(Date: 8.12.2014 – 12.12.2014)
It’s PANDA week! The collaboration meeting on its rounds comes to Jülich. Starting form Tuesday, and going to Friday, we inhabit some of the meeting room at the center. Our task, as PhD students, is to take care of the technical equipment of the sessions. Easy.
I have my talk on Wednesday in the pattern recognition session on the Circle Hough algorithm1. Actually, on the GPU implementation of the algorithm – the actual algorithm is presented in the talk before me by my supervisor. To have something to show, I have to prepare a something. So: Working during the weekends and to late hours on Monday. And during the meeting on Tuesday. But: I manage to finish my algorithm and time it. With a not super correct timing, my GPU algorithm (with out data copies) is up to 150 times faster than the CPU version. But there are some asterisks involved which I need to tackle in the next couple of weeks. Also, I find a strange problem when running my algorithm not on my own institute’s PC, but on the small dedicated GPU server we have. There, for some reason, every 20th or so GPU kernel call results in 1000 x slower execution performance. Strange, but not easily debuggable.
The talk itself is quite alright, beyond the results concerning the GPU Circle Hough, I present also the recent stuff we did concerning the Triplet Finder — this was not shown for some time. A lot of interested questions afterwards, we slightly went over my time slot.
Unfortunately, I did not find time to write stuff for my thesis. First, I created the code for the talk, then I created the presentation, held it, and then started dissecting the code.
Password to the slides, as usual, twice our experiment’s name ↩