<d1b2>
<david.rysk> tnt: I'm just seeing the backlog -- what laptop? I was able to get it working on an intel hd graphics 4000 after installing bleeding-edge mesa/vulkan
<d1b2>
<david.rysk> ah, you're on Haswell? yeah that's oooold... llvmpipe would be the way to go, you'll lose some performance but it should work
<azonenberg>
@david.rysk: last time i checked we had problems running on llvmpipe but lain's queuemanager refactoring may have fixed most of that
<d1b2>
<david.rysk> no wait, haswell is newer than hd 4000, so you might be able to get away with bleeding edge mesa. The "not updated anymore" is odd, apparently it got rolled into mesa
<azonenberg>
(it only had a single queue and we required several, we can now run on arbitrarily feew queues with some mutexing)
<d1b2>
<david.rysk> I installed bleeding-edge mesa on my ivy bridge (hd 4000) system and it worked, but that was a few months ago
<d1b2>
<david.rysk> I can try again in a few weeks
<d1b2>
<david.rysk> ah, I see about the queues -- that should help with Metal
<azonenberg>
Yes thats why we originally did it
<azonenberg>
but it will help on other platforms too
<azonenberg>
The other issue is that several of our shaders for accelerated filters right now do not run on intel or emulated platforms that have a 64k thread max X axis grid size
<azonenberg>
if your waveform is more than about 4M samples
<d1b2>
<david.rysk> yeah iirc older Intel was also queue-limited
<d1b2>
<david.rysk> and ran into that as well
<azonenberg>
the fix is just to break those up into x-y grids instead of 1D
<azonenberg>
but it means refactoring a bunch of code and i havent got around to it yet
<azonenberg>
they can happily run 100M threads but it has to be 64K x 1024 or something
<azonenberg>
not 100M x 1
<d1b2>
<david.rysk> yeah
<d1b2>
<david.rysk> I'm not sure what distro tnt is running; for Ubuntu there are PPAs with bleeding-edge mesa; for arch I got it from the AUR
<azonenberg>
anyway as long as your memory depth is less than 64K * threads per block (usually 32 or a multiple thereof)
<d1b2>
<david.rysk> At first I was confused because the vulkan support was a separate driver, but at some point it got rolled into mesa
<_whitenotifier>
[scopehal] azonenberg a6da8ca - Fixed bug causing message frames to show up as the wrong color
massi has joined #scopehal
<d1b2>
<tnt> Ok, I gave it a shot. Despite the MESA-INTEL: warning: Haswell Vulkan support is incomplete , it does display something. (Right is new, Left is old screenshot of the .scopesession file) https://i.imgur.com/kxBykEo.png