Menachem Adelman
CPU Architect at Google
Israel
Summary
👤
Senior🎓
Top SchoolMenachem Adelman is a CPU architect with a decade of experience at the intersection of algorithms, software and hardware, currently shaping processor design and systems at Google. Previously at Intel he led machine-learning focused ISA and microarchitecture features and implemented high-performance compute kernels, combining hands-on software optimization with architecture modelling. His open-source contributions include performance engineering for the widely used libxsmm library, adding new x86 instruction support and processor-specific codepaths to accelerate dense/sparse matrix transforms and VNNI workloads. With dual summa cum laude degrees in electrical engineering and physics from Technion, he brings rigorous academic training to practical CPU and ML acceleration problems. Notably, he blends deep low-level instruction-set expertise with real-world kernel optimizations that directly improve ML throughput on modern processors.
10 years of coding experience
6 years of employment as a software developer
Master of Science (M.Sc.), Electrical Engineering, summa cum laude, Master of Science (M.Sc.), Electrical Engineering, summa cum laude at Technion - Israel Institute of Technology
Github Skills (21)
avx10
matrix-multiplication10
c1110
c1710
performance-optimization10
simd10
vectorization9
deep-learning9
jit9
data-structures8
data-structure8
machine-learning8
algorithms8
algorithm8
intel8
Programming languages (2)
Github contributions (5)
libxsmm/libxsmm
Dec 2020 - Apr 2021
Library for specialized dense and sparse matrix operations, and deep learning primitives.
Role in this project:
Back-end Developer / Performance Engineer
Back-end Developer / Performance EngineerContributions:2 reviews, 36 commits, 13 PRs in 4 months
Contributions summary:Menachem focused on optimizing the `libxsmm/libxsmm` library, which is designed for high-performance matrix operations and deep learning primitives. The commits involved adding support for new x86 instructions (e.g., `VBROADCASTI64X2`, `VBROADCASTI32X4`, `VINSERTI32X4`) and introducing specialized code paths for specific processor architectures (CLX, SPR) to improve performance, particularly for VNNI-based matrix transformations. The user's work included refactoring and enhancing the core routines used in matrix transpositions, specifically those that are part of the library's performance critical compute kernels.
bfloat16avxsimdlapackavx512
adelmanm/libxsmm
Dec 2020 - Jul 2022
Library targeting Intel Architecture for specialized dense and sparse matrix operations, and deep learning primitives.
Contributions:2 PRs, 25 pushes, 2 branches in 1 year 7 months
cudasparse-matrixdeep-learningsparseintel
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.