High-level Pattern-aware Parallel Software Development

This PhD project tackles a critical challenge in modern computing: enabling developers to write, optimise, and debug code for increasingly complex parallel hardware. Current hardware advancements are outpacing the capabilities of mainstream programming tools, threatening software reliability and the massive investments in hardware-software ecosystems.

Software development based on parallel patterns, where programmers use high-level algorithmic constructs to abstract away the hardware complexity, is our best hope for tackling this software crisis. However, to truly benefit from pattern-based programming, pattern-based software must be well-optimised and easy to maintain, which is not the case right now because existing software development tools cannot fully understand the high-level pattern semantics.

This project asks: What if we could retain and utilise pattern semantics throughout the software toolchain, linking the programmer’s intent directly to the executable binary? To answer this, we will develop novel compiler analysis and optimisation techniques powered by machine learning. We will create new ways to map low-level debugging and profiling data to the programmer’s strategic intentions, enabling developers to interact with hardware efficiently and intuitively. If successful, parallel software development will become faster, simpler, and significantly more reliable.