Repo for benchmarking and profiling code

1. Checkout this repo
2. Make sure to have the following:
   1. Intel LLVM sycl-nightly/20200330
   2. CUDA 10.2
   3. Nsight Compute >= 2020.2
   4. hipSYCL 0.9.0 and dependencies (core, base, etc.)
   5. python3+matplotlib (for plots)
3. Then execute the following:

$ cd benchprof
$ export BENCHPROF_DIR=$PWD
$ cd onemkl
$ source scripts/rng_compile.sh <backend> # backend = host, intelcpu, intelgpu, \
                                          # mkl_curand, mkl_hiprand,            \
                                          # curand, hiprand
$ source scripts/rng_run.sh <backend>
$ python3 python/plot_clock_csv.py

Please report any issues with reproducibility.

The FastCaloSim input files are proprietary ATLAS Experiment data and so cannot be shared publicly. As a result, reproducing our results for this application is not possible.