Advanced Caching topics.  

Book lists 11:

1) Small and Simple to reduce hit time

2) Way prediction to reduce hit time

3) Trace Caches

4) Pipeline Cache

5) Nonblocking caches

6) Multibanked caches

7) Critical word first, early restart

8) Merging write buffer

9) Compiler optimizations to reduce Miss Rate

10) Hardware Prefetching to reduce miss rate/penalty
	
11) Compiler-controlled prefetching to reduce miss rate/penalty

Other cool cache things that aren't mentioned:

Skewed Associative Caches:

Pseudo-Associative Caches:

Victim Caches:

Multi-ported caches:

Variable size blocks:

Multiple-level caches:

Summary:
	Average Access Time = hit time + miss rate * miss penalty.
Remember, though, this can be difficult to calculate in pipelined, OOO machine.
What is miss penalty?  Intuition says it is lost cycles due to waiting for
memory.  What if instructions can bypass blocked load?  What if I have a
prefetching system, so that the results are on their way?  

	Reduce miss rate by
		larger blocks
		Higher associativity
		Bigger Caches
		pseudo-associative caches
		compiler optimizations
		victim caches (depending ...)

	Reduce miss penalty by
		Giving priority to reads over writes
		Merging Write Buffer
		Early restart and critical word first
		Multi-level caches
		sub-block replacement
		Victim caches (depending ...)

	Reducing both
		non-blocking caches
		hardware prefetching (?)
		compiler controlled prefetching (?)

	Reduce Hit time by 
		using small, simple caches
		Avoiding address translation during indexing
		pipelining writes for fast write hits
		Subblock replacement for fast write hits in write through cache
		Trace Cache
		Pipelined Cache Access (book says slow hits!)