Instruction
TILEZERO
Sets every byte of a tile register to zero.
TILEZERO clears one tile register, the usual first step for an accumulator: before the dot products of a new block of C, such as TDPBSSD or TDPBF16PS, add into it. Intel’s manual has it clear the whole register, all 16 rows of 64 bytes, whatever shape the configuration gives the tile; since the bytes outside a tile’s shape are zero anyway, the effect is the same.
Its checks are the fewest of the tile instructions: tile state enabled in XCR0, 64-bit mode, a loaded configuration, and a tile that the configuration gives rows; bit 18 of IA32_XFD set raises #NM. Unlike the loads, stores and multiplications, it does not care how wide the tile’s rows are, and it is one of the few AMX instructions that also work under the ACE palette 2, where tiles hold the results of outer products such as TOP4BSSD.
When it is not needed
LDTILECFG zeroes all tile data as it loads a configuration, and TILERELEASE returns the tiles to their initial state, so code that configures the tiles and then fills each accumulator once needs no TILEZERO for that first block. An accumulator may also start from data: loading the old C with TILELOADD instead adds the product to it.
In compiled code
With -mamx-tile, GCC 14 and Clang 23 at -O2 compile _tile_zero(0) to
tilezero tmm0. Clang 23’s tile type __tile1024i has
__tile_zero(&c), which compiled to a TILEZERO of the register Clang
chose for c.
A use beside AVX-512
Intel’s optimization manual counts TILEZERO as an AMX compute instruction for the processor’s port management: while AMX compute instructions run, the processor closes one of its two AVX-512 FMA ports, and reopening it costs a few cycles each time. Between AMX blocks separated by about 100 to 300 AVX-512 instructions, the manual suggests one or two TILEZEROs in the middle of the AVX-512 code, about 100 instructions apart, to keep the port closed (Section 20.11.3).
Forms
| Instruction | Encoding | Requires |
|---|---|---|
TILEZERO tmm 64-bit mode only
TILEZERO_TMMu32 | VEX.LZ.F2.0F38.W0 49 /r | ACE v1 or AMX_TILE |
Latency and throughput
| Model | tmm |
|---|---|
| Sapphire Rapids | 100 / 0.25p0156 |
Operation
What the forms do, in pseudocode that our tests run against results recorded on real processors. How to read.
on = xcr0 if cr4 >> 18 & 1 else 0 # XCR0, where CR4.OSXSAVE is set
if on >> 17 & 3 != 3: # TILECFG and TILEDATA
fault(UD)
if palette == 0 or not tmm1.rows: # no configuration, or a tile it
fault(UD) # leaves out
if xfd >> 18 & 1: # the OS keeps the tiles from
fault(NM) # this thread
tmm1.write(0) # all 16 rows of 64 bytes
start_row = 0Not run on a processor: none of these processors has AMX.
Operands
| Operand | Kind | Action |
|---|---|---|
TILEZERO tmm 64-bit mode only | ||
tmm | explicit | write |
Flags
None: XED records no flag that the forms test or change.
Exceptions
The exception class is XED's name for the set of exception conditions that a form follows. How to read.
- All forms
AMX-E5
Attributes
What else XED records about the forms. How to read.
- All forms
- may abort a TSX transaction
Intrinsics
The C intrinsics of Clang 23.1.2 that compile to TILEZERO, each with the form it compiles to and the Clang options it needs. The forms come from compiling every intrinsic and decoding the result with Intel XED; all intrinsics A–Z explains the method.
| Intrinsic | Compiles to |
|---|---|
_tile_zero
void _tile_zero(const int tile) |
|
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 |
|---|---|---|
AMX_TILE_BASE | ACE v1 or AMX_TILE | Sapphire Rapids, Emerald Rapids, Granite Rapids, Diamond Rapids |
More AMX-TILE instructions
More ACE instructions
Sources
- Intel XED v2026.08.23 (commit
0bcb6237345c): forms, encodings, operands, flags, exception classes, attributes, ISA sets, CPUID bits and chip model. - Clang's x86 intrinsic headers, LLVM 23.1.2 (commit
85ac56026243): the intrinsics, their signatures and doc comments, and the Clang of that release, which compiled each of them. - 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 4-728 (TILEZERO)
- x86 Ecosystem Advisory Group: AI Compute Extensions (ACE) Specification, version 1.16.2 (July 2026)
- Intel Architecture Instruction Set Extensions and Future Features Programming Reference (319433-062, June 2026)
- Intel 64 and IA-32 Architectures Optimization Reference Manual, Volume 1 (248966-051, October 2026)
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 intrinsics and their signatures come from Clang and the timings from LLVM's scheduling models, both part of the LLVM Project, licensed under the Apache License 2.0 with LLVM Exceptions. The text is our own.