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

Brad Roberts (
Wed, 26 Feb 1997 16:48:11 -0800 (PST)

I don't have the original message with the 50% or 66% numbers in them, but
a little caution needs to be taken here. Don't assume that the binaries
that are on the ftp site are necessairily the same as you're creating with
gcc, or with the same version of gcc. I sent a letter to
and got back a reply that basically said they were using customized and
beta compilers for various platforms. I didn't push or inquire further as
to what, because gcc is enough for me. However, I dropped from 150k
keys/second an a low used p166 with their client to about 140k keys/second
with a slightly modified (shouldn't affect the timings, just logs the date
and time with each line of output) copy of the client.

On Thu, 27 Feb 1997, Marc Lehmann wrote:

> This is long known.. I thought this was because
> elf-shared libraries were somewhat slower, but that cannot account
> for 50%...
> The mapping for elf is more complicated (isn't it?),
> but this, too, should not account for 50%...
> I doubt that gcc -S will help... gcc's output
> is not significant different when using a.out (unless
> you use -fpic also).
> Any ideas? I'd like to have an answer to this question..
> back in the i486 days, the difference was about 10%,
> and nobody cared... but 50% (66% in that case)
> is way to much!
> -----==-
> ----==-- _
> ---==---(_)__ __ ____ __ Marc Lehmann
> --==---/ / _ \/ // /\ \/ /
> -=====/_/_//_/\_,_/ /_/\_\
> The choice of a GNU generation