[Mlir-commits] [mlir] [mlir][AMDGPU] Add gfx950 MFMAs to the amdgpu.mfma op (PR #133553)

[Umang Yadav]

================
@@ -650,10 +650,12 @@ def AMDGPU_MFMAPermBAttr : EnumAttr<AMDGPU_Dialect, AMDGPU_MFMAPermB,
 // mfma
 def MFMAInTypes : AnyTypeOf<[F32, F64, I32, I64,
                              VectorOfLengthAndType<[2], [F32]>,
-                             VectorOfLengthAndType<[4], [F16]>,
-                             VectorOfLengthAndType<[2, 4], [BF16]>,
-                             VectorOfLengthAndType<[4, 8], [I8]>,
-                             VectorOfLengthAndType<[8], [F8E5M2FNUZ, F8E4M3FNUZ, F8E5M2, F8E4M3FN]>]>;
+                             VectorOfLengthAndType<[4, 8], [F16]>,
+                             VectorOfLengthAndType<[2, 4, 8], [BF16]>,
----------------
umangyadav wrote:

Changes looks good. Curious to know why Fp16 and BF16 support different vector lengths ? 

https://github.com/llvm/llvm-project/pull/133553