On Wed, Apr 25, 2012 at 10:01:59AM +0200, Pierre Labastie wrote: > Several comments on this thread : > > - I am building with jhalfs. With my configuration, it passes -03 -pipe > -march=native to all the builds (either 32 or 64 bits) > except gcc, gmp (configure does a better job in finding the right > optimizations), grub > and binutils where I found that I had to remove the -march flag (do not > remember why now). > Never seen any failure. It always build to completion. I have tested > several times on a 32 bit virtual machine, also on 32 bit bare metal, > but maybe not the recent versions on that last one. > > - Almost everybody (I mean three distros: fedora, debian and arch, not > checked the others) builds glibc with -O3 now. The documentations you > are citing have not been modified since the early ages, and might not be > up to date. > FWIW, seems to be a hardware problem : I've just built chapter 6 glibc of current svn using -O3 -pipe and make -j 2. This is a new mobo/cpu/RAM/disk/DVD in an old case and with the old PSU. I suspect the power supply isn't quite powerful enough.
As I said earlier, repeating make -jN *after* a failure succeeded. On the finished system from 20040419, but still in chroot (I was building in /scratch/lfs to debug my initial desktop packages with gcc-4.7 / current glib) I had ICEs at various places in glibc using make -j3 (and one unrepeatable ICE in pcre using -j4). To be fare, the PSU was originally spec'd for an athlon 64 single processor at 2.6 GHz and only 4 GB or RAM - now I've got 8GB and 4 real cores. The total power draw is still mostly < 150 Watts, but maybe the supply to the memory is marginal. Note that memtest86 was fine, but I don't think that stresses the cores, only the memory - compiling glibc appears to stree both at hte same time ;-) ĸen -- das eine Mal als Tragödie, das andere Mal als Farce -- http://linuxfromscratch.org/mailman/listinfo/lfs-dev FAQ: http://www.linuxfromscratch.org/faq/ Unsubscribe: See the above information page
