Intel® oneAPI Deep Neural Network Developer Guide and Reference
A newer version of this document is available. Customers should click here to go to the newest version.
Visible to Intel only — GUID: GUID-82CF3A9C-BF28-45FF-B6A7-838F35E69E36
Visible to Intel only — GUID: GUID-82CF3A9C-BF28-45FF-B6A7-838F35E69E36
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.