x86 snapshot

[Matthew Fluet]

> These are all compile times, right; not run times.  Still, an accross
> the board win except for nucleic.  Any notion on what is happening
> there? 

I'm not entirely sure, but it seems to be related to the copyPropagate
phase.  I looked at the timings for nucleic, barnes-hut, ratio-regions,
and tensor (the ones with approximately the same compile times under the C
backend).  The ratio of time spent in copyPropagate to the total
outputAssembly time for the three were: 

nucleic        0.31
tensor         0.02
ratio-regions  0.02
barnes-hut     0.07

I'll investigate what characteristic of nucleic is really causing the
problem.