[llvm] [AMDGPU] Enable WMMA256bInsts + Wave32 for gfx1200/gfx1201 + SISchedule fix + TargetParser gfx1200 propagation (PR #202093)

[Krzysztof Drewniak via llvm-commits]

krzysz00 wrote:

... I don't know what these "4326" changes are

I want to see _something_ showing _why_ this change should be made. Given the long history of issues on this PR where you didn't understand what your patch was doing / you having things like the Wavefrontsize32 removal that could have led to incorrect codegen. 

In addition, @Malexandra-de  has noted that there are sources that appear to contradict the scheduling change you're trying to make, and I'd like you to address those.

I'm not objecting to the fact that tests changed - that's what's supposed to happen when the code generated by the compiler changes. I also don't think those changes are incorrect, necessarily.

However, I still want you to show us **why** you are trying to make this change. What has caused you to think that this patch is needed? 

I am considering flagging this PR as extractive and resigning as a reviewer if you don't make it much clearer and easier to understand why this change was necessary and what its impacts are.

https://github.com/llvm/llvm-project/pull/202093