关于python调用fortran编译的dll的问题
最近在学习python与fortran的混编，打算使用numpy的ctypeslib模块调用fortran的dll。首先我写了个小例子做测试，混编成功。但我把自己写的代码封装成dll，在python调用，计算结果u_nex, v_nex, eta_nex全是零，即python程序中初始化的值。已调试了一天，还是没发现问题所在。现将代码贴出来，求助大家。
Module proc
Use, Intrinsic :: ISO_C_BINDING
Implicit None
! 将常量、模型参数定义为一个结构体（全局变量），通过子程序参数接口传入子程序
Type, Bind(C) :: arg_params
Real(C_DOUBLE) :: dx, dy, dt, hmin, rho, g ! 水平空间步长，时间歩长，最小截断水深，密度， 重力加速度
Real(C_DOUBLE) :: f, beta, r, taux, tauy, ah ! 科氏力参数，beta系数，底摩擦系数，风阻系数（x，y方向）, 水平涡粘系数
Integer(C_INT) :: mode ! 对流项差分格式
Real(C_DOUBLE) :: slip ! 滑移边界类型
End Type
Contains
! 计算u子程序======================================================================
Subroutine cal_u(mask, h, eta_cur, u_cur, v_cur, u_nex, nx, ny, params) Bind(C)
!!DEC$ ATTRIBUTES DLLEXPORT :: cal_u
Implicit None
! 形参列表nx，ny
Integer(C_INT), Intent(in), Value :: nx, ny
! 形参列表mask, h, eta_cur, u_cur, v_cur, u_nex， 均为假定形状数组
Integer(C_INT), Intent(in) :: mask(0:nx+1, 0:ny+1) ! 水陆掩膜，判断干湿格点（1为湿点，0为干点）
Real(C_DOUBLE), Intent(in) :: h(0:nx+1, 0:ny+1) ! 实时水深
Real(C_DOUBLE), Intent(in) :: eta_cur(0:nx+1, 0:ny+1), u_cur(0:nx+1, 0:ny+1), v_cur(0:nx+1, 0:ny+1)! 第n步eta，u，v
Real(C_DOUBLE), Intent(out) :: u_nex(0:nx+1, 0:ny+1) ! 第n+1步u流速，作为计算结果输出
! 形参列表params
type(arg_params), Intent(in) :: params
! 定义cal_u子程序用到的局部变量
Real(8) :: B_cur(0:nx+1, 0:ny+1), B_nex(0:nx+1, 0:ny+1) ! 第n步示踪物B的浓度，第n+1步示踪物B的浓度
Real(8) :: CuP(0:nx+1, 0:ny+1), CuN(0:nx+1, 0:ny+1)
Real(8) :: CvP(0:nx+1, 0:ny+1), CvN(0:nx+1, 0:ny+1)
Real(8) :: du(0:nx+1, 0:ny+1), upre(0:nx+1, 0:ny+1)
Real(8) :: Rx(0:nx+1, 0:ny+1)
Real(8) :: hu, uu, vu
Real(8) :: pgrdx, tx, corx
Real(8) :: advx(0:nx+1, 0:ny+1), div, div1, div2 ! x方向对流项相关变量
Real(8) :: diffx, term1, term2, term3, term4, s1, s2, h1! x方向扩散项相关变量
Integer :: i, j
! 先计算对流项advx
Do j = 0, ny+1
Do i = 0, nx
CuP(i, j) = 0.25d0*(u_cur(i, j)+u_cur(i+1, j)+abs(u_cur(i, j))+abs(u_cur(i+1, j)))*params%dt/params%dx
CuN(i, j) = 0.25d0*(u_cur(i, j)+u_cur(i+1, j)-abs(u_cur(i, j))-abs(u_cur(i+1, j)))*params%dt/params%dx
CvP(i, j) = 0.25d0*(v_cur(i, j)+v_cur(i+1, j)+abs(v_cur(i, j))+abs(v_cur(i+1, j)))*params%dt/params%dy
CvN(i, j) = 0.25d0*(v_cur(i, j)+v_cur(i+1, j)-abs(v_cur(i, j))-abs(v_cur(i+1, j)))*params%dt/params%dy
End Do
End Do
B_cur(:, :) = u_cur(:, :)
Call advect(CuP, CuN, CvP, CvN, B_cur, B_nex, nx, ny, params%mode)
Do j = 1, ny
Do i = 1, nx
div1 = 0.5d0*(u_cur(i+1, j)-u_cur(i-1, j))/params%dx
div2 = 0.5d0*(v_cur(i, j)+v_cur(i+1, j)-v_cur(i, j-1)-v_cur(i+1, j-1))/params%dy
div = params%dt*B_cur(i, j)*(div1+div2)
advx(i, j)= B_nex(i, j)+div! 计算出advx
End Do
End Do
! 分别计算tx，pgrdx及diffx
Do j = 1, ny
Do i = 1, nx
! 计算u的中间变量
pgrdx = -params%dt*params%g*(eta_cur(i+1, j)-eta_cur(i, j))/params%dx! 计算出pgrdx
hu = 0.5d0*(h(i, j)+h(i+1, j))
uu = u_cur(i, j)
vu = 0.25d0*(v_cur(i, j)+v_cur(i, j-1)+v_cur(i+1, j)+v_cur(i+1, j-1))
diffx = 0.0d0
If (hu>0.0) Then
tx = params%dt*params%taux/(params%rho*hu) ! 计算出tx
Rx(i, j) = params%dt*params%r*sqrt(uu**2+vu**2)/hu! 计算出Rx
! 根据滑移边界类型计算diffx
term1 = h(i+1, j)*(u_cur(i+1, j)-u_cur(i, j))/params%dx
term2 = h(i, j)*(u_cur(i, j)-u_cur(i-1, j))/params%dx
s1 = 1.0d0
s2 = 1.0d0
h1 = h(i, j+1)+h(i+1, j+1)
If (h1<params%hmin) Then
s1 = 0.0d0
s2 = params%slip
h1 = h(i, j)+h(i+1, j)
End If
term3 = 0.25d0*(h(i, j)+h(i+1, j)+h1)*(s1*u_cur(i, j+1)-s2*u_cur(i, j))/params%dy
s1 = 1.0d0
s2 = 1.0d0
h1 = h(i, j-1)+h(i+1, j-1)
If (h1<params%hmin) Then
s1 = 0.0d0
s2 = params%slip
h1 = h(i, j)+h(i+1, j)
End If
term4 = 0.25d0*(h(i, j)+h(i+1, j)+h1)*(s2*u_cur(i, j)-s1*u_cur(i, j-1))/params%dy
diffx = params%dt*params%ah*((term1-term2)/params%dx+(term3-term4)/params%dy)/hu! 计算出diffx
End If
! 第一步，得到不含科氏力项预估的upre
upre(i, j) = u_cur(i, j)+tx+pgrdx+advx(i, j)+diffx
! 第二步，半隐式方法求解科氏力项
corx = params%dt*params%f*vu
du(i, j) = (upre(i, j)-params%beta*u_cur(i, j)+corx)/(1.0d0+params%beta)-u_cur(i, j)! 计算出du
! 第三步，计算u
If (mask(i, j)==1) Then
If ((mask(i+1, j)==1) .or. (du(i, j)>0.0)) u_nex(i, j) = (u_cur(i, j)+du(i, j))/(1.0d0+Rx(i, j))
Else
If ((mask(i+1, j)==1) .and. (du(i, j)<0.0)) u_nex(i, j) = (u_cur(i, j)+du(i, j))/(1.0d0+Rx(i, j))
End If
End Do
End Do
End Subroutine cal_u
! 计算v子程序======================================================================
Subroutine cal_v(mask, h, eta_cur, u_cur, v_cur, v_nex, nx, ny, params) Bind(C)
!!DEC$ ATTRIBUTES DLLEXPORT :: cal_v
Implicit None
! 形参列表nx，ny
Integer(C_INT), Intent(in), Value :: nx, ny
! 形参列表mask, h, eta_cur, u_cur, v_cur, v_nex， 均为假定形状数组
Integer(C_INT), Intent(in) :: mask(0:nx+1, 0:ny+1) ! 水陆掩膜，判断干湿格点（1为湿点，0为干点）
Real(C_DOUBLE), Intent(in) :: h(0:nx+1, 0:ny+1) ! 实时水深
Real(C_DOUBLE), Intent(in) :: eta_cur(0:nx+1, 0:ny+1), u_cur(0:nx+1, 0:ny+1), v_cur(0:nx+1, 0:ny+1)! 第n步eta，u，v
Real(C_DOUBLE), Intent(out) :: v_nex(0:nx+1, 0:ny+1) ! 第n+1步v流速，作为计算结果输出
! 形参列表params
type(arg_params), Intent(in) :: params
! 定义cal_v子程序用到的局部变量
Real(8) :: B_cur(0:nx+1, 0:ny+1), B_nex(0:nx+1, 0:ny+1) ! 第n步示踪物B的浓度，第n+1步示踪物B的浓度
Real(8) :: CuP(0:nx+1, 0:ny+1), CuN(0:nx+1, 0:ny+1)
Real(8) :: CvP(0:nx+1, 0:ny+1), CvN(0:nx+1, 0:ny+1)
Real(8) :: dv(0:nx+1, 0:ny+1), vpre(0:nx+1, 0:ny+1)
Real(8) :: Ry(0:nx+1, 0:ny+1)
Real(8) :: hv, vv, uv
Real(8) :: pgrdy, ty, cory
Real(8) :: advy(0:nx+1, 0:ny+1), div, div1, div2 ! y方向对流项相关变量
Real(8) :: diffy, term1, term2, term3, term4, s1, s2, h1! y方向扩散项相关变量
Integer :: i, j
! 先计算对流项advy
Do j = 0, ny
Do i = 0, nx+1
CuP(i, j) = 0.25d0*(u_cur(i, j)+u_cur(i, j+1)+abs(u_cur(i, j))+abs(u_cur(i, j+1)))*params%dt/params%dx
CuN(i, j) = 0.25d0*(u_cur(i, j)+u_cur(i, j+1)-abs(u_cur(i, j))-abs(u_cur(i, j+1)))*params%dt/params%dx
CvP(i, j) = 0.25d0*(v_cur(i, j)+v_cur(i, j+1)+abs(v_cur(i, j))+abs(v_cur(i, j+1)))*params%dt/params%dy
CvN(i, j) = 0.25d0*(v_cur(i, j)+v_cur(i, j+1)-abs(v_cur(i, j))-abs(v_cur(i, j+1)))*params%dt/params%dy
End Do
End Do
B_cur(:, :) = v_cur(:, :)
Call advect(CuP, CuN, CvP, CvN, B_cur, B_nex, nx, ny, params%mode)
Do j = 1, ny
Do i = 1, nx
div1 = 0.5d0*(u_cur(i, j)+u_cur(i, j+1)-u_cur(i-1, j)-u_cur(i-1, j+1))/params%dx
div2 = 0.5d0*(v_cur(i, j+1)-v_cur(i, j-1))/params%dy
div = params%dt*B_cur(i, j)*(div1+div2)
advy(i, j)= B_nex(i, j)+div! 计算出advy
End Do
End Do
! 分别计算ty，pgrdy及diffy
Do j = 1, ny
Do i = 1, nx
! 计算v的中间变量
pgrdy = -params%dt*params%g*(eta_cur(i, j+1)-eta_cur(i, j))/params%dy! 计算出pgrdy
hv = 0.5d0*(h(i, j)+h(i, j+1))
vv = v_cur(i, j)
uv = 0.25d0*(u_cur(i, j)+u_cur(i, j+1)+u_cur(i-1, j)+u_cur(i-1, j+1))
diffy = 0.0d0
If (hv>0.0) Then
ty = params%dt*params%tauy/(params%rho*hv)! 计算出ty
Ry(i, j) = params%dt*params%r*sqrt(vv**2+uv**2)/hv! 计算出Ry
! 根据滑移边界类型计算diffy
s1 = 1.0d0
s2 = 1.0d0
h1 = h(i+1, j)+h(i+1, j+1)
If (h1<params%hmin) Then
s1 = 0.0d0
s2 = params%slip
h1 = h(i, j)+h(i, j+1)
End If
term1 = 0.25d0*(h(i, j)+h(i, j+1)+h1)*(s1*v_cur(i+1, j)-s2*v_cur(i, j))/params%dx
s1 = 1.0d0
s2 = 1.0d0
h1 = h(i-1, j)+h(i-1, j+1)
If (h1<params%hmin) Then
s1 = 0.0d0
s2 = params%slip
h1 = h(i, j)+h(i, j+1)
End If
term2 = 0.25d0*(h(i, j)+h(i, j+1)+h1)*(s2*v_cur(i, j)-s1*v_cur(i-1, j))/params%dx
term3 = h(i, j+1)*(v_cur(i, j+1)-v_cur(i, j))/params%dy
term4 = h(i, j)*(v_cur(i, j)-v_cur(i-1, j-1))/params%dy
diffy = params%dt*params%ah*((term1-term2)/params%dx+(term3-term4)/params%dy)/hv! 计算出diffy
End If
! 第一步，得到不含科氏力项预估的vpre
vpre(i, j) = v_cur(i, j)+ty+pgrdy+advy(i, j)+diffy
! 第二步，半隐式方法求解科氏力项
cory = -params%dt*params%f*uv
dv(i, j) = (vpre(i, j)-params%beta*v_cur(i, j)+cory)/(1.0d0+params%beta)-v_cur(i, j)! 计算出dv
! 计算v
If (mask(i, j)==1) Then
If ((mask(i, j+1)==1) .or. (dv(i, j)>0.0)) v_nex(i, j) = (v_cur(i, j)+dv(i, j))/(1.0d0+Ry(i, j))
Else
If ((mask(i, j+1)==1) .and. (dv(i, j)<0.0)) v_nex(i, j) = (v_cur(i, j)+dv(i, j))/(1.0d0+Ry(i, j))
End If
End Do
End Do
End subroutine cal_v
! 计算eta子程序=========================================================================
Subroutine cal_eta(h, eta_cur, u_nex, v_nex, eta_nex, nx, ny, params) Bind(C)
!!DEC$ ATTRIBUTES DLLEXPORT :: cal_eta
Implicit None
! 形参列表nx，ny
Integer(C_INT), Intent(in), Value :: nx, ny
! 形参列表mask, h, eta_cur, u_cur, v_cur, v_nex， 均为假定形状数组
Real(C_DOUBLE), Intent(in) :: h(0:nx+1, 0:ny+1) ! 实时水深
Real(C_DOUBLE), Intent(in) :: eta_cur(0:nx+1, 0:ny+1), u_nex(0:nx+1, 0:ny+1), v_nex(0:nx+1, 0:ny+1)! 第n步eta，第n+1步u，v
Real(C_DOUBLE), Intent(out) :: eta_nex(0:nx+1, 0:ny+1) ! 第n+1步u流速，作为计算结果输出
! 形参列表params
type(arg_params), Intent(in) :: params
! 定义cal_eta子程序用到的局部变量
Real(8) :: B_cur(0:nx+1, 0:ny+1), B_nex(0:nx+1, 0:ny+1) ! 第n步示踪物B的浓度，第n+1步示踪物B的浓度
Real(8) :: CuP(0:nx+1, 0:ny+1), CuN(0:nx+1, 0:ny+1)
Real(8) :: CvP(0:nx+1, 0:ny+1), CvN(0:nx+1, 0:ny+1)
Integer :: i, j
Do j = 0, ny+1
Do i = 0, nx+1
CuP(i, j) = 0.5d0*(u_nex(i, j)+abs(u_nex(i, j)))*params%dt/params%dx
CuN(i, j) = 0.5d0*(u_nex(i, j)-abs(u_nex(i, j)))*params%dt/params%dx
CvP(i, j) = 0.5d0*(v_nex(i, j)+abs(v_nex(i, j)))*params%dt/params%dy
CvN(i, j) = 0.5d0*(v_nex(i, j)-abs(v_nex(i, j)))*params%dt/params%dy
End Do
End Do
B_cur(:, :) = h(:, :)
Call advect(CuP, CuN, CvP, CvN, B_cur, B_nex, nx, ny, params%mode)
Do j = 1, ny
Do i = 1, nx
! 计算eta
eta_nex(i, j) = eta_cur(i, j)+B_nex(i, j)
End Do
End Do
End Subroutine cal_eta
! 计算对流项子程序===========================================================================
Subroutine advect(CuP, CuN, CvP, CvN, B_cur, B_nex, nx, ny, mode)
Implicit None
! 形参列表
Integer, Intent(in), Value:: nx, ny, mode
Real(8), Intent(in):: CuP(0:nx+1, 0:ny+1), CuN(0:nx+1, 0:ny+1), CvP(0:nx+1, 0:ny+1), CvN(0:nx+1, 0:ny+1), B_cur(0:nx+1, 0:ny+1)
Real(8), Intent(out) :: B_nex(0:nx+1, 0:ny+1)
! 局部变量
Real(8) :: RxP(0:nx+1, 0:ny+1), RxN(0:nx+1, 0:ny+1)
Real(8) :: RyP(0:nx+1, 0:ny+1), RyN(0:nx+1, 0:ny+1)
Real(8) :: dB, term1, term2, term3, term4
Real(8) :: BwP, BwN, BeP, BeN, BsP, BsN, BnP, BnN
Integer :: i, j
! 数组初始化
RxP(:, :) = 0.0d0
RxN(:, :) = 0.0d0
RyP(:, :) = 0.0d0
RyN(:, :) = 0.0d0
! 计算x方向，y方向的r+
Do j = 1, ny
Do i = 1, nx
dB =B_cur(i+1, j)-B_cur(i, j)
If (abs(dB) > 0.0) RxP(i, j) = (B_cur(i, j)-B_cur(i-1, j))/dB
dB =B_cur(i, j+1)-B_cur(i, j)
If (abs(dB) > 0.0) RyP(i, j) = (B_cur(i, j)-B_cur(i, j-1))/dB
End Do
End Do
! 计算x方向，y方向的r-
Do j = 1, ny
Do i = 0, nx-1
dB = B_cur(i+1, j)-B_cur(i, j)
If (abs(dB) > 0.0) RxN(i, j) = (B_cur(i+2, j)-B_cur(i+1, j))/dB
End Do
End Do
Do j = 0, ny-1
Do i = 1, nx
dB = B_cur(i, j+1)-B_cur(i, j)
If (abs(dB) > 0.0) RyN(i, j) = (B_cur(i, j+2)-B_cur(i, j+1))/dB
End Do
End Do
! 计算示踪物B浓度
Do j = 1, ny
Do i = 1, nx
!x方向
BwP = B_cur(i-1, j)+0.5d0*PSI(RxP(i-1, j), mode)*(1.0d0-CuP(i-1, j))*(B_cur(i, j)-B_cur(i-1, j))
BwN = B_cur(i, j)-0.5d0*PSI(RxN(i-1, j), mode)*(1.0d0+CuN(i-1, j))*(B_cur(i, j)-B_cur(i-1, j))
BeP = B_cur(i, j)+0.5d0*PSI(RxP(i, j), mode)*(1.0d0-CuP(i, j))*(B_cur(i+1, j)-B_cur(i, j))
BeN = B_cur(i+1, j)-0.5d0*PSI(RxN(i, j), mode)*(1.0d0+CuN(i, j))*(B_cur(i+1, j)-B_cur(i, j))
! y方向
BsP = B_cur(i, j-1)+0.5d0*PSI(RyP(i, j-1), mode)*(1.0d0-CvP(i, j-1))*(B_cur(i, j)-B_cur(i, j-1))
BsN = B_cur(i, j)-0.5d0*PSI(RyN(i, j-1), mode)*(1.0d0+CvN(i, j-1))*(B_cur(i, j)-B_cur(i, j-1))
BnP = B_cur(i, j)+0.5d0*PSI(RyP(i, j), mode)*(1.0d0-CvP(i, j))*(B_cur(i, j+1)-B_cur(i, j))
BnN = B_cur(i, j+1)-0.5d0*PSI(RyN(i, j), mode)*(1.0d0+CvN(i, j))*(B_cur(i, j+1)-B_cur(i, j))
term1 = CuP(i-1, j)*BwP+CuN(i-1, j)*BwN
term2 = CuP(i, j)*BeP+CuN(i, j)*BeN
term3 = CvP(i, j-1)*BsP+CvN(i, j-1)*BsN
term4 = CvP(i, j)*BnP+CvN(i, j)*BnN
B_nex(i, j) = term1-term2+term3-term4
End Do
End Do
Contains
Real(8) Function PSI(rr, mmode)
Real(8), Intent(in) :: rr
Integer, Intent(in) :: mmode
Real(8) :: comp1, comp2, comp3
If (mmode == 1) PSI = 0.0d0
If (mmode == 2) PSI = 1.0d0
If (mmode == 3) Then
comp1 = Min(2.0d0*rr, 1.0d0)
comp2 = Min(rr, 2.0d0)
comp3 = Max(comp1, comp2)
PSI = Max(comp3, 0.0d0)
End If
End Function
End Subroutine advect
End Module
多半栽倒在二维数组上面
仅供参考
Python调用C/Fortran混合的动态链接库
http://www.cnblogs.com/pasuka/p/4012508.html pasuka 发表于 2017-6-1 15:29
多半栽倒在二维数组上面
仅供参考
Python调用C/Fortran混合的动态链接库
谢谢，已经拜读过这篇文章，我自己再排查排查吧 已经排查到原因了，我在python程序里把第一个形参mask类型定义错了，应该是dtype=‘int32’，我错误地定义成了‘float64’。现在程序的计算结果正确了！ 学习了，mark一下 问下，结构体Derived Type在Python中是怎么输入的？
页:
[1]