This message reports how many times a kernel was executed on how many equivalent compute units.
One single kernel can be implemented with several compute units in the accelerator. As a result, multiple workloads can be executed on the different compute units concurrently. This message provides some insight into the available workload and the available compute units.
If multiple compute units are present in the device, it is recommended to look at the application timeline to examine compute unit utilization throughout a run with a typical workload. If data transfer is controlling the maximum throughput, not all compute units might be utilized. In this case, consider removing a compute unit and/or consider additional blocks to be implemented on the device.