<html>
<head>
<meta http-equiv="Content-Type" content="text/html;
charset=windows-1252">
</head>
<body text="#000000" bgcolor="#FFFFFF">
<div class="moz-cite-prefix">Hey Christian,</div>
<div class="moz-cite-prefix"><br>
</div>
<div class="moz-cite-prefix">The problem boils down to the fact that
we don't immediately create dma fences when calling
vkQueueSubmit().</div>
<div class="moz-cite-prefix">This is delayed to a thread.</div>
<div class="moz-cite-prefix"><br>
</div>
<div class="moz-cite-prefix">From a single application thread, you
can QueueSubmit() to 2 queues from 2 different devices.</div>
<div class="moz-cite-prefix">Each QueueSubmit to one queue has a
dependency on the previous QueueSubmit on the other queue through
an exported/imported semaphore.</div>
<div class="moz-cite-prefix"><br>
</div>
<div class="moz-cite-prefix">From the API point of view the state of
the semaphore should be changed after each QueueSubmit().</div>
<div class="moz-cite-prefix">The problem is that it's not because of
the thread and because you might have those 2 submission threads
tied to different VkDevice/VkInstance or even different
applications (synchronizing themselves outside the vulkan API).</div>
<div class="moz-cite-prefix"><br>
</div>
<div class="moz-cite-prefix">Hope that makes sense.</div>
<div class="moz-cite-prefix">It's not really easy to explain by
mail, the best explanation is probably reading the test : <a
href="https://gitlab.freedesktop.org/mesa/crucible/blob/master/src/tests/func/sync/semaphore-fd.c#L788">https://gitlab.freedesktop.org/mesa/crucible/blob/master/src/tests/func/sync/semaphore-fd.c#L788</a></div>
<div class="moz-cite-prefix"><br>
</div>
<div class="moz-cite-prefix">Like David mentioned you're not running
into that issue right now, because you only dispatch to the thread
under specific conditions.</div>
<div class="moz-cite-prefix">But I could build a case to force that
and likely run into the same issue.<br>
</div>
<div class="moz-cite-prefix"><br>
</div>
<div class="moz-cite-prefix">-Lionel<br>
</div>
<div class="moz-cite-prefix"><br>
</div>
<div class="moz-cite-prefix">On 02/08/2019 07:33, Koenig, Christian
wrote:<br>
</div>
<blockquote type="cite"
cite="mid:970c65d1-ad63-4492-8ae2-96b603b2c7c8@email.android.com">
<meta http-equiv="Content-Type" content="text/html;
charset=windows-1252">
<meta name="Generator" content="Microsoft Exchange Server">
<!-- converted from text -->
<style><!-- .EmailQuote { margin-left: 1pt; padding-left: 4pt; border-left: #800000 2px solid; } --></style>
<div>
<div dir="auto">Hi Lionel,
<div dir="auto"><br>
</div>
<div dir="auto">Well could you describe once more what the
problem is?</div>
<div dir="auto"><br>
</div>
<div dir="auto">Cause I don't fully understand why a rather
normal tandem submission with two semaphores should fail in
any way.</div>
<div dir="auto"><br>
</div>
<div dir="auto">Regards,</div>
<div dir="auto">Christian.</div>
</div>
<div class="x_gmail_extra"><br>
<div class="x_gmail_quote">Am 02.08.2019 06:28 schrieb Lionel
Landwerlin <a class="moz-txt-link-rfc2396E" href="mailto:lionel.g.landwerlin@intel.com"><lionel.g.landwerlin@intel.com></a>:<br
type="attribution">
</div>
</div>
</div>
<font size="2"><span style="font-size:11pt;">
<div class="PlainText">There aren't CTS tests covering the
issue I was mentioning.<br>
But we could add them.<br>
<br>
I don't have all the details regarding your implementation
but even with <br>
the "semaphore thread", I could see it running into the same
issues.<br>
What if a mix of binary & timeline semaphores are handed
to vkQueueSubmit()?<br>
<br>
For example with queueA & queueB from 2 different
VkDevice :<br>
vkQueueSubmit(queueA, signal semA);<br>
vkQueueSubmit(queueA, wait on [semA, timelineSemB]);
with <br>
timelineSemB triggering a wait before signal.<br>
vkQueueSubmit(queueB, signal semA);<br>
<br>
<br>
-Lionel<br>
<br>
On 02/08/2019 06:18, Zhou, David(ChunMing) wrote:<br>
> Hi Lionel,<br>
><br>
> By the Queue thread is a heavy thread, which is always
resident in driver during application running, our guys
don't like that. So we switch to Semaphore Thread, only when
waitBeforeSignal of timeline happens, we spawn a thread to
handle that wait. So we don't have your this issue.<br>
> By the way, I already pass all your CTS cases for now.
I suggest you to switch to Semaphore Thread instead of Queue
Thread as well. It works very well.<br>
><br>
> -David<br>
><br>
> -----Original Message-----<br>
> From: Lionel Landwerlin
<a class="moz-txt-link-rfc2396E" href="mailto:lionel.g.landwerlin@intel.com"><lionel.g.landwerlin@intel.com></a><br>
> Sent: Friday, August 2, 2019 4:52 AM<br>
> To: dri-devel <a class="moz-txt-link-rfc2396E" href="mailto:dri-devel@lists.freedesktop.org"><dri-devel@lists.freedesktop.org></a>;
Koenig, Christian <a class="moz-txt-link-rfc2396E" href="mailto:Christian.Koenig@amd.com"><Christian.Koenig@amd.com></a>; Zhou,
David(ChunMing) <a class="moz-txt-link-rfc2396E" href="mailto:David1.Zhou@amd.com"><David1.Zhou@amd.com></a>; Jason Ekstrand
<a class="moz-txt-link-rfc2396E" href="mailto:jason@jlekstrand.net"><jason@jlekstrand.net></a><br>
> Subject: Threaded submission & semaphore sharing<br>
><br>
> Hi Christian, David,<br>
><br>
> Sorry to report this so late in the process, but I
think we found an issue not directly related to syncobj
timelines themselves but with a side effect of the threaded
submissions.<br>
><br>
> Essentially we're failing a test in crucible :<br>
> func.sync.semaphore-fd.opaque-fd<br>
> This test create a single binary semaphore, shares it
between 2 VkDevice/VkQueue.<br>
> Then in a loop it proceeds to submit workload
alternating between the 2 VkQueue with one submit depending
on the other.<br>
> It does so by waiting on the VkSemaphore signaled in
the previous iteration and resignaling it.<br>
><br>
> The problem for us is that once things are dispatched
to the submission thread, the ordering of the submission is
lost.<br>
> Because we have 2 devices and they both have their own
submission thread.<br>
><br>
> Jason suggested that we reestablish the ordering by
having semaphores/syncobjs carry an additional uint64_t
payload.<br>
> This 64bit integer would represent be an identifier
that submission threads will WAIT_FOR_AVAILABLE on.<br>
><br>
> The scenario would look like this :<br>
> - vkQueueSubmit(queueA, signal on semA);<br>
> - in the caller thread, this would increment
the syncobj additional u64 payload and return it to
userspace.<br>
> - at some point the submission thread of
queueA submits the workload and signal the syncobj of semA
with value returned in the caller thread of vkQueueSubmit().<br>
> - vkQueueSubmit(queueB, wait on semA);<br>
> - in the caller thread, this would read the
syncobj additional<br>
> u64 payload<br>
> - at some point the submission thread of
queueB will try to submit the work, but first it will
WAIT_FOR_AVAILABLE the u64 value returned in the step above<br>
><br>
> Because we want the binary semaphores to be shared
across processes and would like this to remain a single FD,
the simplest location to store this additional u64 payload
would be the DRM syncobj.<br>
> It would need an additional ioctl to read &
increment the value.<br>
><br>
> What do you think?<br>
><br>
> -Lionel<br>
<br>
<br>
</div>
</span></font>
</blockquote>
<p><br>
</p>
</body>
</html>