[ImportVerilog] Qualifiers unique/unique0/priority are ignored

[rhanqtl]

sv-tests §12.4.2 and §12.5.3.

As stated in the title, qualifiers are ignored in the generated Moore IR.

These qualifiers are used to perform violation checks during simulation and provide optimization opportunities for synthesis.

[rhanqtl]

If there is no existing solution, possible ones are:

1. Wrapping the entire if/case statement with a new Op, say moore.assert_unique:

   moore.assert_unique unique0 {
       // IR of the if/case statement
   }

2. Make moore.assert_unique a side-band

   moore.assert_unique unique0 %cond1, %cond2, ...
   // IR of the if/case statement

3. Introduce moore.if and moore.case

[thomasnormal]

I reproduced this on current main and landed a fix in my fork at commit 846af1252:

• Added semantic regressions:
  • test/Tools/circt-sim/syscall-unique-priority-case-violations.sv
  • test/Tools/circt-sim/syscall-unique-priority-if-violations.sv
• ImportVerilog lowering now emits runtime checks for unique / unique0 / priority for procedural if, case, and pattern case.
• Root cause for missing unique/unique0 case overlap diagnostics: lowering exited on first matched item, so overlap checks on later items were unreachable.
• Fix: continue scanning case items for uniqueness checks while still executing only the first matching item body; default dispatch is now conditioned on no prior match when this scan path is enabled.

Targeted validation run:

• build_test/bin/llvm-lit -sv test/Tools/circt-sim/syscall-unique-priority-case.sv test/Tools/circt-sim/syscall-unique-priority-case-violations.sv test/Tools/circt-sim/syscall-unique-priority-if-violations.sv
• build_test/bin/llvm-lit -sv test/Conversion/ImportVerilog/case-inside.sv test/Conversion/ImportVerilog/case-exhaustive-block-args.sv test/Conversion/ImportVerilog/assertion-guard-if.sv}

[thomasnormal]

Rechecked on current in this workspace: I cannot reproduce qualifier-dropping anymore.\n\nRepro used:\n\n\n\nLowered LLHD contains explicit warning instrumentation:\n- \n- \n- \n- \n\nI also added a focused in-tree regression:\n- \n\nFocused validation:\n- -- Testing: 1 tests, 1 workers --
Testing: 0.. 10.. 20.. 30.. 40.. 50.. 60.. 70.. 80.. 90..

Testing Time: 0.03s

Total Discovered Tests: 1
Passed: 1 (100.00%)\n- -- Testing: 2 tests, 1 workers --
Testing: 0.. 10.. 20.. 30.. 40.. 50.. 60.. 70.. 80.. 90..

Testing Time: 0.05s

Total Discovered Tests: 2
Passed: 2 (100.00%)\n\nAll pass locally.

[thomasnormal]

Clean recheck (without shell expansion artifacts): this no longer reproduces for me on current workspace main.

Repro:

• build_test/bin/circt-verilog --ir-llhd /tmp/unique_priority_check.sv -o -

Observed lowered LLHD includes explicit warning instrumentation:

• sim.fmt.literal "Warning: unique case: multiple items matched"
• sim.fmt.literal "Warning: unique case: no items matched"
• sim.fmt.literal "Warning: priority if: no items matched"
• llvm.call @__moore_warning

Added focused regression:

• test/Conversion/ImportVerilog/unique-priority-qualifier-checks.sv

Focused validation:

• build_test/bin/llvm-lit -sv -j1 test/Conversion/ImportVerilog/unique-priority-qualifier-checks.sv
• build_test/bin/llvm-lit -sv -j1 test/Tools/circt-sim/syscall-unique-priority-case-violations.sv test/Tools/circt-sim/syscall-unique-priority-if-violations.sv

All pass locally.

[thomasnormal]

Rechecked on current main; I cannot reproduce qualifiers being ignored for the covered semantic shapes.

Focused tests pass:

• build_test/bin/llvm-lit -sv -j1 test/Tools/circt-sim/syscall-unique-priority-case-violations.sv test/Tools/circt-sim/syscall-unique-priority-if-violations.sv

Also spot-checked --ir-moore lowering on a mixed unique case + priority if repro and observed explicit runtime-check logic (moore.builtin.severity) emitted for no-match / multi-match checks.

If there is still a failing variant (e.g. specific casez/casex pattern or synthesis-oriented lowering expectation), please share the exact source + command and I can reduce it to an in-tree regression.

[thomasnormal]

Follow-up on this issue: I found and fixed an uncovered semantic bug in else-if chains.

Repro shape

With , CIRCT emitted even though the arm matched.

Same false-positive shape existed for .

Root cause

lowering applied using only the current condition before lowering the full else-if chain, so no-match checking ignored later else-if arms.

Fix (downstream)

• commit:
• added chain-aware tracking for else-if heads and deferred no-match diagnostics to chain completion.
• kept no-match warnings when a chain has no terminal and no arm matches.

Regression test added

Focused validation

All pass in this workspace after the fix.

[thomasnormal]

Follow-up on this issue: I found and fixed an uncovered semantic bug in priority/unique if else-if chains.

Repro shape

priority if (a)
  ...
else if (b)
  ...

With a=0, b=1, CIRCT emitted priority if: no items matched even though the else if arm matched.

Same false-positive shape existed for unique if (...) else if (...).

Root cause

ConditionalStatement lowering applied emitUniquePriorityChecks using only the current if condition before lowering the full else-if chain, so no-match checking ignored later else-if arms.

Fix (downstream)

• commit: 0f45d7b728
• added chain-aware tracking for else-if heads and deferred no-match diagnostics to chain completion.
• kept no-match warnings when a chain has no terminal else and no arm matches.

Regression test added

• test/Tools/circt-sim/syscall-unique-priority-if-elseif-no-spurious-no-match.sv

Focused validation

• llvm-lit -sv -j1 test/Tools/circt-sim/syscall-unique-priority-if-elseif-no-spurious-no-match.sv
• llvm-lit -sv -j1 test/Tools/circt-sim/syscall-unique-priority-if-violations.sv test/Tools/circt-sim/syscall-unique-priority-case-violations.sv

All pass in this workspace after the fix.

[thomasnormal]

Downstream status update: fix is now pushed on as commit (same semantic/code changes + regression test).

[thomasnormal]

Downstream status update: fix is now pushed on thomasnormal/circt main as commit 15d27c65a5 (same semantic/code changes + regression test).

[seldridge]

@thomasnormal, @thomasahle: Due to current CIRCT project policy, we are unable to accept patches from the thomasnormal/circt fork.

For others: please do not upstream any patches from the thomasnormal/circt fork. Additionally, recognize that any comments made by the @thomasnormal account appear to be from one or more AI agents.

For more information, please see this Discourse post: https://discourse.llvm.org/t/circt-project-policy-on-contributions-from-thomasnormal-circt-fork/90075