Add CP code base by alexzucca90 · Pull Request #482 · dwavesystems/dwave-optimization

alexzucca90 · 2026-02-27T23:52:50Z

Basic structure for integrating domains and filtering routines.

wbernoudy · 2026-03-09T23:28:57Z

dwave/optimization/include/dwave-optimization/cp/core/advisor.hpp

+#pragma once
+#include "dwave-optimization/cp/core/index_transform.hpp"
+#include "dwave-optimization/cp/core/propagator.hpp"
+namespace dwave::optimization::cp {


Minor but I would prefer constraint_propagation or similar as a "top-level" name instead of cp. Abbreviating it on less public interfaces or within methods is fine.

to be precise you mean

namespace dwave::optimization::constraint_propagation {

?

Yes. I could be convinced otherwise though!

wbernoudy · 2026-03-09T23:30:55Z

dwave/optimization/include/dwave-optimization/cp/core/advisor.hpp

+#include "dwave-optimization/cp/core/propagator.hpp"
+namespace dwave::optimization::cp {
+
+/// Utility class to help schedule propagators when variable change their domains


Suggested change

/// Utility class to help schedule propagators when variable change their domains

/// Utility class to help schedule propagators when variables change their domains

wbernoudy · 2026-03-09T23:39:28Z

dwave/optimization/include/dwave-optimization/cp/core/advisor.hpp

+/// - map variable index to propagator
+class Advisor {
+ public:
+    Advisor(Propagator* p, ssize_t p_input, std::unique_ptr<IndexTransform> index_transform);


p_input is the topological index of a node?

Actually it seems like it is not used?

Good catch. I added it in case some propagators need to know which variable triggered the propagator. p_input is which input of that propagator the variable is. I'd wait a bit more to remove it for now, just in case I need it soon.

Re-explaining this to make sure I get it: each propagator is associated with two or more variables (i.e. CPVars). When the propagator is called, it wants to be efficient in terms of which of its "output" indices need to be re-evaluated (e.g. shrinking bounds of an output index, if it's a bound consistency propagator). This is the role of the Advisor: each propagator adds an Advisor to each of its associated variables, and that Advisor owns an IndexTransform which translates indices of the associated variable to change output indices.

However, it also seems reasonable that a propagator may want to know not just which of its output indices needs to be re-evaluated, but also which of the associated variables that change came from. So you are allowing for that info to be passed through here.

Now correct where I went wrong 😆

I missed this comment but you are correct!

wbernoudy · 2026-03-09T23:47:05Z

dwave/optimization/include/dwave-optimization/cp/core/cpvar.hpp

+    ssize_t num_propagators() const { return propagators_.size(); }
+
+ protected:
+    bool locked_ = false;


Seems like locked_ is never set by anything? Am I missing a lock() method somewhere?

good catch again, haven't worked on that yet.
And sorry for my work being somewhat nonlinear, I add stuff that I think I need but I don't complete until later..

No worries. Is the idea here to lock the CP model after parsing the dwopt model, and then we can be sure that the model is not modified during the run?

yes, that's right!

wbernoudy · 2026-03-09T23:55:18Z

dwave/optimization/include/dwave-optimization/cp/core/cpvar.hpp

+    virtual ~CPVar() = default;
+
+    // constructor
+    CPVar(const CPModel& model, const dwave::optimization::ArrayNode* node_ptr, int index);


This index is the topological index of the node_ptr, correct? Why not just derive it from the node_ptr directly in the implementation?

good point!

wbernoudy · 2026-03-09T23:58:40Z

dwave/optimization/src/cp/core/cpvar.cpp

+
+    ssize_t max_size, min_size;
+
+    if (node_->size() > 0) {


Suggested change

if (node_->size() > 0) {

if (not node_->dynamic()) {

You also want to handle size 0 arrays here, right?

Yeah good catch!

wbernoudy · 2026-03-09T23:59:30Z

dwave/optimization/src/cp/core/cpvar.cpp

+    assert(this->cp_var_index_ >= 0);
+    assert(static_cast<ssize_t>(state.var_state_.size()) > cp_var_index_);
+
+    // TODO: for now we don't handle dynamically sized nodes


Seems like this does handle dynamic arrays now, as long as they have bounds on size

oh yeah, obsolete comment!

wbernoudy · 2026-03-09T23:59:59Z

dwave/optimization/src/cp/core/cpvar.cpp

+            throw std::invalid_argument("Array has no max size available");
+        }
+
+        if (not sizeinfo.min.has_value()) {


It should be safe to substitute zero in here, though that may not be desirable

Good point, I want to check again how sizeinfo works and I'll likely follow your suggestion

wbernoudy · 2026-03-10T00:01:00Z

dwave/optimization/src/cp/core/cpvar.cpp

+
+    if (node_->size() > 0) {
+        // Static array
+        max_size = node_->size();


You could use sizeinfo.min/max for these and simplify the logic I think

wbernoudy · 2026-03-10T00:17:57Z

dwave/optimization/include/dwave-optimization/cp/core/domain_listener.hpp

+namespace dwave::optimization::cp {
+class DomainListener {
+ public:
+    virtual ~DomainListener() = default;


I think if you're defining a virtual destructor like this, you should also declare the copy and move constructors to follow the rule of three, probably as = delete to prevent object slicing.

The plan is to have other types of listeners (other than CPVar::listener) at some point right?

The point is that if there are other variables types (other than CPVar) then I may need other types of listeners. But to decouple domains from variables, the domains respond to this interface to trigger propagators.

Anyways, yes I need to follow the rule of three! I'll add that!

fastbodin · 2026-03-11T17:58:53Z

dwave/optimization/include/dwave-optimization/cp/core/domain_array.hpp

+    /// Return the minimum size of the array
+    virtual ssize_t min_size() const = 0;
+
+    /// Return the minimum size of the array


fastbodin · 2026-03-11T18:06:55Z

dwave/optimization/include/dwave-optimization/cp/core/interval_array.hpp

+    IntervalArray(StateManager* sm, ssize_t min_size, ssize_t max_size, double lb, double up);
+    IntervalArray(StateManager* sm, ssize_t size, double lb, double up);


I think you mean double ub here?

smharwood · 2026-03-12T16:53:25Z

dwave/optimization/include/dwave-optimization/cp/core/interval_array.hpp

+    // Change double do an object that can be backtracked.
+    // And maybe get


Suggested change

// Change double do an object that can be backtracked.

// And maybe get

Looks like an old TODO that has been addressed?

smharwood · 2026-03-13T16:04:23Z

dwave/optimization/include/dwave-optimization/cp/core/domain_array.hpp

+    /// Return the size of the domain of the index-th variable
+    virtual double size(int index) const = 0;


I might prefer to call this "width" (consistent with the width of a real-valued interval) so that the term "size" is not overloaded so much

hmm, though if we support sparse set domains, something like "cardinality" might be better, though that would not be accurate for a real-valued interval

Add basic binary ops and reduce

Add CP code base

dc9ff26

alexzucca90 requested review from fastbodin, smharwood and wbernoudy February 27, 2026 23:52

alexzucca90 marked this pull request as draft February 27, 2026 23:53

azucca added 17 commits February 27, 2026 16:17

Add licence header

ef5127a

Unify int64_t->ssize_t

612a069

Fix warnings

bca5f13

Fix ssize_t and bool warnings

8ef6bc7

Include header

3e1fa42

Fix int64_t -> ssize_t

abff5d1

Remove std::move in emplace_back

9e0de2b

Fix int64_t -> ssize_t

2d5cf5f

Fix int64_t -> ssize_t

baacb91

Include cstddef header

aae8673

Include common

1b42107

Support dynamically sized arrays in CP

42aa8b4

Support scheduling propagators via indices

aa8e594

Remove printouts

c6794eb

Initialize scheduled_ bool to false

07e4afb

Add getter method for input index

9f24340

Add virtual destructor

dafc3f0

wbernoudy reviewed Mar 10, 2026

View reviewed changes

fastbodin reviewed Mar 11, 2026

View reviewed changes

smharwood reviewed Mar 12, 2026

View reviewed changes

Add basic binary ops and reduce

b3bcf72

smharwood reviewed Mar 13, 2026

View reviewed changes

Merge pull request #2 from alexzucca90/dev/cp_extra

f0ed0e8

Add basic binary ops and reduce

	/// Utility class to help schedule propagators when variable change their domains
	/// Utility class to help schedule propagators when variables change their domains

		IntervalArray(StateManager* sm, ssize_t min_size, ssize_t max_size, double lb, double up);
		IntervalArray(StateManager* sm, ssize_t size, double lb, double up);

		// Change double do an object that can be backtracked.
		// And maybe get

		/// Return the size of the domain of the index-th variable
		virtual double size(int index) const = 0;

Conversation

alexzucca90 commented Feb 27, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

fastbodin Mar 11, 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!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

fastbodin Mar 11, 2026 •

edited

Loading