Behnam Robatmili

         Department of Computer Science, University of Texas at Austin, PHD Student My Curriculum vitae

          beroy@cs.utexas.edu 

RESEARCH INTERESTS

Execution Reuse for Bulk-commit Processors, Jan 2009 - Present, In this study, I explore several techniques for exploiting temporal locality in instruction windows of processor with bulk commit capability.

Criticality-directed Accelation for Bulk-commit Processors, Jan 2009 - Present, In bulk-commit processprs, inter-bulk communication extend between atomically-commited instruction groups. Inter-bulk communications are more likely to happen on the critical path. In this study, we detect and optimize inter-bulk communication for better performance/power.

Extracting Concurrency in Single-thread Applications in a Composable Multicore System, Oct 2007 - Present, Balancing concurrency and communication is a challenge for future distributed processors. The purpose of this project is automatically map the instructions and select the cores for each block of the code. Instruction mapping for each block is done adaptively according to abstracted concurrency hints provided by th compiler.

The Highest Performing Program on TRIPS, Mar 2007 - Aug 2007, Extending the well known algorithm for matrix-multiplcation by Goto to a systolic-like algorithm running on a 2-D TRIPS ALU grid. This algorithm outperforms the implementation Goto's hand tuned code on all of then other general purpose processors so far!

Register Bank Allocation for Specially Partitioned Processors, Mar 2006 - Sep 2007, To choose a register bank for each variable, this algorithm considers the critical path through the program, delays from registers to consuming instructions, and other resource constraints imposed by the hardware.

Policy Search Optimization for Spatial Path Planning, Apr 2007 - Nov 2007, Improving the instruction scheduling for TRIPS instruction blocks using a Reinforment Learning method called NEAT.