Future exascale computing is demanding more and more parallelism from current software if peak computation rates are to be realized. However, exploiting this additional parallelism is not trivial. One approach is to identify finer grained parallelism using data parallel primitives (DPP). Visualization frameworks such as Dax and VTK-m are being developed using DPP for this purpose. Our work presents an exploratory study of how volume rendering maps to current and future super computing architectures. We implement a ray casting and cell projection volume renderer in Dax using DPP and compare their performance on three different hardware architectures. Despite the portability provided by these frameworks, we observe that additional architecture specific modifications are necessary to achieve acceptable performance on some architecture.

Presenter(s)

Hendrik Schroots , University of California, Davis