HP Caliper Advisor Rule Writer Guide

performance data for IA64_INST_RETIRED and another dataset with data for
L1D_READ_MISSES.ALL. However, in this example, there isn’t a dataset that
contains both performance metrics.
Histogram Accessor Function: get_histogram
The syntax is:
get_histogram(histogram_type, metric_list)
get_histogram_nth(histogram_type, metric_list, n)
These accessor functions are used to retrieve histogram (flat profile) performance
metrics.
The data returned by these functions is more complex than that of the get_metrics()
single-valued functions. Because there is a great variety of sampling specifications
possible for every histogram, a rule cannot know what to request. Thus, a different
function is necessary to retrieve histogram data.
The first argument of these functions, histogram_type, is not part of the primary
argument list of metrics to retrieve. Instead, it specifies the histogram type to access.
The first argument can be one of the following:
ALAT (ALAT misses)
DCACHE (data cache misses)
DTLB (data translation lookaside buffer misses)
ICACHE (instruction cache misses)
IP (instruction pointer)
ITLB (instruction translation lookaside buffer misses)
Any histogram of the requested type will match, no matter what its sampling parameters
are. If no histogram of this type exists in any of the datasets, then the None value is
returned.
Three types of metrics are available from these accessor functions, each with its own
syntax:
• Histogram results
• Sampling parameters
• Dataset ID
Histogram Results
This represents a subset of the performance data collected by a histogram measurement
run. Note that in order to preserve the one-to-one correspondence between the metric
list and the returned metric values, this metric is returned as a tuple of tuples.
There are several ways to specify what type of data and how much data to retrieve:
42 HP Caliper Advisor Accessor Functions