Automatic staged compilation

ResearchWorks/Manakin Repository

Search ResearchWorks


Advanced Search

Browse

My Account

Statistics

Related Information

Automatic staged compilation

Show simple item record

dc.contributor.author Philipose, Matthai en_US
dc.date.accessioned 2009-10-06T16:50:11Z
dc.date.available 2009-10-06T16:50:11Z
dc.date.issued 2005 en_US
dc.identifier.other b56437201 en_US
dc.identifier.other 70659527 en_US
dc.identifier.other Thesis 55552 en_US
dc.identifier.uri http://hdl.handle.net/1773/6891
dc.description Thesis (Ph. D.)--University of Washington, 2005. en_US
dc.description.abstract The ability to optimize programs while they execute has become increasingly important in recent years. The primary challenge in such optimization is to keep the run-time overhead of optimization down while maximizing its effectiveness. The widely used solution of Just-In-Time (JIT) compilation keeps run-time overhead low, at considerable engineering cost, by sacrificing performance.The past few years have seen the emergence of staged optimization , which produces run-time optimizations that often have much lower run-time overhead than traditional optimizers, yet do not sacrifice any of their functionality. The key to the technique is a method, called staging, to transfer optimization overhead to static compile time from run time. Unfortunately, developing staged variants of individual optimizations has been highly specialized, labor-intensive work; staging pipelines of optimizations even more so.This dissertation presents a system called the Staged Compilation Framework (SCF), which automatically stages entire pipelines of compiler optimizations at arguably little additional engineering cost beyond building the slower traditional version of the pipeline. SCF harnesses two powerful but traditionally difficult-to-use techniques, partial evaluation and dead-store elimination, to achieve staging. An implementation of SCF shows that staged compilation can speed up pipelines of classical compiler optimizations by up to an order of magnitude, and more commonly by a factor of 4.5 to 5. en_US
dc.format.extent xi, 246 p. en_US
dc.language.iso en_US en_US
dc.rights.uri en_US
dc.subject.other Theses--Computer science and engineering en_US
dc.title Automatic staged compilation en_US
dc.type Thesis en_US


Files in this item

Files Size Format View
3198836.pdf 13.76Mb PDF View/Open

This item appears in the following Collection(s)

Show simple item record