Pooling Functions

Functions
arm_status	arm_avgpool_s8 (const cmsis_nn_context *ctx, const cmsis_nn_pool_params *pool_params, const cmsis_nn_dims *input_dims, const q7_t *src, const cmsis_nn_dims *filter_dims, const cmsis_nn_dims *output_dims, q7_t *dst)
	s8 average pooling function. More...
int32_t	arm_avgpool_s8_get_buffer_size (const int output_x, const int ch_src)
	Get the required buffer size for S8 average pooling function. More...
arm_status	arm_max_pool_s8 (const cmsis_nn_context *ctx, const cmsis_nn_pool_params *pool_params, const cmsis_nn_dims *input_dims, const q7_t *src, const cmsis_nn_dims *filter_dims, const cmsis_nn_dims *output_dims, q7_t *dst)
	s8 max pooling function. More...
void	arm_maxpool_q7_HWC (q7_t *Im_in, const uint16_t dim_im_in, const uint16_t ch_im_in, const uint16_t dim_kernel, const uint16_t padding, const uint16_t stride, const uint16_t dim_im_out, q7_t *bufferA, q7_t *Im_out)
	Q7 max pooling function. More...
void	arm_avepool_q7_HWC (q7_t *Im_in, const uint16_t dim_im_in, const uint16_t ch_im_in, const uint16_t dim_kernel, const uint16_t padding, const uint16_t stride, const uint16_t dim_im_out, q7_t *bufferA, q7_t *Im_out)
	Q7 average pooling function. More...

Description

Perform pooling functions, including max pooling and average pooling

Function Documentation

void arm_avepool_q7_HWC	(	q7_t *	Im_in,
		const uint16_t	dim_im_in,
		const uint16_t	ch_im_in,
		const uint16_t	dim_kernel,
		const uint16_t	padding,
		const uint16_t	stride,
		const uint16_t	dim_im_out,
		q7_t *	bufferA,
		q7_t *	Im_out
	)

Parameters

[in,out]	Im_in	pointer to input tensor
[in]	dim_im_in	input tensor dimention
[in]	ch_im_in	number of input tensor channels
[in]	dim_kernel	filter kernel size
[in]	padding	padding sizes
[in]	stride	convolution stride
[in]	dim_im_out	output tensor dimension
[in,out]	bufferA	pointer to buffer space for input
[in,out]	Im_out	pointer to output tensor

Buffer size:

bufferA size: 2*dim_im_out*ch_im_in

The pooling function is implemented as split x-pooling then y-pooling.

This pooling function is input-destructive. Input data is undefined after calling this function.

References arm_q7_to_q15_no_shift().

arm_status arm_avgpool_s8	(	const cmsis_nn_context *	ctx,
		const cmsis_nn_pool_params *	pool_params,
		const cmsis_nn_dims *	input_dims,
		const q7_t *	input_data,
		const cmsis_nn_dims *	filter_dims,
		const cmsis_nn_dims *	output_dims,
		q7_t *	output_data
	)

Parameters

[in,out]	ctx	Function context (e.g. temporary buffer). Check the function definition file to see if an additional buffer is required. Optional function {API}_get_buffer_size() provides the buffer size if an additional buffer is required.
[in]	pool_params	Pooling parameters
[in]	input_dims	Input (activation) tensor dimensions. Format: [H, W, C_IN] Argument 'N' is not used.
[in]	input_data	Input (activation) data pointer. Data type: int8
[in]	filter_dims	Filter tensor dimensions. Format: [H, W] Argument N and C are not used.
[in]	output_dims	Output tensor dimensions. Format: [H, W, C_OUT] Argument N is not used. C_OUT equals C_IN.
[in,out]	output_data	Output data pointer. Data type: int8

Returns

The function returns ARM_MATH_SUCCESS - Successful operation

• Supported Framework: TensorFlow Lite

References cmsis_nn_pool_params::activation, cmsis_nn_context::buf, cmsis_nn_dims::c, cmsis_nn_tile::h, cmsis_nn_dims::h, MAX, cmsis_nn_activation::max, MIN, cmsis_nn_activation::min, cmsis_nn_pool_params::padding, cmsis_nn_pool_params::stride, cmsis_nn_tile::w, and cmsis_nn_dims::w.

int32_t arm_avgpool_s8_get_buffer_size	(	const int	dim_dst_width,
		const int	ch_src
	)

Parameters

[in]	dim_dst_width	output tensor dimension
[in]	ch_src	number of input tensor channels

Returns

The function returns required buffer size in bytes

arm_status arm_max_pool_s8	(	const cmsis_nn_context *	ctx,
		const cmsis_nn_pool_params *	pool_params,
		const cmsis_nn_dims *	input_dims,
		const q7_t *	input_data,
		const cmsis_nn_dims *	filter_dims,
		const cmsis_nn_dims *	output_dims,
		q7_t *	output_data
	)

Parameters

[in,out]	ctx	Function context (e.g. temporary buffer). Check the function definition file to see if an additional buffer is required. Optional function {API}_get_buffer_size() provides the buffer size if an additional buffer is required.
[in]	pool_params	Pooling parameters
[in]	input_dims	Input (activation) tensor dimensions. Format: [H, W, C_IN] Argument 'N' is not used.
[in]	input_data	Input (activation) data pointer. Data type: int8
[in]	filter_dims	Filter tensor dimensions. Format: [H, W] Argument N and C are not used.
[in]	output_dims	Output tensor dimensions. Format: [H, W, C_OUT] Argument N is not used. C_OUT equals C_IN.
[in,out]	output_data	Output data pointer. Data type: int8

Returns

The function returns ARM_MATH_SUCCESS - Successful operation

• Supported Framework: TensorFlow Lite

References cmsis_nn_pool_params::activation, cmsis_nn_dims::c, clamp_output(), compare_and_replace_if_larger_q7(), cmsis_nn_tile::h, cmsis_nn_dims::h, MAX, cmsis_nn_activation::max, MIN, cmsis_nn_activation::min, cmsis_nn_pool_params::padding, cmsis_nn_pool_params::stride, cmsis_nn_tile::w, and cmsis_nn_dims::w.

void arm_maxpool_q7_HWC	(	q7_t *	Im_in,
		const uint16_t	dim_im_in,
		const uint16_t	ch_im_in,
		const uint16_t	dim_kernel,
		const uint16_t	padding,
		const uint16_t	stride,
		const uint16_t	dim_im_out,
		q7_t *	bufferA,
		q7_t *	Im_out
	)

Parameters

[in,out]	Im_in	pointer to input tensor
[in]	dim_im_in	input tensor dimention
[in]	ch_im_in	number of input tensor channels
[in]	dim_kernel	filter kernel size
[in]	padding	padding sizes
[in]	stride	convolution stride
[in]	dim_im_out	output tensor dimension
[in,out]	bufferA	Not used
[in,out]	Im_out	pointer to output tensor

The pooling function is implemented as split x-pooling then y-pooling.

This pooling function is input-destructive. Input data is undefined after calling this function.

References compare_and_replace_if_larger_q7().