[clang] [llvm] [IPO] Optimise variadic functions (PR #92850)

[Jon Chesterfield via cfe-commits]

JonChesterfield wrote:

[inline-then-fold-variadics.cpp](https://github.com/llvm/llvm-project/pull/92850/commits/15061bfbc2dc06de5bac32628389386cadaa5632#diff-0a9893e04ae7e0a5692ad93dfb73d6efa992953f7e9eebb68c1a3f4acd457e1e) is the motivating example for optimisation - simple variadic functions are removed entirely. Variadic logging wrappers around vfprintf and similar are similarly inlined.

Wasm might choose to use a call to this to replace the backend handling, they should each build the same frame passed by pointer. That would help offset the increase in compiler complexity from this patch. AMDGPU has a requirement for variadic function lowering, at least sufficient for libc. It is much easier to debug in wasm or x64 than on the GPU so I'm hoping this can land as a target independent pass and not end up in a GPU backend.

https://github.com/llvm/llvm-project/pull/92850