The MPI Tools
Interface (MPI_T) provides an opportunity for performance tools to
introspect MPI runtimes and understand their behavior, and also a
mechanism to reconfigure them at runtime. We propose here a plugin
infrastructure for TAU that allows full customization of MPI_T
autotuning and recommendation generation at runtime. Through the TAU
Plugin API, the user can change or toggle the list of plugins to invoke
for specific phases or functions wherein MPI_T autotuning yields highest
benefits. Doing so allows more fine-grained control over the general
case where tuning is phase-agnostic.