From: Christopher Phillips (shrydar_at_jaruth.com)
Date: 2004-11-04 05:12:35
On 4 Nov 2004, at 01:23, Steve Judd wrote:
> Hola CJ,
>
>> Do you still do fullscreen clears rather than writing to the screen
>> with an eor-plotter?
>> Or are you referring to the cost of clearing the eorbuffer?
>
> Actually, I have a very low opinion of eor-plotters (solid polygons).
solid? Pattern fills like the ones your lib supports are pretty easy
to add to an eor-plotter. Texture mapping or gouraud shading's another
issue altogether of course.
> They are slow and obnoxious, with special cases and such to handle, and
> there are much better -- faster, cleaner, more flexible -- ways to do
> it.
They are hard to get working, and at the moment you have much better
performance than me on simple convex objects, but I suspect with a bit
more optimisation a good eor-plotter will win if there's much overdraw.
I only just got the routines in Effluvium debugged/working in time to
release it - there's quite a bit of scope for improving the code.
Congrats on the speed of lib3d btw - I still haven't gotten around to
playing with it, but your quoted 8-10 fps on the Elite craft is v.
impresssive.
>
> Yes, I do fullscreen, or at least full-the-area-containing-data clears.
> Or did, I guess, since this was all years ago. You really have to, if
> you're drawing more than one object. For a single object, my
> recollection
> is that once you factor in overhead it's always faster, unless you're
> drawing just one or two polygons, and only maybe then.
hmm - I overwrite the entire screen every refresh, but each byte is
only touched once regardless of the number of polygons.
>
>>> As to organization, vertical organization is preferred for any
>>> plotting.True for BOBs etc, but I would have thought less applicable
>>> to 3d?
>
> Any routine involving an x,y coordinate benefits, especially line and
> polygon routines.
I use a pair of charmaps myself (32*16), but that's mostly to save me
having one eor-fill routine per character row - it's a memory saving
rather than a speed thing, and one that would not be required with a
hardware eor-buffer such as the one I proposed in an earlier mail in
this thread.
later,
Christopher.
Message was sent through the cbm-hackers mailing list
Archive generated by hypermail pre-2.1.8.