[llvm] [AMDGPU][ASAN] Handle special GVs lowering in amdgpu-sw-lower-lds (PR #161827)

[Jay Foad via llvm-commits]

jayfoad wrote:

> I don't know what "special" refers to here. As far as I'm aware there are no lds variables that are treated inherently differently to others, except for the dynamic case which has the weird aliasing behaviour.

See #114550. Named barriers are represented in IR as addrspace(3) globals with a special type. During codegen they go through special handling which maps them onto a small fixed number of hardware barrier IDs, _instead_ of getting space allocated in LDS. Downstream there are other use cases that use a very similar mechanism.

https://github.com/llvm/llvm-project/pull/161827