Re: Texture Paging

New Message Reply Date view Thread view Subject view Author view

Angus Dorbie (dorbie++at++sgi.com)
Thu, 04 Nov 1999 15:34:37 -0800

'Bwana' Bob Buckley wrote:
>
> Success! I am trying to render to 2 separate pfPipeWindows and in not
> setting them up to share resources I was getting a context switch of
> some sort and each pwin caused a texture page of the same texture. It
> was like it split up texture memory between the 2 pwins and if either of
> them went over their limit then paging was invoked. So, I caused each
> subsequent pwin to attach to the first and set their share masks to
> share darn near everthing, especially the display lists and textures
> (GL_OBJS).
>
> The info on the Octanes scares me. Yech!

Let's draw a distinction between Octane the platform and the graphics
option installed for example MXE (which is ostensibly Maximum IMPACT
with some improvements like bandwidth and GE speed and 4MB TRAM entry).

Cheers,Angus.

>
> Angus Dorbie wrote:
> >
> > oops, forgot to cc everyone.
> >
> > Basically the IMPACT GFX system has some page structure to TRAM which
> > the texture memory manager can't make 100% efficient use of under some
> > circumstances.
> >
> > To get best memory use the pfu's to build the texture list before
> > download. They perform some component sorting for memory useage
> > efficiency on IMPACT.
> >
> > Cheers,ANgus.
> >
> > 'Bwana' Bob Buckley wrote:
> > >
> > > Angus, hopefully, you can tell me why I'm paging textures. I'm running
> > > on a Octane with 4Mb of texture memory and Performer V2.2.6
> > >
> > > The database I have include two distinct pfTextures:
> > >
> > > Texture #1
> > > 256 X 256 X 2 = 131072 bytes
> > > loaded as default (2 bytes per texels)
> > >
> > > Texture #2
> > > 1024 X 512 X 4 = 2097152 bytes
> > > loaded as RGBA_8 (4 bytes per texel)
> > > bilinear min filter
> > > trilinear mip map mag filter.
> > > Hence, 4/3 * 2097152 = 2796203 bytes
> > >
> > > I'm running pfuDownloadTexList which confirms all but the additional
> > > memory required for the mip mapping.
> > >
> > > In channel #1 I load
> > >
> > > Texture #1
> > > Geostate #1
> > > texture enabled
> > > transparency enabled
> > > lighting disabled
> > > default texenv
> > >
> > > Texture #2
> > > Geostate #2
> > > texture enabled
> > > default texenv
> > >
> > > And in channel #2 I again load texture #2 but attached to a different
> > > geostate:
> > >
> > > Texture #2
> > > Geostate #3
> > > texture enabled
> > > lighting disabled
> > > default texenv
> > >
> > > I'm getting texture paging with avg loads equal to 1.0 and kbytes of
> > > 2097.152
> > > I watch the number of loads go from 0 to 24 then it repeats itself
> > > (i.e., after it loads 24 straight times it doesn't load for a single
> > > frame, then starts loading again).
> > > This repeating behavior changes when I run with breaks in cvd.
> > >
> > > Are there any blatant errors I'm making in the way in which this machine
> > > (or other Iris') can handle this situation? I am aware of some of the
> > > banking restrictions of the iRs but know little of the Octanes.
> > >
> > > If nothing is real obvious I can probably put it into a test case and
> > > work from there.
> > >
> > > Help!
> > > Thanks!
> >
> > --
> > "Success is the ability to go from one failure to another with
> > no loss of enthusiasm." - Winston Churchill.
> >
> > Performer + OpenGL examples and tutors: http://www.dorbie.com/
> > -----------------------------------------------------------------------
> > List Archives, FAQ, FTP: http://www.sgi.com/software/performer/
> > Submissions: info-performer++at++sgi.com
> > Admin. requests: info-performer-request++at++sgi.com

-- 
"Success is the ability to go from one failure to another with
 no loss of enthusiasm."    -    Winston Churchill.


Performer + OpenGL examples and tutors: http://www.dorbie.com/

New Message Reply Date view Thread view Subject view Author view

This archive was generated by hypermail 2.0b2 on Thu Nov 04 1999 - 15:34:43 PST

This message has been cleansed for anti-spam protection. Replace '++at++' in any mail addresses with the '@' symbol.