VXLIB_convolve_i8u_c16s_o8u

Functions
VXLIB_STATUS	VXLIB_convolve_i8u_c16s_o8u (const uint8_t src[restrict], const VXLIB_bufParams2D_t *src_addr, uint8_t dst[restrict], const VXLIB_bufParams2D_t *dst_addr, const int16_t conv_mat[restrict], int32_t conv_width, int32_t conv_height, uint32_t conv_scale)
VXLIB_STATUS	VXLIB_convolve_i8u_c16s_o8u_checkParams (const uint8_t src[restrict], const VXLIB_bufParams2D_t *src_addr, const uint8_t dst[restrict], const VXLIB_bufParams2D_t *dst_addr, const int16_t conv_mat[restrict], int32_t conv_width, int32_t conv_height, uint32_t conv_scale)

Detailed Description

Function Documentation

VXLIB_convolve_i8u_c16s_o8u()

VXLIB_STATUS VXLIB_convolve_i8u_c16s_o8u	(	const uint8_t	src[restrict],
		const VXLIB_bufParams2D_t *	src_addr,
		uint8_t	dst[restrict],
		const VXLIB_bufParams2D_t *	dst_addr,
		const int16_t	conv_mat[restrict],
		int32_t	conv_width,
		int32_t	conv_height,
		uint32_t	conv_scale
	)

Description:

User-defined image convolution function of an 8 bit image, producing an 8 bit image

Method:

Convolve the input image with the convolution matrix, saturating the output to [0, 255]

Parameters

[in]	src[]	Pointer to array containing input image (UQ8.0)
[in]	src_addr[]	Pointer to structure containing dimensional information of src
[out]	dst[]	Pointer to array containing output image (UQ8.0)
[in]	dst_addr[]	Pointer to structure containing dimensional information of dst
[in]	conv_mat[]	Pointer to array containing convolution matrix (SQ15.0)
[in]	conv_width	Width of convoltion matrix
[in]	conv_height	Height of convolution matrix
[in]	conv_scale	Scale value to divide the final sum of products (power of 2)

Assumptions:

• I/O buffer pointers are assumed to be not aliased.
• Output height should be == (Input height - (conv_height-1))
• Output width should be == (Input width - conv_width-1) OR (Input width)
• conv_width should be an odd number in the range 3 <= conv_width <= 9
• conv_height should be an odd number in the range 3 <= conv_width <= 9
• conv_scale should be a positive power of 2

Performance Considerations:

• For best performance, the following parameter settings are recommended:
  • Align all pointers to 8 byte boundaries
  • Set all stride values to a multiple of 8
  • Set output width values to a multiple of 16
  • Set widths equal to each other and equal to strides (allows processing over whole image in single loop)
  • Optimized for 3x3, 5x5, 7x7, 9x9 matrix sizes.

VXLIB_convolve_i8u_c16s_o8u_checkParams()

VXLIB_STATUS VXLIB_convolve_i8u_c16s_o8u_checkParams	(	const uint8_t	src[restrict],
		const VXLIB_bufParams2D_t *	src_addr,
		const uint8_t	dst[restrict],
		const VXLIB_bufParams2D_t *	dst_addr,
		const int16_t	conv_mat[restrict],
		int32_t	conv_width,
		int32_t	conv_height,
		uint32_t	conv_scale
	)

Description:

Checks the parameters for programming errors for the VXLIB_convolve_i8u_c16s_o8u function.

Method:

The following checks are made:

• There shall be no NULL pointers
• Output height should be == (Input height - (conv_height-1))
• Output width should be <= Input width
• The strides of each image shall be equal to or greater than the x dimension
• conv_width should be an odd number in the range 3 <= conv_width <= 9
• conv_height should be an odd number in the range 3 <= conv_width <= 9
• conv_scale should be a positive power of 2

Parameters

[in]	src[]	Pointer to array containing input image (UQ8.0)
[in]	src_addr[]	Pointer to structure containing dimensional information of src
[out]	dst[]	Pointer to array containing output image (UQ8.0)
[in]	dst_addr[]	Pointer to structure containing dimensional information of dst
[in]	conv_mat[]	Pointer to array containing convolution matrix (SQ15.0)
[in]	conv_width	Width of convoltion matrix
[in]	conv_height	Height of convolution matrix
[in]	conv_scale	Scale value to divide the final sum of products (power of 2)