On Mon 09 Dec, A.B. Robertson wrote:
>
>                                                       ....    Considering
> that the multiply instruction is now apparently 40 times faster I would
> think that some of the **emulated** floating point instructions would now
> execute at a considerably faster rate. This leads me to another question,
> could someone do this test on a StrongARM processor with RiscOS 3.7 and with
> an emulator which is StrongARM compliant, and compare it to a PowerPC and a
> Pentium of some sort? Then we could see how hobbled (or hopefully not) the
> StrongARM processor is without an FPU.

I'm just trying to post the following to the c.s.a.p news group, and it
seems relevant even though it's under RISC OS rather than unix. My other PC
really *is* a P120 but I won't be able to run this test on it until next
week. I have none of these 'standard' benchmarks here and in fact don't go
for them much; real programs doing real things seem to be a better test.
Here's the article; you may be able to scale the results if you have some
comparative benchmarks for different PC cards and 'real' PCs.  - Ian

------------------------------------------------------------------------

For anyone suffering (as I have for the past few years) from knee-jerk
cynicism regarding the FP capabilities of the RiscPC...

I accidentally ran a comparison between my 486DX2/80 (Aleph 1) and my
StrongARM chip. This was a big number-crunching program, involving much
inversion of matrices, finding of eigenvectors, etc., and I was amazed
to find the SA chip just over 2 times *faster* than the 486DX2/80. This is
surprising because the 486/80 card was 160 times *faster* than my ARM600
when I first put it in. This test is not a short one; it takes 19 minutes
on the SA-110 and 41 min on the PC card (single-tasking, 16 meg).

ADMITTEDLY: (1) the StrongARM MUL instruction has been optimised and the
chip is clocking faster, (2) the PC program was compiled with GCC, whilst
the RISC OS version with Acorn C/C++, (3) my program is now bigger than it
was when I got the 160x result [although it still only takes a 2 MB slot]
and (4) a 486DX is practically stone age by today's standards ("my other PC
is a Pentium 120...") but still...

I mention this because I only bought the PC card for number-crunching and
simply would not have believed that even the StrongARM (magnificent though
it is...) was worth using *at all* for FP code.

Of course, if the SA-110 *really* had a floating point unit...

Regards, Ian.

-- 
Ian Giblin                             mailto:giblin@dm.unipi.it
Dipartimento di Matematica              faxto:+39 50 599524 (shared)
Universita di Pisa                    voiceto:+39 50 599573 (direct)
via Buonarroti 2
I-56127 PISA,  Italy.               http://adams.dm.unipi.it/~giblin/