Notes about compiler benchmarks
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
This file contains informal notes about the peculiarities of
individual benchmark programs, so that I don't keep re-inventing the
wheel.  Just add to this file if you find something new and
interesting.

T18223
~~~~~~
This test is full of casts and coercions.

!8600 led to a 20% decrease in compiler allocation, although the
program was the same size after every step.  I suspect that it was the
patch (part of !8600) that switched off eta-expansion in stable
unfoldings; but I'm not sure.