HP Caliper Advisor Rule Writer Guide

get_metrics_nth(‘CPU_CYCLES’, -2)
(pmu_event, plm) Similar to pmu_event above, except that the
privilege level mask is also given. When the plm
modifier is provided, the entire event description
must be enclosed in parentheses to form a Python
tuple.
Replace plm with USER, KERNEL, BOTH, or
ALL.
Examples:
get_metrics((‘CPU_CYCLES’, ‘USER’))
get_metrics_nth((‘CPU_CYCLES’, ‘USER’), 0)
(pmu_event, plm, threshold) Similar to pmu_event above, except that the
privilege level mask and the threshold are also
given. When the plm and threshold modifiers
are provided, the entire event description must
be enclosed in parentheses to form a Python
tuple.
Replace plm with USER, KERNEL, BOTH, or
ALL.
Replace threshold with an integer value.
Examples:
get_metrics((‘CPU_CYCLES’, ‘ALL’, 4))
get_metrics_nth((‘CPU_CYCLES’, ‘ALL’, 4), -1)
Multiplexed PMU Statistics
This is the type of performance metric available from the HP Caliper cpu measurement.
By multiplexing several measurement sets, HP Caliper can produce a greater breadth
of performance data in fewer runs of an application. Exact PMU counts cannot be
produced, but useful performance statistics are produced.
To retrieve these statistical metrics, use the following three-part metric name:
event_set.metric.statistic
Returns the requested statistic, which is an
integer or floating-point value, depending on the
specific statistic.
Replace event_set with BRPATH, BRPRED,
C2C, CPI, CPUBUS, CSPEC, DISPERSAL,
DSPEC, FP, L1DCACHE, L1ICACHE, L2CACHE,
38 HP Caliper Advisor Accessor Functions