_stream());
+
+        int y_size = output_layout.spatial(1);
+        int x_size = output_layout.spatial(0);
+        int f_size = output_layout.feature();
+        int b_size = output_layout.batch();
+        ASSERT_EQ(output_layout.format, format::bfyx);
+        ASSERT_EQ(y_size, 3);
+        ASSERT_EQ(x_size, 2);
+        ASSERT_EQ(f_size, 1);
+        ASSERT_EQ(b_size, 1);
+
+        VVF<float> output_vec = {
+            { 20.f, 27.f },
+            { 17.f, 19.f },
+            { 34.f, 29.f }
+        };
+        for (int y = 0; y < y_size; ++y) {
+            for (int x = 0; x < x_size; ++x) {
+                ASSERT_EQ(output_vec[y][x], output_ptr[y * x_size + x]);
+            }
+        }
+    }
+}