Render optimization using page alignment techniques

Abstract

To reduce the performance penalty associated with frame buffer memory access times for each page when rendering a primitive having scan lines which cross page boundaries, rendering is constrained to a single page at a time. All pixels mapping to a currently-cached frame buffer page are rendered before loading a different frame buffer page into the cache in order to render other pixels within the primitive. Any pixels within a scan region which map to a different page than the active page are temporarily skipped until all pixels mapping to the current page are completed. Only when no more pixels require rendering within the primitive which map to the currently active frame buffer page is another frame buffer page loaded and all pixels mapping to that page are rendered. In processing a scan region which crosses a page boundary, the next pixel or pixel group to be rendered is examined to determine if it maps to the currently active frame buffer page. If so, it is rendered; if not, however, the position is latched, together with any other necessary state information, and rendering proceeds with the next scan region. Once all pixels for a currently active page are rendered, the latched position is loaded, together with the frame buffer page necessary to render the corresponding pixels, and rendering of the primitive proceeds until complete.