Registering kernels#

class MicrosoftKernelImplementation#: Selects the complete com.microsoft implementation family. NAIVE uses independent scalar correctness oracles; OPTIMIZED uses production kernels.

register_kernels(*, microsoft_implementation=MicrosoftKernelImplementation.OPTIMIZED)#: Legacy name that registers all kernels process-wide. The former ignored sess argument was removed; use register_kernels_for_session() for local registration. New global code can use register_kernels_global() to make the scope explicit.

register_kernel_global(domain, op_type, *, replace=True, microsoft_implementation=MicrosoftKernelImplementation.OPTIMIZED) → bool#: Registers one native kernel in the process-wide dispatch table. Future sessions, and existing sessions that have not resolved the node yet, observe it. Already prepared sessions keep their cached kernel. replace=True replaces an existing factory; replace=False keeps it and returns False. Unknown domain/operator pairs raise ValueError.

register_kernels_global(*, replace=True, microsoft_implementation=MicrosoftKernelImplementation.OPTIMIZED) → int#: Registers all shipped kernels globally and returns the number installed.

register_kernel_for_session(sess, domain, op_type, *, replace=True, microsoft_implementation=MicrosoftKernelImplementation.OPTIMIZED) → bool#: Registers one native compiled kernel on one ReferenceEvaluator. The evaluator owns the registration for its lifetime; no other evaluator or the process-wide table is modified. The evaluator’s cached runtime session is reset so its next run observes the change.

register_kernels_for_session(sess, *, replace=True, microsoft_implementation=MicrosoftKernelImplementation.OPTIMIZED) → int#

Registers all shipped native kernels on one evaluator and returns the number installed. All four explicit APIs are idempotent in resulting state; replace=False also makes duplicate calls no-ops.

from onnx_light.onnx.reference import ReferenceEvaluator
from onnx_light_cpu import register_kernel_for_session, register_kernel_global

register_kernel_global("", "Abs")
local_sess = ReferenceEvaluator(model)
register_kernel_for_session(local_sess, "", "Gemm")

register_backend_test_cases() → None#: Registers the onnx-light-cpu test_cpu_* backend cases in onnx-light’s shared C++ backend test registry.

has_backend_test_cases() → bool#: Returns whether the register_backend_test_cases binding is available.

class BackendCaseResult#: A skipped or failed backend correctness case, including its operator, case name, and reason.

class BackendCorrectnessReport#: The executed and passed case counts and skipped and failed case results.

run_backend_correctness_tests(microsoft_implementation=MicrosoftKernelImplementation.OPTIMIZED) → BackendCorrectnessReport#: Registers onnx-light-cpu kernels and runs applicable onnx-light TestMode.TEST backend cases. Cases use onnx-light’s standard ReferenceEvaluator comparison with their declared tolerances. The report records unsupported cases as skips and execution or comparison errors as failures; a kernel without an applicable correctness case is a failure.