cuda profiler: "Instructions" point

Hello. I have some questions about point “instruction” of signal plot.
my kernel gives this picture:

it takes 672 bytes lmem, 6172 bytes smem, 1242688 bytes cmem, 46 registers. In “README” i have found only “instructions : Number of dynamic instructions (in fetch)”.

  1. What does “Number of dynamic instructions (in fetch)” mean? i want some more detailes. Is it just a total number of all instructions in kernel?

  2. What cause “instruction” number groth and what can i do with my kernel to decrease this number?(in general)
    thanks.

up