#GradientBundleAnalysis
Today I got the concatenation and resampling of saddle point stream traces with existing gradient paths working correctly. I realized that I need the saddle point stream traces to be offset from those seeded from neighboring nodes so that the resulting volumes are still triangular, so there's that to do before I can make surfaces.
Also still need to add the detection of whether a volume straddles more than one #IrreducibleBundle.
Improved the mesh optimization. It now uses the max of local triangle variance for convergence testing rather than the total variance. It's a bit slow though. I sped things up with some openmp inside the variance function, but it's still slow since the majority of the work is inherently serial. Makes me appreciate the benefit of having all your own functions for stream traces and stuff; don't need to worry about one thread making a stream trace and then another thread stealing it by calling getnumzones at the wrong time.
Might be able to speed up the mesh optimization by doing a few iterations of jiggle mesh before I start the actual optimizer. That'll let things move around to fit the constraints in a way cheaper way, then the optimizer just needs to do the fine adjustments.
Also made GBA play nice with Tecplot, announcing data loads and even mesh memory requirements.
That's one thing I need to do for the data loaders is announce the memory required to Tecplot.
Take another whack at some basic openmp parallelism in #Bondalyzer. It might work better in #VS2013.
Personal
Otherwise today was good. #Kaori got another paycheck, I finished digitizing the rest of my file cabinet—it's now useless, I did everything but the essays for the #DOE #CSGF application, we had pineapple fried rice for lunch and yasai itame for dinner.
Remind me to check that the transcripts I ordered from #Mines included undergrad and grad.
via https://dayone.me/29GezkN









