VBCC 09d - bug in code generator ?

[Asman]

Hi

I have problem with vbcc 0.9d (cross - I compiled my own 64 bit version for windows). For following function (simplified a bit to show problem)

void WindowSignals(void)
{
	while (TRUE)
	{
		struct IntuiMessage* pMsg;
		pMsg = (struct IntuiMessage*)GetMsg(g_pWin->UserPort);

		if (pMsg == NULL)
		{
			break;
		}

		ULONG msg_class = pMsg->Class;

		ReplyMsg((struct Message*)pMsg);

		if (IDCMP_CLOSEWINDOW == msg_class)
		{
			return;
		}
	}
}

I have following assembly code

_WindowSignals
	movem.l	l32,-(a7)
l30
	move.l	_g_pWin(a4),a1
	move.l	(86,a1),a1
	move.l	_SysBase(a4),a6
	move.l	a1,a0
	jsr	-372(a6)
	move.l	d0,a3
	beq	l31
	move.l	(20,a3),a2
	move.l	_SysBase(a4),a6
	move.l	a3,a1
	jsr	-378(a6)
	cmp.l	#512,a2
	bne	l30
l31
l32	reg	a2/a3/a6
	movem.l	(a7)+,a2/a3/a6
l34	equ	12
	rts

And of course this code hangs because beq l31 never happens.
For vbcc 0.9b code generated is differ then above

	move.l	d0,a3
	move.l a3,d0
	beq	l31

and works fine.

Anyone noticed this bug too ?

[Leffmann]

Yeah I see it here too, you can fix it by compiling without optimizations.

[Asman]

@Leffmann
Thanks a lot. I tried with -o0 and works fine and code looks

	subq.w	#8,a7
	movem.l	l22,-(a7)
l17
	move.l	_g_pWin(a4),a0
	move.l	(86,a0),a0
	move.l	_SysBase(a4),a6
	jsr	-372(a6)
	move.l	d0,(0+l24,a7)
	bne	l21
l20
	bra	l19

l21
	move.l	(0+l24,a7),a0
	move.w	(24,a0),(4+l24,a7)
	move.l	(0+l24,a7),a1
	move.l	_SysBase(a4),a6
	jsr	-378(a6)

        ;rest code

Of course now move.l dn,(an) set zero flag if dn is 0, so its ok.

Thank you.

[phx]

Could be a known bug. I think it was fixed some time ago. In the list for the next vbcc version I have this:

- m68k: Fixed asm-peephole optimization of move.l dx,ay; move.l ay,dx when
  cc of ay is used.

Please tell me about bugs (by email!). So they may be fixed and you will get an update immediately!

When you have a small source text for reproduction I could confirm that the bug is fixed. The source fragment above is not compilable.

[Leffmann]

It's probably the same bug. It may happen if you use anything other than -O0. Simple way to reproduce it:

void* x();

int func() {
  void* p = x();

  if (p)  return 1;
  else    return 2;
}

[phx]

Confirmed. Then it's fixed.

[Asman]

@phx
Great news. When should I expect a new version of vbcc ?

[phx]

Originally Posted by Asman

When should I expect a new version of vbcc ?

This might take a while. Don't know if we release a new version this year. Maybe I should make another patch for 0.9d, regarding the severity of that bug.

Otherwise just write me a mail and you can get the current development version.