High-Performance Deep Learning via a Single Building Block
This work addresses the maintenance and performance challenges in deep learning library development for researchers and practitioners, offering a more streamlined and efficient approach.
The paper tackles the complexity and lack of generalization in deep learning libraries by introducing a batch-reduce GEMM kernel as a single building block, enabling high-performance implementations of RNNs, CNNs, and MLPs in just 3K lines of code that outperform vendor-optimized libraries on multi-node CPU clusters.
Deep learning (DL) is one of the most prominent branches of machine learning. Due to the immense computational cost of DL workloads, industry and academia have developed DL libraries with highly-specialized kernels for each workload/architecture, leading to numerous, complex code-bases that strive for performance, yet they are hard to maintain and do not generalize. In this work, we introduce the batch-reduce GEMM kernel and show how the most popular DL algorithms can be formulated with this kernel as the basic building-block. Consequently, the DL library-development degenerates to mere (potentially automatic) tuning of loops around this sole optimized kernel. By exploiting our new kernel we implement Recurrent Neural Networks, Convolution Neural Networks and Multilayer Perceptron training and inference primitives in just 3K lines of high-level code. Our primitives outperform vendor-optimized libraries on multi-node CPU clusters, and we also provide proof-of-concept CNN kernels targeting GPUs. Finally, we demonstrate that the batch-reduce GEMM kernel within a tensor compiler yields high-performance CNN primitives, further amplifying the viability of our approach.