Skip to content
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

Set max_trials for VF2Layout in preset pass managers. #10054

Merged
merged 6 commits into from
May 2, 2023
Merged

Set max_trials for VF2Layout in preset pass managers. #10054

Show file tree
Hide file tree
Changes from 4 commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
1814a5a
Set max_trials for VF2Layout in preset pass managers.
kevinhartman May 1, 2023
9c0e5e8
Address review comments.
kevinhartman May 2, 2023
0e315eb
Return tuple of None instead for finer control and a better interface.
kevinhartman May 2, 2023
9a6a4ca
Add deprecation notice to release note.
kevinhartman May 2, 2023
5ef7a94
Remove deprecation until 0.25.0.
kevinhartman May 2, 2023
89148c9
Remove unused import.
kevinhartman May 2, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 15 additions & 4 deletions qiskit/transpiler/preset_passmanagers/builtin_plugins.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -37,7 +37,7 @@ def pass_manager(self, pass_manager_config, optimization_level=None) -> PassMana
else:
routing_pass = BasicSwap(target)

vf2_call_limit = common.get_vf2_call_limit(
vf2_call_limit, vf2_max_trials = common.get_vf2_limits(
optimization_level,
pass_manager_config.layout_method,
pass_manager_config.initial_layout,
Expand All @@ -56,6 +56,7 @@ def pass_manager(self, pass_manager_config, optimization_level=None) -> PassMana
target,
coupling_map,
vf2_call_limit=vf2_call_limit,
vf2_max_trials=vf2_max_trials,
backend_properties=backend_properties,
seed_transpiler=seed_transpiler,
check_trivial=True,
Expand All @@ -67,6 +68,7 @@ def pass_manager(self, pass_manager_config, optimization_level=None) -> PassMana
target,
coupling_map=coupling_map,
vf2_call_limit=vf2_call_limit,
vf2_max_trials=vf2_max_trials,
backend_properties=backend_properties,
seed_transpiler=seed_transpiler,
use_barrier_before_measurement=True,
Expand All @@ -77,6 +79,7 @@ def pass_manager(self, pass_manager_config, optimization_level=None) -> PassMana
target,
coupling_map=coupling_map,
vf2_call_limit=vf2_call_limit,
vf2_max_trials=vf2_max_trials,
backend_properties=backend_properties,
seed_transpiler=seed_transpiler,
use_barrier_before_measurement=True,
Expand All @@ -96,7 +99,7 @@ def pass_manager(self, pass_manager_config, optimization_level=None) -> PassMana
if coupling_map_routing is None:
coupling_map_routing = coupling_map
backend_properties = pass_manager_config.backend_properties
vf2_call_limit = common.get_vf2_call_limit(
vf2_call_limit, vf2_max_trials = common.get_vf2_limits(
optimization_level,
pass_manager_config.layout_method,
pass_manager_config.initial_layout,
Expand All @@ -120,6 +123,7 @@ def pass_manager(self, pass_manager_config, optimization_level=None) -> PassMana
target,
coupling_map,
vf2_call_limit=vf2_call_limit,
vf2_max_trials=vf2_max_trials,
backend_properties=backend_properties,
seed_transpiler=seed_transpiler,
check_trivial=True,
Expand All @@ -131,6 +135,7 @@ def pass_manager(self, pass_manager_config, optimization_level=None) -> PassMana
target,
coupling_map=coupling_map,
vf2_call_limit=vf2_call_limit,
vf2_max_trials=vf2_max_trials,
backend_properties=backend_properties,
seed_transpiler=seed_transpiler,
use_barrier_before_measurement=True,
Expand All @@ -150,7 +155,7 @@ def pass_manager(self, pass_manager_config, optimization_level=None) -> PassMana
if coupling_map_routing is None:
coupling_map_routing = coupling_map
backend_properties = pass_manager_config.backend_properties
vf2_call_limit = common.get_vf2_call_limit(
vf2_call_limit, vf2_max_trials = common.get_vf2_limits(
optimization_level,
pass_manager_config.layout_method,
pass_manager_config.initial_layout,
Expand All @@ -171,6 +176,7 @@ def pass_manager(self, pass_manager_config, optimization_level=None) -> PassMana
target,
coupling_map,
vf2_call_limit=vf2_call_limit,
vf2_max_trials=vf2_max_trials,
backend_properties=backend_properties,
seed_transpiler=seed_transpiler,
check_trivial=True,
Expand All @@ -183,6 +189,7 @@ def pass_manager(self, pass_manager_config, optimization_level=None) -> PassMana
target,
coupling_map=coupling_map,
vf2_call_limit=vf2_call_limit,
vf2_max_trials=vf2_max_trials,
backend_properties=backend_properties,
seed_transpiler=seed_transpiler,
use_barrier_before_measurement=True,
Expand All @@ -194,6 +201,7 @@ def pass_manager(self, pass_manager_config, optimization_level=None) -> PassMana
target,
coupling_map=coupling_map,
vf2_call_limit=vf2_call_limit,
vf2_max_trials=vf2_max_trials,
backend_properties=backend_properties,
seed_transpiler=seed_transpiler,
use_barrier_before_measurement=True,
Expand All @@ -213,7 +221,7 @@ def pass_manager(self, pass_manager_config, optimization_level=None) -> PassMana
if coupling_map_routing is None:
coupling_map_routing = coupling_map
backend_properties = pass_manager_config.backend_properties
vf2_call_limit = common.get_vf2_call_limit(
vf2_call_limit, vf2_max_trials = common.get_vf2_limits(
optimization_level,
pass_manager_config.layout_method,
pass_manager_config.initial_layout,
Expand Down Expand Up @@ -244,6 +252,7 @@ def pass_manager(self, pass_manager_config, optimization_level=None) -> PassMana
target,
coupling_map,
vf2_call_limit=vf2_call_limit,
vf2_max_trials=vf2_max_trials,
backend_properties=backend_properties,
seed_transpiler=seed_transpiler,
check_trivial=True,
Expand All @@ -261,6 +270,7 @@ def pass_manager(self, pass_manager_config, optimization_level=None) -> PassMana
target,
coupling_map=coupling_map,
vf2_call_limit=vf2_call_limit,
vf2_max_trials=vf2_max_trials,
backend_properties=backend_properties,
seed_transpiler=seed_transpiler,
use_barrier_before_measurement=True,
Expand All @@ -277,6 +287,7 @@ def pass_manager(self, pass_manager_config, optimization_level=None) -> PassMana
target,
coupling_map=coupling_map,
vf2_call_limit=vf2_call_limit,
vf2_max_trials=vf2_max_trials,
backend_properties=backend_properties,
seed_transpiler=seed_transpiler,
use_barrier_before_measurement=True,
Expand Down
50 changes: 48 additions & 2 deletions qiskit/transpiler/preset_passmanagers/common.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,6 +51,7 @@
from qiskit.transpiler.passes.layout.vf2_post_layout import VF2PostLayoutStopReason
from qiskit.transpiler.exceptions import TranspilerError
from qiskit.transpiler.layout import Layout
from qiskit.utils.deprecation import deprecate_func

_CONTROL_FLOW_OP_NAMES = {"for_loop", "if_else", "while_loop", "switch_case"}

Expand Down Expand Up @@ -263,6 +264,7 @@ def generate_routing_passmanager(
seed_transpiler=None,
check_trivial=False,
use_barrier_before_measurement=True,
vf2_max_trials=None,
):
"""Generate a routing :class:`~qiskit.transpiler.PassManager`

Expand All @@ -273,7 +275,8 @@ def generate_routing_passmanager(
coupling_map (CouplingMap): The coupling map of the backend to route
for
vf2_call_limit (int): The internal call limit for the vf2 post layout
pass. If this is ``None`` the vf2 post layout will not be run.
pass. If this is ``None`` or ``0`` the vf2 post layout will not be
run.
backend_properties (BackendProperties): Properties of a backend to
synthesize for (e.g. gate fidelities).
seed_transpiler (int): Sets random seed for the stochastic parts of
Expand All @@ -287,6 +290,9 @@ def generate_routing_passmanager(
use_barrier_before_measurement (bool): If true (the default) the
:class:`~.BarrierBeforeFinalMeasurements` transpiler pass will be run prior to the
specified pass in the ``routing_pass`` argument.
vf2_max_trials (int): The maximum number of trials to run VF2 when
evaluating the vf2 post layout
pass. If this is ``None`` or ``0`` the vf2 post layout will not be run.
Returns:
PassManager: The routing pass manager
"""
Expand Down Expand Up @@ -314,14 +320,16 @@ def _swap_condition(property_set):
else:
routing.append([routing_pass], condition=_swap_condition)

if (target is not None or backend_properties is not None) and vf2_call_limit is not None:
is_vf2_fully_bounded = vf2_call_limit and vf2_max_trials
if (target is not None or backend_properties is not None) and is_vf2_fully_bounded:
routing.append(
VF2PostLayout(
target,
coupling_map,
backend_properties,
seed_transpiler,
call_limit=vf2_call_limit,
max_trials=vf2_max_trials,
strict_direction=False,
),
condition=_run_post_layout_condition,
Expand Down Expand Up @@ -539,6 +547,10 @@ def _require_alignment(property_set):
return scheduling


@deprecate_func(
additional_msg="Instead, use ``get_vf2_limits``.",
since="0.24.0",
)
Copy link
Member

Choose a reason for hiding this comment

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

This is a bit too quick, we can't deprecate it until after we're included the alternative in the release, see: https://qiskit.org/documentation/deprecation_policy.html#removing-a-feature so can you remove the decorator and the release note for this. We can do this for 0.25.0, but not for this PR which we'll need to backport to 0.24.0 (alternatively you can set pending=True on the decorator and remove the release not to downgrade the warning emitted to PendingDeprecationWarning)

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Ah that's right, thanks for the reminder. I removed the decorator as well as the release note in 5ef7a94.

def get_vf2_call_limit(
optimization_level: int,
layout_method: Optional[str] = None,
Expand All @@ -554,3 +566,37 @@ def get_vf2_call_limit(
elif optimization_level == 3:
vf2_call_limit = int(3e7) # Set call limit to ~60 sec with rustworkx 0.10.2
return vf2_call_limit


VF2Limits = collections.namedtuple("VF2Limits", ("call_limit", "max_trials"))


def get_vf2_limits(
optimization_level: int,
layout_method: Optional[str] = None,
initial_layout: Optional[Layout] = None,
) -> VF2Limits:
"""Get the VF2 limits for VF2-based layout passes.

Returns:
VF2Limits: An namedtuple with optional elements
``call_limit`` and ``max_trials``.
"""
limits = VF2Limits(None, None)
if layout_method is None and initial_layout is None:
if optimization_level == 1:
limits = VF2Limits(
int(5e4), # Set call limit to ~100ms with rustworkx 0.10.2
2500, # Limits layout scoring to < 600ms on ~400 qubit devices
)
elif optimization_level == 2:
limits = VF2Limits(
int(5e6), # Set call limit to ~10 sec with rustworkx 0.10.2
25000, # Limits layout scoring to < 6 sec on ~400 qubit devices
)
elif optimization_level == 3:
limits = VF2Limits(
int(3e7), # Set call limit to ~60 sec with rustworkx 0.10.2
250000, # Limits layout scoring to < 60 sec on ~400 qubit devices
Copy link
Contributor

@jlapeyre jlapeyre May 2, 2023 •
edited
Loading

Choose a reason for hiding this comment

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

Removing the need to count zeros is good. Maybe 25000 is near the readability threshold, but I'd still opt for making it more readable. Moreso for 250000. Some options are

int(2.5e5)
250_000

int(2.5e4)
25_000

It's not obvious to me whether the LHS or RHS of 30_000_000 == int(3e7) is more readable. Some researchers measure comprehension speed with clocks, but I couldn't find a study on this specific case. I guess in this file it makes sense to use one of either underscores or int uniformly.

Copy link
Member

@mtreinish mtreinish May 2, 2023 •
edited
Loading

Choose a reason for hiding this comment

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

In this case I personally don't view it as a big deal, the number here is just an arbitrary limit that was found by experimentation, the actual numeric value isn't really relevant. My personal preference would be for 2.5e5 as that is how I reason about this in my head because it's more about the order of magnitude between each optimization level. But I can personally see that just as well with how it's written now (not so much by the absolute zero count but by the relative spacing differences). What I view as far more important here is the comment explaining where the value came from because that's what we're really setting is a wall time limit as a second order effect.

)
return limits
1 change: 1 addition & 0 deletions qiskit/transpiler/preset_passmanagers/level1.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -133,6 +133,7 @@ def _vf2_match_not_found(property_set):
call_limit=int(5e4), # Set call limit to ~100ms with rustworkx 0.10.2
properties=backend_properties,
target=target,
max_trials=2500, # Limits layout scoring to < 600ms on ~400 qubit devices
)
)

Expand Down
1 change: 1 addition & 0 deletions qiskit/transpiler/preset_passmanagers/level2.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -118,6 +118,7 @@ def _vf2_match_not_found(property_set):
call_limit=int(5e6), # Set call limit to ~10 sec with rustworkx 0.10.2
properties=backend_properties,
target=target,
max_trials=25000, # Limits layout scoring to < 6 sec on ~400 qubit devices
)
)

Expand Down
1 change: 1 addition & 0 deletions qiskit/transpiler/preset_passmanagers/level3.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -124,6 +124,7 @@ def _vf2_match_not_found(property_set):
call_limit=int(3e7), # Set call limit to ~60 sec with rustworkx 0.10.2
properties=backend_properties,
target=target,
max_trials=250000, # Limits layout scoring to < 60 sec on ~400 qubit devices
)
)

Expand Down
30 changes: 30 additions & 0 deletions releasenotes/notes/preset-pm-vf2-max-trials-958bb8a36fff472f.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,30 @@
---
upgrade:
- |
The maximum number of trials evaluated when searching for the best
layout using :class:`.VF2Layout` and :class:`.VF2PostLayout` is now
limited in
:func:`qiskit.transpiler.preset_passmanagers.level_1_pass_manager`,
:func:`qiskit.transpiler.preset_passmanagers.level_2_pass_manager`,
and
:func:`qiskit.transpiler.preset_passmanagers.level_3_pass_manager`
to ``2,500``, ``25,000``, and ``250,000``, respectively. Previously,
all possible layouts were evaluated. This change was made to prevent
transpilation from hanging during layout scoring for circuits with many
connected components on larger devices, which scales combinatorially
since each connected component must be evaluated in all possible
positions on the device. To perform a full search as
before, manually run :class:`.VF2PostLayout` over the transpiled circuit
in strict mode, specifying ``0`` for ``max_trials``.
deprecations:
- |
The method
:func:`~qiskit.transpiler.preset_passmanagers.common.get_vf2_call_limit`
is deprecated. Instead, use replacement method
:func:`~qiskit.transpiler.preset_passmanagers.common.get_vf2_limits` and
access the ``call_limit`` field of the returned namedtuple.
fixes:
- |
Fixed a potential performance scaling issue with layout scoring in preset
pass managers, which could occur when transpiling circuits with many
connected components on large devices.