record(BACKEND-ROCM): preserve the gfx1100 launch evidence

[VikashLoomba]

Row

BACKEND-ROCM — one records-only unit for issue #2427.

Historical upstream issue #2164 is deleted or unavailable. It remains only as attribution for the original integration request; issue #2427 owns the live landing.

Before starting

• Issue Land the GFX1100-TG200 historical records #2427 is open. Its first body line is exactly Row: BACKEND-ROCM, with BACKEND-ROCM delimited by backticks.
• PR perf(GFX1100-TG200): the 200 tok/s gfx1100 decode campaign and fp8 KV cache decode attention #1936 remains the immutable campaign source at 3a345b5ae5df7cf08f1383b6623b38db9a1335bd. It mixes product changes with the historical records, so this split remains records-only.
• BACKEND-ROCM remains ACTIVE. This pull request changes no product code, checker, test, public document, or live benchmark index.

What changed

Import the complete GFX1100-TG200 campaign evidence and exact gate prompt from PR #1936. The specification now names issue #2427 as the live landing owner and labels #2164 as historical and unavailable. The T5 record now distinguishes the fresh-build 841 assertions across 19 cases from the old 132,094 count read from a stale ROCm 7.14-era binary.

Closes #2427

Evidence

• Final tracked head 6e25501d029a6efebcf44733a4ad5dea7eec02ab contains origin/main ffab931668f650cb79e9bf044ff1da34b7d11863 as an ancestor. The integration merges had no conflicts or manual resolutions.
• Source-identity verification proved 15 evidence files plus tools/tg200-prompt.txt byte-identical to 3a345b5ae5df7cf08f1383b6623b38db9a1335bd. The three reconciled records are the spec, T20, and T5.
• Source commit 09da0553c880a9233dc80aba26ae8aab97aaa825 records the fresh-build 841-assertion, 19-case result and identifies the earlier 132,094 claim as stale-binary provenance.
• The semantic check was red before the edits for the issue/spec/T5/body invariants. Twenty-five independent negative mutations then failed for the intended guarantee and restored every fixture byte-for-byte.
• scripts/check-agent-record.py, scripts/check-symbol-anchors.py, scripts/check-conflict-markers.py, scripts/check-benchmark-index.py, git diff --check, the commit trailer check, and the commit style check exited 0 on that head.
• .agents/completed/issue-index.md stayed at blob 199a10828f22806a6f0fc3b89b01604bafb849a1, matching the frozen PR head and current origin/main. Its historical gfx1100 decode is launch-bound: 15 levers took it to ~103 tok/s and a 3.13x kernel win moved the engine 0.1%, so the next lever must remove dispatches #2164 row is unchanged.
• The landed No gate compiles anything before a push: main landed twice today in a state that does not build #2401 repair removed the ENG-PREFLIGHT-COMPILES baseline failure. Exact head 6e25501d029a6efebcf44733a4ad5dea7eec02ab passed the agent-record, symbol-anchor, conflict-marker, benchmark-index, gate-command, 59-case gate-command mutation, PR-size, commit-trailer, and commit-style gates before its normal push.
• Fresh independent review of exact head 6e25501d029a6efebcf44733a4ad5dea7eec02ab returned findings NONE and verdict PASS. Its full preflight passed every changed-surface gate; only unchanged test_check_windows_portability hit ENOSPC while the filesystem was full, then passed 101/101 on the required serial rerun. Separate PR-size verification passed. Exact-head GitHub checks remain queued; no CI success is claimed.

Speed claims

• This pull request makes no current speed claim. It preserves attributed historical measurements from an unmerged campaign and labels them as such.

Honest gaps

No product change from PR #1936 is included. The fresh current-position rocprofv3 capture and the next dispatch-removal implementation remain campaign work, not part of this records landing. No GPU or oracle run applies to this records-only import.

The exact-head full preflight is not reported as green because of the unchanged Windows-portability suite's transient ENOSPC result. Its 101-case serial rerun passed, and fresh immutable review returned PASS. No completed GitHub check result is claimed yet.

FOLLOWING_AGENTS_PROTOCOL

Following-Agents-Protocol: true
AI-Assisted: true
Assisted-by: AGENT:gpt-5 [codex]

[VikashLoomba]

The records-only head 88b1b1bc8 completed independent immutable review with findings NONE and verdict PASS. The operator preflight ended directly with All gates green. All applicable CI jobs passed except the two Windows jobs.

Both Windows jobs failed the same OpenAI API cases 56, 57, and 59 with status -1073740791. PR #2267, based on the same a6f933b8 main SHA, produced the identical CPU and Vulkan failures, and this PR changes no product or test path.

I validated the live PR body with scripts/agent-pr-body.py --pr 2270. A squash merge was attempted, but GitHub refused it because VikashLoomba lacks MergePullRequest permission. Maintainer merge is the remaining external gate.

[localai-org-maint-bot]

Approved exact head 6e25501d029a6efebcf44733a4ad5dea7eec02ab.

Fresh immutable review returned findings NONE and verdict PASS after complete static inspection and mutation of the imported-record guarantees. The 19-path diff is records and the campaign prompt only. Issue #2427, spec ownership, source identity, the 841/19 correction, T20 arithmetic, T21 adoption, T5 provenance, and the retired-index non-change agree.

Operator focused gates passed: agent records, 431/431 local symbol anchors, conflict markers, benchmark index, gate-command checker and its 59-case suite, PR size, commit trailers, and commit style. The review's full preflight passed every changed-surface gate. Its only red was ENOSPC in the unchanged Windows-portability suite while the filesystem was full; the same 101-case suite then passed serially. The PR body passes scripts/agent-pr-body.py and does not claim queued CI as success.