Skip to content

docs(rfc): extend RFC 0015 engine support claims for bundle sweeps#9296

Draft
adickin-amd wants to merge 2 commits into
developfrom
users/addickin/rfc0015_bundle_sweeps
Draft

docs(rfc): extend RFC 0015 engine support claims for bundle sweeps#9296
adickin-amd wants to merge 2 commits into
developfrom
users/addickin/rfc0015_bundle_sweeps

Conversation

@adickin-amd

@adickin-amd adickin-amd commented Jul 10, 2026

Copy link
Copy Markdown
Contributor

Summary

Extends RFC 0015: Per-Graph Engine Support Claims to cover the template-sweep ("bundle sweep") bundles introduced by RFC 0011, so support claims can attach per sweep case rather than only per single-graph bundle. This unblocks pinning engine-support contracts for sweep-defined graphs, which the integration-test suite already exercises today.

JIRA ID : ALMIOPEN-2303

Risk Assessment

Minimal risk. Documentation-only change to a single RFC markdown file; no product code, build configuration, public API, schema, or tests are affected.

Testing Summary

  • Markdown structure validation: verified fenced code blocks stay balanced and cross-references/section anchors remain consistent after the edits.

Testing Checklist

  • Markdown fence/structure check - balanced code fences, no dangling section references - Status: Passed
  • PR CI - GitHub PR checks - Status: Pending

Technical Changes

  • Generalizes the claim key from "single-graph bundle" to "claimed graph," covering both {Name}.json bundles and template-sweep cases expanded from graph.template.json via sweep.json (§1, §4).
  • Adds §5.4 defining a sweep's machine-owned support.json: keyed engine-first, then an array of { cases, support } groups so a common support footprint compresses across many cases with outliers as separate entries; case ids must exist in sweep.json, each claimed at most once per engine.
  • Drops the unsupported verdict in favor of presence-based supported claims (platform lists per arch); absence uniformly means "not enforced," never "asserted-declined."
  • Restores the end-of-run "unclaimed support" summary for claim-bearing bundles and fixes the resulting cross-references (§5.2, §7.1, §7.4, §13.4).
  • Threads sweep handling through discovery, enforcement_level location (per-case cases[].metadata), the write tool (flatten→overlay→regroup canonical writes), pre-commit validation, support-matrix rendering, and the CI observation harvest (§4, §6, §9, §11, §12, §16).

Add per-case support claims for RFC 0011 template-sweep bundles: a
co-located support.json keyed engine-first, then grouped by shared
support footprint (cases + supported arch/platforms), so a common set
and outliers compress cleanly. Drop the unsupported verdict in favor of
presence-based supported claims (absence = not enforced), and restore
the end-of-run unclaimed-support summary for claim-bearing bundles.
|------------|-----------------|
| A graph that a bundle claims `supported` loses engine support for a claimed `(engine, arch, platform)`. | The claimed engine is *declined* — absent from the resolved support query's ranked list (`GRAPH_NOT_SUPPORTED` when it is the sole/last engine) → **FAIL** (replaces the silent `GTEST_SKIP`). See §5.2 / §7.1 for how "declined" is distinguished from "errored". |
| A graph with a `supported` claim hits an unresolved query (build failure, infrastructure fault) before its verdict can be evaluated. | The claim could not be assessed → **FAIL** ("errored before the claim could be checked; fix the error first"). Distinct from the *declined* verdict above, which is the real-regression case. |
| A graph claimed `unsupported` unexpectedly becomes supported. | Live support query returns supported → **note** ("claimed support != actual support"), not a failure. |

@adickin-amd adickin-amd Jul 10, 2026

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

i removed this because there is no point in putting unsupported in the files. if its absent, its not supported.

@therock-pr-bot

therock-pr-bot Bot commented Jul 10, 2026

Copy link
Copy Markdown

✅ All Checks Passed — Ready for Review

Check Status Details
🌿 Branch Name ✅ Pass
📝 PR Title/Description ✅ Pass
Forbidden Files ✅ Pass
🧪 Unit Test ✅ Pass PR does not contain code files — Unit Test auto-passed
🔎 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.

@therock-pr-bot

Copy link
Copy Markdown

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

Condense the sweep-related prose without changing any rules: merge the
discovery/claim-key bullets, drop duplicated 'no new mechanism' and
preservation wording shared with 9.2, and fold 5.4's engine-key and
claim-group bullets together.
"MIOPEN_ENGINE": {
"gfx942": { "linux": "supported", "windows": "supported" },
"gfx90a": { "linux": "unsupported" }
"gfx942": ["linux", "windows"],

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think I like this better, going to keep this change.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant