On Thu, Oct 27, 2005 at 07:30:09PM +0200, Denis Oliver Kropp wrote:
> Quoting Ville Syrjälä:
> > Attached are some patches that improve field parity and triple buffering
> > in the matrox driver.
>
> Impressive :)
Thanks :)
>
> > I've moved the buffer flipping completely to the kernel which should make
> > it less probable that the hardware flips the buffers when we're not
> > looking.
I was a bit optimistic here. df_andi revealed some nasty tearing. But I
fixed it! Just apply the following the patch and you should be enjoying
absolutely tear-free df_andi @423fps on your TV! I know I am ;)
Index: matrox_crtc2.c
===================================================================
RCS file: /cvs/directfb/DirectFB/gfxdrivers/matrox/matrox_crtc2.c,v
retrieving revision 1.31
diff -u -r1.31 matrox_crtc2.c
--- matrox_crtc2.c 26 Oct 2005 07:26:16 -0000 1.31
+++ matrox_crtc2.c 28 Oct 2005 05:57:46 -0000
@@ -494,7 +534,7 @@
mcrtc2->regs.c2MISC = 0;
/* c2vlinecomp */
- mcrtc2->regs.c2MISC |= (vdisplay + 1) << 16;
+ mcrtc2->regs.c2MISC |= vdisplay << 16;
}
/* c2bpp15halpha */
Now I just wish the embedded-2 mga drm would compile with recent kernels
so I could really stress test this thing with df_gears @>1000fps...
> I'm sure there will be comparable ioctls in other drivers soon,
> or even a new standardized flip/wait ioctl that we've talked about.
>
> This is a BIG step towards it :)
Right. It would be nice to hook in the vsync part next and make double
buffering more useful with a new IDirectFBSurface::WaitForFlip() method.
> > + FBDev *dfb_fbdev = dfb_system_data();
> > + struct matrox_address address;
> > +
> > + bes_calc_regs( mdrv, mbes, &mbes->config, surface );
> > +
> > + address.device = MATROX_ADDRESS_DEVICE_BES;
> > + address.field = -1;
> > + address.count = mbes->last_count + 1;
> > + address.offset[0] = mbes->regs.besA1ORG;
> > + address.offset[1] = mbes->regs.besA2ORG;
> > + address.offset[2] = mbes->regs.besA1CORG;
> > + address.offset[3] = mbes->regs.besA2CORG;
> > + address.offset[4] = mbes->regs.besA1C3ORG;
> > + address.offset[5] = mbes->regs.besA2C3ORG;
> > +
> > + if (ioctl( dfb_fbdev->fd, MATROXFB_ADDRESS, &address ))
> > + D_ERROR( "DirectFB/Matrox/BES Address ioctl() failed!\n" );
> > +
> > + dfb_surface_flip_buffers( surface, address.count ==
> > mbes->last_count );
> > +
> > + mbes->last_count = address.count;
>
> YES, you're using the second argument to dfb_surface_flip_buffers()!
>
> I've never seen it in action after adding it :)
I played with it at one point but then I was doing the flipping in
user-space and then going to the kernel to ask for the vsync count.
Obviosly the user->kernel->user transition took too long and I could not
get reliable results.
> Time to hook the primary layer and implement Flip() with that feature on
> CRTC1 ;)
I hate VGA register. That's my excuse for not doing it :)
--
Ville Syrjälä
[EMAIL PROTECTED]
http://www.sci.fi/~syrjala/
_______________________________________________
directfb-dev mailing list
[email protected]
http://mail.directfb.org/cgi-bin/mailman/listinfo/directfb-dev
