Implement pyqrack interpreter methods for squin dialect #207

david-pl · 2025-04-25T15:02:22Z

Closes #185

github-actions · 2025-04-25T15:04:34Z

☂️ Python Coverage

current status: ✅

Overall Coverage

Lines	Covered	Coverage	Threshold	Status
6844	5887	86%	0%	🟢

New Files

File	Coverage	Status
src/bloqade/pyqrack/squin/_init_.py	100%	🟢
src/bloqade/pyqrack/squin/op.py	100%	🟢
src/bloqade/pyqrack/squin/qubit.py	98%	🟢
src/bloqade/pyqrack/squin/runtime.py	70%	🟢
TOTAL	92%	🟢

Modified Files

File	Coverage	Status
src/bloqade/pyqrack/_init_.py	100%	🟢
src/bloqade/pyqrack/reg.py	95%	🟢
src/bloqade/squin/analysis/nsites/impls.py	93%	🟢
src/bloqade/squin/op/_init_.py	88%	🟢
src/bloqade/squin/op/stmts.py	99%	🟢
src/bloqade/squin/qubit.py	92%	🟢
src/bloqade/squin/wire.py	100%	🟢
TOTAL	95%	🟢

updated for commit: 0b93fd0 by action🐍

codecov · 2025-04-25T15:04:50Z

Codecov Report

Attention: Patch coverage is 79.73568% with 92 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/bloqade/pyqrack/squin/runtime.py	69.69%	90 Missing ⚠️
src/bloqade/pyqrack/squin/qubit.py	98.07%	1 Missing ⚠️
src/bloqade/squin/op/__init__.py	92.30%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

david-pl · 2025-04-29T10:20:04Z

So, the implementation here is somewhat complete. However, I'm still struggling with the wire dialect.

@weinbe58 @johnzl-777 how do you even create a Wire right now? I tried this:

@squin.wired
def main():
    q = squin.qubit.new(1)
    w = squin.wire.unwrap(q[0])
    x = squin.op.x()
    squin.wire.apply(x, w)
    return w

But, of course, squin.qubit.new isn't supported by the wired dialect. Is there something like a NewWire statement that is missing?

src/bloqade/pyqrack/squin/qubit.py

weinbe58 · 2025-05-02T16:43:25Z

@david-pl I can handle the measurement stuff later on thanks you!

weinbe58

I have some suggestions mostly for the runtime.

src/bloqade/pyqrack/squin/qubit.py

weinbe58 · 2025-05-02T18:19:13Z

src/bloqade/pyqrack/squin/runtime.py

+    def broadcast_apply(self, qubits: ilist.IList[PyQrackQubit, Any], **kwargs) -> None:
+        for qbit in qubits:
+            if not qbit.is_active():
+                continue
+
+            self.apply(qbit, **kwargs)


What if the operator accepts two qubits? you should batch the qubits based on the operator size.

You mean something like broadcast(CX, ...)? How would the wrapper for this look, i.e. what's the second argument to broadcast(CX, ...)? Not sure about the syntax here.

src/bloqade/pyqrack/squin/runtime.py

weinbe58 · 2025-05-02T18:29:06Z

src/bloqade/pyqrack/squin/runtime.py

+    def apply(self, *qubits: PyQrackQubit, adjoint: bool = False) -> None:
+        self.op.apply(*qubits, adjoint=adjoint)
+
+        target = qubits[-1]
+
+        # NOTE: just factor * eye(2)
+        m = self.mat(adjoint)
+
+        # TODO: output seems to always be normalized -- no-op?
+        target.sim_reg.mtrx(m, target.addr)


if this is true we might want to error instead of allowing things to run. If we want to do something like rot(scale * op) we can rewrite this to a standard rotation statement.

cc: @Roger-luo

weinbe58 · 2025-05-02T18:33:58Z

src/bloqade/squin/op/__init__.py

+@_wraps(stmts.Mult)
+def mult(lhs: types.Op, rhs: types.Op) -> types.Op: ...
+
+
+@_wraps(stmts.Scale)
+def scale(op: types.Op, factor: complex) -> types.Op: ...


 @_wraps(stmts.Adjoint)
-def adjoint(op: types.Op, *, is_unitary: bool = False) -> types.Op: ...
+def adjoint(op: types.Op) -> types.Op: ...


 @_wraps(stmts.Control)
-def control(op: types.Op, *, n_controls: int, is_unitary: bool = False) -> types.Op: ...
+def control(op: types.Op, *, n_controls: int) -> types.Op: ...


Remove wrappers later when syntax sugar is implemented.

Just out of curiosity: why? What's the hurt in keeping the wrapper?

what syntax sugar are we proposing here for control?

src/bloqade/pyqrack/squin/qubit.py

weinbe58 · 2025-05-02T18:39:26Z

src/bloqade/pyqrack/squin/wire.py

+        w: PyQrackWire = frame.get(stmt.wire)
+        qbit = w.qubit
+        res: int = qbit.sim_reg.m(qbit.addr)
+        qbit.sim_reg.force_m(qbit.addr, False)
+        new_w = PyQrackWire(qbit)
+        return (new_w, res)


I need to make sure this will work in general. I know using force_m can create some issues. Lets leave it for now but I will revisit how to properly reset a qubit with pyqrack.

Co-authored-by: Phillip Weinberg <[email protected]>

david-pl · 2025-05-03T13:56:51Z

Since this PR is getting kind of long, I'm gonna summarize some of the discussions again here. For the most part, I think there was a bit of confusion as I originally only considered a single target for controlled gates and broadcast (which was wrong). That has been fixed.

Here's an overview of the semantics we discussed in some comments and how I chose to implement them for now:

In a Control you can have arbitrarily many controls and targets.
If any of the controls are inactive, no gate is applied to any target.
If the number of controls + the size of the applied operator doesn't match the number of qubits, you get a RuntimeError.
If any of the targets are inactive, they are skipped. The gate is still applied to the remaining, active targets.
The number of targets must match the size (I called the property n_qubits) of the operator that is to be applied to the targets (not just for controls, obviously).
You can broadcast operators of arbitrary size. It's applied to the list of qubits in batches that match the size of the operator from left to right.
That means, you can also broadcast controls, where e.g. CX is then applied to a list of qubits, where each pair is treated as a control and target, respectively. I'm not a huge fan of this since I think it might still be confusing, but I see the value in it.

Pinging @cduck , @Roger-luo and @weinbe58 here. Do you all agree with this?

Also, ready for another round of review from my side!

Roger-luo · 2025-05-03T20:34:31Z

Yes, I agree, I think these are good summarize of the semantics, worth putting them in the docstring of the wrappers later.

just to be pedantic here

The number of targets must match the size (I called the property n_qubits) of the operator that is to be applied to the targets (not just for controls, obviously).

we should call it n_sites because it is not necessarily n_qubits for operators.

weinbe58 · 2025-05-05T17:01:09Z

src/bloqade/pyqrack/squin/qubit.py

+    @interp.impl(qubit.MeasureAny)
+    def measure_any(
+        self, interp: PyQrackInterpreter, frame: interp.Frame, stmt: qubit.MeasureAny
+    ):
+        input = frame.get(stmt.input)
+
+        if isinstance(input, PyQrackQubit) and input.is_active():
+            result = self._measure_qubit
+        elif isinstance(input, ilist.IList):
+            result = []
+            for qbit in input:
+                if not isinstance(qbit, PyQrackQubit):
+                    raise InterpreterError(f"Cannot measure {type(qbit).__name__}")
+
+                result.append(self._measure_qubit(qbit))
+        else:
+            raise InterpreterError(f"Cannot measure {type(input).__name__}")
+
+        return (result,)


This statement is only for syntax sugar not for runtime. can you remove this?

weinbe58

LGTM, if we run into any issues we can polish them later.

@Roger-luo

Seemed sufficiently different from #207, so I created a new branch for it. @Roger-luo two things: * I'm not sure whether this should be a `Fixpoint`. * Should this also be part of the `wire` kernel? --------- Co-authored-by: Phillip Weinberg <[email protected]>

david-pl mentioned this pull request Apr 28, 2025

Rewrite py.Mult to squin.Mult in squin kernel #212

Merged

david-pl force-pushed the david/185-pyqrack-squin branch from 9d2b332 to 19c95dc Compare April 28, 2025 15:11

david-pl force-pushed the david/185-pyqrack-squin branch from e6d79f4 to 0cf3d39 Compare April 29, 2025 11:13

david-pl commented Apr 29, 2025

View reviewed changes

src/bloqade/pyqrack/squin/qubit.py Outdated Show resolved Hide resolved

david-pl added 23 commits April 30, 2025 09:17

Fix identity wrapper argument

1b6d42d

Fix measure wrapper typing

eed5a98

Implement methods for qubit statements

ecc641f

Start implementing operator runtime

2b9da28

Change operator runtime implementation

1dc73e5

Implement mult runtime -- TODO: don't wrap mult

61df76a

Implement kron runtime

b475a76

Scale runtime

a750374

Rework impl of control to work with all supported operators

3153f5b

Runtime for phase and shift operators

22dbc74

Implement Sp/Sn runtime

cb0f0a6

Factor 1/2 in Sn and Sp

d77422a

Fix method name for adjoints

fd08501

Implement simple rotations about X, Y, Z

c328d5e

Fix test and bug

8727acc

Implement (somewhat strange) control apply for Krons

e86c267

Runtime for broadcast

c99bb4d

Implement runtime for U3

3280b47

Implement runtime for CliffordString

b358747

Draft implementation for wire dialect

ca94e55

Check whether qubits are active before applying operator

1d7bfbd

Check is_active in multiple places

f1e6efd

Fix phase runtime and test

0b99977

This was referenced May 2, 2025

Adding noise statements to squin #211

Merged

Remove is_unitary from wrapper. #204

Closed

weinbe58 requested changes May 2, 2025

View reviewed changes

david-pl and others added 7 commits May 2, 2025 21:12

cast qubit measure to bool

c77c205

Co-authored-by: Phillip Weinberg <[email protected]>

Update apply signature for single qubits

6dd413c

Co-authored-by: Phillip Weinberg <[email protected]>

Address PR comments

69b42c5

MeasuremeQubitList returns IList

d298078

Properly check for atom loss in apply methods

a35fb61

Fix tests

fffd492

Fix typing of measure_and_reset

f5eaa8e

david-pl requested a review from weinbe58 May 2, 2025 20:48

david-pl added 3 commits May 3, 2025 15:26

Properly account for operator size in runtime

fba5a2f

Properly deal with nested adjoints

e488852

Fix broadcasting for operators with size larger 1

59f52ab

Add a test for CXX gate

636637f

david-pl added 3 commits May 5, 2025 08:49

Rename n_qubits to n_sites

711e484

Fix applied factor in scale

df4160f

Add some more info to docstrings

96b298c

weinbe58 requested changes May 5, 2025

View reviewed changes

weinbe58 mentioned this pull request May 5, 2025

Implement a better method to reset qubits in pyqrack runtime. #229

Closed

Remove MeasureAny impl

0b93fd0

david-pl requested a review from weinbe58 May 5, 2025 17:34

weinbe58 approved these changes May 5, 2025

View reviewed changes

weinbe58 merged commit cdf3280 into main May 5, 2025
10 of 11 checks passed

weinbe58 deleted the david/185-pyqrack-squin branch May 5, 2025 19:06

david-pl mentioned this pull request May 6, 2025

Fix resetting qubits in pyqrack runtime #233

Merged

Implement pyqrack interpreter methods for squin dialect #207

Implement pyqrack interpreter methods for squin dialect #207

Uh oh!

Conversation

david-pl commented Apr 25, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented Apr 25, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

☂️ Python Coverage

Overall Coverage

New Files

Modified Files

Uh oh!

codecov bot commented Apr 25, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

david-pl commented Apr 29, 2025

Uh oh!

Uh oh!

weinbe58 commented May 2, 2025

Uh oh!

weinbe58 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

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!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

david-pl commented May 3, 2025

Uh oh!

Roger-luo commented May 3, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

weinbe58 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

david-pl commented Apr 25, 2025 •

edited

Loading

github-actions bot commented Apr 25, 2025 •

edited

Loading

codecov bot commented Apr 25, 2025 •

edited

Loading