[clang] [AMDGPU] Mark int64_t and intmax_t as LP64 (PR #214865)

[Yaxun Liu via cfe-commits]

yxsamliu wrote:

This looks right for standalone AMDGCN, but I’m worried about the aux-target path.

For HIP/OpenMP device compilation, `AMDGPUTargetInfo::setAuxTarget()` later calls `copyAuxTarget(Aux)`. My understanding is that the purpose of this is to keep host and device data layout consistent. It copies fields like `LongWidth`, but it does not copy `Int64Type` or `IntMaxType`. So with an aux target like `x86_64-pc-windows-msvc`, we could end up with `long` copied back to 32 bits while `Int64Type` still says `SignedLong`.

Could we either update `setAuxTarget()` to keep these fields consistent with `Aux`, or add a test showing the aux-target path is unchanged?


https://github.com/llvm/llvm-project/pull/214865