HDNN: a cross-platform MLIR dialect for deep neural networks

Abstract This paper presents HDNN, a proof-of-concept MLIR dialect for cross-platform computing specialized in deep neural networks. As target devices, HDNN supports CPUs, GPUs and TPUs. In this paper, we provide comprehensive description of the dialect, outlining how novel approach aims to solve $$P^3$$ P 3 problem parallel programming (portability, productivity, performance). An program is device-agnostic, i.e., only device specifier has be changed run given workload one or another. Moreover, been designed domain-specific language, which ultimately helps productivity. Finally, relies on optimized libraries heavy, performance-critical workloads. evaluated against other state-of-the-art machine learning frameworks all hardware platforms achieving excellent performance. We conclude that ideas concepts used can crucial designing future generation compilers languages overcome challenges forthcoming heterogeneous era.