GSOC 14: Native Julia implementations of massively parallel dense linear algebra routines

Skip to first unread message

Pulkit Singhal

Mar 20, 2014, 3:33:55 PM3/20/14
I'm Pulkit, and I'm pursuing research in parallel algorithms. I'm  interested in this project. I'm familiar with block tiling, cache-aware and cache-oblivious algorithm. But I gather that there are many algorithms that can be implemented. Project page talks about linear solver and singular value decomposition. However it doesn't restricts to these two algorithms. 
There are many algorithms
  • Eigenvalue Problem
  • Linear Solvers
    • LU Factorization
    • Cholskey Factorization
    • Iterative Grahm-Schmidt algorithm
    • Solver for Tridiagonal system using spike algorithm
  • Least Square solver
  • SVD
  • QR Factorization 
So my question is what is the scope of algorithms that we can implement. Surely, the codes need to be benchmarked, so are there any other native implementations which which doesn't use cache-optimized algorithms. Is it OK to compare them with implementation existing outside the scope of julia, for eg. Intel MKL, MATLAB etc.  

Another issue is in regard to interface, how should typical function should look like. Should it be compliant with LAPCAK interface ? 

Jiahao Chen

Mar 20, 2014, 11:05:44 PM3/20/14
Hi Pulkit,

Thanks for your interest in Julia GSoC. Please note that submissions
close tomorrow, Friday, March 21 at 19:00 UTC and we cannot
accept applications after that deadline.

Generally speaking, we would be interested in implementing algorithms
for massively parallel linear algebra. I think our project page makes
it quite clear that we are looking for someone who has interest in
extending the existing Julia interface for linear algebra to also work
on distributed arrays. We are not looking for someone to reimplement
ScaLAPACK. This project requires someone who has sufficient background
in numerical linear algebra and is also sufficiently familiar with
Julia syntax as well as the parallel constructs in Julia (notably
DArrays and SharedArrays).

If you would like to take up this project, I would encourage you to send
in an application based on your email as soon as possible to the GSoC
website, taking into account our suggested guidelines:


Jiahao Chen
Staff Research Scientist
MIT Computer Science and Artificial Intelligence Laboratory
Reply all
Reply to author
0 new messages