Integer shuffles — byte, word, dword (per 128-bit lane, mask or imm8 controlled)

Float shuffles — single / double precision, two source registers, imm8 controlled

128-bit lane shuffles — pick whole 128-bit lanes from two sources, imm8 controlled, k-mask supported

Broadcasts — replicate one element or a 128-/256-bit unit across the destination, k-mask supported

Sign / zero extension — widen each packed byte/word/dword element to a larger element, k-mask supported

Packed equality — all-ones or all-zeros per element, or an opmask in the EVEX form

Predicate compares into an opmask — imm8 picks EQ/LT/LE/FALSE/NEQ/NLT/NLE/TRUE, signed or unsigned