Use alternative CS decomposition when it is missing from LAPACK

[jmbr]

No description provided.

[notmgsk]

; Symbol "HERMITIAN-EIG" not found in the MAGICL package.

[jmbr]

; Symbol "HERMITIAN-EIG" not found in the MAGICL package.

That's only available in MAGICL's master or, preferably, in quil-lang/magicl#65

[notmgsk]

; Symbol "HERMITIAN-EIG" not found in the MAGICL package.

That's only available in MAGICL's master or, preferably, in rigetti/magicl#65

Ah. Then maybe we ought to make a MAGICL release so that I will be picked up here.

[jmbr]

Ah. Then maybe we ought to make a MAGICL release so that I will be picked up here.

Yeah, that's reasonable.

[appleby]

Would it make sense for the csd-compat implementation to live in the magicl package, and have magicl fallback to it if zuncsd isn't available? Or maybe just provide it separately? I only skimmed the diff, but IIUC it looks like csd.lisp only depends on m* from cl-quil, which might also be reasonably moved into magicl.

[jmbr]

The reason this code presently lives in cl-quil is because it implements a specific case of the CSD (currently the one used in Quantum Shannon compilation). If it ever becomes a general-purpose CSD implementation, then I agree we might consider moving it to magicl.

[jmbr]

Passes the test suite. Closes #292.

[appleby]

Should we bump the dependency in cl-quil.asd to magicl 0.6.5?

I guess we're probably waiting on the quicklisp release

[jmbr]

That's right.

…

On Mon, Nov 25, 2019, 08:15 appleby ***@***.***> wrote: Should we bump the dependency in cl-quil.asd to magicl 0.6.5? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#485>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAADE3K4Y6X5WSBQBDWIQ6TQVP2Z5ANCNFSM4JO4FUBQ> .

[appleby]

Passes the test suite. Closes #292.

By "passes the test suite" do you mean you installed a version of LAPACK without ZUNCSD and ran the test suite, or just that the new csd tests you added + rest of the test suite passes when run against an LAPACK that includes ZUNCSD.

[jmbr]

Passes the test suite. Closes #292.

By "passes the test suite" do you mean you installed a version of LAPACK without ZUNCSD and ran the test suite, or just that the new csd tests you added + rest of the test suite passes when run against an LAPACK that includes ZUNCSD.

The former: I used Apple's Accelerate framework (which does not include ZUNCSD) to run the entire test suite (I also double-checked that the new routines were picked up by traceing them).