x86 update

[Matthew Fluet]

> 	 translateChunk starting
> /tmp/fileD5f7bm.0.S 8524: GC
> Starting gc.
> fromSpace 224,886,784  toSpace 224,886,784
> Finished gc.
> time(ms): 1,230
> live(bytes): 38,676,528 (17.2%)
> 	 translateChunk finished in 1.370
...
> Starting gc.
> fromSpace 224,886,784  toSpace 224,886,784
> Out of memory.
> Command exited with non-zero status 1
> 927.38user 7.83system 15:38.02elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
> 0inputs+0outputs (3138major+386559minor)pagefaults 0swaps
> make[1]: Leaving directory `/home/sweeks/mlton/src/mlton'
> make[1]: *** [mlton] Error 1
> make: *** [/home/sweeks/mlton/src/mlton/mlton] Error 2

This is sort of interesting; the longest time for a translateChunk in that
log was 1.610; the 1.370 above was the second lonest, and the third
longest was 0.090 (more than half are 0.0).  In other words, it died on an
extra large chunk.  


> Your guess is right.  The G1 mlton was built with -native-live-transfer 
> false.

Good.  That narrows down some of the possibilities.

> Have you run the full set of regressions with -native-live-transfer
> true?  I didn't.

I had with the new-codegen/old-precodegen combination.  I'll try with the
new/new combination.