[llvm] [CodeGen] Teach ReplaceWithVeclib to use correct calling convention (PR #180773)

[Graham Hunter via llvm-commits]

================
@@ -778,6 +778,9 @@ define <vscale x 4 x float> @llvm_tanh_vscale_f32(<vscale x 4 x float> %in) #0 {
   ret <vscale x 4 x float> %1
 }
 
+declare aarch64_vector_pcs <4 x float> @armpl_vpowq_f32(<4 x float>, <4 x float>)
----------------
huntergr-arm wrote:

Yes, I think adding the injection pass would be a good idea. We'd still have the scalable tests demonstrating what happens when you don't have a CC override at the mapping level.

https://github.com/llvm/llvm-project/pull/180773