Profile guided optimizations can improve application performance significantly
by collecting relevant runtime data and exploiting it to generate highly
optimized code. However, gathering accurate runtime execution profiling
information is typically a very expensive operation.

In this talk, we will discuss the design choices we made for the value profiling
mechanism newly implemented in the Eclipse OpenJ9 runtime profiling infrastructure.
In a nutshell, the Eclipse OpenJ9 value profiler aims to efficiently collect the
most frequently observed values at specific program points in the compiled code
as well as the frequency of execution of those program points. Implemented
entirely in the compiler's intermediate representation, this platform independent
infrastructure serves as a light weight profiler that can collect data from a
running application with minimal impact on performance. Furthermore, it can be
patched to enable or disable data collection which extends its usability to
large number of compiled methods.