Re: Hardware questions...

New Message Reply Date view Thread view Subject view Author view

Avi Bar-Zeev (avi++at++oz.net)
Thu, 30 Jan 1997 20:46:32 -0800 (PST)

To speed up your framebuffer approach, you can open a second, much smalller,
off-screen channel and re-render the scene (capture the first render to a
displaylist). Or, better yet, use the framebuffer-to-texture trick to render
the second channel. By cranking multi-sampling all the way up, you may be
able to use the hardware antialiasing (gausian filtering, right?) to speed
up your calculations.

You might get a big boost with the fb-to-texture trick if your second
channel is 1x1 pixel and you can somehow get the texture quickly MIP-map
processed. But this may be pushing it.

Avi

At 04:34 PM 1/30/97 -0700, Jeff Clark wrote:
>
>Hello all,
>
>I am having difficulty implementing an "automatic gain control" for my
>Performer app. By AGC, I mean the gain on each pixel is a function
>of the average pixel value (intensity... one color component) of the
>current frame (or previous frame, Ill take what I can get). The correct
>result would be constant overall brightness on the display.
>
>I have implemented this in a handful of ways, but all of my attempts EITHER
>make it impossible to maintain my target frame rate OR don't provide
>an adequate AGC performance (I would like an AGC update rate of ~15Hz).
>
>There are two issues involved... here they are along with my attempted
>solutions:
>
>-Come up with an accurate (or approximate) average pixel value at ~15Hz:
>
> -approximate average pixel value using an ogl MINMAX histogram. This
> is fast, but the approximation is unsatisfactory.
>
> -approximate average pixel value by reading a slice of the framebuffer
> (glReadPixels()) every frame. i.e. if I read 32 lines/frame, I read
> the entire FB every 32 frames. So what I have is an "idea" of the
> brightness of the entire scene over the last 32 frames. This introduces
> temporal inaccuracies that are probably unacceptable. This actually
> looks pretty good a lot of the time, but sometimes not...i.e. in the
> worst case a bright source moves into the scene and has no effect on
> the AGC until 31 frames later. At 60 fps, this is an effective AGC
> update of 2 Hz... unacceptable.
>
> -approximate average pixel value using an ogl MINMAX histogram of one
> color component. The penalty to the draw process is comparable to
> using glReadPixels() and, therefore would be done in slices. Now
> I have the same problem as in the last attempt.
>
>-Apply a gain to the entire display based on the result of the first task
> above:
>
> -accumulate the 1 million pixels in the accumulation buffer and set the
> acbuf gain to the desired gain. This, of course, works fine but takes
> up so much draw that I don't have enough time to draw my scene at the
> complexity my simulation requires.
>
>
>SO, anybody have any ideas? (in the context of a 2 pipe iR with Sirius option)
>
>I am starting to think of ways to do it near the end or after the pipeline:
>
>On the gain issue, I wonder if there is a way to apply a gain (analog or
>digital) in the DG... i.e. near the DAC or gamma tables? In the ircombine
>channel attributes window I notice a GAIN field... is this a user register
>that can be changed at 15 Hz? (ASD, if this doesn't exist it sure would be
>a groovy thing to have!) Or, could the Sirius box apply a gain offline
>(and do so quickly) (sorry, I'm Siriusly ignorant)?
>
>On the pixel averaging issue, I wonder again if the Sirius can't be used?
>
>If not Sirius, I would think it would be posible to find a specialized
>hardware box that performs my AGC... i.e. a box that takes analog video
>and outputs a signal with constant overall brightness. Anybody ever heard
>of such a gizmo?
>
>Apologies:
>
>Im sorry that these aren't -exactly- Performer questions, but they are SGI
>questions... and this list is the best resource I know of. Besides, it is
>difficult to get this kind of info from support.
>Im sorry that this post is long enough to cut into your lunch break!
>
>
>Any responses will be GREATLY appreciated... I'm am beginning to run out of
>ideas.
>
>
>=======================================================================
>List Archives, FAQ, FTP: http://www.sgi.com/Technology/Performer/
> Submissions: info-performer++at++sgi.com
> Admin. requests: info-performer-request++at++sgi.com
>
>
Avi Bar-Zeev -- "Every trip begins with a single misstep"
A Real-Time 3-D Animated Character -- avi++at++oz.net http://www.oz.net/~avi
WOLFCHILD-3D Consulting -- avi++at++wolfchild.com
http://www.wolfchild.com/3d

=======================================================================
List Archives, FAQ, FTP: http://www.sgi.com/Technology/Performer/
            Submissions: info-performer++at++sgi.com
        Admin. requests: info-performer-request++at++sgi.com

New Message Reply Date view Thread view Subject view Author view

This archive was generated by hypermail 2.0b2 on Mon Aug 10 1998 - 17:54:30 PDT

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