pub fn compile_to_cuda(source: &str) -> Result<CudaArtifact, ParseError>Expand description
Compile to a single-file CUDA program — the LOGOS → CUDA AOT tier. Lifts the
C tier’s program body (compile_to_c, byte-for-byte ≡ the VM) onto one GPU thread
as __device__ code, over a printf-based Show preamble, plus a host launcher.
Uses the SAME Oracle-optimized bytecode to refuse anything outside the integer/bool
scalar fragment (a GPU thread has no dynamic C runtime), returning a Custom
ParseError — a sound refusal, never wrong code. Returns the emitted
crate::codegen_cuda::CudaArtifact.