[Mesa-dev] New GBM backend for dEQP

[Tapani Pälli]

On 12/20/2016 09:10 PM, Chad Versace wrote:
> On Mon 19 Dec 2016, Tapani Pälli wrote:
>>
>> On 12/17/2016 03:58 AM, Chad Versace wrote:
>>> Happy Christmas to everyone who's busy squashing dEQP bugs.
>>>
>>> I wrote a new GBM backend for dEQP. I even submitted it to dEQP's
>>> upstream Gerrit.  Pyry, dEQP's maintainer, told me over beer earlier
>>> this year that he would accept it if I submitted it, and if it wasn't
>>> too crazy. So, maybe it'll be upstream soon.
>>>
>>> If you want to try it out, you can either fetch the patch from Gerrit:
>>>      $ git fetch https://android.googlesource.com/platform/external/deqp refs/changes/43/315743/1
>>>
>>> View it on Gerrit:
>>>      https://android-review.googlesource.com/#/c/315743/
>>>
>>> Fetch from personal my work-in-progress branch:
>>>      $ git fetch git://git.kiwitree.net/~chadv/deqp refs/heads/wip/gbm
>>>
>>> View it on my cgit:
>>>      http://git.kiwitree.net/cgit/~chadv/deqp/log/?h=wip/gbm
>>>
>>> GBM today does not support pixmaps nor pbuffers (eglCreatePixmapSurface
>>> and eglCreatePbufferSurface), so the dEQP test coverage with GBM does
>>> not have parity with X11. But, on the other hand, you get to run dEQP
>>> without the headache of X11.
>>>
>>> There's probably bugs. No surprises there.
>> Branch did not work 'out of the box' for me:
>>
>> "No rule to make target 'framework/qphelper/.git/index', needed by
>> 'framework/qphelper/qpReleaseInfo.inl'.  Stop."
>>
>> (attached patch makes it work for me)
> Strange. This may be related to a separate fix I submitted to dEQP
> upstream:
>
>      Subject: Fix build when '.git' is a gitfile
>      https://android-review.googlesource.com/#/c/315234/

Yes, that is related. In my case ${DE_GIT_DIR} variable contains just 
'git' so I needed to concat ${CMAKE_SOURCE_DIR} with that so that 
dependency is correctly found, otherwise it thinks that 'git' is under 
qphelper path.

>> One issue is that real users will use X11, Wayland or Android. Would be cool
>> to have a 'switch' to toggle CI to use a particular backend so that most of
>> the time we would run against gbm but then sometimes check that X11 still
>> works etc.
> Yes. I expect full test runs to be faster with GBM than with X11. If
> that's true, then CI should default to running dEQP with GBM. And CI
> should occasionally do a run with X11 to ensure there's no regressions,
> and to also run any pbuffer and pixmap tests that get skipped on the
> GBM run.