This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: egcs-19980508 powerpc-ibm-aix4.1.4.0 results (failures with -p/-pg)
- To: David Edelsohn <dje at watson dot ibm dot com>
- Subject: Re: egcs-19980508 powerpc-ibm-aix4.1.4.0 results (failures with -p/-pg)
- From: Jeffrey A Law <law at cygnus dot com>
- Date: Sun, 07 Jun 1998 00:04:47 -0600
- cc: "Kaveh R. Ghazi" <ghazi at caip dot rutgers dot edu>, meissner at cygnus dot com, egcs at cygnus dot com
- Reply-To: law at cygnus dot com
In message <9806062252.AA35940@rios1.watson.ibm.com>you write:
> Yes, this might affect combining profiling with those other
> languages. I stepped through one of the failures generating a SIGKILL and
> it occurred on the last iteration of the loop when it was storing an
> updated stack value. The memory address was valid (in fact the program
> read from the address a few instructions before).
>
> AIX profiling routines touch the stack. The *way* that GCC
> implements nested functions simply may not be compatible with AIX
> assumptions during profiling. I haven't had time to explore this further
> to understand exactly what is causing the store to blowup. Again, this is
> not a SEGV, the kernel is destroying the entire program context when this
> occurs, not just handing the program an unmaskable signal.
Meissner noted a while back that the profiling code clobbers r11,
which happens to be the static chain pointer. This may account
for the lossage when profiling in the presense of nested functions.
jeff