hiperc
hiperc copied to clipboard
usnistgov
High Performance Computing Strategies for Boundary Value Problems
CircleCI appears to offer GPU build servers. https://circleci.com/docs/enterprise/gpu-configuration/
Portability would be improved by folding host arrays into a container, similar to the `struct CudaData` currently used.
### There's a *missing feature* in HiPerC NVIDIA has a distributed multigrid solver, [amgx](https://github.com/NVIDIA/AMGX), which is of strong relevance to boundary value problems.
Review *Carslaw and Jaeger* to determine whether a true analytical solution is possible, and if the approximate solution is as close as it can get.
Inner loops on CUDA convolution code should run faster using a `#pragma unroll` statement. - [`#pragma unroll N` in CUDA](https://devblogs.nvidia.com/parallelforall/new-compiler-features-cuda-8/) - [`#pragma unroll N` in OpenCL](https://www.khronos.org/registry/OpenCL/extensions/nv/cl_nv_pragma_unroll.txt) - Unavailable for OpenAcc,...