[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: performance comparison of ARM 266 MHz vs i386 800 MHz



On 2006-10-19 23:36 +0200, Bill Allombert wrote:
> On Thu, Oct 19, 2006 at 10:30:29AM +0100, Catalin Marinas wrote:
> > Bill Allombert <ballombe@master.debian.org> wrote:
> > > On Wed, Oct 18, 2006 at 09:29:48AM -0400, Lennart Sorensen wrote:
> > >> So the answer really becomes, that if your software does no floating
> > >> point calculations, then the arm will be quite a fast chip, and
> > >> otherwise it will be very slow.  I would not be surprised if a 266mhz
> > >> arm is at least close to the performance of an 800mhz via eden when
> > >> doing integer calculations.
> > >
> > > Sorry to intervene but I am a bit surprised.
> > >
> > > As far as I remember, ARM has no 32x32->32 unsigned multiply and no
> > > 32x32->64 multiply. Am I wrong ?
> > 
> > MUL   is 32x32->32 both signed and unsigned
> > 
> > In architectures 4 and above:
> > 
> > UMULL is 32x32->64 unsigned
> > SMULL is 32x32->64 signed
> 
> Excellent! Is that supported by the ABI used by Debian ?

No, currently debian-arm is built for arm v3 instruction set

However debian-armel will be built for arm v4 at least, so you can use
these instructions there.

You could build something to specifically use these instructions and
not work on v3 machines (which are now very rare indeed).

Wookey
-- 
Aleph One Ltd, Bottisham, CAMBRIDGE, CB5 9BA, UK  Tel +44 (0) 1223 811679
work: http://www.aleph1.co.uk/                 play: http://wookware.org/



Reply to: