The KNL processors offers unique features concerning memory hierarchy and vectorization capabilities. To improve tool support within these two areas, we present extensions to the Score-P measurement system and the Cube report explorer.
KNL introduced a new memory architecture, utilizing MCDRAM and DDR. To help the user in the decision where to place data structures, we record a MCDRAM candidate metric. In addition we track all MCDRAM allocations through the hbwmalloc API, providing memory metrics like leaked memory or the high-watermark on a per-region basis. For time-line analysis per-process memory statistics are recorded via numastat.
KNL's large vector processing unit needs to be utilized and utilized effectively. The metrics compute-to-data access ratio and VPU intensity are introduced to identify vectorization candidates on a per-region basis.
Taking the hardware structure into account, the distribution of the KNL-specific metrics is visualized in the Cube report explorer.