HN Explorer

7 comments

[ 3.3 ms ] story [ 18.4 ms ] thread
ktpsns
The fastest way are probably rational numbers, as they are natively available in many functional languages or Perl6. That is, 0.1 is represented as the integer tupls (1,10) instead of a floating point number.
munrocket
How many FLOP do you need to add two rationals? We don't have int type in javascript, only bigInt but it not supported in some browsers and probably not fast enough.

UPD: I added Fraction.js to benchmark https://munrocket.github.io/double.js/docs/benchmark.html but it slower and overflows for some reason, .round(31) not helps.

ktpsns
Adding two rationals boils down to adding 2x two integers. Integer arithmetic is in general less complex then floating point arithmetic. But of course that's not true if your hardware supports single instruction double floating point arithmetic (which any relevant 64bit architecture does).
lizmat
Please note that Perl 6 has been renamed to Raku (https://raku.org using the #rakulang tag on social media). So in Perl `say "same" if .1 + .2 == .3` will not say anything, as that condition is not true. In Raku, `say "same" if .1 + .2 == .3` will display "same".
Someone
This doesn’t solve the general problem that binary floating point cannot exactly represent decimal fractions.

The first example

  new Double('0.3').sub(new Double('0.1')).toNumber()
does the equivalent of

  float f = (float)(0.3 - 0.1);
which happens to produce a ‘float32’ that prints as “0.2”, but for ‘float64’ vs ‘float128’.

I expect that adding a sufficient number of zeroes, as in

  double f = (float128)0.00003 - (float128)0.00001);
(actual number of zeroes is too low here, to prevent wrapping on phone screens) will surface the problem again.