This is the mail archive of the
gcc-prs@gcc.gnu.org
mailing list for the GCC project.
target/8816: bad assembly generated
- From: tege at bob dot swox dot se
- To: gcc-gnats at gcc dot gnu dot org
- Date: Thu, 5 Dec 2002 03:42:45 +0100 (MET)
- Subject: target/8816: bad assembly generated
>Number: 8816
>Category: target
>Synopsis: bad assembly generated
>Confidential: no
>Severity: critical
>Priority: medium
>Responsible: unassigned
>State: open
>Class: wrong-code
>Submitter-Id: net
>Arrival-Date: Wed Dec 04 18:46:02 PST 2002
>Closed-Date:
>Last-Modified:
>Originator: Torbjorn Granlund
>Release: 3.2.1
>Organization:
Swox AB
>Environment:
System: SunOS bob.swox.se 5.7 Generic sun4u sparc SUNW,Ultra-60
Architecture: sun4
host: sparc-sun-solaris2.7
build: sparc-sun-solaris2.7
target: sparc-sun-solaris2.7
configured with: /u/gcc/gcc-3.2.1/configure --enable-languages=c,c++
>Description:
Incorrect assembly code is generated for symbol+constant expressions
that are generated as a result of constant propagation.
Assembly sample:
lduh [%o0+%lo(tab)+2], %o0
The last instruction is really dubious, adding a cnstant outside the
lo() expression. The system assembler rightly complains, but the GNU
assembler (version 2.13.1) silently ignores the +2. (I report that
bug separately.)
What makes this bug really serious is that the GNU assembler misses
the incorrect assembly syntax.
>How-To-Repeat:
Compilation command: gcc -O -m64 ~/tco.c
This is tco.c:
unsigned short tab[] = { 0xbad, 0xcafe };
int
foo ()
{
int n = 1;
return tab[n];
}
int
main ()
{
if (foo () != 0xcafe)
abort ();
return 0;
}
>Fix:
>Release-Note:
>Audit-Trail:
>Unformatted: