Optimizing deep neural network inference workloads