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: "Kaveh R. Ghazi" <ghazi at caip dot rutgers dot edu>
- Subject: Re: egcs-19980508 powerpc-ibm-aix4.1.4.0 results (failures with -p/-pg)
- From: David Edelsohn <dje at watson dot ibm dot com>
- Date: Sat, 16 Jan 1999 20:14:27 -0500
- Cc: egcs at cygnus dot com, law at cygnus dot com, meissner at cygnus dot com
>>>>> "Kaveh R Ghazi" writes:
Kaveh> That problem was failures in the testsuite when running -p/-pg on code
Kaveh> which calls nested functions.
Kaveh> Eg: execute/{931002-1.c,921017-1.c,920612-2.c,920501-7.c,920428-2.c}
Yes. Mike thought that the backchain might have been the
problem. When I last looked something about nested functions and
profiling was causing the programs to wander into oblivion during the
nested function.
I think the problem is a poor interaction between the way AIX
generates profiling statistics and GCC's implementation of nested
functions. Something is wrong with the GCC-generated calling convention
for nested functions or AIX is making some unsafe assuption about the
stack. The AIX profiling code goes poking around to find the function
address, updates a counter, and writes some information. In the case of
nested functions, something goes wrong. I have not had time to step
through this code enough to find out what really is going wrong.
David