[Mlir-commits] [mlir] Adds Arm Neon SMMLA, UMMLA, and USMMLA Intrinsics (PR #80511)

[Diego Caballero]

================
@@ -31,3 +31,69 @@ func.func @b_has_2_rows_but_a_has_length_4(%a : vector<4xi32>, %b : vector<2x4xi
     %0 = arm_neon.2d.sdot %a, %b, %b : vector<2x4xi8>, vector<2x4xi8> to vector<4xi32>
     return %0 : vector<4xi32>
 }
+
+// -----
+
+func.func @smmla_invalid_input_types(%a: vector<16xi4>,
+                    %b: vector<16xi4>,
+                    %c: vector<4xi32>) -> vector<4xi32> {
+  // expected-error at +1 {{op operand #1 must be vector of 8-bit signless integer values of length 16, but got 'vector<16xi4>'}}
+  %0 = arm_neon.intr.smmla %c, %a, %b :
+             vector<16xi4> to vector<4xi32>
----------------
dcaballe wrote:

nit: single line?

https://github.com/llvm/llvm-project/pull/80511