[Beignet] build problems: ASSERTION FAILED: Fail to create the gen context

Tue Jul 8 04:05:18 PDT 2014

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA256

Hi,

The master and Release_v0.9.x branch compiles now without problems
even when the lspci Intel VGA does not show up. I am able to create an
OpenCL beignet context, even when the Intel VGA is not listed by
lspci. It seems to catch the device directly from /dev/dri/card0. Here
you see the printed output:

https://gist.github.com/cguentherTUChemnitz/31546f98706432a366ac

Thank you for your effort.
best,
CG

Am 08.07.2014 10:26, schrieb Zhigang Gong:
> On Tue, Jul 08, 2014 at 05:04:22PM +0800, Yichao Yu wrote:
>> Hi,
>> 
>> I am on the master branch and it has always been fine (and
>> nothing seems to be broken by the new commits today). I've seen
>> ur email about the branching policy so I guess the commit you
>> mentioned is also included in master?
> Right. The master branch will always have all the bug fixing
> patches.
> 
>> 
>> As for the AUR packaging issue from someone else I mentioned in
>> the last email, it turns out that the Intel GPU does not show up
>> in lspci for certain BIOS settings although lshw shows both of
>> his NVIDIA and Intel card correctly. I'm not sure what his
>> problem is and whether it is still there with the new runtime
>> detecting mechanism. Do you have any idea about it? (See the AUR
>> comment[1] for more info about his HW.)
>> 
>> 
>> 
>> P.S. He said he had problem posting on this list, any idea about
>> that?
>> 
>> [1] https://aur.archlinux.org/packages/beignet-git/
> 
> His email may be blocked by the list. Anyway, the current master
> and Release_v0.9.x branch both don't need to execute lspci any
> more.
> 
> CC to cguenther, could you confirm that the problem has been solved
> on both master and Release_v0.9.x branch? Thanks.
>> 
>> 
>> On Tue, Jul 8, 2014 at 3:54 PM, Zhigang Gong
>> <zhigang.gong at gmail.com> wrote:
>>> Hi Igor and Yichao,
>>> 
>>> Now you could try branch Release_v0.9.x again which already
>>> fixed this issue completely. We no longer need lspci from now
>>> on. And if that version is good from package maintainer view,
>>> we will consider to make another fix release soon in next week.
>>> Thanks.
>>> 
>>> On Fri, Jul 4, 2014 at 7:53 PM, Igor Gnatenko 
>>> <i.gnatenko.brain at gmail.com> wrote:
>>>> Hi, On Fri, 2014-07-04 at 14:09 +0800, Zhigang Gong wrote:
>>>>> On Fri, Jul 04, 2014 at 07:53:42AM +0200, Adel Gadllah
>>>>> wrote:
>>>>>> Zhigang Gong:
>>>>>>> On Fri, Jul 04, 2014 at 01:33:14AM +0400, Igor Gnatenko
>>>>>>> wrote:
>>>>>>>> Hi,
>>>>>>>> 
>>>>>>>> I've long time no tested beignet builds. Now I've
>>>>>>>> updated it to 0.9.0 in Fedora rawhide[0].
>>>>>>>> 
>>>>>>>> And it failed to build.
>>>>>>>> 
>>>>>>>> Generating ../../utests/../kernels/compiler_ceil.bin 
>>>>>>>> Unknown opt 
>>>>>>>> /builddir/build/BUILD/Beignet-0.9.0-Source/build/backend/src/gbe_bin_generater:
>>>>>>>> option requires an argument -- 't'
>>>>>>>> 
>>>>>>>> ASSERTION FAILED: Fail to create the gen context at 
>>>>>>>> file
>>>>>>>> /builddir/build/BUILD/Beignet-0.9.0-Source/backend/src/backend/gen_program.cpp,
>>>>>>>> function virtual gbe::Kernel*
>>>>>>>> gbe::GenProgram::compileKernel(const gbe::ir::Unit&,
>>>>>>>> const string&, bool), line 167
>>>>>>>> 
>>>>>>>> And etc. I'm attaching FULL build log.
>>>>>>>> 
>>>>>>>> [0]http://pkgs.fedoraproject.org/cgit/beignet.git/commit/?id=0c45e57cbecf72e1a5f243bbc0f375e3f19a2b03
>>>>>>>>
>>>>>>>> 
- --
>>>>>>>> -Igor Gnatenko
>>>>>>> 
>>>>>>>> /builddir/build/BUILD/Beignet-0.9.0-Source/src/GetGenID.sh:
>>>>>>>> line 3: lspci: command not found -- Platform Gen PCI
>>>>>>>> id is
>>>>>>> The root cause is that now beignet will build internal
>>>>>>> binary kernels according current build host's Gen VGA
>>>>>>> PCI id.
>>>>>> This can't work for distributions that want to ship
>>>>>> beignet. The system it gets build on is not the one it
>>>>>> runs on. Also distribution build systems tend to be
>>>>>> virtual machines that do not have any real (or intel) vga
>>>>>> hardware.
>>>>> You are right, this is a problem for distributions. I just
>>>>> realized this issue when I replied to Igor. We will discuss
>>>>> how to make it friendly for package maintainer soon.
>>>> Why do we need find out VGA PCI id?
>>>>> Thanks for pointing this out.
>>>> 
>>>> -- -Igor Gnatenko
>>>> 
>>>> _______________________________________________ Beignet
>>>> mailing list Beignet at lists.freedesktop.org 
>>>> http://lists.freedesktop.org/mailman/listinfo/beignet
>>>> 
>> _______________________________________________ Beignet mailing
>> list Beignet at lists.freedesktop.org 
>> http://lists.freedesktop.org/mailman/listinfo/beignet
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/
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=23xk
-----END PGP SIGNATURE-----