bug-binutils
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Bug gold/13245] PREVAILING_DEF reported too often.


From: hubicka at ucw dot cz
Subject: [Bug gold/13245] PREVAILING_DEF reported too often.
Date: Tue, 04 Oct 2011 09:41:40 +0000

http://sourceware.org/bugzilla/show_bug.cgi?id=13245

--- Comment #12 from hubicka at ucw dot cz 2011-10-04 09:41:40 UTC ---
> http://sourceware.org/bugzilla/show_bug.cgi?id=13245
> 
> --- Comment #11 from rguenther at suse dot de 2011-10-04 09:24:22 UTC ---
> On Sun, 2 Oct 2011, Jan Hubicka wrote:
> 
> > > 
> > > Yes, it does solve the problem here (I have 8GB of RAM and used a
> > > large swapfile on my SSD. It took ~10 minutes to output all ltrans.o 
> > > files).
> > 
> > I am surprised it converge with 8GB of RAM at all...
> > 
> > > 
> > > -fprofile-generate should set -fno-lto automatically IMHO.
> > 
> > I suggested this at one point and richi didn't like it, forgot why.
> > 
> > Thanks for looking into this! It is good to know this is "the last bug" WRT 
> > LTO linking of Mozilla ;)
> 
> Well, it sounded like a hack.  I'd rather have a way to force LTO
> profiling at least.

Instrumentation at linktime is in my longer term TODO, but it will be fun - we
will need to pickle
CFG at compile time and with -fprofile-generate decide on placement of
instrumentation at WPA time
making the actual modifications at ltrans time.
It is not that hard to do, but needs some work - i.e. cleanup what gcov has
already, because gcov
has pretty much all needed, but not readily useable within compiler and not
with the fancier features
like automatic correction.

Also if we get variant of LIPO into mainline, we will be stuck with the
instrumentation at compile time
too.

Honza

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.



reply via email to

[Prev in Thread] Current Thread [Next in Thread]