If you're curious why you would want a vectorized way to sort lists of numbers, one use case is building histograms - it's much easier to build a histogram if you've sorted all your samples first
Actual title: “Vectorized and performance-portable Quicksort” (2022).
Actual sense in which it’s first:
Happily, modern instruction sets (Arm SVE, RISC-V V, x86 AVX-512) include a special instruction suitable for partitioning. Given a separate input of yes/no values (whether an element is less than the pivot), this "compress-store" instruction stores to consecutive memory only the elements whose corresponding input is "yes". We can then logically negate the yes/no values and apply the instruction again to write the elements to the other partition. This strategy has been used in an AVX-512-specific Quicksort. But what about other instruction sets such as AVX2 that don't have compress-store? Previous work has shown how to emulate this instruction using permute instructions.
We build on these techniques to achieve the first vectorized Quicksort that is portable to six instruction sets across three architectures, and in fact outperforms prior architecture-specific sorts.
Id say that it is a lot more likely for the in-house, at least half a decade old library to be correct and performant than 3000 lines of an LLMs mediocre regurgitation of that code.
Well, it came out a while ago, so maybe we can be a bit silly:
There’s something sort of beautiful about mergesort and heapsort. Their names tell you what their main idea is, and how they work is immediately obvious.
Quicksort, on the other hand, has nothing beautiful about it and is named after it’s one redeeming feature (that it is quick for a lot of cases).
Just imagine when candidates will get asked by pre-revenue startups to implement a vectorized version of quick-sort in person in 10 mins, just for a SWE job which they do not use this themselves.
Only the likes of MAG 7, and a couple of hedge-funds would ask to do it since this problem directly applies to them.
Very nice. Let's see Paul Allen's quicksort
(2022)
If you're curious why you would want a vectorized way to sort lists of numbers, one use case is building histograms - it's much easier to build a histogram if you've sorted all your samples first
I wonder what apps have implemented this now that a few years have passed.
Who would've guessed? SIMD is so boring.
Honestly part of me feels that way about way too many things in retrospect about my own life and interests.
time to log off
i read that and thought the same. actual based idea its even inspired me to log off
Actual title: “Vectorized and performance-portable Quicksort” (2022).
Actual sense in which it’s first:
Would they do the same thing today or have an LLM re-implement those 3000 lines of c++ ?
Id say that it is a lot more likely for the in-house, at least half a decade old library to be correct and performant than 3000 lines of an LLMs mediocre regurgitation of that code.
Guys, remember when language features allowed re-usability?
Well, it came out a while ago, so maybe we can be a bit silly:
There’s something sort of beautiful about mergesort and heapsort. Their names tell you what their main idea is, and how they work is immediately obvious.
Quicksort, on the other hand, has nothing beautiful about it and is named after it’s one redeeming feature (that it is quick for a lot of cases).
Definitely needs (2022) in the title, I was a bit confused!
Strange to see it here, the article is quite old.
Since pdqsort, vqsort, and glide sort, the current state-of-the-art are driftsort and ipnsort.
I've integrated them into ClickHouse: https://github.com/ClickHouse/ClickHouse/pull/106650
only sorts numbers? wouldn't radix be much better?
Just imagine when candidates will get asked by pre-revenue startups to implement a vectorized version of quick-sort in person in 10 mins, just for a SWE job which they do not use this themselves.
Only the likes of MAG 7, and a couple of hedge-funds would ask to do it since this problem directly applies to them.
But certainly not pre-revenue startups.
this was made like 4 years back most of the current algorithms use this already !