search

lofar-astron / prefactor

Pre facet calibration pipeline
http://www.astron.nl/citt/prefactor
GNU General Public License v3.0
29 stars 28 forks source link

wsclean segfaults with padding (trim) #37

Closed wndywllms closed 8 years ago

wndywllms commented 8 years ago

Using the padding (trim) options in Initial-Subtract (in commit https://github.com/lofar-astron/prefactor/commit/6d13f127833c503215a22c3d2e05f62483b79ff5), I'm now getting a lot of segfaults from wsclean.

It could be that the new size is not a bsmooth number (I am surprised that this makes it segfault and not just inefficient) -- in one test case I changed the "-size 8601 8601 -trim 5376 5376" to "-size 8640 8640 -trim 5379 5379" so size is bsmooth and it no longer segfaults.

I changed in the pipeline InitSubtract_sort_andcompute script the `imsize*_padso that it is also computed byget_optimum_size` and with that change I had no more segfaults on 24 bands. See https://github.com/wllwen007/prefactor/commit/7ccf41ed9a4111ce8f236a6094654e16edd3dcf2

I guess here that size should be a bsmooth number, but trim does not necessarily have to be (anyone know for sure?) - I've just made them both bsmooth as that was quicker to implement. (presumably imsize_*_pad_stretch should also be bsmooth but I haven't tested the y-stretching)

AHorneffer commented 8 years ago

I find it hard to see why 5376(=3 * 256 * 7) is worse than 5379(=3 * 11 * 163), but using get_optimum_size() in the size calculations seems to be a good idea!