Oracle Java 7-8 Hotspot VM Loop Optimization

Question

I would like to know what is the loop optimization performed by Oracle Java 7 (or 8) Hotspot VM?

Answer 1

• "Range correction" - excludes range checks for loop-invariant arrays. See PhaseIdealLoop :: do_range_check for more details . Optimization is controlled by the -XX:+RangeCheckElimination
• Loop Peeling - breaks the first iteration from the loop and performs it outside the loop body. See the amazing description here PhaseIdealLoop :: do_peeling . This optimization is controlled by the -XX:PartialPeelLoop=true flag -XX:PartialPeelLoop=true
• Loop Predication - excludes checking conditions inside the loop body. Currently, loop predicate optimization is used to remove checking the range of arrays and checking the invariant of the loop (such as checking for errors and checking arrays). The precision of the loop is controlled by the -XX:+UseLoopPredicate . See PhaseIdealLoop :: loop_predication_impl Code
• Loop Unrolling - Used as the first step of the Superword Level Parallelism . See PhaseIdealLoop :: do_unroll . Loop deployment is controlled by the following properties: -XX:LoopMaxUnroll=16 and -XX:LoopUnrollMin=4
• Array Filling - replaces any intrisc fill patterns. See PhaseIdealLoop :: do_intrinsify_fill . JVM Option -XX:+OptimizeFill
• Vectorization - replaces the initialization, copying and arithmetic of an array with vector operations in expanded loops. The Hotspot compiler implements the concept of the Superword Level Parallelism in superword.cpp . See Also JVM Option -XX:+UseSuperWord