| 1 | //! CUDA compute backend with fused PTX kernels. |
| 2 | //! |
| 3 | //! All fused operations dispatch to custom CUDA kernels compiled from |
| 4 | //! `fused_ops.cu` via PTX. GPU synchronization is a no-op on CUDA since |
nothing calls this directly
no outgoing calls
no test coverage detected