[llvm] [AMDGPU] Introduce "amdgpu-sw-lower-lds" pass to lower LDS accesses. (PR #87265)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> Each workgroup needs a different pointer, and a grid can have a lot of work groups. I don't think a global would work.

So why isn't the runtime responsible for setting up this pointer then? It does that effectively for LDS, the allocation is managed as part of the dispatch. That also goes back to my question of why we need to insert explicit free code, instead of just letting the runtime clean it up after as it would need to anyway 



https://github.com/llvm/llvm-project/pull/87265