iro-cuda-ffi-kernels

Reference CUDA kernels for iro-cuda-ffi.

These kernels are compiled with nvcc at build time and demonstrate the expected ABI and wrapper patterns. They are intended as examples and integration tests, not as a production math library.

Notes:

• Requires a CUDA toolkit (nvcc) to build.
• Tests require --features cuda-tests.
• Benchmarks (including cudarc cross-validation) live in iro-cuda-ffi-benchmarks.
• iro-cuda-ffi vs cudarc benchmarks are sanity checks, not a competition.
• Run benchmarks in release mode and serially to avoid GPU contention.

Docs: https://docs.rs/iro-cuda-ffi-kernels