Levon Stepanian
<br>
"Inlining Java Native Calls" - Abstract
<hr>

My research attempts to increase the runtime performance of Java native calls by adding a JNI-inlining
optimization step in the IBM Testarossa JIT compiler (the JIT compiler that ships with the IBM J9 JVM).
<p>

Traditionally, runtime optimizers treat Java native functions in an opaque manner, unable to induce them
to optimizations.  My work takes advantage of an intermediate representation (IR) conversion engine currently
under development at IBM  and will allow the JIT compiler to load the native library containing the function
of
interest, extract the IR for the said native callee, convert it to the JIT compiler's IR and finally modify
the callsite
with the inlined code.
<p>

I hope to attain performance benefits by not only diminishing traditional function invocation overheads,
but those related to JNI invocations and more importantly, overheads related to callbacks from native
code to Java.  The big picture here is to encourage code development with a programmer-friendly
language (Java) by minimizing performance hits associated with the reuse of legacy, high-performance or
architecture dependant native codes.
<p>

My talk will focus on providing a background for my research, as well as related work,
preliminary design and a status report on implementation. (Note: No evaluation yet).
<p>