🐛 Preserve symbolic Euler chains in target synthesis

[simon1hofmann]

🤖 AI text below 🤖

Description

Allow symbolic circSU2 circuits to compile for an SX/X/RZ/CZ target and export
to Qiskit with their original parameters still bindable. Extend direct symbolic
synthesis to compatible X-outer Euler chains as well.

• Reuse ZXZ/ZYZ chain angles directly in the u, zyz, zxz, and zsxx
  bases, and XZX/XYX chain angles in xzx, xyx, and r. Either outer
  rotation may be absent. Reuse the existing emitters without inverse
  trigonometry or conditional expressions, preserving global phase and the
  existing synthesis gate-count bounds.
• Fuse single-qubit runs directly into non-U target bases after placement and
  classical-read cleanup. This avoids the intermediate U phase corrections.
• Retain the existing optimizer for U-based and unrestricted targets. Other
  symbolic chain shapes keep their existing synthesis path; this is not a
  general symbolic atan2 export fix.
• Test controlled rotations with phase-sensitive matrix comparisons, all 56
  compatible chain/basis combinations, and symbolic compile–export–bind cases.
  Check X-outer chains through both target compilation and synthesis-only APIs.

No new dependencies or Python API changes are required.

Results

The two skipped Benchpress Summit circSU2 workloads pass through the MQT
adapter on fake_torino: circular EfficientSU2, three repetitions, default
compilation options. Both runs preserve the input and parameter identities,
export successfully, bind zero/mixed values after export, and pass target
validation.

These measurements were recorded before the XZX/XYX extension:

Qubits	Parameters	CZ	Compilation, two runs	Export, two runs
89	712	609	0.732 / 0.734 s	0.062 / 0.063 s
100	800	651	0.918 / 0.927 s	0.075 / 0.075 s

These are local diagnostic timings, not a cross-tool performance comparison.
The Benchpress skip markers are unchanged pending integration of this fix.

Validation

Based on main dc78630f9; Release build on macOS ARM64 with LLVM/MLIR 23.1.0,
Python 3.14.3 and Qiskit 2.5.2.

Latest validation for 1a9a1da5d, including the XZX/XYX extension:

• C++ optimization and target-synthesis tests: 264 passed (203 + 61).
• Python compiler/API tests in test/python/test_mlir.py: 98 passed, including
  the original small/full-size circSU2 regressions and six new X-outer-chain
  compile/synthesize–export–bind cases.
• Controlled-chain regressions check full matrices at zero, π, 2π, and mixed
  angles, together with basis membership and synthesis gate-count bounds.
• Whole-changed-file C++ lint with Clang-Tidy 23: passed with no findings.
• Repository lint and git diff --check: passed.
• CI for the latest commit is still in progress at the time of this update.

Earlier validation, before the XZX/XYX extension:

• Core Python compiler/export tests: 511 passed.
• Benchpress: 353 passed (101 MQT unit tests and 252 small/medium QASMBench
  functional cases). Timing disabled; excludes large circuits and the eight
  bwt_n21/factor247_n15 topology cases.
• C++ optimization, target synthesis and Euler/fusion tests: 331 passed.
• C++ compiler tests: 222 passed; ConvertsUnknownDeviceFailureToError fails
  because the local QDMI error lacks the expected diagnostic text. This was
  also observed before the fix; a separate native main build was not used to
  establish baseline status.

Native tests ran directly with local shared-library paths. The latest local
build uses deferred GoogleTest discovery to avoid the post-link discovery
failure; no repository build policy was changed.

OpenAI Codex assisted with implementation, tests, review and this description.

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.
• 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]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!