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

["Stefanos A." <stapostol@xxxxxxxxxx>]

Την Sun, 01 Mar 2009 22:55:27 +0200,ο(η) Milan Mimica  
<milan.mimica@xxxxxxxxxx> έγραψε:

> Thomas Fjellstrom wrote:
> > On Sun March 1 2009, Milan Mimica wrote:
> > > Thomas Fjellstrom wrote:
> > > > 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.
> > > Upoading a bitmap from system memory to OpenGL texture is just slow, as
> > > ex_membmp shows it. That's how I see it and never asked why.
> >
> > It's not 4fps on a Quadcore with a 8800GTS slow.
>
> Reminds me, AllegroGL had this comment and I never looked into it:
> /* XXX On NV cards, we want to use BGRA instead of RGBA for speed */

My experience has been that BGRA is ~10% faster than RGBA when uploading  
texture data, as the driver does not have to swizzle the components  
around. This also affects Ati cards. (It's been a long time since the  
test, so things may have changed - but somehow I doubt it.)

Some recent posts on GameDev (or was it Beyond3D?) indicated an  
unoptimized upload speed of ~400MB/sec from system memory to OpenGL. The  
poster was able to improve speed significantly by breaking up the updates  
in smaller calls to glTexSubImage2D, presumably because a single big call  
would thrash the cache.

Which reminds me: you are using glTexSubImage2D instead of glTexImage2D  
for blitting, right? The latter allocates video memory anew and will  
positively destroy performance.