Extension of the CSiBE benchmark

Joel Sherrill joel.sherrill@OARcorp.com
Wed Oct 29 16:15:00 GMT 2003


Arpad Beszedes wrote:

> Dear all,
>
> Inspired by a very positive feedback from the community, we would like 
> to make the CSiBE benchmark for code size 
> (http://sed.inf.u-szeged.hu/csibe/) even more useful by extending its 
> test bed. It is very important the test bed to be representative and 
> contains examples of various fields such as ingeger, floating point, 
> mobile, network, embedded applications and even system code such as 
> kernel modules.
>
> For this, we are seeking for your help: what kind of applications, 
> code snippets, modules, ... would you suggest? The test bed currently 
> includes some media applications, complex integer- and 
> control-compuation applications, text and data manipulation programs, 
> etc. Currently the total size is about 1.2M (binary).
>
> We are planning to add some network protocol implementations such as 
> tcp/ip and parts of the linux kernel. Do you have some ideas on 
> exactly which source code parts would be useful for this purpose? And 
> what kind of other example programs would you like to see?
> Please bear in mind that the example programs should compile for all 
> supported targets (arm, i386, m68k, mips, ppc).
>
You could use pieces of RTEMS or all of it.  The code executive portion 
is intended to be very lean
99% C which depends upon inline assembly to touch HW and inline 
functions instead of macros.  It
includes a port of the FreeBSD TCP/IP stack.

We also have some test programs which seem to map from a code size 
perspective onto what I
would think of as baselines for user applications.  For example, one 
just starts and sysem and
shuts down (no printf) while the next does a printf so you can see the 
impact of doing that on
code size.  Another set show various networking configurations -- basic, 
with httpd, with ftpd.

> Thank you for you help. All suggestions will be very much appreciated.
>
> Best regards,
>   Arpad Beszedes
>
>

--joel sherrill





More information about the Gcc mailing list