[llvm] [X86] Prefer 128-bit vectors for Bulldozer CPUs (PR #225838)

[Simon Pilgrim via llvm-commits]

https://github.com/RKSimon requested changes to this pull request.

Please can you merge against trunk and regenerate the test checks - hopefully I've cleaned up all the tests now, so all you should need to do for any clashing cases is to change the some of bdver check-prefix entries to AVX128 instead of AVX256

https://github.com/llvm/llvm-project/pull/225838