3rd year PhD student @ NU
Ciao! My name is Federico Sossai and I don’t snack during talks :)
I’m a third-year Computer Science PhD student at Northwestern University and part of the ARCANA Lab led by Simone Campanoni.
At the moment I’m working on parallelizing compilers.
In my vision, a powerful langugage allows users to express high-level properties of their algorithms, while a powerful compiler should exploit them and minimize the performance cost of the abstractions that permit such expressiveness.
During the implementation of an algortihm, it’s easy to get distracted by low-level details that pollute our code making it unreadable and inflexible.
When single-core performance is not enough, the problem is exacerbated by manual parallelization techniques that, for example, require data structures to be reshaped or even worse, redesigned.
Automatic parallelization for scientific codes has been studied for decades with remarkable results. However, when programs are not dominated by well-structured array computation we are in dim candlelight.
This is why I’m investigating the following questions:
To answer these questions I’m studying the nature of the loop-carried data dependences that block parallelization and how we can satisfy them in less conventional ways.
I believe that in the advent of an even more heterogeneous future, the importance of a compiler that understands and manipulates parallelism will be hard to overstate.
|
J Giordani, Z Xu, E Colby, A Ning, AR Godala, I Chaturvedi, S Zhu,
Y Chon, G Chan, Z Tan, G Collier, JD Halverson, EA Deiana, J Liang,
F Sossai, Y Su, A Patel, B Pham, N Greiner, S Campanoni, DI August Revisiting Computation for Research: Practices and Trends |
|
T McMichen, N Greiner, P Zhong, F Sossai, A Patel, S Campanoni Representing Data Collections in an SSA Form |
|
S An, L Moneta, S Sengupta, A Hamdan, F Sossai, A Saxena C++ Code Generation for Fast Inference of Deep Learning Models in ROOT/TMVA |
I’ve always had a strong passion for programming. I’m fascinated by how much structure can emerge when spoken ideas become written algorithms, and how these structures come back again and again.
At the University of Padova, I realized how the love for learning did not only belong to computer science but extended to all the disciplines ruled by the rigor of math. For better or worse, calculus, electromagnetism, probability theory, and NP-completeness forever changed how I look at the world, and there’s no way back.
$ cat contacts.json{
"firstname": "Federico",
"lastname": "Sossai",
"email": "${firstname}.${lastname}@gmail.com",
"citizenships": ["Italy"],
"address": {
"building": "Mudd Library",
"room": 3304,
"street": "2233 Tech Drive",
"zip": 60208,
"city": "Evanston",
"state": "IL",
"country": "USA"
}
}