omp_get_num_devices() returns 0 even though I have GPUs

Clang Frontend Using Clang
1

Hi,
I am trying to offload to my GPUs. I have GTX1060 6GB with compute capability of 6.1.

I followed the instructions given in

https://hpc-wiki.info/hpc/Building_LLVM/Clang_with_OpenMP_Offloading_to_NVIDIA_GPUs

Then I compiled my helloworld program using the following environment variables:
CFLAGS=“-fopenmp -fopenmp-targets=nvptx64-sm_61 -pthread -I /usr/lib/gcc/x86_64-linux-gnu/9/include -O2 --libomptarget-nvptx-path=$CLANG_HOME/lib --cuda-gpu-arch=sm_61”

LD_LIBRARY_PATH=$CLANG_HOME/lib:$LD_LIBRARY_PATH

CC=clang

my helloworld program simply printfs omp_get_num_devices() after #including <omp.h>

it compiles fine; but when i run it i get the number of devices as 0.

Does someone have any idea?
Thanks much,
Mani

2

I forgot to mention that I am using the latest clang-llvm-openmp versions of source code (10.0.0).

3

Hi,
I figured it out. I hadnt downloaded the proper NVIDIA drivers.
https://linuxhint.com/ubuntu_nvidia_ppa/
proved useful to me.
Now I am able to detect my GPU.

Thanks much,
mashilamani