On 23.06.2016 00:35, Marek Olšák wrote:
On Thu, Jun 23, 2016 at 12:29 AM, Marek Olšák <[email protected]> wrote:
On Tue, Jun 21, 2016 at 4:47 PM, Nicolai Hähnle <[email protected]> wrote:
On 21.06.2016 14:17, Marek Olšák wrote:
Hi,
This improves u_queue to be more usable in more cases.
With this, the size of the queue (maximum number of jobs waiting) is
configurable as well as the number of threads executing the jobs. The
semaphores are ditched in favor of simpler condvars, and multiple waiters on
fences are allowed as well.
This is a prerequisite for a later series that will add multithreaded
shader compilation into radeonsi.
I like the overall cleanup, but I'd appreciate it if you could rearrange
patch #3 and #5 a little to avoid the temporary deadlock. No need to risk
someone running into that while bisecting.
I think it would suffice to add queue->lock locking around the
queue->kill_threads already in patch #3, but tbh I'd also be fine with the
slightly less-work option of squashing the two patches together.
To be honest, I'm not able to see the deadlock now. Maybe I should
remove the comment.
Strictly speaking, patch 3 doesn't change anything with respect to
locking. Either the deadlock is not there or it has always been there.
If I fix up calloc, free, and snprintf, does it have your Rb?
Yes, go ahead.
Nicolai
Marek
_______________________________________________
mesa-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
