In the histogram mode, the analysis sample post-processes the measurement results. This means that the analysis sample runs the instrumented application, waits until it is finished, loads the collected data from the dump file, and then runs the analysis and shows the histogram.
The dump file is created in the current directory, so it needs to have write permissions. The dump file can contain the measurement results for all iterations of the instrumented workload or for a subset of iterations, depending on the number of iterations and the buffer size. If the buffer size is smaller than the number of iterations, the histogram will only show a subset of all measurements.
Understanding the Histogram
The histogram represents the distribution of measured latencies through multiple bins ranging from minimum to maximum execution time. By looking at the histogram, you can see the relative frequency of measured latencies that fall into different latency ranges, and as a result decide the optimization methods to reduce the maximum latency and improve the determinism of the workload execution. As an example, for short workloads, latency outliers far outside the main peak of the distribution often mean processing of interrupts that interferes with the main flow of workload execution.