# William S. Moses PhD Student, MIT CSAIL

I’m a PhD student at MIT studying computer science and getting up to all sorts of mischief. Previously, I attended MIT as an undergraduate and M.Eng in computer science and physics. Before that, I attended Thomas Jefferson High School for Science and Technology (TJHSST) in Northern Virginia.

 [email protected] 703-638-2387 MIT Stata Center 32 Vassar Street, Cambridge MA 02139

## Papers

Tensor Comprehensions: Framework-Agnostic High-Performance Machine Learning Abstractions Vasilache, Nicolas and Zinenko, Oleksandr and Theodoridis, Theodoros and Goyal, Priya and DeVito, Zachary and Moses, William S and Verdoolaege, Sven and Adams, Andrew and Cohen, Albert. .
@article{tc,
title = {Tensor Comprehensions: Framework-Agnostic High-Performance Machine Learning Abstractions},
author = {Vasilache, Nicolas and Zinenko, Oleksandr and Theodoridis, Theodoros and Goyal, Priya and DeVito, Zachary and Moses, William S and Verdoolaege, Sven and Adams, Andrew and Cohen, Albert},
journal = {arXiv preprint arXiv:1802.04730},
year = {2018},
hackernews = {http://news.ycombinator.com/item?id=16377389},
pdf = {https://arxiv.org/pdf/1802.04730.pdf}
}

OpenMPIR: Implementing OpenMP Tasks with Tapir Stelle, George and Moses, William S. and Olivier, Stephen L. and McCormick, Patrick. LLVM-HPC’17.

Optimizing compilers for task-level parallelism are still in their infancy. This work explores a compiler front end that translates OpenMP tasking semantics to Tapir, an extension to LLVM IR that represents fork-join parallelism. This enables analyses and optimizations that were previously inaccessible to OpenMP codes, as well as the ability to target additional runtimes at code generation. Using a Cilk runtime back end, we compare results to existing OpenMP implementations. Initial performance results for the Barcelona OpenMP task suite show performance improvements over existing implementations.

@inproceedings{openmpir,
author = {Stelle, George and Moses, William S. and Olivier, Stephen L. and McCormick, Patrick},
title = {OpenMPIR: Implementing OpenMP Tasks with Tapir},
booktitle = {Proceedings of the Fourth Workshop on the LLVM Compiler Infrastructure in HPC},
series = {LLVM-HPC'17},
year = {2017},
isbn = {978-1-4503-5565-0},
location = {Denver, CO, USA},
pages = {3:1--3:12},
articleno = {3},
numpages = {12},
url = {http://doi.acm.org/10.1145/3148173.3148186},
doi = {10.1145/3148173.3148186},
acmid = {3148186},
publisher = {ACM},
address = {New York, NY, USA},
pdf = {http://c.wsmoses.com/papers/openmpir.pdf}
}

Tapir: Embedding Fork-Join Parallelism into LLVM’s Intermediate Representation, Best Paper Award Schardl, Tao B. and Moses, William S. and Leiserson, Charles E.. PPoPP ’17.

This paper explores how fork-join parallelism, as supported by concurrency platforms such as Cilk and OpenMP, can be embedded into a compiler’s intermediate representation (IR). Mainstream compilers typically treat parallel linguistic constructs as syntactic sugar for function calls into a parallel runtime. These calls prevent the compiler from performing optimizations across parallel control constructs. Remedying this situation is generally thought to require an extensive reworking of compiler analyses and code transformations to handle parallel semantics.

Tapir is a compiler IR that represents logically parallel tasks asymmetrically in the program’s control flow graph. Tapir allows the compiler to optimize across parallel control constructs with only minor changes to its existing analyses and code transformations. To prototype Tapir in the LLVM compiler, for example, we added or modified about 6000 lines of LLVM’s 4-million-line codebase. Tapir enables LLVM’s existing compiler optimizations for serial code – including loop-invariant-code motion, common-subexpression elimination, and tail-recursion elimination – to work with parallel control constructs such as spawning and parallel loops. Tapir also supports parallel optimizations such as loop scheduling.

@inproceedings{tapir,
author = {Schardl, Tao B. and Moses, William S. and Leiserson, Charles E.},
title = {Tapir: Embedding Fork-Join Parallelism into LLVM's Intermediate Representation},
booktitle = {Proceedings of the 22nd ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming},
series = {PPoPP '17},
month = jan,
year = {2017},
isbn = {978-1-4503-4493-7},
location = {Austin, Texas, USA},
pages = {249--265},
numpages = {17},
pdf = {http://c.wsmoses.com/papers/tapir.pdf},
doi = {10.1145/3018743.3018758},
acmid = {3018758},
publisher = {ACM},
address = {New York, NY, USA},
award = {Best Paper Award},
mitnews = {http://news.mit.edu/2017/optimizing-code-compiler-parallel-programs-0130},
hackernews = {http://news.ycombinator.com/item?id=13568585},
blog = {/tapir}
}

How Should Compilers Represent Fork-Join Parallelism? Moses, William S.. Thesis ’17.

This thesis explores how fork-join parallelism, as supported by concurrency platforms such as Cilk and OpenMP, can be embedded into a compiler’s intermediate representation (IR). Mainstream compilers typically treat parallel linguistic constructs as syntactic sugar for function calls into a parallel runtime. These calls prevent the compiler from performing optimizations across parallel control constructs. Remedying this situation is generally thought to require an extensive reworking of compiler analyses and code transformations to handle parallel semantics.

Tapir is a compiler IR that represents logically parallel tasks asymmetrically in the program’s control flow graph. Tapir allows the compiler to optimize across parallel control constructs with only minor changes to its existing analyses and code transformations. To prototype Tapir in the LLVM compiler, for example, the Tapir team added or modi- fied about 6000 lines of LLVM’s 4-million-line codebase. Tapir enables LLVM’s existing compiler optimizations for serial code — including loop-invariant-code motion, commonsubexpression elimination, and tail-recursion elimination — to work with parallel control constructs such as spawning and parallel loops. Tapir also supports parallel optimizations such as loop scheduling.

@mastersthesis{wmoses-meng,
title = {How {S}hould {C}ompilers {R}epresent {F}ork-{J}oin {P}arallelism?},
author = {Moses, William S.},
booktitle = {Master's Thesis},
series = {Thesis '17},
school = {Massachusetts Institute of Technology},
year = {2017},
month = may,
pdf = {http://c.wsmoses.com/papers/wmoses-meng.pdf},
blog = {/tapir}
}

Extreme Multi-Resolution Visualization: A Challenge on Many Levels Balme, Joanna and Brown-Dymkoski, Eric and Guerrero, Victor and Jones, Stephen and Kessler, Andre and Lichtl, Adam and Lung, Kevin and Moses, William and Museth, Ken and Roberson, Nathan and others. SCVis ’15.
@article{spacex15,
title = {Extreme Multi-Resolution Visualization: A Challenge on Many Levels},
author = {Balme, Joanna and Brown-Dymkoski, Eric and Guerrero, Victor and Jones, Stephen and Kessler, Andre and Lichtl, Adam and Lung, Kevin and Moses, William and Museth, Ken and Roberson, Nathan and others},
booktitle = {SuperComputing Visualization Contest 2015},
series = {SCVis '15},
year = {2015},
pdf = {http://c.wsmoses.com/papers/spacex15.pdf}
}

Computational Complexity of Arranging Music Demaine, Erik D. and Moses, William S.. MOVES ’15.

Music has long been an interesting subject of analysis for mathematicians and has led to many interesting questions in music theory and other fields. For the most part, computer scientists have looked into applying artificial intelligence to music and finding algorithms and data structures to solve various problems in music. Prior work on these algorithms often involves computing various properties of music such as the edit distance between two songs or the optimal fingering. These problems tend to be solvable in polynomial time using dynamic programming and have various application such as the music identification service Shazam or operations on RISM, an online music database.

This paper takes an additional step in this direction, asking what sorts of problems in music cannot be efficiently computed. Specifically, this paper asks how various constraints affect the computational complexity of arranging music originally written for one set of instruments down to a single instrument. The paper then applies these results to other domains including musical choreography (such as ice skating and ballet) as well as creating levels for rhythm games (such as Rock Band). We prove that all of the problems are NP-complete, meaning that there is no efficient algorithm to solve them (assuming the standard conjecture that P != NP).

@incollection{moves15,
author = {Demaine, Erik D. and Moses, William S.},
title = {Computational Complexity of Arranging Music},
booktitle = {Revised Papers from MOVES 2015: Mathematics of Various Entertaining Subjects},
series = {MOVES '15},
publisher = {Princeton University Press},
year = {2015},
pdf = {http://c.wsmoses.com/papers/moves15.pdf}
}

Online Adaptive Frequency Hopping Moses, William and Robertson, Andrew and Dell, John. TJHSST ’14.

Adaptive frequency hopping is one way to maximize the utilization of the wireless spectrum. Yet, when the environment itself is changing, the frequency at which the radio senses can become increasingly less optimal. By having the radio create a model of the environment based off of the sensing data, it is possible to achieve high data rates when the spectrum is not being heavily utilized and maintain a low level of interference at times when it is. The radio was modeled both mathematically and run in simulations. The outcomes of these tests were compared with existing standards such as Bluetooth (random frequency hopping) and IEEE 802.22 (fixed sensing rate). In order to evaluate data rate and interference simultaneously, a metric was created that combined them by taking the product of data rate and ( 1 - interference ). Overall, the online adaptive frequency hopper had a 35% increase in the combined metric over the random frequency hopper and 25% increase over the fixed sensing rate radio.

@misc{oafh,
title = {Online Adaptive Frequency Hopping},
author = {Moses, William and Robertson, Andrew and Dell, John},
booktitle = {TJHSST Teknos 2014},
series = {TJHSST '14},
year = {2014},
pdf = {http://c.wsmoses.com/papers/oafh.pdf}
}


## Presentations

How to Use LLVM To Optimize Parallel Programs US LLVM Developer Meeting, Fall 2018, Oct 18, 2018
Adaptive Value Iteration 6.832 Presentations, May 17, 2018
Quanum Computing for the Common Man 8.371 Presentations, May 7, 2018
Tensor Comprehensions Rework Deep Learning Summit Boston 2018, May 24, 2018
Tensor Comprehensions LLVM Workshop at CGO 2018, Feb 24, 2018
Leveraging LLVM to Optimize Parallel Programs US LLVM Developer Meeting, Fall 2017, Oct 18, 2017
Leveraging LLVM to Optimize Parallel Programs NSF Parlay Meeting, Fall 2016, Sep 29, 2017
Tapir: Embedding Fork-Join Parallelism into LLVM IR MIT Masterworks Poster Symposium, Apr 18, 2017
Tapir: Embedding Fork-Join Parallelism into LLVM IR MIT 6.S898 Class Lecture, Apr 2, 2017
Tapir: Embedding Fork-Join Parallelism into LLVM IR, 2nd place speaker MIT EECSCon 2017, Apr 18, 2017
Tapir: Embedding Fork-Join Parallelism into LLVM IR IBM PL Day 2016, Dec 5, 2017
Embedding Fork-Join Parallelism into LLVM IR Compilers for Parallel Computing 2016, Jul 6, 2016
Computational Complexity of Arranging Music Mathematics of Various of Entertaining Subjects (MOVES) 2015, Aug 3, 2015
Syntactic Simplifications for Reducer Hyperobjects Intel Corporation, Jan 22, 2015

## Projects

– TODO: include projects here

## Coursework

### Junior Fall (2016)

 6.828: Operating Systems A graduate-level class on the design and implementation of operating systems. Topics included: locking, virtual memory, file systems, threads, context switches, kernel interrupts, system calls, IPC, and MMIO devices. Over the semester, implemented a functioning multicore operating system called jos, designed to be a simpler version of xv6. 6.867: Machine Learning A graduate-level class in machine learning. Topics included support vector machines, neural networks (including convolutional and deep networks), decision tree, boosting algorithms, Bayesian methods, and Markov models. 6.UAR: Seminar in Undergraduate Research The first semester in a year-long class in methods for advanced research. 21M.359: Interactive Music Systems A combination humanties and computer science subject focusing on developping interactive systems involving audio. Labs included making games using input devices such as the Microsoft Kinect, audio analysis, and developing a simple guitar hero game. 11.011: The Art and Science of Negotiation An advanced class on negotiation skills with an emphasis on methods in mutual gains.

### Junior Spring (2017):

 6.115: Microcomputer Project Laboratory An advanced laboratory subject on analysis and design of embedded systems from capacitors to assembly and MMIO. Emphasizes construction of complete systems, including labs involving a five-axis robot arm, a fluorescent lamp ballast, a tomographic imaging station (e.g., a CAT scan), and a simple calculator. CMS.335: Short Documentary A humanities subject exploring the world of short documentaries. 6.THM: Thesis A class denoting work on my master's thesis on Tapir and related advances in performance engineering parallel programs

### Sophomore Fall (2015):

 6.172: Performance Engineering An advanced undergraduate course in performance engineering covering both the theory and practice behind technologies in parallelism, cache-behavior, among others. 6.002: Circuits Introductory class in circuit design. 8.370: Quantum Computation A graduate-level class in quantum computation and quantum algorithms. 8.07: Advanced Electrodynamics An advanced undergraduate class in electrodynamics focusing on moving systems and the behavior of materials. 24.241: Logic I An undergraduate class on formal logic.

### Sophomore Spring (2016):

 6.854: Advanced Algorithms A graduate-level course in algorithms. Topics include universal/consistent hashing, dimensionality reduction, multi-commodity flow and other generalizations of max-flow, linear programming and duality, gradient descent, interior points, multiplicative weights, semidefinited programming, and compressed sensing. 6.035: Computer Language Engineering An advanced class on computer languages and compilers. 8.962: General Relativity A graduate-level class on general relativity 8.06: Quantum Mechanics III The most advanced undergraduate course in quantum mechanics 4.352: Advanced Video Making An advanced course in video production.

### Freshman Spring (2015):

 6.886: Performance Engineering for Multicore Systems A seminar class in performance engineering with a focus on performing research in the field 6.046: Design and Analysis Algorithms The most advanced undergraduate course in algorithms spanning subjects from Van Emde Boas trees to network flow. 8.S05: Quantum Mechanics II An intermediate course in Quantum Mechanics, using linear algebra as a basis for describing subjects from the quantum harmonic oscillator, to squeeze states, to the fine structure of Hydrogen. 8.044: Statistical Mechanics An class in statistical mechanics using quantum mechanics to build up more macroscopic effects. 21W.789: Communicating With Mobile Technology A class on how to design, build, and market new computer science start-ups made for mobile platforms. 18.075: Advanced Calculus for Engineers (Complex Analysis) Complex analysis and differential equations taught from the perspective of engineering applications.

### Freshman IAP (2015):

 8.223: Classical Mechanics II An advanced course in classical mechanics covering approximations, Lagrangian mechanics, Hamiltonian Mechanics, among other topics. 18.031: System Functions and the Laplace Transform A supplementary class in differential equations covering system functions and the Laplace transform as methods of solving differential equations and modeling systems. 6.179: Introduction to C/C++ (Instructor) An introductory class to C/C++ that I taught with William Qian with over 200 students. Geared towards those already with a programming background, this provided a fast-paced introduction to C/C++ covering more advacned topics such as template metaprogramming. There were weekly algorithmic PSETS (USACO style) and a capstone final project.

### Freshman Fall (2014):

 6.890: Algorithmic Lower Bounds, Fun With Hardness Proofs A graduate-level computer science class on hardness (NP,PSPACE,EXPSPACE,etc). Covered material ranging from 3SAT to APX and constraint logic. Produced final paper with hopes of publishing in the near future. Final Paper Taught by Erik Demaine. 8.04: Quantum Mechanics I An introductory course in quantum mechanics covering the basic experiments such as the double slit experiment through interference, quantum tunneling, operators, and the hydrogen atom. 5.111: Chemistry An introductory course in chemistry covering topics such as equilibrium, acid-base, orbital shapes, crystal field theory, d-orbitals, thermodynamics, and kinetics. 7.012: Biology An introductory course in biology focusing on genetics and neurobiology. Taught by Eric Lander. 21W.035: Science Writing for the Public A writing class focusing on making science accessible to the public. Projects included: a news article and a researched science article. 12.000: Solving Complex Problems (Terrascope) A seminar class focusing on solving the world's energy problem in the next 100 years. Redulted in heavy research in climate change, statistical modeling and predictions of the future, a website with articles on current and future energy issues and solutions and an hour-long presentation followed by a two-hour question and answer session with experts and executives from oil companies. More information on the class including a recording of the presenation can be found here.

### High School (Sept 2010 - June 2014):

 Advanced Mathematical Techniques for Scientists and Engineers Post-mutivariable calculus math course focusing on mathematical methods with important applications to physics and engineering such as Fourier series, linear algebra, introductory complex analysis, the Gamma function, Bessel functions, Legendre polynomials, and the Riemann Zeta function. Includes a brief introduction to quantum mechanics and statistical mechanics. Quantum Physics and Electrodynamics An introductory course in quantum mechanics covering Stern-Gerlach machines, entanglement, tunneling as well as a more rigorous treatment of E&M through the use of fourier transforms and Greens functions as well as an introduction to relativity. Quantum and Modern Physics Research Lab Mentorship A capstone class focused on research in physics. Researched wireless communication at the Naval Research Laboratory. Research was named a semifinalist in Intel's Science and Talent Search and patent. Computational Physics Introduction to computational techniques used to study and analyze physical phenomena. Advanced Communication Data Streams A laboratory-based class focusing on how data is transmitted. Covered signal analysis topics such as convolutions, filters, and fourier transforms. Artificial Intelligence Class focusing on artificial intelligence, machine learning, and applications thereof. Subjects included genetic algorithms, neural networks, and image processing. Parallel Computing Introduction to parallel algorithms, and techniques. Covers CUDA, MPI, OpenMP, and OpenGL. Labs included n-body simulation, ray tracer, parallel sorts, and huffman encoding.