Using madExcept seems to slow down my app

[csterg]

Hello,
i have been testing the madExcept package along with the FastMM4. It seems that when i use madExcept, the app is being slowed down about 10% (almost the gain i have from FastMM4).
Why is this?
Does madExcept play around with the Memory Manager? Does it use it's own MM?
Thanks
Costas

[madshi]

madExcept doesn't use it's own MM nor does it play around with the memory manager. There are 2 things in madExcept which may affect performance:

(1) Using "RegisterHiddenExceptionHandler" costs performance.
(2) The creation and termination of new threads costs a little bit performance, but not much.

Do you have a simple test project with which I can check the performance problems you're seeing?

Thanks...

[csterg]

Thanks for your fast response: i was expecting to be notified of it by email but somehow i didn't get it so i am back 3 days later... sorry (and i was anxious for the response )

Anyway, let me paste a small example here:

Do you have a simple test project with which I can check the performance problems you're seeing?
Thanks...

Test project:

Code: Select all

procedure TForm1.Button1Click(Sender: TObject);
var
  now: Cardinal;
  s: TStringList;
  i: Integer;
begin
  now := GetTickCount;
  for i:=0 to 1000000 do begin
    s := TStringList.Create;
    s.Add('test');
    s.Free;
  end;
  ShowMessage(IntToStr(GetTickCount-now));
end;

Results (2GHz PC):
Default MM, no madExcept: 1250ms
Default MM, with madExcept: 2550ms
FastMM4, no madExcept: 920ms
FastMM4, with madExcept: 1250ms

In all cases there is optimization and no checks.

Thanks,
Costas

[madshi]

I've just checked out the situation and the reason for the speed loss is quite simple. Try this code *without* madExcept:

Code: Select all

procedure TForm1.Button1Click(Sender: TObject);
var
  now: Cardinal;
  s: TStringList;
  i: Integer;
begin
  IsMultiThread := true;
  now := GetTickCount;
  for i:=0 to 1000000 do begin
    s := TStringList.Create;
    s.Add('test');
    s.Free;
  end;
  ShowMessage(IntToStr(GetTickCount-now));
end;

[csterg]

Well, i see EXACTLY your point: never thought of that
As I understand, madExcept creates an additional thread (for watching) and this sets the IsMultiThread var (in TThread class?), so when i use the madExcept i have the nominal overhead that comes in memory management for multi-threaded apps.

OK, this is perfectly acceptable and understandable: it is very good though that FastMM4 has a 50% increase in these cases (multi-threaded apps), so for multi-threaded apps there is no real 'penalty' for using madExcept (and FastMM4 really compensates in these cases).

Costas

[madshi]

Yes, you got it right. madExcept doesn't cost performance. However, madExcept switches the memory manager to multi threading mode and that is what costs performance.

[madshi]

P.S: Honestly, I was not really aware of this. I naively thought all (or at least most) programs today would be using multiple threads, anyway... But I have to admit, there are probably a lot of programs not using any threads. And madExcept will then cost performance because of the IsMultiThread stuff. But well, there's nothing I can do about that, since madExcept absolutely does need threads.

[csterg]

Well, this is how the combination of madExcept with FastMM4 can persuade even those that build single-threaded apps that madExcept really 'costs' nothing in terms of performance. But the penalty otherwise, is quite big (about doubling the response).
Anyway, a big thanks for this awesome product which really does a terrific job!
Costas