Need to lower memory requirements in distortionCorrect

[KathleenLabrie]

For GMOS LS, distortionCorrect triples the required memory. There might be a way to do better

From Chris - What is going on right now.

distortionCorrect() for GMOS (which I assume is what we're talking about) also mosaics the frames. The input frames are 12 separate extensions so the first thing it has to do is tile the individual CCDs. Since numpy requires that arrays be contiguous in memory this doubles the memory. Then it has to create the mosaicked frames, which is a tripling of the memory. The problem after that is probably one of fragmentation. The mosaicked frames are the largest python objects and it may be that the only place to put them is on top of the heap because the space freed up isn't big enough to be reused by these objects. Having said that, because that primitive creates new AD objects (adoutputs), the input ADs (adinputs) don't go out of scope until the primitive ends so when you mosaic the first AD the memory of its input isn't freed up immediately.

From Chris - A possible solution to look into

It might help if, instead of creating a new adoutputs list each output AD was poked back into the adinputs list to replace the thing it was created from, doing something like replacing the loop with for i, ad in enumerate(adinputs) and instead of adoutputs.append(ad_out) having adinputs[i] = ad_out. This will set the reference count to that object to zero and the memory should get freed by the garbage collector.