[Mesa-dev] [PATCH] i965: Don't tell the hardware about our UAV access.

[Mark Janes]

Emil Velikov <emil.l.velikov at gmail.com> writes:

> Hi all,
>
> On 10 September 2015 at 00:30, Mark Janes <mark.a.janes at intel.com> wrote:
>> Mark Janes <mark.a.janes at intel.com> writes:
>>
>>> Francisco Jerez <currojerez at riseup.net> writes:
>>>
>>>> Mark Janes <mark.a.janes at intel.com> writes:
>>>>
>>>>> When I tested this, I saw an intermittent BSW gpu hang.  I haven't been
>>>>> able to confirm that it is due to the host-mem-barrier test.
>>>>>
>>>> It would probably be useful to know if the hang is due to any of the
>>>> image load/store tests or if it's something unrelated.
>>>
>>> Yes, you are right.  I will take some time tomorrow to catch the
>>> specific test.  Given the low rate of gpu hangs on BSW lately, I expect
>>> it will be due to image load/store tests.  However I need to confirm it.
>>
>> I spent time attempting to reproduce hangs with this patch on BSW today.
>> I forced 3 hangs, but none of them could be attributed to image
>> load/store tests.
>>
> With the above said, should we consider this as inconclusive fix ? If
> so, what is the plan going forward - is this going to the bin or we'll
> beat it into shape and merge it ?
>

I think the fix correctly addresses gpu hangs related to the
host-mem-barrier test.

My initial report wrongly attributed gpu hangs to host-mem-barrier.
Subsequent testing showed that hangs were caused by other tests.  I
provided feedback with partial information because of the time consuming
nature of root-causing gpu hangs.

Curro felt confident in this fix, so I took the time to verify which
tests triggered the hangs.