lopx at gazeta.pl
Wed Apr 30 16:56:50 PDT 2008
>> StackMachnie.java contains methods responsible for stack
>> management/manipulation. ByteCodeMachine.java is the actual bytecode
>> interpreter implementation (with matchAt method containing the big
>> switch). Each opcode is implemented as a separate method.
> That is reasonable. It keeps the switch per se small, and
> (implicitly) asks the JVM to inline the opcode methods that really
Ah, forgot to mention this one. Surprisingly, removing unused cases (for
given matches) that contain just method invocations affects performance.
Can such slight differences in method size affect performance as well ?
Is that because of inlining/thresolds artifacts ? It there a possibility
that neighbour method size plus the size of method being trimmed were on
the verge of inlining threshold at hight level ?
More information about the hotspot-compiler-dev