This is the mail archive of the gcc-bugs@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]

[Bug tree-optimization/80155] [7/8/9 regression] Performance regression with code hoisting enabled


https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80155

--- Comment #41 from rguenther at suse dot de <rguenther at suse dot de> ---
On Wed, 1 Aug 2018, prathamesh3492 at gcc dot gnu.org wrote:

> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80155
> 
> --- Comment #40 from prathamesh3492 at gcc dot gnu.org ---
> ping https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80155#c38
> 
> Thanks,
> Prathamesh

Well, all heuristics will have up and downsides.  In principle
re-ordering sinking and PRE makes sense but given it isn't
only "sinking" side-effects of this need to be watched.  I suppose
similar as to how PRE hoists code it should also sink, removing
the need of a separate (ad-hoc!) sinking pass.  Note that both
passes are confused by dead code.

But yes, in general a live-range shrinking pass is what would
improve the situation in the best possible way given the
exact situation created by PRE & friends can be created by
adjusting the testcase in source.

One complication with the idea of a live-shrinking pass is
that we have TER.  Bernd has done some kind of live-shrinking
pass before that wasn't merged.  I think I pointed you to it
at some point.

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