python - 错误:没有匹配函数来调用‘pybind11::buffer_info::buffer_info

Question

我正在尝试使用 pybind11 和 cppimport 包装一个使用 Armadillo 库的 c++ 函数。但是当我尝试做一些简单的事情时，比如矩阵乘法，我得到了以下错误。

error: no matching function for call to ‘pybind11::buffer_info::buffer_info(double*, long unsigned int, std::__cxx11::string, int, <brace-enclosed initializer list>, <brace-enclosed initializer list>)’
      );
      ^
    In file included from /home/muah/anaconda3/envs/ising/include/python3.6m/pybind11/pytypes.h:13:0,
                     from /home/muah/anaconda3/envs/ising/include/python3.6m/pybind11/cast.h:13,
                     from /home/muah/anaconda3/envs/ising/include/python3.6m/pybind11/attr.h:13,
                     from /home/muah/anaconda3/envs/ising/include/python3.6m/pybind11/pybind11.h:43,
                     from /home/muah/Music/cpp2py/.rendered.code.cpp:3:
    /home/muah/anaconda3/envs/ising/include/python3.6m/pybind11/buffer_info.h:83:5: note: candidate: pybind11::buffer_info::buffer_info(pybind11::buffer_info::private_ctr_tag, void*, pybind11::ssize_t, const string&, pybind11::ssize_t, pybind11::detail::any_container<long int>&&, pybind11::detail::any_container<long int>&&)
         buffer_info(private_ctr_tag, void *ptr, ssize_t itemsize, const std::string &format, ssize_t ndim,
         ^~~~~~~~~~~
    /home/muah/anaconda3/envs/ising/include/python3.6m/pybind11/buffer_info.h:83:5: note:   candidate expects 7 arguments, 6 provided
    /home/muah/anaconda3/envs/ising/include/python3.6m/pybind11/buffer_info.h:59:5: note: candidate: pybind11::buffer_info::buffer_info(pybind11::buffer_info&&)
         buffer_info(buffer_info &&other) {

这是在 armadillo mat 和 numpy 之间进行转换的代码，反之亦然。

#pragma once
#include <armadillo>
#include <pybind11/pybind11.h>
#include <pybind11/numpy.h>

namespace py=pybind11;

typedef py::array_t<double, py::array::f_style | py::array::forcecast> pyarr_d;

inline
arma::mat py_to_mat(pyarr_d& pmat)
{
    py::buffer_info info = pmat.request();
    arma::mat amat;
    if(info.ndim == 1) {
        amat = arma::mat(reinterpret_cast<double*>(info.ptr),info.shape[0],1);
    } else {
        amat = arma::mat(reinterpret_cast<double*>(info.ptr),info.shape[0],info.shape[1]);
    }
    return amat;
}


inline
py::array_t<double> mat_to_py(arma::mat &mat)
{
    py::buffer_info buffer(
        mat.memptr(),
        sizeof(double),
        py::format_descriptor<double>::format(),
        2,
        { mat.n_rows, mat.n_cols },
        { sizeof(double), sizeof(double) * mat.n_rows }
    );
    return py::array_t<double>(buffer);
}

这是 C++ 函数:

<%
cfg['compiler_args'] = ['-std=c++11', '-larmadillo']
setup_pybind11(cfg)
%>

#include <pybind11/pybind11.h>
#include <pybind11/numpy.h>
#include <armadillo>
#include "py2arma.hpp"  

namespace py = pybind11;


py::array matrix_mul(pyarr_d a, pyarr_d b) 
{
    arma::mat A = py_to_mat(a);
    arma::mat B = py_to_mat(b);
    arma::mat C = A * B;
    return mat_to_py(C);
}


PYBIND11_PLUGIN(code) 
{
    py::module m("code", "lala");
    m.def("matrix_mul",(py::array(*)(pyarr_d,pyarr_d))&matrix_mul);
    return m.ptr();
}

这是 python 函数:

import cppimport
import numpy as np

code = cppimport.imp("code")

if __name__ == '__main__':
    xs = np.random.rand(3,3)
    ys = np.random.rand(3,1)

    py_mul = np.dot(xs, ys)
    cpp_mul = code.matrix_mul(xs, ys)
    print(py_mul)
    print(cpp_mul)

此链接包含完整的堆栈跟踪:https://pastebin.com/XuKyQDMQ

我不明白这个错误是什么意思。如何解决这个问题，我是否正确地进行了转换？

编辑:要解决这些问题，请在矩阵形状和 buffer_info 的跨步部分之前添加 py::detail::any_container。并将 mat_to_py 函数的类型从 py::array 更改为 pyarr_d。

这修复了代码的所有问题。

Answer 1

从上述评论的追踪来看，似乎是由于您回溯中的这个片段(第 59 行):


/home/muah/anaconda3/envs/ising/include/python3.6m/pybind11/buffer_info.h:28:5: note: candidate: pybind11::buffer_info::buffer_info(void*, pybind11::ssize_t, const string&, pybind11::ssize_t, pybind11::detail::any_container<long int>, pybind11::detail::any_container<long int>)
buffer_info(void *ptr, ssize_t itemsize, const std::string &format, ssize_t ndim,
^~~~~~~~~~~
/home/muah/anaconda3/envs/ising/include/python3.6m/pybind11/buffer_info.h:28:5: note: no known conversion for argument 6 from ‘<brace-enclosed initializer list>’ to ‘pybind11::detail::any_container<long int>’


我的建议是通过首先显式构建 py::any_container<size_t>(...) 来完成这项工作对于这些论点，这将帮助您缩小推理出错的范围。可能是由于 Armadillo 的尺寸与 ssize_t 不匹配所致来自 pybind11 ？