在某些场合需要在C++实现类似numpy的numpy.transpose(a, axes)功能,但是很多库如NumCpp都没有提供这样的方法,只有二维矩阵的转置,没法进行多维矩阵任意维度的转换。
比较简单的想法就是利用numpy现有的功能,在c++代码里面通过调用python来调用Numpy的transpose。
直接调用Python提供的原生API接口很麻烦,采用了pybind11可以显著简化调用,特别是涉及到传递numpy和list数据上。
直接用numpy的transpose,因为该函数仅仅是改变array的strides,并不影响内存排布,替换的解决方案则是可以使用TensorFlow的transpose函数,可以得到改变内存排布的结果。后面想到了一个直接使用numpy的简单方法:np做transpose之后reshape(-1)变成一维再reshape到结果维度,可以得到期望stride的内存排布。或者类似Pytorch的contiguous使用ascontiguousarray。
代码如下,读者如果运行可能需要适当修改CMakeLists.txt的include和library path,同时export PYTHONPATH包含.py文件的路径。
cpp main.cpp
#include <iostream>
#include <string>
#include <vector>
using namespace std;
#include <pybind11/pybind11.h>
#include <pybind11/stl.h>
#include <pybind11/numpy.h>
#include <pybind11/embed.h> // everything needed for embedding
namespace py = pybind11;
bool Transpose(float* pOutData, float* pInData, vector<int>& inDataShape, vector<int>& perm) {
// start the interpreter and keep it alive
py::scoped_interpreter guard{};
// construct numpy array
py::array_t<float> npInputArray(inDataShape, pInData);
py::module calc = py::module::import("math_test");
auto func = calc.attr("transpose");
py::object result;
try {
result = func(npInputArray, perm);
} catch (std::exception& e) {
cout << "call python transpose failed:" << e.what() << endl;;
return false;
}
py::array_t<float> outArray = result.cast<py::array_t<float>>();
// copy output data
py::buffer_info outBuf = outArray.request();
float* optr = (float*)outBuf.ptr;
memcpy(pOutData, optr, outArray.size() * sizeof(float));
// remove ddata manually, result in double free
// if (!outArray.owndata()) {
// py::buffer_info buf = outArray.request();
// float* ptr = (float*)buf.ptr;
// delete [] ptr;
// }
return true;
}
int main(int argc, char* argv[]) {
vector<float> inVec = {0, 1, 2, 3, 4, 5, 6, 7};
vector<int> shape = {2, 2, 2};
vector<int> perm = {2, 1, 0 };
vector<float> outVec = inVec;
Transpose(outVec.data(), inVec.data(), shape, perm);
cout << "in data:" << endl;
for (int elem : inVec) {
cout << elem << " ";
}
cout << endl;
cout << "out data:" << endl;
for (int elem : outVec) {
cout << elem << " ";
}
cout << endl;
return 0;
}
python math_test.py
def transpose(data, perm):
import numpy as np
result = np.transpose(data, perm)
resultn = result.reshape(-1).reshape(result.shape)
return resultn
CMakeLists.txt
cmake_minimum_required(VERSION 3.10)
project(cmake_study LANGUAGES CXX)
set(CMAKE_CXX_STANDARD 11)
# add_definitions(-D_GLIBCXX_USE_CXX11_ABI=0)
add_executable(main
main.cpp
)
target_include_directories(main
PUBLIC
/usr/include/python3.6m/
/mnt/d/codes/cpp/call_python/pybind11-2.6.1/include
)
target_link_libraries(
main
PUBLIC
/usr/lib/x86_64-linux-gnu/libpython3.6m.so
)
一些坑
这个工程单独是可以work的,也就是单纯的cpp单向调用python是可行的,但是如果在python调用cpp代码,而这个cpp代码通过pyblind再调用Python其他模块时行不通。例如我可能是python 启动了tensorflow,然后再tf cpp插件里面调用了numpy的功能(tf的cpp插件调用python的包不要再调用tensorflow)。
针对原生python c api有如下解决方案(不需要Py_Initialize(); Py_Finalize();):
PyGILState_STATE gstate;
gstate = PyGILState_Ensure();
/* Perform Python actions here. */
result = CallSomeFunction();
/* evaluate result or handle exception */
/* Release the thread. No Python API allowed beyond this point. */
PyGILState_Release(gstate);
经过验证上面的方法对于Pybind11也是适用的,也就是对 py::scoped_interpreter guard{};进行一个替换。
测试代码:
class PyGil {
public:
PyGil() {
gstate = PyGILState_Ensure();
}
~PyGil() {
PyGILState_Release(gstate);
}
private:
PyGILState_STATE gstate;
};
void testPy() {
cout << "test py begin" << endl;
PyGil gil;
// PyGILState_STATE gstate;
// gstate = PyGILState_Ensure();
py::module calc = py::module::import("tensorflow");
py::print("Hello, world!");
// PyGILState_Release(gstate);
cout << "test py end" << endl;
}
pybind11 cpp调用python的其他案例
pybind11调用函数传递字符串的example:(可见直接传递string即可,无需做转换,极大简化了调用过程)
bool TestStr() {
// start the interpreter and keep it alive
py::scoped_interpreter guard{};
string inStr = "hello world";
py::object result;
try {
py::module calc = py::module::import("math_test");
auto func = calc.attr("test_str");
result = func(inStr);
} catch (std::exception& e) {
cout << "call python func failed:" << e.what() << endl;;
return false;
}
string outStr = result.cast<string>();
cout << "out str:" << outStr << endl;
return true;
}
总结
从cpp单向通过pybind11调用python时获取Lock用py::scoped_interpreter guard{};,而如果在Python调用cpp,在这个cpp反向再次调用Python可以用上面PyGil 的方式通过gstate = PyGILState_Ensure(); PyGILState_Release(gstate);来实现。
List/vector, string直接传参就好,numpy数组传递转成py::array_t。
pybind11调用相对模块使用xx.yy方式。
参考资料
https://www.jianshu.com/p/c912a0a59af9
https://stackoverflow.com/questions/44659924/returning-numpy-arrays-via-pybind11
https://gist.github.com/terasakisatoshi/79d1f656be9023cc649732c5162b3fc4
https://pybind11.readthedocs.io/en/stable/advanced/embedding.html