[llvm] [AArch64] Combine concat(binop, binop) into binop(concat, concat) (PR #89911)

[Sjoerd Meijer via llvm-commits]

================
@@ -5,9 +5,13 @@
 define <8 x i16> @concat_add(<4 x i16> %a, <4 x i16> %b, <4 x i16> %c, <4 x i16> %d) {
 ; CHECK-LABEL: concat_add:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    add v2.4h, v2.4h, v3.4h
-; CHECK-NEXT:    add v0.4h, v0.4h, v1.4h
+; CHECK-NEXT:    // kill: def $d1 killed $d1 def $q1
+; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
+; CHECK-NEXT:    // kill: def $d3 killed $d3 def $q3
+; CHECK-NEXT:    // kill: def $d2 killed $d2 def $q2
+; CHECK-NEXT:    mov v1.d[1], v3.d[0]
 ; CHECK-NEXT:    mov v0.d[1], v2.d[0]
+; CHECK-NEXT:    add v0.8h, v0.8h, v1.8h
----------------
sjoerdmeijer wrote:

My guess is that this isn't faster than the old sequence because of the dependency. The other case below is clearly an improvement, so I guess overall this change is beneficial?

https://github.com/llvm/llvm-project/pull/89911