[llvm] [AMDGPU] Promote uniform ops to i32 in GISel (PR #106557)

[Matt Arsenault via llvm-commits]

================
@@ -347,7 +347,8 @@ define amdgpu_ps i8 @s_fshr_i8(i8 inreg %lhs, i8 inreg %rhs, i8 inreg %amt) {
 ;
 ; GFX8-LABEL: s_fshr_i8:
 ; GFX8:       ; %bb.0:
-; GFX8-NEXT:    s_andn2_b32 s3, 7, s2
+; GFX8-NEXT:    s_xor_b32 s3, s2, -1
+; GFX8-NEXT:    s_and_b32 s3, s3, 7
 ; GFX8-NEXT:    s_and_b32 s2, s2, 7
----------------
arsenm wrote:

This is worse 

https://github.com/llvm/llvm-project/pull/106557