Developer Guide and Reference

  • 2022.1
  • 04/11/2022
  • Public Content

Primitive Attributes: floating-point math mode

For some applications, it can be beneficial to allow down-conversions to speedup computations without noticeable impact on accuracy.
This section describes how the default numerical behavior of oneDNN (described in Data Types) can be altered to allow implicit down-conversions of floating-point types.

The floating-point math mode attribute.

When passed to a primitive creation, the dnnl::fpmath_mode primitive attribute specifies which implicit down-conversions are allowed for that given primitive. Only down-conversions from f32 to narrower data-types (f16, bf16) are currently allowed. Furthermore these down-conversions are allowed only during computation, and do not affect the storage datatype (which must remain f32).
The dnnl::fpmath_mode primitive attribute can take 3 types of values:
  • the
    mode disables any down-conversion.
  • the
    mode allows all conversions from f32 to a smaller floating-point datatype (bf16 or f16).
  • a specific datatype (f16 or bf16) which specifically allows down-conversion only from f32 to a datatype at least as accurate as the specified data-type (at least same number of exponent and mantissa bits).
This attribute is ignored if a primitive computation data-type is integral.

A note on default floating-point math mode

The default floating-point mode is
, which means no implicit down-conversion is allowed. However, this default behavior can be changed with the
environment variable, the dnnl_set_default_fpmath_mode (C API) or the dnnl::set_default_fpmath_mode (C++ API) functions.
For builds where Arm Compute Library is enabled, setting
will instruct Compute Library to dispatch bfloat16 kernels where available, provided the hardware supports bfloat16 instructions. Note: this may introduce a drop in accuracy.

Product and Performance Information


Performance varies by use, configuration and other factors. Learn more at