Why switch is faster than if

Because there are special bytecodes that allow efficient switch statement evaluation when there are a lot of cases.

If implemented with IF-statements you would have a check, a jump to the next clause, a check, a jump to the next clause and so on. With switch the JVM loads the value to compare and iterates through the value table to find a match, which is faster in most cases.

A switch statement is not always faster than an if statement. It scales better than a long list of if-else statements as switch can perform a lookup based on all the values. However, for a short condition it won't be any faster and could be slower.

The current JVM has two kinds of switch byte codes: LookupSwitch and TableSwitch.

Each case in a switch statement has an integer offset, if these offsets are contiguous (or mostly contiguous with no large gaps) (case 0: case 1: case 2, etc.), then TableSwitch is used.

If the offsets are spread out with large gaps (case 0: case 400: case 93748:, etc.), then LookupSwitch is used.

The difference, in short, is that TableSwitch is done in constant time because each value within the range of possible values is given a specific byte-code offset. Thus, when you give the statement an offset of 3, it knows to jump ahead 3 to find the correct branch.

Lookup switch uses a binary search to find the correct code branch. This runs in O(log n) time, which is still good, but not the best.

For more information on this, see here: Difference between JVM's LookupSwitch and TableSwitch?

So as far as which one is fastest, use this approach:If you have 3 or more cases whose values are consecutive or nearly consecutive, always use a switch.

If you have 2 cases, use an if statement.

For any other situation, switch is most likely faster, but it's not guaranteed, since the binary-search in LookupSwitch could hit a bad scenario.

Also, keep in mind that the JVM will run JIT optimizations on if statements that will try to place the hottest branch first in the code. This is called "Branch Prediction". For more information on this, see here: https://dzone.com/articles/branch-prediction-in-java

Your experiences may vary. I don't know that the JVM doesn't run a similar optimization on LookupSwitch, but I've learned to trust the JIT optimizations and not try to outsmart the compiler.