Developer Guide and Reference

  • 2022.1
  • 04/11/2022
  • Public Content
Contents

enum dnnl::fpmath_mode

Overview

Floating-point math mode. More…
#include <dnnl.hpp> enum fpmath_mode { strict = dnnl_fpmath_mode_strict, bf16 = dnnl_fpmath_mode_bf16, f16 = dnnl_fpmath_mode_f16, any = dnnl_fpmath_mode_any, };

Detailed Documentation

Floating-point math mode.
Enum Values
strict
Default behavior, no downconversions allowed.
bf16
Implicit f32->bf16 conversions allowed.
f16
Implicit f32->f16 conversions allowed.
any
Implicit f32->f16 or f32->bf16 conversions allowed.

Product and Performance Information

1

Performance varies by use, configuration and other factors. Learn more at www.Intel.com/PerformanceIndex.