SimaKernelQuantization Struct

Quantization fused into a kernel's store. More...

Declaration

struct simaai::neat::SimaKernelQuantization { ... }

Included Headers

#include <SimaKernelTypes.h>

Public Member Attributes Index

float	multiplier = 1.0f
	Positive, finite multiplier. More...
int	zero_point = 0
	INT8 zero point, in [-128, 127]. More...
SimaQuantRounding	rounding = SimaQuantRounding::NearestEven
	Integer rounding mode. More...

Description

Quantization fused into a kernel's store.

The kernel computes saturate_int8(round(binary32(x * multiplier)) + zero_point), with round selected by rounding. multiplier multiplies the value; it is not a dequantization scale. When the Node knows the real value of one unit of x, the INT8 output carries Tensor::semantic.quant with that value divided by multiplier as its scale, and this zero_point.

Definition at line 50 of file SimaKernelTypes.h.

Public Member Attributes

multiplier

float simaai::neat::SimaKernelQuantization::multiplier = 1.0f

Positive, finite multiplier.

Definition at line 51 of file SimaKernelTypes.h.

rounding

SimaQuantRounding simaai::neat::SimaKernelQuantization::rounding = SimaQuantRounding::NearestEven

Integer rounding mode.

Definition at line 53 of file SimaKernelTypes.h.

zero_point

int simaai::neat::SimaKernelQuantization::zero_point = 0

INT8 zero point, in [-128, 127].

Definition at line 52 of file SimaKernelTypes.h.

---

The documentation for this struct was generated from the following file:

• SimaKernelTypes.h

---

Generated via doxygen2docusaurus 2.0.0 by Doxygen 1.15.0.