This lets us experiment with 512-bit vectorization without passing force-vector-width manually. The code generated for a simple integer memset loop is properly vectorized. Disassembly is still broken for it though :(. llvm-svn: 212634
Adam Nemet