Activation functions needed for quantization

These functions are used to accelerate quantization algorithms in Synet Framework. More...

Functions
SIMD_API void	SimdSynetQuantizedHardSigmoid (const uint8_t *src, const float *srcScale, int srcZero, size_t size, const float *scale, const float *shift, uint8_t *dst, const float *dstScale, int dstZero)
	Performs forward propagation of UINT8 quantized HardSigmoid layer. More...
SIMD_API void	SimdSynetQuantizedHswish (const uint8_t *src, const float *srcScale, int srcZero, size_t size, const float *shift, const float *scale, uint8_t *dst, const float *dstScale, int dstZero)
	Performs forward propagation of UINT8 quantized Swish layer. More...
SIMD_API void	SimdSynetQuantizedPreluLayerForward (const uint8_t *src, const float *srcScale, int srcZero, size_t channels, size_t spatial, const float *slope, uint8_t *dst, const float *dstScale, int dstZero, SimdTensorFormatType format)
	Performs forward propagation of UINT8 quantized PReLU layer. More...

Detailed Description

These functions are used to accelerate quantization algorithms in Synet Framework.

Function Documentation

SimdSynetQuantizedHardSigmoid()

void SimdSynetQuantizedHardSigmoid	(	const uint8_t *	src,
		const float *	srcScale,
		int	srcZero,
		size_t	size,
		const float *	scale,
		const float *	shift,
		uint8_t *	dst,
		const float *	dstScale,
		int	dstZero
	)

Performs forward propagation of UINT8 quantized HardSigmoid layer.

Algorithm's details:

value = (src - srcZero)*srcScale[0];
value = Max(Min(value, shift[0]) + shift[0], 0) * scale[0] * value;
dst = RestrictRange(Round(value/dstScale[0]) + dstZero, 0, 255);

Note

This function is used in Synet Framework.

Parameters

[in]	src	- a pointer to UINT8 input tensor.
[in]	srcScale	- a pointer to quantization scale of input tensor.
[in]	srcZero	- a quantization zero parameter of input tensor.
[in]	size	- a size of (input/output) tensors.
[in]	scale	- a pointer to scale parameter. Only scale[0] is used. It is equal to 1/6 in the original paper.
[in]	shift	- a pointer to shift parameter. Only shift[0] is used. It is equal to 1/2 in the original paper.
[out]	dst	- a pointer to UINT8 output tensor.
[in]	dstScale	- a pointer to output quantization scale.
[in]	dstZero	- an output quantization zero.

SimdSynetQuantizedHswish()

void SimdSynetQuantizedHswish	(	const uint8_t *	src,
		const float *	srcScale,
		int	srcZero,
		size_t	size,
		const float *	shift,
		const float *	scale,
		uint8_t *	dst,
		const float *	dstScale,
		int	dstZero
	)

Performs forward propagation of UINT8 quantized Swish layer.

Algorithm's details:

value = (src - srcZero)*srcScale[0];
value = Max(0, Min(value * scale[0] + shift[0], 1));
dst = RestrictRange(Round(value/dstScale[0]) + dstZero, 0, 255);

Note

This function is used in Synet Framework.

Parameters

[in]	src	- a pointer to UINT8 input tensor.
[in]	srcScale	- a pointer to quantization scale of input tensor.
[in]	srcZero	- a quantization zero parameter of input tensor.
[in]	size	- a size of (input/output) tensors.
[in]	shift	- a pointer to shift parameter. Only shift[0] is used. It is equal to 3 in the original paper.
[in]	scale	- a pointer to scale parameter. Only scale[0] is used. It is equal to 1/6 in the original paper.
[out]	dst	- a pointer to UINT8 output tensor.
[in]	dstScale	- a pointer to output quantization scale.
[in]	dstZero	- an output quantization zero.

SimdSynetQuantizedPreluLayerForward()

void SimdSynetQuantizedPreluLayerForward	(	const uint8_t *	src,
		const float *	srcScale,
		int	srcZero,
		size_t	channels,
		size_t	spatial,
		const float *	slope,
		uint8_t *	dst,
		const float *	dstScale,
		int	dstZero,
		SimdTensorFormatType	format
	)

Performs forward propagation of UINT8 quantized PReLU layer.

Algorithm's details:

value = (src - srcZero)*srcScale[0];
value = value > 0 ? value : slope[c]*value;
dst = RestrictRange(Round(value/dstScale[0]) + dstZero, 0, 255);

Note

This function is used in Synet Framework.

Parameters

[in]	src	- a pointer to UINT8 input tensor.
[in]	srcScale	- a pointer to quantization scale of input tensor.
[in]	srcZero	- a quantization zero parameter of input tensor.
[in]	channels	- a number of channels in (input/output) tensors.
[in]	spatial	- a spatial size of (input/output) tensors.
[in]	slope	- a pointer to the 32-bit float array with slope coefficients. The size of the array is equal to channels.
[out]	dst	- a pointer to UINT8 output tensor.
[in]	dstScale	- a pointer to output quantization scale.
[in]	dstZero	- an output quantization zero.
[in]	format	- a format of input and output tensors. It can be SimdTensorFormatNchw or SimdTensorFormatNhwc.