C++ 代码显然乱序执行

Question

代码

这是 Raspberry Pi 上使用 WiringPi 的项目。我有以下三个模板 class 的成员函数，以及 read() 和 write() 的纯虚函数。这个基础 class 然后由提供 read() 和 write() 功能的更专业的 class 子class 编辑（示例如下所示）：

// IChip.hpp (Root abstract class)
class IChip {
    public:
    virtual bool test() noexcept = 0;
};
// End IChip.hpp

// IMemory.hpp (class of interest to the question)
class IMemory: public IChip {
    protected:
    ...
    TAddr m_wordCount;
    TWord m_dataMax;
    // ctor and dtor, and more member fields

    public:
    virtual TWord read(const TAddr addr) const noexcept = 0;
    virtual void write(const TAddr addr, const TWord data) const noexcept = 0;

    // accessors and whatnot ...

    bool march(bool keepGoing = false) noexcept;
    bool checkerboard(bool keepGoing = false) noexcept;

    bool test() noexcept final override;
};
// End IMemory.hpp


// IMemory.cpp
template <typename TAddr, typename TWord>
bool IMemory<TAddr, TWord>::march(bool keepGoing) noexcept {
    bool result = true;
    TAddr i;
    TWord r;
    const uint64_t totalIter = (m_wordCount * 6) - 1;
    uint64_t counter = 0;

    std::cout << "Starting MARCH test." << std::endl;

    for (i = 0; i < m_wordCount; i++) {
        this->write(i, 0);
        std::cout << '\r' << counter << " / " << totalIter << std::flush;
        counter++;
    }

    for (i = 0; i < m_wordCount; i++) {
        r = this->read(i);
        if (r != 0) {
            result = false;
            if (!keepGoing)
                return result;
        }
        this->write(i, m_dataMax);
        std::cout << '\r' << counter << " / " << totalIter << std::flush;
        counter++;
    }

    // 4 more similar loops

    std::cout << std::endl;

    std::cout << "MARCH test done." << std::endl;

    return result;
}

template <typename TAddr, typename TWord>
bool IMemory<TAddr, TWord>::checkerboard(bool keepGoing) noexcept {
    bool result = true;
    TAddr i;
    TWord r;
    TWord curWord;
    const uint64_t totalIter = (m_wordCount * 4) - 1;
    uint64_t counter = 0;

    std::cout << "Starting CHECKERBOARD test." << std::endl;

    curWord = 0;
    for (i = 0; i < m_wordCount; i++) {
        this->write(i, curWord);
        std::cout << '\r' << counter << " / " << totalIter << std::flush;
        counter++;
        curWord = curWord == 0 ? m_dataMax : 0;
    }

    curWord = 0;
    for (i = 0; i < m_wordCount; i++) {
        r = this->read(i);
        if (r != curWord) {
            result = false;
            if (!keepGoing)
                return result;
        }
        std::cout << '\r' << counter << " / " << totalIter << std::flush;
        counter++;
        curWord = curWord == 0 ? m_dataMax : 0;
    }

    // 2 more similar loops ...

    std::cout << std::endl;
    std::cout << "CHECKERBOARD test done." << std::endl;

    return result;
}

template <typename TAddr, typename TWord>
bool IMemory<TAddr, TWord>::test() noexcept {
    bool march_result = this->march();
    bool checkerboard_result = this->checkerboard();
    bool result = march_result && checkerboard_result;

    std::cout << "MARCH: " << (march_result ? "Passed" : "Failed") << std::endl;
    std::cout << "CHECKERBOARD: " << (checkerboard_result ? "Passed" : "Failed") << std::endl;
    return result;
}

// Explicit instantiation
template class IMemory<uint16_t, uint8_t>;
// End IMemory.cpp


// Sample read() and write() from HM62256, a subclass of IMemory<uint16_t, uint8_t>
// These really just bitbang onto / read data from pins with appropriate timings for each chip.
// m_data and m_address are instances of a Bus class, that is just a wrapper around an array of pins, provides bit-banging and reading functionality.

uint8_t HM62256::read(uint16_t addr) const noexcept {
    uint8_t result = 0;

    m_data->setMode(INPUT);
    m_address->write(addr);
    digitalWrite(m_CSPin, LOW);
    digitalWrite(m_OEPin, LOW);
    delayMicroseconds(1);
    result = m_data->read();
    digitalWrite(m_OEPin, HIGH);
    digitalWrite(m_CSPin, HIGH);
    delayMicroseconds(1);
    return result;
}

void HM62256::write(uint16_t addr, uint8_t data) const noexcept {
    digitalWrite(m_OEPin, HIGH);
    delayMicroseconds(1);

    m_address->write(addr);
    delayMicroseconds(1);

    m_data->setMode(OUTPUT);
    m_data->write(data);

    digitalWrite(m_CSPin, LOW);
    digitalWrite(m_WEPin, LOW);
    delayMicroseconds(1);

    digitalWrite(m_WEPin, HIGH);
    digitalWrite(m_CSPin, HIGH);
    delayMicroseconds(1);
}

// main.cpp
void hm62256_test() {
    const uint8_t ADDR_PINS[] = {4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18};
    const uint8_t DATA_PINS[] = {19, 20, 21, 22, 23, 24, 25, 26};
    Chiptools::Memory::HM62256 *device = new Chiptools::Memory::HM62256(ADDR_PINS, DATA_PINS, 2, 3, 27);

    device->setup();
    bool result = device->test();
    std::cout << "Device " << ( result ? "passed all" : "failed some") << " tests." << std::endl;
    delete device;
}

int main(int argc, char *argv[]) {
    wiringPiSetupGpio();
    hm62256_test();
}

输出

现在当我运行这个时，有时它工作得很好:

Starting MARCH test.
196607 / 196607
MARCH test done.
Starting CHECKERBOARD test.
131071 / 131071
CHECKERBOARD test done.
MARCH: Passed
CHECKERBOARD: Passed
Device passed all tests.

但随机我会得到这个输出：

Starting MARCH test.
67113 / 196607Starting CHECKERBOARD test.
33604 / 131071MARCH: Failed
CHECKERBOARD: Failed
Device failed some tests.

工具链信息
- gcc 8.3.0 arm-linux / C++14
- CMake 3.16.3
- 没有线程。
- 编译器和链接器标志：

set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -fsanitize=address,leak,undefined")
set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -fsanitize=address,leak,undefined  -static-libasan")

问题和我尝试过的方法

我有几十个筹码。所有芯片都可以在 TL866ii 编程器/测试器上正常工作。所有这些芯片都会发生这种情况。这样就排除了芯片是问题的根源。

好吧，起初我想也许我没有正确地刷新 cout 流，但是 AFAIK std::endl 确实刷新了输出，所以不是那样。

接下来，我设置了几个断点：(A)就在march()之前returns，(B) 就在调用 checkerboard() 的地方（test() 中的第 2 行），(C) 在 checkerboard().[=26= 中的第 1 行]

当输出符合预期时，按此顺序命中断点 A、B、C。
当输出不符合预期时，按此顺序命中断点B，C，A。

看起来正在发生的事情是，有时 checkerboard() 被调用而 march() 仍然是运行ning，导致随机 GPIO 输出，此时一个或两个测试失败。

虽然我正在寻找解决方案，但我更感兴趣的是了解正在发生的事情。我会认为，由于我的代码没有使用多线程，并且根据我对 C++ 标准的理解，在执行下一条语句之前，语句会一个接一个地执行到完成。我知道一些编译器实现会重新排序语句以进行优化，但 AFAIK 它不应该影响我的代码的语义。我可能错了，因为这些东西让我难以理解。

Answer 1

这可能不是答案，但评论太长了。

A 是否在“March test done”行之后的 return 语句中？

我基于此输出做出以下评论：

Starting MARCH test.
67113 / 196607Starting CHECKERBOARD test.
33604 / 131071MARCH: Failed
CHECKERBOARD: Failed
Device failed some tests.

似乎正在发生的事情是您的 MARCH 测试在第 3 个循环中失败，因此 returning 提前（在循环内）。然后，您的棋盘测试在第二个循环中失败，并且提前 returns。如果 A 位于我提到的位置，那么我认为命中该断点只是运气或编译器怪癖。

也就是说，从逻辑上讲，我不希望在发生故障时命中断点 A，仅针对 B 和 C。我认为 A 最后被击中可能是因为它的程序是如何编译的，也许还有一些奇怪的优化。或者也许调试器在程序集中放置断点的位置；它可能只是在无论如何都会被调用的最终指令上。尝试将断点放在 return 之前的 std::cout 行，看看它是否仍然命中。

为了扩展您的评论，这是我在问题输出中看到的内容：

Starting MARCH test.
67113 / 196607 [march() returns early] [checkerboard() starts] Starting CHECKERBOARD test.
33604 / 131071 [checkerboard() returns early] [test() reports results] MARCH: Failed
CHECKERBOARD: Failed
Device failed some tests.

总而言之，我认为如果您更改 return 行，输出将符合您的期望：

if (!keepGoing) 
    return result;

像这样：

if (!keepGoing) {
    std::cout << std::endl;
    std::cout << "MARCH test failed." << std::endl;
    return result;
}

我希望产生这样的输出：

Starting MARCH test.
67113 / 196607
MARCH test failed
Starting CHECKERBOARD test.
33604 / 131071
CHECKERBOARD test failed
MARCH: Failed
CHECKERBOARD: Failed
Device failed some tests.

C++ 代码显然乱序执行

C++ code apparently executing out of sequence

c++

templating

raspberry-pi

wiringpi