GCC quality (was Re: AMD-64 optimizations, some (new) code)

Torbjorn Granlund tege at swox.com
Tue Sep 27 20:04:59 CEST 2005

Ashod Nakashian <saghmos at xter.net> writes:

  I'm quite surprised. 4.0 was supposedly well tested and most of the 
  sources of old bugs were replaced with fresh code, also well designed 
  and tested. I expected the 4th version to be much more stable than the 
  now-old 3 version. I'm sure you know what you are talking about though, 
  so I'll take your word.
When old code is replaced by new code, many bugs are typically

I think the entire software community has a very serious attitude
problem leading to quality problems with almost all software.  GCC,
the Linux kernel, Windows, etc, are examples.

  > I have had more problems with 3.4.x than with 3.3.x.  But with
  > the track record of gcc, it is not safe to use any gcc release.
  So 3.3.x is the most stable one?
I wouldn't outright nominate a certian version as the most stable.
I have found 3.3.x reasonably stable and giving good GMP code.  But
  By the way, the same problems you speak of are very serious issues in 
  Intel's C++ compilers as well. In particular, I've written MP math 
  function and when optimizations were enabled, Intel's generated binaries 
  would simply either hang (at initialization, before calling main) or 
  crash instanty. In some cases it would generate wrong code, by that was 
  more rare than hangs and crashes. I've found these problems in at least 
  7 and 8. In this respect version 6 is more stable than version 7 than 
  version 8. I have version 9, but its requirements to install it are too 
  much to pollute my workstation.

I believe Intel's compiler is written with SPEC in mind; it might
be unsuitable for any other programs.  :-)

  I thought it would be interesting to put a $400+ professional and 
  commercial compiler in prespective with an open-source, FREE compiler.

Yes, and HP's compilers (targeting HPPA and IA64) are also very
buggy.  But let's not argue that a certian piece of Free Software
is good enough, just because there are commercial horror examples!
Is GNU/Linux's buggyness OK just because Windows is worse?  (And is
it really still worse?)


More information about the gmp-discuss mailing list