alpaka, LLAMA and More - Solutions for Exascale Performance Portability

Heterogeneous hardware landscapes will define the Exascale era. At the same time, keeping scientific libraries and applications portable across different hardware setups while maintaining high performance is no trivial matter. Vendor-provided programming platforms often cannot target accelerators from other vendors, and different hardware types like CPUs, GPUs and FPGAs require differently tuned algorithms for optimal performance.

A solution to these issues can be found in abstraction layers that provide the user with a single programming interface while still maintaining portability and performance. In this talk, we introduce the Caravan HPC ecosystem. With the alpaka abstraction library for accelerator programming at its core and many sibling libraries for related use cases --- such as the memory access abstraction layer LLAMA or the C++ primitives library vikunja --- the Caravan ecosystem is an ideal choice for scientists and programmers setting out to tackle the challenges of the Exascale era.