Enable building a functional Sire on ppc64le architecture

[ptosco]

This PR implements a few tweaks to get Sire to build and pass all unit tests on a ppc64le architecture.
More details:

• build/build_sire.py: replaced os.system with subprocess.run in This avoids that a new shell is spawned to run the subprocess and allows keeping the environment of the calling shell. I had already committed similar changes to unit tests with the same purpose.
• compile_sire.sh: added ppc64le support
• corelib/CMakeLists.txt: added ppc64le support; added support for Debug builds; avoid attempting to compile cpuid on ppc64le platforms where it is not supported
• corelib/build/cmake/FindVector.cmake: removed unused SSE4 flag
• corelib/src/apps/testmulti/speedtest.cpp: avoid that the build fails if _mm_alloc and _mm_free are not defined because no intrinsics headers were included
• corelib/src/libs/SireBase/cpuid.cpp: added support for getting the number of CPUs with native platform-specific methods in the absence of libcpuid
• corelib/src/libs/SireMaths/multi*.h: all of these files used the ! instead of the ~ operator for bitwise not, which is wrong and leads to 20/50 unit tests failing in SireMM. This has gone unnoticed so far as only happens when a non-SSE/non-AVX build is done, which never happens these days on x86 platforms. I also replaced the nested loop with a simpler and more efficient construct. Now all tests pass.
• wrapper/CMakeLists.txt: added a comment

[chryswoods]

Thanks @ptosco - this looks fantastic and much appreciated. On a quick look I am very much in favour of merging (particularly as you’ve fixed a long standing bug that caused many unit test failures - feel a little embarrassed about that!).

I’m on holiday until next Tuesday so can’t review this until then. I’m happy for @lohedges to merge this tomorrow or Monday if he’s happy with it, as it looks great.

[lohedges]

Hi @ptosco. Many thanks for this, it looks great. I'll try to test local builds on Linux and macOS today and will accept when I'm done. Obviously, I won't be able to test on ppc64le mysef, but I'm happy to trust that this is working. I'll also check for any knock-on effects to BioSimSpace.

Thanks again, your efforts are always greatly appreciated.

[lohedges]

All good on Linux on macOS over here so I'm happy to merge. Next to deal with any issues with the pipelines build. (It's not been run in a while so I'm hoping that there are no new conda-related issues.)

[lohedges]

Huzzah, the build worked without issue 👍

Given the intrascale looping speedups, the stable sorting of potential terms and this pull, it's probably time to think about 2020.1.0 releases for Sire and BioSimSpace. @ptosco: are there any other pull requests that you have in the pipeline? If so, I'll hold off until you submit them. Otherwise I can probably get a new release sorted next week.

Cheers.

[ptosco]

@lohedges Hi Lester, no, not really, I think that's all for now. Thank you for the super-quick merge!

[lohedges]

Just a quick update. I still plan on making 2020.1.0 releases for Sire and BioSimSpace. @ptosco, I was also planning on updating MiniConda to Python 3.8 and upgrading the Conda dependencies to their latest versions. They way I normally choose suitable versions is to go to the conda-forge page for each dependency, then choose the most recent version that is available for all operating systems that we support. If there is a known bug in a particular version, then I'll roll back the version for all operating systems. Is there anything that I should be aware of for the Windows build, e.g. certain versions of particular dependencies not working. Also, do I only need to care about the version for win-64, or do you build for win-32 too?

[ptosco]

@lohedges I am not aware of particular Windows issues with Python 3.8 - as long as the dependencies are available on conda-forge (sometimes the Windows packages lag behind a bit). And, win64 only. Thanks!

[lohedges]

Hmmm, it turns out that there's no easy way to do this. OpenMM hasn't been built for Python 3.8 (apart from in conda-forge testing) so we're stuck with 3.7 for now. Also, when updating dependencies for Sire I need to be careful that they're compatible with all dependencies of BioSimSpace, particularly rdkit. Wish there was an easier way.

I'll push an update once I've got a working environment. If you could then test on Windows that would be great. If all is okay, I'll go ahead and create a new release.