[Mlir-commits] [mlir] [mlir][arith] add wide integer emulation support for subi (PR #133248)

[Jakub Kuderski]

================
@@ -130,6 +130,44 @@ func.func @addi_vector_a_b(%a : vector<4xi64>, %b : vector<4xi64>) -> vector<4xi
     return %x : vector<4xi64>
 }
 
+// CHECK-LABEL: func @subi_scalar_a_b
+// CHECK-SAME:    ([[ARG0:%.+]]: vector<2xi32>, [[ARG1:%.+]]: vector<2xi32>) -> vector<2xi32>
+// CHECK-NEXT:    [[LOW0:%.+]]   = vector.extract [[ARG0]][0] : i32 from vector<2xi32>
+// CHECK-NEXT:    [[HIGH0:%.+]]  = vector.extract [[ARG0]][1] : i32 from vector<2xi32>
+// CHECK-NEXT:    [[LOW1:%.+]]   = vector.extract [[ARG1]][0] : i32 from vector<2xi32>
+// CHECK-NEXT:    [[HIGH1:%.+]]  = vector.extract [[ARG1]][1] : i32 from vector<2xi32>
+// CHECK-NEXT:    [[SUB_L:%.+]]  = arith.subi [[LOW0]], [[LOW1]] : i32
+// CHECK-NEXT:    [[ULT:%.+]]    = arith.cmpi ult, [[LOW0]], [[LOW1]] : i32
+// CHECK-NEXT:    [[CARRY:%.+]]  = arith.extui [[ULT]] : i1 to i32
+// CHECK-NEXT:    [[SUB_H0:%.+]] = arith.subi [[HIGH0]], [[CARRY]] : i32
+// CHECK-NEXT:    [[SUB_H1:%.+]] = arith.subi [[SUB_H0]], [[HIGH1]] : i32
+// CHECK:         [[INS0:%.+]]   = vector.insert [[SUB_L]], {{%.+}} [0] : i32 into vector<2xi32>
+// CHECK-NEXT:    [[INS1:%.+]]   = vector.insert [[SUB_H1]], [[INS0]] [1] : i32 into vector<2xi32>
+// CHECK-NEXT:    return [[INS1]] : vector<2xi32>
+func.func @subi_scalar_a_b(%a : i64, %b : i64) -> i64 {
----------------
kuhar wrote:

nit: I don't think we need the `_a_b` suffix

https://github.com/llvm/llvm-project/pull/133248