Project Background

In collaboration with a Chinese automotive OEM's autonomous driving team, we executed a project on software performance modeling and code optimization. Our focus was to drastically reduce the system's CPU utilization and end-to-end latency, directly boosting its technical competitiveness.

Consulting Process

Partnering with the client's experts, we diagnosed performance bottlenecks in their autonomous driving system and defined a clear optimization roadmap. Our technical work focused on decoupling algorithms from the platform scheduler for better modularity and boosting performance by optimizing or replacing critical algorithms. We also reduced computational complexity and increased concurrency to hit performance goals. A key part of the engagement was upskilling the client's team through hands-on training in performance modeling and high-performance coding, helping them build a fast-feedback performance engineering workflow.

Collaborative Achievements

Through this consulting partnership, we successfully achieved the optimization targets for CPU load and end-to-end latency. Furthermore, we helped the client establish a comprehensive, end-to-end software performance engineering capability, covering the entire lifecycle from design and implementation to testing and feedback.

Project Background

The client's codebase was becoming bloated and hard to maintain, prompting a migration and refactoring effort from C to C++. Lacking experience, their initial attempts ran into major issues: while they managed to shrink the code size, performance dropped dramatically, blocking the product from being deployed.

Consulting Process

Partnered with the client's experts to thoroughly analyze and diagnose their system, pinpointing performance bottlenecks and improper C++ usage, helping to identify performance bottlenecks and instances of improper C++ usage. Also implemented performance testing tools and established unit testing practices. Based on the client's real-world scenarios, we abstracted foundational components and assisted them in building a core component library. To address specific project pain points, we conducted targeted training for their developers, imparting in-depth C++ knowledge and best practices. At the architectural level, we adjusted the client's system to ensure it achieved an optimal balance of performance, maintainability, and testability. In terms of standards, we helped the client refine their coding standards by removing incorrect or unenforceable rules while optimizing existing ones, ensuring that the standards remained applicable for the project. Furthermore, we fine-tuned their automated code analysis tools and rules to guarantee that these standards were correctly and consistently enforced.

Collaborative Achievements

The pilot project doubled the company’s performance and cut the code size to just one-third, surpassing all targets and dramatically increasing the team's confidence. In the follow-up phase, we upskilled their developers, built a core C++ library, and successfully refactored the rest of the system.