Structural Select isn't very memory efficient on full, iso-valued objects

[eriknw]

I would expect some select operations to be more memory efficient then they currently are.

Here are a couple examples where I get OutOfMemory error:

• Large, full iso-valued Matrix as input and using a mask with few elements (i.e., whose indices comfortably fit in memory); for example, when selecting with TRIL.
• Large, full iso-valued Vector as input with select operator "ROWLE" with small thunk value (i.e., final indices comfortably fit in memory).

I stumbled upon the first one when writing an algorithm. The second one was simply exploration and not important to me.

[DrTimothyAldenDavis]

My GrB_select method currently does the select as T = select(A,...), which does not exploit the mask, followed by C=T. That's why you're seeing the out-of-memory error. T is very large and is explicitly constructed. I realize that's not optimal. I have this marked as a "TODO":

GraphBLAS/Source/GB_selector.c

Line 14 in 4bb44bb

// TODO: GB_selector does not exploit the mask.

It's on my radar to speed up GrB_select when a sparse mask is present.