[RutarH13] Software techniques for negating skid and approximating cache miss measurements

Modern hardware counters are used to find program instructions that cause most cache misses for example, and the way is to measure how many times a counter overflow happens on a particular instruction. However, when an overflow happens as an interrupt, the exact instruction causing the interrupt may be incorrect, a problem that Intel calls a “skid”.

The solution is to consider surrounding instructions as the set of probabilities. Then the overlap of these probabilities will show the most likely instruction.

The problem and solution are hardware dependent.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s