it divides 32 bit into 3 branches instead of 4.

your lookup is fine but i am using 11 bits table instead of 8.

i am using BDS2006 there is probably slow down by state push/popping by asm directive.

While you can use a lookup table ofĪny power of 2, I usually use a table size of 256 or 64K entries.įirst we create the lookup table: lookup = new int įor (int i = 1 i = 0x1000000) Īlso did some testing in C++ and my implementation of BSR is slower than lookup table I chose the latter, but it doesn't really matter - the end result is slow in all cases.Ī faster solution for this is to use a lookup table. Note that we need to modify this constant a bit to get the right results (as a result of floating point errors) or add a small number to get the correct results. Of course, we can optimize this by making 1.0/Math.Log(2) a constant. It requires 2 Log operations, that are in itself quite slow already. This implementation is trivial, but also very slow. The trivial implementation of a 2log in C# is: (int)(Math.Log(x) / Math.Log(2)) Also I do the 'add' operation to ensure the whole loop isn't magically removed by the JIT'ter.

run your tests at least 3 times and make sure the counters are big enough to hit multiple seconds. The complete source code to test all implementations is added at the bottom, so you can see for yourself.Īnd of course. For me it's key to get this as fast as possible. What is the same is the fact that this is a real performance drainer (as shown by performance tests of various applications). Especially the log2(int) operation is very useful.īecause I find myself using log2 all the time, I don't want to give a specific application I'm working on here. In almost all of these contexts they are used in the low-level code that is called billions of times. To name a few: compression algorithms, 3d engines and machine learning. The log2(int) and log2(float) operations are very useful in a lot of different contexts. Are there any other (and/or faster) implementations of a basic 2log?

Calculate log base 2