Calculate the sum of two vectors (C = A + B) in parallel using OpenMP.
A skeleton code is provided in sum(.c|.F90). Fill in the computational part
and calculate it in parallel in GPU using OpenMP offloading. Try both teams,
parallel, distribute, for / do constructs as well as loop construct.
Use compiler diagnostics to investigate differences between the two versions:
- C:
cc -fopenmp -fsave-loopmark sum.c - Fortran:
ftn -hmsgs -hlist=m -fopenmp sum.F90
Try also setting CRAY_ACC_DEBUG environment variable to emit runtime debug messages for offload activity:
export CRAY_ACC_DEBUG=3 (value 1, 2, or 3; the higher value increases the verbosity of the messages).
Try to run the code also in CPU only nodes, do you get the same results both with GPU and CPU?