本文最后更新于 258 天前,其中的信息可能已经过时,如有错误请发送邮件到wuxianglongblog@163.com
Cython:class 和 cdef class,使用 C++
class 和 cdef class
class
定义属性变量比较自由,cdef class
可以定义 cdef
class
使用 __init__
初始化,cdef class
在使用 __init__
之前用 __cinit__
对 C
相关的参数进行初始化。
cdef class
中的方法可以是 def, cdef, cpdef
三种,只有 public
的属性才可以被访问,不可以添加新的属性。
__dealloc__
函数类似析构函数,负责释放申请的内存。
Cython
属性可以使用关键词 property
来定义,然后定义 __get__
和 __set__
方法来进行获取和设置:
property name:
def __get__(self):
return something
def __set__(self):
set_something
使用 C++ 类
使用 C++
类时要加上 cppclass
关键词,在编译时 setup
中要加上 language="c++"
的选项。
假设我们有这样一个 C++
类:
%%file particle_extern.h
#ifndef _PARTICLE_EXTERN_H_
#define _PARTICLE_EXTERN_H_
class Particle {
public:
Particle() :
mass(0), charge(0) {}
Particle(float m, float c, float *p, float *v);
~Particle() {}
float getMass() {return mass; }
void setMass(float m) { mass = m; }
float getCharge() { return charge; }
const float *getVel() { return vel; }
const float *getPos() { return pos; }
void applyImpulse(float *f, float t);
private:
float mass, charge;
float pos[3], vel[3];
};
#endif
Overwriting particle_extern.h
%%file particle_extern.cpp
#include "particle_extern.h"
Particle::Particle(float m, float c, float *p, float *v) :
mass(m), charge(c)
{
for (int i=0; i<3; ++i) {
pos[i] = p[i]; vel[i] = v[i];
}
}
void Particle::applyImpulse(float *f, float t)
{
float newvi;
for(int i=0; i<3; ++i) {
newvi = vel[i] + t / mass * f[i];
pos[i] = (newvi + vel[i]) * t / 2.;
vel[i] = newvi;
}
}
Overwriting particle_extern.cpp
在 Cython
中调用这个类:
%%file particle.pyx
import numpy as np
cdef extern from "particle_extern.h":
cppclass _Particle "Particle":
_Particle(float m, float c, float *p, float *v)
float getMass()
void setMass(float m)
float getCharge()
const float *getVel()
const float *getPos()
void applyImpulse(float *f, float t)
cdef class Particle:
cdef _Particle *thisptr # ptr to C++ instance
def __cinit__(self, m, c, float[::1] p, float[::1] v):
if p.shape[0] != 3 or v.shape[0] != 3:
raise ValueError("...")
self.thisptr = new _Particle(m, c, &p[0], &v[0])
def __dealloc__(self):
del self.thisptr
def apply_impulse(self, float[::1] v, float t):
self.thisptr.applyImpulse(&v[0], t)
def __repr__(self):
args = ', '.join('%s=%s' % (n, getattr(self, n)) for n in ('mass', 'charge', 'pos', 'vel'))
return 'particle.Particle(%s)' % args
property charge:
def __get__(self):
return self.thisptr.getCharge()
property mass: # Cython-style properties.
def __get__(self):
return self.thisptr.getMass()
def __set__(self, m):
self.thisptr.setMass(m)
property vel:
def __get__(self):
cdef const float *_vel = self.thisptr.getVel()
cdef float[::1] arr = np.empty((3,), dtype=np.float32)
for i in range(3):
arr[i] = _vel[i]
return np.asarray(arr)
property pos:
def __get__(self):
cdef const float *_pos = self.thisptr.getPos()
cdef float[::1] arr = np.empty((3,), dtype=np.float32)
for i in range(3):
arr[i] = _pos[i]
return np.asarray(arr)
Overwriting particle.pyx
首先从头文件声明这个类:
cdef extern from "particle_extern.h":
cppclass _Particle "Particle":
_Particle(float m, float c, float *p, float *v)
float getMass()
void setMass(float m)
float getCharge()
const float *getVel()
const float *getPos()
void applyImpulse(float *f, float t)
这里要使用 cppclass
关键词,并且为了方便,我们将 Particle
类的名字在 Cython
中重命名为 _Particle
。
cdef class Particle:
cdef _Particle *thisptr
def __cinit__(self, m, c, float[::1] p, float[::1] v):
if p.shape[0] != 3 or v.shape[0] != 3:
raise ValueError("...")
self.thisptr = new _Particle(m, c, &p[0], &v[0])
为了使用这个类,我们需要定义一个该类的指针,然后用指针指向一个 _Particle
对象。
%%file setup.py
from distutils.core import setup
from distutils.extension import Extension
from Cython.Distutils import build_ext
ext = Extension("particle", ["particle.pyx", "particle_extern.cpp"], language="c++")
setup(
cmdclass = {'build_ext': build_ext},
ext_modules = [ext],
)
Overwriting setup.py
要加上 language="c++"
的选项,然后编译:
!python setup.py build_ext -i
running build_ext
cythoning particle.pyx to particle.cpp
building 'particle' extension
C:\Anaconda\Scripts\gcc.bat -DMS_WIN64 -mdll -O -Wall -IC:\Anaconda\include -IC:\Anaconda\PC -c particle.cpp -o build\temp.win-amd64-2.7\Release\particle.o
C:\Anaconda\Scripts\gcc.bat -DMS_WIN64 -mdll -O -Wall -IC:\Anaconda\include -IC:\Anaconda\PC -c particle_extern.cpp -o build\temp.win-amd64-2.7\Release\particle_extern.o
writing build\temp.win-amd64-2.7\Release\particle.def
C:\Anaconda\Scripts\g++.bat -DMS_WIN64 -shared -s build\temp.win-amd64-2.7\Release\particle.o build\temp.win-amd64-2.7\Release\particle_extern.o build\temp.win-amd64-2.7\Release\particle.def -LC:\Anaconda\libs -LC:\Anaconda\PCbuild\amd64 -lpython27 -lmsvcr90 -o "C:\Users\lijin\Documents\Git\python-tutorial\07. interfacing with other languages\particle.pyd"
particle.cpp: In function 'void __Pyx_RaiseArgtupleInvalid(const char*, int, Py_ssize_t, Py_ssize_t, Py_ssize_t)':
particle.cpp:14931:59: warning: unknown conversion type character 'z' in format [-Wformat]
particle.cpp:14931:59: warning: format '%s' expects argument of type 'char*', but argument 5 has type 'Py_ssize_t {aka long long int}' [-Wformat]
particle.cpp:14931:59: warning: unknown conversion type character 'z' in format [-Wformat]
particle.cpp:14931:59: warning: too many arguments for format [-Wformat-extra-args]
particle.cpp: In function 'int __Pyx_BufFmt_ProcessTypeChunk(__Pyx_BufFmt_Context*)':
particle.cpp:15498:78: warning: unknown conversion type character 'z' in format [-Wformat]
particle.cpp:15498:78: warning: unknown conversion type character 'z' in format [-Wformat]
particle.cpp:15498:78: warning: too many arguments for format [-Wformat-extra-args]
particle.cpp:15550:67: warning: unknown conversion type character 'z' in format [-Wformat]
particle.cpp:15550:67: warning: unknown conversion type character 'z' in format [-Wformat]
particle.cpp:15550:67: warning: too many arguments for format [-Wformat-extra-args]
particle.cpp: In function 'PyObject* __pyx_buffmt_parse_array(__Pyx_BufFmt_Context*, const char**)':
particle.cpp:15612:69: warning: unknown conversion type character 'z' in format [-Wformat]
particle.cpp:15612:69: warning: format '%d' expects argument of type 'int', but argument 3 has type 'size_t {aka long long unsigned int}' [-Wformat]
particle.cpp:15612:69: warning: too many arguments for format [-Wformat-extra-args]
particle.cpp: In function 'int __Pyx_GetBufferAndValidate(Py_buffer*, PyObject*, __Pyx_TypeInfo*, int, int, int, __Pyx_BufFmt_StackElem*)':
particle.cpp:15797:73: warning: unknown conversion type character 'z' in format [-Wformat]
particle.cpp:15797:73: warning: format '%s' expects argument of type 'char*', but argument 3 has type 'Py_ssize_t {aka long long int}' [-Wformat]
particle.cpp:15797:73: warning: unknown conversion type character 'z' in format [-Wformat]
particle.cpp:15797:73: warning: too many arguments for format [-Wformat-extra-args]
particle.cpp: In function 'void __Pyx_RaiseTooManyValuesError(Py_ssize_t)':
particle.cpp:16216:94: warning: unknown conversion type character 'z' in format [-Wformat]
particle.cpp:16216:94: warning: too many arguments for format [-Wformat-extra-args]
particle.cpp: In function 'void __Pyx_RaiseNeedMoreValuesError(Py_ssize_t)':
particle.cpp:16222:48: warning: unknown conversion type character 'z' in format [-Wformat]
particle.cpp:16222:48: warning: format '%s' expects argument of type 'char*', but argument 3 has type 'Py_ssize_t {aka long long int}' [-Wformat]
particle.cpp:16222:48: warning: too many arguments for format [-Wformat-extra-args]
particle.cpp: In function 'int __Pyx_ValidateAndInit_memviewslice(int*, int, int, int, __Pyx_TypeInfo*, __Pyx_BufFmt_StackElem*, __Pyx_memviewslice*, PyObject*)':
particle.cpp:16941:50: warning: unknown conversion type character 'z' in format [-Wformat]
particle.cpp:16941:50: warning: format '%s' expects argument of type 'char*', but argument 3 has type 'Py_ssize_t {aka long long int}' [-Wformat]
particle.cpp:16941:50: warning: unknown conversion type character 'z' in format [-Wformat]
particle.cpp:16941:50: warning: too many arguments for format [-Wformat-extra-args]
import particle
注意这里类型要设成 float32
,因为 C++
程序中接受的是 float
类型,默认是 float64(double)
类型:
import numpy as np
pos = vel = np.arange(3., dtype='float32')
mass = 1.0
charge = 2.0
p = particle.Particle(mass, charge, pos, vel)
p
particle.Particle(mass=1.0, charge=2.0, pos=[ 0. 1. 2.], vel=[ 0. 1. 2.])
调用 apply_impulse
方法:
p.apply_impulse(np.arange(3., dtype='float32'), 1.0)
p
particle.Particle(mass=1.0, charge=2.0, pos=[ 0. 1.5 3. ], vel=[ 0. 2. 4.])
查看质量:
p.mass
1.0
修改质量:
p.mass = 3.0
查看 charge
:
p.charge
2.0
因为 charge
没有定义 __set__
方法,所以它是只读的属性,不能进行修改。
import zipfile
f = zipfile.ZipFile('07-05-particle.zip','w',zipfile.ZIP_DEFLATED)
names = ['particle.pyx',
'particle_extern.cpp',
'particle_extern.h',
'setup.py']
for name in names:
f.write(name)
f.close()
!rm -f setup*.*
!rm -f particle*.*
!rm -rf build