Erik Rigtorp
Erik Rigtorp
There can only be one writer thread.
@X-Ryl669 1. That will work, but it as a stronger operation than needed and will be slower. 2. Only a single writer is supported.
@MengRao You're solution will work on TSO memory model, it will not work for ARM. Even for amd64 your solution and mine will not work if the assignment operator uses...
@MengRao It's also needed to align and pad the seqlock to avoid false sharing. If you have array/vector of seqlocks with multiple writer threads for different seqlocks you need to...
> > @MengRao It's also needed to align and pad the seqlock to avoid false sharing. If you have array/vector of seqlocks with multiple writer threads for different seqlocks you...
@X-Ryl669 Yes it's possible to implement other types of locks, but then it's no longer a seqlock. The whole point here was to implement seqlocks.
> will this implementation work on x86_64 or only on x86 ?? Both
@infrastation I fixed the manpage and rebased!
Isn't this only used for backward compatibility with pre C++20 allocators? In C++20 `construct`/`destroy` is no longer an extension point?
There seems to be a misunderstanding of the pre-conditions of calling `pop()`. I updated the documentation here: https://github.com/rigtorp/SPSCQueue/commit/40661138be2245d7d52854e4fb3b7f6bb4bf02ad This is a concurrent queue intended to be used by two threads....