Runtime State Based Code Re-Optimization

Abstract

A method and an apparatus to execute a code compiled from a source code to access an untyped variable are described. An optimized access code may be compiled in the code with speculative optimization via a type prediction of runtime value of the untyped variable. Invalidity of the type prediction may be dynamically detected for future runtime values of the untyped variable. The code may be updated with an access code compiled for the access without the speculative optimization based on the invalidity detection. The updated code can be executed for the access to the untyped variable without executing the optimized access code.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS[0001]This application is related to, and claims the benefits of U.S. Provisional Patent Application No. 61 / 681,530, filed Aug. 9, 2012, entitled “DYNAMIC CODE OPTIMIZATION”, by Filip Pizlo et al., which is hereby incorporated by reference herein in its entirety.FIELD OF INVENTION[0002]The present invention relates generally to optimized runtime compilation. More particularly, this invention relates to timely re-optimization of dynamic language based code based on runtime state.BACKGROUND[0003]Dynamic languages that lack type annotations of any kind are increasingly popular. For example, JavaScript has become the lingua franca of the World Wide Web, while Ruby, Python, Perl, and PHP have become standard languages for server-side infrastructure. However, runtime of dynamic language based code may require significant overhead to execute.[0004]For example, lack of type information may prohibit optimal compilation to avoid dynamic type conversion f...

AI Technical Summary

Problems solved by technology

However, runtime of dynamic language based code may require significant overhead to execute.

For example, lack of type information may prohibit optimal compilation to avoid dynamic type conversion for a variable.

Typical runtime profiling, however, relies on recording types of values encountered and requires complicated data structures and extensive profiling operations.

As a result, effectiveness of the compilation is adversely affected for overall performance improvement.

Further, when current runtime values encountered no longer match the previous profiling results, optimally compiled code based on the previous profiling results may cost more processing cycles to execute than a corresponding non-optimally compiled code.

Such discrepancies between current runtime values and previous profiling results can severely tax execution performance.

As a result, runtime performance may be degraded with the overhead incurred by data loading / updating through the interface layer.

Therefore, existing runtime compilation and execution for dynamically typed code tends to introduce performance cost which may be unnecessary, wasteful and avoidable.

Images