[cgmillette]

Motivation

The docs for trap() said a taken trap makes the launch report an error or abort the process, and that it lowers to s_trap 2. Neither holds everywhere, so a kernel author reading them could expect a recoverable error where there is none.

Technical Details

Docs and comments only; no emitted IR or HIP source changes.

• On Linux (ROCr) the queue reports an error and the process aborts.
• On Windows the HIP runtime never turns the trap into an error and the launch never returns.
• Under a debugger the wave stops at the trap.
• On gfx11 the AMDGPU backend emits its simulated trap: s_trap 2, then s_sendmsg, then an s_sethalt loop. Other targets get s_trap 2.

Updated: the IRBuilder.trap docstring, the tile.trap entry in dsl_docs/reference/op_vocabulary.md, the rocke_b_trap comment in cpp/include/rocke/ir.h, and the module docstring of tests/core/test_trap.py. Each now says that a check the process must survive belongs on the host, before the launch.

Test Plan

• Engine and bindings build; ctest.
• tools/check_byte_identity.py (default flavor plus the gfx1250 llvm23 pass).
• tests/core/test_trap.py with both engines.
• pre-commit over the branch.
• The gfx11 and other-target trap sequences were checked by compiling an llvm.trap kernel with llc for gfx90a, gfx942, gfx950, gfx1100 and gfx1201.

Test Result

All green; byte identity GREEN for every family.

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

ISSUE ID : AICK-2906

🤖 Generated with Claude Code

[therock-pr-bot]

✅ All Checks Passed — Ready for Review

Check	Status	Details
📝 PR Description	✅ Pass	—
⛔ Forbidden Files	✅ Pass	—
🧪 Unit Test	✅ Pass	—
🔎 pre-commit	✅ Pass	—
🚫 Draft PR	🔜 To Be Enabled	—
🚩 Feature Flag	🔜 To Be Enabled	—
📊 Code Coverage	🔜 To Be Enabled	—
🤖 therock-pr-bot	✅ Pass	—

🎉 All checks passed! This PR is ready for review.

📖 Need help? See the Policy FAQ for details on every check and how to fix failures.

🙋 Wish to Override Policy?

[therock-pr-bot]

🎉 All checks passed! This PR is ready for review.

[CongMa13]

No new issues found.

[CongMa13]

No new P0 or P1 issues were identified during this review. The changes are approved.

[cgmillette]

Automated review: Model - qwen3.6:35b-a3b-coding-mtp-q4_K_M, effort: max

PR #14866 — say what a taken trap does on each platform (cgmillette, AICK-2906) — 🟡 First-round: gfx11 overstated; CI running

Prior item	Status
CongMa13 (11-02) APPROVED: no P0/P1 issues	🟢 APPROVED bb8d18b8 • at this head; no items raised
Bots / CI	🟡 Running • running: Multi-Arch CI re-run after the org label; 3 jobs • fails: Multi-Arch CI Summary — cancelled run's, re-run pending (read) • 30 ✓ incl. hipDNN superbuilds · 6 cancelled
Our checks	🟢 Clean • merge vs 2a155d53 clean; rocke/ tree = head's (ran) • build + ctest 34 ✓ / 1 skip; byte identity GREEN ×4 flavors (ran) • goldens PASS, 0 changed; code tokens unchanged; test_trap 6 ✓ (ran)

#	Sev	file:line	Finding
N1	🟡	ir.py:5527, op_vocabulary.md:196, ir.h:1813, test_trap.py:7	"gfx11" is too broad: the simulated trap is gfx1100–gfx1103 only • impact: gfx1150–gfx1153 (e.g. gfx1151) emit plain s_trap 2 (ran) • Windows hang evidence is gfx110X CI only (read) • fix: write gfx110x (gfx1100–gfx1103) in all four files

Utility: 🟢 Positive — corrects per-platform trap docs; comments only, IR unchanged.

---

Legend: 🔴 blocking · 🟠 should-fix · 🟡 minor nit · 🟢 info/good · ⚪ retracted

First-round review: first pass by a local qwen3.6 model, verified by the orchestrator (Claude Opus 5.5). Checks: merge with develop, build + ctest, byte identity, goldens, comment-only token diff, test_trap, trap ISA per target, CI at head. Flags whether a closer look is needed; not an approval or rejection.

[assistant-librarian]

TheRock Submodule Bump Activity

Newest first

• 2026-11-03 — Included in TheRock via ROCm/TheRock#9698.