Update LLVM version to the latest we can (LLVM 6 -> ???)

[Quincunx271]

Currently, OptSched is assuming LLVM 6, but some things only work with LLVM 7, and some configurations need higher versions (e.g. the GPU build needs LLVM 9). It would make sense to make the LLVM version the highest version we can support.

Things to check:

• ROCm latest LLVM version support
• Flang latest LLVM version support
• Whether this will cause any issues with in-progress papers or projects.

We would want the highest version of LLVM that we could use.

[Quincunx271]

• ROCm 4.2.0 is based on LLVM 12.0.0: https://github.com/RadeonOpenCompute/llvm-project/blob/rocm-4.2.0/llvm/CMakeLists.txt#L19
• Classic Flang (which we are currently using) supports a latest of LLVM 10.0.0: https://github.com/flang-compiler/flang/wiki/Building-Flang#dependencies
  • However, there seems to be a new Flang which is now part of the LLVM project itself. In that case, we may be able to use this new Flang, which should work with newer LLVMs: https://flang.llvm.org/docs/

For the final point about issues with in-progress papers or projects, I think that can be worked around by updating LLVM in a separate branch, only merging it when the in-progress papers / projects are finished or when those in-progress papers / projects are okay to live in a separate branch using the older LLVM.

[Quincunx271]

Related: #170

[Quincunx271]

AFAICT, no current work is based on LLVM 6, but there is current work based on LLVM 7.1.0. It may be worth migrating only to LLVM 7.1.0 first, then dropping support for lower LLVM versions later (perhaps on some schedule)