ENH add an introsort implementation

[hayd]

This is an implementation from the pseudo-code in Musser's paper.

I haven't done any performance testing, no doubt there is optimization to do here e.g. @inbounds and something better for index_of_median_of_three (which atm is terrible)...

[hayd]

this guard should rather be on hi - lo > 1 (which can cause an exception due to log2(...) == -Inf).

[hayd]

Hmmm, I wonder what this comment meant...

[kmsquire]

It's referring to the line below, and I think that's right. Care to submit a pull request?

[hayd]

So this line should just be hi-li > 1, will see if that works and push an update commit here.

Any thoughts on optimizations for index_of_median_of_three ?

[stevengj]

What's the advantage of introsort?

[kmsquire]

I haven't looked at this yet, but generally, introsort uses quick sort,
but keeps track of recursion depth, and if it reaches a certain depth,
switches to a different algorithm with guaranteed worst case depth of O(log
N).

It's a good algorithm to have available.

On Monday, October 27, 2014, Steven G. Johnson notifications@github.com
wrote:

What's the advantage of introsort?

—
Reply to this email directly or view it on GitHub
#7 (comment)
.

[xiaodaigh]

it's been more than 5 years. close it, merge it, or start another repo

[nalimilan]

There's an unresolved comment above that somebody needs to address (and also needs a rebase). (BTW, necro-posting generally isn't useful if you're not volunteering to work on PRs yourself.)