Re: [Dri-devel] Pageflipping for Rage 128

[Ian R. <id...@us...>]

Eric Anholt wrote:
> The last few days I've been looking at the radeon pageflipping code and
> copying it to Rage 128.  It's been pretty easy due to the disabled
> fullscreen pageflip code already being there.  I've got a patch now that
> seems to work for the trivial stuff I've tried, but I would like someone
> else to look at it since I'm clueless, particularly in the h/w driver
> (is that the current official name of the *_dri.so?).  The current patch

I've always called it the client-side driver.  Afterall, there is 
another part to the 3D driver that lives in the kernel. :)  I don't 
think it matters too much.  Client-side driver, h/w driver, 3D 
driver...people will know what you mean.

> I have is here:
> http://people.freebsd.org/~anholt/dri/files/r128-pageflip-4.diff
> 
> On my Rage 128 Mobility M4 MF:
> glxgears at 16 bit went from 735->733 fps (too much deviation to see a difference)
> glxgears at 24 bit went from 458->661 fps

I haven't had a chance to look at it yet, but if it works, then it's 
probably okay.  You might wait to commit it until after I merge the 
texmem branch into the trunk.  I refactored some code in this area in 
the MGA, R200, and Radeon drivers.  You might want to use that for the 
R128 also.

> Among the things I noticed in trunk code while working on this:
> 
> There are clipping issues on R128 with two 3d clients.  To see it, drag
> the corner of a glxgears over another running glxgears.

I don't see that problem on R200, so it's probably R128 specific.  Have 
you  had a chance to look into it more?

> Texture problems are visible in mesa fire demo, also in various ways in
> tuxracer.  Here are links to hardware and software
> (LIBGL_ALWAYS_INDIRECT=1) screenshots of the fire demo:
> http://people.freebsd.org/~anholt/dri/shots/r128fire.png
> http://people.freebsd.org/~anholt/dri/shots/swfire.png

I see the same problem on R200.  It looks like there's too much fog 
happening.

> There is a block of garbage that looks 64x64 in tuxracer where the
> cursor used to be before it got turned off.

Hmm...sounds like the cursor isn't really getting turned off.  I seem to 
remember a problem like this once a long time ago w/a different driver.

> In radeon_dri.c:RADEONDisablePageFlip() it talks about "bumping the
> window counters".  Are they actually being bumped?  I see that
> radeon_lock.c is watching for the counters to get bumped.

I think that comment is incorrect.  The 2D driver communicates to the 3D 
driver that page-flipping is disabled by setting pfAllowPageFlip to 0. 
The 2D driver internally uses "window counting" to know when to set 
pfAllowPageFlip to 0 or 1.