these programs create snapshots of l1 processes
and analyze their heap storage.

we use these programs to tune the *performance*
of the code l1 generates to analyzes heaps.

s0 creates a snapshot of the process in which it
executes (snap0).  it uses a l1 names file
(../l1.names) to traverse the snapshot heap and report
statistics.  it also emits a new names file
(./stat.names) that contains the subset of l1.names
needed to traverse the heap.

s0 requires the l1 -r flag because its uses slimns
to compute stat.names.

s1 is very similar to s0: it creates a snapshot of the
process in which it executes (snap1), traverses the
snapshot heap, and reports the same statistics as s0.
the difference is that s1 loads stat.names instead of
l1.names, which substantially reduces its code
allocation (5.7MB vs 76.9MB), and thus the size of its
snapshot and the time required to traverse it.

s2 re-runs either of the heap analyses of s0 or s1
on previously computed snap0 or snap1 snapshots.
by default it re-runs s1.  in both cases it uses
the stat.names namespace.

s2 is a benchmark for iterative performance
experiments: s2 on snap1 is a quick experiment; s2 on
snap0 is a longer one.  s0 and s1 together bootstrap
these experiments.

the scripts expect to run in the directory
they occupy.

usage:

% cd <l1 source directory>
% make
% dwarf2cqct < l1 > l1.names
% cd perf
% ../l1 -r ./s0.cqct
% ../l1 ./s1.cqct

% ../l1 ./s2.cqct            # quick experiment
% ../l1 ./s2.cqct snap0      # longer one