[llvm] [AArch64] Improve cost model for legal subvec insert/extract (PR #81135)

[Sander de Smalen via llvm-commits]

================
@@ -45,27 +49,43 @@ define void @vector_insert_extract_idxzero_128b() #1 {
 ; TYPE_BASED_ONLY-NEXT:  Cost Model: Invalid cost for instruction: %insert_nxv16i1_nxv2i1 = call <vscale x 16 x i1> @llvm.vector.insert.nxv16i1.nxv2i1(<vscale x 16 x i1> undef, <vscale x 2 x i1> undef, i64 0)
 ; TYPE_BASED_ONLY-NEXT:  Cost Model: Invalid cost for instruction: %extract_nxv4i1_nxv16i1 = call <vscale x 4 x i1> @llvm.vector.extract.nxv4i1.nxv16i1(<vscale x 16 x i1> undef, i64 0)
 ; TYPE_BASED_ONLY-NEXT:  Cost Model: Invalid cost for instruction: %extract_v8i1_nxv8i1 = call <8 x i1> @llvm.vector.extract.v8i1.nxv8i1(<vscale x 8 x i1> undef, i64 0)
+; TYPE_BASED_ONLY-NEXT:  Cost Model: Invalid cost for instruction: %insert_v2f32_nxv2f32 = call <vscale x 2 x float> @llvm.vector.insert.nxv2f32.v2f32(<vscale x 2 x float> undef, <2 x float> undef, i64 0)
----------------
sdesmalen-arm wrote:

Perhaps not something to address in this patch, but I think it's wrong for this case to return an invalid cost, even when it cannot deduce anything for the operands. At worst it's a high cost for using spills and fills. But invalid cost basically says that we can't lower the general case, which isn't true.

https://github.com/llvm/llvm-project/pull/81135