hoangt / liftracc Goto Github PK

liftracc is a library framework to create accelerated replacements of existig
libraries. The new library can be used instead of the old, unaccelerated
version. One technique to use such libraries is 'shared library interposing'.
Through a flexible plugin system it is possible to integrate existing versions
of accelerated sublibraries. Also it is possible to write the needed
programmcode directly inside a plugin to control the hardware accelerator
(e.g. grafik card, accelerator board).

The framework provides mechanisms to dynamically select between different
implementations of one function (provided through the plugins). Thus it is
possible to use different accelerators on different problem sizes. Research
around the liftracc project has shown that you can get certain speedup from
such dynamic selection.

In this repository, as an example, the framework is used to build an
accelerated version of the blas (basic linear algebra subprogramms) interface.
A function collection of mathematical vector an matrix functions.

For further information on how to install, use and extend the framework have a
look at 'liftracc/library/documentation/html/index.html' after cloning.

directory structure:
  liftracc/benchmarks/
   - linpack benchmark in order to test the library
  liftracc/blas_libs/
   - bash scripts for easy sublibrary installation
  liftracc/common/
   - some python scripts
   - cmake modules
   - xml files
  liftracc/library/
   - liftracc source code
   - auto generation scripts
   - profiling scripts
  liftracc/library/documentation/html/
   - library documentation
  liftracc/testapps/
   - some test applications