Slow sparse kronecker products

[ghost]

I have found that the kronecker products are much slower compared to matlab.
Is there something that can be done?

In Julia

ddx(n) = spdiagm((-ones(n), ones(n)),[0 1],n,n+1) 
D3 = ddx(128);
tic(); A = kron(kron(D3,speye(128)),speye(128)); toc()

elapsed time: 1.625884532 seconds

In matlab

ddx = @(n) spdiags([-ones(n,1), ones(n,1)],[0,1],n,n+1);
D3 = ddx(128);
tic; A = kron(kron(D3,speye(128)),speye(128)); toc

Elapsed time is 0.395843 seconds.

[ViralBShah]

There are some other bugs in kron too. I am fixing them now.

[ViralBShah]

The speedup comes from devectorizing one more inner loop, and allocating much fewer objects. For me the original time was 1.7 seconds, but now is faster than matlab:

julia> tic(); A = kron(kron(D3,speye(128)),speye(128)); toc()
elapsed time: 0.090981935 seconds
0.090981935

[ViralBShah]

BTW, there were no bugs in kron, but a bug in show that I fixed in an earlier commit.

[timholy]

Nice 20x speedup, Viral!

[StefanKarpinski]

Seriously, that's not shabby at all.

[ViralBShah]

This one turned out to be surprisingly easy. It is getting harder to find such "opportunities"!