Recently, I am analyzing a performance issue, in which CU has collected lots of PMC events, and i found something strange.
In one physical core,
its DTLB_LOAD_MISSES.WALK_DURATION is 45054431.81;
its DTLB_STORE_MISSES.WALK_DURATION is 25549.30;
its CPU_CLK_UNHALTED.THREAD_P is 11567496.63;
with these 3 numbers, I don't know how to get the ratio of DTLB miss percentage.
Could you please teach me how to use these numbers?
For more complete information about compiler optimizations, see our Optimization Notice.