gcc -O2

I'm using

    # gcc -v
    Reading specs from /usr/lib/gcc-lib/ia64-cygnus-linux/2.9-ia64-000216-final/specs
    gcc version 2.9-ia64-000216-final

as bundled in Turbolinux 0823

   # uname -a
   Linux impromptu 2.4.0test7-000823-42 #1 Sat Aug 26 16:31:32 PDT 2000 ia64 unknown

I've diagnosed a seg fault that seems to stem from a bad optimization by gcc -O2. I can't find any FAQ that discusses known problems with the ia64 gcc. Is there one? Is -O2 considered generally usable? or is it well known that one ought to avoid it in this release? How 'bout vanilla -O? (my particular test case works OK with -O but it's hardly an exhaustive test...).

Where should I direct gcc bug reports -- on this list, or to the general gcc bug reporting mechanism (a query on GNATS on gcc.gnu.org shows no matches on the search string "ia64"... Mmmm.... :-)).

Ought I avoid gcc and skip directly to the SGI sgicc?

Steve

That compiler is too old. The current one is 0925 + patch.022.

H.J.

H . J . Lu wrote:

| On Mon, Oct 02, 2000 at 01:20:09PM -0400, Steve Tynor wrote:
| > I'm using 
| > 
| >     # gcc -v
| >     Reading specs from /usr/lib/gcc-lib/ia64-cygnus-linux/2.9-ia64-000216-final/specs
| >     gcc version 2.9-ia64-000216-final
| > 
| 
| That compiler is too old. The current one is 0925 + patch.022.

Aha! The cygnus toolchain link on the linuxia64.org web page just
points at a tarball dated something like 1 May 2000 -- where am I expected to find this 0925 version? Should I be cvs up'ing from cygnus? Are there rpm's available?

Thanks!
Steve

All optimizing compilers have bugs. You shouldn't panic when you find one. The -O2 option is definitely usable. Entire linux distributions have been built with it.

There are some known problems with the C++ support in the 000216 release. There was a lot of C++ front end development work happening at the time, and we ended up with an FSF gcc snapshot that has a few instabilities in the C++ front end. Otherwise, the release has held up very well. There haven't been very many problems with it.

Bug reports sent here are OK. Bug reports sent to the usual FSF addresses are OK, if you make it clear that you are using one of the unofficial releases that the ia64-linux project is using.

There is a new toolchain release that you can try if you are interested. I expect that the next series of linux distributions will use this new toolchain. It is located on ftp.cygnus.com in pub/ia64-linux.

I am not too interested in bug reports against the old toolchain at this point, but I will help if I can.

Jim

Jim Wilson wrote:

| There is a new toolchain release that you can try if you are interested.
| I expect that the next series of linux distributions will use this new
| toolchain.  It is located on ftp.cygnus.com in pub/ia64-linux.

Excellent -- I'll grab that. If the bug still remains, I'll submit a bug report.

Perhaps someone can fix the link in the www.linuxia64.org homepage to point to a this? The current link points to what appears to be a 010500 snapshot (http://www.cygnus.com/ia64 which resovles to ftp://ftp.cygnus.com/private/intel/).

Thanks for the help,
Steve

It isn't clear to me what to do about this. The www.linuxia64.org page points to the current release that is in use by all existing linux distributions, and is known to be OK.

The new release is not ABI compatible with the old release, and there are no linux distributions yet that use the new release. Also, the new release is still getting the bugs worked out, and may not be as reliable as the old release yet. Certainly, I'll want the link changes when this becomes the recommended compiler, but we aren't quite there yet.

Jim