Compile-time local memory usage

Hello,
I was wondering how we could query for the local memory usage done by an implementation. As an example, when using group algorithms, DPC++ will reserve a certain amount of local memory to establish inter-sub-group communication. One the CUDA back-end, that value is properly reported by `ptxas` when compiling the kernel.
Maybe we could introduce a kernel query such as `sycl::info::kernel_device_specific::private_mem_size`, but for local Memory ? Relying only on `sycl::info::device::local_mem_size` in a program can be unsafe.