I build things close to the metal. My interests centre on computer architecture and systems programming — from OS kernels and memory allocators to CUDA kernels and distributed solvers. I care about how software maps onto hardware, and how to extract maximum performance from modern architectures.
Currently at Australian National University, previously Beijing University of Technology (BJUT) and University College Dublin (UCD), where I graduated with First Class Honours.
- 🔭 Studying at ANU · applying for PhD positions
- 🌱 Deep in computer architecture, OS and HPC
- 📫 jingyu.lu@anu.edu.au
🖥️ Systems Programming
|
Concurrent train scheduler with deadlock prevention via resource ordering. Fine-grained mutex strategy minimises lock contention across critical sections. 
|
Heap allocator with explicit free lists and boundary-tag coalescing, paired with a mark-and-sweep GC. Balanced for throughput and fragmentation ratio.
|
Kernel-level extensions to MIT's xv6: system calls, scheduling algorithms, virtual memory management, and file system features — all from scratch.
|
⚡ Parallel & High-Performance Computing
|
GPU solver for advection equations using CUDA shared memory tiling. Warp-aligned data staging cuts global memory bandwidth and maximises on-chip reuse. 
|
Multi-node advection solver with halo exchange. Non-blocking MPI communication pipelines computation and communication across distributed processes.
|
🔐 LLM Safety
|
Contributing to an open-source LLM vulnerability scanner — probing models for jailbreaks, hallucination patterns, and safety boundary failures across multiple attack surfaces. 
|
| Low-level |
|
| Systems |
|
| ML / AI |
|
"Premature optimisation is the root of all evil — but late optimisation is the root of all latency."