[Beignet] Unit test fail (and regression?) on i7-4702HQ

[Yichao Yu]

On Mon, Oct 20, 2014 at 9:25 PM, Zhigang Gong <zhigang.gong at gmail.com> wrote:
> It's weird that only that case failed. Did you check whether a GPU
> hang occurred?
> And did you try to reboot your system to run the unit tests again?
> Sometimes, if the GPU hang once, it may not be able to fully recover to normal
> state without a reboot. Just in case.

Hmmm. I guess that is probably what happens my program seems to hang
the GPU for a while (screen freeze) before it through that error and
I've seen some random small white patches on the screen a few times
after that. I will reboot soon and report back...

>
> On Tue, Oct 21, 2014 at 1:03 AM, Yichao Yu <yyc1992 at gmail.com> wrote:
>> On Mon, Oct 20, 2014 at 1:01 PM, Yichao Yu <yyc1992 at gmail.com> wrote:
>>> On Mon, Oct 20, 2014 at 11:14 AM, Zhigang Gong <zhigang.gong at gmail.com> wrote:
>>>> Could you try to run all unit test cases with 0.9.3 version and share
>>>> the full output here?
>>>> Thanks.
>>>
>>> Here it is[1]
>>
>> The output includes both stdout and stderr. The failing test is here[2]
>>
>> [2] https://gist.github.com/yuyichao/4ae6cb022bd0870a14c7#file-utests-log-L579
>>
>>>
>>> Configured with
>>>
>>> cmake .. -DCMAKE_BUILD_TYPE=RelWithDebInfo
>>> -DPYTHON_EXECUTABLE=/usr/bin/python2 -DCMAKE_INSTALL_PREFIX=/usr
>>>
>>> Compiled with gcc 4.9.1
>>>
>>> [1] https://gist.github.com/yuyichao/4ae6cb022bd0870a14c7
>>>
>>>>
>>>> On Mon, Oct 20, 2014 at 11:10 PM, Yichao Yu <yyc1992 at gmail.com> wrote:
>>>>> On Mon, Oct 20, 2014 at 8:05 AM, Zhigang Gong <zhigang.gong at gmail.com> wrote:
>>>>>> I can't reproduce it with current git master git-8fe98a on my MacBook
>>>>>> Air i5-4260U.
>>>>>> Could you help to bisect this utest regression on your system?
>>>>>> Thanks.
>>>>>
>>>>> I haven't tryed bisecting with my own program (which has a good
>>>>> commit) but it seems that that test fails here with 0.9.3 (and the
>>>>> corresponding commit in master branch) as well (and 0.9.2 doesn't want
>>>>> to compile with my llvm 3.5.0)
>>>>>
>>>>> Is there any other info I can provide to figure out the problem?
>>>>>
>>>>> Yichao Yu
>>>>>
>>>>>>
>>>>>> On Mon, Oct 20, 2014 at 7:48 PM, Yichao Yu <yyc1992 at gmail.com> wrote:
>>>>>>> Hi,
>>>>>>>
>>>>>>> With current master, the unit test compiler_atomic_functions fails with output
>>>>>>>
>>>>>>> compiler_atomic_functions()    [FAILED]
>>>>>>>     Error: ((int *)buf_data[0])[i] == cpu_dst[i]
>>>>>>>   at file /home/yuyichao/projects/archcn/repo/beignet-git/src/beignet/utests/compiler_atomic_functions.cpp,
>>>>>>> function compiler_atomic_functions, line 92
>>>>>>>
>>>>>>> on my laptop with i7-4702HQ.
>>>>>>>
>>>>>>> I was running the unit tests because I've noticed that some of my code
>>>>>>> which runs fine with beignet a few commits ago fails with the current
>>>>>>> master. The code runs with beignet 0.9.3 (and IIRC some later commit
>>>>>>> at the master branch as well) but it fails on the current master with
>>>>>>> the following error after freezing the screen for a few seconds.
>>>>>>>
>>>>>>> python: /home/yuyichao/projects/archcn/repo/beignet-git/src/beignet/src/intel/intel_gpgpu.c:703:
>>>>>>> intel_gpgpu_check_binded_buf_address: Assertion
>>>>>>> `gpgpu->binded_buf[i]->offset != 0' failed.
>>>>>>>
>>>>>>> Kernel version 3.17.0
>>>>>>> Kernel patch for haswell applied.
>>>>>>> cmd_parser turned off just in case (doesn't work when enabled wither)
>>>>>>>
>>>>>>> Yichao Yu
>>>>>>> _______________________________________________
>>>>>>> Beignet mailing list
>>>>>>> Beignet at lists.freedesktop.org
>>>>>>> http://lists.freedesktop.org/mailman/listinfo/beignet