Row-wise and column-wise scaling of a sparse matrix runs out of memory

[pitsianis]

As the subject says, when I try to scale the values of a large sparse matrix using .*, a dense matrix is built, and OutOfMemoryError() is raised!

For instance,

n = 10_000_000
v = rand(n)
A = sprand(n,n,2/n)
v .* A  # ERROR: OutOfMemoryError()
A .* v' # ERROR: OutOfMemoryError()

Using Diagonal or spdiagm will do the computation,

julia> @btime $A * Diagonal($v);
  97.998 ms (6 allocations: 381.50 MiB)

julia> @btime Diagonal($v) * $A;
  134.449 ms (6 allocations: 381.50 MiB)

julia> @btime spdiagm($v) * $A;
  1.794 s (31 allocations: 1.31 GiB)

julia> @btime $A * spdiagm($v);
  522.747 ms (31 allocations: 1.31 GiB)

where spdiagm(v) forms a sparse diagonal matrix at a great expense in time and space.

See discussion here

[pitsianis]

This is related to a previous issue