This test is definitely imperfect (should have no erroneous failures,
only erroneous passes), but should serve as a good smoke test
if matrix alignment is ever broken again. Looking at the generated LLVM,
there is a bunch of weird choices for alignment chosen that might be
worth looking into.
It's also worth noting that the failure mode of this test is a #GP
exception, which I don't know how well the test runner handles in a
larger test corpus.
