This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [patch] Predictive commoning


> Hello,
> 
> > > 2 - This information would be lost if we ever decide to "recalculate" or
> > > reset it after some CFG manipulations or code removal.  Recreated info
> > > would not include any temps you created here. Up until now, its easy to
> > > recreate accurately if we ever needed to.
> > 
> > For recomputing we can always have two flags - ie
> > SSA_NAME_NO_OVERLAP_LIVERANGES and SSA_NAME_OCCURS_IN_ABNORMAL_PHI.
> > This would also make it more obvious what is going on.
> > 
> > We've discussed this some time ago with Zdenek (I also didn't like the
> > change) and there was good testcase sort of justifying this. Zdenek,
> > could you remind me, please?
> 
> I do not remember such a testcase :-(.  I also do not like the change
> too much, so I am open to proposals (something along the lines of
> SSA_NAME_NO_OVERLAP_LIVERANGES seems like the best choice to me at the
> moment, but if we ever get rid of SSA_NAME_OCCURS_IN_ABNORMAL_PHI
> completely, introducing this flag would mean that we would still need
> to preserve the code to enforce it).

Well, without the testcase we might pretty much just skip this part from
patch and wait for it to appear and if it does adding the neccesary
bits.

Honza


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]