We dene the resolution of the function to be the

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: seen Measurement 1 range from 317 and 606, and Measurement 5 range from 301 to 326. On the other hand, the other four measurements only vary by at most a few cycles from one run to another. Clearly Measurement 1 is an overestimate, because it includes the cost of loading the measurement code and data structures into cache. Furthermore, it is the most subject to wide variations. Measurement 5 includes the cost of loading procB into the cache. This is also subject to significant variations. In most real applications, the same code is executed repeatedly. As a result, the time to load the code into the instruction cache will be relatively insignificant. Our example measurements are somewhat artificial in that the effects of instruction cache misses were proportionally greater than what would occur in a real application. To measure the time required by a procedure P where the effects of instruction cache misses are minimized we can execute the following code: 1 2 3 4 5 6 7 double time_P_warm() { P(); /* Warm up the cache */ start_counter(); P(); ret...
View Full Document

This note was uploaded on 09/02/2010 for the course ELECTRICAL 360 taught by Professor Schultz during the Spring '10 term at BYU.

Ask a homework question - tutors are online