removed casapy-free CASA and pyfftw as default build options

Author: caseyjlaw

README.md

@@ -17,50 +17,48 @@ Requirements:
 -- python (probably version >2.6..?)
 -- numpy, matplotlib
 -- CASA 4.0 or higher
--- patchelf (optional)
--- pyFFTW (optional)
+
+Optional (requires some hacking):
+-- patchelf (to get casapy-free CASA)
+-- pyFFTW (for accelerated fft)
 
 Usage:
 
-For full installation (see below), start a python session:
-$ python
-> import leanpipedt
-> d = leanpipedt.pipe_thread(filename='data.ms', scan=0, nskip=0, iterint=100, nints=100, spw=[0], chans=range(5,60), selectpol=['RR','LL'], searchtype='imgall', filtershape='z', secondaryfilter='fullim', dmarr=[0,100], dtarr=[1,2], size=512*58, res=58, nthreads=8, gainfile='cal.g1', bpfile='cal.b1')
+For baseline mode, start a casapy session:
+$ casapy
+CASA> import leanpipedt
+CASA> d = leanpipedt.pipe_thread(filename='data.ms', scan=0, nskip=0, iterint=100, nints=100, spw=[0], chans=range(5,60), selectpol=['RR','LL'], searchtype='imgall', filtershape='z', secondaryfilter='fullim', dmarr=[0,100], dtarr=[1,2], size=512*58, res=58, nthreads=8, gainfile='cal.g1', bpfile='cal.b1')
 
 This command will read the first 100 interations from scan 0 (first scan) of 'data.ms'. It will look for the first spectral window and save channels 5,59 (inclusive) for two, orthogonal, circular polarizations. The data will have calibration applied from the gain and bp files. A zero-mean will be subtracted, data dedispersed, and resampled for DM=0 and 100 pc/cm3 and time widths of 1 and 2 integrations. The image search will use a uv grid cell size of 58 lambda and an image size of 512 pixels square (appropriate 2 pixels per beam for L-band, VLA images; covers twice the FWHM).
 
-An alternative version (without patchelf dependency) requires running leanpipedt from within a casapy session.
-
 Files:
 -- leanpipedt.py: master script that defines search pipeline.
 -- applycals2.py and applytelcal.py: script to parse CASA calibration tables (gain and bp) or telcalfile. Called by leanpipedt.py.
 -- leanpipedt_cython.pyx: Cython-accelerated utility functions, including dedispersion of visibilities.
 -- qimg_cython.pyx: Cython-accelerated imaging functions.
 -- setup.py: script to compile Cython into shared-object libraries.
--- tpipe.py: old version of search script (class-based structure, includes Miriad format data support).
+-- tpipe.py: deprecated version of search script (class-based structure, includes Miriad format data support).
 
 Build Instructions:
 1) Install CASA
 
-2) Build casapy-free CASA (optional; not possible on OSX yet)
+(optional 2) Build casapy-free CASA (not possible on OSX yet)
 This step builds python modules to import CASA into any Python session (no "casapy" session needed).
 This requires fixing some links to libraries with patchelf.
 Full instructions at http://newton.cx/~peter/2014/02/casa-in-python-without-casapy. 
 See also the casapatch.sh script, which must be edited and run to build new CASA python modules.
 Once complete, you should be able to simply type "import casac" and get access to CASA's ms and table tools.
 Finally, download "casautil.py" at https://github.com/pkgw/pwpy/blob/master/intflib/casautil.py.
-
-If you skip this step, you will need to remove references to "casautil" and the "ms" and "tb" definitions in leanpipedt.py and applycals2.py. Running from inside casapy will automatically define those objects in the Python namespace.
+To use this, you will need to uncomment references to "casautil" and the "ms" and "tb" definitions in leanpipedt.py and applycals2.py.
 
 3) Install Cython
 See http://cython.org to build Cython, an optimizing static compiler for Python.
 
-4) Install pyFFTW (optional)
+(Optional 4) Install pyFFTW
 Get pyFFTW at https://github.com/hgomersall/pyFFTW. Note that there is an unsupported library with similar name at https://pypi.python.org/pypi/PyFFTW3/0.2.1.
 This supposedly requires Cython 0.15, FFTW 3.3, Python 2.7, Numpy 1.6. Although I've made it work with Python 2.6.
 FFTW is an optimized FFT library and can boost performance by a factor of 2. Can be tricky to do this stage, so skipping is ok for many use cases.
-
-If you skip this step, you will need to edit leanpipedt.py, leanpipedt_cython.pyx, and qimg_cython.pyx to redefine "fft". Instead of "fft" being an object from pyFFTW, it will be from numpy.fft.
+To use this, you will need to change commented lines in leanpipedt.py and qimg_cython.pyx to redefine "fft" and add references to pyfftw.
 
 5) Build accelerated functions
 Edit "setup.py" file to define how Cython will compile (basically setting filename/function to be compiled).

applycals2.py

@@ -1,8 +1,9 @@
 import numpy as n
 import os, glob
-import casautil
 
-tb = casautil.tools.table()
+# optional import for casapy-free CASA
+#import casautil
+#tb = casautil.tools.table()
 
 class solutions():
     """ Container for CASA caltable(s).

casapatch.sh

@@ -17,13 +17,16 @@
 import leanpipedt_cython as lib
 #import leanpipe_external as lib
 import qimg_cython as qimg
-#from numpy import fft as fft
-import pyfftw.interfaces.NUMPY_fft as fft
-import casautil
 
-# set up tools
-ms = casautil.tools.ms()
-qa = casautil.tools.quanta()
+# Optional imports that have extra dependencies
+# can choose fft from numpy or pyfftw.
+from numpy import fft as fft
+#import pyfftw.interfaces.NUMPY_fft as fft
+#
+# optionally can use CASA outside of casapy session. requires hacking CASA shared-object libraries...
+#import casautil
+#ms = casautil.tools.ms()
+#qa = casautil.tools.quanta()
 
 def numpyview(arr, datatype, shape):
     """ Takes mp.Array and returns numpy array with shape of data in MS.

leanpipedt.py

@@ -2,9 +2,9 @@ import numpy as n
 cimport numpy as n
 cimport cython
 # can choose between numpy and pyfftw
-#from numpy import fft
-import pyfftw
-import pyfftw.interfaces.NUMPY_fft as fft
+from numpy import fft
+#import pyfftw
+#import pyfftw.interfaces.NUMPY_fft as fft
 
 CTYPE = n.long
 ctypedef n.long_t CTYPE_t
@@ -213,7 +213,7 @@ cpdef imgallfull(n.ndarray[n.float32_t, ndim=2] u, n.ndarray[n.float32_t, ndim=2
                     for p in xrange(len3):
                         grid[t, cellu, cellv] = data[t,i,j,p] + grid[t, cellu, cellv]
 
-    pyfftw.interfaces.cache.enable()
+#    pyfftw.interfaces.cache.enable()
     for t in xrange(len0):
         grid[t] = fft.ifft2(grid[t])
 # too slow
@@ -267,7 +267,7 @@ cpdef imgallfullfilter(n.ndarray[n.float32_t, ndim=2] u, n.ndarray[n.float32_t,
                         grid[t, cellu, cellv] = data[t,i,j,p] + grid[t, cellu, cellv]
 
     # make images and filter based on threshold
-    pyfftw.interfaces.cache.enable()
+#    pyfftw.interfaces.cache.enable()
     candints = []; candims = []; candsnrs = []
     for t in xrange(len0):
         im = fft.ifft2(grid[t]).real
@@ -325,7 +325,7 @@ cpdef imgallfullfilterxy(n.ndarray[n.float32_t, ndim=2] u, n.ndarray[n.float32_t
                         grid[t, cellu, cellv] = data[t,i,j,p] + grid[t, cellu, cellv]
 
     # make images and filter based on threshold
-    pyfftw.interfaces.cache.enable()
+#    pyfftw.interfaces.cache.enable()
     candints = []; candims = []; candsnrs = []
     for t in xrange(len0):
         im = fft.ifft2(grid[t]).real
@@ -391,7 +391,7 @@ cpdef imgallfullfilterxy2(n.ndarray[n.float32_t, ndim=2] u, n.ndarray[n.float32_
                         grid[t, cellu, cellv] = data[t,i,j,p] + grid[t, cellu, cellv]
 
     # make images and filter based on threshold
-    pyfftw.interfaces.cache.enable()
+#    pyfftw.interfaces.cache.enable()
     candints = []; candims = []; candsnrs = []
     for t in xrange(len0):
         im = fft.ifft2(grid[t]).real*int(ndimx*ndimy)/float(nonzeros)
@@ -449,7 +449,7 @@ cpdef imgallfullfilterminxy(n.ndarray[n.float32_t, ndim=2] u, n.ndarray[n.float3
                         grid[t, cellu, cellv] = data[t,i,j,p] + grid[t, cellu, cellv]
 
     # make images and filter based on threshold
-    pyfftw.interfaces.cache.enable()
+#    pyfftw.interfaces.cache.enable()
     candints = []; candims = []; candsnrs = []
     for t in xrange(len0):
         im = fft.ifft2(grid[t]).real
@@ -523,7 +523,7 @@ cpdef imgallfullw(n.ndarray[n.float32_t, ndim=2] u, n.ndarray[n.float32_t, ndim=
 #                            grid[t, cellu-ksize/2:cellu+ksize/2+1, cellv-ksize/2:cellv+ksize/2+1] = uvker*data[t,i,j,p,None,None] + grid[t, cellu-ksize/2:cellu+ksize/2+1, cellv-ksize/2:cellv+ksize/2+1]
 #                            grid[t, cellu, cellv] = data[t,i,j,p] + grid[t, cellu, cellv]    # no conv gridding
 
-    pyfftw.interfaces.cache.enable()
+#    pyfftw.interfaces.cache.enable()
     for t in xrange(len0):
         grid[t] = fft.ifft2(grid[t])