possibly.. although sprinkling queue_barrier() calls (which is at
least useful for debugging, although I think I won't use it in the end
after debugging) hasn't found the issue yet. I did at least find an
issue w/ fence handling (I was grabbing the fence # potentially before
the batch was
I think your issue is that you have self-releasing jobs with the
cleanup callback and you automatically lose fences that way, so there
is no way to wait for completion.
Since you have only 1 thread with N jobs at most, I suggest you keep
N+1 fences around (a ring of fences) that you reuse for new
On Mon, Jul 18, 2016 at 4:34 PM, Nicolai Hähnle wrote:
> On 18.07.2016 22:25, Rob Clark wrote:
>>
>> Helper to block until all previous jobs are complete.
>> ---
>> So I think this might end up being useful to me in some cases.. but
>> the implementation only works for a
On 18.07.2016 22:25, Rob Clark wrote:
Helper to block until all previous jobs are complete.
---
So I think this might end up being useful to me in some cases.. but
the implementation only works for a single threaded queue (which is
all I need). I could also just put a helper in my driver code.
Helper to block until all previous jobs are complete.
---
So I think this might end up being useful to me in some cases.. but
the implementation only works for a single threaded queue (which is
all I need). I could also just put a helper in my driver code.
Opinions?