enum dnnl_fpmath_mode_t

Overview

Floating-point math mode. More…

#include <dnnl_types.h>

enum dnnl_fpmath_mode_t
{
    dnnl_fpmath_mode_strict,
    dnnl_fpmath_mode_bf16,
    dnnl_fpmath_mode_f16,
    dnnl_fpmath_mode_any,
};

Detailed Documentation

Floating-point math mode.

Enum Values

dnnl_fpmath_mode_strict

Default behavior, no downconversions allowed.

dnnl_fpmath_mode_bf16

Implicit f32->bf16 conversions allowed.

dnnl_fpmath_mode_f16

Implicit f32->f16 conversions allowed.

dnnl_fpmath_mode_any

Implicit f32->f16 or f32->bf16 conversions allowed.