ImageCL: An Image Processing Language for Performance Portability on Heterogeneous Systems

TL;DR

ImageCL is a high-level language and compiler that simplifies writing portable, high-performance image processing code for heterogeneous systems by automating optimizations and tuning for different hardware devices.

Contribution

We introduce ImageCL, a domain-specific language and source-to-source compiler that automates performance tuning and enhances portability across diverse hardware platforms.

Findings

Achieved up to 4.57x speedup over existing solutions.

Generated optimized code for multiple devices from a single source.

Demonstrated effectiveness on three image processing benchmarks.

Abstract

Modern computer systems typically conbine multicore CPUs with accelerators like GPUs for inproved performance and energy efficiency. However, these sys- tems suffer from poor performance portability, code tuned for one device must be retuned to achieve high performance on another. Image processing is increas- ing in importance , with applications ranging from seismology and medicine to Photoshop. Based on our experience with medical image processing, we propose ImageCL, a high-level domain-specific language and source-to-source compiler, targeting heterogeneous hardware. ImageCL resembles OpenCL, but abstracts away per- formance optimization details, allowing the programmer to focus on algorithm development, rather than performance tuning. The latter is left to our source-to- source compiler and auto-tuner. From high-level ImageCL kernels, our source- to-source compiler can generate multiple OpenCL implementations with different optimizations applied. We rely on auto-tuning rather than machine models or ex- pert programmer knowledge to determine which optimizations to apply, making our tuning procedure highly robust. Furthermore, we can generate high perform- ing implementations for different devices from a single source code, thereby im- proving performance portability. We evaluate our approach on three image processing benchmarks, on different GPU and CPU devices, and are able to outperform other state of the art solutions in several cases, achieving speedups of up to 4.57x.