Some SIMD optimization cases can occasionally exceed scalar versions by more than lane width multiply. Often because of branch elision, but also less dependency

Some SIMD optimization cases can occasionally exceed scalar versions by more than lane width multiply. Often because of branch elision, but also less dependency chains stalls due to the larger register file. There are also some special instructions that can be very useful in SIMD case (packing, unpacking, permutation, etc.).