Re: a.out binaries that are 66% faster than ELF

Paul Gortmaker (gpg109@rsphy1.anu.edu.au)
Wed, 26 Feb 1997 14:36:55 +1100 (EST)


Hugues Talbot wrote:

> Yes, a.out is known to be generally faster than ELF, because
> ELF requires the use of an extra register. x86 processors being
> register-starved, I'm not surprised that you get the sort of result

That is only for things that go to shared libraries, which are built
with "-fPIC" (i.e. libc.so.5 etc.). Then you lose the ebx register
to get position independent code. By linking it statically I took care
of that possibility. It is the same as the fact that there is no speed
difference between an a.out kernel and an ELF kernel. Besides, this
app doesn't do libc calls in the core calculation, so static vs
non-static shouldn't matter.

> You can usually get the speed back using the
>
> -fomit-frame-pointer

That was/is already on as default in the makefile for the rc5 client.
Thanks for the reply, but there must be something else funny going on.

Paul.