Fix an alignment error in `llvm::expandAtomicRMWToCmpXchg` without breaking the build...

Summary: Divide the primitive size in bits by eight so the initial load's alignment is in bytes as expected. Tested with the included unit test.

Reviewers: rengolin, jfb

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D11804

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@244229 91177308-0d34-0410-b5e6-96231b3b80d8