使用Boost::Compute传递两值到自定义函数时遇编译错误求助

我在Windows 10 + Visual Studio 2022环境下，通过vcpkg安装了OpenCL和Boost::Compute，想要用Boost::Compute实现两个向量元素的乘积计算，但编译时出现错误。

代码实现 #

#include <random>
#include <vector>
#include <boost/compute/core.hpp>
#include <boost/compute/lambda.hpp>
#include <boost/compute/algorithm/transform.hpp>
#include <boost/compute/container/vector.hpp>

std::mt19937 rng{ std::random_device()() };
std::uniform_int_distribution<> dist(1);

namespace compute = boost::compute;
using namespace boost::compute::lambda;

int GetRandomNumber()
{
    return dist(rng);
}

int main(int argc, char* argv[])
{
    const auto device = compute::system::default_device();
    const compute::context context(device);
    compute::command_queue commandQueue(context, device);

    std::vector<int> hostVector1(10000);
    std::vector<int> hostVector2(10000);
    std::vector<int> hostResultVector(10000);
    std::ranges::generate(hostVector1, GetRandomNumber);
    std::ranges::generate(hostVector2, GetRandomNumber);

    compute::vector<int> deviceVector1(hostVector1.size(), context);
    compute::vector<int> deviceVector2(hostVector2.size(), context);
    compute::vector<int> deviceResultVector(hostResultVector.size(), context);

    compute::copy(hostVector1.begin(), hostVector1.end(), deviceVector1.begin(), commandQueue);
    compute::copy(hostVector2.begin(), hostVector2.end(), deviceVector2.begin(), commandQueue);

    transform(deviceVector1.begin(), deviceVector1.end(), deviceVector2.begin(), deviceResultVector.begin(), _1 * _2, commandQueue);

    compute::copy(deviceResultVector.begin(), deviceResultVector.end(), hostResultVector.begin(), commandQueue);

    return 0;
}

编译错误信息 #

1>C:\Users\Dave\source\repos\BoostComputeTest\BoostComputeTest\vcpkg_installed\x64-windows\x64-windows\include\boost\compute\iterator\transform_iterator.hpp(177,53): error C2039: 'get_buffer': is not a member of 'boost::compute::zip_iterator<boost::tuples::tuple<boost::compute::buffer_iterator<T>,boost::compute::buffer_iterator<T>,boost::tuples::null_type,boost::tuples::null_type,boost::tuples::null_type,boost::tuples::null_type,boost::tuples::null_type,boost::tuples::null_type,boost::tuples::null_type,boost::tuples::null_type>>'
1>C:\Users\Dave\source\repos\BoostComputeTest\BoostComputeTest\vcpkg_installed\x64-windows\x64-windows\include\boost\compute\iterator\transform_iterator.hpp(177,53): error C2039:         with
1>C:\Users\Dave\source\repos\BoostComputeTest\BoostComputeTest\vcpkg_installed\x64-windows\x64-windows\include\boost\compute\iterator\transform_iterator.hpp(177,53): error C2039:         [
1>C:\Users\Dave\source\repos\BoostComputeTest\BoostComputeTest\vcpkg_installed\x64-windows\x64-windows\include\boost\compute\iterator\transform_iterator.hpp(177,53): error C2039:             T=int
1>C:\Users\Dave\source\repos\BoostComputeTest\BoostComputeTest\vcpkg_installed\x64-windows\x64-windows\include\boost\compute\iterator\transform_iterator.hpp(177,53): error C2039:         ]
1>(compiling source file 'Main.cpp')
1>    C:\Users\Dave\source\repos\BoostComputeTest\BoostComputeTest\vcpkg_installed\x64-windows\x64-windows\include\boost\compute\iterator\zip_iterator.hpp(172,7):

问题分析与解决 #

这个错误属于Boost::Compute库内部的接口兼容性问题：当使用双输入参数的transform算法时，库内部会生成zip_iterator来封装两个输入迭代器，但该迭代器并未实现get_buffer方法，导致编译失败。你可以通过以下两种方式解决当前问题，同时也可以向Boost项目提交bug报告修复该底层问题。

• 改用for_each配合zip_iterator实现
  替换原代码中的transform调用：

  compute::for_each(
      compute::make_zip_iterator(boost::make_tuple(deviceVector1.begin(), deviceVector2.begin())),
      compute::make_zip_iterator(boost::make_tuple(deviceVector1.end(), deviceVector2.end())),
      (_1[0] * _1[1]) >> deviceResultVector.begin()[_index],
      commandQueue
  );
  

• 使用自定义OpenCL内核实现
  直接编写内核代码执行元素乘积：

  const char multiply_kernel[] = R"(
      __kernel void multiply(__global const int* a, __global const int* b, __global int* result)
      {
          int i = get_global_id(0);
          result[i] = a[i] * b[i];
      }
  )";
  
  compute::program program = compute::program::create_with_source(multiply_kernel, context);
  program.build();
  
  compute::kernel kernel(program, "multiply");
  kernel.set_arg(0, deviceVector1.get_buffer());
  kernel.set_arg(1, deviceVector2.get_buffer());
  kernel.set_arg(2, deviceResultVector.get_buffer());
  
  commandQueue.enqueue_1d_range_kernel(kernel, 0, deviceVector1.size(), 0);
  

内容的提问来源于stack exchange，提问作者Dave F