[kuhar]

Motivation

Make ISA targets statically pluggable and simplify integration of downstream NPI targets without adding them to central target switches.

Technical Details

• Add a constexpr descriptor to each ISA that owns its names, public and ELF bindings, decoder factory, and execution capability.
• Compose one validated immutable registry per final image from its selected provider headers, and use it for decoder creation and code-object lookup.
• Add canonical-name decoder creation with recoverable unavailable-target errors.
• Remove fine-grained gfx1250 execution registration markers and replace them with one generated immutable callback table while preserving model-only decoding.

Issue Tracking

N/A

Test Plan

Run ctest and pytest.

Test Result

Passed.

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/rocm-systems/blob/develop/CONTRIBUTING.md.

[Copilot]

Copilot wasn't able to review this pull request because it exceeds the maximum number of lines (20,000). Try reducing the number of changed lines and requesting a review from Copilot again.

[therock-pr-bot]

❌ PR Check — Action Required

Check	Status	Details
🌿 Branch Name	✅ Pass	—
📝 PR Title/Description	❌ Fail	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 • JIRA ID (on separate line) ROCM-25757 • ISSUE ID : TESTUTO-3334 • ISSUE ID #3334 • ISSUE ID - TESTAUTO-3433 • ISSUE ID (on separate line) AIRUNTIME-2352 • 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	✅ Pass	—
🔎 pre-commit	✅ Pass	—
🚫 Draft PR	🔜 To Be Enabled	—
🚩 Feature Flag	🔜 To Be Enabled	—
📊 Code Coverage	🔜 To Be Enabled	—

⚠️ 1 policy check(s) failed. Please address the issues above before this PR can be Reviewed.

🚫 Please fix the failed policies

• ❌ PR Title/Description

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

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

[amd-arosa]

I sent you two patches to align with the style guide.
There might be something funny happening with create_decoder_for_target, but it was pre-existing.
but other than that LGTM.

[amd-eochoalo]

I think this is good. Not sure if it would be interesting / good to link kTargetDescriptor to isa_properties somehow in the future (but probably not worth while on this PR at the moment).

[atgutier]

Approved with a minors from the agent:

MINOR

[MINOR] emulation/rocjitsu/lib/rocjitsu/src/rocjitsu/isa/arch/amdgpu/
gfx1250/CMakeLists.txt:71 — Full gfx1250 provider drops its object
libraries’ link requirements

Reasoning: rocjitsu_isa_gfx1250 exposes the model and execution objects
through $<TARGET_OBJECTS:...>, but it does not link or otherwise propagate
the underlying object-library targets. Those targets receive util and
simdojo_headers requirements through rj_add_object_library().

This differs from normal object-library providers: the registry helper both
includes their objects and links the provider target so its usage
requirements propagate. Because the full gfx1250 provider is an
INTERFACE_LIBRARY, the registry helper cannot discover the underlying model
and execution targets. Current broad in-tree consumers independently link
the missing dependencies, masking the composition gap. A narrow or
downstream consumer can therefore depend on undocumented transitive setup
or fail to link as these object libraries acquire additional non-header-
only dependencies.

Evidence:

• gfx1250/CMakeLists.txt:35-64 creates the model and execution object
  libraries through rj_add_object_library().

• gfx1250/CMakeLists.txt:71-77 forwards only their object files.

• cmake/rj_add_object_library.cmake:27-30 attaches common dependencies to
  each object library.

• cmake/rj_add_object_library.cmake:101-104 extracts objects only when the
  registered provider is itself an object library.

• cmake/rj_add_object_library.cmake:134-139 republishes the full interface
  provider, whose interface currently contains no link dependencies.

• docs/isa-target-providers.md:66-87 presents the provider/registry
  combination as the consumer-facing composition boundary.

Suggested fix: retain the explicit $<TARGET_OBJECTS:...> entries, but
propagate both underlying targets’ usage requirements:

target_link_libraries(
rocjitsu_isa_gfx1250
INTERFACE
rocjitsu_isa_gfx1250_model
rocjitsu_isa_gfx1250_exec
)

Alternatively, extend the provider helper to record and forward
implementation dependencies. Add a narrow final-image link test that
consumes the full gfx1250 registry without also inheriting the simulator’s
broad dependency list.

Confidence: medium-high. The usage-requirement propagation gap is definite;
existing production targets currently mask it, so I did not reproduce an
in-tree link failure.