Hacker News Reader

Comment by gpderetta | Hacker News Reader

Another simple trick is duplicating the buffer pointer field and buffer size field for consumer and producer (instead of having a single vector field). Paradoxically this makes the queue object smaller as it make use of the wasted padding used for cache alignment.

View on HN

> Another simple trick is duplicating the buffer pointer field and buffer size field for consumer and producer

Do you mean something like this?

    struct Inner {
        int\* data;
        size_t size;
        size_t cachedIdx;
    };

    struct ringbuffer3 {
        // No vector<int> [...] here, saves a cache line
        alignas(64) std::atomic<size_t> readIdx_{0};
        alignas(64) Inner writeInner;
        alignas(64) std::atomic<size_t> writeIdx_{0};
        alignas(64) Inner readInner;

      // Constructor ...
    };

Am I understanding correctly?

(edit: formatting)

Yes, except that you do not want to alignas writeInner and readInner. The size of your whole ringbuffer should be 128 bytes.

It might also be better for performance since the two cores can RFO the buffer pointer cache lines from each other.

Neither producer or consumer ever writes into that line though, so it should stay in S state forever.

Reply on news.ycombinator.com