ArrayArgSwitch

In the Fortran array A(I,J) is stored with I varying fastest

Dimension A2d(27,31),A1d(27*31)

Equivalence (A1d(1),A2d(27,31))

A1d(I+(31)*(J-1))=A2d(I,J)

If the Array A(I,J) was written to the disk with the commands

OPEN(1,FILE=’Test1.unf’,FORM=’UNFORMATTED’)

Do j=1,31

Write(1)(A(I,j),i=1,27)

Enddo

CLOSE(1)

The one dimensional array can be read and used with the code

OPEN(2,FILE=’Test2.unf’,FORM=’UNFORMATTED’)

Dimension A1d(27)

Do j=1,31

READ(1)(Aid(I),i=1,27)

Call proc(A1d,27)

Write(2)(A1d(I),i=1,27)

Enddo

CLOSE(2)

The segment of code above uses much less memory than the first two segments. In ..\..\Fourier\DFT2-FFT2.docx, only the simple transform uses the entire array. The intermediate transform requires a single re-order, while the fft requires “essentially” two re-orders. If the procedure is a fast Fourier transform in the x direction. The result is

dd(x,y)àDd(fx,y)

If Dd will fit in memory, the easiest way to make the y transform is coded into reorder.for

open(1,file=iin,form='unformatted')

do j=1,ny

read(1)(dd(i,j),i=1,nx)

enddo

close(1)

open(2,file=iiout,form='unformatted')

do i=1,nx

write(2)(dd(i,j),j=1,ny)

enddo

This is both simple and fast Dd(fx,y) àDD(fy,fx). On a computer with 4g of memory, the nalues of nx and ny could be 10,000 each and the allocation preceeding the step above still worked. – On older computer with less memory about 1000 is the limit. The FFT, however, has the speed necessary to handle enormous arrays in nx and ny. If the allocation fails, it makes sense to use a direct access file to achieve the same result.

Two re-orders are needed for hh(x,y)à HH(fx,fy)

The simple reorder requires 0.22 seconds, while the DA reorder requires 0.34 seconds. This is worse but …

C:\Public\TwoDFT>twodft

n2px is power of 2 > max(nx,nfx)

enter XINT, nx, nfx, n2px

20,200,100,256

enter YINT, ny, nfy, n2py

10,100,100,128

FFT transform

sec 0.2200000