Matt Fishman

Looks like there is some package compatibility issue in the tests.

One suggestion I would have to make this first PR simpler to review would be to just support an SVD implementation that transfers to CPU, and then add in other...

> > One suggestion I would have to make this first PR simpler to review would be to just support an SVD implementation that transfers to CPU, and then add...

> > Yes, I understand that. I just don't like that interface. How would you choose the CPU SVD backend? Would we need to define a new `"*_to_cpu_svd"` for every...

Sorry, I didn't see you referring to `default_svd_alg(T)`. I forget how that is handled for `Metal` but it could just be handled the same way, presumably it chooses the default...

> As far as I can tell, the `RUN NDTensors tests` CI check is failing because latest version of `AMDGPU.jl` that supports Julia 1.6 is 0.2.17. > > > ─AMDGPU...

Looks good! Thanks @wbernoudy and @kmp5VT.

@ogauthe @emstoudenmire I'm merging this. The status is that you should be able to perform arbitrary broadcasting operations (adding, scalar multiplication, permutations) of block sparse arrays, as well as perform...

Thanks @shinaoka, this will be useful to have. I would instead vote for generalizing the function signature of: ```julia function contract(::Algorithm"densitymatrix", A::MPO, ψ::MPS; kwargs...) # [...] end ``` to: ```julia...

Thanks for the report. Very likely this is due to differences in the performance of BLAS and LAPACK on those two systems, I would recommend comparing BLAS/LAPACK functionality like matrix...