FreePascal:

    procedure x;
    var a: integer;
       b: pinteger;
    begin
      b := @a;
      b^ := 1;
      b^ := 2;
    end;
    

becomes

    project1.lpr:15                           begin
    0000000000401090 50                       push   %rax
    project1.lpr:17                           b^ := 1;
    0000000000401091 c7042401000000           movl   $0x1,(%rsp)
    project1.lpr:18                           b^ := 2;
    0000000000401098 c7042402000000           movl   $0x2,(%rsp)
    project1.lpr:19                           end;
    000000000040109F 59                       pop    %rcx
    00000000004010A0 c3                       ret    

Perhaps Delphi, too

Although I do not know how much is by design and how much comes from bugs in the optimizer. But they said that one reason why they maintain their own optimizer rather than switching to LLVM. LLVM would "break" this function