Skip to content

hitobject-construct-outside-allowed-stages

Status: stub. The full-length analysis is queued for a v1.0.x patch release per ADR 0018, section 5, criterion #6. The companion rule page at docs/rules/hitobject-construct-outside-allowed-stages.md contains the canonical detection logic + GPU reasoning.

TL;DR

The SER programming model assumes that dx::HitObject is constructed at well-defined hardware points where the RT subsystem can hand a coherent traversal record back to the shader. On NVIDIA Ada Lovelace, those points correspond to the SM-side handoff slots that the RT cores expose; the hardware does not present the same handoff in any-hit (where the traversal is suspended mid-leaf, not committed) or intersection (where the procedural primitive's existence is still being evaluated). On future AMD and Intel SER implementations, the same hardware-scoped restriction applies.

What the rule fires on

A dx::HitObject::TraceRay, dx::HitObject::FromRayQuery, or dx::HitObject::MakeMiss constructor call from a stage other than the SM 6.9 SER spec's allowed set: raygeneration, closest-hit, and miss (with stage-specific restrictions per the spec's table). The rule reads the entry-point stage from Slang reflection and matches each constructor against the spec's allowed-stages set, firing when a call originates from any other stage (any-hit, intersection, callable, compute, mesh, amplification, vertex, pixel, etc.).

See the What it detects section of the rule page for the full pattern definition.

Why it matters

The full GPU-mechanism analysis lives in the Why it matters on a GPU section of the companion rule page.

Examples

The bad / good code snippets are kept canonical on the rule page; see hitobject-construct-outside-allowed-stages.md -> Examples.

See also


This is a v1.0-ship stub. Full analysis pending; track issue link TBD.

© 2026 NelCit — Apache-2.0 (code), CC-BY-4.0 (docs).