preflight_device_arch

metrale_gpu_runtime::cuda_backend::arch_preflight

Function preflight_device_arch 

Source 
pub fn preflight_device_arch(
    ordinal: usize,
    compiled_arch: Option<&str>,
) -> Result<()>
Expand description

2026-09-25: Fail if kernels built for compiled_arch cannot run on GPU ordinal. Call it before MetraleCudaBackend::new, which loads the modules (init_gpu_backend does, through kernel_gate::gate_device_arch).

compiled_arch is None when the target records no architecture (preflight_arch); that is logged as a warning and returns Ok without a device query.