core dumps in optimized case with gcc 3.0 and 3.0.1 (regression against 2.95)

[Nuesser, Wilhelm]

Hi,

when compiling our software ( binary size approx. 40 MB) with gcc 3.0 and
3.0.1 with -O2 on
	Linux ia32, 
	RedHat 6.2 and SuSE 7.0 (i.e. glibc 2.1.3),
	kernel 2.2.19 and 2.4.4
we find core dumps which do not appear when compiling with 2.95.[2,3] 
(with or without optimization) or with gcc 3* without optimizations.

We suspect gcc bugs, since our software run on many 32 and 64 bit platforms,
but - of course -
we are not sure. We have no idea what is actually going on so we can not
produce any testcase
at the moment.

We find the following behaviour:
when stepping through the optimized code - using -O2 -g  - and checking for
our own
variables/structs we can not find any strange or corrupted values but the
program
crashes in one memcpy inside an inlined function. 
Again: just before calling memcpy all pointers etc seem to be OK.  Also,
memcpy seems 
to get the correct parameters. This seems to rule out something like memory
corruption 
from inside our own program.... After the core the pointers are corrupted.

Can you give us any hints on how to analyse this further, e.g. compiler
switches for optimization etc.

Regards
Willi Nüßer
SAP LinuxLab

PS:
please cc me since I´m not on the list