This directory support the fftlib specializer for interfacing
to the FTDOCK program.  Contents include:

    fftlib.py         SEJITS enabled FFTW interface module
    flush_cache       A script to flush asp caches
    ftdock_Grid3D.py  The only Ftdock file I needed to change
    pyfftlib.py       A "plug compatible" numpy version of fftlib.py
    README            This README file
    run_tests.sh      A shell script to test fftlib (uses asp)
    templates         The template files used in fftlib.py
    tests             programs to test fftlib, numpy and FFTW

The rest of this file provides details about the fftlib specializer
and how to use these modules in Ftdock

-------------------------------------------

fftlib: The FFTW interface module

Exports two modules:  fftlib.py and pyfftlib.py

The FFTW export specializers is sort of working.  I use the 
modifier "sort of" since there are several minor issues I want 
to resolve.   They are fully described in the comments to the 
code so we can be sure to fix them later.  The only one that the 
user of this module needs to be aware of is the fact I must
put the location of the directory where the template file 
is located as a string in the fftlib module (file fftlib.py): 

    installDir="\home\vagrant\fftlib"

There has to be a way around this, but I don't know what it is.
Of course, I also had to modify PYTHONPATH in the runscript for 
ftdock to tell it where to find my fft library

    #!/bin/bash
    PYTHONPATH=`pwd`:../allcomb:../fftlib:../asp:${PYTHONPATH} python 
    ftdock_main.py --m1 1CHG.pdb --m2 1HPT.pdb --gr 8.0 --st 8.0 
    --ar 180.00 --ts 5

There are two modules in fftlib

     fftlib.py    Calls fftw for forward and inverse ffts
     pyfftlib.py  Calls the analogous numpy fft routines

They export the same interface so in code you just have to 
change the import line to get one or the other.  For example
in ftdock_Grid3D.py I added the line

     import fftlib as fft

You can recover the numpy fft solution just by changing that to

    import pyfftlib as fft

The other change I had to make to ftdock_Grid3D.py is in the calls
to the FFTs themselves.  Instead of just calling numpy

     return np.fft.fftn(self.grid)

I had to explicit insert the library object

     return fft.lib().fftn_R2C(self.grid)

Like wise for the inverse fft I typed

     return np.real_if_close(fftlib().ifftn_C2R(self.grid),tol=100000)

Note since I'm not doing an explicit complex to real fft I don't need
the "real_if_close" function.  I leave it there, however to support
easy transition between the all-numpy and the FFTW versions of 
the fftlib modules.

Currently, I am using the un-normalized FFTs from fftw.  This does 
change the answers a bit, but I think in a way that is not relevant
to the function of the application.  When you run ftdock with the
all numpy interface to fftlib (i.e. pyfftlib) I get the following 
results

Begin...
Finished reading PDB:  hydrolase zymogen (serine proteinase)   01-mar-75   1chg
Finished reading PDB:  serine protease inhibitor
(31000.000000000011, 4, 7, 4, 0.0, 1.0, 0.0)
(31000.000000000007, 5, 4, 2, 0.0, -1.0, 1.0)
(30000.000000000007, 4, 7, 4, 0.0, 1.0, -1.0)
(28000.000000000033, 5, 5, 2, -1.0, 0.0, -1.0)
(28000.0, 4, 6, 3, -1.0, -1.0, 1.0)
End...

This compares to the answer with the "pyfftlib" module

Begin...
Finished reading PDB:  hydrolase zymogen (serine proteinase)   01-mar-75   1chg
Finished reading PDB:  serine protease inhibitor
(31.000000000000068, 4, 7, 4, 0.0, 1.0, 0.0)
(30.999999999999989, 5, 4, 2, 0.0, -1.0, 1.0)
(30.000000000000057, 4, 7, 4, 0.0, 1.0, -1.0)
(28.000000000000021, 5, 5, 2, -1.0, 0.0, -1.0)
(27.999999999999972, 4, 6, 3, -1.0, -1.0, 1.0)
End...

Let me know if you need me to normalize the ffts.