Update version to v0.3.11, Fix INTERFACE64, gcc version

[mattip]

Fixes gh-42. There must be a better way to get the matrix values into the environment, but this should work.

[mattip]

Merging this will change the hashes for the already-uploaded packages. Should I go ahead and get the HEAD from OpenBLAS so we can check their fix for numpy/numpy#16774 ?

[mattip]

... and of course I have a typo.

[mattip]

gcc fails to compile 0.3.11 since the mingw version is 8.1.0 which suffers from this bug:

config/i386/avx512fintrin.h (_mm512_abs_pd, _mm512_mask_abs_pd): Use
__m512d type for __A argument rather than __m512

It doesn't appear that there are later versions of mingw available.

[mattip]

[skip travis]

[mattip]

Now getting

undefined reference to `dgetri_`

in 32-bit tests. When do we drop windows 32-bits?

[tylerjereddy]

When do we drop windows 32-bits?

My impression was that Python 32-bit interpreter installs on Windows were still very common for whatever reason. I think Matthew Brett may have been saying that?

[charris]

I think Matthew Brett may have been saying that?

That's also my recollection.

[mattip]

I was blaming the 32-bit build too soon. I set fail-fast to false so all builds run to completion, and they all give the undefined reference to dgetri_ error. I am trying to reproduce locally.

[matthew-brett]

About importance of Windows 32-bit bit - yes, that is usually me quoting Steve Dower : pypa/manylinux#118 (comment) . I haven't investigated the wheel download statistics though.

[mattip]

something is still off with the interface64 builds, the internal OpenBLAS linktest is failing.

[mattip]

self-merging to move this chain of fixes forward