Array#float_sum (like math.fsum of Python)

Issue #10298 has been updated by Takeshi Nishimatsu.

Timing.
Of course it is fast.

$ uname -sprsv
Darwin 13.3.0 Darwin Kernel Version 13.3.0: Tue Jun  3 21:27:35 PDT 
2014; root:xnu-2422.110.17~1/RELEASE_X86_64 i386
$ sysctl -n machdep.cpu.brand_string
Intel(R) Core(TM) i5-3210M CPU @ 2.50GHz
$ ./miniruby --version
ruby 2.2.0dev (2014-09-29 trunk 47735) [x86_64-darwin13]
$ /usr/bin/time ./miniruby -e 'p Array.new(10000000,0.1).reduce(:+)'
999999.9998389754
        0.66 real         0.62 user         0.03 sys
$ /usr/bin/time ./miniruby -e 'sum=0.0; c=0.0; 
Array.new(10000000,0.1).each{|x|y=x-c; t=sum+y; c=(t-sum)-y; sum=t}; p 
sum'
1000000.0
        1.47 real         1.44 user         0.02 sys
$ /usr/bin/time ./miniruby -e 'p Array.new(10000000, 0.1).float_sum'
1000000.0
        0.10 real         0.06 user         0.03 sys


Feature #10298: Array#float_sum (like math.fsum of Python)

  • Author: Takeshi Nishimatsu
  • Status: Open
  • Priority: Low
  • Assignee:
  • Category: math
  • Target version:

Here, I propose Array#float_sum in array.c (or math.c).
Array#float_sum returns an accurate total summation of Float
elements in an array using the Kahan summation algorithm
Kahan summation algorithm - Wikipedia .
This algorithm can significantly reduce the numerical
error in the total obtained by adding a sequence of
finite precision floating point numbers, compared to the
obvious approach. Python already have math.fsum
9.2. math — Mathematical functions — Python 2.7.18 documentation .

[0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1].float_sum   #=> 1.0
[].float_sum                                                   #=> 0.0
Array.new( 10, 0.1).float_sum    #=>  1.0
Array.new(100, 0.1).float_sum    #=> 10.0
# cf.
Array.new( 10, 0.1).reduce(:+)   #=>  0.9999999999999999
Array.new(100, 0.1).reduce(:+)   #=>  9.99999999999998

The name of method can be fsum, sum_float, etc., though
I propose float_sum.

This Array#float_sum is inspired by Feature #9834
Float#{next_float,prev_float}.

—Files--------------------------------
array.float_sum.patch (1.32 KB)