The current way to detect hostcalls by looking for "__ockl_hostcall_internal()" function in the module seems to be not reliable enough. The LTO may rename the "__ockl_hostcall_internal()" function when an application is compiled with "-fgpu-rdc", and MetadataStreamer pass to fail to detect hostcalls, therefore it does not set the "hidden_hostcall_buffer" kernel argument.
This change adds new function attribute: "amdgpu-hostcalls" that can be used to detect whether GPU functions use any host calls.