[Mlir-commits] [mlir] [MLIR][AMDGPU] Add a wrapper for global LDS load intrinsics in AMDGPU (PR #133498)

[Krzysztof Drewniak]

================
@@ -668,6 +668,12 @@ def WMMAInTypes : AnyTypeOf<[VectorOfLengthAndType<
 def WMMAOutTypes : AnyTypeOf<[VectorOfLengthAndType<[4, 8], [F32, I32]>,
                               VectorOfLengthAndType<[4, 8, 16], [F16, BF16]>]>;
 
+// MI300: limit to only 4 bytes per elements.
+def GlobalLoadTypes : AnyTypeOf<[F16, F32, I8, SI8, UI8, I16, I32,
----------------
krzysz00 wrote:

Let's not do that

This works for 1, 2, or 4 bytes, and can be a check in `::verify`

https://github.com/llvm/llvm-project/pull/133498