More on $PCT
------------

Pct library behavior is controlled throught the PCT variable.  Syntax for the
value of this key environment variable is as follows.  If $PCT contains:
  t       truncate counts -- do not add to pre-existing ones
  p       name pct file simply "pc.pct" in the current working directory
  v       verbose mode -- print out some diagnostics
  l       log mode instead of histogram mode
  cN      circular log mode, at most N records into the past
  gB      binning granuarity == B program bytes per bin
  wW      width in bytes of the counter integer -- 1, 2, 4
  qQ      sampling time quantum is Q microseconds
  Ix      x=V|R|P => use ITIMER_VIRT, ITIMER_REAL, ITIMER_PROF timers
  rR      random intervals a mean inter-sample time of R*Q
  oPATH then if PATH is a directory, pctfile="PATH/basename(ARGV[0]).pct"
             else pctfile="PATH", which needs to be write()-able/creat()-able.
	NOTE: Whitespace after the 'o' is significant.  PATH is considered to
	      be the substring from the very first char after 'o' to the end
	      of the string.

	      Also note that, currently, directory names need a trailing '/'

For example,
	$ PCT='vtp' ./myprogram
generates the output
	./myprogram: PC counting activated
	[ whatever other output myprogram has.. ]
and a file:
	pc.pct
which is truncate()ed before being used.

As another example, assuming a directory "/home/foobar/pct"
	$ PCT='t o/tmp/pct/' ./myprogram
writes counts to "/tmp/pct/myprogram.pct".  "pct-ln" is pretty smart.  E.g.
"pct-ln /tmp/pct/myprogram.pct | pct-merge" will likely do something useful.

This helps one keep profiling data organized, but easily accessible.  It also
helps ensure that file storage for profile counts is mmap()d to local, rather
than remote files (which just helps avoid inaccuracies due to high overhead for
counter updates).