[llvm] Scalarize binop of splats build vector/free (PR #65773)

[Matt Arsenault via llvm-commits]

================
@@ -95,9 +95,10 @@ define amdgpu_kernel void @truncate_high_elt_extract_vector(ptr addrspace(1) noc
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
 ; SI-NEXT:    s_sext_i32_i16 s4, s4
 ; SI-NEXT:    s_sext_i32_i16 s5, s5
-; SI-NEXT:    s_mul_i32 s5, s5, s4
-; SI-NEXT:    s_lshr_b32 s4, s5, 16
+; SI-NEXT:    s_mul_i32 s4, s5, s4
+; SI-NEXT:    s_lshr_b32 s5, s4, 16
 ; SI-NEXT:    v_mov_b32_e32 v0, s4
+; SI-NEXT:    v_alignbit_b32 v0, s5, v0, 16
----------------
arsenm wrote:

Regression 

https://github.com/llvm/llvm-project/pull/65773