问题描述
我有一个 3D X/Y/Z 点列表 foundPoints 和单个 3D X/Y/Z 点 rayOrigin。我希望从列表中返回离 rayOrigin 最远的点。该列表可能包含数千个点，因此我想对它进行多线程处理。
我有一个现有的单线程函数，它可以正常工作并且是 O(n)。我在多线程上所做的第一次尝试是 O(2n) 因为我要浏览两次列表，我完全承认它并不漂亮但我正在努力寻找一种方法来使用多线程但在 O (n)。我怀疑答案将与 PLINQ 有关，但我非常愿意接受建议。
这是我的单线程代码
static List<double> getFarthestPoint(List<List<double>> foundPoints,double[] rayOrigin)
{
//No need to iterate if we have one point
if (foundPoints.Count == 1)
{
return foundPoints[0];
}
//Run through a bunch of output points and compare to eyePoin
List<double> farthestPoint = foundPoints[0];
double maxdistance = getPointdistance(rayOrigin,farthestPoint);
double pointdistance;
foreach (List<double> foundPoint in foundPoints)
{
pointdistance = getPointdistance(rayOrigin,foundPoint);
if (pointdistance > maxdistance)
{
maxdistance = pointdistance;
farthestPoint = foundPoint;
}
}
//Return max distance point
return farthestPoint;
}
这是我平庸的多线程代码
static List<double> getFarthestPointMultiThread(List<List<double>> foundPoints,double[] rayOrigin)
{
//No need to iterate if we have one point
if (foundPoints.Count == 1)
{
return foundPoints[0];
}
//This is multi-threaded but still O(2n) is there a way to multi-thread an O(n)?
Dictionary<Double,List<Double>> calculateddistances = new Dictionary<double,List<double>>();
//Multi thread to create dictionary of all distances
Parallel.For(0,foundPoints.Count,i =>
calculateddistances.Add(getPointdistance(rayOrigin,foundPoints[i]),foundPoints[i]));
//return point of max distance
return calculateddistances[calculateddistances.Keys.AsParallel().Max()];
}
解决方法
我对 PLINQ 不是很熟悉，但是这个调用链应该允许你分配工作量：
return foundPoints
.AsParallel()
.Select(point => (point: point,distance: getPointDistance(rayOrigin,point)))
.Aggregate((p1,p2) => p1.distance > p2.distance ? p1 : p2)
.point;
感谢 @PeterDuniho 指出我之前的基准测试设置中的一个主要缺陷。
以下是结果（2000 万点，100 次迭代，在 i7-3820 @ 3.6 GHz 上运行）：
1 core 2 cores 4 cores 8 cores
------------------- ------------------- ------------------- -------------------
Average: 1673 ms Average: 932 ms Average: 581 ms Average: 454 ms
------------------- ------------------- ------------------- -------------------
Batch 0: 1698 ms Batch 0: 1004 ms Batch 0: 595 ms Batch 0: 401 ms
Batch 1: 1670 ms Batch 1: 966 ms Batch 1: 652 ms Batch 1: 381 ms
Batch 2: 1624 ms Batch 2: 872 ms Batch 2: 624 ms Batch 2: 424 ms
Batch 3: 1639 ms Batch 3: 887 ms Batch 3: 688 ms Batch 3: 446 ms
Batch 4: 1662 ms Batch 4: 870 ms Batch 4: 596 ms Batch 4: 372 ms
Batch 5: 1650 ms Batch 5: 891 ms Batch 5: 638 ms Batch 5: 460 ms
Batch 6: 1679 ms Batch 6: 861 ms Batch 6: 594 ms Batch 6: 450 ms
Batch 7: 1699 ms Batch 7: 922 ms Batch 7: 619 ms Batch 7: 440 ms
Batch 8: 1663 ms Batch 8: 906 ms Batch 8: 576 ms Batch 8: 361 ms
Batch 9: 1662 ms Batch 9: 848 ms Batch 9: 599 ms Batch 9: 460 ms
Batch 10: 1622 ms Batch 10: 905 ms Batch 10: 605 ms Batch 10: 362 ms
Batch 11: 1646 ms Batch 11: 969 ms Batch 11: 586 ms Batch 11: 459 ms
Batch 12: 1675 ms Batch 12: 940 ms Batch 12: 573 ms Batch 12: 405 ms
Batch 13: 1598 ms Batch 13: 887 ms Batch 13: 620 ms Batch 13: 462 ms
Batch 14: 1618 ms Batch 14: 869 ms Batch 14: 572 ms Batch 14: 418 ms
Batch 15: 1597 ms Batch 15: 851 ms Batch 15: 567 ms Batch 15: 515 ms
Batch 16: 1611 ms Batch 16: 858 ms Batch 16: 587 ms Batch 16: 529 ms
Batch 17: 1602 ms Batch 17: 877 ms Batch 17: 590 ms Batch 17: 445 ms
Batch 18: 1644 ms Batch 18: 869 ms Batch 18: 607 ms Batch 18: 359 ms
Batch 19: 1615 ms Batch 19: 1021 ms Batch 19: 648 ms Batch 19: 490 ms
Batch 20: 1700 ms Batch 20: 957 ms Batch 20: 530 ms Batch 20: 454 ms
Batch 21: 1642 ms Batch 21: 902 ms Batch 21: 583 ms Batch 21: 433 ms
Batch 22: 1680 ms Batch 22: 857 ms Batch 22: 567 ms Batch 22: 513 ms
Batch 23: 1615 ms Batch 23: 894 ms Batch 23: 562 ms Batch 23: 436 ms
Batch 24: 1680 ms Batch 24: 907 ms Batch 24: 584 ms Batch 24: 371 ms
Batch 25: 1611 ms Batch 25: 851 ms Batch 25: 569 ms Batch 25: 519 ms
Batch 26: 1590 ms Batch 26: 865 ms Batch 26: 514 ms Batch 26: 385 ms
Batch 27: 1686 ms Batch 27: 934 ms Batch 27: 634 ms Batch 27: 521 ms
Batch 28: 1729 ms Batch 28: 931 ms Batch 28: 582 ms Batch 28: 415 ms
Batch 29: 1655 ms Batch 29: 863 ms Batch 29: 560 ms Batch 29: 442 ms
Batch 30: 1700 ms Batch 30: 939 ms Batch 30: 553 ms Batch 30: 420 ms
Batch 31: 1651 ms Batch 31: 892 ms Batch 31: 565 ms Batch 31: 401 ms
Batch 32: 1613 ms Batch 32: 861 ms Batch 32: 562 ms Batch 32: 511 ms
Batch 33: 1629 ms Batch 33: 841 ms Batch 33: 543 ms Batch 33: 549 ms
Batch 34: 1652 ms Batch 34: 918 ms Batch 34: 545 ms Batch 34: 551 ms
Batch 35: 1714 ms Batch 35: 954 ms Batch 35: 554 ms Batch 35: 375 ms
Batch 36: 1709 ms Batch 36: 906 ms Batch 36: 566 ms Batch 36: 415 ms
Batch 37: 1665 ms Batch 37: 898 ms Batch 37: 588 ms Batch 37: 366 ms
Batch 38: 1659 ms Batch 38: 886 ms Batch 38: 585 ms Batch 38: 500 ms
Batch 39: 1620 ms Batch 39: 903 ms Batch 39: 536 ms Batch 39: 421 ms
Batch 40: 1627 ms Batch 40: 851 ms Batch 40: 572 ms Batch 40: 542 ms
Batch 41: 1607 ms Batch 41: 864 ms Batch 41: 560 ms Batch 41: 605 ms
Batch 42: 1678 ms Batch 42: 926 ms Batch 42: 575 ms Batch 42: 561 ms
Batch 43: 1679 ms Batch 43: 890 ms Batch 43: 560 ms Batch 43: 612 ms
Batch 44: 1707 ms Batch 44: 941 ms Batch 44: 556 ms Batch 44: 531 ms
Batch 45: 1863 ms Batch 45: 1001 ms Batch 45: 567 ms Batch 45: 483 ms
Batch 46: 1667 ms Batch 46: 968 ms Batch 46: 634 ms Batch 46: 576 ms
Batch 47: 1640 ms Batch 47: 968 ms Batch 47: 567 ms Batch 47: 398 ms
Batch 48: 1666 ms Batch 48: 952 ms Batch 48: 551 ms Batch 48: 556 ms
Batch 49: 1636 ms Batch 49: 969 ms Batch 49: 566 ms Batch 49: 442 ms
Batch 50: 1675 ms Batch 50: 984 ms Batch 50: 558 ms Batch 50: 421 ms
Batch 51: 1806 ms Batch 51: 951 ms Batch 51: 571 ms Batch 51: 365 ms
Batch 52: 1768 ms Batch 52: 881 ms Batch 52: 575 ms Batch 52: 478 ms
Batch 53: 1828 ms Batch 53: 859 ms Batch 53: 550 ms Batch 53: 460 ms
Batch 54: 1743 ms Batch 54: 915 ms Batch 54: 540 ms Batch 54: 508 ms
Batch 55: 1637 ms Batch 55: 1066 ms Batch 55: 563 ms Batch 55: 488 ms
Batch 56: 1668 ms Batch 56: 1059 ms Batch 56: 570 ms Batch 56: 478 ms
Batch 57: 1760 ms Batch 57: 1110 ms Batch 57: 505 ms Batch 57: 437 ms
Batch 58: 1814 ms Batch 58: 874 ms Batch 58: 626 ms Batch 58: 426 ms
Batch 59: 1855 ms Batch 59: 893 ms Batch 59: 555 ms Batch 59: 440 ms
Batch 60: 1818 ms Batch 60: 927 ms Batch 60: 570 ms Batch 60: 432 ms
Batch 61: 1701 ms Batch 61: 932 ms Batch 61: 551 ms Batch 61: 452 ms
Batch 62: 1691 ms Batch 62: 875 ms Batch 62: 537 ms Batch 62: 389 ms
Batch 63: 1743 ms Batch 63: 966 ms Batch 63: 592 ms Batch 63: 521 ms
Batch 64: 1698 ms Batch 64: 1010 ms Batch 64: 611 ms Batch 64: 430 ms
Batch 65: 1744 ms Batch 65: 964 ms Batch 65: 563 ms Batch 65: 505 ms
Batch 66: 1687 ms Batch 66: 918 ms Batch 66: 553 ms Batch 66: 382 ms
Batch 67: 1656 ms Batch 67: 1021 ms Batch 67: 567 ms Batch 67: 457 ms
Batch 68: 1684 ms Batch 68: 965 ms Batch 68: 548 ms Batch 68: 480 ms
Batch 69: 1675 ms Batch 69: 891 ms Batch 69: 548 ms Batch 69: 516 ms
Batch 70: 1700 ms Batch 70: 893 ms Batch 70: 560 ms Batch 70: 429 ms
Batch 71: 1687 ms Batch 71: 971 ms Batch 71: 531 ms Batch 71: 486 ms
Batch 72: 1654 ms Batch 72: 1145 ms Batch 72: 555 ms Batch 72: 394 ms
Batch 73: 1669 ms Batch 73: 924 ms Batch 73: 554 ms Batch 73: 477 ms
Batch 74: 1637 ms Batch 74: 1023 ms Batch 74: 591 ms Batch 74: 481 ms
Batch 75: 1653 ms Batch 75: 1108 ms Batch 75: 639 ms Batch 75: 404 ms
Batch 76: 1618 ms Batch 76: 978 ms Batch 76: 627 ms Batch 76: 464 ms
Batch 77: 1711 ms Batch 77: 969 ms Batch 77: 620 ms Batch 77: 392 ms
Batch 78: 1647 ms Batch 78: 953 ms Batch 78: 593 ms Batch 78: 501 ms
Batch 79: 1724 ms Batch 79: 898 ms Batch 79: 574 ms Batch 79: 385 ms
Batch 80: 1680 ms Batch 80: 945 ms Batch 80: 622 ms Batch 80: 381 ms
Batch 81: 1630 ms Batch 81: 973 ms Batch 81: 555 ms Batch 81: 448 ms
Batch 82: 1618 ms Batch 82: 906 ms Batch 82: 828 ms Batch 82: 441 ms
Batch 83: 1703 ms Batch 83: 958 ms Batch 83: 580 ms Batch 83: 436 ms
Batch 84: 1683 ms Batch 84: 920 ms Batch 84: 636 ms Batch 84: 467 ms
Batch 85: 1618 ms Batch 85: 1032 ms Batch 85: 588 ms Batch 85: 427 ms
Batch 86: 1695 ms Batch 86: 947 ms Batch 86: 547 ms Batch 86: 410 ms
Batch 87: 1691 ms Batch 87: 1008 ms Batch 87: 561 ms Batch 87: 393 ms
Batch 88: 1764 ms Batch 88: 953 ms Batch 88: 587 ms Batch 88: 510 ms
Batch 89: 1704 ms Batch 89: 935 ms Batch 89: 568 ms Batch 89: 464 ms
Batch 90: 1631 ms Batch 90: 962 ms Batch 90: 570 ms Batch 90: 435 ms
Batch 91: 1613 ms Batch 91: 983 ms Batch 91: 569 ms Batch 91: 439 ms
Batch 92: 1604 ms Batch 92: 930 ms Batch 92: 582 ms Batch 92: 428 ms
Batch 93: 1592 ms Batch 93: 926 ms Batch 93: 507 ms Batch 93: 518 ms
Batch 94: 1596 ms Batch 94: 946 ms Batch 94: 613 ms Batch 94: 444 ms
Batch 95: 1688 ms Batch 95: 894 ms Batch 95: 572 ms Batch 95: 492 ms
Batch 96: 1613 ms Batch 96: 939 ms Batch 96: 586 ms Batch 96: 490 ms
Batch 97: 1651 ms Batch 97: 915 ms Batch 97: 606 ms Batch 97: 489 ms
Batch 98: 1689 ms Batch 98: 946 ms Batch 98: 639 ms Batch 98: 482 ms
Batch 99: 1680 ms Batch 99: 953 ms Batch 99: 614 ms Batch 99: 400 ms
------------------- ------------------- ------------------- -------------------
测试代码：
double CalculateDistance(List<double> p1,List<double> p2) {
double dx = p2[0] - p1[0];
double dy = p2[1] - p1[1];
double dz = p2[2] - p1[2];
return Math.Sqrt(dx * dx + dy * dy + dz * dz);
}
var initTime = Stopwatch.StartNew();
var origin = new List<double> { 20,40,60 };
var points = Enumerable
.Range(0,20000000)
.Select(i => new List<double> { i,i,i })
.ToList();
WriteLine($"Initialization: {initTime.ElapsedMilliseconds} ms");
var times = new List<long>();
for(int i = 0; i < 100; i++) {
var sw = Stopwatch.StartNew();
points
.AsParallel()
.WithDegreeOfParallelism(8)
.Select(point => (point: point,distance: CalculateDistance(point,origin)))
.Aggregate((p1,p2) => p1.distance > p2.distance ? p1 : p2);
long time = sw.ElapsedMilliseconds;
WriteLine($"Batch {i}: {time} ms");
times.Add(time);
}
WriteLine($"Average: {times.Average()} ms");