Working With gprofng and Threaded Applications

6 Working With `gprofng` and Threaded Applications

Collect performance data for multithreaded applications.

By default, the performance data for a multithreaded application is aggregated over all threads. If the data for individual threads, or a set of threads, is needed, then filters can be used to analyze that information.

To list all the threads for which performance data has been captured, use the -thread_list and -threads options:

gprofng display text -thread_list -threads experiment-directory-name.er

Exp Sel Total
=== === =====
  1 all     3
Objects sorted by metric: Exclusive Total CPU Time

Excl. Total    Name
CPU
 sec.      %
5.775 100.00   <Total>
2.773  48.01   Process 1, Thread 3
2.722  47.13   Process 1, Thread 2
0.280   4.85   Process 1, Thread 1

In the example output you can see that the CPU times for three threads were used during that experiment. The thread number can be used to review information for both a specific thread or a group of threads.

To review the performance data for a specific thread, use the -thread_select option with a thread number:

gprofng display text -limit 5 -thread_select 1 -functions experiment-directory-name.er

Print limit set to 5
Exp Sel Total
=== === =====
  1 1       3
Functions sorted by metric: Exclusive Total CPU Time

Excl. Total   Incl. Total    Name
CPU           CPU
 sec.      %   sec.      %
0.280 100.00  0.280 100.00   <Total>
0.126  44.84  0.267  95.37   init_data
0.068  24.20  0.104  37.01   erand48_r
0.038  13.52  0.142  50.53   drand48
0.036  12.81  0.036  12.81   __drand48_iterate