README

El Dorado Hills, CA, January 2008

OpenKL is a dynamic run-time for parallel constraint solvers. 

Writing parallel constrains solvers is hard enough, but tuning them for specific microarchitectures from Intel, AMD, IBM, NVIDIA, ARM, or TI is down right impossible.

Each micro-architecture creates unique computational trade-offs as all machines exhibit different data flow constaints. It is not uncommon for a solver to run at only 5% of the peak performance of a machine.

The goal of OpenKL is to extend the high level linear algebra framework defined by MTL4, which is a standard C++ library for linear algebra, and add the dynamic algorithm selection to adjust the execution to the best possible performance given the underlying hardware.