Compiling

Compiling part looks like this:

g++ -std=c++1y -O3 -g3 -Wall -c -fmessage-length=0 -march=native -mavx -fopenmp -pthread -fPIC -MMD -MP -MF"src/foo.d" -MT"src/foo.d" -o "src/foo.o" "../src/foo.cpp"

Linking part looks like this:

g++ -L/usr/local/cuda/lib64 -o "your_app_name" ./src/foo.o -lOpenCL -lpthread -lgomp

Both g++7 - g++10 can run it.