Order arrays of pairs of uint32? about hpc-algorithms HOT 3 OPEN

You're welcome! Glad you're finding it useful. What level of performance gains are you experiencing? A custom version of LSD Radix Sort can be created which would process an array of UInt32's, where even indexes hold keys to be sorted by and odd indexes hold the data items associated with each key. This would most likely lead to the highest performance. Javascript is not my expertise, and I don't know if an array of pairs is slower to access than an single array. A faster version of LSD Radix Sort has been developed ( https://duvanenko.tech.blog/2019/02/27/lsd-radix-sort-performance-improvements/) in C++ and C#, which can be ported to Javascript to provide higher performance. -Victor

Hi Hannes, LSD Radix Sort that you pointed out by Andy Adinets and Duane Merill is the latest GPU-based algorithm. They use some of my work (see the second page) and point to one of my blog entries: " We remark that Duvanenko recently made an independent, similar observation regarding the upfront consolidation of digit histograms for sequential implementations of LSD radix sorting [6]" It's a technique to split LSD Radix Sort into two phases: Histogram and Permutation, to reduce the number of passes over the array almost in half. I have only scanned Andy's and Duane's paper so far, and do not yet fully understand their technique. WebGPU is an exciting effort. I appreciate you writing to me, as there aren't many people that understand these particular algorithms. Doing key/value pair sorting can be done by moving both items together, which drops performance in half, as now there is twice as much data to move around. But, maybe it's possible to come up with a technique that doesn't move the value. Maybe the index of the value is moved instead. This would pay off for 64-bit values and arrays smaller than 4 GigaElements. It doesn't pay off for 32-bit values, since the index is also 32-bits. Take a look at my other repos: (C++) https://github.com/DragonSpit/ParallelAlgorithms (C#) https://github.com/DragonSpit/HPCsharp I have also written a book, but it is only for CPU's: https://www.amazon.com/Practical-Parallel-Algorithms-Sorting-Multicore-ebook/dp/B0C3TZPRKZ/ref=sr_1_2?crid=23440PVJS5U12&keywords=duvanenko&qid=1699848491&sprefix=duvanenko%2Caps%2C94&sr=8-2 It offers deeper and more updated explanations for some of these CPU-targetted algorithms, but not the OneSweep of Andy's and Duane's.