Runtime Dispatch¶
RL-Kernel routes operators through KernelRegistry. Callers request an operator by
logical type, and the registry selects the first available backend for the current device.
Dispatch Flow¶
- Detect platform from
device_ctx. - Load the priority list for the requested operator type.
- Try each backend in priority order.
- Cache successfully constructed operator instances.
- Skip backends that already failed in the current process.
WS2 Attention uses the stricter KernelRegistry.get_attention_op(contract) path. In addition to
platform priority, this path requires a backend capability descriptor and checks the requested
role, mode, dtype, TP/CP layout, LSE export, deterministic merge, packed varlen, and KV-cache
semantics. Incompatible candidates produce explicit rejection reasons and are never used as an
undeclared fallback. See WS2 CP-aware Attention contract.
WS2 TP-aware logprob uses the stricter KernelRegistry.get_logprob_op(contract) path. In
addition to platform priority, this path requires a backend capability descriptor and checks
the requested role, dtype, TP/CP layout, padded-vs-real vocab masking, inactive-token
support, vocab-domain LSE export, and deterministic TP merge semantics. Incompatible
candidates produce explicit rejection reasons and are never used as an undeclared fallback.
The contract objects and their normative reduction semantics are documented in
rl_engine.kernels.logprob_contract.
LogP Priority¶
| Platform | Priority |
|---|---|
| CUDA | CUDA generic LogP by default; experimental SM90 fused LogP only when explicitly enabled, FlashInfer, Triton generic, PyTorch native |
| ROCm | AITER, Triton generic, PyTorch native |
| CPU | PyTorch native |
For CUDA devices with compute capability 9.0 or newer, the registry only inserts
the legacy SM90 LogP backend when RL_KERNEL_ENABLE_EXPERIMENTAL_SM90_LOGP=1 is
set. The fused linear logp SM90 backend is gated separately and remains the
default linear logp backend when the extension is built on Hopper.
Relevant Files¶
rl_engine/kernels/registry.pyrl_engine/platforms/device.pyrl_engine/kernels/ops/