Array streaming for array programming

Abstract

A barrier to efficient array programming, for example in Python/NumPy, is that algorithms written as pure array operations completely without loops, while most efficient on small input, can lead to explosions in memory use. The present paper presents a solution to this problem using array streaming, implemented in the automatic parallelisation high-performance framework Bohrium. This makes it possible to use array programming in Python/NumPy code directly, even when the apparent memory requirement exceeds the machine capacity, since the automatic streaming eliminates the temporary memory overhead by performing calculations in per-thread registers. Using Bohrium, we automatically fuse, stream, JIT-compile, and execute NumPy array operations on GPGPUs without modification to the user programs. We present performance evaluations of three benchmarks, all of which show dramatic reductions in memory use from streaming, yielding corresponding improvements in speed and utilisation of GPGPU-cores. The fusion step is implemented using the theoretical framework presented in Kristensen et al. (2016), using a streaming-maximising cost function. The streaming-enabled Bohrium effortlessly runs programs on input sizes several orders of magnitude beyond sizes that crash on pure NumPy due to exhausting system memory.