Robert Hundt

Distinguished Engineer @ Google

About

I am the Senior Tech Lead for AI/ML Compilers and Performance at Google. Currently reimagining PyTorch / TPU (it will be awesome). Among other things, I have started and TL/M'ed many compiler projects at Google, such as an x86 server compiler optimization effort and an ARM compiler for Android (both gcc-based). I used Open64 for inter-procedural optimization and hunted for Javascript memory leaks with the Closure compiler. My team built the open-source LLVM-based CUDA compiler GPUCC and a compiler for a RISC-like network processor (Lanai). I built and led the whole TPU SW team, which included the XLA compiler, related runtime(s), performance, and tools, and co-started the Cloud TPU and PyTorch/XLA projects. Recently I did the XLS toolchain for High-Level Synthesis. I also wrote about quantum computing with focus on algorithms and their simulation (buy the book, now in its 2nd edition ;-). Before joining Google, I worked for half a decade on HP's Itanium compilers, focusing on inter-procedural optimization, loop optimizations, and automatic parallelization. All posts here are my own!

Country

United States

City

Mountain View

Industry

Computer Software

Skill

Fire Eating, Performance Analysis, Compilers, Optimization, Analysis, X86, Debugging, High Performance Computing, Debuggers, C++, Assembly, C, Python, Software Development, Object Oriented Design, Java, Software Engineering, GCC, Compiler Development, System Software

Experience

Distinguished Engineer

Google

2018-4 - Present · 8 yrs 6 mos

Sunnyvale, CA

Senior Tech Lead (IC, again) for ML Compilers, runtime, performance. OpenXLA. ODML. Prior: Platforms Infrastructure Team (Mgr, 150+ SWEs). - EDA Infra, - BMC and in-band mgmt, - SW quality, HW reliability, - ML for Platforms Accelerated HW Synthesis - XLS - HLS toolchain and Co-Design for FPGA, ASIC

Principal Engineer

Google

2013-11 - 2018-4 · 4 yrs 6 mos

Mountain View, CA

SW Lead for Google Cloud TPU - a monster ML Supercomputer SW Lead for Google TPU - Tensor Processing Unit Uber TL for XLA - a TensorFlow JIT and AOT Compiler for CPU, GPU, TPU M for Platforms Performance Team TLM for Open-Source CUDA Compiler TLM for Open-Source Lanai LLVM backend Managing Teams of 80+

Senior Staff Software Engineer

Google

2008-10 - 2013-10 · 5 yrs 1 mo

Mountain View, CA, 94043

Datacenter and Compiler Research TLM for GMail Performance Team (Backend, JS, Chrome) TLM for Datacenter Performance Team TLM for Google-wide Profiling Infrastructure TLM for x86 Server Compiler Optimization (FDO, LIPO, MAO, more) TLM for ARM/Android Compiler Optimization

Staff Software Engineer

Google

2007-10 - 2008-9 · 1 yr

Mountain View, CA, 94043

TLM for x86 Compiler Optimization (gcc) Reached >12% geomean in first 9 month

Senior Software Engineer

Google

2007-1 - 2007-9 · 9 mos

Mountain View, CA, 94043

Tried to use open64 for inter-procedural analysis and optimization Meh

Book Author

Cambridge University Press

2020-3 - 2022-3 · 2 yrs 1 mo

Get a copy: https://www.amazon.com/dp/1009098179

Compiler Engineer

Hewlett Packard Enterprise

2000-1 - 2007-1 · 7 yrs 1 mo

Member of high-level optimizer team for HP's Itanium platforms. Infrastructure, scalar optimizations, loop optimizations, and mainly inter-procedural optimizations, such as data layout transformations and advice. Automatic parallelization, automatic compiler bug triaging. Design and implementation of HP Caliper, HP's performance analysis tool. Technical authority for dynamic binary instrumentation on HP's Itanium platforms.

Senior Software Engineer

Spectra Precision terraSat

1992-1 - 2000-1 · 8 yrs 1 mo

Design, Implementation and hands-on management for TerraSat's main product line of the time - GeoGenius, a static GPS data processing and visualization engine. And lots and lots of other stuff. Co-owner of the company (25 developers). Fun artifact: http://sup.xenya.si/sup/info/magellan(thalesnavigation)/reference_manuals/AOS/AOS.pdf