Mail Archives: djgpp/1999/08/19/13:19:43
On 19 Aug 99, at 10:12, Anton Helm wrote:
> At 03:39 PM 8/18/99 +0200, you wrote:
> >Hi,
>
> >Could someone tell me where I can find the user-visible changes
> >(changes in usage) when I move from 2.81 to 2.95 ?
>
> I would be interrested in finding out what's the difference between
> v2.95 and v2.95.1 which is already on our local GNU mirror
> (source codes, not DJGPP binaries!!!).
see
http://gcc.gnu.org
> Yesterday I had some discussions with a colleague about v2.95 and
> he was very disappointed (although he used a different wording
> which I'm not going to repeat here in the NG).
>
> He is working on a C++ program that computes a sequence of
> binary numbers e.g., 100111011... (don't ask me for details,
> I don't know). The program used to compile and run on a number
> of unix systems both with native and GCC (2.8.1) and also DJGPP
> and had equal results.
> Now with GCC v2.95 he is getting different sequences of 1 and 0
> depending on the level of Optimization. I don't remember exactly
> but I think he said the one matching the previous output is not
> from the -O2 compiled binary. We have reproduced this effect on
> DJGPP, Linux and Solaris running GCC v2.95.
>
It's wild guess but You may have run into pointer aliasing problems.
You may check it by compiling sources using option (gcc-2.95.X only)
-fno-strict-aliasing
Search gcc mailing list for these problems (see: http://gcc.gnu.org)
- Raw text -