'Re: [Dri-users] Status of Savage Driver'

[prev in list] [next in list] [prev in thread] [next in thread] 

List:       dri-users
Subject:    Re: [Dri-users] Status of Savage Driver
From:       Alex Deucher <alexdeucher () gmail ! com>
Date:       2005-11-30 19:28:06
Message-ID: a728f9f90511301128l2d6b908ag79305018a0284505 () mail ! gmail ! com
[Download RAW message or body]

On 11/30/05, Robert S. Kerr <rskerr@dsbox.com> wrote:
> Alex, Felix,
>
> Thanks both of you for your help here.
>
> I'm planning to try again and see what I find.
>
> The apps I was running previously (and will be running again) are an
> xserver and MythTV.  MythTV does video playback so there is use of
> streams and Xv scaling going on in there I'm pretty sure.  The problem
> before was that replaying video was taking way more CPU with the DRI
> enabled than without.  To explore some more I found a couple of really
> simple apps that did similar operations (I don't remember the name, but
> I think one was called xvtest or something like that.  It just displayed
> an image (single frame animation?) and used Xv to scale it to the window
> size.)  Dragging that window around caused a CPU spike.  It might have
> been all windows but I don't remember.
>

Ah!  try turning off BCIforXv.  theoretically using the BCI for image
transfers should be faster, but it's always been pretty buggy, plus I
think in that particular instance it's using MMIO rather than DMA.

> Its entirely possible that I botched up the X install last time around
> too.  Everything seemed to work okay though.  The logs showed DRI being
> enabled, but if I remember correctly, getting everything installed took
> a couple of tries, it could be that some part of my work messed up the
> rest of the X install.
>
> The last e-mail I sent was 28 FEB 05, so I probably had the shadow
> status stuff as I was using CVS at the time.  If you check this link:
>
> http://archive.netbsd.se/?ml=dri-users&a=2005-03&t=729479
>
> You can see what I wrote then and your responses.  Not all of it is
> making sense to me now as I've forgotten some of what I'd learned about
> this stuff and I don't know diddley about how X drivers work or I'd try
> to hack this out myself, but I remember at the time, your responses made
> sense to me.  You mentioned something about DRM not using IRQs and that
> being the likely source, but then also mentioned that if the driver used
> BCI instead of MMIO it might be faster.  At the time you both mentioned
> that you had other things you were working on and so wouldn't be doing
> anything in this area anytime soon.
>
> I'll have some time Thursday and Friday to play around with this stuff
> again and will be happy to report my progress.  Is the users list the
> right place?

yes.

>
> Rob
>
> Felix Kühling wrote:
> > Am Mittwoch, den 30.11.2005, 12:33 -0500 schrieb Alex Deucher:
> >
> >> On 11/30/05, Robert S. Kerr <rskerr@dsbox.com> wrote:
> >>
> >>> Hello,
> >>>
> >>> I'm writing to ask about the status of the savage driver and DRI in X.org.
> >>>
> >>> I have a Linux machine that I use for MythTV (a homebrew PVR app)
> >>> running FC3 with an onboard Prosavage DDR chip in it.
> >>>
> >>> A while back (FEB '05), I spent the time to get DRI working on my X.org
> >>> install.  It worked great, but at the time caused other performance
> >>> issues when dragging windows that made the benefits of accelerated
> >>> openGL not worth it.  The issue was that when dragging windows around on
> >>> the screen, or under Myth when I play video, there is a spike in my
> >>> Xserver's CPU utilization.  Without DRI, Xserver utilization never broke
> >>> 10% or so.  When DRI was enabled, it would jump to as high as 50%
> >>> utilization.
> >>>
> >>>
> >> I don't see how the DRI would affect regular 2D performance unless
> >> some app you are using uses the drm or GL somehow.
> >>
> >
> > The tiled frame-buffer might have a bad performance impact on direct
> > access to the frame buffer. I had the impression that software fallbacks
> > were very slow in the Savage and my guess was that the translation from
> > linear for tiled memory layout in hardware may be the cause.
> >
> > Also the Xv implementation is new in current CVS compared to Xorg 6.8.
> > It adds smooth scaling (with some caveats) on the Savage 4, ProSavage
> > and Twister. It may also have different performance compared to Xorg
> > 6.8.
> >
> > Another thing: I did some work on shadow status in January which might
> > have an impact on performance under some circumstances. Robert, I'm not
> > sure if your bad experience was before or after those changes.
> >
> > Regards,
> >   Felix
> >
> >
> >>> At the time, Felix Kuhling mentioned that the driver was essentially
> >>> very inefficient and didn't make use of IRQs or DMA to speed things up.
> >>>
> >>> My question is if this has been resolved in recent work with DRI and
> >>> x.org has resolved either of these two issues.  I've looked through the
> >>> archives but haven't seen any direct reference, but I noticed that there
> >>> has been a fair amount of savage work in the interim.
> >>>
> >>> Can anyone provide some insight?
> >>>
> >>> Also, when I compiled DRI in the past, I had to get the CVS version of
> >>> X.org, CVS Mesa, and CVS DRM and build all the parts separately.  I've
> >>> noticed that X.org now includes versions of Mesa and DRM in its tree.
> >>> Can I use these to compile DRM for the Savage?
> >>>
> >>>
> >> the drm and mesa were always included in monolithic xorg releases,
> >> however, savage DRI support in mesa and drm was not merged in until
> >> recently. you can now use either.  the version of mesa in xorg cvs is
> >> 6.4.x, while cvs mesa and drm as the latest and greatest.
> >>
> >> Alex
> >>
> >>
> >>> Rob
> >>>
> >>>
> >>>
> >>>
> >>>
> >> -------------------------------------------------------
> >> This SF.net email is sponsored by: Splunk Inc. Do you grep through log files
> >> for problems?  Stop!  Download the new AJAX search engine that makes
> >> searching your log files as easy as surfing the  web.  DOWNLOAD SPLUNK!
> >> http://ads.osdn.com/?ad_idv37&alloc_id865&op=click
> >> --
> >> _______________________________________________
> >> Dri-users mailing list
> >> Dri-users@lists.sourceforge.net
> >> https://lists.sourceforge.net/lists/listinfo/dri-users
> >>
> >>
> >>
>
>
>
>


-------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc. Do you grep through log files
for problems?  Stop!  Download the new AJAX search engine that makes
searching your log files as easy as surfing the  web.  DOWNLOAD SPLUNK!
http://ads.osdn.com/?ad_idv37&alloc_id865&opÌk
--
_______________________________________________
Dri-users mailing list
Dri-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/dri-users

[prev in list] [next in list] [prev in thread] [next in thread]
Configure | About | News | Add a list | Sponsored by KoreLogic