Re: [Mesa3d-dev] Status of intel (i915 driver) pageFlip implementation

[Keith W. <ke...@tu...>]

Actually the problems I had were to do with synchronization of the flip 
- the hardware I played with (long time ago now) would delay the flip an 
unpredicable amount of time and therefore you'd get a subsequent flash 
of the clear operation appearing on the screen.

If you've found a way around this or if it's no longer true, then great.

Other things that need to be considered are the tiling issues between 
front and back buffer, currently these are set up as static regions with 
the fence registers, ideally you want the backbuffer tiled and if you're 
flipping that means you need the frontbuffer tiled also.

I've never had the time to look at support for region-based rendering on 
the intel chipsets, but it is certainly worth investigating if someone 
does have the time to put into it.

Keith

Joan Bertran wrote:
> with 6 Megabytes of frame buffer (1536 x 1024 at 32bpp) it takes 10 ms 
> to do swap buffers with a 855GM
> 
> If measuring the windows driver, which supports flip and copy
> with a simple benchmark (one square with texture of 1024x1024) in 
> fullscreen mode 1280x1024 32bpp
> on a "slow memory like 855GM", Flip works much faster (100% .. 200%) 
> than Copy. 160fps vs 55fps
> 
> Do you know how could i get more info about intel graphics to better 
> understand
> this topic (caching?), perhaps to find a solution ?
> 
> How long can it take to update the frame buffer address in a intel card ?
> 
> If in windows the flip works fast perhaps there is something missing.
> Please any benchmark you are willing to run with the windows driver, 
> tell me.
> 
> On 7/24/06, *Ian Romanick* < id...@us... <mailto:id...@us...>> wrote:
> 
>     -----BEGIN PGP SIGNED MESSAGE-----
>     Hash: SHA1
> 
>     Joan Bertran wrote:
> 
>      > Taking into account that intel graphics cards
>      > are shared memory, so they don't have the bandwith
>      > of other gpu's the glSwapBuffers can take a
>      > precious time to finish if pageFlip is not
>      > available.
> 
>     keithw did some experiments in this area some years ago.  It turns that
>     that, due to the way the Intel chipsets do cachine, it's *faster* to do
>     copies than flips.  I seem to recall the break-even point being for full
>     screen apps larger than 1280x800 or something similar.  Even at that
>     point the difference was very small.
> 
>     There are other things that could be done to the Intel driver ( e.g.,
>     modify it to do region-based rendering) that would improve performance
>     by a larger margin in more situations than adding page flipping support.
>     -----BEGIN PGP SIGNATURE-----
>     Version: GnuPG v1.4.2.2 (GNU/Linux)
> 
>     iD8DBQFExM/sX1gOwKyEAw8RApfcAJ9XEVVk2mf/w0iAoL8D/ImvA28frgCfVr1W
>     HXKD6a4ec0/q8qXK2qfq8Og=
>     =m92m
>     -----END PGP SIGNATURE-----
> 
> 
> 
> ------------------------------------------------------------------------
> 
> -------------------------------------------------------------------------
> Take Surveys. Earn Cash. Influence the Future of IT
> Join SourceForge.net's Techsay panel and you'll get the chance to share your
> opinions on IT & business topics through brief surveys -- and earn cash
> http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> Mesa3d-dev mailing list
> Mes...@li...
> https://lists.sourceforge.net/lists/listinfo/mesa3d-dev