GCC 2.9.5, Solaris 2.8, SPARC - alignment issue - PLEASE HELP!

GCC 2.9.5, Solaris 2.8, SPARC - alignment issue - PLEASE HELP!

[Rob Newberry]

Hi,

I have some legacy code, and I really, really, really need to not modify
it.  This code has alignment issues, but I would expect the compiler (or
the OS maybe) to handle this for me.

Here's some example of the code that causes problems:

        char    key[256];
        long    value = 1;

	// this causes a SIGSEGV on SPARC
        *((long*)&key[2]) = value;

I know this is cruddy, but I would _EXPECT_ the compiler to essentially
catch this and (more-or-less) rewrite it as something similar to:

	memcpy( &key[2], &value, sizeof(long));

I have tried compiling with STRICT_ALIGNMENT=1 and
SLOW_UNALIGNED_ACCESS=1, but neither is helping.  It always seg-faults.

Is there a solution to my problem on Sun OS 5.8 with GCC?

I am amenable to mucking with the Makefile or changing some kind of OS
configuration.  The thing that I can't do is modify the code.

(Actually, I can modify the code, but the problem is that I'm not sure I
can find all the places this could happen in the code, so I'd like to
simply let the compiler do it.  If there is a way to have the compiler
generate an error every time it sees misalignment, that would also be OK,
but absolutely less optimal.)

PLEASE HELP!

Rob

---------------------------------------------------------------------
Rob Newberry
Director of Fajita Technology
Group Logic, Inc.

Re: GCC 2.9.5, Solaris 2.8, SPARC - alignment issue - PLEASE HELP!

[Alexandre Oliva]

On Jul 24, 2001, Rob Newberry <robnewberry@grouplogic.com> wrote:

>         char    key[256];

> 	// this causes a SIGSEGV on SPARC
>         *((long*)&key[2]) = value;

> I know this is cruddy, but I would _EXPECT_ the compiler to essentially
> catch this and (more-or-less) rewrite it as something similar to:

> 	memcpy( &key[2], &value, sizeof(long));

Except that, when you tell it it's accessing a long, it's free to
assume it can do it efficiently, without using the byte access
instructions.  If you mean to do byte-per-byte access, use memcpy
explicitly.

> Is there a solution to my problem on Sun OS 5.8 with GCC?

It might be possible to use signal handlers to trap bus errors and
adjust unaligned accesses.

-- 
Alexandre Oliva   Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
Red Hat GCC Developer                  aoliva@{cygnus.com, redhat.com}
CS PhD student at IC-Unicamp        oliva@{lsd.ic.unicamp.br, gnu.org}
Free Software Evangelist    *Please* write to mailing lists, not to me