Clean way to compile with Machine-specific optimizations [Re: Machine-specific optimizations]
Cesar Eduardo Barros <cesarb@nitnet.com.br> writes:
> I know this theme has been repeated a lot here, but I still think that using
> machine-specific optimizations can make a difference. Specifically, there are a
> few packages (libgmp, gnupg, bzip2) where it could make a lot of difference.
> Some packages use every tiny bit of extra compiler optimization, while others
> have subarch-dependent assembly code which makes a lot of difference.
>
> So, is there any plan to use them (like recompiling the package on the user's
> machine)?
I'm worked a lot on my little private autobuilder system which aim is
to setup a local mirror from source on a need to have basis.
The simplest way is to just tell it to install a package and it will
do so by downloading all needed sources and compile them in the right
order (including build-depends). Anything present will of cause be
used and reused and removed if not needed anymore.
The high end setup would be to have it run as a full buildsystem
building according to your package list and priorities given in the
background and whenever you feel like it you use apt-get update from
the generated local packages.
I successfully build and installed some smaller packages and my
current big testcase is kde, which should rule out all obvious bugs
once it builds.
Thats pretty much the now state.
BUT how do I set local maschine specific optimisation in a clean way?
Is there anything that works on all packages except replacing the
gcc/g++ with a wraper? Even that won't work for gcc, since it uses its
internal xgcc intermidiat compiler.
Any better ideas?
May the Source be with you.
Goswin
Reply to: