Matthias Springer's Homepage

Matthias Springer

Deep Learning Compiler Engineer at NVIDIA

Previously: Software Engineer at Google
Previously: Programming Research Group, Tokyo Institute of Technology (Ph.D.)
Previously: Hasso Plattner Institute, University of Potsdam (M.Sc., B.Sc.)
Previously: University of California, San Diego (Visiting Student)

Profile

I am an ML compiler compiler engineer at NVIDIA. I have been working with and contributing to the open-source MLIR compiler project since 2021. My research at university was concerned with code generation, domain-specific languages (DSLs), programming models and memory access optimizations for SIMD/GPU architectures. I am passionate about compilers, high-performance computing (especially GPUs), programming language design and parallel computing.

Curriculum Vitae

Publications

matthias-springer on GitHub

MLIR: Mixed-Level Intermediate Representation (LLVM Project)

LLVM Dev Meeting 2024

Matthias Springer. The State of Pattern-Based IR Rewriting in MLIR (Slides). In 2024 LLVM Developers' Meeting (Technical Talk). Santa Clara, CA, USA. October 2024.

EuroLLVM 2024

Matthias Springer. Computing Bounds of SSA Values in MLIR (Slides, Video Recording). In 2024 European LLVM Developers' Meeting (Technical Talk). Vienna, Austria. April 11, 2024.

LLVM Dev Meeting 2023

Matthias Springer and Martin Erhart. MLIR Bufferization: From Tensors to MemRefs (Slides). In 2023 LLVM Developers' Meeting (Tutorial). Santa Clara, CA, USA. October 12, 2023.

LCPC 2022

Nicolas Vasilache, Oleksandr Zinenko, Aart J. C. Bik, Mahesh Ravishankar, Thomas Raoux, Alexander Belyaev, Matthias Springer, Tobias Gysi, Diego Caballero, Stephan Herhut, Stella Laurenzo and Albert Cohen: Structured Operations: Modular Design of Code Generators for Tensor Compilers. In Languages and Compilers for Parallel Computing (LCPC) 2022, Lecture Notes in Computer Science, vol 13829. Springer, Cham.

Tech Report

Nicolas Vasilache, Oleksandr Zinenko, Aart J. C. Bik, Mahesh Ravishankar, Thomas Raoux, Alexander Belyaev, Matthias Springer, Tobias Gysi, Diego Caballero, Stephan Herhut, Stella Laurenzo and Albert Cohen: Composable and Modular Code Generation in MLIR: A Structured and Retargetable Approach to Tensor Compiler Construction. arXiv 2202.03293, 2022.

Open Design Meeting

Matthias Springer and Nicolas Vasilache. One-Shot Function Bufferization of Tensor Programs (Slides, Video Recording). MLIR Open Design Meeting (virtual). January 13, 2022.

RFCs

Ikra: Object-oriented High Performance Computing on GPUs and CPUs.
(with Hidehiko Masuhara)

Ikra-Ruby: Array-based GPU Computing in Ruby

Ikra-Cpp: A C++/CUDA DSL for Object-oriented Programming

DynaSOAr: A CUDA framework and dyn. mem. allocator for Single-Method Multiple-Objects (SMMO) Applications

Doctoral Thesis

Matthias Springer. Memory-Efficient Object-Oriented Programming on GPUs (Defense Slides, Exam Slides). Programming Research Group, Department of Mathematical and Computing Sciences, Graduate School of Information Science and Engineering, Tokyo Institute of Technology.

ISMM 2019

Matthias Springer and Hidehiko Masuhara. Massively Parallel GPU Memory Compaction (Presentation). In Proceedings of the 2019 ACM SIGPLAN International Symposium on Memory Management (ISMM), Phoenix, AZ, USA, June 23, 2019. Best Student Presentation Award.

ACM SRC

Matthias Springer. CompactGpu: Massively Parallel Memory Defragmentation on GPUs (Extended Abstract, Poster, Presentation). ACM Student Research Competition at ACM SIGPLAN Conference on Programming Language Design and Implementation (PLDI) 2019. Phoenix, AZ, USA, June 22, 2019. First Place in Graduate Category.

ECOOP 2019

Matthias Springer and Hidehiko Masuhara. DynaSOAr: A Parallel Memory Allocator for Object-oriented Programming on GPUs with Efficient Memory Access (Artifact, Poster, Presentation). In Proceedings of the European Conference on Object-oriented Programming (ECOOP) 2019, London, United Kingdom, July 15-19, 2019.

ACM SRC

Matthias Springer. SoaAlloc: Accelerating Single-Method Multiple-Objects Applications on GPUs (Extended Abstract, Poster, Presentation, SRC Finals 2019 Paper). ACM Student Research Competition at ACM SIGPLAN Conference on Systems, Programming, Languages and Applications: Software for Humanity (SPLASH) 2018, Boston, MA, USA, November, 2018. First Place in Graduate Category.

ARRAY 2018

Matthias Springer, Yaozhu Sun and Hidehiko Masuhara. Inner Array Inlining for Structure of Arrays Layout (Presentation). In Proceedings of the 5th ACM SIGPLAN International Workshop on Libraries, Languages, and Compilers for Array Programming (ARRAY) 2018, co-located with PLDI 2018, Philadelphia, PA, USA, June 19, 2018, ACM DL.

WPMVP 2018

Matthias Springer and Hidehiko Masuhara. Ikra-Cpp: A C++/CUDA DSL for Object-oriented Programming with Structure-of-Arrays Data Layout (Presentation). In Proceedings of the 4th Workshop on Programming Models for SIMD/Vector Processing (WPMVP) 2018, co-located with PPoPP 2018, Vienna, Austria, February 24, 2018, ACM DL.

ACM SRC

Matthias Springer. A C++/CUDA DSL for Object-oriented Programming with Structure-of-Arrays Data Layout (Extended Abstract, Poster, Presentation). ACM Student Research Competition at International Symposium on Code Generation and Optimization (CGO) 2018, Vienna, Austria, February 25, 2018. Third Place in Graduate Category.

(unreviewed)

Matthias Springer and Hidehiko Masuhara. An Embedded C++ DSL for Object-oriented Programming with Structure-of-Arrays Layout (Poster). 5th Google Compiler & Prog. Language Summit, Munich, Germany, December 6, 2017.

ARRAY 2017

Matthias Springer, Peter Wauligmann and Hidehiko Masuhara. Modular Array-based GPU Computing in a Dynamically-typed Language (Presentation). In Proceedings of the 4th ACM SIGPLAN International Workshop on Libraries, Languages, and Compilers for Array Programming (ARRAY) 2017, co-located with PLDI 2017, Barcelona, Spain, June 18, 2017, ACM DL.

PPL 2017

Matthias Springer, Peter Wauligmann and Hidehiko Masuhara. Iterative Stencil Computations in Ruby on GPUs (Poster). 19th JSSST Workshop on Programming and Programming Languages (第19回プログラミングおよびプログラミング言語ワークショップ), Usui-shi, Yamanashi-ken, Japan, March 8, 2017.

ARRAY 2016

Matthias Springer, and Hidehiko Masuhara. Object Support in an Array-based GPGPU Extension for Ruby (Presentation). In Proceedings of the 3rd ACM SIGPLAN International Workshop on Libraries, Languages, and Compilers for Array Programming (ARRAY) 2016, co-located with PLDI 2016, Santa Barbara, CA, USA, June 14, 2016, ACM DL.

PPL 2016

Matthias Springer and Hidehiko Masuhara. Leveraging Object-oriented Abstractions in a Ruby-to-CUDA JIT Translator (Poster). 18th JSSST Workshop on Programming and Programming Languages (第18回プログラミングおよびプログラミング言語ワークショップ), Tamano-shi, Ookayama-ken, Japan, March 7, 2016.

Dart2Java: A source-to-source transpiler from Dart to Java, focusing on generic type specialization.
(Software Engineering Intern Project, with Andrew Krieger and Stanislav Manilov)

Source Code, Presentation

ICOOOLPS 2017

Matthias Springer, Andrew Krieger, Stanislav Manilov and Hidehiko Masuhara. dart2java: Running Dart in Java-based Environments (Presentation). In Proceedings of the Workshop on Implementation, Compilation, Optimization of Object-Oriented Languages, Programs and Systems (ICOOOLPS) 2017, co-located with ECOOP 2017, Barcelona, Spain, June 19, 2017, ACM DL.

Matriona: A module system for the Smalltalk programming languages, based on class nesting and class parameterization, with a hierarchical name lookup mechanism, and inspired by Newspeak.
(with Fabio Niephaus, Robert Hirschfeld and Hidehiko Masuhara)

Source Code

Master's Thesis

Matthias Springer. Nested Class Modularity in Squeak/Smalltalk (Disputation Slides). Software Architecture Group, Hasso Plattner Institute, University of Potsdam. August 17, 2015.

MODULARITY 2016

Matthias Springer, Fabio Niephaus, Robert Hirschfeld and Hidehiko Masuhara. Matriona: Class Nesting with Parameterization in Squeak/Smalltalk (Presentation). In Proceedings of the Conference on Modularity 2016, Málaga, Spain, March 14-17, 2016, ACM DL.

LASSY 2016

Matthias Springer, Hidehiko Masuhara and Robert Hirschfeld. Hierarchical Layer-based Class Extensions in Squeak/Smalltalk (Presentation). In Proceedings of the Workshop on Live Adaptation of Software SYstems (LASSY) 2016, co-located with Modularity 2016, Málaga, Spain, March 14, 2016, ACM DL.

Context-oriented Programming: A technique similar to aspect-oriented programming for modularizing heterogeneous cross-cutting concerns. Various projects/ideas.

COP 2016

Matthias Springer, Hidehiko Masuhara and Robert Hirschfeld. Classes as Layers: Rewriting Design Patterns with COP - Alternative Implementations of Decorator, Observer, Visitor (ECOOP 2016 Poster, Presentation). In Proceedings of the Workshop on Context-oriented Programming (COP) 2016, co-located with ECOOP 2016, Rome, Italy, July 19, 2016, ACM DL.

JIP
Matthias Springer, Hidehiko Masuhara and Robert Hirschfeld. A Layer-based Approach to Hierarchical Dynamically-scoped Open Classes. In IPSJ (Information Processing Society of Japan) Journal of Information Processing, Vol. 25 (Mar. 2017), pp. 296-307. Presented at IPSJ-PRO Matsumoto Summer Workshop on Parallel/Distributed/Cooperative Processing (SWoPP 2016) (Presentation)
ContextAmber: A library for context-oriented programming, implemented in Amber Smalltalk.

(with Jens Lincke and Robert Hirschfeld)

ContextAmber with Athens Demo, Source Code

COP 2015
Matthias Springer, Jens Lincke and Robert Hirschfeld. Efficient Layered Method Execution in ContextAmber (Presentation). In Proceedings of the Workshop on Context-oriented Programming (COP) 2015, co-located with ECOOP 2015, Prague, Czech Republic, July 5, 2015, ACM DL.

GQ-Fast: A database system for relationship queries (graph analytics queries).

(with Chunbin Lin, Benjamin Mandel and Yannis Papakonstantinou)

arXiv Article, Project Website

VLDB 2017

Chunbin Lin, Benjamin Mandel, Yannis Papakonstantinou and Matthias Springer. Fast In-Memory SQL Analytics on Typed Graphs. In Proceedings of the VLDB (Very Large Data Bases) Endowment, Volume 10, 2016, Munich, Germany, Aug 28 - Sept 1, 2017, ACM DL.

Implementation of the Athens vector graphics library in Amber Smalltalk, a Smalltalk execution environment running entirely in the web browser, using HTML5 Canvas. Developed a Morphic-like framework for building user interfaces on top of Athens. (Google Summer of Code 2013 project): Live Demo, Morphic Demo, Project Development Blog, Source Code, Video Demo

MagLev: A Ruby implementation running on top of a GemStone/S 64 Smalltalk implementation. Worked on support for Ruby singleton classes and various smaller optimizations. Source Code

(Bachelor's Project, supervised by Software Architecture Group, supported by GemTalk Systems)
- MagLev Database Explorer: An IDE running entirely in a web browser for exploring Ruby/Smalltalk objects persisted in a GemStone/S 64 image, writing Ruby/Smalltalk code, and debugging Rails/Sinatra applications interactively. Built with Amber Smalltalk, Ruby on Rails, and Twitter Bootstrap. Source Code, Video Demo
Bachelor's Thesis

Matthias Springer. Inter-language Collaboration in an Object-oriented Virtual Machine (arXiv Article). Software Architecture Group, Hasso Plattner Institute, University of Potsdam. July 1, 2013.