Bernard Leclerc (bleclerc++at++cae.ca)
Thu, 31 Oct 1996 10:15:11 -0500
> While trying to figure out why an app runs slow when using
> an MCO to split the screen into 3 sections, we've discovered
> that our 2 RM machine chokes when the pixel depth is small.
> As long as the pixel depth is medium the app runs nicely.
Small/medium/large pixel depth refer to the number of bits per pixel. It
relate to the capability to simultaneously configure the frame buffer for
multi-sampling, accumulation buffer, stencil, z-buffer, etc.
With a small pixel depth, you might not be able to obtain any sub-samples
for anti-aliasing. With a medium pixel depth, you'll obtain 8 sub-samples.
Really, the pixel depth doesn't affect the performance of your application
if you're able to obtain the frame buffer configuration you need.
> So, we are faced with either using only 2 screens, use 3
> screens and buying 2 more RM's or figure out what we can do
> to the program to make it run fast enough with small pixels :)
The fact that your program runs faster with 2 screens than with 3 screens
indicate that you're exceeding the RM's fill-rate capacity. To verify this
statement, try using the MCO with differents VOFs. If I recall correctly,
you're using a resolution of 960x680. Try using 3++at++800x600_60. You'll still
have only a small pixel depth available but the total number of pixels
will drop from 1,958,400 to 1,440,000. If it's not enough, try
6++at++640x480_60 and only drive 3 channels out of 6 (for a total of 921,600
pixels) to verify if you're limited by the fill-rate of your 2 RMs.
Raster managers work in parallel to fill the channels on your screen. When
configuring the MCO for 2 screens, the 2 RMs each fills one channel. When
configuring 3 channels, the 2 RMs now have a larger amount a pixels to
fill - it takes more time.
> My program is based on multichannel.c example. What kind of
> things can I turn off to gain speed? Things that might be
> emulated by performer under small pixels, but done in hardware
> at medium ?
The only thing that could be emulated in software when using a small pixel
depth is the accumulation buffer. I don't think you're using it.
As far as I know, there's nothing you can turn off to gain speed unless
you're willing to drop textures. And even that might not be enough.
> We are using Irix 5.3 and performer 2.0. The machine is a desk
> side onyx , 2 CPU's, 2 RM's , one MCO.
Again, I see only 2 solutions: reduce the size of the channels, or add 2
raster managers.
--
Bernard Leclerc CAE Electronics Ltd., 8585 Cote De Liesse
Technical Leader Saint-Laurent, Quebec, Canada, H4L-4X4
3-D Graphics Applications tel: +1 514 341 2000 extension 2275
bleclerc++at++cae.ca fax: +1 514 340 5496
=======================================================================
List Archives, FAQ, FTP: http://www.sgi.com/Technology/Performer/
Submissions: info-performer++at++sgi.com
Admin. requests: info-performer-request++at++sgi.com
This archive was generated by hypermail 2.0b2 on Mon Aug 10 1998 - 17:53:50 PDT