Neighbor context caching in block processing pipelines

A method, comprising:

• inputting blocks of pixels from an input frame to a block processing pipeline, wherein the pipeline includes a plurality of stages each configured to perform one or more operations on a block, wherein said inputting the blocks spaces the blocks in the pipeline so that adjacent blocks on a row are not concurrently at adjacent stages of the pipeline;
• wherein the input frame is subdivided into rows of blocks, wherein the rows of blocks are divided into a plurality of row groups each including two or more rows, wherein said inputting blocks of pixels comprises, after inputting a block from a bottom row of a row group, inputting a next block from a top row of the row group;
• performing, by each of one or more of the plurality of stages of the pipeline, for each block on the top row of a row group processed at the stage:
• receiving, from an upstream stage of the pipeline, information from a neighbor block on the bottom row of a previous row group;
• storing the information from the neighbor block to a first buffer in local memory of the stage, wherein said storing overwrites oldest neighbor block information in the first buffer;
• processing the block according to information from one or more neighbor blocks stored in the local memory;
• and
• outputting the processed block and the information from the neighbor block to the next stage in the pipeline or outputting the processed block to an external memory.

A device, comprising:

• a memory;
• and
• an apparatus configured to process video frames and to store the processed video frames as frame data to the memory, the apparatus comprising a block processing pipeline that implements a plurality of stages each comprising one or more pipeline units, each pipeline unit configured to perform one or more operations on a block of pixels passing through the pipeline;
• wherein each frame is subdivided into rows and columns of blocks of pixels, wherein the apparatus is configured to input the blocks of pixels from each frame to the block processing pipeline so that adjacent blocks on a row are not concurrently at adjacent stages of the pipeline;
• wherein one or more of the pipeline units in the block processing pipeline are each configured to:
• receive a block for processing at the pipeline unit;
• process the block according to information from one or more previously processed neighbor blocks stored in one or more buffers in a local memory for the pipeline unit;
• store information from the processed block to a first buffer in the local memory, wherein said storing overwrites oldest information from a previously processed block in the first buffer;
• and
• output information from the processed block to one or more other pipeline units in the pipeline.