FCMOV is a floating point conditional move opcode of the Intel x86 architecture, first introduced in Pentium Pro processors. It copies the contents of one of the floating point stack register, depending on the contents of EFLAGS integer flag register, to the ST(0) (top of stack) register. There are 8 variants of the instruction selected by the condition codes that need be set for the instruction to perform the move.
Similar to the CMOV instruction, FCMOV allows to perform some conditional operations without the usual branching overhead.[1] However, it has a higher latency than conditional branch instructions.[2] Therefore, it is most useful for simple yet unpredictable comparison or conditional operations, where it can provide substantial performance gains.
The instruction is usually used with the FCOMI instruction or the FCOM-FSTSW-SAHF idiom to set the relevant condition codes based on the result of a floating point comparison.
This table shows the variants of the FCMOV instructions.[1] In the Opcode column, i denotes the number of the floating point stack register used as the second operand (the first is always the ST(o) register - the top of the stack). Add it to the given number to obtain the corresponding opcode.
| Opcode | Mnemonic | Meaning | Condition |
|---|---|---|---|
| DA C0+i | FCMOVB | Move if below | Carry flag set |
| DA C8+i | FCMOVE | Move if equal | Zero flag set |
| DA D0+i | FCMOVBE | Move if below or equal | Carry flag or Zero flag set |
| DA D8+i | FCMOVU | Move if unordered | Parity flag set |
| DB C0+i | FCMOVNB | Move if not below | Carry flag cleared |
| DB C8+i | FCMOVNE | Move if not equal | Zero flag cleared |
| DB D0+i | FCMOVNBE | Move if not below or equal | Carry flag and Zero flag cleared |
| DB D8+i | FCMOVNU | Move if not unordered | Parity flag cleared |