Basically there are two variants of unused reuse opportunities that the Freja tools detect: reuse within a loop, and reuse between different loops. The former is addressed by applying a technique called blocking or tiling, and the latter is handled by bringing the two loops closer together or actually merging the loop bodies completely.

Reuse come in two distinct flavors: spatial reuse, where the benefit comes from that subsequent data is already fetched into the cache, and temporal reuse, where the very same data is revisited multiple times. Minimizing the time before the revisit will lower miss ratios.

Addressing these issues will usually cause loop hierarchies to be partially turned around and broken up. Loop fusion will also necessarily change the features of the program. This changed scene should be re-analyzed for unnecessary memory accesses, poor layout and poor access patterns.

	Tip
	Bring related loops closer together, and optimally merge related parts of their loop bodies.

	Tip
	Look for spatial reuse, and change loop nesting or block with respect to the data structure that displays long reuse distance.

	Tip
	Look for temporal reuse, and change algorithms to perform as many iterations as possible for a given data chunk before moving along.

	Tip
	After merging loop bodies or transforming loop structures, rerun analysis to find advice for the new program structure.