PPC64LE: Fix build issue on gcc (#288)

* ppc64le: Several improvements for VSX

  - fix the build on gcc
  - fix accuracy tests
  - fix the build with C++ when 'vector' compiler token is combined,
    NOTE: the generated header is using '__vector' instead of 'vector'
  - fix unaligned memory load
  - set rounding to the nearest even, make it equivalent to X86
  - improve the performance of conversions
  - improve the performance of shuffles and reverses
  - cleanup the whole VSX implementation

* Travis CI: Add ppc64le support