Real-time volume rendering system

摘要

A volume rendering processor processes volume data in sections. Within each section, groups of volume elements or "voxels" are read and processed in parallel. Each group in a beam of voxels is read and processed until a section boundary in a first dimension is reached, each beam in a slice of voxels is read and processed until a section boundary in a second dimension is reached, and each slice of voxels is read and processed until a boundary of the volume data set in the third dimension is reached. The section boundary in either the first or second dimension may coincide with a boundary of the volume data set in the same dimension. Rendering pipelines are associated with respective voxels within each group, and each pipeline calculates intermediate values for all the associated voxels within the volume data. Within each pipeline, intermediate values for use in subsequent calculations for voxels associated with the pipeline are retained. Intermediate values to be used in calculations for voxels associated with a neighboring pipeline are transferred to the neighboring pipeline, and intermediate values calculated by a neighboring pipeline to be used for calculations for voxels associated with the one pipeline are received from the neighboring pipeline. During processing of the boundary portion of each section, intermediate values are stored into a memory, and values stored in the memory for a preceding section are retrieved and used for calculations within the processor. A section memory has two independent banks that are alternately read and written to. First-in-first-out (FIFO) buffers are employed to store section data when the section memory banks are switching between reading and writing operations.