[hipDNN] Ragged tensor frontend, backend and schema

[jerehartAMD]

Motivation

This PR is based on RFC 0014, adding in ragged tensor support to the hipDNN frontend, backend and flatbuffer schema. It introduces the isRaggedTensorEnabled parameter to graphs to reject plugins without ragged tensor support.

In addition, it adds the alignment parameter to TensorAttributes, setting byte alignment requirements for

Technical Details

• Schemas: tensor_attributes.fbs gains ragged_offset_tensor_uid (null = non-ragged) and alignment (default 16); graph.fbs gains is_ragged_tensor_enabled; execution_plan.fbs gains tensor_alignments[]
  (parallel to tensor_uids). All additive/defaulted ⇒ wire-compatible.
• C-API attributes: ..._IS_RAGGED_TENSOR_ENABLED_EXT (610), ..._TENSOR_BYTE_ALIGNMENT (1308), ..._TENSOR_RAGGED_OFFSET_DESC (1309); wired through GraphDescriptor and TensorDescriptor.
• Frontend: TensorAttributes gets set/get_ragged_offset, has_ragged_offset, set/get_alignment (validate rejects alignment < 1); Graph auto-detects the ragged flag by scanning node tensors at build time.
• Plugin versioning: engine ABI minor bumped 1→2; ragged graphs require plugin ≥ 1.2.0, override graphs ≥ 1.1.0, else baseline. Below-floor plugins filtered from the applicable set.
• Alignment enforcement: executeOpGraph checks each device pointer against the plan's per-UID alignment before dispatch (BAD_PARAM on violation); skips null pointers and unrecorded UIDs.
• Plan back-compat fix: PLAN_SERIALIZATION_VERSION 1→2; deserialize now accepts any known version and requires tensor_alignments only for v2+ (v1 plans load with empty alignments). Fixes the two deserialize
  regressions.
• JSON: round-trips the new fields, defaulting alignment to 16.

Unlike the ragged surface, alignment enforcement is not gated behind a feature flag — it runs on every executeOpGraph. Every tensor now carries a default alignment of 16 bytes, so any supplied device
pointer that isn't 16-byte aligned will fail execute with HIPDNN_STATUS_BAD_PARAM. Reviewers should confirm no in-tree path relies on sub-16-byte-aligned buffers.

Note: This PR can be updated to also give alignment its own graph enablement flag, if that's desired

Compatibility

All schema additions are defaulted/additive (no ABI break); plan format is now version-aware. Existing plugins keep serving non-ragged/non-override graphs; ragged graphs require a 1.2.0+ plugin.

Test Plan

~1,000+ lines added: unit coverage for descriptor plumbing, alignment enforcement, ragged version gating, and plan round-trip (incl. legacy); frontend and JSON round-trip tests; and a new integration test
that lowers a ragged graph through the real backend and asserts the flag lands in the serialized image.

Test Result

Old and new tests pass

Submission Checklist

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

[codecov-commenter]

Codecov Report

❌ Patch coverage is 91.23506% with 22 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
.../hipdnn/frontend/include/hipdnn_frontend/Graph.hpp	66.67%	7 Missing and 3 partials ⚠️
...ects/hipdnn/backend/src/BackendEnumStringUtils.hpp	0.00%	3 Missing and 3 partials ⚠️
...ipdnn/backend/src/descriptors/TensorDescriptor.cpp	95.65%	1 Missing and 1 partial ⚠️
...backend/src/plugin/EnginePluginResourceManager.cpp	95.35%	0 Missing and 2 partials ⚠️
...clude/hipdnn_frontend/detail/DescriptorHelpers.hpp	80.00%	0 Missing and 1 partial ⚠️
...end/include/hipdnn_frontend/detail/GraphPacker.hpp	85.71%	0 Missing and 1 partial ⚠️

❌ Your project status has failed because the head coverage (77.89%) is below the target coverage (80.00%). You can increase the head coverage or adjust the target coverage.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #8728      +/-   ##
===========================================
+ Coverage    71.27%   71.43%   +0.16%     
===========================================
  Files         2613     2612       -1     
  Lines       409461   407441    -2020     
  Branches     61203    60802     -401     
===========================================
- Hits        291826   291041     -785     
+ Misses       96184    95103    -1081     
+ Partials     21451    21297     -154     

Flag	Coverage Δ		*Carryforward flag
TensileLite	76.83% <ø> (+0.16%)	⬆️	Carriedforward from 34b245e
hipBLAS	90.81% <ø> (ø)		Carriedforward from 34b245e
hipBLASLt	41.36% <ø> (-0.01%)	⬇️	Carriedforward from 34b245e
hipCUB	82.68% <ø> (ø)		Carriedforward from 34b245e
hipDNN	86.54% <91.24%> (+0.02%)	⬆️
hipFFT	50.79% <ø> (+0.62%)	⬆️	Carriedforward from 34b245e
hipRAND	76.12% <ø> (ø)		Carriedforward from 34b245e
hipSOLVER	69.18% <ø> (ø)		Carriedforward from 34b245e
hipSPARSE	86.55% <ø> (ø)		Carriedforward from 34b245e
rocBLAS	48.08% <ø> (ø)		Carriedforward from 34b245e
rocFFT	47.22% <ø> (+0.93%)	⬆️	Carriedforward from 34b245e
rocRAND	57.07% <ø> (ø)		Carriedforward from 34b245e
rocSOLVER	77.89% <ø> (+0.97%)	⬆️	Carriedforward from 34b245e
rocSPARSE	72.37% <ø> (ø)		Carriedforward from 34b245e
rocThrust	91.34% <ø> (-0.01%)	⬇️	Carriedforward from 34b245e

*This pull request uses carry forward flags. Click here to find out more.

Files with missing lines	Coverage Δ
...ackend/src/descriptors/ExecutionPlanDescriptor.cpp	86.38% <100.00%> (+1.26%)	⬆️
...hipdnn/backend/src/descriptors/GraphDescriptor.cpp	92.89% <100.00%> (+0.42%)	⬆️
...de/hipdnn_flatbuffers_sdk/utilities/json/Graph.hpp	90.59% <100.00%> (+0.17%)	⬆️
...latbuffers_sdk/utilities/json/TensorAttributes.hpp	55.33% <100.00%> (+7.68%)	⬆️
...de/hipdnn_frontend/attributes/TensorAttributes.hpp	100.00% <100.00%> (ø)
...sdk/include/hipdnn_plugin_sdk/engine_api_version.h	100.00% <ø> (ø)
...clude/hipdnn_frontend/detail/DescriptorHelpers.hpp	88.54% <80.00%> (-0.28%)	⬇️
...end/include/hipdnn_frontend/detail/GraphPacker.hpp	88.30% <85.71%> (-0.21%)	⬇️
...ipdnn/backend/src/descriptors/TensorDescriptor.cpp	87.60% <95.65%> (+0.83%)	⬆️
...backend/src/plugin/EnginePluginResourceManager.cpp	83.60% <95.35%> (+0.67%)	⬆️
... and 2 more

... and 330 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[therock-pr-bot]

❌ PR Check — Action Required

Check	Status	Details
🌿 Branch Name	✅ Pass	—
📝 PR Title/Description	❌ Fail	Error: Title does not follow Conventional Commits style. Expected: start with a valid type (feat, fix, docs, …). Desired format: type(optional-scope): short description ─── Error: PR description must reference a JIRA ID, ISSUE ID, or a GitHub closing keyword. Expected: include a JIRA ID / ISSUE ID line (separator : or -, or omitted; value may be a JIRA key, a number with/without #, or a link), OR a closing keyword + issue reference. Accepted examples: • JIRA ID : TESTAUTO-6039 • JIRA ID - #330 • JIRA ID #330 • ISSUE ID : TESTUTO-3334 • ISSUE ID #3334 • ISSUE ID - TESTAUTO-3433 • ISSUE ID : https://github.com/<org_name>/<repo_name>/issues/1234 • Closes #10 • Fixes octo-org/octo-repo#100 • Resolves: #123 • #123 • https://github.com/<org_name>/<repo_name>/issues/123 Current: no valid JIRA/ISSUE/closing-keyword reference found
⛔ Forbidden Files	✅ Pass	—
🧪 Unit Test	❌ Fail	Error: Source/code files changed without an accompanying unit test. Expected: add at least one test file named like test_<name>.py / test_<name>.cpp (or <name>_test.*). Current: code file(s) changed: projects/hipdnn/backend/include/HipdnnBackendAttributeName.h, projects/hipdnn/backend/src/BackendEnumStringUtils.hpp, projects/hipdnn/backend/src/descriptors/ExecutionPlanDescriptor.cpp, projects/hipdnn/backend/src/descriptors/ExecutionPlanDescriptor.hpp, projects/hipdnn/backend/src/descriptors/GraphDescriptor.cpp (+25 more); no test file found
🔎 pre-commit	✅ Pass	—
🚫 Draft PR	🔜 To Be Enabled	—
🚩 Feature Flag	🔜 To Be Enabled	—
📊 Code Coverage	🔜 To Be Enabled	—

⚠️ 2 policy check(s) failed. Please address the issues above before this PR can be Reviewed.

🚫 Please fix the failed policies

• ❌ PR Title/Description
• ❌ Unit Test

The Not ready to Review label was added to this PR. Once all policies pass, the label is removed automatically.

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

[therock-pr-bot]

🚫 Please fix the failed policies before requesting reviews.

The following policy checks failed:

• ❌ PR Title/Description
• ❌ Unit Test

The Not ready to Review label has been added to this PR.
Once all policies pass, the label will be removed automatically.