[Bug ipa/65028] [5 Regression] 450.soplex in SPEC CPU 2006 is miscompiled

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65028

Jan Hubicka <hubicka at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mjambor at suse dot cz

--- Comment #2 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
OK, the code dies realtively soon:

Breakpoint 1, main (argc=2, argv=0x7fffffffec28) at example.cc:116
116     {
(gdb) n
169        NameSet                rownames;
(gdb) 
170        NameSet                colnames;
(gdb) 
186        for(optidx = 1; optidx < argc; optidx++)
(gdb) 
188           if (*argv[optidx] != '-')
(gdb) 
254        if ((argc - optidx) < 1)
(gdb) 
267        MySoPlex work(type, representation);
(gdb) 
262        Param::setEpsilon(epsilon);
(gdb) 
263        Param::setVerbose(verbose);
(gdb) 
265        std::cout.setf(std::ios::scientific | std::ios::showpoint);
(gdb) 
267        MySoPlex work(type, representation);
(gdb) 
262        Param::setEpsilon(epsilon);
(gdb) 
267        MySoPlex work(type, representation);
(gdb) 
213              timelimit = atof(&argv[optidx][2]);

Clearly I do not pass any option, so there is no need for code getting there at
all.

-fno-ipa-cp fixes the bug, so I suspect it is a bug in ipa-cp WRT redirection
to local functions with clonnes...
Martin, perhaps you can spot it faster?

Honza