Compaq Computer Corporation, Inc.

Apple Inc.

David William Goodwin

Scott James Forstall

Gregory B Novick

Cahya Adiansyah Masputra

Gregory R Chapman

John Joseph Iarocci

Gordon J Freedman

Richard Warren Schreyer

Neil Gareth Crane

Damien Pascal Sorresso

Lucia Elena Ballard

Henry Graham Mason

Darren S Litzinger

Christopher Marcellino

Curtis Rothert

Paul Geoffrey Lowney

Robert S Cohn

Charles Srisuwananukorn

Michael Van Milligan

David B Myszewski

Eric Albert

Joseph Sokol

Matthew G Waston

Thomas Brogan Duffy

Norman Rubin

A method for permitting software optimization tools, software instrumenting tools and other analysis tools to re-write executables having mixed instructions and data uses a data structure having an entry for each multi-bit word in an executable file. Each entry of the data structure includes a number of flags that are set to identify the type of the multi-bit word in the associated line of the executable file. The types include instruction, data and unclassified. Each entry also includes a flag that indicates that the multi-bit word should not be optimized and a flag indicating that the multi-bit word is a problem branch. The no-optimize and problem branch flags may be used to identify multi-bit words that may be either branch instructions or data, and to ensure that such multi-bit words are not affected by optimization or other rewriting of the executable. In addition, a problem fall through flag is provided to maintain program flow for possible fall through code segments.

Mechanism for re-writing an executable having mixed code and data

A profile feedback optimization system is provided. The system accepts as input an original application and produces an optimized version of the application. The system first instruments the original application through an instrumentation process such that when it executes, it generates profile data. During instrumentation, the system configures images in the application to trap any function calls to critical functions. This allows the instrumented version of the application to execute transparently and appear as the original version. The instrumentation process converts each original image of the application to a version that references other instrumented images. The instrumentation process also configures the operating system to execute the instrumented version instead of the original version when the user selects the application for execution. Once instrumented, the instrumented application is executed. During execution, profile data produced from the instrumented images is saved in a database. Once profile data has been generated, an optimization process is used to optimize the original version of the application based upon the profile data. The optimized version is also configured to execute just as the original application, but references optimized images instead of original images.

User transparent mechanism for profile feedback optimization

Growing community of inventors

Waltham, MA, United States of America

David William Goodwin