xdimsum

Lindsey Davis wrote on Jan 22, 2002

Hello Christophe,

    Mkmask is borrowed directly from the old dimsum package. Although I have
replaced a set of imreplace calls with a single imexpr call, I have not
changed the definition or use of flagval, at least not intentionally!  I think
the "# Set flag ..." comment in the mkmask script is misleading.  Flagval
is just a number outside the usual data range for the normalized noise image.

    I am more concerned about the floating point error out of imexpr or
possibly imarith ? I have had reports of floating point errors out of imexpr on
DecAplha but only in expressions involving exponentiation. This problem
disappeared with different ("later") versions of the operating system. I have
also had reports of occasional "high performance trap" errors in imexpr, and
other tasks, again on a DecAlpha, but have not been able to duplicate them here.
Are you running xdimsum on a DecAlpha ? Is it possible your input images
contain funny for example very large numbers numbers. 

    If you would like me to look into this further please send me fits files
of your input to mkmask and a listing of your mkmask parameters. You can
ftp fits files to 140.252.1.1, login as anonymous, cd to pub and transfer the
files in binary mode.


					       Regards,
					       Lindsey Davis

---

prme wrote on Jan 23, 2002

Christophe -

I changed the flag value in mkmask from (minval - 1) to 
2*minval - maxval = minval - (maxval - minval) 
after running into roundoff problems when minval was a big
number.  I neglected to update the comment line, however.

Peter Eisenhardt

---

Last post on Jan 23, 2002