Can anyone explain how to use OpenACC in GCC? - c++

Question

I am using WSL Ubuntu. And use GCC to compile my C++ code.
I am trying to use OpenACC to parallelize my code. I heard that I can use OpenACC with GCC.
So after I install Ubuntu, I installed GCC.
And then I typed as follows:
$ g++ -fopenacc -o t testfile.cpp
As far as I know, -fopenacc should enable me to use OpenACC inside my code.
However, I got an message saying:
lto-wrapper: fatal error: could not find accel/nvptx-none/mkoffload in /usr/lib/gcc/x86_64-linux-gnu/9/:/usr/lib/gcc/x86_64-linux-gnu/9/:/usr/lib/gcc/x86_64-linux-gnu/:/usr/lib/gcc/x86_64-linux-gnu/9/:/usr/lib/gcc/x86_64-linux-gnu/ (consider using ‘-B’)
So I guess I missed something.
Does anyone know how to use OpenACC with GCC?
Edited:
I tried:
sudo apt-get install gcc-9-offload-nvptx
And then the error message is now changed to:
lto1: error: ‘-fcf-protection=full’ is not supported for this target
mkoffload: fatal error: x86_64-linux-gnu-accel-nvptx-none-gcc-9 returned 1 exit status
compilation terminated.
lto-wrapper: fatal error: /usr/lib/gcc/x86_64-linux-gnu/9//accel/nvptx-none/mkoffload returned 1 exit status
compilation terminated.
/usr/bin/ld: error: lto-wrapper failed
collect2: error: ld returned 1 exit status

Answer 1

I got one step further with:
gcc-10 -O2 -ffast-math -fopenacc -fopt-info-all-omp -fcf-protection=none -foffload=nvptx laplace2dkernels.c
but just to find the next problem:
gcc-10: fatal error: GCC is not configured to support nvptx as offload target
compilation terminated
So it seems they only compiled it for AMD GCN.
If you use
gcc-10 -O2 -ffast-math -fopenacc -fopt-info-all-omp -fcf-protection=none -foffload=nvptx-none laplace2dkernels.c
>laplace2dkernels.c:74:9: optimized: assigned OpenACC seq loop parallelism
>/usr/bin/ld: /tmp/ccYd0cZe.crtoffloadtable.o: warning: relocation against >`__offload_vars_end' in read-only section `.rodata'
It compiles but then when running it misses:
libgomp: while loading libgomp-plugin-gcn.so.1: libgomp-plugin-gcn.so.1: cannot open shared object file: No such file or directory
I do not need to install those libraries because I want to get it to run with my Nvidia card.
With the PGI compilers everything worked fine.
NOTE: I am also using WSL and special NVIDIA drivers to allow for GPU passthrough.

Answer 2

Here's what worked for me for a Tiny C++ stencil code with OpenACC directives, after installing gcc-10-offload-nvptx:
g++ -c -g -O -fopenacc -foffload=nvptx-none -fcf-protection=none -foffload=-misa=sm_35 -fno-stack-protector stencil_acc.cxx
g++ -o stencil_acc -g -O -fopenacc -foffload=nvptx-none -fcf-protection=none -foffload=-misa=sm_35 -fno-stack-protector util.o stencil_acc.o