Profiling and Timing
Gaining insight into what the hardware is doing can be priceless. We like to think of profiling, or performance monitoring tools, as flashlights in an otherwise dark interior of a computer system. It feels as though the more complex the internals of computers get, the darker they get without flashlights.
When we think of systems with Intel® Xeon Phi™ coprocessors, we can think about wanting insights into the activities on each thread or core or coprocessor, and wanting to understand the communication traffic between MPI ranks.
For insight into the activities of a processor or coprocessor, Intel supports event-monitoring registers. On the coprocessor these are similar to some counters on a processor, but with additional ...