Linear solves are consistently slower and more memory-heavy?

[mkasiulis]

Hi,

I've been trying to use LazyArrays for large linear solves, but it seems that even examples from the readme are both slower and more memory-heavy than just base solve for explicit arrays.

For instance,

A = randn(5,5); b = randn(5); c = similar(b);
@btime c .= @~ A \ b
@btime c .= applied(\, A, b)
@btime c .= Ldiv(A, b)
@btime c .= A \ b

returns, on one example run:

3.877 μs (33 allocations: 1.70 KiB)
3.851 μs (32 allocations: 1.67 KiB)
1.280 μs (11 allocations: 832 bytes)
388.098 ns (4 allocations: 464 bytes)

I am not surprised that it takes longer to use lazy arrays, but is there any way to decrease memory use here?

[dlfivefifty]

I guess you are saying the first two should do the same thing as the 3rd? Probably just missing special cases