This talk presents a scalable method for exposing and exploiting hidden localities in production stencil applications. An end-to-end framework automatically transforms stencil-based GPU programs to exploit inter-kernel data locality. The CUDA-to-CUDA transformation collectively replaces the user-written kernels by auto-generated kernels optimized for data reuse. The transformation is based on two basic operations, kernel fusion and fission, and relies on a series of automated steps: gathering metadata, generating graphs expressing dependencies and precedency constraints, searching for optimal kernel fissions/fusions, and generation of optimized code. We show how the automatic transformations were practical and effective in exploiting exposed data localities for a variety of real-world applications with large codebases that contain dozens of kernels and data arrays.