Lessons in Programming Model Comparisons Using OpenMP and CUDA for Targetting GPUs

" summary

Comparison of two GPU programming models for Fortran: OpenMP target directives and CUDA, using a Riemann problem implementation from the GenASiS astrophysics code as the test case. The work measures effort and performance trade-offs on Summit (which supports both) and considers portability implications for Frontier’s OpenMP and HIP-translated CUDA-like models, reporting practical lessons about productivity and portability.

AI-written summary · about