>From gcc report, while builing decoder.cc in m5.opt, the following phases consumes a lot of memory:
parser : 4.59 ( 2%) usr 0.64 ( 3%) sys 4.91 ( 1%) wall 343691 kB (12%) ggc integration : 4.40 ( 2%) usr 0.62 ( 3%) sys 6.75 ( 2%) wall 422895 kB (15%) ggc tree operand scan : 1.37 ( 1%) usr 1.04 ( 6%) sys 2.22 ( 1%) wall 144875 kB ( 5%) ggc expand : 20.22 ( 9%) usr 0.38 ( 2%) sys 22.51 ( 5%) wall 249566 kB ( 9%) ggc symout : 13.25 ( 6%) usr 0.30 ( 2%) sys 24.96 ( 6%) wall 265239 kB ( 9%) ggc var-tracking emit : 16.57 ( 7%) usr 0.14 ( 1%) sys 16.90 ( 4%) wall 122478 kB ( 4%) ggc On 2/25/12, Hammond, Simon David (-EXP) <[email protected]> wrote: > Hi guys, > > We have been hitting this problem repeatedly on RedHat-based machines. > Debian and Ubuntu fine, RedHat and Fedora are not. At times we need over > 25GB in the machine to get a compile to run properly. > > We have tried changing compiler versions on the RedHat boxes but this > doesn't seem to help. > > If you find you get really stuck it might be worth trying a VM of another > operating system? > > > > ________________________________ > From: [email protected] [[email protected]] on behalf of > Steve Reinhardt [[email protected]] > Sent: Saturday, February 25, 2012 11:23 AM > To: gem5 users mailing list > Subject: [EXTERNAL] Re: [gem5-users] Error in compiling X86 decoder.cc > > Interesting... I also have an 8GB machine and hadn't noticed any trouble in > compiling (though of course it does take a long time). > > AFAIK, no one has done anything on this front since the email thread you > cite, though we did get a pretty concrete proposal for a workable solution > that "merely" needs to be implemented: > > http://www.mail-archive.com/[email protected]/msg04744.html > > A short-term fix would be to get rid of the "--pipe" option to gcc so that > the compiler and assembler don't get run concurrently. > > Steve > > > On Fri, Feb 24, 2012 at 2:24 PM, Max Kelner > <[email protected]<mailto:[email protected]>> wrote: > It seems it has been proposed in: > > http://www.mail-archive.com/[email protected]/msg04607.html > > but I don't know if there's been any more recent effort. > > > On Fri, Feb 24, 2012 at 5:17 PM, Paul Rosenfeld > <[email protected]<mailto:[email protected]>> wrote: > To be fair x86 is no ordinary architecture -- a mere ordinary machine can't > compile it. > > I mean to read 150k lines of source code or whatever and then try to > optimize them is no easy task. I wonder if anyone has proposed trying to > generate several files that make up the decoder instead of a single > monolithic .cc file. I don't know if that is feasible or if that would help > memory usage. > > On Fri, Feb 24, 2012 at 5:10 PM, Max Kelner > <[email protected]<mailto:[email protected]>> wrote: > > To the earlier email, I was able to compile gem5.fast, so it is likely the > compiler optimizations that are consuming all of the extra memory. I > monitored the build with a top window, and I had two active processes: as, > which used 2.1 GB, and cc1plus, which used 5.7 GB (out of 8 GB total). > > Has no one been able to compile gem5.opt for x86 on an ordinary machine? > > On Fri, Feb 24, 2012 at 4:46 PM, Paul Rosenfeld > <[email protected]<mailto:[email protected]>> wrote: > Yeah I've been able to compile x86 on our big servers (48gb memory), but > have had trouble on ordinary machines. I'd definitely keep a top window open > and monitor the memory usage. That decoder file is certainly a lot for a > compiler to chew on .... > > > > On Fri, Feb 24, 2012 at 4:35 PM, Nilay Vaish > <[email protected]<mailto:[email protected]>> wrote: > On Fri, 24 Feb 2012, Max Kelner wrote: > > I've been trying to compile gem5.opt for x86, using both the unmodified > gem5-stable repo (with 'scons build/X86_SE/gem5.opt') and the unmodified > gem5 repo (with 'scons build/X86/gem5.opt'). I've tried on gcc 4.3.2, > 4.5.2, and 4.6.1, and in all cases the build fails when it gets to > decoder.cc after a very long time. The error is always something like: > > [ CXX] X86/arch/x86/decoder.cc -> .o > build/X86/arch/x86/decoder.cc:120069:3: fatal error: error writing to -: > Broken pipe > compilation terminated. > g++: Internal error: Killed (program as) > Please submit a full bug report. > See <http://gcc.gnu.org/bugs.html> for instructions. > scons: *** [build/X86/arch/x86/decoder.o] Error 1 > scons: building terminated because of errors. > > I've seen several other related errors on both the users and dev mailing > list, but I haven't seen any solutions. On the dev list, I saw a discussion > about splitting decoder.cc into several files, so that gcc could compile > it, but I haven't come across any followup. > > Does anyone have an techniques for getting past this point in the compile? > > Thanks, > Max > > Other system details: > The system has 8GB memory > Red Hat (Enterprise 4) > Python 2.6.7 > > > Are you able to compile gem5.fast or gem5.debug? It definitely seems like > gcc runs out of memory while compiling. Try compiling on a machine with more > memory / swap space if you can. > > -- > Nilay > _______________________________________________ > gem5-users mailing list > [email protected]<mailto:[email protected]> > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users > > > _______________________________________________ > gem5-users mailing list > [email protected]<mailto:[email protected]> > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users > > > > -- > Max Kelner > College of Engineering, > Cornell University > [email protected]<mailto:[email protected]> > (419) 512-1098<tel:%28419%29%20512-1098> > > > _______________________________________________ > gem5-users mailing list > [email protected]<mailto:[email protected]> > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users > > > _______________________________________________ > gem5-users mailing list > [email protected]<mailto:[email protected]> > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users > > > > -- > Max Kelner > College of Engineering, > Cornell University > [email protected]<mailto:[email protected]> > (419) 512-1098<tel:%28419%29%20512-1098> > > > _______________________________________________ > gem5-users mailing list > [email protected]<mailto:[email protected]> > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users > > -- -- // Naderan *Mahmood; _______________________________________________ gem5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
