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] Avoid ssa update in cfg cleanup


On Sat, 2006-02-04 at 22:28 +0100, Zdenek Dvorak wrote:
> Hello,
> 
> > > The key observation is that no new vops should ever appear in this way,
> > > only the old ones may disappear.  If new vops appear, it means that we
> > > either had incorrect alias information before (very bad), or that by
> > > making the copy propagation, we are losing information (not good,
> > > either).  
> > 
> > While not good, it happens right now.  DOM, for example, will propagate
> > things that will lose alias information in some cases.
> > 
> > Sadly, i believe doing this in your patch will have some negative
> > generated code performance impact, but maybe i am wrong.
> 
> as the testing shows, none of the DOM (or other) testcases we have in
> testsuite seems to be affected by the change.  I will run SPEC's
> overnight to get some idea whether there are performance loses or not.

Thanks.

I can construct testcases for you where DOM will do propagate things
that will cause us to lose NMT information, if you want them, but i'm
only concerned whether it actually matters in practice.



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