Caliper: A Performance Analysis Toolbox in a Library

Caliper is a library to integrate performance profiling capabilities into applications. To use Caliper, developers mark code regions of interest using Caliper's annotation API. Applications can then enable performance profiling at runtime with Caliper's configuration API. Alternatively, you can configure Caliper through environment variables or config files.

Caliper can be used for lightweight always-on profiling or advanced performance engineering use cases, such as tracing, monitoring, and auto-tuning. It is primarily aimed at HPC applications, but works for any C/C++/Fortran program on Unix/Linux.

Features include:

• Low-overhead source-code annotation API
• Configuration API to control performance measurements from within an application
• Recording program metadata for analyzing collections of runs
• Flexible key:value data model to capture application-specific features for performance analysis
• Fully threadsafe implementation, support for parallel programming models like MPI
• Synchronous (event-based) and asynchronous (sampling) performance data collection
• Trace and profile recording
• Connection to third-party tools, e.g. NVidia NVProf or Intel(R) VTune(tm)
• Measurement and profiling functionality such as timers, PAPI hardware counters, and Linux perf_events
• Memory allocation annotations: associate performance measurements with named memory regions

Documentation

Extensive documentation is available here: https://llnl.github.io/Caliper/

Usage examples of the C++, C, and Fortran annotation and ConfigManager APIs are provided in the examples directory.

See the "Getting started" section below for a brief tutorial.

Building and installing

Building and installing Caliper requires cmake 3.1+ and a current C++11-compatible Compiler. Clone Caliper from github and proceed as follows:

 $ git clone https://github.com/LLNL/Caliper.git
 $ cd Caliper
 $ mkdir build && cd build
 $ cmake -DCMAKE_INSTALL_PREFIX= ..
 $ make
 $ make install

Link Caliper to a program by adding

libcaliper

g++ -o app app.o -L/lib64 -lcaliper

There are many build flags to enable optional features, such as

-DWITH_MPI

Getting started

Typically, we integrate Caliper into a program by marking source-code sections of interest with descriptive annotations. Performance profiling can then be enabled through the Caliper ConfigManager API or environment variables. Alternatively, third-party tools can connect to Caliper and access information provided by the source-code annotations.

Source-code annotations

Caliper's source-code annotation API allows you to mark source-code regions of interest in your program. Much of Caliper's functionality depends on these region annotations.

Caliper provides macros and functions for C, C++, and Fortran to mark functions, loops, or sections of source-code. For example, use

CALI_CXX_MARK_FUNCTION

#include 

void foo()
{
    CALI_CXX_MARK_FUNCTION;
    // ...
}

You can mark arbitrary code regions with the

CALI_MARK_BEGIN

CALI_MARK_END

cali_begin_region()

cali_end_region()

#include 

// ...
CALI_MARK_BEGIN("my region");
// ...
CALI_MARK_END("my region");

The cxx-example, c-example, and fortran-example example apps show how to use Caliper in C++, C, and Fortran, respectively.

Recording performance data

With the source-code annotations in place, we can run performance measurements. By default, Caliper does not record data - we have to activate performance profiling at runtime. An easy way to do this is to use one of Caliper's built-in measurement configurations. For example, the

runtime-report

CALI_CONFIG

$ cd Caliper/build
$ make cxx-example
$ CALI_CONFIG=runtime-report ./examples/apps/cxx-example
Path       Min time/rank Max time/rank Avg time/rank Time %
main            0.000119      0.000119      0.000119  7.079120
  mainloop      0.000067      0.000067      0.000067  3.985723
    foo         0.000646      0.000646      0.000646 38.429506
  init          0.000017      0.000017      0.000017  1.011303

The runtime-report config works for MPI and non-MPI programs. It reports the minimum, maximum, and average exclusive time (seconds) spent in each marked code region across MPI ranks (the values are identical in non-MPI programs).

You can customize the report with additional options. Some options enable additional Caliper functionality, such as profiling MPI and CUDA functions in addition to the user-defined regions, or additional metrics like memory usage. Other measurement configurations besides runtime-report include:

• loop-report: Print summary and time-series information for loops.
• mpi-report: Print time spent in MPI functions.
• callpath-sample-report: Print a time spent in functions using call-path sampling.
• event-trace: Record a trace of region enter/exit events in .cali format.
• hatchet-region-profile: Record a region time profile for processing with the hatchet library or cali-query.

See the "Builtin configurations" section in the documentation to learn more about different configurations and their options.

You can also create entirely custom measurement configurations by selecting and configuring Caliper services manually. See the "Manual configuration" section in the documentation to learn more.

ConfigManager API

A distinctive Caliper feature is the ability to enable performance measurements programmatically with the ConfigManager API. For example, we often let users activate performance measurements with a command-line argument.

With the C++ ConfigManager API, built-in performance measurement and reporting configurations can be activated within a program using a short configuration string. This configuration string can be hard-coded in the program or provided by the user in some form, e.g. as a command-line parameter or in the programs's configuration file.

To use the ConfigManager API, create a

cali::ConfigManager

add()

start()

flush()

#include 
// ...
cali::ConfigManager mgr;
mgr.add("runtime-report");
// ...
mgr.start(); // start requested performance measurement channels
// ... (program execution)
mgr.flush(); // write performance results

The

cxx-example

-P

-P runtime-report

$ ./examples/apps/cxx-example -P runtime-report
Path       Min time/rank Max time/rank Avg time/rank Time %
main            0.000129      0.000129      0.000129  5.952930
  mainloop      0.000080      0.000080      0.000080  3.691740
    foo         0.000719      0.000719      0.000719 33.179511
  init          0.000021      0.000021      0.000021  0.969082

See the Caliper documentation for more examples and the full API and configuration reference.

Authors

Caliper was created by David Boehme, [email protected]

A complete list of contributors is available on GitHub.

Major contributors include:

• Alfredo Gimenez (libpfm support, memory allocation tracking)
• David Poliakoff (GOTCHA support)

Citing Caliper

To reference Caliper in a publication, please cite the following paper:

• David Boehme, Todd Gamblin, David Beckingsale, Peer-Timo Bremer, Alfredo Gimenez, Matthew LeGendre, Olga Pearce, and Martin Schulz. Caliper: Performance Introspection for HPC Software Stacks. In Supercomputing 2016 (SC16), Salt Lake City, Utah, November 13-18, 2016. LLNL-CONF-699263.

Release

Caliper is released under a BSD 3-clause license. See LICENSE for details.

LLNL-CODE-678900