Fix out-of-bounds-errors in GPU Comm schemes + add tests for MemcpyPackInfo
- When using incomplete stencils, accesses into the
streams_array in UniformGPUScheme and NonuniformGPUScheme used to go out-of-bounds for higher-order velocities. This is fixed by usingStencil::idxto translate the direction to a linear index. - Extend API of
MemcpyPackInfoto permit customization of the number of ghost layers to be communicated - Add a test app for UniformGpuScheme + MemcpyPackInfo testing this (previously untested) pack info for several stencils and GL counts
- Adapt CMake definitions of GPU test suite to run with HIP also