Dog-ish ARM performance?

Hi,

I have a chance to repurpose some wireless displays
for an application, here.  Presently run WinCE 5 (?).

I believe there is a PXA270 @ 300MHz under the hood.

But, the performance appears *abysmal*.  Should I
attribute this to MS's typical ineptitude at extracting
performance from processors?  (I have no first-hand
experience with WinCE but to assume it follows boldly
in the MS tradition  :< )

I don't want to invest time reverse engineering the
devices only to discover the underlying iron is the
actual "problem".

(highest performance task envisioned would be
streaming video to the display)

Any pointers on a suitable place to compare apples and
oranges in terms of performance criteria?

Thanks!
--don

D Yuniskis <not.going.to.be@seen.com> wrote:
> Hi,

> I have a chance to repurpose some wireless displays
> for an application, here.  Presently run WinCE 5 (?).

> I believe there is a PXA270 @ 300MHz under the hood.

An Intel XScale PXA270 shouldn't suck that badly however as with
anything it depends what the rest of the system design is like.

There is one big thing, you must tell the compiler that it's compiling
for an XScale as there is a set of conditional instructions which are
used for return from functions, which unlike other ARM implementations,
XScales take as long to not execute as to execute. Telling the compiler
that it's compiling for an XScale will cause it not to generate these
instructions.

> (highest performance task envisioned would be
> streaming video to the display)

If you are needing to stream compressed then uncompressing on the
device then I'd give up on these devices and find a Cortex-A8 or
Cortex-A9 based tablet that you can hack.

-p
-- 
Paul Gotch
--------------------------------------------------------------------

Hi Paul,

On 5/14/2011 2:49 PM, Paul Gotch wrote:
> D Yuniskis<not.going.to.be@seen.com>  wrote:
>> I have a chance to repurpose some wireless displays
>> for an application, here.  Presently run WinCE 5 (?).
>> I believe there is a PXA270 @ 300MHz under the hood.
>
> An Intel XScale PXA270 shouldn't suck that badly however as with
> anything it depends what the rest of the system design is like.
>
> There is one big thing, you must tell the compiler that it's compiling
> for an XScale as there is a set of conditional instructions which are
> used for return from functions, which unlike other ARM implementations,

I wouldn't currently have any control over that as the device is
just running "existing code" (though I will make note if I opt to
pursue this iron).

> XScales take as long to not execute as to execute. Telling the compiler
> that it's compiling for an XScale will cause it not to generate these
> instructions.
>
>> (highest performance task envisioned would be
>> streaming video to the display)
>
> If you are needing to stream compressed then uncompressing on the
> device then I'd give up on these devices and find a Cortex-A8 or
> Cortex-A9 based tablet that you can hack.

<frown>  Sadly, "The price is right"  :<

Thanks for your comments!  I'm going to see if I can find a
prebuilt binary that claims to run on CE to "display videos"
and see how badly (?) it performs.  Of course, finding something
that will drive VGA instead of QVGA may be a challenge (I think
most CE stuff is geared towards phones, etc.?)

But, your comments make sense in light of the types of applications
these *appear* to have been targeted towards (kiosk, etc.).  Maybe
just pull the displays, touch panels and packaging and plan on
redesigning the core electronics (?).  Unfortunately, this was
at the ideal power (consumption) point for my needs...

Thx,
--don

Hi Paul,

[followup]

On 5/14/2011 2:49 PM, Paul Gotch wrote:
> D Yuniskis<not.going.to.be@seen.com>  wrote:

>> I believe there is a PXA270 @ 300MHz under the hood.
>
> An Intel XScale PXA270 shouldn't suck that badly however as with
> anything it depends what the rest of the system design is like.
>
>> (highest performance task envisioned would be
>> streaming video to the display)
>
> If you are needing to stream compressed then uncompressing on the
> device then I'd give up on these devices and find a Cortex-A8 or
> Cortex-A9 based tablet that you can hack.

Apparently the device contains a 2700 to handle/assist with
the display.  I believe these have some (crude) acceleration
capabilities... (?)

I'll poke around and see if I can find a tool to (effectively)
give me the "probe()" results for the hardware (never played
with WinCE before and not real keen on investing much in it
now!  :> ).  It may be easier to just peel the thing apart and
poke around inside (though I am always fearful of mangling
antennas, displays, touchpanels, etc. in the process)