如何快速实现并行地图

Question

我有以下代码：-

extension Collection {

    // EZSE : A parralelized map for collections, operation is non blocking
    public func pmap<R>(_ each: (Self.Iterator.Element) -> R) -> [R?] {
        let indices = indicesArray()
        var res = [R?](repeating: nil, count: indices.count)

        DispatchQueue.concurrentPerform(iterations: indices.count) { (index) in
            let elementIndex = indices[index]
            res[index] = each(self[elementIndex])
        }

        // Above code is non blocking so partial exec on most runs
        return res
    }

    /// EZSE : Helper method to get an array of collection indices
    private func indicesArray() -> [Self.Index] {
        var indicesArray: [Self.Index] = []
        var nextIndex = startIndex
        while nextIndex != endIndex {
            indicesArray.append(nextIndex)
            nextIndex = index(after: nextIndex)
        }
        return indicesArray
    }
}

在此处的return语句res中，它经常在部分执行完成的情况下返回。有道理，并发 Perform 是非阻塞的。我不确定如何继续等待它。我应该使用调度组/期望之类的东西还是有一些更简单更优雅的方法？本质上，我正在快速寻找一个简单的等待通知抽象。

Answer 1

@user28434 的回答很棒，使用concurrentPerform 可以更快：

extension Collection {
    func parallelMap<R>(_ transform: @escaping (Element) -> R) -> [R] {
        var res: [R?] = .init(repeating: nil, count: count)

        let lock = NSRecursiveLock()
        DispatchQueue.concurrentPerform(iterations: count) { i in
            let result = transform(self[index(startIndex, offsetBy: i)])
            lock.lock()
            res[i] = result
            lock.unlock()
        }

        return res.map({ $0! })
    }
}

它与在我正在运行的任务上使用 OperationQueue() 相比（在 10,000 个项目上并行运行我的某些功能）平均使用 100 次运行：

• concurrentPerform 平均耗时 0.060 秒。
• OperationQueue() 平均耗时 0.087 秒。

Answer 2

你可以试试这样的：

// EZSE : A parralelized map for collections, operation is non blocking
public func pmap<R>(_ each: @escaping (Self.Iterator.Element) -> R) -> [R] {
    let totalCount = indices.count
    var res = ContiguousArray<R?>(repeating: nil, count: totalCount)
    
    let queue = OperationQueue()
    queue.maxConcurrentOperationCount = totalCount
    
    let lock = NSRecursiveLock()
    indices
        .enumerated()
        .forEach { index, elementIndex in
            queue.addOperation {
                let temp = each(self[elementIndex])
                lock.lock()
                res[index] = temp
                lock.unlock()
            }
    }
    queue.waitUntilAllOperationsAreFinished()
    
    return res.map({ $0! })
}