Is VGF2P8AFFINEINVQB the longest x86 instruction mnemonic?

Question

Trivia Question

At 17 characters, is VGF2P8AFFINEINVQB - Galois Field Affine Transformation Inverse, the longest x86 instruction mnemonic? Is there a length limit?

Answer 1

Sure, probably yes at the moment, VGF2P8AFFINEINVQB at 17 bytes is the longest mnemonic in Intel's docs.

There's no length limit; assemblers are just software, each one with a different parser. If you were adding this instruction to an old assembler that had some hard-coded token-length limit, you could make up an alternate name. The CPU only cares about machine code¹.

One way to confirm if it's the longest Intel currently documents would be to check a database of instructions, like the one that comes with XED. https://uops.info/xml.html . That should contain them all, unless some are constructed via suffixes.

---

Other long instructions I see from visually scanning down the left-hand column of https://www.felixcloutier.com/x86/ include AESDECWIDE128KL (15 bytes), but that doesn't have a V VEX form and that would only get it up to 16 byte. But that page doesn't use a monospace font, and some entry titles aren't the full name of an instruction (e.g. vbroadcast).

Compares with an infix predicate can get somewhat long, but I think the longest is VCMPUNORD_SPS (13 bytes).

VFMADDSUB132PD and VPMULTISHIFTQB are both 14. VBROADCASTF32X8 is 15.

---

Footnote 1: Intel's manuals don't bother to define much about asm syntax rules, but do lay out rules for machine code.

Machine code has a max instruction length of 15 byte which can at least hypothetically force breaking up a complex instruction into multiple instructions using a temporary register, e.g. for an LEA or mov-immediate. (Maybe for real with APX for 4-byte EVEX prefixes for integer instructions, like lock add r31d, [fs: r30+rax*4 + 0x123456], 12345678 which needs a ModRM+SIB + disp32 and an imm32, so 10 bytes for that, plus 4-byte EVEX plus opcode. So that's 15 byte without segment or lock prefixes.)