Hello, i use nsight compute to profile the kernel, and something very strange was found. Like below.
l2_tex_read_hit_num = 5129
l2_tex_write_hit_num = 3
l2_text_read_sum = 5130
l2_tex_write_sum = 3
According to my understanding, l2 read hit rate maybe l2_tex_read_hit_num / (l2_text_read_sum+l2_tex_write_sum) = 0.9992 = 99.92% is not equal to nsight compute metric l2_tex_read_hit_rate=99.86%. so what is the way the L2 read hit rate is calculated??
nsight compute version 2019.5.0 (Build 27346997) (public-release)
device: jetson xavier jetpack 4.4