Venkat V. Easwar

Bandwidth efficient image transformations on a multiprocessor

Image transformation computations are typically the most computationally intensive operations in raster image processing. Raster image processing is the process of converting a page description language like postscript to a bitmap. Researchers have traditionally assumed a uniprocessor model when optimizing the execution time associated with image transformation. In the case of multiprocessing environments, in addition to computational speed, other factors like the frame-buffer access bottleneck, processor utilization, etc. have to be considered. A block based approach that takes these factors into consideration will be presented along with techniques to choose optimal block dimensions.We will focus on the single-chip multi-processing environment provided on Texas Instruments TMS320C80 which has a RISC processor and four Digital Signal Processors on the same chip. We will compare the performance of the block based approach with a conventional scan-line based approach.