SPECCPU 2006 Characterization Results
This pages present the SPECCPU 2006 characterization results. The data were originally obtained by the authors.
In most of the experiments, we use sampling to collect the data. So the data collected may not be absolutely precise. However, the inaccuracy should be trivial in most cases.
The authors give out theses data in the hope that they may be useful to someone. But they should not be held responsible for any damage that might be caused by these data.
Auto-parallization effect of the ICC compiler
Shows how does ICC's auto-parallelization perform.
The effect of shared cache on performance
How does the shared cache of Intel Core 2 Duo effect the performance of benchmarks.
Demonstrates how much time each function in each bechmark takes.
Demonstrates the instruction-level characteristics, including FP/INT/branch/load/store functions... etc.
Shows where the CPU cycles are spent, e.g. different pipeline stalls.
Prefetch effect of ICC and Open64
How many prefetch instructions are generated, how much cache miss latency is hidden.
Per-function characterization, including instruction breakup, cache miss behavior, and branch behavior.