Thanks for the correction. I hadn't considered bulk memory operations to be part of SIMD operation but it makes sense -- they operate on a larger grain than word-size so they can do the same operation with less micro-ops overhead.