llvm.org GIT mirror llvm / release_90 docs / AMDGPUInstructionSyntax.rst
release_90

Tree @release_90 (Download .tar.gz)

AMDGPUInstructionSyntax.rst @release_90view markup · raw · history · blame

AMDGPU Instruction Syntax

Instructions

Syntax

An instruction has the following syntax:

<opcode mnemonic>    <operand0>, <operand1>,...    <modifier0> <modifier1>...

:doc:`Operands<AMDGPUOperandSyntax>` are normally comma-separated while :doc:`modifiers<AMDGPUModifierSyntax>` are space-separated.

The order of operands and modifiers is fixed. Most modifiers are optional and may be omitted.

Type and Size Suffices

Instructions which operate with data have an implied type of data operands. This data type is specified as a suffix of instruction mnemonic.

There are instructions which have 2 type suffices: the first is the data type of the destination operand, the second is the data type of source data operand(s).

Note that data type specified by an instruction does not apply to other kinds of operands such as addresses, offsets and so on.

The following table enumerates the most frequently used type suffices.

Type Suffices Packed instruction? Data Type
_b512, _b256, _b128, _b64, _b32, _b16, _b8 No Bits.
_u64, _u32, _u16, _u8 No Unsigned integer.
_i64, _i32, _i16, _i8 No Signed integer.
_f64, _f32, _f16 No Floating-point.
_b16, _u16, _i16, _f16 Yes Packed.

Instructions which have no type suffices are assumed to operate with typeless data. The size of data is specified by size suffices:

Size Suffix Implied data type Required register size in dwords
- b32 1
x2 b64 2
x3 b96 3
x4 b128 4
x8 b256 8
x16 b512 16
x b32 1
xy b64 2
xyz b96 3
xyzw b128 4
d16_x b16 1
d16_xy b16x2 2 for GFX8.0, 1 for GFX8.1 and GFX9
d16_xyz b16x3 3 for GFX8.0, 2 for GFX8.1 and GFX9
d16_xyzw b16x4 4 for GFX8.0, 2 for GFX8.1 and GFX9

Warning

There are exceptions from rules described above. Operands which have type different from type specified by the opcode are :ref:`tagged<amdgpu_syn_instruction_operand_tags>` in the description.

Examples of instructions with different types of source and destination operands:

s_bcnt0_i32_b64
v_cvt_f32_u32

Examples of instructions with one data type:

v_max3_f32
v_max3_i16

Examples of instructions which operate with packed data:

v_pk_add_u16
v_pk_add_i16
v_pk_add_f16

Examples of typeless instructions which operate on b128 data:

buffer_store_dwordx4
flat_load_dwordx4

Encoding Suffices

Most VOP1, VOP2 and VOPC instructions have several variants: they may also be encoded in VOP3, DPP and SDWA formats.

The assembler will automatically use optimal encoding based on instruction operands. To force specific encoding, one can add a suffix to the opcode of the instruction:

Encoding Encoding Suffix
Native 32-bit encoding (VOP1, VOP2 or VOPC) _e32
VOP3 (64-bit) encoding _e64
DPP encoding _dpp
SDWA encoding _sdwa

These suffices are used in this reference to indicate the assumed encoding. When no suffix is specified, a native encoding is implied.

Operands

Syntax

Syntax of most operands is described :doc:`in this document<AMDGPUOperandSyntax>`.

For detailed information about operands follow operand links in GPU-specific documents:

Modifiers

Syntax

Syntax of modifiers is described :doc:`in this document<AMDGPUModifierSyntax>`.

Information about modifiers supported for individual instructions may be found in GPU-specific documents: