This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Large resources required to build profiled gcc
- From: Daniel Berlin <dan at dberlin dot org>
- To: lucier at math dot purdue dot edu
- Cc: gcc at gcc dot gnu dot org
- Date: Fri, 22 Mar 2002 16:42:14 -0500 (EST)
- Subject: Re: Large resources required to build profiled gcc
On Fri, 22 Mar 2002 lucier@math.purdue.edu wrote:
> I'm trying to build a profiled gcc on sparcv9-sun-solaris2.8
> with BOOT_CFLAGS=' -O2 -pg -g -m64 -mcmodel=medlow' and
> BOOT_LDFLAGS=' -O2 -pg -g -m64 -mcmodel=medlow'.
>
> A parallel build failed with the remark:
>
> ./gencheck > tmp-check.h
> ./genpreds > tmp-preds.h
> No space for profiling buffer(s)
> /bin/sh ../../gcc/move-if-change tmp-mlib.h multilib.h
> multilib.h is unchanged
> echo timestamp > s-mlib
> ./gengenrtl -h > tmp-genrtl.h
> No space for profiling buffer(s)
> Segmentation Fault
> Segmentation Fault
>
> So I reran it serially, and got this information from top:
>
> 23017 lucier 1 20 0 3075M 1680M cpu/1 0:46 20.77% gengenrtl
>
> It's fairly remarkable that gengenrtl takes 3 GB of swap space,
> no wonder it ran out of virtual memory.
>
> Is there some way to reduce these resource requirements? If not, only
> people with fairly large machines will be able to build/test/run these
> profiled programs.
>
This smells like a bug in compiling gengenrtl more than being for real.
I've screwed up things before that cause gengenrtl to end up running
forever, rather than segfault.
Whenever it did this, it would just keep eating up memory.
> Brad
>