Visible to Intel only — GUID: GUID-C36BD30A-44D9-43EE-A928-55F37A7D597C
Visible to Intel only — GUID: GUID-C36BD30A-44D9-43EE-A928-55F37A7D597C
enum dnnl_data_type_t
Overview
Data type specification. More…
#include <dnnl_common_types.h>
enum dnnl_data_type_t
{
dnnl_data_type_undef = 0,
dnnl_f16 = 1,
dnnl_bf16 = 2,
dnnl_f32 = 3,
dnnl_s32 = 4,
dnnl_s8 = 5,
dnnl_u8 = 6,
dnnl_f64 = 7,
dnnl_boolean = 8,
dnnl_f8_e5m2 = 9,
dnnl_f8_e4m3 = 10,
dnnl_s4 = 11,
dnnl_u4 = 12,
dnnl_data_type_max = 0x7fff,
};
Detailed Documentation
Data type specification.
Enum Values
dnnl_data_type_undef
Undefined data type, used for empty memory descriptors.
dnnl_f16
16-bit/half-precision floating point.
dnnl_bf16
non-standard 16-bit (bfloat16 w/ 7 bit mantissa) floating point.
dnnl_f32
32-bit/single-precision floating point.
dnnl_s32
32-bit signed integer.
dnnl_s8
8-bit signed integer.
dnnl_u8
8-bit unsigned integer.
dnnl_f64
64-bit/double-precision floating point.
dnnl_boolean
Boolean data type. Size is C++ implementation defined.
dnnl_f8_e5m2
OFP8 standard 8-bit floating-point with a 5-bit exponent and a 2-bit mantissa.
dnnl_f8_e4m3
OFP8 standard 8-bit floating-point with a 4-bit exponent and a 3-bit mantissa.
dnnl_s4
4-bit signed integer.
dnnl_u4
4-bit unsigned integer.
dnnl_data_type_max
Parameter to allow internal only data_types without undefined behavior.
This parameter is chosen to be valid for so long as sizeof(int) >= 2.