pub const fn cuda_capabilities() -> &'static [OperationCapability]Expand description
Return the CUDA backend operation capability descriptor table.
The table states what this backend serves natively; it is written down rather than derived
from the kernel catalogue’s accepts list, and tests/seam_capability.rs holds each entry to
the engine’s observed behaviour. docs/guides/devices-and-gpu.md renders the same table and
a default-features test checks the two agree.
§Examples
use tenferro_core_ops::PrimitiveOpKind;
use tenferro_gpu::cuda::cuda_capabilities;
use tenferro_tensor::{DType, SupportLevel};
let add = cuda_capabilities()
.iter()
.find(|entry| entry.op == PrimitiveOpKind::Add && entry.dtype == DType::I32)
.unwrap();
assert_eq!(add.result, SupportLevel::Native);