Count overruns when profiling; change units to ns.
authorPaul Eggert <eggert@cs.ucla.edu>
Tue, 2 Oct 2012 06:30:40 +0000 (23:30 -0700)
committerPaul Eggert <eggert@cs.ucla.edu>
Tue, 2 Oct 2012 06:30:40 +0000 (23:30 -0700)
commitb3ecad33ae977f9dd0bdd3e0a10cd8603f54a398
tree9641b9d035dc733b9f3600557becdcb0ffc31290
parent8cb51b66fcb7a0954437108f8a1d05d6af6dcf89
Count overruns when profiling; change units to ns.

* lisp/profiler.el (profiler-sampling-interval): Change units
from ms to ns, multiplying the default by 1000000 so that
it remains 1 ms.
(profiler-report-cpu-line-format): Give enough room for
the maximum counters on 64-bit hosts.
(profiler-report-render-calltree-1): Call them "CPU samples",
not "Time (ms)", since they are not milliseconds now (and
never really were).
* src/profiler.c (handle_profiler_signal): Count sampling intervals, not ms.
Give extra weight to samples after overruns, to attempt to count
the time more accurately.
(setup_cpu_timer): Change sampling interval units from ms to ns, since
the underlying primitives nominally do ns.
(Fprofiler_cpu_start): Document the change.  Mention that
the sampling intervals are only approximate.
lisp/ChangeLog
lisp/profiler.el
src/ChangeLog
src/profiler.c