浅谈 Cache

先来看一段代码：

#include <Windows.h>
#include <stdio.h>

int x, y, r1, r2;

HANDLE Event1, Event2, EventMain1, EventMain2;

DWORD
_stdcall
ThreadProc1 (
    LPVOID lpThreadParameter
    )
{
    while (TRUE) {
        WaitForSingleObject(Event1, INFINITE);

        while (rand() % 8 != 0);

        x = 1;
        r1 = y;

        SetEvent(EventMain1);
    }
}

DWORD
_stdcall
ThreadProc2 (
    LPVOID lpThreadParameter
    )
{
    while (TRUE) {
        WaitForSingleObject(Event2, INFINITE);

        while (rand() % 8 != 0);

        y = 1;
        r2 = x;

        SetEvent(EventMain2);
    }
}

void main()
{
    HANDLE Thread1, Thread2;

    Thread1 = CreateThread(NULL, 0, ThreadProc1, NULL, 0, NULL);
    Thread2 = CreateThread(NULL, 0, ThreadProc2, NULL, 0, NULL);
    SetThreadAffinityMask(Thread1, 1);
    SetThreadAffinityMask(Thread2, 2);

    Event1 = CreateEvent(NULL, FALSE, FALSE, NULL);
    Event2 = CreateEvent(NULL, FALSE, FALSE, NULL);
    EventMain1 = CreateEvent(NULL, FALSE, FALSE, NULL);
    EventMain2 = CreateEvent(NULL, FALSE, FALSE, NULL);

    int dectected = 0;
    for (int i = 0; ;i++) {
        x = 0;
        y = 0;

        SetEvent(Event1);
        SetEvent(Event2);

        WaitForSingleObject(EventMain1, INFINITE);
        WaitForSingleObject(EventMain2, INFINITE);

        if (r1 == 0 && r2 == 0) {
            printf("detected %d, i %d\n", dectected++, i);
        }
    }
}

描述一下这段代码的重点，起两个线程在两个cpu 核心上执行，分别执行 x = 1, r1= y; y = 1, r2 = x；无论哪个线程先执行，或一起执行，按照常理执行结果一般为 r1 = 1 或 r2 = 1 或者两个都为 1，也就是在 main 函数中的判断 r1 和 r2 同为 0 的判断是不可能成立的，打印也不会出现。

当你真实在一台至少双核机器上运行时，你应该不会相信你的眼睛。

detected 305, i 202654
detected 306, i 204704
detected 307, i 205029
detected 308, i 206454
detected 309, i 207135
detected 310, i 208178
detected 311, i 208610
detected 312, i 208879
detected 313, i 208896
detected 314, i 208960
detected 315, i 209003
detected 316, i 209869
detected 317, i 210573
detected 318, i 211259
detected 319, i 211275
detected 320, i 211487
detected 321, i 212646
detected 322, i 212963
detected 323, i 213317
detected 324, i 213382
detected 325, i 213473
detected 326, i 213588
detected 327, i 214940
detected 328, i 215055
detected 329, i 215271

（由于贴图不方便，直接把本机的打印复制下来了）

从打印中可以得出，在两个线程运行完后，r1== r2 == 0 的情况居然出现了，难道 CPU 在逗我们么。

CPU 没有逗我们，是我们不够了解它。（待续）