The stats_miss program is used to generate some commonly used performance metrics from the concise and detailed statistics output files produced by RSIM, averaging across all processors or caches. The statistics displayed include the total number of read misses, the average absolute read miss latency from the point of address generation, the average read miss latency overlapped by the ILP processor between the point of address generation and graduation, and statistics about the number of references to various levels of the cache hierarchy. The output from stats_miss also includes information about the MSHR (miss status holding register) utilization at the L1 and L2 caches, as well as the average absolute latency beyond the L2 cache seen by read misses in the system.
Multiple simulation runs can be specified with a single invocation of stats_miss; each run is processed separately. The stats_miss program takes as input the the simulation output file names (without the _stat or _err suffixes), as well as the phase of interest for the program. The processor phase is the same as the phase used in the stats utility. However, the stats_miss utility also introduces the concept of the cache phase. The cache phase is the number of sets of statistics generated by StatReportAll before the specific call to StatReportAll that is of interest. Thus, the first call to StatReportAll leads to cache phase 0, and so on.
The syntax for invoking stats_miss to process the files associated with two different runs of an application (here app1_stat and app1_err, along with app1opt_stat and app1opt_err) is:
stats_miss -cache 1 app1 app1opt 2
where the cache phase of interest is assumed to be 1 and the processor phase of interest is assumed to be 2. If the cache phase and processor phase are the same for the files specified, the cache phase does not need to be specified.