Visualizing the Impact of the Cache on Program Execution
Cache behavior of a program has an ever-growing strong
impact on its execution time. Characterizing the behavior by visible patterns
is considered a way to pinpoint the bottleneck against performance.
We present a framework of visualization for trace distributions to extract the
useful cache behavior patterns. We focus on cache misses, reuse distances,
temporal or spatial localities, etc. The histograms of these distribution
patterns measure the behavior in quantity, revealing effective program
optimizations. The performance bottlenecks are exposed as hot spots highlighted
in the source code, showing the exact locations to apply suitable
optimizations. The impact of the source-level program optimizations, again, can
be verified by the visualization tool.
The work has been presented in the Information Visualization (IV'01) 2001 conference.
More details are published in the CCAI journal.