GitLab

- Introduce TMAStoreArrive and TMAStoreWait operations for CUDA TMA store synchronization
- Add new builtin operations in op/builtin.cc and op/builtin.h
- Implement TMAStoreSyncInjector to automatically inject TMA store synchronization calls
- Update CUDA codegen to support new TMA store synchronization intrinsics
- Add Python language bindings for new TMA store synchronization operations