https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65482
--- Comment #3 from Michael Meissner <meissner at gcc dot gnu.org> --- The -mallow-movmisalign option has always been a debug option. It was added in the early power7 days. However, after benchmarking, we discovered that vectors of 32-bit elements (i.e. vector int and vector float) were actually slower if we enabled the movmisalign patterns, so I disabled it, and never revisited it. I think I had discussions with the then Haifa team that were working on vectorization support, and I tried to argue that we should do tests without the option, even if it means some tests won't get vectorized. In the power8 days, the hardware had better support for unaligned vector support, and we added -mefficient-unaligned-vsx. We probably should redo the tests that use -mallow-misalign to be power8 tests.
