pub fn av1_encode_capable(dev: &GpuDevice) -> boolExpand description
Whether an AV1 encoder can actually be constructed for this device — the
authoritative, build-aware capability check. It runs the same
select_encoder dispatch a per-chunk worker uses, pinned to the device’s
vendor + index, so true means a worker leased to this GPU will encode
rather than hard-fail. Used to drop AV1-incapable cards (e.g. a pre-Ada
NVIDIA that decodes via NVDEC but has no AV1 encode silicon) from the
multi-GPU encode pool, so a mixed-vendor host encodes on the capable cards
instead of aborting when a chunk leases to an incapable one.
The probe constructs + immediately drops a real encoder, so the verdict is cached per GPU index (queried once per process).