Homepage of

Saeed Maleki


  • As my first project in UIUC (Spring 09-Spring 11), I worked on compiler optimization and code tuning for the compiler part of the Blue Waters project. This work has been supervised by Prof. David Padua and Prof. Maria Garzaran. The focus was mostly on vectorization for two architectures including IBM and Intel processors. We studied the compiler behavior for various benchmarks to evaluate the compilers' vectorization ability including Intel C/C++ Compiler (for Intel platform), XLC/C++ compiler (for IBM platform) and GCC (for both of them). An outcome of this work was a paper in PACT'11 conference entitled "An Evaluation of Vectorizing Compilers" and a tutorial presented at SC'10, CGO'11 and PLDI'11 entitled "Program Optimization through Loop Vectorization". Also, the developers of IBM XLC compiler are updating and improving the compiler based on our studies and recommendations to the.
  • I worked as a visiting scholar on SPIRAL project at Carnegie Mellon University in summer 2011. In that project, I worked on an automatic highly optimized code-generator for Digital Signal Processing (DSP) algorithms such as Fourier Transformation for ARM processors. The ARM processor that we used was a dual-core Cortex A9 with NEON vector unit. Therefore, we utilized these vector units and parallelized the code by using PThreads. Our performance results for Fourier transform are very promising in compare to FFTW results for ARM although SPIRAL can generate many other DSP algorithms and FFTW only generates Fourier Transform.
  • I worked on designing compilation techniques for high level languages to compile the same code for multi-core processors, GPUs and clusters. The result of this work was a paper in IPDPS'12 conference entitled "Performance Portability with the Chapel Language".
  • My current research is about designing an interface for parallelizing graph problems  for distributed systems. Right now I am writing a paper about parallelizing single/many-source shortest path problem and betweenness centrality problem which are very challenging to parallelize on a distributed machine.


Office address: 4213 Siebel Center for Computer Science, 201 N Goodwin Ave, Urbana, IL 61801

Email:
mylastname+1 <at>  illinois.edu

Cell Phone:
Available per request by email

Copyright © 2009, Saeed Maleki
Last Update:
September, 2011