Remove measurement insertion from mapping pass by atgeller · Pull Request #5463 · NVIDIA/cuda-quantum

atgeller · 2026-09-21T19:32:00Z

Motivation

Mapping.cpp inserted implicit measurements for unmeasured qubits as a side effect of mapping. This is unpleasant for several reasons:

That is add-measurements' job, having duplicate functionality is never good.
Doing this inside the mapper meant the measurement set depended on when mapping happened to run.
This prevents mapping from being used composable, as this measurement insertion is a whole program transformation.

This PR removes these implicit measurements, making add-measurements the single place where implicit sampling measurements are added.

Changes

Mapping.cpp: drop the implicit-measurement insertion. The mapping lit tests now carry explicit measurements of their own.
Compiler.cpp: move the sample-only add-measurements call to before executeMainPipeline. With the mapper no longer measuring, running it afterwards measured physical qubits and emitted an empty mapping_reorder_idx, so sample results came back in physical rather than program order ('01' for '10'). This affects every target whose pipeline has qubit-mapping but no add-measurements of its own: oqc, iqm, anyon.
add-measurements: new use-evince option (default false). add-deallocs does not insert deallocations in a stable order, whereas implicit measurements are assumed to be added in allocation order. inject-implicit-output emits quake.evinces in allocation order, which is preserved, providing a workaround. It should be noted that using this new option as the default across the codebase causes no issues, and may be a good future direction to take the pass to avoid ordering issues.
ArgumentConversion.cpp: createInitFunc replaces the callee's alloca with a subveq of the caller's veq, so inlined state-producing kernels leave behind deallocations and compiler-generated evince ops that act on qubits the caller owns. Both are now dropped via a shared traceToAllocation helper; otherwise they become deallocations and observation points in the middle of the caller's circuit.

Behavior change

Measurements are now added before the main pipeline runs, so ancilla qubits introduced by multicontrol decomposition are no longer measured and no longer appear in sample bitstrings. Affected expectations in test_kernel_builder.py and QuantinuumTester.cpp are updated.

Signed-off-by: Adam Geller <adgeller@nvidia.com>

github-actions · 2026-09-21T20:08:02Z

CI Summary (`push`) — ✅ passed

Run #37705585697 · ✅ 8 · ⏩ 8 · ❌ 0 · ⛔ 0

Top-level jobs (16)

Job	Result
`binaries`	⏩ skipped
`build_and_test`	✅ success
`changes`	✅ success
`config_devdeps`	✅ success
`config_source_build`	⏩ skipped
`config_wheeldeps`	✅ success
`devdeps`	✅ success
`docker_image`	⏩ skipped
`documentation`	✅ success
`gen_code_coverage`	⏩ skipped
`metadata`	✅ success
`python_devel_wheel`	⏩ skipped
`python_metapackages`	⏩ skipped
`python_wheels`	⏩ skipped
`source_build`	⏩ skipped
`wheeldeps`	✅ success

⏩ Skipped jobs (8) — intentionally skipped on PR builds; run on merge_group / workflow_dispatch

Job
`binaries`
`config_source_build`
`docker_image`
`gen_code_coverage`
`python_devel_wheel`
`python_metapackages`
`python_wheels`
`source_build`

All sub-jobs (45) — every matrix leg, with links

Job	Status	Link
Build and test (amd64, gcc12, openmpi) / Dev environment (Debug)	✅ success	view
Build and test (amd64, gcc12, openmpi) / Dev environment (Python)	✅ success	view
Build and test (amd64, llvm, openmpi) / Dev environment (Debug)	✅ success	view
Build and test (amd64, llvm, openmpi) / Dev environment (Python)	✅ success	view
Build and test (arm64, llvm, openmpi) / Dev environment (Debug)	✅ success	view
Build and test (arm64, llvm, openmpi) / Dev environment (Python)	✅ success	view
CI Summary	❔ in_progress	view
Check for stable CUDA-Q changes	✅ success	view
Configure build (devdeps)	✅ success	view
Configure build (source_build)	⏩ skipped	view
Configure build (wheeldeps)	✅ success	view
Create CUDA Quantum installer	⏩ skipped	view
Create Docker images	⏩ skipped	view
Create Python metapackages	⏩ skipped	view
Create Python wheels	⏩ skipped	view
Create shared Python development wheel	⏩ skipped	view
Documentation / Build and check docs	✅ success	view
Gen code coverage	⏩ skipped	view
Load dependencies (amd64, gcc12) / Caching	✅ success	view
Load dependencies (amd64, gcc12) / Finalize	✅ success	view
Load dependencies (amd64, gcc12) / Metadata	✅ success	view
Load dependencies (amd64, llvm) / Caching	✅ success	view
Load dependencies (amd64, llvm) / Finalize	✅ success	view
Load dependencies (amd64, llvm) / Metadata	✅ success	view
Load dependencies (arm64, gcc12) / Caching	✅ success	view
Load dependencies (arm64, gcc12) / Finalize	✅ success	view
Load dependencies (arm64, gcc12) / Metadata	✅ success	view
Load dependencies (arm64, llvm) / Caching	✅ success	view
Load dependencies (arm64, llvm) / Finalize	✅ success	view
Load dependencies (arm64, llvm) / Metadata	✅ success	view
Load source build cache	⏩ skipped	view
Load wheel dependencies (amd64, 12.6) / Caching	✅ success	view
Load wheel dependencies (amd64, 12.6) / Finalize	✅ success	view
Load wheel dependencies (amd64, 12.6) / Metadata	✅ success	view
Load wheel dependencies (amd64, 13.0) / Caching	✅ success	view
Load wheel dependencies (amd64, 13.0) / Finalize	✅ success	view
Load wheel dependencies (amd64, 13.0) / Metadata	✅ success	view
Load wheel dependencies (arm64, 12.6) / Caching	✅ success	view
Load wheel dependencies (arm64, 12.6) / Finalize	✅ success	view
Load wheel dependencies (arm64, 12.6) / Metadata	✅ success	view
Load wheel dependencies (arm64, 13.0) / Caching	✅ success	view
Load wheel dependencies (arm64, 13.0) / Finalize	✅ success	view
Load wheel dependencies (arm64, 13.0) / Metadata	✅ success	view
Prepare cache clean-up	❔ in_progress	view
Retrieve PR info	✅ success	view

✅ Required checks (7/7) — declared in .github/required-checks.yml for push

Required check	Status	Link
Build and test (amd64, llvm, openmpi) / Dev environment (Debug)	✅ success	view
Build and test (amd64, llvm, openmpi) / Dev environment (Python)	✅ success	view
Build and test (arm64, llvm, openmpi) / Dev environment (Debug)	✅ success	view
Build and test (arm64, llvm, openmpi) / Dev environment (Python)	✅ success	view
Build and test (amd64, gcc12, openmpi) / Dev environment (Debug)	✅ success	view
Build and test (amd64, gcc12, openmpi) / Dev environment (Python)	✅ success	view
Documentation / Build and check docs	✅ success	view

Signed-off-by: Adam Geller <adgeller@nvidia.com>

…match allocation order when deallocs are present. Signed-off-by: Adam Geller <adgeller@nvidia.com>

Signed-off-by: Adam Geller <adgeller@nvidia.com>

sacpis

Overall LGTM. Thanks @atgeller. Left few comments.

sacpis · 2026-10-08T00:56:31Z

+    if (useEvince && analysis.hasEvinces()) {
+      if (failed(addMeasurementsBefore(func, analysis.evinces)))
+        signalPassFailure();


The following repro.cpp

#include <cudaq.h> struct test { void operator()(bool b) __qpu__ { cudaq::qubit q; x(q); if (b) { cudaq::qubit a; x(a); } } }; int main() { cudaq::sample(10, test{}, true).dump(); }

shows output as

{ 1:10 }

Shouldn't it be { 11:10 }? I think in evince mode, we should also measure in front of the deallocations of qubits that are not covered by an evince (for example, nested allocations)? Or the other option is to have inject-implicit-output create evinces for nested allocations too.

We don't really want to spend time beefing up automatic injection of evince operations. The entire pass is a hack to work around other more serious problems.

Which is to say, beefing up that pass is a waste of time as the pass will be removed entirely very soon. Which means, this approach for adding the measurements is a temporary band-aid. Which is to say, the insertion of phantom measurements is either wrong or misguided even though it is papering over gaps in other passes.

This is a short-term bandaid to try to reduce the amount of implicit behavior. In this case using the add-measurements pass instead of inserting the measurements in the mapper, so it's at least an explicit pass. @schweitzpgi is working on a longer term solution to remove the need for implicit behavior.

sacpis · 2026-10-08T00:59:45Z

+  // The caller owns the qubits, so drop deallocations of the argument.
+  initFunc.walk([&](cudaq::quake::DeallocOp dealloc) {
+    if (isa<BlockArgument>(traceToAllocation(dealloc.getReference())))
+      dealloc.erase();
+  });
+
+  // The caller is also responsible for making its own qubits observable. Drop
+  // the compiler-generated evince ops on the argument, which would otherwise
+  // be observation points in the middle of the caller's circuit.
+  initFunc.walk([&](cudaq::quake::EvinceOp evince) {
+    if (evince.getCompilerGenerated() &&
+        llvm::all_of(evince.getArgs(), [&](Value v) {
+          return isa<BlockArgument>(traceToAllocation(v));
+        }))
+      evince.erase();
+  });


I don't see any test for the new evince and dealloc removal in state kernel init functions. Let's add a regression case or an end-to-end sample test with a get_state / state-from-kernel argument on an emulated remote target?

sacpis · 2026-10-08T01:02:08Z

+module {
+  func.func @reference() attributes {"cudaq-entrypoint", "cudaq-kernel", no_this} {
+    %0 = quake.alloca !quake.veq<2>
+    %1 = quake.alloca !quake.ref
+    %2 = quake.extract_ref %0[0] : (!quake.veq<2>) -> !quake.ref
+    quake.h %2 : (!quake.ref) -> ()
+    quake.x %1 : (!quake.ref) -> ()
+    return
+  }
+}


Ah seems like it does not cover nested allocations or early returns in evince mode. Should we add tests for an allocation made inside cc.if or cc.scope? And multiple returns?

It would be better to fall back to using deallocations than to try to build the house of cards around evince even higher. (The entire purpose of evince is not to be a workaround for removing, adding, or teleporting measurement operations around in the IR.)

sacpis · 2026-10-08T01:03:44Z

  let description = [{
    This pass adds measurement operations to the kernels that have no 
    measurements.

    This pass should be run in the sampling context, so as to comply with the 
    spec - If no measurements appear in the kernel, then the `__global__` 
    register is formed with implicit measurements being added for all the 
    qubits defined in the kernel, and the measurements all occur at the end of 
    the kernel.



Let's add a short paragraph describing the new mode use-evince? Also, when the dealloc or return paths are used instead.

schweitzpgi

+1 Thanks for grinding through this, Adam.

Remove measurement insertion from mapping pass

29b1341

Signed-off-by: Adam Geller <adgeller@nvidia.com>

atgeller added 2 commits September 22, 2026 12:46

Add measurements to tests

aa0809f

Signed-off-by: Adam Geller <adgeller@nvidia.com>

Merge branch 'main' into remove_mapping_measures

a2a878f

atgeller marked this pull request as ready for review September 23, 2026 21:41

github-actions Bot added the core compiler label Sep 23, 2026

atgeller marked this pull request as draft September 23, 2026 22:10

atgeller added 4 commits October 5, 2026 15:49

Move add-measurements earlier to avoid pipeline issues

bbc8e20

Signed-off-by: Adam Geller <adgeller@nvidia.com>

Fix more test strings

a9a7689

Signed-off-by: Adam Geller <adgeller@nvidia.com>

Merge branch 'main' into remove_mapping_measures

f138248

Add option to add measurements before evinces instead of deallocs to …

4b514de

…match allocation order when deallocs are present. Signed-off-by: Adam Geller <adgeller@nvidia.com>

atgeller marked this pull request as ready for review October 8, 2026 00:00

Merge branch 'main' into remove_mapping_measures

3d78ecd

github-actions Bot added runtime CUDA quantum runtime python-lang Anything related to the Python CUDA Quantum language implementation labels Oct 8, 2026

atgeller requested review from 1tnguyen, khalatepradnya, sacpis, schweitzpgi and taalexander October 8, 2026 00:00

Fix comment and test case

2659edb

Signed-off-by: Adam Geller <adgeller@nvidia.com>

atgeller requested a review from cabreraam October 8, 2026 00:09

sacpis approved these changes Oct 8, 2026

View reviewed changes

schweitzpgi approved these changes Oct 8, 2026

View reviewed changes

Navigation Menu

Repository navigation

Remove measurement insertion from mapping pass - #5463

Remove measurement insertion from mapping pass#5463
atgeller wants to merge 9 commits into
NVIDIA:mainfrom
atgeller:remove_mapping_measures

atgeller commented Sep 21, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Sep 21, 2026 •

edited

Loading

Uh oh!

sacpis left a comment

Uh oh!

sacpis Oct 8, 2026

Uh oh!

schweitzpgi Oct 8, 2026 •

edited

Loading

Uh oh!

atgeller Oct 8, 2026

Uh oh!

sacpis Oct 8, 2026

Uh oh!

sacpis Oct 8, 2026

Uh oh!

schweitzpgi Oct 8, 2026 •

edited

Loading

Uh oh!

sacpis Oct 8, 2026

Uh oh!

schweitzpgi left a comment

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

atgeller commented Sep 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Motivation

Changes

Behavior change

Uh oh!

github-actions Bot commented Sep 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

CI Summary (push) — ✅ passed

Uh oh!

sacpis left a comment

Choose a reason for hiding this comment

Uh oh!

sacpis Oct 8, 2026

Choose a reason for hiding this comment

Uh oh!

schweitzpgi Oct 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

atgeller Oct 8, 2026

Choose a reason for hiding this comment

Uh oh!

sacpis Oct 8, 2026

Choose a reason for hiding this comment

Uh oh!

sacpis Oct 8, 2026

Choose a reason for hiding this comment

Uh oh!

schweitzpgi Oct 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

sacpis Oct 8, 2026

Choose a reason for hiding this comment

Uh oh!

schweitzpgi left a comment

Choose a reason for hiding this comment

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

atgeller commented Sep 21, 2026 •

edited

Loading

github-actions Bot commented Sep 21, 2026 •

edited

Loading

CI Summary (`push`) — ✅ passed

schweitzpgi Oct 8, 2026 •

edited

Loading

schweitzpgi Oct 8, 2026 •

edited

Loading