Instruction
OUTS
Writes a byte, word or doubleword from memory at [RSI] to the I/O port in DX and steps RSI; REP repeats it RCX times.
- Extensions
- Mnemonics
- OUTS, OUTSB, OUTSW, OUTSD
- Forms
- 6 (6 Legacy)
- Data
- Intel XED v2026.08.23
OUTS writes one element, a byte, word or doubleword, from [RSI] to the
I/O port in DX, then moves RSI to the next element: up while the
direction flag is clear, down while it is set. With a REP prefix it
repeats RCX times, so rep outsw sends a buffer to a device’s data
port, which OUT would do one register at a time.
INS runs the other way; its page covers what the two
share, including Intel’s caution about devices that cannot keep up.
What is particular to OUTS
- The segment may change. The source is DS:rSI, and a segment prefix replaces DS, where INS’s ES is fixed. In 64-bit mode DS adds nothing, but an FS or GS prefix adds that segment’s base.
- No quadword. With REX.W the element is still a doubleword, both manuals say, and RSI steps by 4.
- The port first. The permission check of OUT comes before the read of memory. At CPL 3 with IOPL 0, Zen 5 raised #GP for OUTSB, OUTSW and OUTSD, with REX.W, in the REP forms with RCX = 0, and with RSI = 8, in page 0, which the probe leaves unmapped, where a page fault might have been expected. RSI, RCX and memory were left alone.
- Not buffered. Intel’s Volume 1 treats OUTS and REP OUTS as OUT: the next instruction waits until the write and every earlier store have completed (Table 20-1).
- Stopping half way. An interrupt or exception during REP OUTS leaves RSI and RCX at the next element, so that the run resumes where it stopped. Intel’s Volume 3 adds that an I/O breakpoint on the port fires after the first element.
From C
glibc’s <sys/io.h> has outsb, outsw and outsl, which take a
port, a buffer and a count. At -O2, GCC 14 and Clang 23 move the count
into RCX and the port into EDX, while the buffer, the second argument,
already arrives in RSI; the inline assembly then emits CLD and REP
OUTSW, clearing the direction flag itself (see CLD).
AMD’s Zen 5 guide lists OUTS as microcoded.
Forms
| Instruction | Encoding | Requires |
|---|---|---|
OUTSB
OUTSB | 6E | Base instruction set |
OUTSW
OUTSW | 66 6F | Base instruction set |
OUTSD
OUTSD | 6FREX.W 6F | Base instruction set |
REP OUTSB
REP_OUTSB | F3 6EF2 6E | Base instruction set |
REP OUTSW
REP_OUTSW | 66 F3 6F66 F2 6F | Base instruction set |
REP OUTSD
REP_OUTSD | F3 6FF3 REX.W 6FF2 6FF2 REX.W 6F | Base instruction set |
Latency and throughput
| Model | OUTSBOUTSB | OUTSWOUTSW | OUTSDOUTSD |
|---|---|---|---|
| Sandy Bridge | 100 / 0.33p015 | 100 / 0.33p015 | 100 / 0.33p015 |
| Haswell, Broadwell, Skylake, Skylake-SP, Ice Lake | 100 / 0.25p0156 | 100 / 0.25p0156 | 100 / 0.25p0156 |
| Alder Lake, Sapphire Rapids | 100 / 17p0 p1 p2 p3 p4 p5 p6 p7 p8 p9 p10 p11 | 100 / 18p0 p1 p2 p3 p4 p5 p6 p7 p8 p9 p10 p11 | 100 / 20p0 p1 p2 p3 p4 p5 p6 p7 p8 p9 p10 p11 |
| Lunar Lake | 100 / 0.17p012345 | 100 / 0.17p012345 | 100 / 0.17p012345 |
| Bonnell | 74 / 37p01 | 74 / 37p01 | 74 / 37p01 |
| Silvermont | 100 / 1FPC0 | 100 / 1FPC0 | 100 / 1FPC0 |
| Piledriver | 100 / 0.5EX0 EX1 | 100 / 0.5EX0 EX1 | 100 / 0.5EX0 EX1 |
| Zen, Zen 2 | 100 / 0.25 | 100 / 0.25 | 100 / 0.25 |
| Zen 3, Zen 4 | 100 / 25ALU0 ALU1 ALU2 ALU3 | 100 / 25ALU0 ALU1 ALU2 ALU3 | 100 / 25ALU0 ALU1 ALU2 ALU3 |
| Jaguar | 100 / 0.5ALU0 ALU1 | 100 / 0.5ALU0 ALU1 | 100 / 0.5ALU0 ALU1 |
Operation
What the forms do, in pseudocode that our tests run against results recorded on real processors. How to read.
n = min(OSIZE, 32) # REX.W: still a doubleword
if CPL > iopl and not io_allowed(dx, n):
fault(GP, 0) # as in the probe's cases: CPL 3,
# IOPL 0, the port not allowed
io_write(dx, n, load(aSI, n)) # DS:rSI; an FS or GS prefix adds
# that segment's base
step = -(n // 8) if df else n // 8
aSI = low(aSI + step, ASIZE)Run on AMD Zen 5 (EPYC 9645): 9 cases. Every one raised an exception, at CPL 3, where the probe runs.
n = min(OSIZE, 32) # REX.W: still a doubleword
if CPL > iopl and not io_allowed(dx, n):
fault(GP, 0) # even when aCX is 0 (Zen 5)
step = -(n // 8) if df else n // 8
s = aSI
for i in range(aCX): # one port write per element
io_write(dx, n, load(s, n))
s = low(s + step, ASIZE)
aSI = s
aCX = 0Run on AMD Zen 5 (EPYC 9645): 10 cases. Every one raised an exception, at CPL 3, where the probe runs.
Operands
Flags
| DF | IOPL |
|---|---|
| All forms | |
| T | T |
Attributes
What else XED records about the forms. How to read.
- All forms
- may abort a TSX transaction
- REP forms
- repeated: count in rCX
CPUID and processors
Each form belongs to an XED ISA set. A form can be used when the processor reports every CPUID bit of one of its ISA set's alternatives.
| ISA set | CPUID | Processors |
|---|---|---|
I186 | No CPUID bit: part of the base instruction set. | Every x86-64 processor in XED’s model and Zen 3 to Zen 6 |
I386 | No CPUID bit: part of the base instruction set. | Every x86-64 processor in XED’s model and Zen 3 to Zen 6 |
More General-purpose instructions
Sources
- Intel XED v2026.08.23 (commit
0bcb6237345c): forms, encodings, operands, flags, exception classes, attributes, ISA sets, CPUID bits and chip model. - The CPU data of x86-64.net: Zen 3 to Zen 6, from GCC’s
-marchfeature sets and AMD’s Processor Programming References, which the page of each processor cites. - LLVM 23.1.2's x86 scheduling models (commit
85ac56026243): the latency, reciprocal throughput and ports of each form, as llvm-mca reports them. - Intel 64 and IA-32 Architectures Software Developer's Manual, Volume 2 (325383-093, September 2026): Volume 2B, page 4-173 (OUTS/
OUTSB/ OUTSW/ OUTSD) - Intel 64 and IA-32 Architectures Software Developer's Manual, Volume 1 (253665-093, September 2026)
- Intel 64 and IA-32 Architectures Software Developer's Manual, Volume 3 (325384-093, September 2026)
- AMD64 Architecture Programmer's Manual, Volume 3: General-Purpose and System Instructions (24594, rev. 3.39, October 2026)
- Software Optimization Guide for the AMD Zen5 Microarchitecture (58455, rev. 1.00, August 2024), Wayback Machine copy of AMD's ZIP
The tables are derived from Intel XED, Copyright Intel Corporation, licensed under the Apache License 2.0; x86-64.net converted and reformatted them. The timings come from LLVM's scheduling models, part of the LLVM Project, licensed under the Apache License 2.0 with LLVM Exceptions. The text is our own.