[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 #1 from Roland Scheidegger <sroland at vmware.com> ---
I can't say I'm terribly surprised this commit broke it. That it might not work
on BE certainly came to my mind when I did this, despite my efforts... I
suspect the affected format didn't work with SoA fetch before, but this wasn't
much of an issue because it was commonly only used with vertex fetch, which
used the AoS fetch. Albeit just before this commit, I did some changes to SoA
fetch which might also potentially break BE (but you'd only see it with texture
format tests and such before this commit).
It is also possible the vector_justify parameter for lp_build_gather() isn't
quite set correctly somewhere. Fixing llvmpipe on big-endian arch was done by
Richard Sandiford <r.sandiford at uk.ibm.com> a couple years ago (introducing that
vector_justify) and it worked as long as noone touched the BE affecting code I
suppose...
I don't think there's too many developers with BE boxes, unfortunately (well
there aren't many BE boxes in the first place nowadays...). Someone would have
to look into it - glxgears should be only using R32G32B32F vertex format, with
some luck it's a special case (due to it being native 3 channel, npot) which
can be fixed rather easily, but there might be more fundamental issues with SoA
fetch on BE.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20170407/543f15ad/attachment-0001.html>