Status meeting March 15th, 2012¶
Manfred from CMU is joining the effort.
Incorporating CUDA code in cmsExp (Soon)¶
Soon added connecting code between cmsExp and Cuda and he has started testing the calculation on Cuda (leveraging the extraction).
Also wrote a simple transportation code. Comparing the performances CPU vs GPU. On GPU, he is using 148 threads. Working
on tweaking the blocks vs threads counts. Next step is to quantify these performances.
Port of the standalone example to CUDA (Philippe).¶
CMS field valuation is the same as the texture. But the storage is float rather than double.
However Marc point it is likely to be okay.
Search for equation of (magnetic) motion (Marc).¶
I have started looking but I am not sure that what I found is applicable yet (particle propagated in gravity field). Astronomy simulation paper. Conserve energy.
Daniel in at least one case, we have used a time varying field (in the context of accelerator simulation).
Details of how Geant4 does the task of stepping particles (Marc/Jim).¶
One question to be asked to John Apostolakis is what is the overall algorithm to calculate the step/movement. In particular is there a way to do more 'throw' away calculation. Marc will setup a phone meeting with John A.
Philippe will test the use of texture for the magnetic field lookup and evaluate the speed-up and accuracy.
Soon will do a systemic evaluation of the performance of the code already ported to CUDA.