将 std::accumulate 与 constexpr 一起使用

Question

原来的

给定一个std::array<std::variant<>>我得到总字节大小（例如{float,float,uint32_t}=4+4+4=12 ）然后我想得到数组的形式一个void* （传递给 Vulkan）。

我不相信我需要为此分配堆，尽管我坚持如何将计算出的大小作为模板参数传递。

Repl： https://repl.it/@JonathanWoollet/HumongousDistinctDefinitions

错误：

> clang++-7 -pthread -std=c++17 -o main main.cpp
main.cpp:42:30: error: constexpr variable 'size' must be initialized by a
      constant expression
    constexpr uint32_t const size = arrSize<num>(arr);
                             ^      ~~~~~~~~~~~~~~~~~
main.cpp:17:34: note: non-constexpr function 'accumulate<const
      std::variant<unsigned int, float> *, unsigned long, (lambda at
      main.cpp:19:9)>' cannot be used in a constant expression
    return static_cast<uint32_t>(std::accumulate(arr.cbegin(),arr.cend(),
                                 ^
main.cpp:42:37: note: in call to 'arrSize(arr)'
    constexpr uint32_t const size = arrSize<num>(arr);
                                    ^
/usr/bin/../lib/gcc/x86_64-linux-gnu/8/../../../../include/c++/8/bits/stl_numeric.h:146:5: note: 
      declared here
    accumulate(_InputIterator __first, _InputIterator __last, _Tp __init,
    ^
main.cpp:44:21: error: no matching function for call to 'toVoidPtr'
    void* voidArr = toVoidPtr<num,size>(arr);
                    ^~~~~~~~~~~~~~~~~~~
main.cpp:25:7: note: candidate template ignored: invalid explicitly-specified
      argument for template parameter 'Size'
void* toVoidPtr(std::array<std::variant<uint32_t,float>,Num> const& arr) {
      ^
2 errors generated.
compiler exit status 1

（用 C++20 编译会导致同样的错误）

我已经看到编译时相当于 std::accumulate()但https://en.cppreference.com/w/cpp/algorithm/accumulate清楚地显示了一个constexpr定义：

template< class InputIt, class T, class BinaryOperation >
constexpr T accumulate( InputIt first, InputIt last, T init,
                        BinaryOperation op );

我通过以下方式获得尺寸：

template <uint32_t Num>
constexpr uint32_t arrSize(std::array<std::variant<uint32_t,float>,Num> const& arr) {
    auto var_size = [](auto const& var) -> size_t {
        using T = std::decay_t<decltype(var)>;
        return sizeof(T);
    };
    return static_cast<uint32_t>(std::accumulate(arr.cbegin(),arr.cend(),
        std::size_t{ 0 },
        [var_size](std::size_t acc, auto const var) { return acc + std::visit(var_size,var); }
    ));
}

我通过以下方式转换为void* ：

template <uint32_t Num, uint32_t Size>
void* toVoidPtr(std::array<std::variant<uint32_t,float>,Num> const& arr) {
    size_t byteCounter = size_t{0};
    std::array<std::byte,Size> bytes;
    std::for_each(arr.cbegin(),arr.cend(), [&](auto const& var) {
        std::visit([&] (auto const& var) {
            using T = std::decay_t<decltype(var)>;
            std::memcpy(bytes.data()+byteCounter,static_cast<void const*>(&var),sizeof(T));
            byteCounter += sizeof(T);
        },var);
    });
    return static_cast<void*>(bytes.data);
}

我想遵循这样的命令：

int main() {
    std::cout << "Hello World!\n";

    const uint32_t num = 3U;
    std::array<std::variant<uint32_t,float>,num> arr = { 1.34F, 3.76F, 124U };
    constexpr uint32_t const size = arrSize<num>(arr);
    std::cout << "size: " << size << std::endl;
    void* voidArr = toVoidPtr<num,size>(arr);
}

我不确定我在这里错过了什么，而且我似乎无法弄清楚。 我真的很感激任何帮助。

（我遗漏了什么请发表评论，我会尝试添加它）

---

使用gcc 10.2.0和c++20

按照Werner Henze的建议更新为使用兼容的构建工具，并将我的本地项目更新为使用gcc 10.2.0 Henze 和c++20 ，我得到了不同的错误。

[build] In file included from C:\Users\jonat\Projects\gpu_blas\c++\test\ExampleTests.cpp:2:
[build] c:\users\jonat\projects\gpu_blas\c++\example.hpp: In instantiation of 'ComputeApp<NumPushConstants>::ComputeApp(const char*, uint32_t, const uint32_t*, float**, std::array<std::variant<unsigned int, float>, NumPushConstants>, std::array<unsigned int, 3>, std::array<unsigned int, 3>) [with unsigned int NumPushConstants = 1; uint32_t = unsigned int]':
[build] C:\Users\jonat\Projects\gpu_blas\c++\test\ExampleTests.cpp:41:5:   required from here
[build] c:\users\jonat\projects\gpu_blas\c++\example.hpp:367:95:   in 'constexpr' expansion of 'Utility::pushConstantsSize<1>(pushConstant)'
[build] c:\users\jonat\projects\gpu_blas\c++\example.hpp:102:53:   in 'constexpr' expansion of 'std::accumulate<const std::variant<unsigned int, float>*, long long unsigned int, Utility::pushConstantsSize<1>::<lambda(std::size_t, auto:43)> >((& pushConstants)->std::array<std::variant<unsigned int, float>, 1>::cbegin(), (& pushConstants)->std::array<std::variant<unsigned int, float>, 1>::cend(), 0, <lambda closure object>Utility::pushConstantsSize<1>::<lambda(std::size_t, auto:43)>{<lambda closure object>Utility::pushConstantsSize<1>::<lambda(const auto:42&)>()})'
[build] c:\users\jonat\projects\gpu_blas\c++\example.hpp:367:38: error: 'pushConstant' is not a constant expression
[build]   367 |             constexpr uint32_t const sizeTester = Utility::pushConstantsSize<NumPushConstants>(pushConstant);
[build]       |                                      ^~~~~~~~~~
[build] c:\users\jonat\projects\gpu_blas\c++\example.hpp: In instantiation of 'ComputeApp<NumPushConstants>::ComputeApp(const char*, uint32_t, const uint32_t*, float**, std::array<std::variant<unsigned int, float>, NumPushConstants>, std::array<unsigned int, 3>, std::array<unsigned int, 3>) [with unsigned int NumPushConstants = 3; uint32_t = unsigned int]':
[build] C:\Users\jonat\Projects\gpu_blas\c++\test\ExampleTests.cpp:849:5:   required from here
[build] c:\users\jonat\projects\gpu_blas\c++\example.hpp:367:95:   in 'constexpr' expansion of 'Utility::pushConstantsSize<3>(pushConstant)'
[build] c:\users\jonat\projects\gpu_blas\c++\example.hpp:102:53:   in 'constexpr' expansion of 'std::accumulate<const std::variant<unsigned int, float>*, long long unsigned int, Utility::pushConstantsSize<3>::<lambda(std::size_t, auto:43)> >((& pushConstants)->std::array<std::variant<unsigned int, float>, 3>::cbegin(), (& pushConstants)->std::array<std::variant<unsigned int, float>, 3>::cend(), 0, <lambda closure object>Utility::pushConstantsSize<3>::<lambda(std::size_t, auto:43)>{<lambda closure object>Utility::pushConstantsSize<3>::<lambda(const auto:42&)>()})'
[build] c:\users\jonat\projects\gpu_blas\c++\example.hpp:367:38: error: 'pushConstant' is not a constant expression
[build] mingw32-make[2]: *** [test\CMakeFiles\ExampleTests.dir\build.make:82: test/CMakeFiles/ExampleTests.dir/ExampleTests.cpp.obj] Error 1
[build] mingw32-make[1]: *** [CMakeFiles\Makefile2:315: test/CMakeFiles/ExampleTests.dir/all] Error 2
[build] mingw32-make: *** [makefile:159: all] Error 2
[build] Build finished with exit code 2

我在这里玩了更多，但我找不到问题所在，我再次非常感谢任何帮助。

由于我无法提供 Repl 以最低限度地重现此问题，因此这里是 CMake 项目的链接，当前分支产生这些错误： https://github.com/JonathanWoollett-Light/gpu_blas/tree/Compiler-update/c%2B %2B

Answer 1

cppreference accumulate说accumulate只是constexpr since C++20, not in C++17. 所以你需要切换到 C++20。

如果您想知道哪个编译器已经支持constexpr accumulate ，您可以查看cppreference compiler support并搜索“constexpr for numeric algorithms”/P1645R1。 例如，对于 clang，它需要 libc++ 12。

Answer 2

原来的

给定一个std::array<std::variant<>>我得到总字节大小（例如{float,float,uint32_t}=4+4+4=12 ），然后我想以以下形式获取数组一个void* （传递给 Vulkan）。

我不相信我需要为此进行堆分配，尽管我坚持如何将计算出的大小作为模板参数传递。

副本：https://repl.it/@JonathanWoollet/HumongousDistinctDefinitions

错误：

> clang++-7 -pthread -std=c++17 -o main main.cpp
main.cpp:42:30: error: constexpr variable 'size' must be initialized by a
      constant expression
    constexpr uint32_t const size = arrSize<num>(arr);
                             ^      ~~~~~~~~~~~~~~~~~
main.cpp:17:34: note: non-constexpr function 'accumulate<const
      std::variant<unsigned int, float> *, unsigned long, (lambda at
      main.cpp:19:9)>' cannot be used in a constant expression
    return static_cast<uint32_t>(std::accumulate(arr.cbegin(),arr.cend(),
                                 ^
main.cpp:42:37: note: in call to 'arrSize(arr)'
    constexpr uint32_t const size = arrSize<num>(arr);
                                    ^
/usr/bin/../lib/gcc/x86_64-linux-gnu/8/../../../../include/c++/8/bits/stl_numeric.h:146:5: note: 
      declared here
    accumulate(_InputIterator __first, _InputIterator __last, _Tp __init,
    ^
main.cpp:44:21: error: no matching function for call to 'toVoidPtr'
    void* voidArr = toVoidPtr<num,size>(arr);
                    ^~~~~~~~~~~~~~~~~~~
main.cpp:25:7: note: candidate template ignored: invalid explicitly-specified
      argument for template parameter 'Size'
void* toVoidPtr(std::array<std::variant<uint32_t,float>,Num> const& arr) {
      ^
2 errors generated.
compiler exit status 1

（用 C++20 编译会导致同样的错误）

我已经看到编译时等效于 std::accumulate()但https://en.cppreference.com/w/cpp/algorithm/accumulate清楚地显示了constexpr定义：

template< class InputIt, class T, class BinaryOperation >
constexpr T accumulate( InputIt first, InputIt last, T init,
                        BinaryOperation op );

我通过以下方式获得尺寸：

template <uint32_t Num>
constexpr uint32_t arrSize(std::array<std::variant<uint32_t,float>,Num> const& arr) {
    auto var_size = [](auto const& var) -> size_t {
        using T = std::decay_t<decltype(var)>;
        return sizeof(T);
    };
    return static_cast<uint32_t>(std::accumulate(arr.cbegin(),arr.cend(),
        std::size_t{ 0 },
        [var_size](std::size_t acc, auto const var) { return acc + std::visit(var_size,var); }
    ));
}

我通过以下方式转换为void* ：

template <uint32_t Num, uint32_t Size>
void* toVoidPtr(std::array<std::variant<uint32_t,float>,Num> const& arr) {
    size_t byteCounter = size_t{0};
    std::array<std::byte,Size> bytes;
    std::for_each(arr.cbegin(),arr.cend(), [&](auto const& var) {
        std::visit([&] (auto const& var) {
            using T = std::decay_t<decltype(var)>;
            std::memcpy(bytes.data()+byteCounter,static_cast<void const*>(&var),sizeof(T));
            byteCounter += sizeof(T);
        },var);
    });
    return static_cast<void*>(bytes.data);
}

我想遵循如下命令：

int main() {
    std::cout << "Hello World!\n";

    const uint32_t num = 3U;
    std::array<std::variant<uint32_t,float>,num> arr = { 1.34F, 3.76F, 124U };
    constexpr uint32_t const size = arrSize<num>(arr);
    std::cout << "size: " << size << std::endl;
    void* voidArr = toVoidPtr<num,size>(arr);
}

我不确定我在这里缺少什么，我似乎无法弄清楚。 我真的很感激任何帮助。

（我缺少任何东西，请发表评论，我会尝试添加它）

---

使用gcc 10.2.0 & c++20

按照Werner Henze的建议更新以使用兼容的构建工具，并更新我的本地项目以使用gcc 10.2.0 Henze 和c++20后，我得到了不同的错误。

[build] In file included from C:\Users\jonat\Projects\gpu_blas\c++\test\ExampleTests.cpp:2:
[build] c:\users\jonat\projects\gpu_blas\c++\example.hpp: In instantiation of 'ComputeApp<NumPushConstants>::ComputeApp(const char*, uint32_t, const uint32_t*, float**, std::array<std::variant<unsigned int, float>, NumPushConstants>, std::array<unsigned int, 3>, std::array<unsigned int, 3>) [with unsigned int NumPushConstants = 1; uint32_t = unsigned int]':
[build] C:\Users\jonat\Projects\gpu_blas\c++\test\ExampleTests.cpp:41:5:   required from here
[build] c:\users\jonat\projects\gpu_blas\c++\example.hpp:367:95:   in 'constexpr' expansion of 'Utility::pushConstantsSize<1>(pushConstant)'
[build] c:\users\jonat\projects\gpu_blas\c++\example.hpp:102:53:   in 'constexpr' expansion of 'std::accumulate<const std::variant<unsigned int, float>*, long long unsigned int, Utility::pushConstantsSize<1>::<lambda(std::size_t, auto:43)> >((& pushConstants)->std::array<std::variant<unsigned int, float>, 1>::cbegin(), (& pushConstants)->std::array<std::variant<unsigned int, float>, 1>::cend(), 0, <lambda closure object>Utility::pushConstantsSize<1>::<lambda(std::size_t, auto:43)>{<lambda closure object>Utility::pushConstantsSize<1>::<lambda(const auto:42&)>()})'
[build] c:\users\jonat\projects\gpu_blas\c++\example.hpp:367:38: error: 'pushConstant' is not a constant expression
[build]   367 |             constexpr uint32_t const sizeTester = Utility::pushConstantsSize<NumPushConstants>(pushConstant);
[build]       |                                      ^~~~~~~~~~
[build] c:\users\jonat\projects\gpu_blas\c++\example.hpp: In instantiation of 'ComputeApp<NumPushConstants>::ComputeApp(const char*, uint32_t, const uint32_t*, float**, std::array<std::variant<unsigned int, float>, NumPushConstants>, std::array<unsigned int, 3>, std::array<unsigned int, 3>) [with unsigned int NumPushConstants = 3; uint32_t = unsigned int]':
[build] C:\Users\jonat\Projects\gpu_blas\c++\test\ExampleTests.cpp:849:5:   required from here
[build] c:\users\jonat\projects\gpu_blas\c++\example.hpp:367:95:   in 'constexpr' expansion of 'Utility::pushConstantsSize<3>(pushConstant)'
[build] c:\users\jonat\projects\gpu_blas\c++\example.hpp:102:53:   in 'constexpr' expansion of 'std::accumulate<const std::variant<unsigned int, float>*, long long unsigned int, Utility::pushConstantsSize<3>::<lambda(std::size_t, auto:43)> >((& pushConstants)->std::array<std::variant<unsigned int, float>, 3>::cbegin(), (& pushConstants)->std::array<std::variant<unsigned int, float>, 3>::cend(), 0, <lambda closure object>Utility::pushConstantsSize<3>::<lambda(std::size_t, auto:43)>{<lambda closure object>Utility::pushConstantsSize<3>::<lambda(const auto:42&)>()})'
[build] c:\users\jonat\projects\gpu_blas\c++\example.hpp:367:38: error: 'pushConstant' is not a constant expression
[build] mingw32-make[2]: *** [test\CMakeFiles\ExampleTests.dir\build.make:82: test/CMakeFiles/ExampleTests.dir/ExampleTests.cpp.obj] Error 1
[build] mingw32-make[1]: *** [CMakeFiles\Makefile2:315: test/CMakeFiles/ExampleTests.dir/all] Error 2
[build] mingw32-make: *** [makefile:159: all] Error 2
[build] Build finished with exit code 2

我在这里做了更多的游戏，但我找不到问题，再次感谢任何帮助。

由于我无法提供 Repl 以最小限度地重现此问题，因此这里是 CMake 项目的链接，当前分支会产生这些错误： https://github.com/JonathanWoollett-Light/gpu_blas/tree/Compiler-update/c%2B %2B