Hi, I look at Vtune logs and documents. L1 bound in top-down metrics is defined and calculated as follows
L1_Bound = 100 * max(0, (EXE_ACTIVITY.BOUND_ON_LOADS - MEMORY_ACTIVITY.STALLS_L1D_MISS) / CPU_CLK_UNHALTED.THREAD)
Are EXE_ACTIVITY.BOUND_ON_LOADS, and MEMORY_ACTIVITY.STALLS_L1D_MISS counted in slots not cycles? If yes, shouldn't we divide it by the total number of slots?
The problem is that when I get these event counters, I can get these
EXE_ACTIVITY.BOUND_ON_LOADS 69010262370.0
MEMORY_ACTIVITY.STALLS_L1D_MISS 64751552228.0
CPU_CLK_UNHALTED.THREAD 200677632402.0
So L1_Bound = 2.2% but Vtune reports 21%?
Is there a factor of 10 missing?!!
連結已複製
0 回應
