TODO: Get code coverage to 95%

[DilumAluthge]

For reference, here is the current code coverage:

[DilumAluthge]

This is done now.

[DilumAluthge]

Our coverage has fallen below 100%, so I'm reopening this.

[chriselrod]

There are a lot of branches based on the size of the arrays. Ideally, we'd be able to enable coverage for our tests over different array sizes.

For the 20x20 case that is tested with coverage enabled, the matrices are small enough to take the fast-path, and thus skip almost all the code.

[DilumAluthge]

I'm going to change the goal to 95% since I think that there are some lines that we'll never be able to cover.

[DilumAluthge]

@chriselrod We're at 40% now. Take a look at Codecov - do you think that we need to add more matrix sizes (and/or bigger matrix sizes) to test/matmul_coverage.jl? Or do we need to just start writing unit tests to cover the uncovered lines?

[chriselrod]

Two things that should be simple and help with this file:

1. Add a test dependency on something like StaticArrays.jl (or StrideArrays after it is registered). Testing with mutable statically sized arrays (meaning StaticArrays.MArray) would let us hit inlineloopmul!. It requires small statically sized arrays to be called. It should work out of the box with with StaticArrays, so that's worth testing to confirm itself.
2. Bigger matrices. In that file, you can see that none of the matrices were big enough to pack both A and B, so related code (i.e., matmul_st_pack_A_and_B!, matmul_pack_A_and_B!, and sync_mul!) weren't tested.

[DilumAluthge]

Bigger matrices. In that file, you can see that none of the matrices were big enough to pack both A and B, so related code (i.e., matmul_st_pack_A_and_B!, matmul_pack_A_and_B!, and sync_mul!) weren't tested.

#49