"Chris M. Thomasson" <cristom@charter.net> wrote in message news:hgfmj4$8m0$1@ger.gmane.org...
"Tim Blechmann" <tim@klingt.org> wrote in message news:4B23A083.7080008@klingt.org...
i just went through the replies ... (maybe i should upcase some parts in the documentation, that the implementation focus on WORST CASE, not AVERAGE CASE performance ... people keep complaining that the stack/fifo may be outperformed be blocking algorithms, which is both true and irrelevant for me, as these implementations are soft real-time safe (and could be made hard real-time safe).
Agreed. Are you planning to augment your library with algorithms for different needs? Like MPSC, SPMC and so forth? I think that could be useful, because it looks like implementing the most generic MPMC case won't give you the best results.
i just added a lock-free single-producer/single-consumer ringbuffer to my git repository, branch topic/spsc_ringbuffer ... it implements an algorithm, that is commonly found in open source audio applications ... performs about 8 times faster than the m-s fifo queue (which is based on linked lists).
FWIW Tim, here is a very simple pseudo-code implementation of a nice single-producer/consumer bounded queue: _______________________________________________________________ template<template T, size_t T_depth> class spsc_queue { atomic<T*> m_buffer[T_depth]; // = { NULL } size_t m_head; // = 0 size_t m_tail; // = 0
public: bool push(T* object) { if (m_buffer[m_head].load(memory_order_relaxed)) { return false; }
m_buffer[m_head].store(memory_order_release);
Ummm, that should be m_buffer[m_head].store(object, memory_order_release); of course!!! Sorry about that.