[llvm] [AArch64] Avoid GPR trip when moving truncated i32 vector elements (PR #114541)

[Paul Walker via llvm-commits]

================
@@ -11,9 +11,10 @@ define <4 x float> @test(ptr %lhs_panel, ptr %rhs_panel, <4 x float> %a) {
 ; CHECK-NEXT:    .cfi_def_cfa_offset 16
 ; CHECK-NEXT:    .cfi_offset b8, -16
 ; CHECK-NEXT:    fmov x8, d0
+; CHECK-NEXT:    fmov w9, s0
 ; CHECK-NEXT:    ldr q16, [x1]
 ; CHECK-NEXT:    ldr q8, [x0]
-; CHECK-NEXT:    lsr x9, x8, #32
+; CHECK-NEXT:    lsr x8, x8, #32
----------------
paulwalker-arm wrote:

Do you know why this cases is worse?

https://github.com/llvm/llvm-project/pull/114541