The ASEArch team are developing three illustrative testcases which will compare the performance achievable on different hardware platforms, and document the programming effort required to obtain it.
- 3D Laplace solver using Jacobi iteration on a structured grid (available here on CCP Forge, with report here)
- 3D ADI (alternating direction implicit) solver on a structured grid -- this involves the parallel solution of sets of independent tridiagonal equations in each coordinate direction. Two talks related to this work were given at NVIDIA's GPU Tecnology Conference 2014 (talk 1, talk2)
- a Lattice Boltzmann code (paper at International OpenCL Workshop)
In addition, there is related work (funded by AWE) at Bristol, Warwick and Oxford on a structured grid CFD testcase called Cloverleaf. This also involves implementations on a range of different platforms, using different software implementations (OpenMP, CUDA, OpenCL, OpenACC). The Oxford work shows how different implementations can be generated automatically using OPS, a high-level API abstraction being developed under an EPSRC-funded project.