On Wed, 2 Feb 2005 18:57:02 -0500, Daniel Phillips <[EMAIL PROTECTED]> wrote: > On Wednesday 02 February 2005 15:51, Rodolphe Ortalo wrote: > > On Wednesday 02 February 2005 06:23, Daniel Phillips wrote: > > > It seems to me that only the ring buffer is needed for commands and > > > vertices, and indirect is only needed to transfer textures. I can > > > see why you'd also want to support indirect commands and vertices > > > but that still looks like an expendable goody. > > > > I have exactly the opposite opinion and would rather keep indirect > > alone than direct. > > Do you have the time to explain why you think indirect is not needed > > for commands? > > Please see my geometry throughput estimate in a different thread, which > is based on just what I see as the simplest workable arrangement. > Indirect commands have to be seen as a goody, and a bunch of potential > complexities have to be considered. For example, you can't tell just > from the ring buffer pointers when a good time to refill the ring > buffer might be, and it's tempting to elaborate the interrupt scheme to > compensate for this, so complexity breeds complexity.
I am aware of one architecture which can periodically write the "read pointer" to a location in host memory, so that the host doesn't incur any bus overhead to computer free ring buffer space. _______________________________________________ Open-graphics mailing list [email protected] http://lists.duskglow.com/mailman/listinfo/open-graphics List service provided by Duskglow Consulting, LLC (www.duskglow.com)
