azonenberg changed the topic of #scopehal to: libscopehal, libscopeprotocols, and glscopeclient development and testing | https://github.com/glscopeclient/scopehal-apps | Logs: https://libera.irclog.whitequark.org/scopehal
Degi_ has joined #scopehal
Degi has quit [Ping timeout: 276 seconds]
Degi_ is now known as Degi
Fridtjof has quit [Ping timeout: 264 seconds]
Stary has quit [Ping timeout: 260 seconds]
<d1b2> <tnt> Oh wow ... just got the same vk::DeviceLostError on a RTX4070 ...
<d1b2> <tnt> [21066.610073] NVRM: Xid (PCI:0000:0b:00): 109, pid=84166, name=ngscopeclient, Ch 00000020, errorString CTX SWITCH TIMEOUT, Info 0x19c00f
<d1b2> <tnt> So I was still in immediate presentation mode when this was triggered, but still, that shouldn't happen. I'm wondering if there is some race somewhere and it's just that on powerful cards the chance of hitting it are very slim because they render so much faster than the vsync in normal mode while for the intel ones, you're usually refreshing slower than vsync.
bvernoux has joined #scopehal
azonenberg has quit [Ping timeout: 250 seconds]
azonenberg has joined #scopehal
Stary has joined #scopehal
Fridtjof has joined #scopehal
<azonenberg> @tnt: Very possible
<d1b2> <tnt> Is there a way to draw the eye mask without any actual data ?
<d1b2> <tnt> It was rendering badly under ngscopeclient on my nvidia setup, but rendering fine in glscopeclient. But I didn't have time to fuss around, I needed to collect the data so I just used glscopeclient. Now I'd like to look more at what was failing but I don't have a data source.
<azonenberg> i think you might need a single UI for it to set the displayed UI width in pixels right
<azonenberg> not sure if completely empty is doable
<azonenberg> but this is a WIP
<azonenberg> also, you should now be able to save a scopesession in glscopeclient and load it in ngscopeclient for analysis
<azonenberg> this is read only, you cannot save a scopesession in ngscopeclient yet (at least, not a useful one - i think it can write a dummy file header or something)
<tnt> Including the data ?
<azonenberg> Yes
<azonenberg> It does not save the relative sizes and placements of waveform groups
<azonenberg> they'll just show up as top level tabs you need to rearrange as you see fit
<tnt> Oh. Then I might actually have some data I can use.
<azonenberg> but all filters, waveform areas, data, and instrument config should load correctly as of this past weekend
<tnt> Somehow I had in my head it was loading state/config but not data.
<azonenberg> That was true for a few hours saturday afternoon :p
<azonenberg> you might have seen my update then lol
<azonenberg> anyway, as of now you cannot save a scopesession in ngscopeclient but loading should work
<azonenberg> saving is WIP but a few days off given how busy i am
<tnt> Yup checked and it loaded my scopesession from like 1 year ago :)
<d1b2> <tnt> Ok, mask alsodrawn badly on intel.
<d1b2> <tnt> Oh, looking at the code now : Are the polygons supposed to be convex in the eye mask definition ?
<d1b2> <tnt> No easy fix unfortunately. The old cairo code handled arbitrary polygon fills, but ImGUI drawing list doesn't have such support, it can only do convex polygons.
<bvernoux> It seems there is several native ImGUI API to draw rectangle, triangle with option to fill them ...
<tnt> bvernoux: it uses AddConvexPolyFilled currently.
<bvernoux> you can use something like window->DrawList->AddRect()
<tnt> Which is clearly limited to Convex polygons.
<bvernoux> with ImGuiContext& g = *GImGui;
<bvernoux> ImGuiWindow* window = g.CurrentWindow;
<bvernoux> you can also just write an Image too
<bvernoux> draw_list->AddImage
<bvernoux> or maybe you need that https://github.com/ocornut/imgui/issues/760 Flood fill generic polygon (not strictly convex); methods? #760
<tnt> Yeah, that'd be the "correct" solution, but apparently they decided not to include that in ImGui so you need an external lib.
<bvernoux> other way is to do it directly with Vulkan API
<bvernoux> but it is probably not simpler ;)
<tnt> Definitely not.
<bvernoux> an ext lib is the good way to do it in that case
<bvernoux> especially if it does not have external dependencies
<bvernoux> to be discussed with @azonenberg as maybe he has already done something about that
<tnt> bvernoux: did you receive your new rig btw ?
JSharp has quit [*.net *.split]
JSharp_ has joined #scopehal
<azonenberg> Polygons were always suppsed to be convex
<azonenberg> if not the mask was probably bugged
<azonenberg> as i am not aware of any protocol that specifies a non convex polygon
<azonenberg> We also need to do some retooling eventually because right now we use cairo rendering for mask violation checks and imgui for displaying the mask
<azonenberg> so it may not be 1:1 and we pull in cairo, a fairly large dependency, for that one tiny flood fill
<d1b2> <tnt> I was implementing this : https://i.imgur.com/0vwnDfe.png
<d1b2> <tnt> But I can manually divide it.
<azonenberg> yeah that is a near term hack. also we probably want a separate pulse mask test filter at some point that works on a single waveform vs an eye pattern?
<azonenberg> long term open a ticket for non convex polygon support
<azonenberg> no idea if it will happen any time soon
<azonenberg> but its nice to have a request
<tnt> On scolehal-apps ?
<azonenberg> Hmm, yeah probably on -apps
<azonenberg> may require changes to the lib too TBD
<_whitenotifier-8> [scopehal-apps] smunaut opened issue #587: Support non-convex polygons in EyeMasks - https://github.com/glscopeclient/scopehal-apps/issues/587
<bvernoux> tnt, I will receive my new PC tomorrow but I need to install everything Dual Boot Win11 Pro on dedicated SSD and on other dedicated SSD Ubuntu 22.04 LTS after
<bvernoux> tnt, I have finally chosen the ASUS TUF Gaming GeForce RTX 4070 12GB GDDR6X OC Edition
<d1b2> <tnt> Any reason for that particular card ?
<bvernoux> with Ryzen 9 7950X3D + 96GB DDR5 + ROG CROSSHAIR X670E HERO (mainly for USB4, PCIE5, DDR5...)
<d1b2> <david.rysk> X3D... be careful with those voltage issues
<bvernoux> yes the reason is the price is great and the power consumption/temperature is very low with better than on Reference card RT4070
<bvernoux> Yes the Bios shall be update to latest version to avoid burning the 7950X3D
<bvernoux> I have a special warranty of 1 years for any destruction even if I destruct it myself ;)
<bvernoux> all was mounted & tested (with burn in test) by pcspecialist (in UK) even if I have bought in to pcspecialist.fr
<bvernoux> They have very nice price not far from raw price for each parts bought alone but with everything mounted/tested and with a warranty of 3 years ...
<bvernoux> I was tempted by the Radeon RX 7900 XT which is better on paper than RTX 4070 but the heat and power is 2x more for not a lot of gain in practice especially for a workstation