The Compilation phase is responsible for converting (aka compiling) a deep neural network into a sequence of hardware layers that are optimized for a given NVDLA configuration. Having a compiled network optimized for specific hardware configuration improves performance by reducing model size, load and run times. Compilation is a 2-step process that consists of: Parsing and Compiling.
But where is it?