Utilities

• Compile-time constant tensors
• Adding Torch Custom Ops in cuDNN Frontend
• CUDA Graphs
• Custom Execution Plan
• Deviceless Ahead-of-time Compilation
• Dynamic Shapes and Kernel Cache
• Framework integration performance: getting the best per-call performance from cuDNN Frontend
• First-class cudnn.Handle — design
• Python-native cudnn.pygraph and pluggable execution backends