enum dnnl::fpmath_mode¶
Overview¶
Floating-point math mode. More…
#include <dnnl_common.hpp> enum fpmath_mode { strict = dnnl_fpmath_mode_strict, bf16 = dnnl_fpmath_mode_bf16, f16 = dnnl_fpmath_mode_f16, tf32 = dnnl_fpmath_mode_tf32, 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.
tf32
Implicit f32->tf32 conversions allowed.
any
Implicit f32->f16, f32->tf32 or f32->bf16 conversions allowed.