Performance Benchmarks about bignumber.js

[kyledrake]

JSBN so far seems to be faster (aside from the fluke Chrome benchmark), it would be fun to try to optimize the BigNumber.js source and see if it could be made competitive in this particular benchmark!

[josdejong]

JSBN seems to do worse on Chrome 34?

May be interesting as well to compare big number calculations with regular floating point Number calculations? Like:

var a = 1024;
var b = 9999;
(a + b) * b;

[weilu]

34 beta. Yea it's strange. Out of curiosity I did include the browser float calculations when I was toying with it. It was so uncomparable that I couldn't event see bars of other tests at all. Try it ;)

[MikeMcl]

The source code is already pretty well optimsed across various platforms, and I'm happy with its competitiveness as it is.

bignumber.js can be made faster by adding more specialised code paths (i.e. more code) and by storing values in a different format (e.g. a higher base), but having experimented with such I found the benefits didn't outweigh the costs.

jsbn is great, but it's integers only.

[kyledrake]

Totally understandable. The code quality is the reason we were looking at this, wouldn't want you to sacrifice that. Thank you for your time!

Would you be able to make an asm.js optimized version? Does anything like that exist?

[MikeMcl]

@werian

I am working on a faster version but not optimised for asm.js.

I had the idea of re-writing the whole thing in the asm.js JavaScript subset, and looked into it last year briefly and came to the conclusion that it wasn't viable, and that it wasn't what asm.js was for.

The library could probably be optimised further using typed arrays but I am not looking to go down that route at the moment as I want to keep the broad platform compatibility, and some compilers may be making that optimisation behind the scenes anyway.