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
<d1b2> <johnsel> dayum that looks very good indeed
<azonenberg> :)
<azonenberg> next step is some file dialog tweaks, #538
<azonenberg> starting that tonight if i have time
Degi_ has joined #scopehal
Degi has quit [Ping timeout: 268 seconds]
Degi_ is now known as Degi
massi has joined #scopehal
<d1b2> <johnsel> well don't forget about my CI blockers either, I am about finished with my game so I will have time come this weekend
t4nk_fn has quit [*.net *.split]
t4nk_freenode has joined #scopehal
<azonenberg> Yeah I'm going to be working on the CI infrastructure now into this weekend too
<azonenberg> the VM server is rebooted with the new ram and PSU
<azonenberg> i still have to set up the open source version of XOA
<azonenberg> And i still have a shed to finish :p
<azonenberg> tomorrow is supposed to be dry-ish weather so before all of the turkey day festivities i'm gonna try and get at least one coat of paint up
<azonenberg> and probably do some work on the roof, i still have about 1/3 of the surface not shingled and covered by a tarp
<azonenberg> So i probably won't be getting much coding done between that and the feast
<azonenberg> but friday into the weekend are open
<_whitenotifier> [scopehal-apps] azonenberg opened issue #549: ngscopeclient: preference settings for protocol analyzer colors - https://github.com/glscopeclient/scopehal-apps/issues/549
<_whitenotifier> [scopehal-apps] azonenberg labeled issue #549: ngscopeclient: preference settings for protocol analyzer colors - https://github.com/glscopeclient/scopehal-apps/issues/549
<_whitenotifier> [scopehal-apps] azonenberg labeled issue #549: ngscopeclient: preference settings for protocol analyzer colors - https://github.com/glscopeclient/scopehal-apps/issues/549
<_whitenotifier> [scopehal-apps] azonenberg pushed 2 commits to master [+2/-0/±8] https://github.com/glscopeclient/scopehal-apps/compare/767b82e5ca38...828b7e9544f0
<_whitenotifier> [scopehal-apps] azonenberg f9dfb51 - Initial implementation of NFDFileBrowser. See #538. Still need preference + fullscreen mode selector to use it or IGFD as appropriate
<_whitenotifier> [scopehal-apps] azonenberg 828b7e9 - Added preferences for file browser selection. Fixes #538.
<_whitenotifier> [scopehal-apps] azonenberg closed issue #538: Look into how to create a *modeless*, cross platform, native file dialog - https://github.com/glscopeclient/scopehal-apps/issues/538
bvernoux has joined #scopehal
<d1b2> <johnsel> cool, that should work
massi has quit [Remote host closed the connection]
bvernoux1 has joined #scopehal
bvernoux has quit [Ping timeout: 264 seconds]
bvernoux1 has quit [Quit: Leaving]
<Degi> @azonenberg Do you know how much time Sonnet takes per time step per simulation cell?
<Degi> And how many cells the average simulation is?
<azonenberg> Degi: it's not time domain, it's frequency domain
<azonenberg> you evaluate the response at a number of frequency points and then it does some interpolation and curve fitting to figure out what happens in between
<azonenberg> (by default it's adaptive, if it doesn't like the fit it will run at more points, but you can also do a linear sweep without that or a single frequency)
<azonenberg> Run time is not linear in cell count either
<azonenberg> it does a FFT then solves a big linear system of equations
<azonenberg> as far as cell count goes, a few random examples
<azonenberg> my most recent connector launch model with fine meshing was 17.6K cells and took a total of 28 mins 40 sec
<azonenberg> one i did a while back for stackup design was 7.5K cells and took 1 hr 40 mins (note that it actually took longer despite less cells, i think this might have to do with the much more complex stackup? this was also 18.52 instead of 18.56 so there may have been code changes)
<azonenberg> one i did for DDR3 design was 51.7K cells and took 18 hours
<azonenberg> one i did for the AKL-PT5 filter was 3.68K cells and took a minute 29 sec
<azonenberg> another one i did of the entire PT5 including tip resistors and everything was 18.7K cells and took 49 min 32 sec
<azonenberg> Some of the longer sims needed more frequency points to converge
<azonenberg> also, the mapping of actual grid squares (fixed pitch across the entire design) to simulation cells has an impact
<azonenberg> some of the coupling calculations have to be done for each grid square, but multiple grid squares are combined into one sim cell
<azonenberg> so you end up with some operations like coupling scaling with the number of grid squares while others like the matrix solve scale with the number of sim cells
<Degi> Hmm interesting
<Degi> I wonder how that compares to a FDTD simulation with a gaussian or delta shaped excitation
<azonenberg> Also, right now the core of their matrix solver is a bunch of hand tuned SSE code written circa 2005
<azonenberg> I'm working with them to update it to use AVX, which gives significant speed boosts on modern hardware
<azonenberg> they've promised me an engineering build any day now for the past two weeks
<Degi> Hmm nice
<azonenberg> so whenever it's stable enough i should get an alpha to play with
<azonenberg> (you know a vendor likes you when they ask you for an NDA so they can share source code, lol)
<Degi> Neat
<azonenberg> The new avx math functions are my own work, although their team handled the integration into the existing code and the glue to detect whether avx is available or not etc
<azonenberg> anyway, point is all of these performance figures are based on what was current at the time i ran the sim in question
<azonenberg> and are likely invalid already, if not they will be soon :p