[Mesa-dev] r600g: various fixes to get r600_clear and depthtesting working

[Jerome Glisse]

On Tue, May 25, 2010 at 03:43:56PM +0200, Bas Nieuwenhuizen wrote:
> On Tue, 2010-05-25 at 09:51 +0200, Jerome Glisse wrote:
> > On Tue, May 25, 2010 at 12:35:45AM +0200, Bas Nieuwenhuizen wrote:
> > > Hello,
> > > 
> > > I wrote two patches that fix various things to get r600g running on my
> > > computer and to get r600_clear and depthtesting working.
> > > 
> > > I wrote them starting from Marek's r600g branch.
> > > 
> > > Feel free to criticize as this is my first contribution to a large
> > > project. 
> > > 
> > > - Bas Nieuwenhuizen
> > > 
> > > patches:
> > > http://www.basnieuwenhuizen.nl/0001-r600g-various-fixes.patch
> > 
> > I don't think change to r600_texture_destroy is right, why did you
> > remove the unreference there ?
> 
> r600_texture_destroy was called from pipe_resource_reference because the
> reference count was 0. And then r600_texture_destroy calls
> pipe_resource_reference again with a reference count of 0 which triggers
> an assert in pipe_reference.
> 
> > > http://www.basnieuwenhuizen.nl/0002-r600g-fixed-some-corruption-added-code-for-depthtest.patch
> > 
> > Is the wait idle needed on your hw ? If so can you tell me which gpu you
> > are using.
> 
> It is needed on my hw, which is a rv730, HD4670.
> 
> The corruption that it fixed was that some blocks of the background were
> drawn in the color of the other primitives that are drawn and some
> flicker. Some very short flat shaded programs don't trigger it.
> 
> I got the idea from the classic driver where r700TryDrawPrims calls
> r700WaitForIdleClean after the draw.
> 
> - Bas Nieuwenhuizen
> 

I will check with my HD4670, it's strange as i did quite intensive
testing of the winsys through radeondb. Maybe you are affected by
the barrier issue (got a fix in my tree for that).

I am debugging my shader rework, as soon as it's working again i
will push

Cheers,
Jerome