Re: CPU results on Freescale i.MX

[David Laight <david%l8s.co.uk@localhost>]

On Mon, Jan 24, 2011 at 06:43:52AM -0500, Christopher Berardi wrote:
> I wrote a simple program to 'stress-test' a cpu. Basically it is a program 
> that
> computes a couple of sqrt roots a billion or so times, which I have used as
> a (crude?) benchtest. 
...
> Genesi Efika MX (Smarttop) with Freescale i.MX 515 (ARMv7 processor rev5 
> (v7l))
> running Ubuntu 10.10
>     real 37m15.173s
...
> Lenovo T61 with Intel T9300 (Intel Core2 Duo @ 2.5GHz)
> running Ubuntu 10.04
>     real 15m43.094s
..
> Homebuilt with AMD (AMD Athalon dual-core @ 2.5GHz)
> running NetBSD 5.1
>     real 182.95s (3m02.95s)

Something odd is going on.
The ARM will be using soft float - so should be much slower than
either x86 system even at the same clock speed!

You then need to look at exactly which form of FP sqrt() the x86
systema are using. Could be soft, could be the i386 fp stack, or
could be the SSE(n).
In general the latter is fastest.
However gcc may not generate the sse instructions for netbsd because
we still, build binaries that will run on a 486.

        David

-- 
David Laight: david%l8s.co.uk@localhost