Akari has quit [Quit: segmentation fault (core dumped)]
Akari has joined #zink
<ajax>
zmike: in !12934, what do you mean by "unresolve"?
<zmike>
ajax: it's for tilers; basically a vk ext that will do a copy from single sample -> multisample at the start of renderpass
<kusma>
replicate is usually the terminology for that...
<ajax>
got it, thanks
nsneck has joined #zink
<hch12907>
anyone mind reviewing 13004?
<zmike>
yea
<zmike>
I did
<zmike>
but I need to test it too
<hch12907>
ah cool, take your time
<zmike>
prob get to it today at some point
<hch12907>
pretty sure there'd be some corner cases here and there, that part of the codegen is hairy...
<zmike>
yea we always break something or other when it changes
<hch12907>
but FYI my tests with Haswell show that we are not accidentally disabling extensions (except those that were mistakenly enabled of course)
<zmike>
good
<zmike>
thanks for checking it out
<ajax>
is there a tool that watches your vulkan app and draws graphs of the render passes it creates?
<zmike>
there's gfxreconstruct, which is like apitrace for vulkan
<zmike>
dunno exactly what you mean by graphs?
<zmike>
hch12907: tested
<ajax>
zmike: one node for every draw command, input arcs are your textures / ubos / whatever, output arcs are render targets. and then for a whole frame you can draw a digraph whose last node is the swapchain image.
<zmike>
ah
<zmike>
I'm not aware of any such thing? maybe bnieuwenhuizen or JoshuaAshton or jekstrand would know
<hch12907>
zmike: thanks
<bnieuwenhuizen>
not aware of any
<bnieuwenhuizen>
might be able to do something by dumping the renderdoc trace to xml
<bnieuwenhuizen>
but good luck getting a nice graph with huger descriptorsets and VK_KHR_descriptor_indexing
<zmike>
good news: zink doesn't do either of those things
<bnieuwenhuizen>
not even with bindless?
<zmike>
nope, not really
<bnieuwenhuizen>
huh, how do you do bindless then?
<zmike>
mesa only exposes 32 bindless texture units
<zmike>
so juggling that many doesn't require a huge set at all
<bnieuwenhuizen>
uh what
* bnieuwenhuizen
never knew
<zmike>
yeah drivers that implement it for mesa only have to allocate up to 1024 descriptors in a set
<zmike>
and even that's probably way overkill
<bnieuwenhuizen>
all the discussions I've seen about descriptor indexing is how 500k is not enough
<zmike>
bindless texture != descriptor indexing
<bnieuwenhuizen>
yep, apparently
<zmike>
zink only uses descriptor indexing features to implement bindles
<zmike>
not the actual huge sets
<ajax>
it probably says something about me that you thought "dump to xml" and i thought "dump to graphviz"
<zmike>
basically the same thing
<zmike>
baryluk: any chance you wanna check out that exploder geexlab demo? I've been running it here for about 5 minutes now and it's completely fine