1) LIS.cond2s contains some suggestions about trivial improvements which
   would improve the quality of the code by a fair bit.


2) LIS.csieve is a disassembly of the sieve benchmark run through NORCROFTs
   C compiler - which does a dataflow analysis to see if it can keep
   variables in registers throughout a procedure.  The code for the
   identical version of the sieve in Imp is in LIS.sieve - NORCROFT's
   code wins hands-down, although the Imp code could be improved by
   simple live/dead analysis at a local level, with delayed storing of
   variables and moving invariants out of short loops.


3) After observing the quality of the NorCroft C, I put the conditional-test
   file through it (after translating to C) - it produced almost exactly
   the code I generated by hand...