Block-Based Lossless Compression of 3D Geometric Data

The patent divides geometric data into fixed-size compression blocks (e.g., 1024 bytes) that can be independently processed. Each block contains a header indicating the number of primitives and their data format, allowing the rendering system to load and decompress only the necessary blocks rather than handling all geometric data at once, thus reducing memory bandwidth consumption while maintaining rendering speed.

The patent performs compression of geometric data in advance before rendering. By pre-compressing the geometric data into compact blocks with efficient formats (such as using 16-bit integers instead of 32-bit floats where precision allows), the system reduces the amount of data that needs to be transferred from memory during rendering, thereby reducing memory bandwidth consumption while maintaining fast rendering performance.

The patent introduces a decompression unit as an intermediary component between memory and the rendering pipeline. This dedicated hardware unit automatically handles the decompression of geometric data blocks, transparently converting compressed data into the format required by the rendering engine. This intermediary approach reduces memory bandwidth consumption while preventing data access complexity from affecting the main rendering logic.

The patent extracts the compression and decompression operations from the main rendering pipeline into separate, dedicated processing units. By isolating these complex operations into specialized hardware blocks with their own control logic and data paths, the system reduces memory bandwidth consumption through efficient compression while containing the access complexity within dedicated modules rather than propagating it throughout the entire rendering system.

The patent segments geometric data into fixed-size compression blocks that can be processed in parallel. By organizing data into uniform blocks with standardized headers and formats, the system enables multiple blocks to be decompressed simultaneously by multiple processing units, thereby improving memory bandwidth efficiency while preventing decompression processing time from increasing linearly with scene complexity.

The patent employs multiple compression formats with different compression ratios and decompression complexities, selecting the appropriate format based on the specific geometric data characteristics. By changing compression parameters (such as precision requirements, block size, and format type) to match the scene requirements, the system improves memory bandwidth efficiency for complex scenes while maintaining acceptable decompression processing times by using simpler formats when appropriate.