github
Advanced Search
  • Home
  • Pricing and Signup
  • Explore GitHub
  • Blog
  • Login

snb / labench

  • Admin
  • Watch Unwatch
  • Fork
  • Your Fork
  • Pull Request
  • Download Source
    • 1
    • 1
  • Source
  • Commits
  • Network (1)
  • Issues (0)
  • Downloads (0)
  • Wiki (2)
  • Graphs
  • Branch: master

click here to add a description

click here to add a homepage

  • Branches (1)
    • master ✓
  • Tags (0)
Sending Request…
Enable Donations

Pledgie Donations

Once activated, we'll place the following badge in your repository's detail box:
Pledgie_example
This service is courtesy of Pledgie.

Tool to measure performance of linear algebra libraries, such as BLAS and LAPACK — Read more

  cancel

  cancel
  • Private
  • Read-Only
  • HTTP Read-Only

This URL has Read+Write access

Print out times once again 
snb (author)
Thu Sep 10 05:28:50 -0700 2009
commit  37f04547142027eeba58bfdf8161da9c0a6030fd
tree    6fdff83122cde51f098fc46b761df267b4092ac6
parent  c8c7ec9eaf3f75cf50b60848b3726a18f187d465
labench /
name age
history
message
file .gitignore Tue Sep 08 03:58:39 -0700 2009 More compiled files we can ignore [snb]
file Makefile Loading commit data...
file README.md Wed Sep 09 09:33:55 -0700 2009 More clarification about building [snb]
file TODO Thu Sep 10 05:23:31 -0700 2009 Matrix functions now return timing_duration str... [snb]
file defs.h
file main.c Thu Sep 10 05:28:50 -0700 2009 Print out times once again [snb]
file matrix.c Thu Sep 10 05:23:31 -0700 2009 Matrix functions now return timing_duration str... [snb]
file matrix.h
file timing.c Thu Sep 10 05:22:56 -0700 2009 Unused include; fixed some typos [snb]
file timing.h
README.md

This is labench, an oh-so-catchy name for linear algebra benchmarking. To build the tests, just run make, or gmake on systems where the make command is not GNU make. You will probably have to modify the Makefile to choose which libraries you want to test. After compiling, you will end up with a number of binaries for testing double and single precision versions of the test routines using each of the library versions specified in the Makefile. So far I have tested labench only on Mac OS X 10.6 and Ubuntu Linux 9.04. It might work on other systems, but no promises.

This project was started because I was interested in comparing the performance of different implementations of common linear algebra libraries such as BLAS and LAPACK. Right now it is possible to measure the CPU and wall clock time of multiplying dense general matrices together with the sgemm and dgemm routines from level 3 BLAS, and the speed of finding eigenvalues and eigenvectors of a dense general matrix using sgeev and dgeev LAPACK routines.

Another tool for benchmarking numerical computing performance is HPCC, the High Performance Computing Challenge. HPCC even includes a dgemm test. My goal with labench was to focus specifically on making it easy to compare performance of different implementations of the same libraries, so one can decide which library is the best choice to use on a given machine. I also hope to make it very easy to extend labench to support performance testing of other BLAS and LAPACK functions, and possibly also other numerical libraries.

labench is fairly basic right now, but take a look at the TODO file to see what sort of things are planned for its future.

-- Nick Barkas snb@FreeBSD.org

Blog | Support | Training | Contact | API | Status | Twitter | Help | Security
© 2010 GitHub Inc. All rights reserved. | Terms of Service | Privacy Policy
Powered by the Dedicated Servers and
Cloud Computing of Rackspace Hosting®
Dedicated Server