Re: linux 2.1.117 built with egcs-1.1 19980823

[Andi Kleen <ak at muc dot de>]

On Tue, Aug 25, 1998 at 02:56:02PM +0200, David S. Miller wrote:
>    Date: Tue, 25 Aug 1998 14:50:25 +0200
>    From: Andi Kleen <ak@muc.de>
> 
>    > (Andi, as a fun experiment, rebuild the kernel with
>    >  "-fno-schedule-insns -fschedule-insns2", this turns off
>    >  the first scheduling pass, yet keeps the second pass enabled which
>    >  runs after register allocation and therefore can't increase
>    >  register pressure and create more spills)
> 
>    I tried this, it seems to cut off another 4K:
> 
> Amazing, especially since Jeff mentioned that on Intel the first
> scheduling pass is not even supposed to run :-)

Hmm... 


> 
>    I'll look at some more function diffs later today to see if I can
>    spot something obvious.
> 
> Another recent addition to egcs which can generate thicker code in
> some cases is GCSE.  So if you have more cycles to burn, -fno-gcse
> might be interesting too. :-)

I tried -fno-gcse earlier on Jeff's suggestion and it gave a few K difference
too. Currently it is about 32K difference (to egcs -fno-gcse all alignments 
set to 2), 1.6K difference in the data segment, about 1K difference in BSS.

I have not run benchmarks to test if the egcs kernel runs faster, but then
32K aren't too bad. Linus adds that regulary just by careless use of inline
functions ;)

-Andi