[Bug 90749] [BDW Bisected]dEQP-GLES3.functional.rasterization.fbo.rbo_multisample_max.primitives.lines_wide fails

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=90749

--- Comment #10 from Iago Toral <itoral at igalia.com> ---
(In reply to Ian Romanick from comment #9)
> (In reply to Iago Toral from comment #2)
> > From the OpenGL ES 3.1 spec:
> > 
> > "13.4.2.1 Wide Lines
> > The actual width of lines is determined by rounding the supplied width to
> > the nearest integer, then clamping it to the implementation-dependent
> > maximum line width."
> > 
> > This is what the bad commit implements. In section "13.4.4 Line Multisample
> > Rasterization", the spec talks about multisampled scenarios in particular
> > (which is what the failed test uses), but nothing there seems to contradict
> > the above statement.
> > 
> > More over, the test that fails tests multiple line widths, but only fails
> > for the one we advertise as maximum (7.375). If I change that value to
> > 7.250, 7.125 or 7.000, the dEQP test passes just fine. This, as per ther
> > statement in section 13.4.2.1 that I pasted above, is unexpected, since all
> > these line widths are supposed to be rounded to 7.0 and produce the exact
> > same rendering. The fact that dEQP expects a different rendering for 7.375
> > points to the test being bogus. In fact, not rounding the line-width if
> > multisampling is enabled seems to fix the test, but as I say, that seems to
> > be against the spec, so my conclusion is that this dEQP test is incorrect.
> 
> I think you could read the spec either way.  Section 13.4.2.1 talks about
> rendering aliased wide lines, and section 13.4.4 talks about rendering
> general multisampled lines.  Look at similar text in the desktop OpenGL 4.4
> spec:
> 
>     The actual width of non-antialiased lines is determined by rounding the
>     supplied width to the nearest integer, then clamping it to the
>     implementation-dependent maximum non-antialiased line width.
> 
> When ES removed antialiased lines, they removed "non-antialised" but
> probably should not have.  I believe removing the quantization for
> multisample and antialiased lines is correct.

Oh, good catch! I just tested this and works without causing any regressions in
any of the dEQP wide line rasterization tests. I sent this patch for review:
http://lists.freedesktop.org/archives/mesa-dev/2015-June/086031.html

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/intel-3d-bugs/attachments/20150609/d46b6d5c/attachment.html>