Flotsam: Insanely Fast Floating-Point Number Serialization for Java, Javascript

Sort of similar to this base64 encoder I wrote to speed up ThreeJS float data streams. Seems more complex though that the stuff that I did.

I had good performance gains versus JSON as evidenced by this popular JSPref:

http://jsperf.com/json-vs-base64

This is a problem that only "walled garden" languages like Java and JavaScript would ever even have in the first place. If you're not trapped in a gilded cage, you just print raw bytes directly to a socket and then read data back the same way. The only concern is byte ordering, which is easy.

Regarding this specific approach, "only 20% overhead" sounds pretty good, but base 64 encoding has "only 33% overhead", is completely general, and likely already has faster implementations than this.

Hah, I did almost the same thing recently! Precision and rounding was not that important though so i just went with log2 to get the exponent, normalize and cast to int to get the mantissa. Then I realized I could do it properly by just using a typed array :)

This could open up some nice possibilities for browser based and Node.js driven distributed computing.

Do the performance gains depend on how many digits before or after the decimal point there are?