[daip] IMAGR ver. 31DEC07 now works for ARCH=SUL!

[Eric Greisen]

Andrzej Marecki writes:
 > On Thu, Jan 18, 2007 at 11:36:23AM -0700, Eric Greisen wrote:
 > 
 > > It is not now safe so I think I will get rid of the routine. 
 > 
 > [...]
 > 
 > OK, so I see that last night the MNJ removed two routines and IMAGR seems to
 > work now although the images produced with ver. 31DEC06 and 31DEC07 (as it
 > is on 19.01) are slightly different. These differences are small yet they
 > are visible when the two images are TVLOAD-ed to two channels and then
 > TVBLINK-ed. This is a bit strange as this effect does not take place for
 > ARCH=LINUX - here the images produced with ver. 31DEC06 and 31DEC07 are
 > bit-by-bit exactly the same. By the way, the dirty beams made with ver. 
 > 31DEC06 and 31DEC07 under Solaris are identical. Thus, it looks that this is
 > clean process that runs in a slightly different manner for ARCH=SUL
 > depending on the version.
 > 
 > Do you plan to further investigate this issue and possible make some 
 > additional changes?
 > 

I think not.  I would worry that the old version of QCLNSU was not
entirely reliable even before we started adding large offsets to the
integer addresses.  Thus it may have made minor errors in the inner
Clean loop causing the choice of Clean components to change slightly.
The outer loops were fine, so whatever Clean components were taken
were correctly subtracted from the uv data before makingh the next
residual image.  The Clean algorithm suffers from being adaptive and
so is likely to wander to slightly different answers depending on very
minor choice differences.  

Eric Greisen