A Taxonomy of Modern GPGPU Programming Methods: On the Benefits of a Unified Specification

• Published in: IEEE transactions on computer-aided design of integrated circuits and systems Vol. 41; no. 6; pp. 1649 - 1662
• Main Authors: Capodieci, Nicola, Cavicchioli, Roberto, Marongiu, Andrea
• Format: Journal Article
• Language: English
• Published: New York IEEE 01-06-2022; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Computer architecture; Encoding; General-purpose GPU (GPGPU); Graphics processing units; Hardware; Kernel; Knobs; parallel programming tools; Pipelining (computers); Programming; Semantics; Specifications; State-of-the-art reviews; Taxonomy
• ISSN: 0278-0070; 1937-4151
• DOI: 10.1109/TCAD.2021.3082863

Several application programming interfaces (APIs) and frameworks have been proposed to simplify the development of general-purpose GPU (GPGPU) applications. GPGPU application development typically involves specific customization for the target operating systems and hardware devices. The effort to port applications from one API to the other (or to develop multitarget applications) is complicated by the availability of a plethora of specifications, which in essence offers very similar underlying functionality. In this work we provide an in-depth study of six state-of-the-art GPGPU APIs. From these we derive a taxonomy of the common semantics and propose a unified specification. We describe a methodology to translate this unified specification into different target APIs. This simplifies cross-platform application development and provides a clean framework for benchmarking. Our proposed unified specification is called GPGPU unified specification and translation (GUST) and it captures common functionality found in compute-only APIs (e.g., compute unified device architecture and open computing language), in the compute pipeline of traditional graphic-oriented APIs (e.g., open graphic language and Direct3D11) and in last-generation bare-metal APIs (e.g., Vulkan and Direct3D12). The proposed translation methodology solves differences between specific APIs in a transparent manner, without hiding available tuning knobs for compute kernel optimizations and fostering best programming practices in a simple manner.