Buffer Port Data Types - 2025.1 English - UG1603

AI Engine-ML Kernel and Graph Programming Guide (UG1603)

Document ID: UG1603
Release Date: 2025-05-29
Version: 2025.1 English

Table 1. AIE-ML Supported Buffer Port Data Types
Input Buffer Port Types	Output Buffer Port Types
`input_buffer<int4>`	`output_buffer<int4>`
`input_buffer<int8>`	`output_buffer<int8>`
`input_buffer<int16>`	`output_buffer<int16>`
`input_buffer<int32>`	`output_buffer<int32>`
`input_buffer<int64>`	`output_buffer<int64>`
`input_buffer<uint4>`	`output_buffer<uint4>`
`input_buffer<uint8>`	`output_buffer<uint8>`
`input_buffer<uint16>`	`output_buffer<uint16>`
`input_buffer<uint32>`	`output_buffer<uint32>`
`input_buffer<uint64>`	`output_buffer<uint64>`
`input_bufferd<cint16>`	`output_buffer<cint16>`
`input_bufferd<cint32>`	`output_buffer<cint32>`
`input_buffer<float>`	`output_buffer<float>`
`input_buffer<cfloat>`	`output_buffer<cfloat>`
`input_buffer<bfloat16>`	`output_buffer<bfloat16>`

Table 2. AIE-ML v2 Supported Buffer Port Data Types
Input Buffer Port Types	Output Buffer Port Types
`input_buffer<int4>`	`output_buffer<int4>`
`input_buffer<int8>`	`output_buffer<int8>`
`input_buffer<int16>`	`output_buffer<int16>`
`input_buffer<int32>`	`output_buffer<int32>`
`input_buffer<int64>`	`output_buffer<int64>`
`input_buffer<uint4>`	`output_buffer<uint4>`
`input_buffer<uint8>`	`output_buffer<uint8>`
`input_buffer<uint16>`	`output_buffer<uint16>`
`input_buffer<uint32>`	`output_buffer<uint32>`
`input_buffer<uint64>`	`output_buffer<uint64>`
`input_bufferd<cint16>`	`output_buffer<cint16>`
`input_bufferd<cint32>`	`output_buffer<cint32>`
`input_buffer<float>`	`output_buffer<float>`
`input_buffer<cfloat>`	`output_buffer<cfloat>`
`input_buffer<bfloat16>`	`output_buffer<bfloat16>`
`input_buffer<float16>`	`output_buffer<float16>`
`input_buffer<mx9>` ¹	`output_buffer<mx9>`
In all of these buffer declarations, the number of elements is the number of samples. For `mx9` block floating-point datatypes, an element is considered to be a block of 16 floating-point values encoded into an `mx9` block. In future releases this will be changed to the number of encoded floating-point values encoded.