org.graalvm.compiler.lir

Class SwitchStrategy

java.lang.Object

org.graalvm.compiler.lir.SwitchStrategy

Direct Known Subclasses:

SwitchStrategy.BinaryStrategy, SwitchStrategy.RangesStrategy, SwitchStrategy.SequentialStrategy

public abstract class SwitchStrategy
extends Object

This class encapsulates different strategies on how to generate code for switch instructions. The SwitchStrategy.getBestStrategy(double[], JavaConstant[], LabelRef[]) method can be used to get strategy with the smallest average effort (average number of comparisons until a decision is reached). The strategy returned by this method will have its averageEffort set, while a strategy constructed directly will not.

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	SwitchStrategy.BaseSwitchClosure Backends can subclass this abstract class and generate code for switch strategies by implementing the SwitchStrategy.BaseSwitchClosure.conditionalJump(int, Condition, Label) method.
static class	SwitchStrategy.BinaryStrategy This strategy recursively subdivides the list of keys to create a binary search based on probabilities.
static class	SwitchStrategy.RangesStrategy This strategy divides the keys into ranges of successive keys with the same target and creates comparisons for these ranges.
static class	SwitchStrategy.SequentialStrategy This strategy orders the keys according to their probability and creates one equality comparison per key.

Field Summary

Fields

Modifier and Type	Field and Description
double[]	keyProbabilities

Constructor Summary

Constructors

Constructor and Description
SwitchStrategy(double[] keyProbabilities)

Method Summary

Modifier and Type	Method and Description
double	getAverageEffort()
static SwitchStrategy	getBestStrategy(double[] keyProbabilities, JavaConstant[] keyConstants, LabelRef[] keyTargets) Creates all switch strategies for the given switch, evaluates them (based on average effort) and returns the best one.
abstract Constant[]	getKeyConstants()
protected void	registerDefaultEffort(int depth) Tells the system that the default successor is reached after depth number of comparisons, which is used to calculate average effort.
protected void	registerEffort(int rangeStart, int rangeEnd, int depth) Tells the system that the given (inclusive) range of keys is reached after depth number of comparisons, which is used to calculate the average effort.
abstract void	run(org.graalvm.compiler.lir.SwitchStrategy.SwitchClosure closure)
String	toString()

Methods inherited from class java.lang.Object

clone, equals, getClass, hashCode, notify, notifyAll, wait, wait, wait

Field Detail

keyProbabilities

public final double[] keyProbabilities

Constructor Detail

SwitchStrategy

public SwitchStrategy(double[] keyProbabilities)

Method Detail

getKeyConstants

public abstract Constant[] getKeyConstants()

getAverageEffort

public double getAverageEffort()

registerEffort

protected void registerEffort(int rangeStart,
                              int rangeEnd,
                              int depth)

Tells the system that the given (inclusive) range of keys is reached after depth number of comparisons, which is used to calculate the average effort.

registerDefaultEffort

protected void registerDefaultEffort(int depth)

Tells the system that the default successor is reached after depth number of comparisons, which is used to calculate average effort.

toString

public String toString()

Overrides:

toString in class Object

run

public abstract void run(org.graalvm.compiler.lir.SwitchStrategy.SwitchClosure closure)

getBestStrategy

public static SwitchStrategy getBestStrategy(double[] keyProbabilities,
                                             JavaConstant[] keyConstants,
                                             LabelRef[] keyTargets)

Creates all switch strategies for the given switch, evaluates them (based on average effort) and returns the best one.