Home

Byfl documentation

Introductory material

• README file
• Installation instructions
• Usage instructions

Usage examples

• NuT: Monte Carlo neutrino-transport code
  • Description of Byfl instrumentation
  • Spreadsheet of Byfl measurement output
  • Source code
• SNAP: SN (discrete ordinates) application proxy
  • Description of Byfl instrumentation
  • Spreadsheet of Byfl measurement output
  • Source code

Compiler wrapper scripts

These all contain pretty much the same information. Take your pick.

• bf-clang wraps the Clang C compiler.
• bf-clang++ wraps the Clang C++ compiler.
• bf-flang wraps the Flang Fortran compiler.

In addition to the above, Byfl also provides more manual control of instrumentation via the following script:

• bf-inst instruments LLVM bitcode produced by any compiler.

The idea is that you generate LLVM bitcode (e.g., the -emit-llvm option in Clang), pass it to bf-inst, and bf-inst instruments the code using Byfl, and writes a new LLVM bitcode file.

Output postprocessors

Byfl programs generate a binary .byfl file as output. The following programs convert .byfl files to various other formats:

• bfbin2xmlss converts a .byfl file to a formatted XML Spreadsheet spreadsheet that can be imported into LibreOffice Calc, Microsoft Excel 2003+, and Numbers for Mac.
• bfbin2csv outputs a .byfl file in comma-separated value format for ease of parsing
• bfbin2sqlite3 converts a .byfl file to a SQLite 3 database. SQLite 3 databases are standalone files that can be used as is or imported into a heavyweight database management system.
• bfbin2hdf5 recodes a .byfl file in HDF5 format, a format often used for scientific data.
• bfbin2hpctk generates an HPCToolkit database from a .byfl file's Functions table. These databases can be loaded into the hpcviewer GUI and compared side-by-side with non-Byfl-produced HPCToolkit databases.
• bfbin2cgrind generates a Callgrind file from a .byfl file's Functions table. Callgrind files can be loaded into the KCachegrind GUI for visualization and analysis.

All of the above are implemented using a callback-based API that Byfl provides. The API includes a single C function that parses a .byfl file and calls user-defined functions for each table, column header, and data value it encounters. You can use the API to write your own .byfl output postprocessors:

• bf_process_byfl_file

SWIG wrappers are installed if SWIG is available. These let you write .byfl postprocessors in a variety of scripting languages. The SWIG version of the Byfl API is not based on callbacks because SWIG doesn't currently support callback functions. The following example shows how to process a .byfl file from Python:

• bfbin2py installation instructions and Python script

Publications

For now, this is the best paper to cite in research publications that refer to Byfl:

• Scott Pakin and Patrick McCormick, "Hardware-independent application characterization". 2013 IEEE International Symposium on Workload Characterization (IISWC), Portland, Oregon, USA, 22-24 Sept. 2013, pp. 111–112. ISBN: 978-1-4799-0553-9, DOI: 10.1109/IISWC.2013.6704676.

More Byfl-related publications can be found on the Byfl publications page.

Current build status

Byfl uses Travis CI for a third-party check that the code builds cleanly. The following is Byfl's current build status:

Byfl branch	Build status
master