Module backend 

OneApiExecutable — compile an IR graph for the Intel oneAPI Level Zero backend and execute it.

Two execution paths share one legalized graph (the rlx-vulkan primitive set, so the same rewrite/legalize decompositions apply):

• run_host — a value-map interpreter that evaluates every node through the rlx-cpu reference. This is the path the macOS dev box / CI take (no Level Zero device), and it makes the backend fully correct without Intel hardware.
• run_l0 — the native path: a USM-shared f32 arena + per-op SPIR-V kernel dispatch (with a CPU host-fallback, against the same arena, for ops with no native kernel yet). Selected only when a live device and embedded kernels are both present — neither is true off an Intel build host, so it is compiled-but-dormant here, pending hardware validation on Arc / Data Center Max.

Structs

OneApiExecutable

Constants

SUPPORTED_OPS

OpKinds this backend lowers (claim set). Identical to rlx-vulkan’s: the rewrite pass decomposes everything else into this primitive set, and the CPU reference covers every entry, so a legalized graph always executes.