Bug #21851: Performance difference between / operator and fdiv

• Status: Closed
• Assignee: -
• Target version: -
• ruby -v: ruby 4.0.1 (2026-01-13 revision e04267a14b) +PRISM [x64-mingw-ucrt]
• Backport: 3.2: UNKNOWN, 3.3: UNKNOWN, 3.4: UNKNOWN, 4.0: UNKNOWN

[mbrown1intel (Michael Brown)]

Benchmarking fdiv against the / operator, it seems like fdiv is significantly slower. As far as I'm aware there's no functional difference between these two methods.

I first ran this on 2.6, and then checked against 4.0 to make sure it hadn't been fixed in the meantime, and saw the same results.

require 'benchmark'

n = 500_000_000
c = nil

Benchmark.bm do |benchmark|
  benchmark.report("to_f ") do
    n.times do |i|
      c = n / i.to_f
    end
  end

  benchmark.report("fdiv ") do
    n.times do |i|
      c = n.fdiv(i)
    end
  end
end

           user     system      total        real
to_f  37.672000   0.110000  37.782000 ( 38.714876)
fdiv 113.016000   0.265000 113.281000 (116.428442)

[byroot (Jean Boussier)]

Profile with / .to_f: https://share.firefox.dev/4rbNEEW

Profile with fdiv: https://share.firefox.dev/4qeMwzK

In short, fdiv end up in the gcd path, which is indeed much slower.

We could certainly recognize this case and add a fastpath, not sure I'd consider this a bug though.

[mame (Yusuke Endoh)]

• Related to Bug #13078: with Integer a and b, a.fdiv(b) sometimes inaccurate than Rational(a, b).to_f added

[mame (Yusuke Endoh)]

It seems Integer#fdiv uses gcd for accuracy in #13078. I'm not entirely sure why they only aimed to make fdiv accurate but not /.

Anyway, is this difference causing any real-world issues? If so, it's worth considering a fix. If just for performance consistency, doing nothing might be good enough.

[mbrown1intel (Michael Brown)]

No, this isn't causing me any issues, I just would prefer fdiv for my own codebase for readability. But with that large of a performance difference I think I would stick with "/".

[nobu (Nobuyoshi Nakada)]

GH-16144

[nobu (Nobuyoshi Nakada)]

• Status changed from Open to Closed

Applied in changeset git|74a4e2e66d14fed7d6ade09345a54ed14bc106cc.

---

[Bug #21851] Divide as double if accurate