native codegen bug

[Matthew Fluet]

Fixed and checked in.

BTW, here's the time to stabilize to a new version of mlton on the linux
cluster (i.e., given an existing stable version of mlton-compile and
runtime, mllex, mllyacc, and mlprof, and new sources, build the new
runtime, use the stable mlton-compile and runtime to build a new
mlton-compile, use the new mlton-compile and runtime to build another
mlton-compile, and use this stable mlton-compile to build new mllex,
mllyacc, and mlprof).

[fluet@lennon mlton]$ /usr/bin/time ./bin/remote-mlton-stabilize cfs12
...
MLton starting
   Compile SML starting
      pre codegen starting
         gcc -DNODEBUG -DMLton_safe=TRUE -DMLton_detectOverflow=TRUE \
             -I/tmp/mlton.27206/bin/../mlton-stable/include -o \
             /tmp/fileqzI10v /tmp/fileWXSlUd.c \
             -L/tmp/mlton.27206/bin/../mlton-stable/lib -lmlton -lm -lgmp
         /tmp/fileqzI10v /tmp/fileKRSfsv
      pre codegen finished in 96.13 + 139.12 (59% GC)
      x86 code gen starting
      x86 code gen finished in 134.69 + 55.27 (29% GC)
   Compile SML finished in 230.82 + 194.39 (46% GC)
...
MLton finished in 259.24 + 194.40 (43% GC)
446.68user 7.01system 7:36.36elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (48430major+655478minor)pagefaults 0swaps
MLton starting
   Compile SML starting
      pre codegen starting
         gcc -DNODEBUG -DMLton_safe=TRUE -DExn_keepHistory=FALSE \
             -DMLton_detectOverflow=TRUE \
             -I/tmp/mlton.27206/bin/../mlton/include -o /tmp/fileZ4rE9m \
             /tmp/fileve5aDO.c -L/tmp/mlton.27206/bin/../mlton/lib -lmlton \
             -lm -lgmp
         /tmp/fileZ4rE9m /tmp/fileHgTqXA
      pre codegen finished in 95.46 + 137.01 (59% GC)
      x86 code gen starting
      x86 code gen finished in 135.24 + 54.84 (29% GC)
   Compile SML finished in 230.70 + 191.85 (45% GC)
...
MLton finished in 259.30 + 191.86 (43% GC)
444.24user 6.98system 7:32.29elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (48306major+651936minor)pagefaults 0swaps
...
5.91user 1.82system 17:47.09elapsed 0%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (5490major+1687minor)pagefaults 0swaps

So, all told, less than 20minutes for the full round trip.  Not bad.