Preferred Terminology

• Use “Hybridizer” to refer to the compiler toolchain; avoid generic “translator”.
• Refer to “managed IL (MSIL) / Java bytecode” as inputs; “native backends” as outputs.
• Use “CUDA backend”, “vector backends (AVX/AVX512/NEON/POWER)”, “OMP+CUDA”.
• Use “kernel” for GPU functions; “vectorized function” for CPU SIMD.
• Use “data marshalling” for host↔backend data movement and transformation.