From:	Eric Backus <ericb AT lsid DOT hp DOT com>
Subject:	Re: -O and -O2
To:	jon AT halsp DOT hitachi DOT com
Date:	Tue, 8 Dec 92 14:26:52 PST
Cc:	ESCHN705 AT RZ DOT Braunschweig DOT PTB DOT DBP DOT de, djgpp AT sun DOT soe DOT clarkson DOT edu,
	bug-gcc AT prep DOT ai DOT mit DOT edu
Mailer:	Elm [revision: 66.25]

> The following was posted to the djgpp mailing list by 
> 	ESCHN705 AT RZ DOT Braunschweig DOT PTB DOT DBP DOT de
> I tried it using gcc 2.2.2, running under Iteractive Unix version 2.2.
> It produces errors of essentially the same kind as ESCHN705 reported.
> 
> When compiled with option -O it produces (the program is named fail.c):
> $ fail
>  0 Bus error
> 
> When compiled with option -O2 it produces:
> $ fail
>  0  1 Bus error
> 
> This looks like a bug in gcc; but I don't have time to do more than
> report it.

I don't think so.  As I pointed out to the original poster, this
program sometimes uses a negative index into the array, thus probably
overwriting some other local variable.

>    for (lact=0; lact <= maxdim; ++lact)
>      { printf("%2d ",lact); fflush(stdout);
>        for (lm=0; lm <= 2*lact; ++lm)
>          for (ln=0; ln <= 2*lact; ++ln) morg[lm][ln] = 0;
> 
>        for (lm=0; lm <= lact; ++lm)
>          { morg[lm][lm] =
>            morg[2*lact-lm][2*lact-lm] =
>              -0.5*((2*lm+1)*lact-lm*lm)*(alpha1+alpha2) -
>              (lact-lm)*(lact-lm)*alpha3;
>            morg[lm][2+lm] = morg[2+lm][lm] =
>            morg[2*lact-lm][2*(lact-1)-lm] = morg[2*(lact-1)-lm][2*lact-lm] =

The problem is here ---------^   and here ---------^

>              -0.25*sqrt((lm+1.0)*(lm+2.0)*(2.0*lact-lm)*(2.0*lact-1.0-lm));
>          }
>      }

When lact = 0 and lm = 0, the index is -2.  When lact = 1 and lm = 1,
the index is -1.  In both cases, something gets overwritten.
--
				Eric Backus
				ericb AT lsid DOT hp DOT com
				(206) 335-2495

- Raw text -