Software Engineer III, TPU Compiler

Experience driving progress, solving problems, and mentoring more junior team members; deeper expertise and applied knowledge within relevant area.

Minimum qualifications

• Bachelor’s degree or equivalent practical experience.
• 2 years of experience with software development in one or more programming languages, or 1 year of experience with an advanced degree.
• 1 year of experience with ML infrastructure (e.g., model deployment, model evaluation, optimization, data processing, debugging).
• Experience in C++.
• Experience with performance, systems data analysis, visualization tools, or debugging.

Preferred qualifications

• Master's degree or PhD in Computer Science or related technical fields and 5 years of experience with data structures/algorithms.
• 2 years of experience with data structures or algorithms.
• Experience in Machine Learning and High Performance Computing (HPC).
• Experience optimizing distributed programs at large-scale and experience with Compilers and Compiler construction.
• Excellent debugging and programming concurrent/parallel computations, and excellent working on accelerators including but not limited to VLIW and Vector machines, GPUs, or DSPs.

Our team develops the Accelerated Linear Algebra (XLA) Tensor Processing Units (TPU) parallelizing compiler used to partition, optimize, and run large-scale machine learning models across multiple TPU accelerators for internal Google, Google DeepMind and external customers. The software stack includes the partitioner to share work across multiple TPUs, scheduling optimizations, and code generation.

Responsibilities

• Write product or system development code for the TPU compiler (in C++).
• Participate in, or lead design reviews with peers and stakeholders to decide amongst available technologies.
• Contribute to a compiler which scales-out machine learning models across accelerators like TPU/Graphics Processing Unit (GPU) at Google and Cloud.
• Conduct static and runtime performance analysis of important large-scale production models.
• Design and implement performance optimizations and critical features, which increase the velocity of important production teams.