[Rcpp-devel] Google perftools profiler works best, AFAICT

[Paul Johnson]

Here's the writeup with examples on the Google profiler tool as
applied to an R package that uses RcppArmadillo.

http://pj.freefaculty.org/blog/?p=140  "R package profiling with
Google Preftools"

I had forgotten until yesterday, but Dirk E. advised me in 2009 to try
this out.  After having invested a couple of hours in each of these
alternatives, I agree the Google perftools works fairly well,
certainly better than the others I've tried this week (better than
gprof, sprof, oprofile, and valgrind).

Here's the payoff picture, the graph from one function to another with
time percentages.

http://pj.freefaculty.org/blog-attachments/pprof5994.0-LD_PRELOAD.pdf

The major unsolved problem is that many of the functions are not
displayed with human readable names.

I had some limited success with the new oprofile, but I can't make
much sense of the output (yet).

In case you want to try sprof, forget it (at least on Debian and
Ubuntu), for the reasons outlined here:

http://pj.freefaculty.org/blog/?p=136

Even after you weave your way through the puzzling sequence of
environment commands, it is still impossible to analyze the output
file.

I also tried valgrind, but with R shared libraries in packages, well,
all I can say is "apparently useless for profiling."  In the past,
I've had treat success with valgrind for analysis of memory usage, but
for profiling, well, limited in value, so far as I can tell. In case
you wonder, the suggestion that we should try valgrind is in this
writeup
(http://www.admin-magazine.com/HPC/Articles/HPC-Profiling-Is-the-Key-to-Survival)

Found a Nice debate on gprof alternatives:
http://stackoverflow.com/questions/1777556/alternatives-to-gprof

Its kinda dominated by one big post against gprof and any other
compiler I've heard of. :)

pj
-- 
Paul E. Johnson
Professor, Political Science      Assoc. Director
1541 Lilac Lane, Room 504      Center for Research Methods
University of Kansas                 University of Kansas
http://pj.freefaculty.org               http://quant.ku.edu