Re: PXA255, same hw speed, 4.2 vs 4.1, slower performance
- From: Valter Minute <vminute@xxxxxxxxxxxxxxxxxxxxxxxxxxx>
- Date: Wed, 28 Sep 2005 00:57:23 -0700
Jeff DeFouw <jeffd@xxxxxxxxxxxxxxx> wrote in
news:e1dxQC6wFHA.736@xxxxxxxxxxxxxxxxxxxx:
[...]
Are you doing your comparison with release builds in both cases?
Debug build disables compiler optimization and this could have an
heavvy impact on performance in some routines.
Intel provides a C/C++ compiler that should (I never tested it)
improve performance, it may be that the 4.1 has been compiled using
it, or contains some optimized code.
I suppose that you don't have the source code of the 4.1 BSP.
> 1. The first surprise was that a WaitForSingleObject(event, 0)
> loop in an application takes twice as long on the new board. Is
> this function slower in CE 4.2 vs 4.1, or is there an optimization
> that I can do that affects this function? I don't really need
> this function to be that fast but I thought it might be a sign.
>
Check how many threads and how many processes are running on your
system. It may be that the old board has an image with less
application or less device drivers.
You may use kernel tracker to understand if a process (or a driver)
is taking CPU time (for example to poll a device or something like
that) or if a process is spawning many threads.
> 2. The display driver is 50% faster on the old board using a test
> program drawing a zigzag with the GDI (20% faster with our real
> application). There's no hardware graphics acceleration on either
> board and they use the same driver base. Does this have something
> to do with 4.2 or is it likely that the old board company has
> discovered some additional software optimizations and added it to
> their display driver?
The display driver for the PXA has some assembly routines that will
be built only if a preprocessor macro has been defined, check that
this macro is declared in your driver SOURCES build file (you should
find a CDEFINES=$(CDEFINES) -D<macro name>).
--
Valter Minute
(the reply address of this message is invalid,
but you may be able to guess the right one)
.
- References:
- PXA255, same hw speed, 4.2 vs 4.1, slower performance
- From: Jeff DeFouw
- PXA255, same hw speed, 4.2 vs 4.1, slower performance
- Prev by Date: Re: device driver compilation problem
- Next by Date: KernelIOControl from an application
- Previous by thread: Re: PXA255, same hw speed, 4.2 vs 4.1, slower performance
- Next by thread: Re: PXA255, same hw speed, 4.2 vs 4.1, slower performance
- Index(es):
Relevant Pages
|
Loading
