On Mon, Jul 7, 2014 at 4:09 PM, Jan Hubicka <hubi...@ucw.cz> wrote: > Hi, > AIX as' .set pseudoop has somewhat unexpected behaviour. It seems to be > implemented by syntactically replacin all appereances of the alias by > its target (that can be an expressoin) but it is still possible to globalize > the alias by .globl operation. > > This breaks the assumption we make about aliases; for example when alias is > declared static but it is alias of global/weak symbol, we assume it is static > symbol, but in AIX implementation it is handled as global. This breaks > horrible way for local aliases. > > Another unexpected behaviour is that .set output before definition of alias > target is silently compiled into something else (NULL?). We are not fully > consistent about outputing aliases after targets. > > This patch avoids use of .set for aliases. Instead > rs6000_xcoff_declare_object_name and rs6000_xcoff_declare_function_name simply > walk and outputs all associated aliases at the time they output function. > I.e. > > var: > <var data> > .set var, aliasvar > > Is now done as: > var: > aliasvar: > <var data> > > This avoids the surprises we was hitting on AIX and makes aliases to work > consistently. The way it is hooked into output machinery is not the most > beautiful, but I do not see better way except for redesigning the output > interface (adding new macro for this type of aliases) that seems to be > unnecesay > unless we find more targets that would benefit from aliases output this way. > > The testcases localalias/globalalias was added to check sanity of alias > implementation and this patch makes them to work. > > Bootstrapped/regtested rs6000-aix OK?
Hi, Honza This looks really good and mostly is working. With the patch, GCC on AIX now responds that "ifunc" is supported. The C and C++ attr-ifunc testcases now run and fail. g++.dg/ipa/devirt-10 and devirt-15 now fail. Otherwise, the results look very good. Thanks, David
