elaps'd

elaps'd (say: elapsed) is a tool for analyzing the run-time of your application. It currently supports multi-threaded applications written in C, C++ or FORTRAN. But wait, there is more: it also provides a web-based for visual analysis of the recorded data:

Currently, one has to manually utilize elaps'd in the application. However, the plan is to automate this in future, so the user only has to label functions to be measured.

TL;DR Usage Summary

Definitions: Kernel -> The function you want to measure; Device -> The device the function runs on (e.g. CPU, GPU, CPU + GPU, ...).

1. Build with make
2. Have a look at the examples
3. Incorporate elaps'd into your application and generate some DB's
4. Use the analyzer by issuing python analyzer.py and opening http://localhost:8080 in your browser

Large Developer's Request: Please report Bugs/Improvement Wishes at GitHub.

Requirements

• For the backend, i.e. the measuring part

  • C++ compiler with C++03 support
  • SQLite (>= 3.6)
  • OpenMP
  • A C or FORTRAN compiler if you want to build these APIs

• For the frontend, i.e. the visualization part

  • Python
  • apsw for your version of SQLite
  • Flask
  • HTML5 Browser (best: Google Chrome or Chromium)

Build

1. Open Makefile.include in your favorite editor and adapt settings as necessary. No changes are needed most of the time.
2. Type make in the root of elaps'd directory. This builds the library libelapsd.so inside lib and the examples inside examples.
3. If you want to install it, copy headers and library (sorry about the manual step! I'll change this in near future).

Usage

Create Measurements (Backend)

This is a short summary on how to use the backend. Please have a look at the provided examples for more details.

1. Include elapsd/elapsd.h for C++, elapsd/celapsd.h for C. For FORTRAN, use elapsd/m_elapsd.mod. All these are in the include directory.
2. Initialize elaps'd. C++ users simply do e.g. ENHANCE::elapsd e("elapsd.db", "Experiment Name"); (watch out for the namespace!). C users call elapsdInit, so do FORTRAN users.
3. Use addKernel and addDevice function calls to add kernels and devices.
4. Use startTimer and stopTimer to start and stop measurements. Please consider that startTimer must be called prior stopTimer and that to every startTimer a stopTimer belongs.
5. Once you finished all measurements call std::cout << e << "\n" for summarized output and/or the commitToDB() function to write the data to SQLite for further analysis.

Use the Analyzer (Frontend)

1. Copy the generated DB to static/db which is inside the analyzer directory. This step will be relaxed in near future letting you configure the path to the DB's.
2. Start the analyzer's integrated Flask-based webserver by calling python analyzer.py.
3. Open your favorite browser (I recommend Google Chrome or Chromium for best user experience) and point it to http://localhost:8080
4. Select the DB you want to work with and start analyzing!

Changelog

2013-09-03 -- v0.3

• Added prediction library with Lagrange model
• Extended analyzer to visualize predictions

2013-07-30 -- v0.2

• Added ability to store additional parameters for an experiment

2013-05-31 -- v0.1

• Initial release

Author information

• Author: Sebastian Dreßler
• Company: Zuse Institute Berlin
• Mail: [lastname][at][zib][dot][de]

elaps'd is developed as part of the ENHANCE project (for further info visit http://www.enhance-project.de) which is funded by German ministry for education and reserach (BMBF). BMBF grant no.: 01IH11004A-G

License

elaps'd an all self-written parts of it is licensed under the 3-clause BSD license. The full license text is provided in the file LICENSE.

Disclaimer

This software is currently in a very early stadium. There is no guarantee that measurements are consistently correct and that there are no major bugs.