[Mlir-commits] [mlir] [mlir][ArmSME] Support lowering masked vector.outerproduct ops to SME (PR #69604)

[Benjamin Maxwell]

================
@@ -97,3 +97,32 @@ func.func @arm_sme_move_tile_slice_to_vector__bad_result_type(%tile : vector<[4]
   %0 = arm_sme.move_tile_slice_to_vector %tile[%tile_slice_index] : vector<[2]xf64> from vector<[4]x[4]xf32>
   return %0 : vector<[2]xf64>
 }
+
+// -----
+
+// expected-note at +1 {{prior use here}}
+func.func @arm_sme_outproduct__bad_mask_type(%vecA: vector<3xf32>, %vecB: vector<[2]xf32>, %maskA: vector<5xi1>, %maskB: vector<[2]xi1>)-> vector<3x[2]xf32>
+{
+  // expected-error at +1 {{use of value '%maskA' expects different type than prior uses}}
+  %0 = arm_sme.outerproduct %vecA, %vecB masks(%maskA, %maskB) : vector<3xf32>, vector<[2]xf32>, vector<3x[2]xf32>
+  return %0 : vector<3x[2]xf32>
+}
----------------
MacDue wrote:

It is an attempt to test the constraint used to infer the types of these masks, I couldn't figure out how to get it to use the custom error. 

https://github.com/llvm/llvm-project/pull/69604