[PATCH] [ARM] Remove dead code for target architectures with a '-' in the name

[Bradley Smith]

  No, I'm saying that the string 'armv7-a' will be parsed as 'armv7' then the 'a' will fall into the 2nd part of the triple, hence ignored for the purpose of the architectures. Any armv*-* architecture will never be parsed properly and hence shouldn't be used at all, this patch simply removes the code that makes it look like it can handle armv*-* when it can't.

  The Darwin specific stuff appears to be in ToolChains.cpp, but these cpu/architecture tables are used for -mcpu, hence armv*-* would be allowed.

  As for testing, given my above comment it doesn't make sense to test handling of armv*-* since we know that it can't be handled properly regardless. (Ideally, we'd want to fault armv*-* but I'm not sure clang is capable of that).

http://llvm-reviews.chandlerc.com/D1695