[clang] [llvm] clang: Use TargetID parsing from AMDGPUTargetParser (PR #209845)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> It seems to fail the check-offload tests. Do we need any other changes? Thanks!
> 
> ```
> clang: error: invalid target ID 'gfx90a'; format is a processor name followed by an optional colon-delimited list of features followed by an enable/disable sign (e.g., 'gfx908:sramecc+:xnack-')
> ```
> 
> https://lab.llvm.org/buildbot/#/builders/234/builds/1391

The offload tests seem to have been only partially / prematurely converted to use the new triples. They're mixing "amdgpu-amd-amdhsa" with a separate -mcpu. The parser expects if you're using the new amdgpu name, you must have a valid subarch. e.g., here fails using -fopenmp-targets=amdgpu-amd-amdhsa -mcpu=gfx90a. These tests should have waited for #206482

https://github.com/llvm/llvm-project/pull/209845