The Blackjack project is developing metrics for evaluating compilers for scientific computing.
The evaluation is performed by using representative applications
and implementing revelant micro-benchmarks,
as well as testing and analyzing the productivity, correctness, and performance
of multiple commercially available and freely available open source compiler systems. The main two goals are:
- to test whether compiler technology can automatically select the
appropriate optimizations based on a learned characterization of the
target system; and
- to ensure the compiler technology has a dynamic
runtime environment that can dynamically improve the performance of a
program during runtime and/or provide information that can be used by
the compiler to optimize for future runs of the program.
