ATI Radeon LY Mobility M6: DRM does not work - locking issue ?
eta at lclark.edu
Thu Sep 9 01:11:18 PDT 2004
On Thu, 2004-09-09 at 00:19, Vladimir Grebenschikov wrote:
> В чт, 09/09/2004 в 08:55 +0200, Jose M Rodriguez пишет:
> > On Thu, 09 Sep 2004 10:37:44 +0400, Vladimir Grebenschikov <vova at fbsd.ru>
> > wrote:
> > > В ср, 08/09/2004 в 16:43 -0700, Eric Anholt пишет:
> > >> On Wed, 2004-09-08 at 05:23, Jose M Rodriguez wrote:
> > >> > On Wed, 08 Sep 2004 13:04:13 +0400, Vladimir Grebenschikov
> > >> <vova at fbsd.ru>
> > >> > wrote:
> > >> > ...
> > >
> > >> > Options "AGPMode" "4" # Set AGP Speed DANGEROUS but very effective
> > >>
> > >> I would be interested in reports of "very effective." The effectiveness
> > >> that I've heard from anyone so far is "not measurable" to 1-2%.
> > >
> > > Seems to be enabled successful (both mode 4 and fast-write):
> > >
> > > (**) RADEON(0): Using AGP 4x mode
> > > (**) RADEON(0): Enabling AGP Fast Write
> > > (II) RADEON(0): Depth moves disabled by default
> > >
> > > But can't see any difference with glxgears as fps metter.
> > > it is gives about 470 fps with default window size on
> > > ATI M6 16Mb + PentiumM 1700MHz
> > >
> > No. you need textures, not 3D. 3D commands can't waste I/O bandwirth.
> Ok, can you advise simple tool to make tests (I know there is GL games,
> but may be something simplier) ?
Commands do use bandwidth -- remember the mapping of"vertex/indirect and
ring buffers in the Xorg.0.log? Those are in AGP. Now, let's take for
example ipers. On my highly untuned system, I'm seeing 400,000
polys/sec. Each vertex has at least 3 coordinates plus two texture
coordinates, and I bet a color as well, so let's say (8 * 4) bytes/vert
* 4 vert/poly (I think it's drawing quads) * > 400,000 polys/sec.
That's 51MB/sec -- not that big compared to even my 1x AGP, and this is
an app designed to spew lots of vertices, but still is bandwidth used.
And that's with (I'm pretty sure) a bad compile of Mesa, and we've had
to disable some of the optimizations for producing vertices quickly in
exchange for new features (3d textures).
However, as always, glxgears is still the wrong thing to test with.
> > And even not
> > loading glx & dri modules.
> Not helps, still
> (WW) RADEON(0): Static buffer allocation failed -- need at least 17325
> kB video memory
Sounds like you made a mistake in configuration -- that only happens if
the DRI was enabled.
Eric Anholt eta at lclark.edu
http://people.freebsd.org/~anholt/ anholt at FreeBSD.org
More information about the freebsd-current