[Bug 30007] Regression in r300g

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=30007

--- Comment #25 from Marek Olšák <maraeo at gmail.com> 2010-09-12 21:37:52 PDT ---
Fredrik, thanks for the feedback. I've pushed your patch to the piglit
repository.

(In reply to comment #23)
> The r500 should be able to execute this, but there are two things that are
> currently preventing correct executions

However the reporter has r400, so the hardware just cannot do it. I am nearly
sure fglrx would fall back to software, which is not particularly useful for
us. Specifying kernelSize at compile time should be enough for the compiler to
unroll the loop.

(In reply to comment #24)
> Oops, I submitted before I was done.  This shader would be better, because the
> loop could be unrolled even if kernelSize is different in each execution:

Yeah, the shader could work provided the instruction scheduler generates no
more than 4 TEX indirections. I'd personally lean towards making kernelSize a
compile-time constant. So instead of calling glUniform1f(kernelSize, i), let's
call glUseProgram(shader[i]);

Fredrik, please, would it be hard to implement this in kwin?

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