Colloquia: Vijay Menon/Intel Preserving and Exploiting Information in an Optimizing Compiler TAY 3.128

Contact Name: 
Jenna Whitney
Date: 
Nov 2, 2006 12:30pm - 1:30pm

There is a signup schedule for this event.

Type of Talk:

Colloquia

Speaker Name: Vijay Menon

Speaker Affiliation: In

tel

Date: Thursday November 2 2006

Start Time: 12:30 p.m.

Location: TAY 3.128

Host: Keshav Pingali

Talk Title

: Preserving and Exploiting Information in an Optimizing Compiler

T

alk Abstract:
I present a novel framework for representing safety inform

ation in aggressively optimizing compilers for safe languages such as Java

and C# being developed at Intel''s Programming Systems Lab. We introduce

a verifiable low-level program representation to embed propagate and pres

erve safety information during compilation. Our representation precisely e

ncodes safety information via static single-assignment proof variables that
are first-class constructs in the program.

I argue that our represe

ntation allows a compiler to both (1) express aggressively optimized machin

e-independent code and (2) leverage existing compiler infrastructure to pre

serve safety information during optimization. I demonstrate that this appr

oach supports standard compiler optimizations requires minimal changes to

the implementation of those optimizations and does not artificially impede
those optimizations to preserve safety.

I also discuss ongoing work
where we use this framework to enable speculative motion of dangerous (i.e

. potentially faulting) instructions in safe languages such as Java and C#

. In contrast with earlier work our framework requires no hardware or ope

rating system support. Instead we leverage the properties already provide

d by a safe language to define fault safety a more precise notion of safet

y that guarantees that a dangerous operation (e.g. a memory load) will not
fault at a particular program point.

I discuss experiences integrat

ing this work into the machine independent global optimizer of StarJIT a h

igh-performance just-in-time compiler that performs aggressive control-flow
data-flow and algebraic optimizations and is competitive with top produc

tion systems.

Speaker Bio:
Vijay Menon is a research staff membe

r at the Intel Santa Clara labs. He received his bachelor''s degree from th

e University of California Berkeley and his PhD from Cornell where he wor

ked with Keshav Pingali on fractal symbolic analysis. He is known for his c

ontributions to software transactional memory and JIT compiling. He will gi

ve two talks this Thursday on recent work that has appeared in PLDI and POP

L.