Status meeting October 11, 2012¶
Collaboration with KISTI.¶
Tim Tautges is collaborating with S. Korean on 'Direct Accelerated Geometry Monte Carlo' implementation related Geant4.
KISTI is planning to hire a new PostDoc to replace the person that was going to work on Geometry.
KISTI's next year proposal should come out mid-November at which time they should know how much funding they will have for next year.
Philippe: Learning about the ray tracing libraries provided by Nvidia.
Soon: I am finishing putting Otto's geometry package into our prototype while preserving the OpenCL part of his code. One difference is that OpenCL does specify where the pointer's address is located. Big plus is the voxelized navigation. Tested with CMS geometry, 10000 solids, ran but the performance is low.
Soon: I also added RK-Nystrom and RK-fellberg to our example.
Soon: Now I think we are almost ready to come back and take a look at optimization.
Dongwook: Tested generating random number on the GPU (Gaussian not implemented yet). Added EM physics electron ionization to extend the interaction length. Still missing the AlongStep part (only PostStep ; post interaction length and energy loss). Multiple scattering for electrons. For AlongStep the code need Geometry information and navigator information.
Discussion: Eventually we would need to have reproducibility on the same hardware with the same original seed we should get the exact same result.
Marc 'volunteered' to look at random number reproducibility and stability of the thread scheduling.
Dongwook there is ensight, an eclipse cuda development environment.