On Wednesday 27 July 2005 04:54 pm, Ian Romanick wrote:
> Patrick McFarland wrote:
> > Even if we violate precision/range stuff, being able to accelerate
> > simplistic shaders would be quite useful. Its better than not having a
> > software implementation of the shader pipeline.
>
> The problem is that most shaders that use ARB_fp or NV_fp aren't
> simplistic enough.  It would be a *lot* of work to benefit 1% of
> real-world shaders.

I think ATI really screwed R200 owners then. The shader pipeline ultimately is 
useless.

> > Also, what stops you from splitting up a shader, and running the peices
> > back to back over multiple passes? Can't you emulate longer shaders doing
> > that?
>
> So, I looked into this really deeply in the past for other things.  The
> problem is it gets *very* hard to deal with framebuffer blend modes.  If
> you have an arbitrary triangle list, triangles in the list may overlap.
>  If you have a framebuffer blend mode other than dst=src, you can't
> multipass it (generally) without breaking up the triangle list and
> sending one triangle at a time.  It would not surprise me at all if the
> performance there was close to that of a good software implementation.

So, how many games use blend modes other than dst=src? Also, even if it isn't 
faster than a good software implementation, its still less work done by the 
CPU. I own a pretty outdated P3 550, and I'd rather have any sort of boost I 
can get.

> This, BTW, is what ATI's "fbuffer" in all about.

I'm trying to find more information about this "fbuffer", but Google isn't 
being too friendly.

-- 
Patrick "Diablo-D3" McFarland || [EMAIL PROTECTED]
"Computer games don't affect kids; I mean if Pac-Man affected us as kids, we'd 
all be running around in darkened rooms, munching magic pills and listening to
repetitive electronic music." -- Kristian Wilson, Nintendo, Inc, 1989

Attachment: pgpHhlHkgPi9u.pgp
Description: PGP signature

Reply via email to