Imperfect factors increase totol computes counts

[tuvaluhk]

Hi,@angshuman-parashar @tanner-andrulis
I‘m using spatial imperfect factors like N=6,2 to simulate a 3-level arch of 1 mainmemory, 6 buffer instances, and 6 macc unit as below.
arch.yaml
mapping.yaml
problem.yaml

I'm expecting the mapping would split the dim-N into 8 pieces and computes 6 pieces at first round and then compute the next 2 pieces.
However, the sim result shows that total computes is 768=8*12*8. I wonder if there happens a padding in second round compute.
timeloop-model.stats.txt
In a mean time, the result also shows a reasonable scalar reads and updates of the tensor.
I can't find too much explanation of this feature, so I open a issue here. Is anyone familiar to this, any help would be appreciate.