Explore the current state of JVM Escape Analysis and its downstream optimizations in this informative conference talk from the Jfokus VM Tech Summit 2020. Delve into the intricacies of escape analysis (EA) and its role in optimizing JVM workloads. Learn about scalar replacement of objects and how it differs from stack allocation. Examine how C2, Graal, and OpenJ9 utilize EA data for various optimizations. Discover potential improvements for C2 in the OpenJDK community to enhance JVM efficiency through post-EA optimizations. Gain insights from Charlie Gracie, a Principal Software Engineer at Microsoft with extensive experience in virtual machine development, as he shares his analysis of EA and its impact on heap pressure. Cover topics such as partial escape analysis, EA consuming optimizations, limitations of scalar replacement, and potential implementations of stack allocation in C2. Understand the performance improvements and current limitations of these optimizations, and explore future steps for enhancing JVM performance.
Current State of JVM Escape Analysis and Downstream Optimizations
Overview
Syllabus
Intro
Overview
What is Escape Analysis?
Partial Escape Analysis
EA Consuming Optimizations
Hotspot C2 EA and optimizations
Example of non escaping objects
Limitations of scalar replacement
How common is this issue?
Simple change to the experiment
Could we make scalar replacement work?
Considering the previous example...
Stack allocate if scalar replacement fails
What if C2 had stack allocation?
How did we implement this in C2
Bax.ackNode to the rescue
We had to...
GC roat scanning of stack allocated objects
Overlapping live object ranges
Current Limitations
Performance improvements
When and where can we see this patch?
Next steps
Taught by
Java
