Add cirq.google.optimized_for_xmon() #647

Strilanc · 2018-07-12T06:54:20Z

This is an all-in-one method for producing xmon-optimized circuits.

For example, it decomposes this:

0: ───iSwap───────────iSwap───M───
      │               │       │
1: ───iSwap───iSwap───iSwap───M───
              │               │
2: ───iSwap───iSwap───iSwap───M───
      │               │       │
3: ───iSwap───────────iSwap───M───

into this:

0: ───W(0.25)^-0.5───@───W(0.25)^0.5────@───W(0.75)^0.5──────────────────────────────────────────@───W(0.25)^-0.5───@───W(0.25)^0.5────M───
                     │                  │                                                        │                  │                  │
1: ───W(0.25)^0.5────@───W(0.25)^-0.5───@───W(0.75)^0.5────@───W(0.25)^0.5────@───W(0.75)^-0.5───@───W(0.25)^-0.5───@───W(0.25)^0.5────M───
                                                           │                  │                                                        │
2: ───W(0.25)^-0.5───@───W(0.25)^0.5────@───W(0.75)^-0.5───@───W(0.25)^-0.5───@───W(0.75)^0.5────@───W(0.25)^0.5────@───W(0.25)^-0.5───M───
                     │                  │                                                        │                  │                  │
3: ───W(0.25)^0.5────@───W(0.25)^-0.5───@───W(0.75)^-0.5─────────────────────────────────────────@───W(0.25)^0.5────@───W(0.25)^-0.5───M───

Obsoletes #488
Obsoletes #644
Obsoletes #645

This PR is just to keep track of experimentation into how to make the optimizers better I will peel useful sub-PRs off of it until it is a no-op.

- Use same style as EjectZ optimizer - Add with_bits_flipped to MeasurementGate - Fixes to_unitary_matrix ignoring bit flips from measurement gates Fixes #642

…_cleanup # Conflicts: # cirq/google/eject_z.py

# Conflicts: # cirq/google/eject_z.py

…_cleanup # Conflicts: # cirq/circuits/circuit.py # cirq/circuits/optimization_pass.py # cirq/google/eject_full_w.py # cirq/google/eject_z.py

- Slipped by because of a silent merge conflict overwriting updated methods with old ones - The old methods were calling the new "canonicalize_up_to_measurement_phase" with the wrong number of arguments, triggering a type error, triggering a path expecting a type error for a *different reason* (i.e. non-unitary operations in the circuit) - Removed the non-unitary fallback path and renamed the method to make it clear only measurement-implies-terminal circuits were expected - Simplified the W-over-partial-W derivation - Fixed false-positive assertions that were now correctly failing - Added a test that fails for the old behavior but not for the new behavior Fixes #684

# Conflicts: # cirq/google/eject_z_test.py

- Updated the cost metric to include presence-of-non-xmon - Converting gates to xmon before returning

# Conflicts: # cirq/google/merge_interactions.py # cirq/google/merge_interactions_test.py Signed-off-by: Craig Gidney <craiggidney@google.com>

…_cleanup

…alent - Use it in eject_z_test and eject_full_w_test Fixes #670

# Conflicts: # cirq/google/eject_full_w_test.py # cirq/google/eject_z_test.py

# Conflicts: # cirq/google/merge_rotations_test.py

…_cleanup # Conflicts: # cirq/google/eject_full_w_test.py # cirq/google/eject_z_test.py # cirq/google/xmon_gates.py # cirq/testing/circuit_compare.py

…_cleanup # Conflicts: # cirq/google/merge_interactions.py

cduck · 2018-07-18T00:51:43Z

cirq/google/optimize_test.py

+
+    after = cg.optimized_for_xmon(before)
+
+    assert len(after) <= 22


Is the optimized length currently 22?

Yes. But I'm happy if it goes down, too.

cduck · 2018-07-18T00:54:35Z

cirq/google/optimize_test.py

+    before = cirq.Circuit.from_ops(
+        cirq.CCZ(cirq.GridQubit(5, 5),
+                 cirq.GridQubit(5, 6),
+                 cirq.GridQubit(5, 7)))


No terminal measurements? Would it be useful to have separate tests for both cases (with measurements and without)?

Intentional.

cduck

LGTM.

[WIP] optimization playground

0f4a4b5

This PR is just to keep track of experimentation into how to make the optimizers better I will peel useful sub-PRs off of it until it is a no-op.

googlebot added the cla: yes Makes googlebot stop complaining. label Jul 12, 2018

Strilanc added 28 commits July 14, 2018 19:42

Implement EjectFullW optimizer

f1073c0

- Use same style as EjectZ optimizer - Add with_bits_flipped to MeasurementGate - Fixes to_unitary_matrix ignoring bit flips from measurement gates Fixes #642

doc

de00b46

Merge branch 'master' of github.com:quantumlib/cirq into optimization…

cbdb073

…_cleanup # Conflicts: # cirq/google/eject_z.py

r

01c98b2

Merge branch 'diff3' into optimization_cleanup

977b02b

# Conflicts: # cirq/google/eject_z.py

Define cirq.google.optimize_for_xmon

ee07218

Merge branch 'master' of github.com:quantumlib/cirq into optimization…

fc727bc

…_cleanup # Conflicts: # cirq/circuits/circuit.py # cirq/circuits/optimization_pass.py # cirq/google/eject_full_w.py # cirq/google/eject_z.py

condition

4a46caf

cover

36ed1e0

Merge branch 'master' of github.com:quantumlib/cirq into fix-w

18c1f77

Merge branch 'fix-w' into optimization_cleanup

b25a890

# Conflicts: # cirq/google/eject_z_test.py

fix

a6b0a74

Merge branch 'fix-w' into optimization_cleanup

08b7416

test_ccz

c0bfb40

Refactor drop_negligible into an OptimizationPass using batch_remove

72d9aee

Fix merge_interactions failing to convert into xmon gates

7cd58e6

- Updated the cost metric to include presence-of-non-xmon - Converting gates to xmon before returning

Merge branch 'master' of github.com:quantumlib/cirq into optimize-1

e9c5acb

fix tests

185b517

Merge branch 'optimize-0' into optimization_cleanup

07c1fde

Merge branch 'optimize-1' into optimization_cleanup

094b3cb

# Conflicts: # cirq/google/merge_interactions.py # cirq/google/merge_interactions_test.py Signed-off-by: Craig Gidney <craiggidney@google.com>

cut bool

3fca893

naming things

86dc1d0

Merge branch 'optimize-1' into optimization_cleanup

091aebe

Cleanup imports in merge_rotations_test

27cadfb

Merge branch 'master' of github.com:quantumlib/cirq into optimization…

6a64583

…_cleanup

Add cirq.testing.assert_circuits_with_terminal_measurements_are_equiv…

1febf3d

…alent - Use it in eject_z_test and eject_full_w_test Fixes #670

Strilanc added 5 commits July 17, 2018 01:50

lint

9de3e03

compare_unitaries

ff8a4fc

dindent

dc26303

Merge branch 'optimize-3' into optimization_cleanup

c511210

# Conflicts: # cirq/google/eject_full_w_test.py # cirq/google/eject_z_test.py

Merge branch 'optimize-2' into optimization_cleanup

d8aa6c5

# Conflicts: # cirq/google/merge_rotations_test.py

Strilanc changed the title ~~[WIP] optimization playground~~ [AFTER #699, #696, #695, #691] Add cirq.google.optimized_for_xmon() Jul 17, 2018

Strilanc changed the title ~~[AFTER #699, #696, #695, #691] Add cirq.google.optimized_for_xmon()~~ [AFTER #700, #699, #696, #695, #691] Add cirq.google.optimized_for_xmon() Jul 17, 2018

Strilanc added 3 commits July 17, 2018 02:39

fix

75b9501

Fix

2fb7103

lint

37ef6ff

Strilanc requested review from cduck and dabacon July 17, 2018 17:51

Strilanc changed the title ~~[AFTER #700, #699, #696, #695, #691] Add cirq.google.optimized_for_xmon()~~ [AFTER #696] Add cirq.google.optimized_for_xmon() Jul 17, 2018

Strilanc added 2 commits July 17, 2018 16:15

Merge branch 'master' of github.com:quantumlib/cirq into optimization…

9007754

…_cleanup # Conflicts: # cirq/google/eject_full_w_test.py # cirq/google/eject_z_test.py # cirq/google/xmon_gates.py # cirq/testing/circuit_compare.py

Merge branch 'master' of github.com:quantumlib/cirq into optimization…

296bdea

…_cleanup # Conflicts: # cirq/google/merge_interactions.py

Strilanc changed the title ~~[AFTER #696] Add cirq.google.optimized_for_xmon()~~ Add cirq.google.optimized_for_xmon() Jul 18, 2018

cduck reviewed Jul 18, 2018

View reviewed changes

cduck approved these changes Jul 18, 2018

View reviewed changes

Strilanc merged commit 7d0ca0d into master Jul 18, 2018

Strilanc deleted the optimization_cleanup branch July 18, 2018 00:59

This was referenced Jul 18, 2018

MergeInteractions step should do better at single-qubit gate outputs #488

Closed

Add 'followup' optimizations to PointOptimizer #644

Closed

PointOptimizer should automatically drop emptied moments #645

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Add cirq.google.optimized_for_xmon() #647

Add cirq.google.optimized_for_xmon() #647

Uh oh!

Strilanc commented Jul 12, 2018 •

edited

Loading

Uh oh!

cduck Jul 18, 2018

Uh oh!

Strilanc Jul 18, 2018

Uh oh!

cduck Jul 18, 2018

Uh oh!

Strilanc Jul 18, 2018

Uh oh!

cduck left a comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants


		after = cg.optimized_for_xmon(before)

		assert len(after) <= 22

Add cirq.google.optimized_for_xmon() #647

Add cirq.google.optimized_for_xmon() #647

Uh oh!

Conversation

Strilanc commented Jul 12, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

cduck Jul 18, 2018

Choose a reason for hiding this comment

Uh oh!

Strilanc Jul 18, 2018

Choose a reason for hiding this comment

Uh oh!

cduck Jul 18, 2018

Choose a reason for hiding this comment

Uh oh!

Strilanc Jul 18, 2018

Choose a reason for hiding this comment

Uh oh!

cduck left a comment

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Strilanc commented Jul 12, 2018 •

edited

Loading