Collaborative caching allows software to use hints to influence
cache management in hardware. Recent studies have shown that such
collaboration can achieve optimal cache replacement, at least in theory.

Pacman is a practical solution for collaborative caching in loop-based
code. Pacman uses profiling to analyze the patterns of optimal cache
management in order to determine which data to cache and at what time.
It then splits each loop into different parts at compile time so the
loop boundary can be adjusted at run time to selectively store active
program data in the available cache space. Pacman requires a single bit
at the load and store instructions. Some of the current hardware has
partial support. We will present results using both simulated and real
systems.