Index: llvm/test/CodeGen/AArch64/bitfield-insert.ll
===================================================================
--- llvm/test/CodeGen/AArch64/bitfield-insert.ll
+++ llvm/test/CodeGen/AArch64/bitfield-insert.ll
@@ -285,6 +285,36 @@
   ret i32 %shl.4
 }
 
+; Optimal codegen is to use bfi, which simplifies away two instrutions (%mask and %bit-field-pos-op).
+; As a comparsion, 'test_orr_not_bfi' shows when orr is better than bfi.
+define i64 @test_bfi_not_orr(i64 %0, i64 %1) {
+; CHECK-LABEL: test_bfi_not_orr:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    and x8, x1, #0xff
+; CHECK-NEXT:    bfi x8, x0, #8, #8
+; CHECK-NEXT:    mov x0, x8
+; CHECK-NEXT:    ret
+  %bfi_dst = and i64 %1, 255
+  %mask = and i64 %0, 255
+  %bit-field-pos-op = shl i64 %mask, 8
+  %or_res = or i64 %bit-field-pos-op, %bfi_dst
+  ret i64 %or_res
+}
+
+; Optimal codegen should fold the left shift (%3) into orr.
+define i64 @test_orr_not_bfi(i64 %0) {
+; CHECK-LABEL: test_orr_not_bfi:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    and x8, x0, #0xff
+; CHECK-NEXT:    bfi x8, x0, #8, #8
+; CHECK-NEXT:    mov x0, x8
+; CHECK-NEXT:    ret
+  %2 = and i64 %0, 255
+  %3 = shl i64 %2, 8
+  %4 = or i64 %2, %3
+  ret i64 %4
+}
+
 define void @test_nouseful_strb(i32* %ptr32, i8* %ptr8, i32 %x)  {
 ; CHECK-LABEL: test_nouseful_strb:
 ; CHECK:       // %bb.0: // %entry