[llvm] [AMDGPU] Split struct kernel arguments (PR #133786)

[Yaxun Liu via llvm-commits]

yxsamliu wrote:

> I'd rather avoid having an ABI breaking pass, and I definitely want to avoid spreading the handling of non-byref kernel arguments. This has overlap with the existing IR expansion of non-byref arguments

For HIP programs that launches the kernel, there is no ABI change since the arguments passed to the kernel launching triple chevron do not change. This is an internal optimization of kernel launching procedure by skipping unused fields in struct type kernel arguments so that only used fields are preloaded. In a sense, it is similar to the arguments promotion IPO, but happening between the compiler and the runtime through the newly added kernel argument metadata. Basically, it tells runtime which chunk of kernel arguments are to be kept and then preloaded. Since the optimization only happens late in the LLVM pipeline, it won't proliferate non-byref kernel arguments in FE or middle end.

https://github.com/llvm/llvm-project/pull/133786