<div dir="ltr">Hi<div><br></div><div>Just another gentle ping to see where you are with this?</div><div><br></div><div>Cheers</div><div><br></div><div>Mike</div></div><br><div class="gmail_quote"><div dir="ltr">On Wed, 12 Oct 2016 at 01:40 Michel Dänzer <<a href="mailto:michel@daenzer.net">michel@daenzer.net</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">On 11/10/16 09:04 PM, Christian König wrote:<br class="gmail_msg">
> Am 11.10.2016 um 05:58 schrieb Michel Dänzer:<br class="gmail_msg">
>> On 07/10/16 09:34 PM, Mike Lothian wrote:<br class="gmail_msg">
>>> This has discussion has gone a little quiet<br class="gmail_msg">
>>><br class="gmail_msg">
>>> Was there any agreement about what needed doing to get this working<br class="gmail_msg">
>>> for i965/amdgpu?<br class="gmail_msg">
>> Christian, do you see anything which could prevent the solution I<br class="gmail_msg">
>> outlined from working?<br class="gmail_msg">
><br class="gmail_msg">
> I thought about that approach as well, but unfortunately it also has a<br class="gmail_msg">
> couple of downsides. Especially keeping the exclusive fence set while we<br class="gmail_msg">
> actually don't need it isn't really clean either.<br class="gmail_msg">
<br class="gmail_msg">
I was wondering if it's possible to have a singleton pseudo exclusive<br class="gmail_msg">
fence which is used for all BOs. That might keep the overhead acceptably<br class="gmail_msg">
low.<br class="gmail_msg">
<br class="gmail_msg">
<br class="gmail_msg">
> I'm currently a bit busy with other tasks and so put Nayan on a road to<br class="gmail_msg">
> get a bit into the kernel driver (he asked for that anyway).<br class="gmail_msg">
><br class="gmail_msg">
> Implementing the simple workaround to sync when we export the DMA-buf<br class="gmail_msg">
> should be something like 20 lines of code and fortunately Nayan has an<br class="gmail_msg">
> I+A system and so can actually test it.<br class="gmail_msg">
><br class="gmail_msg">
> If it turns out to be more problematic or somebody really starts to need<br class="gmail_msg">
> it I'm going to hack on that myself a bit more.<br class="gmail_msg">
<br class="gmail_msg">
If you mean only syncing when a DMA-buf is exported, that's not enough,<br class="gmail_msg">
as I explained before. The BOs being shared are long-lived, and<br class="gmail_msg">
synchronization between GPUs is required for every command stream<br class="gmail_msg">
submission.<br class="gmail_msg">
<br class="gmail_msg">
<br class="gmail_msg">
--<br class="gmail_msg">
Earthling Michel Dänzer | <a href="http://www.amd.com" rel="noreferrer" class="gmail_msg" target="_blank">http://www.amd.com</a><br class="gmail_msg">
Libre software enthusiast | Mesa and X developer<br class="gmail_msg">
</blockquote></div>