Here we will port the nbody sample code in oneAPI HPC Toolkit located at
oneAPI-samples/DirectProgramming/C++SYCL/N-BodyMethods/Nbody/ from C++ to Fortran.
See README2023.md and Makefile2023 for previous versions of the documents based on oneAPI 2023 which mostly discusses compiler options and steps for running on Intel DevCloud.
In this document, we will focus on oneAPI 2024 and discuss steps to run do concurrent offload to a client intel GPU on Ubuntu operating system.
Following https://www.intel.com/content/www/us/en/docs/fortran-compiler/developer-guide-reference/2024-0/fopenmp-qopenmp-target-do-concurrent.html, the compiler options used are:
-fopenmp-target-do-concurrent -fiopenmp -fopenmp-targets=spir64
-
Install intel oneAPI 2024 (base and HPC including fortran compiler)
-
Assuming the installation is on the default location, set the environment:
source /opt/intel/oneapi/setvars.sh -
Install runtime libraries following: https://dgpu-docs.intel.com/driver/client/overview.html
-
Build and run using
make && ./nbody
===============================
Initialize Gravity Simulation
nPart = 16000
nSteps = 10
dt = 0.1000000
-------------------------------------
s kenergy time (s)
-------------------------------------
1 3.9802892E-03 0.2495000
2 3.9802892E-03 8.1000002E-03
3 3.9802892E-03 7.3000002E-03
4 3.9802892E-03 7.7000000E-03
5 3.9802892E-03 7.4000000E-03
6 3.9802892E-03 8.8000000E-03
7 3.9802892E-03 7.4000000E-03
8 3.9802892E-03 7.4000000E-03
9 3.9802892E-03 7.1999999E-03
10 3.9802892E-03 7.8999996E-03
Total Time (s): 0.3187000
=============================
To check that GPU is being used install intel-gpu-tools and check the output of intel_gpu_top while the code is running.