IntelGPU backend: ArgMax operation fix (#2413)

src/ngraph/runtime/intelgpu/intelgpu_op_custom_kernels.cpp

@@ -1345,7 +1345,7 @@ void runtime::intelgpu::do_arg_max_min_operation(cldnn::topology& topology,
     vector<size_t> gws;
 
     const string operation_sign = is_max ? " > " : " < ";
-    const string infinity = is_max ? "-INFINITY" : "INFINITY";
+    const string infinity = get_opencl_type_min_max_value(input_type, is_max);
     const string var_name = operation_name + "_val";
 
     size_t current_input = 0;

src/ngraph/runtime/intelgpu/unit_test.manifest

 all_2x2x3_eliminate_dims_0_1
-argmax_4D_axis_3_i64_in_i32
 argmin_trivial_in_double
-argmin_trivial_in_i32
 avg_pool_2d_2channel_2image_padded_only_above_do_not_include_in_computation
 avg_pool_2d_2channel_2image_padded_only_above_include_in_computation
 avg_pool_3d_uneven_strided_padded