我使用gnu FFTW 3.3.8来计算具有复杂输入的一维FFT,由于某些原因,它无法按如下所述工作。
当使用N <= 16的FFTW(来自http://www.fftw.org/的FFTW 3.3.8)时,代码运行良好(N是数组长度,所有元素都等于1)。对于N> = 22,输入和输出均返回零,在我看来,这两个值之间例如16
program hello
implicit none
integer N
parameter (N = 20)
double complex, dimension (N) :: in
double complex, dimension (N) :: out
real pi, one
integer i
double precision xone
xone=1.0000000000
pi=4.0*atan(one)
! generate input
do i=1,N
in(i)=xone
end do
call calc_fft(N, in, out)
! print output
do i=1,N
write(*,*)real (in(i)), real (out(i))
end do
! output data into a file
open(1, file = 'dataM.dat', status='new')
do i = 1,N
write(1,*) in(i), out(i)
end do
close(1)
end program hello
subroutine calc_fft(N,in,out)
integer N
double complex, dimension (N) :: in
double complex, dimension (N) :: out
integer*8 plan
integer i
call dfftw_plan_dft_1d(plan,N,in,out,FFTW_FORWARD,FFTW_ESTIMATE)
call dfftw_execute_dft(plan, in, out)
call dfftw_destroy_plan(plan)
end subroutine
gfortran testfftw.f90 -L / usr / lib64 / -lfftw3
N = 16时的输入/输出
1.0000000000000000 16.000000000000000
1.0000000000000000 0.0000000000000000
1.0000000000000000 0.0000000000000000
1.0000000000000000 0.0000000000000000
1.0000000000000000 0.0000000000000000
1.0000000000000000 0.0000000000000000
1.0000000000000000 0.0000000000000000
1.0000000000000000 0.0000000000000000
1.0000000000000000 0.0000000000000000
1.0000000000000000 0.0000000000000000
1.0000000000000000 0.0000000000000000
1.0000000000000000 0.0000000000000000
1.0000000000000000 0.0000000000000000
1.0000000000000000 0.0000000000000000
1.0000000000000000 0.0000000000000000
1.0000000000000000 0.0000000000000000
gfortran testfftw.f90 -L/usr/lib64/ -lfftw3
Input / Output for N=20
0.0000000000000000 0.0000000000000000
0.0000000000000000 0.0000000000000000
0.0000000000000000 0.0000000000000000
0.0000000000000000 0.0000000000000000
0.0000000000000000 0.0000000000000000
0.0000000000000000 0.0000000000000000
0.0000000000000000 0.0000000000000000
0.0000000000000000 0.0000000000000000
0.0000000000000000 0.0000000000000000
0.0000000000000000 0.0000000000000000
0.0000000000000000 0.0000000000000000
0.0000000000000000 0.0000000000000000
0.0000000000000000 0.0000000000000000
0.0000000000000000 0.0000000000000000
0.0000000000000000 0.0000000000000000
0.0000000000000000 0.0000000000000000
0.0000000000000000 0.0000000000000000
0.0000000000000000 0.0000000000000000
0.0000000000000000 0.0000000000000000
0.0000000000000000 0.0000000000000000
我正在使用Scientific Linux 7.6版(氮)
uname -a :
Linux localhost.localdomain 3.10.0-957.el7.x86_64 #1 SMP Tue Oct 30 14:13:26 CDT 2018 x86_64 x86_64 x86_64 GNU/Linux
gcc-gfortran-4.8.5
Intel(R)CoreTM i3-3240 CPU @ 3.40GHz
具有4GB的内存
我也尝试了具有类似效果的另一个CPU
答案 0 :(得分:1)
一个大问题似乎是calc_fft()
没有implicit none
,因此隐式键入适用...
subroutine calc_fft(N,in,out)
integer N
double complex, dimension (N) :: in
double complex, dimension (N) :: out
...
如果我们将implicit none
添加为
subroutine calc_fft(N,in,out)
implicit none !<--
integer N
double complex, dimension (N) :: in
double complex, dimension (N) :: out
...
然后gfortran给出消息
testfftw.f90:37:67:
call dfftw_plan_dft_1d(plan,N,in,out,FFTW_FORWARD,FFTW_ESTIMATE)
1
Error: Symbol 'fftw_estimate' at (1) has no IMPLICIT type
testfftw.f90:37:53:
call dfftw_plan_dft_1d(plan,N,in,out,FFTW_FORWARD,FFTW_ESTIMATE)
1
Error: Symbol 'fftw_forward' at (1) has no IMPLICIT type
此处,FFTW_FORWARD
和FFTW_ESTIMATE
是一些需要通过FFTW头文件定义的参数(否则,这些参数被视为默认的real
变量,而没有{{1} }!)。
implicit none
然后我们重新编译为
subroutine calc_fft(N, in, out)
implicit none
include 'fftw3.f' !<--
integer N
并获得预期的结果。 (包含文件的位置可能因计算机/操作系统而异,ScientificLinux7似乎在gfortran testfftw.f90 -L/usr/lib64 -I/usr/include -lfftw3
中包含了它们。请参阅this page,并在必要时将它们安装为/usr/include
。此外,要获得最佳性能,使用不同的FFTW软件包可能要比从# yum install fftw-devel
获得的软件包更好,但这是另一回事...)