Wolfsort: An ultra-fast hybrid radix sort algorithm

by scandum • 6 years ago

102 27 6 years ago

Wolfsort requires 8 times the array size in swap memory for the O(n) partitioning process.
Wolfsort is primarily a proof of concept, currently it only properly supports unsigned 32 bit integers.

A lot of sorts can beat a generic sort on an array of numbers by gathering statistics, paying better attention to locality, implementation details like use of SIMD or some combination of those three.

cheez5• 6 years ago

Indeed, I wonder if someone can't write an ensemble algo of sorts that gets chosen based on some AI (not to hand wave)

pdq• 6 years ago

See Hybrid Sorting algorithms. Here's Introsort as an example:

https://en.wikipedia.org/wiki/Introsort

HippoBaro• 6 years ago

Historically software was always low on memory, and subsequently sorting algorithms were developed that bent over backwards in order to use as little additional memory as possible. Nowadays most modern systems have several gigabytes of memory that are not used and are just sitting idle.

True, but on these systems, most of the time, users don't care about performance. And when they do, the dataset is not large enough for this kind of algorithms to shine anyway.

nine_k• 6 years ago

Also, memory is way slower than the CPU cache(s). So operating within a small subset of the available memory is still beneficial if you need the top performance.

gridlockd• 6 years ago

There often is only a narrow range where (data + overhead) is larger than cache, but (data + 0) is smaller than cache.

Cached reads/writes arent't free either, so if overhead drastically reduces those, you may come out on top 100% of the time.

secondcoming• 6 years ago

Who doesn't care about performance? If your browser took 1 minute to start up you'd uninstall it.

shin_lao• 6 years ago

It seems to me that it's using a strategy similar to SpreadSort, or am I wrong?

https://en.wikipedia.org/wiki/Spreadsort

nedoka• 6 years ago

Similar strategy. Wolfsort is stable however while SpreadSort does not appear to be so.

I wonder if wolfsort is stable and faster because it utilizes quadsort?

https://github.com/scandum/quadsort

aciswhat• 6 years ago

It does, according to the Wolfsort's readme

usr1106• 6 years ago

Wolfsort is primarily a proof of concept, currently it only properly supports unsigned 32 bit integers.

Which makes it pretty useless outside of theoretical exercises(that is not saying that theoretical exercises would be useless). What would be a practical use case to sort integers? Well, the integers can be keys to some kind of records, but if your records remain unsorted accessing them is very expensive even if you have the sorted list of keys available.

Edit: Calculating the median is a practical case. Not sure how often that is needed in performance-critical contexts.

joshuamorton• 6 years ago

Of course if you want the median you do so in linear time with quickselect/k-select.

zelly2• 6 years ago

All radix sorts only work for integers. This isn't unique to OP's algorithm.

kd5bjo• 6 years ago

This is true in only the strictest sense, as computers only deal with integers:

Any set of tuples/sequences can be lexicographically radix-sorted if its components can be; strings are a special case, where the components are the individual character codes.

Floating point numbers are integers with the high-order bits used for the exponent. When normalized, they can be treated as integers for sorting.

All radix sort requires is a recursively partitionable domain, and it performs optimally when the partitioning scheme matches the data distribution.

dthul• 6 years ago

A bit OT: I very much appreciate well written C code for its simplicity and "rawness", but how come that many C codebases seem to place no importance on code comments?

This repo contains a few thousand lines of code without so much as a single comment. In my experience that tends to happen much less in other languages. And it's not like this C code is obvious.

croutonix• 6 years ago

Comments increase compile time!

scaredginger• 6 years ago

Not to mention file sizes!

bane• 6 years ago

C has an interesting culture of terse variable and function names, and limited comments. It comes from an entirely different mindset and time period in software. It predates most "modern" concepts of software re-usability that eventually turned into enterprise OOP practices and was crystalized in Java and C#.

It's not great anymore. The software world has moved on to understanding how important these things are (and even often takes things too far), but C seems a bit stuck in its ways.

vii• 6 years ago

The value of comments and references depends on the software context. If inexperienced people are expected to be editing or reading the code they're more valuable. However, they are expensive to keep in sync with changes to the implementation and the value has to balanced against this cost.

Java and C# did not replace C. They're typically used for different tasks.

Objectively speaking, do we have evidence these enterprise OOP practices lead to better software?

Wolfsort: An ultra-fast hybrid radix sort algorithm

Related Stories