Abstract

Writing GPU kernel code which optimally exploits parallelism and the GPU architecture is the most challenging and time-consuming aspect of GPU software development. Programmers have to identify algorithms suitable for parallelization and while implementing them reason about deadlocks, synchronization, race conditions, shared memory layout, plurality of state, granularity, throughput, latency and memory bottlenecks. This means that new languages with professional tooling which increase the productivity of GPU software development, whilst retaining the full flexibility of the underlying GPU programming model CUDA or OpenCL, are of tremendous value. In this talk we introduce the upcoming version 2 of Alea.GPU, a high productivity GPU development tool chain for .NET. We show how GPU scripting, dynamic compilation and unique features of the F# language can be leveraged to reduce the development effort for cross platform GPU accelerated applications. Finally we look at our new reactive dataflow model, which simplifies GPU computing further.