[llvm] [AMDGPU] Optimize LDS DMA soft waitcnt (PR #138802)

[Pierre van Houtryve via llvm-commits]

https://github.com/Pierre-vh commented:

This hasn't fully clicked for me yet, so bear with me a bit.

Do I understand correctly that the end goal here to change the rule of synchronization to also include global->lds direct loads, so that they cannot reorder with normal LDS load?

Why do we suddenly need to do that ? Is this a tailored change for a specific case?
I'd like to see the reasoning in memory model terms as to why global->lds loads should be considered as normal loads and fall under the usual synchronize-with rules (and thus a new wait is needed).

https://github.com/llvm/llvm-project/pull/138802