♻️ Generalize MLIR compiler target facts by burgholzer · Pull Request #2218 · munich-quantum-toolkit/core

burgholzer · 2026-08-23T17:32:58Z

🤖 AI text below 🤖

Description

Define a complete, immutable MLIR compiler-target contract before serializing it as a typed MQT dialect attribute.

Require connectivity to be all-to-all or an explicit coupling graph.
Require native-operation support to be unrestricted or explicitly enumerated.
Reject missing or unrepresentable QDMI target facts during target inference.
Represent operation arity as an exact width or a positive variadic minimum.
Represent gphase as a fixed zero-arity operation and preserve its observable phase effects during native synthesis.
Recognize structurally controlled QCO operations through variadic base-gate capabilities.
Let the bundled DDSIM device advertise arbitrary positive controls for every canonical standard gate with one or more targets, including multi-target gates, through one exact QDMI 1.3 compatibility marker.
Support homogeneous fixed operations above two qubits without treating format-level constructs as compiler-target operations.
Use deterministic placement for all-to-all targets and mapping for explicit coupling graphs.
Preserve the full nonnegative site-ID domain and cheap shared target snapshots.

The DDSIM custom property is private to MQT Core and can be removed when QDMI standardizes the same operation capability. Controlled aliases such as mcx and mcp remain available to existing adapters but are not enumerated as compiler-target capabilities.

This PR is the bottom layer of the stack. #2323 adds target-aware controlled-operation decomposition, and #2215 adds typed target serialization on top of it. This PR does not add payload capabilities, replaceable-driver work, or a dependency on the QDMI v1.4 development API.

Target synthesis relies on the canonical QCO cleanup pipeline to remove empty control modifiers. Doxygen comments in every C++ file touched by this PR consistently use ///.

Validation

Compiler unit tests: 143 passed.
Mapping unit tests: 86 passed.
Native-synthesis unit tests: 26 passed.
DDSIM device tests: 52 passed.
Focused Python MLIR and Qiskit translation tests: 279 passed with the bundled superconducting-device catalog selected explicitly.
Generated Python stubs are current.
The release build passes.
Full CTest reached 4,156 passes and one expected skip; its only failure was the obsolete CUSTOM1 unsupported-property probe. The corrected isolated test passes.
Sphinx documentation passes with warnings treated as errors.
uvx nox -s lint passes.
uvx nox -s cpp-lint -- origin/main reports no clang-format or clang-tidy findings.
git diff --check passes.
An independent final audit found no concrete issues.

All 13 commits were rebased onto current main, signed, and verified. Hosted checks for the new head are pending.

Implementation, validation, and pull-request text were prepared with OpenAI Codex.

Checklist

The pull request only contains commits that are focused and relevant to this change.
I have added appropriate tests that cover the new/changed functionality.
I have updated the documentation to reflect these changes.
I have added entries to the changelog for any noteworthy additions, changes, fixes, or removals.
I have added migration instructions to the upgrade guide (if needed).
The changes follow the project's style guidelines and introduce no new warnings.
The changes are fully tested and pass the CI checks.
I have reviewed my own code changes.

If PR contains AI-assisted content:

Any agent that created, edited, or submitted GitHub content was explicitly authorized for that scope, as required by our AI Usage Guidelines.
Every agent-authored or agent-edited public text body begins with the visible disclosure 🤖 *AI text below* 🤖 (titles are exempt).
I have disclosed AI assistance in the PR description.
I confirm that I have personally reviewed and understood all AI-generated content, and accept full responsibility for it.

codecov · 2026-08-23T17:40:22Z

Codecov Report

❌ Patch coverage is 93.76694% with 23 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
mlir/lib/Compiler/QDMIAdapter.cpp	90.1%	11 Missing ⚠️
mlir/lib/Compiler/Target.cpp	94.7%	9 Missing ⚠️
...QCO/Transforms/NativeSynthesis/TargetSynthesis.cpp	82.3%	3 Missing ⚠️

📢 Thoughts on this report? Let us know!

burgholzer

This one feels close to ready.

coderabbitai · 2026-08-24T12:40:06Z

Important

Approval pending

CodeRabbit has no unresolved comments, but it has not reviewed the latest commit.

Use the checkbox below to review the latest commit. CodeRabbit will approve the changes if it finds no blocking issues.

🔍 Trigger review

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

_{Comment @coderabbitai help to get the list of available commands.}

denialhaag

I just had a free minute to go through this, and it looks really clean to me!

simon1hofmann

This also looks quite good, just a few comments about the documentation and two concerns flagged by another LLM review run.

burgholzer · 2026-08-25T09:41:09Z

Thanks @denialhaag and @simon1hofmann 🙏🏼
Given how this is at the very bottom of an 11 PR stack, I'd like to only touch this once more and freeze the implementation afterwards. Are there any further comments on the above from your side?
Quick signal would be good. Then I'll kick off the implementation and the rebase.

simon1hofmann · 2026-08-25T10:16:33Z

Thanks @denialhaag and @simon1hofmann 🙏🏼

Given how this is at the very bottom of an 11 PR stack, I'd like to only touch this once more and freeze the implementation afterwards. Are there any further comments on the above from your side?

Quick signal would be good. Then I'll kick off the implementation and the rebase.

Thanks for addressing my comments, I agree with all your answers.

denialhaag · 2026-08-25T17:29:33Z

Thanks @denialhaag and @simon1hofmann 🙏🏼
Given how this is at the very bottom of an 11 PR stack, I'd like to only touch this once more and freeze the implementation afterwards. Are there any further comments on the above from your side?
Quick signal would be good. Then I'll kick off the implementation and the rebase.

Thanks for addressing my comments, I agree with all your answers.

Go also from my side!

burgholzer

Two more things to address here before this can go in.

burgholzer · 2026-09-01T22:58:45Z

I believe I am happy with this now. Looking forward to what you think @simon1hofmann 😌

simon1hofmann

This looks quite clean and I really like the changes here!
I just have two comments that should be checked before this can go in.

simon1hofmann · 2026-09-02T10:42:41Z

@burgholzer I fixed the two open comments and this PR should we ready now.

burgholzer · 2026-09-02T10:49:56Z

@burgholzer I fixed the two open comments and this PR should we ready now.

One (hopefully) last request for changes. Then this can go in 👍🏼

Represent unknown, unrestricted, and explicit connectivity and native-operation facts separately. Use site and arity vocabulary, and make target passes request facts only when needed. Assisted-by: OpenAI Codex

Assisted-by: OpenAI Codex

Keep unavailable QDMI v1.3 properties unknown while preserving the compile-and-execute integration with explicit simulator capabilities. Assisted-by: OpenAI Codex

Replace ordered sets with the project's LLVM containers and in-place canonicalization. Assisted-by: OpenAI Codex

Copy public collection views into compact LLVM storage and document the explicit DDSIM synthesis basis used until QDMI v1.4 reports all-to-all connectivity. Assisted-by: GPT-5.6 Sol via Codex

Avoid topology reconciliation when a single-site program does not need unknown connectivity, preserve the full nonnegative site-ID domain, and document explicit DDSIM capabilities. Assisted-by: GPT-5.6 Sol via Codex

QDMI 1.3 cannot enumerate the DDSIM all-to-all topology or homogeneous operation loci compactly. Advertise an exact custom marker and let the compiler adapter consume it without changing the fail-closed default for other devices. Assisted-by: GPT-5.6 Sol via Codex

Parallel Windows CTest workers could delete each other's inputs because the QIR and invalid-input scripts shared one output directory. Give the invalid-input test its own scratch path. Assisted-by: GPT-5.6 Sol via Codex

Keep the generic Python overload checks independent of DDSIM metadata now that CUSTOM1 carries compiler-target facts. Assisted-by: GPT-5.6 Sol via Codex

Dispatch explicit connectivity to mapping and use deterministic placement for all-to-all or safe unknown-connectivity programs. Reject unknown multi-site placement before mutation. Keep the QDMI 1.3 DDSIM metadata marker explicitly temporary until QDMI standardizes the equivalent facts. Assisted-by: GPT-5.6 Sol via Codex

Require complete topology and native-operation facts at target construction instead of propagating unusable unknown states. Model global phase as fixed zero arity and simulator gate families as positive variadic capabilities. DDSIM reports arbitrary controls through one exact QDMI 1.3 compatibility marker, so compiler support covers every canonical base gate without enumerating controlled aliases. Assisted-by: GPT-5.6 Sol via Codex

Rely on the QCO cleanup pipeline to canonicalize empty control modifiers before target synthesis. Keep target documentation consistent with the current Doxygen style and query an unused custom-property slot in generic DDSIM tests. Assisted-by: GPT-5.6 Sol via Codex

Treat variadic X and Z capabilities as CX and CZ support in the gate cache. This keeps synthesis-basis resolution consistent with controlled-operation support. Assisted-by: GPT-5.6 Sol via Codex

Global-phase normalization can leave an empty control shell beside the lowered phase operation. Remove that shell before target support planning so the standalone synthesis pass accepts P-only targets. Assisted-by: GPT-5.6 Sol via Codex

Use LLVM early-increment iteration to preserve reverse deletion without manual iterator bookkeeping. Assisted-by: GPT-5.6 Sol via Codex

Keep an explicit pointer to the previous operation before erasing the current node. Reverse early-increment iteration retains a dangling base iterator after deletion.\n\nAssisted-by: GPT-5.6 Sol via Codex

Adapt the mapping test added on main to the explicit connectivity and native-operation facts required by the generalized compiler target API.\n\nAssisted-by: GPT-5.6 Sol via Codex

Adapt the fixed-point test added on main to the explicit connectivity and native-operation facts required by the generalized compiler target API. Assisted-by: GPT-5.6 Sol via Codex

burgholzer added refactor Anything related to code refactoring c++ Anything related to C++ code python Anything related to Python code MLIR Anything related to MLIR labels Aug 23, 2026

burgholzer self-assigned this Aug 23, 2026

burgholzer mentioned this pull request Aug 23, 2026

✨ Record typed compiler targets in MQT IR #2215

Merged

12 tasks

burgholzer added this to the MLIR Support milestone Aug 23, 2026

burgholzer commented Aug 24, 2026

View reviewed changes

Comment thread docs/qdmi/ddsim_device.md Outdated

Comment thread docs/qdmi/ddsim_device.md Outdated

Comment thread mlir/include/mlir/Compiler/Target.h Outdated

Comment thread mlir/include/mlir/Compiler/Target.h Outdated

Comment thread mlir/include/mlir/Compiler/Target.h Outdated

burgholzer mentioned this pull request Aug 24, 2026

✨ Legalize control flow for selected payloads #2162

Draft

12 tasks

burgholzer force-pushed the codex/generalize-compiler-target branch from 9b961b2 to 8fafa97 Compare August 24, 2026 13:52

mergify Bot added the conflict label Aug 24, 2026

burgholzer force-pushed the codex/generalize-compiler-target branch from 8fafa97 to a6f7fd1 Compare August 24, 2026 23:47

burgholzer removed the conflict label Aug 24, 2026

denialhaag approved these changes Aug 25, 2026

View reviewed changes

simon1hofmann requested changes Aug 25, 2026

View reviewed changes

mergify Bot added the conflict label Aug 26, 2026

burgholzer modified the milestones: MLIR Support, v4.0.0 - IEEE Quantum Week Sep 1, 2026

burgholzer force-pushed the codex/generalize-compiler-target branch from a6f7fd1 to fe41d2a Compare September 1, 2026 06:13

burgholzer added fix Fix for something that isn't working and removed conflict labels Sep 1, 2026

burgholzer mentioned this pull request Sep 1, 2026

⬆️ Adopt QDMI 1.4 in MQT Core #2093

Open

burgholzer force-pushed the codex/generalize-compiler-target branch from fe41d2a to 363f8b5 Compare September 1, 2026 08:44

burgholzer commented Sep 1, 2026

View reviewed changes

Comment thread mlir/lib/Dialect/QCO/Transforms/Mapping/Mapping.cpp

Comment thread docs/qdmi/ddsim_device.md

burgholzer mentioned this pull request Sep 1, 2026

♻️ Separate target placement from routing #2316

Merged

12 tasks

burgholzer requested a review from simon1hofmann September 1, 2026 22:58

simon1hofmann approved these changes Sep 2, 2026

View reviewed changes

Comment thread mlir/lib/Compiler/Target.cpp

Comment thread mlir/lib/Dialect/QCO/Transforms/NativeSynthesis/TargetSynthesis.cpp

simon1hofmann force-pushed the codex/generalize-compiler-target branch from 9355f74 to 2b887ba Compare September 2, 2026 14:19

burgholzer and others added 19 commits September 2, 2026 18:00

♻️ Generalize compiler target facts

781bf60

Represent unknown, unrestricted, and explicit connectivity and native-operation facts separately. Use site and arity vocabulary, and make target passes request facts only when needed. Assisted-by: OpenAI Codex

🎨 pre-commit fixes

7bef032

📝 Fold compiler target work into launch changelog

9308dc3

Assisted-by: OpenAI Codex

🧪 State DDSIM target facts explicitly

4ca2178

Keep unavailable QDMI v1.3 properties unknown while preserving the compile-and-execute integration with explicit simulator capabilities. Assisted-by: OpenAI Codex

⚡️ Use compact LLVM target checks

73c5008

Replace ordered sets with the project's LLVM containers and in-place canonicalization. Assisted-by: OpenAI Codex

♻️ Use LLVM views for target collections

c18c964

Copy public collection views into compact LLVM storage and document the explicit DDSIM synthesis basis used until QDMI v1.4 reports all-to-all connectivity. Assisted-by: GPT-5.6 Sol via Codex

🐛 Preserve compiler target fact semantics

d860d65

Avoid topology reconciliation when a single-site program does not need unknown connectivity, preserve the full nonnegative site-ID domain, and document explicit DDSIM capabilities. Assisted-by: GPT-5.6 Sol via Codex

🧪 Isolate mqt-cc test scratch files

5ed0f5c

Parallel Windows CTest workers could delete each other's inputs because the QIR and invalid-input scripts shared one output directory. Give the invalid-input test its own scratch path. Assisted-by: GPT-5.6 Sol via Codex

🧪 Test an unused custom property slot

56ecd62

Keep the generic Python overload checks independent of DDSIM metadata now that CUSTOM1 carries compiler-target facts. Assisted-by: GPT-5.6 Sol via Codex

🐛 Derive controlled gates from variadic bases

14e0132

Treat variadic X and Z capabilities as CX and CZ support in the gate cache. This keeps synthesis-basis resolution consistent with controlled-operation support. Assisted-by: GPT-5.6 Sol via Codex

♻️ Move empty control cleanup into phase normalization

64bad12

♻️ Simplify reverse operation cleanup

2ecba6e

Use LLVM early-increment iteration to preserve reverse deletion without manual iterator bookkeeping. Assisted-by: GPT-5.6 Sol via Codex

🐛 Preserve reverse deletion iterators

808be13

Keep an explicit pointer to the previous operation before erasing the current node. Reverse early-increment iteration retains a dangling base iterator after deletion.\n\nAssisted-by: GPT-5.6 Sol via Codex

🧪 Update mapping target construction

ba81d4b

Adapt the mapping test added on main to the explicit connectivity and native-operation facts required by the generalized compiler target API.\n\nAssisted-by: GPT-5.6 Sol via Codex

simon1hofmann force-pushed the codex/generalize-compiler-target branch from 2b887ba to ba81d4b Compare September 2, 2026 16:04

🧪 Update compiler target construction

1bc668b

Adapt the fixed-point test added on main to the explicit connectivity and native-operation facts required by the generalized compiler target API. Assisted-by: GPT-5.6 Sol via Codex

simon1hofmann merged commit 156921e into main Sep 2, 2026
27 checks passed

simon1hofmann deleted the codex/generalize-compiler-target branch September 2, 2026 17:11

burgholzer mentioned this pull request Sep 4, 2026

♻️ Define target compilation with payload specifications #2219

Open

12 tasks

Uh oh!

Uh oh!

Conversation

burgholzer commented Aug 23, 2026 • edited by simon1hofmann Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Validation

Checklist

Uh oh!

codecov Bot commented Aug 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

burgholzer left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

coderabbitai Bot commented Aug 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Approval pending

Uh oh!

denialhaag left a comment

Choose a reason for hiding this comment

Uh oh!

simon1hofmann left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

burgholzer commented Aug 25, 2026

Uh oh!

simon1hofmann commented Aug 25, 2026

Uh oh!

denialhaag commented Aug 25, 2026

Uh oh!

burgholzer left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

burgholzer commented Sep 1, 2026

Uh oh!

simon1hofmann left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

simon1hofmann commented Sep 2, 2026

Uh oh!

burgholzer commented Sep 2, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

burgholzer commented Aug 23, 2026 •

edited by simon1hofmann

Loading

codecov Bot commented Aug 23, 2026 •

edited

Loading

coderabbitai Bot commented Aug 24, 2026 •

edited

Loading