I'm working with an all-RAM system (where a host processor pokes
the linux kernel and initial ramdisk into RAM, then releases the
reset line). The kernel is 2.4.17 or so.
I have only 64 MB (or 32 or 16, depending), so every megabyte is precious.
The C++ app I'm running on this system is 5MB, kinda fat.
I'd like to save a couple megabytes somehow, either by compressing
the executable, or by using execute-in-place.
I made a cramfs image of the app; that shrunk it from 5MB down to 2MB.
You can't mount cramfs images from tmpfs, so I made a little ramfs,
and loopback mounted it from there, then measured ram usage on a 64MB system
with the app both running and not running.
/bin/free reported (kb):
in tmpfs, not running: 47888
in tmpfs, running: 47244
in cramfs, not running:46972
in cramfs, running: 46336
Looks like I lose! Using cramfs to compress the app appears to
*waste* a megabyte over just using the noncompressed app.
>From what I can tell, there's no support for execute-in-place
except on MIPS with a hacked gcc.
OK, I've exhausted my stock of clever ideas; does anyone have
a suggestion about how to save RAM in my situation?
(gzexe won't do it, nor will upx, as they all create a temporary
uncompressed file on execution...)
Thanks,
Dan