optimization/1603: Optimization of any kind produces bad code.

[jcmcknny at uiuc dot edu]

>Number:         1603
>Category:       optimization
>Synopsis:       Optimization of any kind produces bad code.
>Confidential:   no
>Severity:       serious
>Priority:       high
>Responsible:    unassigned
>State:          open
>Class:          wrong-code
>Submitter-Id:   net
>Arrival-Date:   Wed Jan 10 00:56:00 PST 2001
>Closed-Date:
>Last-Modified:
>Originator:     jcmcknny@uiuc.edu
>Release:        2.96
>Organization:
>Environment:
Redhat7.0 with latest updates or RedHat6.0 as compiled by RH7(with updates)
>Description:
Completely wrong code being made with any optimizations as compiled by 2.96 with my code.  I get a seg fault 11 on any intel machine when compiling my code with 2.96 when -O1 -O2 or -O3 is used.  Each opt has a different problem and sometimes the compiler doens't even finish and gets an error.

I'll make note of this error the next time it occurs.

With -O2 or -O1 sometimes it gets to my print of an program error and gives a statement(values of certain variables) and clearly if that was the right value it never should have gotten to that program error printout.

With -O3 I get a seg fault usually at some point(init.c:3400 for future reference) and it shouldn't have failed.  gdb shows nothing wrong with the addresses.
>How-To-Repeat:
Run 2.96gcc compiled program(my program) on any intel machine with opts -O1, -O2, or -O3.  -O0 works fine.

I'll be happy to send my program, but it's quite large.  If you request it, I ask it be taken as confidential.

I plan on just reverting to a working compiler since I can compile using egcs-2.91.66 and that works fine on any of my intels.
>Fix:

>Release-Note:
>Audit-Trail:
>Unformatted: