Array enumeration bug with opmisation level 2 for target 80386 ?
lazarus / 64-FPC development

----------------------------------------------------------------------
Nitorami · 2018-07-29 11:48

I have tracked down an obstinate bug in my program, and isolated it in
a small piece of code, see below. Can someone please confirm with
FPC3.0.4, win32, and the parameters -O2 -Op80386 -Mdelphi (or
-Mobjfpc)

Procedure TMyClass.TestModes runs a loop over the 4 array elements of
fOutputs and prints their length. The correct result should be
Quote10
10
10
10
But with optimization level 2, and optimization target processor
80386, I get this
Quote10
10
0
Runtime error 204 at $004034A0
  $004034A0
  $00401651
The issue is rather sensitive to small changes elsewhere in the code,
and may be somewhat random. I did not manage to reproduce it with
optimization off, nor when optimising for a different target processor
than 80386, so I guess it is a compiler bug.

Code: Pascal  [Select][+][-]window.onload = function(){var x1 =
document.getElementById("main_content_section"); if (x1) { var x =
document.getElementsByClassName("geshi");for (var i = 0; i type TPoint
= record x,y: integer; end; type  TOutput = object    OName:
Ansistring;    OConns  : integer;    TargetList    : array of TPoint; 
  procedure Init;  end;   TMyClass = Class     fOutputs: array [0..3]
of TOutput;     procedure TestModes;    constructor create; 
end; //-------------------------------procedure TOutput.Init;begin 
OName := '';  OConns := 0;  SetLength
(TargetList,10);end; //-------------------------------constructor
TMyClass.Create;var n: integer;begin  for n := 0 to high (fOutputs) do
fOutputs[n].Init;end; procedure TMyClass.TestModes;var  OP:
TOutput;begin  for OP in fOutputs do begin  //Enumeration seems to go
wrong    writeln (length(OP.Targetlist)); 
end;end; //-------------------------------var C1: TMyClass;begin  C1
:= TMyClass.Create;  C1.TestModes;  C1.Free;end.

----------------------------------------------------------------------
marcov · 2018-07-29 12:46

Yes. Trunk seems to have it also, please file it as bug on
bugs.freepascal.org

----------------------------------------------------------------------
ASerge · 2018-07-29 13:13

Confirm. Win32. Found the difference

No optimization, without -Op80386. OK
Code: ASM  [Select][+][-]window.onload = function(){var x1 =
document.getElementById("main_content_section"); if (x1) { var x =
document.getElementsByClassName("geshi");for (var i = 0; i No
optimization, without -Op80386:# [31] for n := 0 to high (fOutputs) do
fOutputs[n].Init;        movl    $0,-12(%ebp)        subl   
$1,-12(%ebp)        .balign 4,0x90.Lj40:        addl    $1,-12(%ebp) 
     movl    -8(%ebp),%edx        movl    -12(%ebp),%eax ; This OK   
   imull   $12,%eax,%eax        leal    4(%edx,%eax),%eax        call 
  P$PROGRAM$_$TOUTPUT_$__$$_INIT
No optimization, with -Op80386. OK
Code: ASM  [Select][+][-]window.onload = function(){var x1 =
document.getElementById("main_content_section"); if (x1) { var x =
document.getElementsByClassName("geshi");for (var i = 0; i # [31] for
n := 0 to high (fOutputs) do fOutputs[n].Init;        movl   
$0,-12(%ebp)        subl    $1,-12(%ebp)        .balign 4,0x90.Lj40: 
     addl    $1,-12(%ebp)        movl    -8(%ebp),%edx        movl   
-12(%ebp),%eax        imull   $12,%eax,%eax  ; This OK        leal   
4(%edx,%eax),%eax        call    P$PROGRAM$_$TOUTPUT_$__$$_INIT
Optimization, without -Op80386. OK
Code: ASM  [Select][+][-]window.onload = function(){var x1 =
document.getElementById("main_content_section"); if (x1) { var x =
document.getElementsByClassName("geshi");for (var i = 0; i # [31] for
n := 0 to high (fOutputs) do fOutputs[n].Init;        movl   
$0,-12(%ebp)        subl    $1,-12(%ebp)        .balign 4,0x90.Lj40: 
     addl    $1,-12(%ebp)        movl    -8(%ebp),%edx        movl   
-12(%ebp),%eax        imull   $12,%eax,%eax  ; This OK        leal   
4(%edx,%eax),%eax        call    P$PROGRAM$_$TOUTPUT_$__$$_INIT
Optimization, -Op80386. Error
Code: ASM  [Select][+][-]window.onload = function(){var x1 =
document.getElementById("main_content_section"); if (x1) { var x =
document.getElementsByClassName("geshi");for (var i = 0; i # [31] for
n := 0 to high (fOutputs) do fOutputs[n].Init;        movl   
$0,-12(%ebp)        subl    $1,-12(%ebp)        .balign 4,0x90.Lj40: 
     addl    $1,-12(%ebp)        movl    -8(%ebp),%edx        movl   
-12(%ebp),%eax        leal    (,%eax,4),%eax  ; This eax*4       
leal    (%eax,%eax,8),%eax; This eax*4 + (eax*4)*8 = eax*36       
leal    4(%edx,%eax),%eax        call   
P$PROGRAM$_$TOUTPUT_$__$$_INIT
With optimization and -Op80386 line 9,10 do multiplication by 36, not
by 12.

----------------------------------------------------------------------
Nitorami · 2018-07-29 13:23

Thanks. I'll file a bug report.

https://bugs.freepascal.org/view.php?id=34051