From cTuning.org

Jump to: navigation, search

NEWS: Since 2015, we moved all related developments, benchmarks, data sets and tools to our new Collective Knowledge Framework!


logo_unidapt_f.gif

Universal Adaptation Framework

Statically enabling run-time optimization and adaptation
Web shortcut: http://cTuning.org/unidapt
Navigation: cTuning.org > CTools 

NOTES:

cTuning Google discussions list

UNIDAPT concept has been developed during 2004-2006 by Grigori Fursin to statically enable run-time optimizations and self-tuning binaries through cloning of program hot spots, applying various aggressive optimizations to clones for different optimization cases (that may improve performance/power/fault-tolerance, etc), statically integrating low-overhead program/system behaviour monitoring routines (using hardware counters) and selecting appropriate versions at run-time as a reaction to different program behavior, architectural changes or contentions.

For the first time, Grigori utilized his Interactive Compilation Interface for PathScale compiler with loop vectorization, tiling, unrolling, interchange, fission/fusion, pipelining, prefetching and array padding to make static self-tuning binaries that can automatically learn from the past experience and adapt/react to various environments, run-time behavior and contentions that is important to improve efficiency and cost of both embedded systems and HPC data centers (cloud computing).

This technique opened up many research possibilities, has been used in multiple research projects in collaboration with UPC, ICT, IBM, CAPS Enterprise, STMicro, has been supported by MILEPOST, HiPEAC and Google Summer of Code grants, has been referenced in patents and has been extended to speed up iterative compilation (FCOP2005, FCOP2006), enable transparent continuous collective optimization (FT2010, FT2009,FMPP2007), enable portable program characterization techniques based on reactions to optimizations (FT2010, FT2009), enable predictive scheduling for heterogeneous multicore systems (JGVP2009), enable adaptive libraries based on dataset characterization using machine learning and decision trees (LCWP2009) among many other usages based on continuous transparent run-time program optimization and adaptation as a reaction to dynamic changes in program behavior and environment. Since 2007 it is being actively extended by Google Inc. for data centers.

We are gradually working to move this framework to mainline GCC combined with ICI (though source-to-source adaptation framework can still be useful). We are working to provide a unified view of heterogeneous architectures and optimizations with a high-level abstraction layer (architectures, compilers, run-time systems) to automate and simplify program development and optimization for heterogeneous multi-core systems. We would like to use this framework to automatically detect contentions in computing systems and react to such changes. We also hope to provide a unified view of heterogeneous architectures (CPU/GPU, CELL-like, FPGA, accelerators), optimizations and data movement/partitioning with a high-level abstraction layer (architectures, compilers, run-time systems) to automate and simplify program development and optimization for heterogeneous multi-core systems.

You are welcome to join the project, provide feedback and help with developments.

unidapt-framework.gif figure_unidapt.gif figure_unidapt_hetero.gif

You are welcome to join us and participate in discussions, developments or provide feedback and suggestions to extend UNIDAPT Framework.

Locations of visitors to this page

Tweet