cuDLA is an extension of the CUDA programming model to encompass DLA work as well. It provides familiar CUDA programming model to work with DLA, reduces SW complexity (without need for EGLStreams) to interoperate between DLA – GPU for your inference pipelines.