Progressive material caching

Abstract

With minimal preprocessing of trees of materials in textures, this approach progressively caches outputs of some material nodes that are evaluated while rendering a scene. A node may be the root of a subtree, and the node may generate a value of a texel based on values of other nodes in the subtree. The subtree of a cacheable node contains no shading-point dependent node as a descendant other than texture coordinates. When reaching the cacheable node during a material evaluation for rendering, a descriptor of the node is used as a key to store or reuse the texel value of the cacheable node in a materials cache in memory. By defining cacheable nodes at highest possible levels in a tree, repeated evaluation of most nodes in the tree is avoided. For multithreaded acceleration, a materials cache is designed for thread-safe operation with an atomic instruction.