Java program runs slower when code that is never executed is commented out

Question

I observed some strange behaviour in one of my Java programs. I have tried to strip the code down as much as possible while still being able to replicate the behaviour. Code in

Answer 1

The commented code affects how inlining is handled. If functionB gets longer/bigger (more bytecode instructions) it will not be inlined into functionA.

So @J3D1 was able to use VMOptions to manually switch off inlining for functionB(): -XX:CompileCommand=dontinline,com.jd.benchmarking.StrangeBeh‌​aviour::functionB This appears to eliminate the delay with the shorter function.

with the vm options you can display inlining -XX:+UnlockDiagnosticVMOptions -XX:+PrintInlining

The bigger Version, wont inline functionB

@ 8   StrangeBehaviour::functionB (326 bytes)   callee is too large
@ 21   StrangeBehaviour::functionA (12 bytes)
  @ 8   StrangeBehaviour::functionB (326 bytes)   callee is too large
@ 35   StrangeBehaviour::functionA (12 bytes)
  @ 8   StrangeBehaviour::functionB (326 bytes)   callee is too large

The shorter Version will try to inline functionB, causing a few of further attempts.

@ 8   StrangeBehaviour::functionB (318 bytes)   inline (hot)
 @ 21   StrangeBehaviour::functionA (12 bytes)   inline (hot)
   @ 8   StrangeBehaviour::functionB (318 bytes)   inline (hot)
     @ 35   StrangeBehaviour::functionA (12 bytes)   recursive inlining is too deep
 @ 35   StrangeBehaviour::functionA (12 bytes)   inline (hot)
   @ 8   StrangeBehaviour::functionB (318 bytes)   inline (hot)
     @ 21   StrangeBehaviour::functionA (12 bytes)   recursive inlining is too deep
     @ 35   StrangeBehaviour::functionA (12 bytes)   recursive inlining is too deep
@ 21   StrangeBehaviour::functionA (12 bytes)   inline (hot)
 @ 8   StrangeBehaviour::functionB (318 bytes)   inline (hot)
   @ 35   StrangeBehaviour::functionA (12 bytes)   inline (hot)
     @ 8   StrangeBehaviour::functionB (318 bytes)   recursive inlining is too deep
@ 35   StrangeBehaviour::functionA (12 bytes)   inline (hot)
 @ 8   StrangeBehaviour::functionB (318 bytes)   inline (hot)
   @ 21   StrangeBehaviour::functionA (12 bytes)   inline (hot)
    @ 8   StrangeBehaviour::functionB (318 bytes)   recursive inlining is too deep
   @ 35   StrangeBehaviour::functionA (12 bytes)   inline (hot)
     @ 8   StrangeBehaviour::functionB (318 bytes)   recursive inlining is too deep

Mostly guessing, but the bigger/inlined bytecode will cause problems with branch prediction and caching