[BusyBox] Using busybox as a space optimization benchmark for compilers

Aaron Lehmann aaronl at vitelus.com
Tue Aug 13 05:41:03 UTC 2002


I compiled busybox with several different compilers and the results
are somewhat encouraging:

2.7.2.3                                         350900
gcc 2.95.4                                      309544
gcc-3.2 (GCC) 3.2 20020809 (Debian prerelease)  305228
gcc (GCC) 3.3 20020809 (experimental)           296588
gcc (GCC) 3.3 20020809 (experimental) -fnew-ra  300684


Obviously, unsupported flags are not given to the older compilers.
Most significantly, gcc 2.7.2.3 was passed -O2 instead of -Os by the
makefiles.

I'm excited about these results, and they let me recommend the latest
GCC to anyone looking to build the smallest busybox possible.

The only thing that concerns me is that turning on the experimental
GCC's new register allocator caused an increase in binary size.
However, the new register allocator is a work in progress, and I will
check back on it and submit bug reports to the GCC developers if this
issue is not addressed by the time the code matures.

I'd also be interested how proprietary compilers like Intel's fare,
but considering the fact that such compilers are proprietary it's not
an experiment I'm going to perform any time soon.



More information about the busybox mailing list