[Mesa-dev] [Bug 100613] Regression in Mesa 17 on s390x (zSystems)

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=100613

--- Comment #45 from Roland Scheidegger <sroland at vmware.com> ---
(In reply to Ben Crocker from comment #44)
> By the way, draw-vertices-2101010 is failing across all architectures:
> X86 and PPC64LE as well as PPC64/S390x.
Yes, but this isn't related to swizzling (at least on x86, unless ppc64 has
additional errors), but is rather due to GL weirdness with converting signed
numbers. Pre-GL 4.2 there's 2 different formulas depending on if it's used as a
vertex format or a pixel format, and the test expects the result of the formula
used for vertex formats. GL 4.2 and up ditched this formula entirely
(obviously, two formulas don't make all that much sense if pixel and vertex
data is all in the same buffers anyway...), and that's the one we always use,
no matter the GL version.
The difference in the formulas usually isn't all that large, but with just 2
bits the error is huge: the four values you can represent are [-1.0, -0.333,
0.333, 1.0] for the old formula and [-1.0, -1.0, 0.0, 1.0] for the new one,
hence the failure (Expected: ... 0.333000, Observed: ... 0.000000).

Note that at least some hw/drivers will ignore these old GL rules too (after
all, GL 4.2 is supposed to be backwards compatible, and clearly with the
different formula gone it can't quite be), therefore we don't consider this a
driver bug, rather a test bug:
https://bugs.freedesktop.org/show_bug.cgi?id=59150

-- 
You are receiving this mail because:
You are the assignee for the bug.
You are the QA Contact for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20170901/cad343da/attachment-0001.html>