[Dri-devel] Re: [PATCH] Savage streams re-work

[Alex D. <ag...@ya...>]

--- Felix Kühling <fx...@gm...> wrote:
> On Fri, 19 Mar 2004 12:51:14 -0800 (PST)
> Alex Deucher <ag...@ya...> wrote:
> 
> > I've re-worked the non-Xv streams code and a few other things.  I
> added
> > a new streams reset function that seems to fix the occasional
> console
> > corruption I got on my IX.  I enabled the COB on savage4 and newer
> > chips and added an option to disable it.  Tim's driver had it
> disabled
> > and had a note that there was a hardware bug in the COB on savage4
> and
> > prosavage; S3 enabled the COB.  either way seems to work fine for
> me. 
> > With the COB enabled my frame rates with glxgears improved quite a
> bit:
> > 
> > depth  COB   No-COB
> > 16     ~560  ~450
> > 24     ~270  ~240
> 
> Yes, I'm testing your patch on the ProSavageDDR and now I get about
> the
> same glxgears fps as with DMA command transfer. So BCI+COB seems to
> work
> pretty efficiently. I'm going to start working on DMA vertex
> transfers
> this weekend. I'm curious how that will affect performance.
> 
> There's still one problem. The hardware cursor is corrupted when 3D
> apps
> are active. It looks like the COB is scribbling over the memory that
> stores the hardware cursor bitmasks. With glxgears the pattern
> changes
> very frequently, with flightgear it takes between half a second and a
> second between "updates". ;-)

UGH... I looked into the cursor/cob problem. apparently different
versions of the savage 4 and newer savages have a different meaning for
the cobIndex bitfield of register 0x48C14 and I'm not sure which are
affected.  The old code also didn't align the cob properly.  Try the
new patch I've attached.   If you are still having problems with the
cursor, try setting the cobIndex to 2 rather than 7 for your chip.

> 
> > 
> > I also converted the BCI Xv stuff into an option, BCIforXv, but it
> > still leaves artifacts even with the new streams setup and is
> disabled
> > by default.  I still haven't tracked down how S3 does interpolation
> in
> > their driver with the old streams engine.
> 
> This one works without problems. I don't see any difference with or
> without BCI. No cursor corruption either. But I think this line is
> wrong: ;-)

Do you see any corruption when you move the output window around?  It
leaves artifacts on savage4.  If not, perhaps we can enable the BCI for
Xv on prosavages and twisters.

> 
>         xf86DrvMsg(pScrn->scrnIndex, X_CONFIG,
>                    "Option: %s use of the BCI for
> Xv\n",(psav->BCIforXv?"Disable":"Enable"));
> 
> It says BCI for Xv is disabled when it is actually enabled and vice
> versa.

Whoops!  cut and paste error. fixed now. new patch here:
http://www.botchco.com/alex/savage/streams_cleanup_big-v2.diff

Alex

> 
> > 
> > Please test and let me know how it goes.  If all is well I'll
> commit
> > it.
> > You can also download the patch here:
> > http://www.botchco.com/alex/savage/streams_cleanup_big.diff
> > 
> > Alex
> 
> Felix


__________________________________
Do you Yahoo!?
Yahoo! Mail - More reliable, more storage, less spam
http://mail.yahoo.com