On Wed, Jun 19, 2013 at 09:23:34PM -0500, Bruce Dubbs wrote: > I've noticed some things about binutils lately. I suppose it is just > getting bigger and more complex, but the SBU time for it on my system is > now 124 seconds with the build size 401 M. > > Looking back, I see: > > SVN-20130616 124 sec 401M binutils-2.23.2 gcc-4.8.1 > LFS-7.3 113 sec 412M binutils-2.23.1 gcc-4.7.2 > LFS-7.2 110 sec 390M binutils-2.22 gcc-4.7.1 > LFS-7.1 104 sec 373M binutils-2.22 gcc-4.6.2 > LFS-7.0 101 sec 351M binutils-2.21.1 gcc-4.6.1 > LFS-6.8 missing > LFS-6.7 91 sec 312M binutils-2.10.1 gcc-4.5.1 > > The changes could also be due to gcc. Has anyone else noticed these > increases? > Ever since I started. Well, ever since gcc-4 anyway, and probably also in the days of gcc-3. That's one of the reasons I've had to keep upgrading my hardware ;-)
Selected figures I still have for binutils pass 1, all x86_64, all probably passing CFLAGS=-O2 (so, overwriting the default -O2 -g here, which explains why mine are so much smaller) - I _did_ do a build without passing CFLAGS last year, honestly, but it must have been on the intel machine. I reworked my buildscripts around LFS-7.2, before that my timings were elapsed whole seconds and I won't swear that the space is always accurate. Also, the timings depend a little on what I was building _from_ : sometimes a very similar recent SVN, sometimes an older release. In recent builds which I'm keeping, I retime the SBU after the system has been booted so I can have a reliable SBU for BLFS edits : typically, the time for binutils pass 1 is about 10% slower than from system I used to build it - unless that system was very recent. old uniprocessor athlon64 "3200+", only 1GB DRAM LFS 6.6 binutils-2.20, secs=178, 199.8 MiB LFS-7.0 binutils-2.21.1a, secs=200, 217.9 MiB LFS-7.1 binutils-2.22, secs=211, 233.9 MiB another old uniprocessor athlon64, slightly faster, 2GB DRAM, with 10K rpm disk, now defunct LFS-svn-20100817 binutils-2.20.1, secs=168, 200.3 MiB LFS-svn-20101229 binutils-2.21, secs=146, 216.4 MiB - maybe something got faster that time ! LFS-6.8 binutils-2.21, secs=143, 216.4 MiB LFS-7.0 binutils-2.21.1a, secs=143, 217.1 MiB looks as if I built those two from the same host system! phenom, x86_64 (only the figures for -j 1) LFS-7.1 binutils-2.22, secs=189, 233.5 MiB LFS-7.3 binutils-2.23.1, secs=192.984, 245.1 MiB LFS-svn-20130424 binutils-2.23.2, secs=122.577, 245.9 MiB These are all with the ondemand cpufreq driver - for the Phenom I've now got that running significantly faster by tweaking one of the control knobs (the sampling_down_factor) and that shows in the last entry above. But I haven't got round to measuring the power consumption! I guess my main conclusion is that each minor release of gcc takes longer to compile simple C code. Also, any variation of less than 5% time is probably just noise - unless you run the realtime kernel. ĸen -- das eine Mal als Tragödie, das andere Mal als Farce -- http://linuxfromscratch.org/mailman/listinfo/blfs-dev FAQ: http://www.linuxfromscratch.org/blfs/faq.html Unsubscribe: See the above information page
