Floating-point intrinsics - 2023.2 English

Vitis Tutorials: AI Engine (XD100)

Document ID
Release Date
2023.2 English

There is a limited set of intrinsics with which a multitude of operations can be performed. All of them return either a vector<float,8> or vector<cfloat,4>, 256-bit vectors.

The basic addition, subtraction, and negation functions are as follows:

  • fpadd

  • fpadd_abs

  • fpsub

  • fpsub_abs

  • fpneg

  • fpneg_abs

  • fpabs

The simple multiplier function is available with the following options:

  • fpmul

  • fpabs_mul

  • fpneg_mul

  • fpneg_abs_mul

The multiplication accumulation/subtraction function has the following options:

  • fpmac

  • fpmac_abs

  • fpmsc

  • fpmsc_abs

On top of these various intrinsics, you have a fully configurable version multiplier and multiply-accumulate:

  • fpmul_conf

  • fpmac_conf