Yang Gao
HPC Director at SenseTime 商汤科技
Beijing, China
Summary
👤
Senior🎓
Top SchoolYang Gao is an HPC director based in Beijing with 11 years of experience leading cross-platform performance teams across x86, ARM, GPU and FPGA ecosystems. He heads SenseBrain’s x86/ARM, GPU and FPGA groups and is the driving force behind the Parrots Primitive Library (PPL), a high-performance deep learning inference library optimized for ARM, x86 and mobile GPUs. His background includes building HPC architectures and performance-tuning compute-dense systems at Baidu and Alibaba, and low-level benchmarking work such as implementing SIMD/AVX/FMA/AVX-512 paths and SMTL-based parallelism in the cpufp project. Yang combines hands-on assembly/ISA-level optimization with team leadership across Beijing, Shanghai and Shenzhen, and has practical experience tuning for a broad set of accelerators from NVIDIA and AMD GPUs to mobile GPUs and DSPs. He is known for squeezing peak floating-point performance out of hardware and translating those gains into production-grade inference throughput.
12 years of coding experience
2 years of employment as a software developer
Master Degree(MS), Computer Science, Master Degree(MS), Computer Science at Graduate University of Chinese Academy of SciencesBachelor's degree, Computer Science, Bachelor's degree, Computer Science at Beijing University of Technology
Github Skills (13)
c1710
avx10
benchmarking10
assembly10
c1110
sse10
x86-6410
multithreading10
assemble10
assembler10
x8610
float329
floating-point9
Programming languages (2)
Github contributions (5)
pigirons/cpufp
Feb 2016 - Dec 2022
A CPU tool for benchmarking the peak of floating points
Role in this project:
Performance Engineer
Performance EngineerContributions:15 commits, 17 PRs, 103 pushes in 6 years 10 months
Contributions summary:Yang primarily focused on optimizing CPU performance by implementing and integrating instruction set architectures (ISAs) like SSE, AVX, FMA, and AVX-512. They developed and integrated a static multi-threading library (SMTL) to manage parallel execution for floating-point benchmarks. The user's contributions include adding and refining benchmark implementations, and correcting a multi-threading bug to improve the benchmark's accuracy and efficiency. Through these commits, the user aimed to benchmark peak floating-point performance.
benchmarkingfloating-point
pigirons/sgemm_hsw
Sep 2014 - Jun 2021
This is an implementation of sgemm_kernel on L1d cache.
Contributions:8 commits, 4 pushes in 6 years 10 months
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.