[Bug 31830] [bisected] broken shadows in Unigine Sanctuary and Lightsmark

bugzilla-daemon at freedesktop.org bugzilla-daemon at freedesktop.org
Thu Jan 27 01:18:41 PST 2011


https://bugs.freedesktop.org/show_bug.cgi?id=31830

--- Comment #11 from Pavel Ondračka <drakkk at centrum.cz> 2011-01-27 01:18:41 PST ---
(In reply to comment #10)
> (In reply to comment #3)
> > What penumbra quality number are you using in Lightsmark? Are you seeing any
> > error messages like this:
> > 
> > r300: ERROR: FS input generic 18 unassigned, not enough hardware slots.
> > r300: ERROR: FS input generic 19 unassigned, not enough hardware slots.
> > r300: ERROR: FS input generic 20 unassigned, not enough hardware slots.
> > r300: ERROR: FS input WPOS unassigned, not enough hardware slots.
> > 
> 
> In Lightsmark, I've been able to identify the failing shader and a few
> instruction transformations that change the way the shadows are rendered.  I've
> spent a lot of time looking at this, and all the transformations in question
> appear to be correct.  It's possible that above errors are the reason why the
> shadows are misrendering.  The last error about WPOS also appears in sanctuary. 
> 
> My hypothesis is that with these kinds of errors, the behavior of a shader is
> undefined.  This means that prior to the commit in question, the scenes were by
> luck being rendered correctly, or close enough so that it wasn't noticeable. 
> Then, the modifications made to the shader by the reg_rename pass changed the
> state of the GPU in such a way that the misrendering became more obvious.
> 
> Maybe Marek can comment on whether or not this is a plausible hypothesis.

Does this also explain the slowdown mentioned in comment 2? Because it is in
all scenes, not only the "penumbra shadows" one and it is really weird that
RADEON_DEBUG=noopt actually makes things faster.

-- 
Configure bugmail: https://bugs.freedesktop.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.


More information about the dri-devel mailing list