Re: [AD] Memory Bitmap to Display blit slow?

[ Thread Index | Date Index | More lists.liballeg.org/allegro-developers Archives ]


On Sat, February 28, 2009 2:58 pm, Thomas Fjellstrom said:
> I've been noticing this for a while now, but under linux, with the nvidia
> binary drivers, bliting a memory bitmap to a regular allegro 5 display is
> horrendously slow.
>
> Using SiegeLord's ex_prim example, when you change to software mode fps drops
> to 4fps on my (q6600+8800GTS) system. SiegeLord mentioned once that if you
> remove the final blit, the fps goes back up to normal, so it is not the
> software primitive rendering causing the slowdown. I'm confused as to why it
> would be so slow.
>
> I tried setting the new bitmap format before creating the memory bitmap to the
> current display flags, but if it did help, it wasn't by much, maybe 1fps. So
> it doesn't seem to be a expensive conversion issue.  that leaves texture
> uploads being really slow, or the copy from the bitmap to a texture compatible
> data structure being the cause.
>
> Anyone else have any ideas? As it is, bliting memory bitmaps to a display is
> unusable on my system right now.

I believe it was Peter who already filed a bug report on the slowness of
the memory blenders. They aren't optimized at all. They are very very
generic. They could be made much faster, which would be easy but
time consuming considering all the different pixel formats. Assuming
We do blending with only 1 format (converting to and from that format
with the conversion functions that exist), we would still need ~16 optimized
memory blenders. So if anyone wants to do that go ahead.

> Also, in a related note, when I turn the same example's blending off, the text
> disappears and the primitives change brightness.

This is expected, since the text uses alpha blending and the primitives
are normally translucent.

Trent :{)>




Mail converted by MHonArc 2.6.19+ http://listengine.tuxfamily.org/