I was also wondering if MV or TI had found a better workaround that we
are not aware of.
Thanks,
Craig
On Thursday 28 May 2009 5:13:30 pm Eric Nelson wrote:
> Craig Matsuura wrote:
> > So you said MV claims there is a fix. Do you know what the fix is?
If we
> > can run at 1280x720 with out the starvation we would be interested in
> > trying it. However we never had anything that complete rid us of the
> > problem.
>
> Don't get me wrong. I certainly don't speak for (or against) MV.
> I just think I first heard about the fix on the linux-davinci mailing
list
> in the context of someone using MontaVista.
>
> We've been using the git kernel forever.
>
> We're running 1280x800 with a full OSD layer (as 16-bit RGB565) without
> the DMA under-run being a problem. That's not to say that we can't make
> the problem happen if we try by stressing the USB, SD card and ethernet
> channels concurrently, but it's not a problem because the combinations
> don't happen in our customers' production apps.
>
> > Thanks,
> > Craig
> >
> > On Thursday 28 May 2009 2:59:09 pm Eric Nelson wrote:
> > > Craig Matsuura wrote:
> > > > We actually referred to it a tear. It is a starvation of the FIFO's
> > > > on the OSD0, the VID0 does not have this problem as they have
larger
> >
> > FIFO's
> >
> > > > (This is straight from the silicon engineers at TI).
> > > >
> > > >
> > > > We provided a partial solution to TI and unfortunately MV was of no
> >
> > help
> >
> > > > when we addressed it with them. I don't recall the exact fix, but
> >
> > it had
> >
> > > > something to do with a PBBBR register and provided this to TI. No
> >
> > matter
> >
> > > > who or how it was fixed if you put a heavy load on the kernel the
> > > > video on the OSD will start giving you a tearing like effect (or
> > > > shift). I think we did a dd if=/dev/zero of=/dev/null and it was
> > > > really bad.
> > >
> > > Yep. That's the one I was referring to.
> > >
> > > We saw it when doing heavy ethernet and/or SD-card traffic.
> > >
> > > > Even if we had other layers off if had the problem at 1280x720. So
> > > > we ended up on the vid0 layer.
> > > >
> > > >
> > > > Thanks,
> > > > Craig
> > > >
> > > > On Thursday 28 May 2009 2:34:38 pm Eric Nelson wrote:
> > > > > Craig Matsuura wrote:
> > > > > > Thanks for the quick response. Actually from my understanding
> >
> > on the
> >
> > > > > > tearing issues (this is direct from TI Engineers), the FIFO on
> > > > > > the OSD0 are too small when displaying at 1280x720 so there is
> > > > > > no
> >
> > way to
> >
> > > > > > use the OSD at 1280x720.
> > > > >
> > > > > I think you're referring to something other than tearing. At
> > > > > resolutions over around 1024x768, you'll get DMA starvation
in the
> > > > > video output on the OSD. This produces a really weird behavior
> > > > > where the OSD layer shifts left for the rest of the frame.
> > > > >
> > > > > We've only seen it when enabling VID0 with OSD0 and OSD1 set
up as
> > > > > an overlay with transparency (I forget what TI calls this mode).
> > > > >
> > > > > There's a partial fix for both the MV and git kernels that
> > > > > involves changing around the DMA priority for the video output.
> > > > > There's also a 'spraa' or 'sprue' doc that discusses things
on the
> > > > > TI web-site.
> > > > >
> > > > > > How does directfb access the dma blitting?
> > > > >
> > > > > If memory serves, the DFB patches use /dev/mem to allocate the
> > > > > memory directly by the app (instead of using cmem or the dav-dma
> > > > > allocation routines), then hands the physical memory addresses to
> > > > > the dav-dma
> > > >
> > > > driver.
> > > >
> > > > > You'll need to configure DFB to steer clear of the RAM used
by the
> > > > > codec engine and DSP.
> > > > >
> > > > > > Craig
> > > > > >
> > > > > > On Thursday 28 May 2009 1:48:18 pm Eric Nelson wrote:
> > > > > > > Hello Craig,
> > > > > > >
> > > > > > > Craig Matsuura wrote:
> > > > > > > > I am also curious about the progress. I am also interested
> > > > > > > > in speeding up blitting in directfb. I am currently on
> > > > > > > > directfb 1.2, I am using the fbdev not the mem mapping as I
> > > > > > > > need it for an app that runs on the vid0 in rgb24 bit mode.
> > > > > > >
> > > > > > > The latest driver can be found in our Git repository
> > > > > > > (file drivers/misc/dav-dma.c)
> > > > > > > http://git.boundarydevices.com/git-repos?p=linux-bd
> > > > > > >
> > > > > > > > I can not run on the osd0 as resolutions higher than
640x480
> > > > > > > > on the OSD will cause video tearing. I run the vid0 at
> > > > > > > > 1280x720 (720p).
> > > > > > >
> > > > > > > Using DMA for blts will help with tearing, but you'll likely
> > > > > > > also
> > > >
> > > > need
> > > >
> > > > > > > to sync with the vertical refresh interrupt.
> > > > > > >
> > > > > > > > I have contemplated using the /dev/mem but was unsure if it
> > > > > > > > was really worth the effort. I am also using the DSP with
> > > > > > > > Codec
> > > >
> > > > Engine 2
> > > >
> > > > > > > > from
> > > > > >
> > > > > > TI so
> > > > > >
> > > > > > > > I can not use the c64 acceleration.
> > > > > > >
> > > > > > > The "blt" operations in dav-dma should inter-operate with CE2
> >
> > just
> >
> > > > > > > fine without the acceleration code in DFB.
> > > > > > >
> > > > > > > > On Wednesday 02 January 2008 8:36:20 pm Sriram Neelakandan
> >
> > wrote:
> > > > > > > > > Hi Eric,
> > > > > > > > >
> > > > > > > > > I stumbled on this posting recently..
> >
> >
http://mail.directfb.org/pipermail/directfb-users/2007-November/000123.ht
> >
> > > > > > > >ml
> > > > > > > >
> > > > > > > > > I am looking to use the driver ..
> > > > > > > > > Do u have any latest updates ?
> > > > > > > > > Also it will be of great help, if you could post a sample
> >
> > user
> >
> > > > > > > > > space program, that demos the usage of your driver.
> > > > > > >
> > > > > > > Sorry for that, Sriram. I haven't had time to keep up
with the
> > > > > > > Direct-FB mailing list and didn't notice your post.
> > > > > > >
> > > > > > > I also haven't touched base with the DFB team to know whether
> > > > > > > they've been keeping the DMA blt support up-to-date.
> > > > > > >
> > > > > > > When integrated with Direct-FB, the standard DFB demos will
> > > > > > > work
> > > >
> > > > (only
> > > >
> > > > > > > faster).
> > > > > > >
> > > > > > > Eric
> > > > > >
> > > > > > --
> >
> >
------------------------------------------------------------------------
> >
> > > > > > Craig Matsuura - Principal Engineer
> > > > > > Control4
> > > > > > 11734 South Election Road - Suite 200
> > > > > > Salt Lake City, UT 84020-6432
> > > > > > PH: 801-523-3161
> > > > > > FX: 801-523-3199
> >
> > --
> >
> >
> >
------------------------------------------------------------------------
> >
> >
> > Craig Matsuura - Principal Engineer
> >
> > Control4
> > 11734 South Election Road - Suite 200
> > Salt Lake City, UT 84020-6432
> > PH: 801-523-3161
> > FX: 801-523-3199
> >
> >
> >
------------------------------------------------------------------------
> >
> > _______________________________________________
> > directfb-users mailing list
> > [email protected]
> > http://mail.directfb.org/cgi-bin/mailman/listinfo/directfb-users
--
------------------------------------------------------------------------
Craig Matsuura - Principal Engineer
Control4
11734 South Election Road - Suite 200
Salt Lake City, UT 84020-6432
PH: 801-523-3161
FX: 801-523-3199
------------------------------------------------------------------------
_______________________________________________
directfb-users mailing list
[email protected]
http://mail.directfb.org/cgi-bin/mailman/listinfo/directfb-users
