Repo for benchmarking and profiling code
- Checkout this repo
- Make sure to have the following:
- Intel LLVM
sycl-nightly/20200330
- CUDA 10.2
- Nsight Compute >= 2020.2
- hipSYCL 0.9.0 and dependencies (core, base, etc.)
- python3+matplotlib (for plots)
- Intel LLVM
- 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.