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

[Alexandra Munkes via llvm-commits]

================
@@ -1570,7 +1572,8 @@ def FeatureGFX12 : GCNSubtargetFeatureGeneration<"GFX12",
    FeatureIEEEMinimumMaximumInsts, FeatureSALUMinimumMaximumInsts,
    FeatureMinimum3Maximum3F32, FeatureMinimum3Maximum3F16,
    FeatureAgentScopeFineGrainedRemoteMemoryAtomics, FeatureFlatOffsetBits24,
-   FeatureFlatSignedOffset, FeatureInstCacheLineSize128
+   FeatureFlatSignedOffset, FeatureInstCacheLineSize128,
+   FeatureWMMA256bInsts
----------------
Malexandra-de wrote:

These instructions are not compatible with gfx1200 or gfx1201. The instructions specified by FeatureWMMA256bInsts require duplicated matrix elements:
https://github.com/llvm/llvm-project/blob/main/llvm/lib/Target/AMDGPU/AMDGPU.td#L832C6-L832C20
See also the ISA Guide for RDNA3:
https://docs.amd.com/api/khub/documents/UkT_UPQL21KfKAMUBFnZTw/content?Ft-Calling-App=ft%2Fturnkey-portal&Ft-Calling-App-Version=5.3.12#%5B%7B%22num%22%3A357%2C%22gen%22%3A0%7D%2C%7B%22name%22%3A%22XYZ%22%7D%2C0%2C129.86683%2Cnull%5D
Compare to the relevant section for RDNA4/gfx1200/gfx1201 where the duplicated elements are missing:
https://docs.amd.com/api/khub/documents/uQpkEvk3pv~kfAb2x~j4uw/content?Ft-Calling-App=ft%2Fturnkey-portal&Ft-Calling-App-Version=5.3.12#%5B%7B%22num%22%3A456%2C%22gen%22%3A0%7D%2C%7B%22name%22%3A%22XYZ%22%7D%2C0%2C547.122%2Cnull%5D

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