Differences

This shows you the differences between two versions of the page.

--- perf [2014/03/29 19:09] – [Perf] mikael.ohman
+++ perf [2014/08/22 00:11] (current) – [Perf] mikael.ohman
@@ Line 1: / Line 1: @@
 ====== Measuring performance ======
 In general, to obtain useful information from binaries, certain compile flags should be used.
 ===== Perf =====
@@ Line 11: / Line 10: @@
 which will generate a perf.data file.
-Then you can visualize the results in several ways. A good, simple to understand method is to use gprof2dot to generate a complete callgraph:
+Then you can visualize the results in several ways. A good, simple to understand method is to use [[http://code.google.com/p/jrfonseca/wiki/Gprof2Dot|Gprof2Dot]] to generate a complete callgraph:
-''perf script | ./gprof2dof.py -f perf | dot -Tsvg  -o output.svg''
+''perf script | ./gprof2dot.py -f perf | dot -Tsvg  -o output.svg''
+Or use the ncurses program
+''perf report -G %%--%%sort comm,dso''
+Perf has very small overhead, but only does statistical sampling.
+//Users of perf will likely first need to turn off the restrictions before they are able to run ''perf record'' as normal users://
+''sudo sh -c %%"%%echo 0 > /proc/sys/kernel/kptr_restrict%%"%%''
 ===== Callgrind =====
 Callgrind is a tool in valgrind, and should only be used on medium. This has a huge overhead, so expect OOFEM to run over a hundred times slower through valgrind.
 Simply run
 ''valgrind --tool=callgrind ./oofem -f myinputfile.in''
 and you will produce a new file named ''callgrind.out.123456'' where the numbers at the end are randomized.
 Open this file in Kcachegrind to visualize the results.