ACS: Concurrent Kernel Execution on Irregular, Input-Dependent Computational Graphs

Abstract

GPUs are widely used to accelerate many important classes of workloads today. However, we observe that several important emerging classes of workloads, including simulation engines for deep reinforcement learning and dynamic neural networks, are unable to fully utilize the massive parallelism that GPUs offer. These applications tend to have kernels that are small in size, i.e., have few thread blocks that do not saturate compute resources. Executing independent kernels concurrently is a promising approach to improve parallelism and utilization. However, this inter-kernel concurrency is difficult to leverage in such workloads with existing approaches: First, the inter-kernel dependencies and computational graph are input-dependent and vary each time the application is executed. Second, the computational graphs tend to be irregular, requiring fine-grain scheduling and synchronization; thus incurring significant synchronization overheads if kernel execution is parallelized. In this work, we propose ACS, a framework that enables lightweight detection of inter-kernel dependencies and low overhead kernel scheduling at runtime. The key idea behind ACS is to perform inter-kernel dependency checks for a small window of kernels at runtime, similar to out-of order instruction scheduling. This enables concurrent execution of kernels in applications whose computational graphs are input dependent and require fine-grained scheduling. We propose ACS-SW, a software-only open-source implementation of ACS and ACS-HW, a hardware-software cooperative implementation. ACS-HW further reduces synchronization overheads by reducing communication between the CPU and GPU. We evaluate ACS for deep RL simulation and dynamic DNNs on both real hardware and a GPU simulator. We demonstrate speedups of up to 2.19x (1.56x on average) by improving GPU utilization with concurrent kernel execution.