[Mlir-commits] [mlir] [mlir][gpu] Allow insert/extract of signed ints in gpu.mma_matrix (PR #221800)

[Vito Secona]

================
@@ -38,6 +38,12 @@ include "mlir/Interfaces/SideEffectInterfaces.td"
 class GPU_Op<string mnemonic, list<Trait> traits = []> :
     Op<GPU_Dialect, mnemonic, traits>;
 
+// Scalar element types supported by gpu.mma_matrix
+def GPU_MMAMatrixElementType :
+    AnyTypeOf<[SI8, UI8, AnyI32, F16, F32, F64]> {
+  let cppFunctionName = "isValidMMAMatrixElementType";
+}
+
----------------
secona wrote:

I think this'd be better in GPUBase.td instead

https://github.com/llvm/llvm-project/pull/221800