Fortran在自动并行时出错了,请高手支招,急急...

主题：Fortran在自动并行时出错了,请高手支招,急急...

cxcctv [专家分：240] 发布于 2008-05-01 09:41:00

我想在主函数里二个循环实现自动并行,我用的是IVF9.1,Openmp2.5,我的CPU是Intel (R) Core (TM)2 Duo ,我更改了二处属性:
1.  Fortran--optimization--parallelizaion==yes,threshold for auto-parallelization==0
2.  language--process openmp direcives==generate parallel code
  可是一直出现如下的错误:
  1.当在配置管理器的Debug改成win32 时,出现了ifort: Command line warning: ignoring option '/Qopenmp'; no argument required

  2  当我把把配置管理器的Debug改成X64时,它又出现了:Intel(R) Fortran Compiler for 'x64 not installed'  可我我用的XP64位系统,这是为何?请高手支招!!

  我的主函数为:
program p70
use  hanshu !这是一个函数模块
use omp_lib
implicit none
integer::nels,nxe,neq,nn,nr,nip=4,nodof=1,nod=4,ndof=4,loaded_nodes,i,k,  &
          iel,ndim=2,fixed_nodes
real::det,aa,bb    ; character (len=15) :: element = 'quadrilateral'
!----------------------------- dynamic arrays----------------------------------
real,allocatable::kv(:),kvh(:),loads(:),points(:,:),coord(:,:),jac(:,:),  &
                   der(:,:),deriv(:,:),weights(:),kp(:,:),g_coord(:,:),    &
                   value(:),kay(:,:),disps(:),perms(:)
integer,allocatable::nf(:,:),g(:),num(:),g_num(:,:),g_g(:,:),kdiag(:),    &
                      node(:),no(:)
!-------------------------input and initialisation----------------------------
open(10,file='p70.dat',status =   'old',   action = 'read')
open(11,file='water result.txt',status='replace',   action = 'write')
read (10,*)nels,nxe,aa,bb; nn=(nxe+1)*(nels/nxe+1)
allocate(nf(nodof,nn),points(nip,ndim),g(ndof),g_coord(ndim,nn),          &
          coord(nod,ndim),jac(ndim,ndim),weights(nip),der(ndim,nod),       &
          deriv(ndim,nod),kp(ndof,ndof),num(nod),g_num(nod,nels),          &
          g_g(ndof,nels),kay(ndim,ndim),perms(ndim))
call timestamp
read(10,*)perms
kay=0.0; do i=1,ndim; kay(i,i)=perms(i); end do
read(10,*)nr
nf=1; if(nr>0)read(10,*)(k,nf(:,k),i=1,nr); call formnf(nf); neq=maxval(nf)
allocate (kdiag(neq))
! ----------loop the elements to set up global geometry and kdiag -------------
kdiag=0

  elements_1: do iel=1,nels
               call geometry_4qx(iel,nxe,aa,bb,coord,num)
               g_num(:,iel)=num
               g_coord(:,num)=transpose(coord)
               call num_to_g (num, nf , g ); g_g(:,iel)=g; call fkdiag(kdiag,g)
end do elements_1

write(11,'(a)')"Global coordinates"
do k = 1 , nn
   write(11,'(a,i5,a,3e12.4)')"Node    ",k,"    ",g_coord(:,k); end do
write(11,'(a)')"Global node numbers"
do k=1,nels
   write(11,'(a,i5,a,27i3)')"Element ",k,"      ",g_num(:,k); end do

kdiag(1)=1; do i=2,neq; kdiag(i)=kdiag(i)+kdiag(i-1); end do
write(11,'(2(a,i5),/)')                                                    &
   "There are ",neq,"  equations and the skyline storage is ",kdiag(neq)

allocate(kv(kdiag(neq)),kvh(kdiag(neq)),loads(0:neq),disps(0:neq))
kv=0.0; loads=0.0
call sample(element,points,weights)
!------- element conductivity integration and assembly--------------------
elements_2: do iel=1,nels
         kp= 0.0
         num=g_num(:,iel); coord=transpose(g_coord(:,num)); g=g_g(:,iel)
         integrating_pts_1: do i=1,nip
             call shape_der(der,points,i); jac=matmul(der,coord)
             det=determinant(jac); call invert(jac); deriv=matmul(jac,der)
             kp=kp+matmul(matmul(transpose(deriv),kay),deriv)*det*weights(i)
         end do integrating_pts_1
         call fsparv (kv,kp,g,kdiag)
end do elements_2
kvh=kv
read (10,*)loaded_nodes
if(loaded_nodes/=0)read(10,*)(k,loads(nf(:,k)),i=1,loaded_nodes)
read (10,*)fixed_nodes
if(fixed_nodes/=0)then
      allocate(node(fixed_nodes),no(fixed_nodes),value(fixed_nodes))
      read(10,*)(node(i),value(i),i=1,fixed_nodes)
      do i=1,fixed_nodes; no(i)=nf(1,node(i)); end do
      kv(kdiag(no))=kv(kdiag(no))+1.e20; loads(no)=kv(kdiag(no))*value
end if

!--------------------------equation solution-----------------------------------
call sparin(kv,kdiag); call spabac(kv,loads,kdiag)
!------------------------  retrieve flow rate ---------------------------------

call linmul_sky(kvh,loads,disps,kdiag)
  write(11,'(a)')"The nodal values are:"
write(11,'(a)')"             Potentials   Flow rates"
do k=1,nn
   write(11,'(i5,a,2f12.2)')k,"   ",loads(nf(1,k)),disps(nf(1,k)); end do
write(11,'(a)')"      Inflow     Outflow"
write(11,'(2f12.2)')sum(disps(1:),mask=disps<0.),sum(disps(1:),mask=disps>0.)
  call timestamp
end program p70

最后更新于：2008-05-01 09:43:00

本帖地址： http://bbs.pfan.cn/post/274838.html

回复列表（共9个回复）

沙发

cxcctv [专家分：240] 发布于 2008-05-01 09:48:00

我的模块函数是:
  module hanshu
    contains
     subroutine formnf(nf)
   ! reform nf
   implicit none
   integer,intent(in out)::nf(:,:)
    integer:: i,j,m
    m=0
    do j=1,ubound(nf,2)
       do i=1,ubound(nf,1)
          if(nf(i,j)/=0) then
             m=m+1; nf(i,j)=m
           end if
       end do
    end do
    return
  end subroutine formnf

  subroutine geometry_4qx(iel,nxe,aa,bb,coord,num)
! coordinates and nodal vectors for equal four node quad
! elements, numbering in x
implicit none
integer,intent(in)::iel,nxe; real,intent(in)::aa,bb
real,intent(out)::coord(:,:); integer,intent(out)::num(:)
integer :: ip,iq    ; iq=(iel-1)/nxe+1; ip=iel-(iq-1)*nxe
   num=(/iq*(nxe+1)+ip,(iq-1)*(nxe+1)+ip,   &
         (iq-1)*(nxe+1)+ip+1, iq*(nxe+1)+ip+1/)
   coord(1:2,1)=(ip-1)*aa; coord(3:4,1)=ip*aa
   coord(1:4:3,2)=-iq*bb; coord(2:3,2)=-(iq-1)*bb
return
end subroutine geometry_4qx

subroutine num_to_g(num,nf,g)
!finds the g vector from num and nf
implicit none
integer,intent(in)::num(:),nf(:,:)  ; integer,intent(out):: g(:)
integer::i,k,nod,nodof ; nod=ubound(num,1) ; nodof=ubound(nf,1)
  do i = 1 , nod
      k = i*nodof  ; g(k-nodof+1:k) = nf( : , num(i) )
  end do
return
end subroutine num_to_g

subroutine fkdiag(kdiag,g)
! finds the maximum bandwidth for each freedom
implicit none
integer,intent(in)::g(:); integer,intent(out)::kdiag(:)
integer::idof,i,iwp1,j,im,k,x
  idof=size(g)
     do i = 1,idof
     iwp1=1
     if(g(i)/=0) then
        do j=1,idof
           if(g(j)/=0) then
              im=g(i)-g(j)+1
              if(im>iwp1) iwp1=im
               end if
        end do
        k=g(i);   if(iwp1>kdiag(k))kdiag(k)=iwp1
        end if
     end do
   return
end subroutine fkdiag

subroutine sample(element,ps,ws)
           implicit none
            character(*),intent(in) ::element
            real,intent(out) :: ps(:,:) ,ws(:)
            real xi
            xi=1./sqrt(3.)
              if(element =='quadrilateral') then
                ps(1,1)=-xi;ps(1,2)=xi
                ps(2,1)=xi ;ps(2,2)=-xi
                ps(3,1)=-xi;ps(3,2)=xi
                ps(4,1)=xi ;ps(4,2)=-xi
                ws=1.0
              end if
            return
         end subroutine sample

         subroutine shape_der(der,ps,k)
          implicit none
             real,intent(out):: der(:,:)
             real,intent(in) :: ps(:,:)
             integer,intent(in) ::k
             real :: x1,x2,x3,x4

             x1=0.25*(1.-ps(k,2))
             x2=0.25*(1.+ps(k,2))
             x3=0.25*(1.-ps(k,1))
             x4=0.25*(1.+ps(k,1))

             der(1,1)=-x1; der(1,2)=-x2; der(1,3)=x2; der(1,4)=x1
             der(2,1)=-x3; der(2,2)=x3; der(2,3)=x4; der(2,4)=-x4
            return
         end subroutine shape_der

         function determinant (jac) result(det)
! returns the determinant of a 1x1 2x2 3x3 jacobian matrix
implicit none    ; real :: det
real,intent(in)::jac(:,:); integer:: it ; it = ubound(jac,1)
select case (it)
   case (1)
    det=1.0
   case (2)
    det=jac(1,1)*jac(2,2) - jac(1,2) * jac(2,1)
   case (3)
    det= jac(1,1)*(jac(2,2) * jac(3,3) -jac(3,2) * jac(2,3))
    det= det-jac(1,2)*(jac(2,1)*jac(3,3)-jac(3,1)*jac(2,3))
    det= det+jac(1,3)*(jac(2,1)*jac(3,2)-jac(3,1)*jac(2,2))
   case default
    print*,' wrong dimension for jacobian matrix'
end select
return
end function determinant

板凳

cxcctv [专家分：240] 发布于 2008-05-01 09:49:00

  接上面的:subroutine invert(matrix)
  ! invert a small square matrix onto itself
  implicit none
  real,intent(in out)::matrix(:,:)
  integer::i,k,n; real::con  ; n= ubound(matrix,1)
  do k=1,n
     con=matrix(k,k); matrix(k,k)=1.
     matrix(k,:)=matrix(k,:)/con
     do i=1,n
        if(i/=k) then
           con=matrix(i,k); matrix(i,k)=0.0
           matrix(i,:)=matrix(i,:) - matrix(k,:)*con
        end if
     end do
  end do
  return
end subroutine invert

subroutine fsparv(bk,km,g,kdiag)
! assembly 装配of element 单元matrices into 整体刚度矩阵skyline global matrix
implicit none
real,intent(in)::km(:,:); integer,intent(in)::g(:),kdiag(:)
real,intent(out)::bk(:) ;  integer::i,idof,k,j,iw,ival
idof=ubound(g,1)
   do i=1,idof
      k=g(i)
      if(k/=0) then
         do j=1,idof
            if(g(j)/=0) then
               iw=k-g(j)
               if(iw>=0) then
                   ival=kdiag(k)-iw
                   bk(ival)=bk(ival)+km(i,j)
                end if
            end if
         end do
      end if
   end do
return
end subroutine fsparv

subroutine sparin(a,kdiag)
! Choleski factorisation of variable bandwidth matrix a
! stored as a vector and overwritten
implicit none
real,intent(in out)::a(:);integer,intent(in)::kdiag(:)
integer::n,i,ki,l,kj,j,ll,m,k; real::x
n=ubound(kdiag,1)  ; a(1)=sqrt(a(1))
do i=2,n
    ki=kdiag(i)-i;  l=kdiag(i-1)-ki+1
    do j=l,i
       x=a(ki+j);  kj=kdiag(j)-j
       if(j/=1) then
          ll=kdiag(j-1)-kj+1; ll=max0(l,ll)
          if(ll/=j) then
              m=j-1
              do k=ll,m ; x=x-a(ki+k)*a(kj+k) ; end do
          end if
       end if
       a(ki+j)=x/a(kj+j)
    end do
    a(ki+i)=sqrt(x)
end do
return
end subroutine sparin

subroutine spabac(a,b,kdiag)
! Choleski forward and backward substitution combined
! variable bandwidth factorised matrix a stored as a vector
implicit none
real,intent(in)::a(:);real,intent(in out)::b(0:);integer,intent(in)::kdiag(:)
integer::n,i,ki,l,m,j,it,k; real::x
n=ubound(kdiag,1)
b(1)=b(1)/a(1)
  do i=2,n
     ki=kdiag(i)-i;  l=kdiag(i-1)-ki+1 ; x=b(i)
     if(l/=i) then
        m=i-1
        do j=l,m ; x=x-a(ki+j)*b(j); end do
     end if
     b(i)=x/a(ki+i)
  end do
  do it=2,n
     i=n+2-it; ki=kdiag(i)-i; x=b(i)/a(ki+i); b(i)=x; l=kdiag(i-1)-ki+1
     if(l/=i) then
       m=i-1
       do k=l,m; b(k)=b(k)-x*a(ki+k); end do
     end if
  end do
b(1)=b(1)/a(1)
return
end subroutine spabac

3 楼

cxcctv [专家分：240] 发布于 2008-05-01 09:49:00

接上:

   subroutine linmul_sky(bp,disps,loads,kdiag)
! skyline product of symmetric matrix and a vector
implicit none
real,intent(in)::bp(:),disps(0:);real,intent(out)::loads(0:)
integer,intent(in)::kdiag(:); integer::n,i,j,low,lup,k; real::x
  n=ubound(disps,1)
  do i = 1 , n
     x = .0 ; lup=kdiag(i)
     if(i==1)low=lup; if(i/=1)low=kdiag(i-1)+1
     do j = low , lup
        x = x + bp(j) * disps(i + j - lup)
     end do
     loads(i) = x
     if(i == 1) cycle   ; lup = lup - 1
     do j = low , lup
        k = i + j -lup - 1
        loads(k) = loads(k) + bp(j)*disps(i)
     end do
    end do
     return
     end subroutine linmul_sky

   subroutine timestamp ( )
implicit none
  character ( len = 8 ) ampm
  integer d
  character ( len = 8 ) date
  integer h
  integer m
  integer mm
  character ( len = 9 ), parameter, dimension(12) :: month = (/ &
    'January  ', 'February ', 'March    ', 'April    ', &
    'May      ', 'June     ', 'July     ', 'August   ', &
    'September', 'October  ', 'November ', 'December ' /)
  integer n
  integer s
  character ( len = 10 )  time
  integer values(8)
  integer y
  character ( len = 5 ) zone

   call date_and_time ( date, time, zone, values )

  y = values(1)
  m = values(2)
  d = values(3)
  h = values(5)
  n = values(6)
  s = values(7)
  mm = values(8)

  if ( h < 12 ) then
    ampm = 'AM'
    else if ( h == 12 ) then
      if ( n == 0 .and. s == 0 ) then
      ampm = 'Noon'
      else
      ampm = 'PM'
    end if
  else
    h = h - 12
    if ( h < 12 ) then
      ampm = 'PM'
    else if ( h == 12 ) then
      if ( n == 0 .and. s == 0 ) then
        ampm = 'Midnight'
      else
        ampm = 'AM'
      end if
    end if
  end if

  write ( *, '(a,1x,i2,1x,i4,2x,i2,a1,i2.2,a1,i2.2,a1,i3.3,1x,a)' ) &
    trim ( month(m) ), d, y, h, ':', n, ':', s, '.', mm, trim ( ampm )
  return
end subroutine timestamp

4 楼

f2003 [专家分：7960] 发布于 2008-05-01 12:26:00

1，你的程序里没有使用openmp，除了use了一下，其他什么都没有。所以ivf说ignore。
大概是这样吧。
可以做个实验，里面加个简单omp do，看看还报不报。

2，你没有安装ivf的64位版本。

5 楼

cxcctv [专家分：240] 发布于 2008-05-01 15:00:00

好的,我等会把IVF卸掉再装一下,当初我也觉得装得有点问题!

6 楼

cxcctv [专家分：240] 发布于 2008-05-01 17:03:00

  f2003,你好,我重装了IVF9.1,能够运行Debug里的X64和win32,而且在BuildLog.htm时的提示是:
   ------ Build started: Project: water, Configuration: Debug|Win32 ------

Compiling with Intel Fortran 9.1 C:\Program Files (x86)\Intel\Compiler\Fortran\9.1\IA32\...

ifort /nologo /Zi /O3 /QaxT /QxT /Qparallel /Qpar_threshold:0 /assume:buffered_io /Qopenmp /module:"Debug/" /object:"Debug/" /traceback /check:bounds /libs:static /threads /dbglibs /c  /Qvc8 /Qlocation,link,"C:\Program Files (x86)\Microsoft Visual Studio 8\VC\bin" "D:\fortran\water\water\water.F90"

D:\fortran\water\water\water.F90(315) : (col. 2) remark: LOOP WAS AUTO-PARALLELIZED.
D:\fortran\water\water\water.F90(317) : (col. 2) remark: LOOP WAS AUTO-PARALLELIZED.
D:\fortran\water\water\water.F90(317) : (col. 67) remark: LOOP WAS AUTO-PARALLELIZED.
D:\fortran\water\water\water.F90(320) : (col. 2) remark: LOOP WAS AUTO-PARALLELIZED.
D:\fortran\water\water\water.F90(341) : (col. 2) remark: LOOP WAS AUTO-PARALLELIZED.
D:\fortran\water\water\water.F90(341) : (col. 10) remark: LOOP WAS AUTO-PARALLELIZED.
D:\fortran\water\water\water.F90(345) : (col. 10) remark: LOOP WAS AUTO-PARALLELIZED.
D:\fortran\water\water\water.F90(346) : (col. 28) remark: LOOP WAS AUTO-PARALLELIZED.
D:\fortran\water\water\water.F90(348) : (col. 48) remark: LOOP WAS AUTO-PARALLELIZED.
D:\fortran\water\water\water.F90(349) : (col. 60) remark: LOOP WAS AUTO-PARALLELIZED.
D:\fortran\water\water\water.F90(354) : (col. 2) remark: LOOP WAS AUTO-PARALLELIZED.
D:\fortran\water\water\water.F90(315) : (col. 2) remark: LOOP WAS VECTORIZED.
D:\fortran\water\water\water.F90(317) : (col. 2) remark: LOOP WAS VECTORIZED.
D:\fortran\water\water\water.F90(317) : (col. 67) remark: LOOP WAS VECTORIZED.
D:\fortran\water\water\water.F90(345) : (col. 10) remark: LOOP WAS VECTORIZED.
D:\fortran\water\water\water.F90(26) : (col. 4) remark: LOOP WAS AUTO-PARALLELIZED.
D:\fortran\water\water\water.F90(76) : (col. 17) remark: LOOP WAS AUTO-PARALLELIZED.

water build succeeded.

现在有几个问题:
  1.  我用omp_get_wtime()测试了一下运行的时间
                           win32             x64
自动并行选项关闭时:       0.00220          0.00146
自动并行选项开启时        0.00195          0.00146

   在工程上,很多应用程序是串行程序,如果我把这些程序改并行的,应该能提高效率,基于这个目的,我编写了这个串行代码,然后试着把它改成并行,可是手动插入并行代码时出现了很多错误,我调试了好长时间,可是并不理想,于是我想到了自动并行,并比较它们的差异,现在这个数据不知道能不能说明问题?并行成功了??

  2.  可是照我原来的设计,它自动并行后会在自动在代码里的加上一些语句如:!$omp ...但是没有出现这些语句,不知道我这个设想对不对?

  3  :为什么在这些代码的上面会有"Build started: Project: water, Configuration: Debug|Win32",我明明是调成x64,但是出现的是这个,不知道问题出在哪里.
   4:  我在重新安装IVF9.1时,在安装说明时提示此软件包含有Intel (R)Extended memory 64 technology,请问这个是不是你所说的IVF64位的版本?

    问题有点多啊,麻烦您啊!!

7 楼

f2003 [专家分：7960] 发布于 2008-05-01 22:10:00

4）对。ivf有3个版本，ia32,x64(也叫em64t)和ia64。后两个是64位的，但ia64是指安腾处理器，平常很少见，与我们的pc机不兼容。所以一般说64位ivf版本就是x64的，其运行需要64位的windows。

2）你理解错了，自动并行不会修改源码的。自动并行是生成并行可执行二进制代码，也并非通过转化成omp程序来生成二进制代码，而是直接通过调用一些底层的线程库。omp是比较高层的并行方法。
程序员应该首选omp，而不是自动并行或者自行调用底层的线程库。

1）自动并行有苛刻的条件，速度平均有个15%-30%的提高。ivf10据说较大改善了自动并行，不过现阶段不要对编译器自动并行寄予太大希望。
　　
3）我在linux下使用ifort，windows下不熟悉，我只能猜。你的win是32位的？还有，我听说ivf不是一个完整的编译器？连接器依靠vs，要是系统里没有64位的连接器，显然还是制造不了64位程序。我的说法不一定正确，听听其他人的意见。

我还补充几条：
５）每本并行优化书都可能提到：等串行程序编完了然后再并行化是一个错误的观点。那时候可能已经太复杂了。程序员从一开始就要think in parallel并行化思维。数据依赖性对并行化影响较大，要把无关的数据和操作分开，组织好数据，为并行打好基础。我觉得现代oop程序更易于并行化，而老的Fortran77要并行化反而不易，尤其是那些数据大量堆垒、操作复杂、大量使用common的程序。

６）有兴趣到这里看看，专门的并行讨论区。http://forum.csdn.net/SList/IntelMulti-core/UnClosedList/

７）你的程序多处已经多处自动并行化、矢量化（即使用了sse指令）了，与很多编译器相比已经不错了。我认为，只是我个人爱好，编译器就要最新版本的ivf，即使命令行也无所谓。然后再找个其它编译器如cvf,gfortran，pgi，帮助验证正确性。

８）你的编译选项中QaT可以去掉，如果你的程序只在自己的机器上运行不需要兼容其他cpu。
ivf编译选项里那个pgo优化有点意思，也能提高点速度，值得玩玩。

９）intel的一些资源：
intel优化白皮书http://www.intel.com/design/processor/manuals/248966.pdf
intel一些其他白皮书http://www.intel.com/cd/software/products/asmo-na/eng/338676.htm
isn论坛http://softwarecommunity.intel.com/isn/Community/en%2DUS/Forums/

１０）我也是菜鸟，大家一起切磋，共同提高。

8 楼

cxcctv [专家分：240] 发布于 2008-05-02 08:51:00

谢谢你!原以为把串行写出来后改成并行就行,现在看来不行了,不能满足课题组的要求,看样子我的程序要动大手术了,过几天我把并行程序写出后大家共同讨论!!

9 楼

cxcctv [专家分：240] 发布于 2008-05-02 09:29:00

f2003大侠的这句话:"我听说ivf不是一个完整的编译器？连接器依靠vs，要是系统里没有64位的连接器，显然还是制造不了64位程序".

不知道有没有高手知道这个连接器? 多谢高手们留言!

[em2]

我来回复

您尚未登录，请登录后再回复。点此登录或注册

主题：Fortran在自动并行时出错了,请高手支招,急急...

回复列表（共9个回复）

我来回复

程序员工具箱 new

代码片段

本版新帖

主题：Fortran在自动并行时出错了,请高手支招,急急...

回复列表 （共9个回复）

我来回复

程序员工具箱 new

代码片段

本版新帖

回复列表（共9个回复）