![]() |
AI Engine-ML v2 Intrinsics User Guide
v2025.1
|
Topics | |
| Brain Floating-Point Vector Types | |
| Complex Floating-Point Vector Types | |
| Complex Vector Types | |
| Floating-Point Vector Types | |
| Integer Vector Types | |
| MX Vector Types | |
| Microscaling Formats (MX) vector data-types are stored in the vector registers within the AIE-ML v2. | |
| Sparse Vector Types | |
Vector data-types pack multiple scalar data elements into a wider vector. There are vector data-types with standard sized elements (4-bit, 8-bit, 16-bit, 32-bit).
The general naming convention for these data-types is shown below.
v{NumLanes}[cu]int{SizeofElement}
v{NumLanes}bfloat16
v{NumLanes}float[8,16]
v{NumLanes}mx{6,9}
Disambiguation :
For example, v32uint4, v16int8, v8int16, v128int4, v32bfloat16, v32float16
The total data-width of the vector data-types can be 128-bits, 256-bits, 512-bits, or 1024-bits.
| Scalar type | Supported vector lanes |
|---|---|
| int4/uint4 | 32/64/128/256 |
| int8/uint8 | 16/32/64/128 |
| int16/uint16 | 8/16/32/64 |
| int32/uint32 | 4/8/16/32 |
| cint16 | 4/8/16/32 |
| cint32 | 2/4/8/16 |
| cfloat | 2/4/8/16/32 |
| bfloat8 | 8/16/32/64/128/256 |
| bfloat16 | 8/16/32/64/128 |
| cbfloat16 | 4/8/16/32/64 |
| float8 | 2/4/8/16/32/64/128/ |
| float16 | 2/4/8/16/32/64/128 |
| float | 2/4/8/16/32/64 |
| mx4 | 16/64/128/256 |
| mx6 | 16/64/128/256 |
| mx9 | 16/32/64/128/256 |
| int4/uint4_sparse | 64/128 |
| int8/uint8_sparse | 128/256 |
| int16/uint16_sparse | 64/128 |
| float8_sparse | 128/256 |
| float16_sparse | 64/128 |
| bfloat8_sparse | 128/256 |
| bfloat16_sparse | 64/128 |