VPC Function

Function Description

The VPC module provides the following functions:

Cropping: Crops a required area out of the input image.
Resizing
- In the VPC, image resizing can be classified into 8K image resizing and non-8K image resizing modes based on resolution:
  - 8K image resizing: used to process the input image whose width or height is within the range of 4096–8192 pixels
  - Non-8K image resizing: used to process the input image whose resolution is within the range of 32 x 6 pixels to 4096 x 4096 pixels
- Single-image cropping/resizing (supporting uncompressed format and HFBC compressed format) and single-image multi-ROI cropping/resizing (supporting uncompressed format and HFBC compressed format):
  HFBC is a compressed image format for the VDEC output. In this format, the VDEC has better processing performance.
- Other resizing modes: for example, original image resizing and proportional image resizing.
Overlaying: Crops an image out of an input image, resizes the cropped image, and places it in a specified area of the output image. The output image may be a blank image (when the output buffer allocated by the user is empty) or an existing image (when an image has been read into the output buffer allocated by the user). Note that the overlaying concept here refers only to the case when the output image is an existing image.
Stitching: Crops multiple images out of an input image, resizes the cropped images, and place them in a specified area of the output image.
Format conversion
- Converts an RGB, YUV422, or YUV444 image to a YUV420 image.
- Converts a color image to a grayscale image. For the output image, only the data of the Y component is used.

Restriction

The following table describes the 8K image resizing and non-8K image resizing modes in the VPC based on resolution.

Input Image Resolution	Input Image Format	Width Stride x Height Stride Alignment for the Input Image	VPC Function	Output Image Resolution	Output Image Format	Width Stride x Height Stride Alignment for the Output Image
Width or height within the range of 4096–8192 pixels (excluding 4096)	YUV420SP (NV12 and NV21)	2 x 2 alignment	8K image resizing	16 x 16 to 4096 x 4096	For details, see the outputFormat parameter in Table 1.	2 x 2 alignment
32 x 6 to 4096 x 4096 pixels (including 4096)	For details, see the inputFormat parameter in Table 1.	For details about the width stride alignment, see the widthStride parameter in Table 1. The height stride is 2-pixel aligned.	Non-8K image resizing	32 x 6 to 4096 x 4096	For details, see the outputFormat parameter in Table 1.	16 x 2 alignment

Resizing ratio of the width or height: [1/32, 16]
8K image resizing: Resizing is supported, the format conversion between YUV420SP NV12 and YUV420SP NV21 is supported, but cropping is not supported.
Buffer restrictions:
- The start address of the buffer must be 16-byte aligned.
- In the same task, the virtual addresses of the input and output buffers must be in the same 4 GB space.

VPC Function Diagram

Figure 1 VPC function diagram (cropping+resizing+overlaying)

Figure 2 VPC function diagram (stitching)

**Table 1** Concepts
Concept	Description
Width stride	Step for a row of pixels, indicating the width of the input image after alignment. The width stride calculation of an RGB image is different from that of a YUV image. The minimum width stride is 32, and the maximum width stride is 4096 x 4 (the image width is 4096 and the image format is ARGB). YUV400SP, YUV420SP, YUV422SP, and YUV444SP: Align the width of the input image to a 16-pixel boundary. YUV422 packed: Multiply the width of the input image by 2 and align the result to a 16-pixel boundary. YUV444 packed and RGB888: Multiply the width of the input image by 3 and align the result to a 16-pixel boundary. XRGB8888: Multiply the width of the input image by 4 and align the result to a 16-pixel boundary. HFBC format: The wide stride equals the width of the input image.
Height stride	Number of lines in the buffer of an image, indicating the height of the input image after alignment. Value: The height of the input image must be 2-pixel aligned. The minimum height stride is 6, and the maximum height stride is 4096.
Top/Bottom/Left/Right offset	You can configure the top offset, bottom offset, left offset, and right offset to implement the following functions: (1) Specify the position of the cropped area or overwritten area. (2) Control the width and height of the cropped area or overwritten area by using the following formulas: Right offset – Left offset + 1 = Width; Bottom offset – Top offset + 1 = Height. Left offset: horizontal offset of points 1 and 3 in the cropped/overwritten area relative to point 0 in the input/output image Right offset: horizontal offset of points 2 and 4 in the cropped/overwritten area relative to point 0 in the input/output image Top offset: vertical offset of points 1 and 2 in the cropped/overwritten area relative to point 0 in the input/output image Bottom offset: vertical offset of points 3 and 4 in the cropped/overwritten area relative to point 0 in the input/output image
Cropped area	Image area to be cropped. The minimum resolution is 10 x 6, and the maximum resolution is 4096 x 4096.
Overwritten area	Area specified by the user in the output image. The minimum resolution is 10 x 6, and the maximum resolution is 4096 x 4096. Restrictions: For the overwritten area, the left offset and the top offset must be even numbers, and the right offset and the bottom offset must be odd numbers. The cropped area cannot be larger than the input image, and the overwritten area cannot be larger than the output image. The overwritten area can be directly mapped on the leftmost side of the output image, that is, the left offset of the output image is 0. The maximum number of overwritten areas is 256. The left offset of the overwritten area relative to the output image is 16-pixel aligned. It is recommended that the width of the output overwritten area be 16-pixel aligned. If the width is not 16-pixel aligned, invalid extra data is written to make the width 16-pixel aligned.

Performance Specifications

For non-8K image resizing, the VPC performance involves different scenarios such as image cropping and resizing. When the resolution changes during image processing, the performance is calculated based on the larger resolution. For example, if the resolution of the image after resizing is greater than that before resizing, the former is used to calculate the performance specifications. If the resolution of the overwritten area is greater than that of the cropped area, the former is used to calculate the performance specifications. For YUV420 SP images, the performance specifications in typical scenarios are as follows:

Scenario	Total Frame Rate
1080p x n channels (n < 4, one channel corresponds to one thread)	n x 360 fps
1080p x n channels (n ≥ 4, one channel corresponds to one thread)	1440 fps
4K x n channels (n < 4, one channel corresponds to one thread)	n x 90 fps
4K x n channels (n ≥ 4, one channel corresponds to one thread)	360 fps

For 8K image resizing, the VPC performance is closely related to the output resolution. A higher output resolution indicates longer processing time and lower performance. The following table lists the performance specifications in typical scenarios (output resolution of 1080p or 4K and image format of YUV420 SP).

Scenario	Total Frame Rate
1080p x n channels (n = 1, one channel corresponds to one thread)	4 fps
1080p x n channels (n ≥ 4, one channel corresponds to one thread)	16 fps
4K x n channels (n = 1, one channel corresponds to one thread)	1 fps
4K x n channels (n ≥ 4, one channel corresponds to one thread)	4 fps

Reference

The following figure shows the component layout of the RBG and YUV images. Two YUV420SP images are used as examples for SP format images and an ARGB image is used as an example for packed and RGB images.