enum dnnl_quantization_mode_t

enum dnnl_quantization_mode_t#

Overview#

Quantization kind. More…

#include <dnnl_types.h>

enum dnnl_quantization_mode_t
{
    dnnl_quantization_mode_undef,
    dnnl_quantization_mode_static_sazp,
    dnnl_quantization_mode_dynamic_mx,
};

Detailed Documentation#

Quantization kind.

Enum Values#

dnnl_quantization_mode_undef

used for unspecified quantization kind

dnnl_quantization_mode_static_sazp

static quantization mode: quantization parameter is computed ahead of time with scale applied after zero-point (\(x_{f32} = scale * (x_{quant} - zp)\)) and passed to oneDNN as an input.

dnnl_quantization_mode_dynamic_mx

dynamic quantization mode following OCP MX spec: quantization parameter is computed by oneDNN following the OCP MX spec formula and written as an output.