<div dir="ltr"><div class="gmail_extra"><div class="gmail_quote">On Mon, Mar 19, 2018 at 4:53 PM, Alex Deucher <span dir="ltr"><<a href="mailto:alexdeucher@gmail.com" target="_blank">alexdeucher@gmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><span class="">On Mon, Mar 19, 2018 at 4:26 PM, Marek Olšák <<a href="mailto:maraeo@gmail.com">maraeo@gmail.com</a>> wrote:<br>
> When Mesa wants a buffer in VRAM, it always sets VRAM. It relies on BO move<br>
> throttling to prevent unnecessary BO moves.<br>
><br>
> My questions are:<br>
> - what should Mesa do differently for tiny VRAM?<br>
> - what is a tiny VRAM?<br>
> - if VRAM is tiny, which allocations should we put there?<br>
<br>
</span>Right now, all mesa would have to do is query the kernel to see if the<br>
device supports s/g display. If it does, it can do whatever makes<br>
sense in the long term once we've had more time to benchmark, etc.,<br>
but to start off with, I'd say just make it GTT or VRAM|GTT.<br>
Otherwise we have the kernel second guessing mesa and I'd like to<br>
avoid that. I want the kernel to respect that userspace asks for as<br>
much as possible.<br></blockquote><div><br></div><div>Sounds good.</div><div><br></div><div>Marek<br></div></div></div></div>