Around 19 o'clock on Sep 23, Andy Isaacson wrote:
> I just plain don't understand why PseudoColor emulation has to be so > expensive. Can't you just do it in the same manner as the ShadowFB > layer, whereby the screen representation is updated only every 10 ms or > whatever? Yes, that's the usual implementation (although there are others). The problem is applications expect that updates to the colormap are cheap, but sending the entire frame buffer over the AGP bus is quite expensive, as you say, optimizations are possible, but it takes some clever code to take advantage of any kind of pipelining on the AGP bus while skipping the large unmodified parts of the screen. One thought I just had was that Xlib/Xnest style hacks could give every client a private colormap. This would limit the damage to that clients windows instead of forcing a scan of the whole screen. The RandR question remains though -- if this approach is sufficient to resolve the needs of 80% of the clients, is the ability to swap which visuals are mapped directly to the hardware interesting enough to spend significant effort on in the near future? Or should we just leave RandR as the resize/rotate extension and leave hardware visual swaps for some future extension if we ever need it ? Keith Packard XFree86 Core Team HP Cambridge Research Lab _______________________________________________ Xpert mailing list [EMAIL PROTECTED] http://XFree86.Org/mailman/listinfo/xpert
