Generating Profiling Datasets

Caliper

Caliper can be installed using Spack or manually from its GitHub repository. Instructions to build Caliper manually can be found in its documentation.

To record performance profiles using Caliper, you need to include cali.h and call the cali_init() function in your source code. You also need to link the Caliper library in your executable or load it using LD_PRELOAD. Information about basic Caliper usage can be found in the Caliper documentation.

To generate profiling data, you can use Caliper’s built-in profiling configurations customized for thicket: hatchet-region-profile and spot or hatchet-sample-profile. The former generates a profile based on user annotations in the code while the latter generates a call path profile (similar to HPCToolkit’s output). If you want to use one of the built-in configurations, you should set the CALI_CONFIG environment variable (e.g. CALI_CONFIG=hatchet-sample-profile).

You can read more about Caliper services in the Caliper documentation. Thicket can currently only read .cali files, that is a native Caliper output.

Adiak

Adiak can be used with Caliper to record program metadata. You can use Adiak, a C/C++ library to record environment information (user, launchdata, system name, etc.) and program configuration (input problem description, problem size, etc.). To build Caliper with Adiak support, -DWITH_ADIAK=On is required. Adiak proides built-in fucntions to collect common environment metadata that enables performance comparisons across different runs. Some common metadata that can be used with thicket are launchdate or clustername, where a user can use this metadata information to organize the performance data with the help of thicket’s capabilities.

adiak_user(); /* user name */
adiak_uid(); /* user id */
adiak_launchdate(); /* program start time (UNIX timestamp) */
adiak_executable(); /* executable name */
adiak_executablepath(); /* full executable file path */
adiak_cmdline(); /* command line parameters */
adiak_hostname(); /* current host name */
adiak_clustername(); /* cluster name */
adiak_job_size(); /* MPI job size */
adiak_hostlist(); /* all host names in this MPI job */
adiak_walltime(); /* wall-clock job runtime */
adiak_cputime(); /* job cpu runtime */
adiak_systime(); /* job sys runtime */

adiak::value() records key:value pairs with overloads for many data types

#include <adiak.hpp>

vector<int> ints { 1, 2, 3, 4 };

adiak::value(“myvec”, ints);
adiak::value(“myint”, 42);
adiak::value(“mydouble”, 3.14);
adiak::value(“mystring”, “hi”);
adiak::value(“mypath”, adiak::path(“/dev/null”));
adiak::value(“compiler”, adiak::version(“gcc@8.3.0”));

adiak_nameval() uses printf()-style descriptors to determine data types

#include <adiak.h>

int ints[] = { 1, 2, 3, 4 };

adiak_nameval(“myvec”, adiak_general, NULL, “[%d]”, ints, 4);
adiak_nameval(“myint”, adiak_general, NULL, “%d”, 42);
adiak_nameval(“mydouble”, adiak_general, NULL, “%f”, 3.14);
adiak_nameval(“mystring”, adiak_general, NULL, “%s”, “hi”);
adiak_nameval(“mypath”, adiak_general, NULL, “%p”, “/dev/null”);
adiak_nameval(“compiler”, adiak_general, NULL, “%v”, “gcc@8.3.0”);

You can learn more about the Adiak library in the Adiak documentation.