The proposed design environment will be built on open DARPA-funded standards and interfaces that will ensure support of diverse adaptive computing architectures as well as compatibility with existing compiler efforts and upgradability to new compilation and implementation techniques that may be developed on future programs. Specific innovations are:
Adaptive-computing specific compilation
Development of a parallelizing compiler that analyzes the user application
to identify opportunities for computational speed-up such as configuration
of the adaptive computer to support execution of operations that can be
performed in parallel. The compiler will automate several optimizations
performed manually on current ACS design projects, saving several man-years
of design effort per project in the future. These will include the
difficult problem of determining hardware configurations that can be
re-used in different parts of the same application.
Development of a module generation library and assembler tool that
can eliminate the need for traditional logic synthesis and place
and route tools for the configurable units in the adaptive computer
thereby achieving orders of magnitude reduction in compilation time.
Retargetability to multiple application descriptions and ACS
architectures
The compiler will support multiple ACS architectures by guiding the code
optimizations with knowledge of the target architecture specifications as
well as with estimations of the cost of different computational operations
on the target architecture. An open design environment will be developed
that uses DARPA-funded representations (SUIF, MPI) to link the compiler
to multiple high-level languages (including C and MATLAB) and visual design
entry systems as well as to multiple back-end implementation tools for the
configurable computing units (e.g. FPGAs), general-purpose processors and
micro-controllers in adaptive computers.
Adaptive computing design environment demonstration
The proposed design environment will be validated on ACS defense challenge
problems through the on-going SLAAC program. Demonstrate measurable benefits
of compiler in achieving higher performance and improved reliability for
COTS adaptive computers over micro-processor based alternatives.