Re: [PATCH v3 2/4] videobuf2-dma-streaming: new videobuf2 memory allocator

[Marek Szyprowski <m.szyprowski@samsung.com>]

On 1/7/2013 9:15 PM, Mauro Carvalho Chehab wrote:
> Em Mon, 7 Jan 2013 12:40:50 -0700
> Jonathan Corbet <corbet@lwn.net> escreveu:
>
> > On Mon, 7 Jan 2013 00:09:47 +0100
> > Alessandro Rubini <rubini@gnudd.com> wrote:
> >
> > > I don't expect you'll see serious performance differences on the PC. I
> > > think ARM users will have better benefits, due to the different cache
> > > architecture.  You told me Jon measured meaningful figures on a Marvel
> > > CPU.
> >
> > It made the difference between 10 frames per second with the CPU running
> > flat out and 30fps mostly idle.  I think that probably counts as
> > meaningful, yeah...:)
>
> Couldn't this performance difference be due to the usage of GFP_DMA inside
> the VB2 code, like Federico's new patch series is proposing?
>
> If not, why are there a so large performance penalty?

Nope, this was caused rather by a very poor CPU access to non-cached (aka
'coherent') memory and the way the video data has been accessed/read 
with CPU.

Best regards
-- 
Marek Szyprowski
Samsung Poland R&D Center