Skip to content

Add traits for Quantum dialect ops and remove AnyAttr usage #8083

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 27 commits into from
Sep 5, 2025
Merged

Add traits for Quantum dialect ops and remove AnyAttr usage #8083

merged 27 commits into from
Sep 5, 2025

Conversation

@mudit2812
Copy link
Contributor

@mudit2812 mudit2812 commented Aug 13, 2025 •
edited
Loading

Context:
The xDSL quantum dialect had a lot of missing traits and very permissive ins/outs definitions. This PR resolves both these issues.

Description of the Change:

  • Add traits to the Quantum dialect
  • Remove all uses of AnyAttr from the Quantum dialect.

Benefits:

Possible Drawbacks:

Related GitHub Issues:

[sc-97692] [sc-97620]

mudit2812
mudit2812 commented Aug 14, 2025
View reviewed changes
@mudit2812 mudit2812 added the unified compiler Issues and pull requests related to PennyLane's xDSL python compiler. label Aug 22, 2025
@mudit2812 mudit2812 changed the base branch from master to tensor-memref-constraints August 25, 2025 20:43
Base automatically changed from tensor-memref-constraints to master August 26, 2025 16:15
mudit2812 and others added 9 commits August 28, 2025 10:47
@mudit2812
Squashed commit of the following:
0c35f34 
commit 154de3c
Author: Marcus Edwards <[email protected]>
Date:   Thu Aug 28 09:01:10 2025 -0400

    Decomposition rule (#8070)

    ------------------------------------------------------------------------------------------------------------

    **Context:**
    We would like to provide a way for users to specify that they are using
    a compute/uncompute pattern [U, V, U†] so that it may be efficiently
    controlled by applying controls to the "middle" operator V instead of
    the entire pattern.

    **Description of the Change:**
    Adds a decomposition rule and registers it under "C(ChangeOpBasis)".
    This rule applies ctrl only to the middle op in a compute/uncompute
    pattern.
    <img width="283" height="57" alt="image"
    src="https://github.com/user-attachments/assets/228b8a09-3b56-47bd-a014-d41667e0cb98"
    />

    **Benefits:** We can now decompose controlled compute/uncompute patterns
    efficiently.

    **Possible Drawbacks:**

    **Related ShortCut Stories:** [sc-97007]

    ---------

    Co-authored-by: Christina Lee <[email protected]>
    Co-authored-by: Astral Cai <[email protected]>
    Co-authored-by: Pietropaolo Frisoni <[email protected]>
    Co-authored-by: ANT0N <[email protected]>
    Co-authored-by: Austin Huang <[email protected]>

commit a77cf19
Author: Andrija Paurevic <[email protected]>
Date:   Thu Aug 28 08:31:11 2025 -0400

    feat: freeze `dict` fields of `ExecutionConfig` using new custom `FrozenMapping` object (#8046)

    **Context:**

    Currently we can mutate the dict fields of the `ExecutionConfig`,

    ```python
    import pennylane as qml
    from pennylane.devices.execution_config import ExecutionConfig

    config = ExecutionConfig(
        device_options={"hi": "bye"}, gradient_keyword_arguments={"foo": "bar"}
    )
    config.device_options["hi"] = "hello" # Should not be able to mutate this!
    ```
    This is bad.

    **Description of the Change:**

    Creates a new `FrozenMapping` object that prevents this.

    **Benefits:** Fully immutable execution config. This ensures our
    configuration isn't being mutated along the way which could lead to
    unexpected results. Furthermore, this helps out with the upgrade to jax
    0.7.0 as it is now hashable.

    **Possible Drawbacks:** Might impact ecosystem?

    [sc-96529]

    ---------

    Co-authored-by: Christina Lee <[email protected]>
    Co-authored-by: Astral Cai <[email protected]>
    Co-authored-by: Yushao Chen (Jerry) <[email protected]>

commit a17daca
Author: ringo-but-quantum <[email protected]>
Date:   Thu Aug 28 09:51:46 2025 +0000

    [no ci] bump nightly version

commit b9a0cfa
Author: Mehrdad Malek <[email protected]>
Date:   Wed Aug 27 19:09:57 2025 -0400

    Add elementwise operationss to Stablehlo dialect in Python compiler  (#8084)

    **Context:**
    Adds All the missing element-wise operations to the stablehlo dialect in
    python compiler.

    **Description of the Change:**
    We organized elementwise operations into logical modules (unary.py,
    binary.py, other.py)

    New Elementwise Operation Classes:
    - Unary Elementwise Operations (20 operations):
    `ConvertOp`, `CosineOp`, `ExponentialMinusOneOp`, `ExponentialOp`,
    `FloorOp`, `ImagOp`, `IsFiniteOp`, `LogOp`, `LogPlusOneOp`,
    `LogisticOp`, `NegateOp`, `RealOp`, `RoundNearestAfzOp`,
    `RoundNearestEvenOp`, `RsqrtOp`, `SignOp`, `SineOp`, `SqrtOp`, `TanOp`,
    `TanhOp`
    - Binary Elementwise Operations (6 operations):
    `ComplexOp`, `DivideOp`, `MaximumOp`, `MinimumOp`, `PowerOp`,
    `RemainderOp`
    - Other Operations (5 operations):
    `ClampOp`, `CompareOp`, `MapOp`, `ReducePrecisionOp`, `SelectOp`

    New Traits:
    - `Elementwise`
    - `SameOperandsAndResultShape`
    - `SameOperandsElementType`

    New Constraint:
    - `NestedTupleOfConstraint`

    **Benefits:**

    **Possible Drawbacks:**

    **Related GitHub Issues:**

    ---------

    Co-authored-by: Mudit Pandey <[email protected]>

commit 2e42d23
Author: Astral Cai <[email protected]>
Date:   Wed Aug 27 16:15:06 2025 -0400

    Refactor decompositions to use dynamic wire allocation (#8024)

    **Context:**

    **Description of the Change:**
    Updates decompositions for controlled operators to use dynamic wire
    allocation
    No updates have been made to templates.

    **Benefits:**

    **Possible Drawbacks:**

    **Related GitHub Issues:**
    [sc-94401]

    ---------

    Co-authored-by: Isaac De Vlugt <[email protected]>
    Co-authored-by: Mudit Pandey <[email protected]>
    Co-authored-by: Marcus Edwards <[email protected]>
    Co-authored-by: Yushao Chen (Jerry) <[email protected]>

commit de04d85
Author: David Wierichs <[email protected]>
Date:   Wed Aug 27 22:02:03 2025 +0200

    Stop queueing from within `qml.pauli.PauliWord.operation` and `qml.pauli.PauliSentence.operation` (#8136)

    **Context:**
    Queuing behaviour is inconsistent due to caching, see bug report.

    **Description of the Change:**

    Stop queuing altogether from within the mentioned methods.

    **Benefits:**
    Consistent queuing behaviour.

    **Possible Drawbacks:**
    Stopping the queuing system for each call to `.operation` might be a
    performance regression (because in many cases there was no queuing
    anyways, so now we just do extra work of toggling the recording status,
    I suppose).

    **Related GitHub Issues:**
    Fixes #8135
    [sc-98282]

commit 08d58a7
Author: Simone Gasperini <[email protected]>
Date:   Wed Aug 27 15:51:24 2025 -0400

    Implement strategy for smart shots distribution (#7988)

    When estimating the expectation value of a Hamiltonian written as a
    linear combination of several terms, we usually need to partition it
    into groups of mutually-commuting terms which can be measured
    simultaneously. Given a fixed total number of measurement shots, a
    crucial task is to determine the optimal allocation of these shots
    across the measurements for each commuting group. The goal of this
    strategy is to finally combine the individual expectation values to
    produce the best estimate of the whole original Hamiltonian. For
    example, we might want to weight the number of shots for each group
    based on the coefficients of its terms.

    The aim of this PR is to set the ground for this "smart" shots
    distribution and to integrate this strategy into the already existing
    `split_non_commuting` transform. Here we focus on the simple case of a
    circuit measuring the expectation value of a single Hamiltonian, but the
    idea can be generalized for multiple Hamiltonians as well.

    Let's show an example considering the following case:
    ```python
    import pennylane as qml

    coeffs = [10, 0.1, 20, 100, 0.2]
    observables = [
        qml.X(0) @ qml.Y(1),
        qml.Z(0) @ qml.Z(2),
        qml.Y(1),
        qml.X(1) @ qml.X(2),
        qml.Z(0) @ qml.Z(1) @ qml.Z(2)
    ]
    shots = 10000

    ham = qml.Hamiltonian(coeffs=coeffs, observables=observables)
    tape = qml.tape.QuantumScript(measurements=[qml.expval(ham)], shots=shots)
    ```
    We want to allow the user to pass a custom function to the
    `split_non_commuting` transform in order to define a `shot_distribution`
    strategy. For example:
    ```python
    from pennylane.transforms import split_non_commuting

    def my_shot_distribution(total_shots, coeffs_per_group):
        sum_per_group = np.array([np.sum(np.abs(coeffs)) for coeffs in coeffs_per_group])
        prob_shots = sum_per_group / np.sum(sum_per_group)
        return total_shots * prob_shots

    new_tapes, _ = split_non_commuting(tape, shot_distribution=my_shot_distribution)

    for new_tape in new_tapes:
        print(new_tape.measurements, new_tape.shots)
    ```
    ```
    [expval(X(0) @ Y(1)), expval(Y(1))] Shots(total=2302)
    [expval(Z(0) @ Z(2)), expval(Z(0) @ Z(1) @ Z(2))] Shots(total=23)
    [expval(X(1) @ X(2))] Shots(total=7674)
    ```

    [sc-96319]

    ---------

    Co-authored-by: Christina Lee <[email protected]>
    Co-authored-by: Yushao Chen (Jerry) <[email protected]>
    Co-authored-by: Isaac De Vlugt <[email protected]>
    Co-authored-by: Isaac De Vlugt <[email protected]>

commit 57e07c7
Author: Marcus Edwards <[email protected]>
Date:   Wed Aug 27 14:44:16 2025 -0400

    Efficiently controlled ops conjugation (#8023)

    ------------------------------------------------------------------------------------------------------------

    **Context:** We would like to provide a way for users to specify that
    they are using a compute / uncompute pattern U, V, U† so that it may be
    efficiently controlled by applying controls to the "middle" operator V
    instead of the entire pattern.

    **Description of the Change:** Adds `ChangeOpBasis` and
    `change_op_basis` which allow users to represent a compute / uncompute
    pattern using `qml.change_op_basis(U, V, U†)`. A follow-on will add a
    decomposition rule for controlled Conjugations which just applies
    controls to V.

    **Benefits:** Decompositions using this pattern are made more efficient.

    **Possible Drawbacks:** We need to re-factor existing decompositions to
    use `ChangeOpBasis` and `change_op_basis`.

    **Related ShortCut Stories:** [sc-97008]

    ---------

    Co-authored-by: Christina Lee <[email protected]>
    Co-authored-by: Astral Cai <[email protected]>
    Co-authored-by: Pietropaolo Frisoni <[email protected]>
    Co-authored-by: ANT0N <[email protected]>

commit 62b73ce
Author: Astral Cai <[email protected]>
Date:   Wed Aug 27 12:02:43 2025 -0400

    matrix and cond now dequeues operators in arguments to the qfunc (#8119)

    **Context:**
    Applying the same thing in
    #8094 to `matrix` and
    `cond`

    **Description of the Change:**

    **Benefits:**

    **Possible Drawbacks:**

    **Related GitHub Issues:**
    [sc-98196]

    ---------

    Co-authored-by: Yushao Chen (Jerry) <[email protected]>

commit 10747ae
Author: ringo-but-quantum <[email protected]>
Date:   Wed Aug 27 09:51:35 2025 +0000

    [no ci] bump nightly version

commit 392d005
Author: ringo-but-quantum <[email protected]>
Date:   Tue Aug 26 21:28:49 2025 +0000

    [no ci] bump nightly version

commit 97052b0
Author: Christina Lee <[email protected]>
Date:   Tue Aug 26 17:28:23 2025 -0400

    [Capture] support for shot vectors and dynamic number of shots (#7652)

    **Context:**

    Program capture only allowed for shots to come from the device and only
    staticly known shots. Catalyst currently allows for dynamic numbers of
    shots. In order for program capture to serve as the frontend for
    catalyst, capture needs to be able to handle overridden shots, shot
    vectors, and abstract shots.

    **Description of the Change:**

    Adds a `shots_len` metadata to the qnode primitive. Then the first
    `shots_len` of the args describe the shots. `shots_len=0` means analytic
    and shots > 1 means a shot vector.

    **Benefits:**

    * Integration with `qml.set_shots`
    * Dynamic numbers of shots support
    * Shot vectors suport

    An example of what we can now capture is:
    ```
    jax.config.update("jax_dynamic_shapes", True)
    qml.capture.enable()

    @qml.qnode(qml.device('default.qubit', wires=1))
    def c():
        return qml.sample(wires=0), qml.expval(qml.Z(0))

    def w(num_shots):
        return qml.set_shots(c, (num_shots, 10))()

    jaxpr = jax.make_jaxpr(w)(3)
    jaxpr
    ```
    ```
    { lambda ; a:i32[]. let
        b:i32[] = mul a 1:i32[]
        c:i32[] = add 0:i32[] b
        _:i32[] = add c 10:i32[]
        d:i32[a] e:f32[] f:i32[10] g:f32[] = qnode[
          device=<default.qubit device (wires=1) at 0x17be6e480>
          execution_config=ExecutionConfig(grad_on_execution=False, use_device_gradient=True, use_device_jacobian_product=False, gradient_method='backprop', gradient_keyword_arguments={}, device_options={'max_workers': None, 'rng': Generator(PCG64) at 0x17CDB74C0, 'prng_key': None}, interface=<Interface.JAX: 'jax'>, derivative_order=1, mcm_config=MCMConfig(mcm_method='deferred', postselect_mode=None), convert_to_numpy=True, executor_backend=<class 'pennylane.concurrency.executors.native.multiproc.MPPoolExec'>)
          n_consts=0
          qfunc_jaxpr={ lambda ; . let
              h:AbstractMeasurement(n_wires=1) = sample_wires 0:i32[]
              i:AbstractOperator() = PauliZ[n_wires=1] 0:i32[]
              j:AbstractMeasurement(n_wires=None) = expval_obs i
            in (h, j) }
          qnode=<QNode: device='<default.qubit device (wires=1) at 0x17be6e480>', interface='jax', diff_method='best'>
          shots_len=2
        ] a 10:i32[]
      in (d, e, f, g) }
    ```

    **Possible Drawbacks:**

    Not integrated with native plxpr execution.

    The qnode primitive is getting clunky.

    **Related GitHub Issues:**

    [sc-97095]

    ---------

    Co-authored-by: JerryChen97 <[email protected]>

commit 2e5690c
Author: Yushao Chen (Jerry) <[email protected]>
Date:   Tue Aug 26 15:58:54 2025 -0400

    Make `qnode.shots` Public Again (#8073)

    **Context:**
    Note that long ago we abandoned the `shots` specification at `QNode`.
    However, now with many things and contexts changed drastically, we would
    like to bring back a public UI for convenience and friendliness.

    **Description of the Change:**
    - [x] a new public property `shots` has been created. Its setter is
    blocked by errors due to exclusive access of `qml.set_shots`
    - [x] `QNode` now can accept `shots` kwarg as the default shots; this
    will take precedence over the deprecated `device.shots`

    **Benefits:**

    **Possible Drawbacks:**

    **Related GitHub Issues:**
    [sc-97581]

    ---------

    Co-authored-by: Isaac De Vlugt <[email protected]>
    Co-authored-by: Christina Lee <[email protected]>

commit 6993524
Author: Yushao Chen (Jerry) <[email protected]>
Date:   Tue Aug 26 15:32:42 2025 -0400

    Skip custom measurement for ionq plugin (#8126)

    **Context:**
    With deprecation of device.shots on ionq plugin, it does not support
    analytic mode when executed. Therefore certain measurement test should
    be skipped. See
    https://github.com/PennyLaneAI/PennyLane-IonQ/actions/runs/17146349703/job/48643340932?pr=149
    for a quick look why this appears redundant.

    **Description of the Change:**
    Added a skipper for ionq device.
    Used the skipper for
    `ennylane/devices/tests/test_measurements.py::TestCustomMeasurement::test_custom_measurement`

    **Benefits:**

    **Possible Drawbacks:**
    [sc-98197]
    **Related GitHub Issues:**

commit c256b7e
Author: Christina Lee <[email protected]>
Date:   Tue Aug 26 12:29:14 2025 -0400

    more informative draw error for deep circuits (#8139)

    **Context:**

    Unhelpful error message is raised when drawing very deep circuits.

    **Description of the Change:**

    Adds context to the error message.

    **Benefits:**

    **Possible Drawbacks:**

    **Related GitHub Issues:**

    Fixes #8137 [sc-98283]

commit a346335
Author: Mudit Pandey <[email protected]>
Date:   Tue Aug 26 11:50:49 2025 -0400

    [Unified compiler] Add constraints for tensors and memrefs that allow richer constraints (#8120)

    Add new constraints for providing ways to constrain the element type,
    and either shape or rank of the containers. The new constraints allow
    constraining the exact shape of the containers, or the rank. The rank
    can be constrained to a single rank, or to a set of possible ranks.

    [sc-98077]

    ---------

    Co-authored-by: Joey Carter <[email protected]>

commit ed7aacd
Author: ringo-but-quantum <[email protected]>
Date:   Tue Aug 26 09:51:47 2025 +0000

    [no ci] bump nightly version

commit 73066dd
Author: Diksha Dhawan <[email protected]>
Date:   Mon Aug 25 21:11:20 2025 -0400

    Added a template for estimating the resources for trotterization of a Hamiltonian (#7910)

    **Context:**
    Added templates for estimating the resources for trotterization of a
    Hamiltonian based on Trotter-Suzuki formula.

    **Description of the Change:**

    **Benefits:**

    **Possible Drawbacks:**

    **Related GitHub Issues:**

    [sc-94207]

    ---------

    Co-authored-by: Jay Soni <[email protected]>
    Co-authored-by: Soran Jahangiri <[email protected]>

commit 45d6d13
Author: Mehrdad Malek <[email protected]>
Date:   Mon Aug 25 17:04:00 2025 -0400

    Extend the StableHLO dialect in the Python Compiler (#8036)

    **Context:**
    Currently StableHLO dialect in xdsl contains a small subset of
    operations available in the OpenXLA repo. A full list of categorized
    operations can be found
    [here](https://github.com/openxla/stablehlo/blob/main/docs/interpreter_status.md#stablehlo-ops-categories)

    **Description of the Change:**
    Creates a local shim which mirrors the existing upstream operations as
    well as allowing us to extend the stablehlo to include more operation.

    **Benefits:**

    **Possible Drawbacks:**
    This PR should be upstreamed once the StableHLO migration from xdsl is
    complete

    **Related GitHub Issues:**
@mudit2812
Merge branch 'master' into dialect-type-fixes
e8120c2
@mudit2812
Add filecheck tests for allocation/deallocation, insertion/extraction
a3e18d6
@mudit2812
Add filecheck tests for state prep operations
6cfafe3
@mudit2812
Add filecheck tests for miscelleneous operations
0262a1b
@mudit2812
Make run_filecheck logs better
120c49d
@mudit2812
Add tests for observables and measurements
6ec330d
@mudit2812
Add gate/MCM tests
94ea69b
@mudit2812
Merge branch 'master' into dialect-type-fixes
4357a85
@mudit2812 mudit2812 marked this pull request as ready for review September 3, 2025 20:57
@github-actions
Copy link
Contributor

github-actions bot commented Sep 3, 2025

Hello. You may have forgotten to update the changelog!
Please edit doc/releases/changelog-dev.md with:

  • A one-to-two sentence description of the change. You may include a small working example for new features.
  • A link back to this PR.
  • Your name (or GitHub username) in the contributors section.

@codecov
Copy link

codecov bot commented Sep 3, 2025 •
edited
Loading

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.69%. Comparing base (e83288f) to head (aa872d4).
⚠️ Report is 6 commits behind head on master.

Additional details and impacted files 
@@           Coverage Diff           @@
##           master    #8083   +/-   ##
=======================================
  Coverage   99.69%   99.69%           
=======================================
  Files         548      548           
  Lines       56845    56845           
=======================================
  Hits        56672    56672           
  Misses        173      173

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

mehrdad2m
mehrdad2m requested changes Sep 5, 2025
View reviewed changes
Copy link
Contributor

@mehrdad2m mehrdad2m left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks Great. Thanks @mudit2812! Just a couple of minor comments!

mehrdad2m
mehrdad2m approved these changes Sep 5, 2025
View reviewed changes
Copy link
Contributor

@mehrdad2m mehrdad2m left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🎉

PietropaoloFrisoni
PietropaoloFrisoni reviewed Sep 5, 2025
View reviewed changes
Copy link
Contributor

@PietropaoloFrisoni PietropaoloFrisoni left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks @mudit2812, looks good. I just have some minor questions

PietropaoloFrisoni
PietropaoloFrisoni approved these changes Sep 5, 2025
View reviewed changes
Copy link
Contributor

@PietropaoloFrisoni PietropaoloFrisoni left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks @mudit2812 , happy to approve.

When you have a chance, please don't forget the PR(s) for visualization : )

@mudit2812 mudit2812 added this pull request to the merge queue Sep 5, 2025
Merged via the queue into master with commit 16c29ac Sep 5, 2025
52 checks passed
@mudit2812 mudit2812 deleted the dialect-type-fixes branch September 5, 2025 22:39
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@mehrdad2m mehrdad2m mehrdad2m approved these changes

+1 more reviewer

@PietropaoloFrisoni PietropaoloFrisoni PietropaoloFrisoni approved these changes

Reviewers whose approvals may not affect merge requirements

Assignees

No one assigned

Labels

unified compiler Issues and pull requests related to PennyLane's xDSL python compiler.

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

4 participants

@mudit2812 @mehrdad2m @PietropaoloFrisoni