Well, it includes some masking instructions that allow algorithms to use vectorization in cases where it's not natively processing vectors. The benefit should b

Well, it includes some masking instructions that allow algorithms to use vectorization in cases where it's not natively processing vectors. The benefit should be similar to the thread-local control flow stack introduced in the Volta microarchitecture of Nvidia.