Flags and Condition Codes

Condition Encoding

Conditional instructions encode a four-bit condition code. The zero-valued condition is T, which is also used for canonical aliases such as JMP for Jcc.T.

The condition-code bits are the low four meaningful bits of FLAGS, as shown in the Programming Model section.

Condition Code Encoding

Bits	Name	Aliases	Expression
0000	T	-	true
0001	F	-	false
0010	EQ	Z	FLAGS.Z == 1
0011	NE	NZ	FLAGS.Z == 0
0100	ULT	C	FLAGS.C == 1
0101	UGE	NC	FLAGS.C == 0
0110	MI	N	FLAGS.N == 1
0111	PL	NN	FLAGS.N == 0
1000	VS	V	FLAGS.V == 1
1001	VC	NV	FLAGS.V == 0
1010	ULE	-	FLAGS.C == 1 || FLAGS.Z == 1
1011	UGT	-	FLAGS.C == 0 && FLAGS.Z == 0
1100	LT	-	FLAGS.N != FLAGS.V
1101	GE	-	FLAGS.N == FLAGS.V
1110	LE	-	FLAGS.Z == 1 || FLAGS.N != FLAGS.V
1111	GT	-	FLAGS.Z == 0 && FLAGS.N == FLAGS.V

Flag Meanings

Integer condition codes are held in FLAGS. Integer instructions leave FLAGS unchanged unless their instruction description explicitly defines a FLAGS write. Every FLAGS-writing instruction replaces the complete FLAGS.Z, FLAGS.N, FLAGS.C, and FLAGS.V image; partial FLAGS writes do not exist. An instruction that does not write FLAGS preserves the complete image.

For an operand width of n bits, ordinary integer ALU results are evaluated modulo 2^n unless the instruction explicitly defines a wider intermediate.

Explicit FLAGS-writing integer instructions include CMP, TEST, ADC, SBB, INCF, DECF, BTEST, BSET, BCLR, BCHG, SETF, WRFLAGS, CMPXCHG, SEGLEA, and the bounds-check instructions. CMPJcc and TESTJcc compute temporary condition flags for their branch decision and do not update architectural FLAGS. ADD, SUB, logic operations, INC, DEC, shifts, rotates, moves, extensions, min/max, and multiply/divide instructions leave FLAGS unchanged.

Integer Condition-Code Bits

Bit	Name	Meaning
FLAGS.Z	zero	Set when the result value is zero.
FLAGS.N	negative	Set from the most significant result bit at the operand size.
FLAGS.C	carry/borrow	Set on unsigned carry out for addition or unsigned borrow for subtraction.
FLAGS.V	overflow	Set on signed overflow or an explicitly reported exceptional condition.

Conditional Consumers

Conditional branches, calls, traps, and sets consume the condition-code table without recomputing FLAGS. An ordinary conditional consumer observes the FLAGS image produced by the last committed FLAGS-writing instruction. REPcc instead tests a temporary flag image derived from the body instruction's repeat-observed value, as defined in Repeated Scalar Execution. That test leaves architectural FLAGS unchanged.

A false condition suppresses the conditional action unless the instruction description explicitly defines a different commit rule, such as a repeat instruction's terminating iteration rule.

Floating-Point Interactions

Floating-point comparison instructions may update integer condition codes only when their instruction definition says so. FFLAGS holds floating-point exception flags, and FSTATUS holds the corresponding exception-condition enables and rounding state. IEEE-754 FFLAGS causes use independent accrued semantics. The complete-image rule applies to integer FLAGS.

Integer conditional consumers do not read FFLAGS directly; floating-point instructions that want integer control-flow consumers must materialize the selected condition into FLAGS or an Rn value.