Fix shootout benchmarks

[brson]

Several of them do not report the correct results, per Isaac Gouy:

I saw that there'd been changes to mozilla/rust/src/test/bench so I took a look and ran the programs with Rust 0.3

There are a few trivial errors --

1) binarytrees prints "long lived trees of depth" instead of "long lived tree of depth"


2) chameneosredux -- the whitespace printed by show_number should be the same as printed by show_color

See the attached file.

3) spectralnorm prints an extra \n

4) threadring prints an extra \n


and less trivial errors


5) fasta -- strangely, output stops half-way through line 169 -- with another 8166 lines still to do.

6) the latest binarytrees fails with input N=20 on a 4GB machine.


stretch tree of depth 21     check: -1
2097152     trees of depth 4     check: -2097152
524288     trees of depth 6     check: -524288

[catamorphism]

Confirmed that (1) and (5), at least, are still issues. I'm not sure I understand (2), (3), and (4).

[Thiez]

The spectralnorm problem was solved in 9738c2a by @pcwalton

[Thiez]

This shouldn't be closed; as @brson pointed out at #7193, the last problem (binarytrees) is not yet fixed.

[toddaaro]

Visiting for triage. This should be dealt with, but it should be considered low priority.