[llvm] [AMDGPU] Optimize rotate/funnel shift pattern matching in instruction selection (PR #149817)

[Matt Arsenault via llvm-commits]

================
@@ -430,19 +433,20 @@ define amdgpu_kernel void @s_insertelement_v2i16_0_reghi_both_multi_use_1(ptr ad
 ; CI-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    s_load_dword s2, s[2:3], 0x0
-; CI-NEXT:    v_mov_b32_e32 v1, s1
-; CI-NEXT:    v_mov_b32_e32 v2, s4
 ; CI-NEXT:    v_mov_b32_e32 v0, s0
+; CI-NEXT:    v_mov_b32_e32 v1, s1
 ; CI-NEXT:    s_lshr_b32 s0, s4, 16
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
-; CI-NEXT:    s_lshr_b32 s1, s2, 16
-; CI-NEXT:    v_alignbit_b32 v2, s1, v2, 16
+; CI-NEXT:    s_and_b32 s1, s2, 0xffff0000
+; CI-NEXT:    s_or_b32 s1, s0, s1
+; CI-NEXT:    v_mov_b32_e32 v2, s1
+; CI-NEXT:    s_lshr_b32 s2, s2, 16
----------------
arsenm wrote:

Seems like a regression? 

https://github.com/llvm/llvm-project/pull/149817