[Mlir-commits] [mlir] [mlir][Vector] Remove more special case uses for extractelement/insertelement (PR #130166)

[Kunwar Grover]

================
@@ -187,7 +187,7 @@ func.func @broadcast_vec2d_from_vec0d(%arg0: vector<f32>) -> vector<3x2xf32> {
 //       CHECK: %[[T0:.*]] = builtin.unrealized_conversion_cast %[[A]] : vector<f32> to vector<1xf32>
 //       CHECK: %[[T1:.*]] = ub.poison : vector<3x2xf32>
 //       CHECK: %[[T2:.*]] = builtin.unrealized_conversion_cast %[[T1]] : vector<3x2xf32> to !llvm.array<3 x vector<2xf32>>
-//       CHECK: %[[T4:.*]] = llvm.mlir.constant(0 : index) : i64
+//       CHECK: %[[T4:.*]] = llvm.mlir.constant(0 : i64) : i64
----------------
Groverkss wrote:

This is not because of the index type, but the lowering difference between vector.extract and vector.extractelement. vector.extract lowering to llvm is better, and actually uses the type converter on index while lowering. vector.extractelement doesn't and is a worse lowering. This is just a side effect of vector.extract lowering being better.

https://github.com/llvm/llvm-project/pull/130166