This repository is a proof of concept showing that an OpenMP offloading backend
for C++ standard parallelism can be implemented in llvm-project/libcxx.
To install the LLVM compiler with the patch applied, modify the source and descitnation directories in install_llvm.sh and type
$ source install_llvm.sh
There are three examples using std::for_each(std::par_unseq,...). To compile
and run them, select a file from src and compile it by specifying APP, for
instace
$ mkdir bin
$ make APP=foreach_std_vector OMP=1 && bin/foreach_std_vector
Convince yourself that the program runs on the GPU with
$ LIBOMPTARGET_DEBUG=-1 ./bin/foreach_std_vector
or
$ ./bin/foreach_std_vector & watch -n 0.1 rocm-smi
========================= ROCm System Management Interface =========================
=================================== Concise Info ===================================
GPU Temp (DieEdge) AvgPwr SCLK MCLK Fan Perf PwrCap VRAM% GPU%
0 35.0c 77.0W 1725Mhz 800Mhz 14.51% auto 225.0W 9% 100%
1 30.0c 17.0W 930Mhz 350Mhz 14.51% auto 225.0W 0% 0%
2 30.0c 19.0W 930Mhz 350Mhz 14.51% auto 225.0W 0% 0%
3 31.0c 23.0W 930Mhz 350Mhz 14.51% auto 225.0W 0% 0%
4 28.0c 21.0W 930Mhz 350Mhz 14.51% auto 225.0W 0% 0%
5 29.0c 17.0W 930Mhz 350Mhz 14.51% auto 225.0W 0% 0%
6 28.0c 22.0W 930Mhz 350Mhz 14.51% auto 225.0W 0% 0%
7 28.0c 20.0W 930Mhz 350Mhz 14.51% auto 225.0W 0% 0%
====================================================================================
=============================== End of ROCm SMI Log ================================