Compiler flags and memory usage

John Reiser jreiser at BitWagon.com
Sun Jun 15 18:43:30 UTC 2014


> These compilations have been particularly annoying because it was
> nearly impossible to compile on my PC without blocks dependent by
> *high* memory usage.

How much RAM and swap and free space does the build machine have?
("sed 5q  < /proc/meminfo;  cat /proc/swaps;  df -h")

The BuildRequires of coin-or-lemon-devel seems to be not yet available in f20,
so "rpmbuild -bc seqan.spec" fails.  Using the seqan-src-1.4.1.tar.gz
that is named in https://bugzilla.redhat.com/show_bug.cgi?id=1105730
and invoking by hand the 'cmake' command from seqan.spec, then a
plain "make -k" using -O2 [but no "-j" parallelization], I compiled all-but-one
of the source files.  On razers3/razers.cpp, the complaint "OpenMP not found"
caused failure.  Perhaps OpenMP should be a requirement?

On my 3.4GHz box with 8GB RAM, compiling finished in less than 30 minutes.
While compiling, I watched the display from /usr/bin/top.  On pair_align.cpp
the VIRT for c++ reached just under 3000000 (3M * 1KB ==> 3GB).
Memory usage for other files seemed to be 500MB (VIRT 500000) or less.
So you need 3GB of RAM for g++ on pair_align.cpp, which means
4GB RAM on the box.  If you don't have 4GB of RAM then you must have
swap space (paging space) to make up the difference; and compiling
that file will take much longer (estimate: 10 to 20 times as long)
because of the paging.



More information about the devel mailing list