The names given to individual instructions in AssemblyLanguage. Different CPU families have different op codes, and assembler has mnemonic names for the different instructions (which to the CPU are just 1s and 0s).
(From a slashdot poll on favourite op-codes):
My favorite in the intel world is LOADALL, 0x0F05 on 286 and 0x0F07 on 386. LOADALL was put into the A & B steppings for test purposes, but was removed in the production C stepping of the 286. Then M$ sued intel and they were forced to put it back in all later steppings. They also put it in most, but not all, 386's but with the different opcode, just to fuck over M$. The 486 and later don't have it, because virus writers the world over would take advantage of it.
The reason its my favorite is that M$ used LOADALL as the key to getting around protected and real mode operation built into the 286 when it was developing DOS-286, a precursor to the OS/2 project. Because it allowed any M$ compiled program to switch from protected mode to real mode and back, it made multi-tasking easier than writing a real task handler. M$ was 12+ months and US$20 million into developing DOS-286 when they started testing on the production IBM PC-XT, with the production 286's. At that point there were 200,000+ units shipped, and IBM refused to do a recall. So M$ had to scrap the project and team up with IBM to work on OS/2. It cost them the lead when the Macintosh came out the next year, and as history has shown, they never really recovered :-)
(I think this came from the same slashdot poll)
The best op ever has to be the PowerPC's EIEIO (Ensure Inline Execution of I/O). Someone at IBM has a sense of humor.