Tudy and I are working on a branch of the LHC (Large Hadron Collider) BPM (beam position measurement) analysis code to be made parallel using GPUs. The branch can be found here on GitHub. This code performs harmonic analysis of turn-by-turn data from a circular acceleration (measured data or from numerical simulation). It is heavily used in the optics re-construction for the LHC and the same tools are being exported to the lower-energy machines at CERN.
Analysis of all LHC BPMs is a major real time computational bottleneck. As such, an effort has been underway to decrease the real computational time of the SUSSIX codes by paralellizing it. This has been achieved using the OpenMP API to parallelize both the Fortran and C implementations of the code. We can then safely assume that the code can also be successfully paralellized to run off the GPU.
For Assignment 1, I profiled sussix4drivexxNoO.f and Drive_God_lin.c.. The most time was spent on two subroutines in the Fortran portion of the program.
time seconds seconds calls ms/call ms/call name 47.04 5.80 5.80 311575 0.02 0.02 zfunr_ 25.63 8.96 3.16 522 6.05 6.05 ordres_ 8.43 10.00 1.04 312021 0.00 0.00 cfft_ 3.89 10.48 0.48 312559 0.00 0.02 tunelasr_
zfunr and ordres take the majority of the total run time so we'll focus on parallelizing these two subroutines first.