info_outline
X Note: By applying to this position you will have an opportunity to share your preferred working location from the following:
Austin, TX, USA; Mountain View, CA, USA.
Minimum qualifications: - Bachelor's degree in Computer Science or equivalent practical experience.
- 8 years of experience with software development in one or more programming languages (e.g., Python, C, C , Java, JavaScript).
- 5 years of experience with compiler development or construction.
- Experience with low level virtual machines (LLVM) development, modifying LLVM as a contributor.
Preferred qualifications: - Master's degree or PhD in Computer Science, or a related technical field.
- Experience developing compilers using the Clang/LLVM/Multi-Level Intermediate Representation (MLIR) infrastructure, specifically for Very Long Instruction Word (VLIW) processors or Machine Learning (ML) accelerators.
- Experience in the design or management of large-scale systems and High-Performance Computing (HPC) environments.
About the jobThe Machine Learning Compiler (MPACT) team in Google Research is developing open-source, retargetable compiler infrastructure in the Low Level Virtual Machines (LLVM) and MLIR frameworks to streamline iterative processor and system co-design for the Google Tensor Processing Unit (TPU) family of processors, and to enable C/C /Cuda High Performance Computing (HPC) codes to run efficiently on TPUs.
US: $207000 - $301000 (USD) 20% bonus target equity benefits
Learn more about benefits at Google .
Responsibilities - Build, release, and support Clang/LLVM/MLIR-based compilers, debuggers, simulators, and performance analysis tools for Google TPU accelerators.
- Optimize the TPU software ecosystem to support conventional High-Performance Computing (HPC) using C, C , and Compute Unified Device Architecture (CUDA).
- Write TPU-specific backend code generation and optimization components, ensuring support for new hardware designs.
- Create MLIR components to translate and optimize OpenHLO, CUDA, and PyTorch programs for efficient TPU execution.