Concurrent queues are, arguably, one of the most important data structures in high-load applications, which require them to be extremely fast and scalable. Achieving these properties is non-trivial. The early solutions, such as the classic queue by Michael and Scott, store elements in a concurrent linked list. Reputedly, this design is non-scalable and memory-inefficient. Modern solutions utilize the Fetch-and-Add instruction to improve the algorithm’s scalability and store elements in arrays to reduce the memory pressure. One of the most famous and fast such algorithms is LCRQ. The main disadvantage of its design is that it relies on the atomic CAS2 instruction, which is unavailable in most modern programming languages, such as Java, Kotlin, or Go, let alone some architectures.

This paper presents the LPRQ algorithm, a portable modification of the original LCRQ design that eliminates all CAS2 usages. In contrast, it performs the synchronization utilizing only the standard Compare-and-Swap and Fetch-and-Add atomic instructions. Our experiments show that LPRQ provides the same performance as the classic LCRQ algorithm, outrunning the fastest of the existing solutions that do not use CAS2 by up to 1.6$\times$.