Add flamegraph visualizations to visualize T-complexity of algorithms #732

tanujkhattar · 2024-03-02T17:21:14Z

This PR adds tools to generate flamegraphs using then https://github.com/brendangregg/FlameGraph/blob/master/flamegraph.pl perl script. There are a couple of open questions

What's the best way to invoke the perl script from Qualtran? Right now, I just copied the flamegraph.pl file into qualtran/drawing directory but this is probably not the right way? We also need to figure out if using the tool causes any licensing issues. cc @mpharrigan
The flame graphs added in this PR are only visualizing T-complexity traces (see plots below). The one in the THC paper (Figure 12) have number of qubits used on the y-axis. It's not clear to me how to generate these plots using the existing tools.

This PR also makes a BUNCH of improvements and bugfixes to cirq interop in order to make things work. We'll probably have to pull out pieces into separate PRs or have more detailed discussions. Some improvements made in this PR are

_extract_bloq_from_op is now smarter to automatically map known operations from Cirq to Bloqs. This includes mapping ControlledGate to Controlled Bloq (this has some issues, so it's currently commented out. but the idea is to support this as well); _InverseCompositeGate to Adjoint and cirq gates corresponding to all bloqs in basic_gates/
_t_complexity_ protocol is broken after the recent unification and introduction of Adjoint. Specifically, Adjoint passes adjoint=True parameter to subbloq but there's only one subbloq (MultiAnd) that accepts this parameter. Thus, using Adjoint(bloq).t_complexity() for any other bloq results in an error because the subbloq._t_complexity_() does not accept an adjoint parameter. It looks like this was merged as an idea but never followed up to add the parameter to all other Bloqs? The temporary fix in this PR is to comment out then _t_complexity_ of MultiAnd and rely on decomposition strategy instead.
Add pretty_name to a bunch of subbloqs.

Some Flume Graph plots for THC (bloq native) and Hubbard model (cirq style native) are as follows:

Hubbard model Qubitization

THC

…repare as well

fdmalone · 2024-03-02T22:05:42Z

What's the best way to invoke the perl script from Qualtran? Right now, I just copied the flamegraph.pl file into qualtran/drawing directory but this is probably not the right way? We also need to figure out if using the tool causes any licensing issues. cc @mpharrigan

I imagine we probably want to keep it separate and just point users to the external tool?

The flame graphs added in this PR are only visualizing T-complexity traces (see plots below). The one in the THC paper (Figure 12) have number of qubits used on the y-axis. It's not clear to me how to generate these plots using the existing tools.

I think we need to find the mathematica notebook / whatever was used and maybe just write something ourselves? Not sure how hard it would be.

…ame_graph and remove special casing for Adjoint

…ame_graph

…into flame_graph

…eir Bloq versions

…rq_to_bloq_interop_improve

…ame_graph

qualtran/cirq_interop/_cirq_to_bloq.py

qualtran/drawing/flame_graph.py

tanujkhattar · 2024-04-01T18:07:04Z

I'll clean it and update it soon

…ame_graph

…temp file

tanujkhattar · 2024-04-02T00:51:16Z

@mpharrigan I've cleaned up the PR and addressed most of the pending comments. This is ready for a re-review.

…ame_graph

mpharrigan · 2024-04-03T19:02:53Z

qualtran/drawing/flame_graph.py

+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+
+"""Classes for drawing bloqs with FlameGraph."""


"this relies on third party flamegraph.pl

mpharrigan · 2024-04-03T19:05:02Z

qualtran/drawing/flame_graph.py

+def _pretty_name(bloq: Bloq) -> str:
+    from qualtran.serialization.bloq import _iter_fields
+
+    ret = bloq.pretty_name()
+    if bloq.pretty_name.__qualname__.startswith('Bloq.'):
+        for field in _iter_fields(bloq):
+            ret += f'[{_pretty_arg(getattr(bloq, field.name))}]'
+    return ret


this seems like a hack around the issues discussed in #791. Can you add a description of what you'd like in the flame graph and link to the issue here. Is it the same thing we'd want in the call graph drawings?

Left a detailed comment on the linked issue

mpharrigan · 2024-04-03T19:05:43Z

qualtran/drawing/flame_graph.py

+    return ret
+
+
+@functools.lru_cache(maxsize=None)


i'm surprised graph is hashable since it is mutable. could be dangerous

mpharrigan · 2024-04-03T19:05:53Z

qualtran/drawing/flame_graph.py

+    sigma = _compute_sigma(bloq, graph)
+    return t_counts_from_sigma(sigma)


why not t_complexity(bloq).t?

It doesn't sum up costs for rotation bloqs anymore, since we moved ceil(1.149 * log2(1.0 / eps) + 9.2) to TComplexity.rotation_cost and but we don't track the eps in the TComplexity object.

mpharrigan · 2024-04-03T19:07:20Z

qualtran/drawing/flame_graph.py

+
+
+def _populate_flame_graph_data(
+    bloq: Bloq, graph: nx.DiGraph, graph_t: nx.DiGraph, prefix: List[str]


document arguments -- prefix is modified

mpharrigan · 2024-04-03T19:08:01Z

qualtran/drawing/flame_graph.py

+                _t_counts_for_bloq(succ, graph_t) * graph.get_edge_data(bloq, succ)['n']
+            )
+            data += curr_data * graph.get_edge_data(bloq, succ)['n']
+        # assert total_t_counts == succ_t_counts, f'{bloq=}, {total_t_counts=}, {succ_t_counts=}'


uncomment or remove

mpharrigan · 2024-04-03T19:08:21Z

qualtran/drawing/flame_graph.py

+    file_path: Union[None, pathlib.Path, str] = None,
+    keep: Optional[Sequence['Bloq']] = _keep_if_small,
+    **kwargs,
+) -> List[str]:


mpharrigan · 2024-04-03T19:59:33Z

qualtran/drawing/flame_graph.py

+
+
+def get_flame_graph_svg_data(
+    *bloqs: Bloq, file_path: Union[None, pathlib.Path, str] = None, **kwargs


mpharrigan · 2024-04-03T20:00:27Z

qualtran/drawing/flame_graph.py

+    data = get_flame_graph_data(*bloqs, **kwargs)
+
+    data_file = tempfile.NamedTemporaryFile(mode='w')
+    flame_graph_path = pathlib.Path(__file__).resolve().parent.parent / "third_party/flamegraph.pl"


I'm pretty sure this won't work if you pip install qualtran.

Do you have suggestions for what to change? can we fix this in a followup PR?

…ame_graph

…me_graph script

tanujkhattar · 2024-04-03T22:25:47Z

@mpharrigan I've addressed all comments, PTAL

tanujkhattar added 5 commits March 2, 2024 11:10

Add flamegraph visualizations to visualize T-complexity of algorithms

01c1803

Add flamegraph.pl script

967ac5d

Fix pylint and failing tests

c3597ed

Merge branch 'main' into flame_graph

358883a

Use qubitization of quantum walks infrastructure for THC Select and P…

f064ca8

…repare as well

This was referenced Mar 4, 2024

Future and upgrades to T-Complexity protocol #735

Closed

Temporarily fix the failing T-Complexity for Adjoint #736

Merged

tanujkhattar added 3 commits March 4, 2024 14:11

Merge branch 'main' of https://github.com/quantumlib/Qualtran into fl…

32c32ab

…ame_graph and remove special casing for Adjoint

Merge branch 'main' of https://github.com/quantumlib/Qualtran into fl…

e69ff29

…ame_graph

Fix pylint error

36d49ce

tanujkhattar added 9 commits March 4, 2024 17:59

Merge branch 'main' of https://github.com/quantumlib/Qualtran into fl…

9de89da

…ame_graph

Merge commit

9e82868

Remove redundant _t_complexity_ override for THCRotations Bloq

d1e1cdd

Merge branch 'main' into flame_graph

8a54976

Add back space to _bloq_to_cirq.py

eda30a9

Merge branch 'flame_graph' of https://github.com/tanujkhattar/Qualtran …

f7f6f7d

…into flame_graph

Fix infinite recursion bug in t_complexity protocol

0ae4fe9

Fix pylint

24bc6d0

Fix cirq to bloq interop to automatically replace basic gates with th…

b7a4fd2

…eir Bloq versions

tanujkhattar mentioned this pull request Mar 5, 2024

Update cirq to bloq interop to automatically replace basic gates with their Bloq versions #746

Merged

tanujkhattar added 4 commits March 5, 2024 12:53

Fix failing tests by updating T-complexity of CZPowGate

d68b0a1

Merge branch 'main' of https://github.com/quantumlib/Qualtran into ci…

a1e45f0

…rq_to_bloq_interop_improve

Merge branch 'main' of https://github.com/quantumlib/Qualtran into fl…

eb9d518

…ame_graph

Merge branch 'cirq_to_bloq_interop_improve' into flame_graph

149941f

mpharrigan reviewed Mar 5, 2024

View reviewed changes

qualtran/cirq_interop/_cirq_to_bloq.py Outdated Show resolved Hide resolved

mpharrigan reviewed Mar 5, 2024

View reviewed changes

qualtran/drawing/flame_graph.py Outdated Show resolved Hide resolved

tanujkhattar added 7 commits April 1, 2024 12:00

Merge branch 'main' of https://github.com/quantumlib/Qualtran into fl…

e054aeb

…ame_graph

Fix merge conflicts

f0a5198

Move flame graph script to third_party and use subprocess.run with a …

dafe2f8

…temp file

Restore bloqs to main

7eb49db

Fix formatting

4d5980a

Revert more unrelated changes, make improvements to the flame graph code

fd01889

Add tests for flame_graph.py

6ec8a5f

tanujkhattar assigned mpharrigan Apr 2, 2024

tanujkhattar added 2 commits April 1, 2024 17:57

Use sorted instead of set to compared unordered lists

72a1f27

Merge branch 'main' of https://github.com/quantumlib/Qualtran into fl…

3f5d26e

…ame_graph

tanujkhattar mentioned this pull request Apr 2, 2024

bloq.call_graph() should preserve ordering of successors across calls / build_call_graph should return a Dict instead of Set #845

Closed

mpharrigan reviewed Apr 3, 2024

View reviewed changes

tanujkhattar mentioned this pull request Apr 3, 2024

How bloqs choose how to display themselves #791

Closed

tanujkhattar added 2 commits April 3, 2024 14:47

Merge branch 'main' of https://github.com/quantumlib/Qualtran into fl…

f932674

…ame_graph

Add lots of docstrings and a LICENSE and METADATA for third_party fla…

0b844c6

…me_graph script

mpharrigan approved these changes Apr 3, 2024

View reviewed changes

tanujkhattar enabled auto-merge (squash) April 3, 2024 23:17

Merge branch 'main' into flame_graph

6021909

tanujkhattar merged commit d102606 into quantumlib:main Apr 3, 2024
6 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add flamegraph visualizations to visualize T-complexity of algorithms #732

Add flamegraph visualizations to visualize T-complexity of algorithms #732

tanujkhattar commented Mar 2, 2024 •

edited

Loading

fdmalone commented Mar 2, 2024

tanujkhattar commented Apr 1, 2024

tanujkhattar commented Apr 2, 2024

mpharrigan Apr 3, 2024

mpharrigan Apr 3, 2024

tanujkhattar Apr 3, 2024

mpharrigan Apr 3, 2024

mpharrigan Apr 3, 2024

tanujkhattar Apr 3, 2024

mpharrigan Apr 3, 2024

mpharrigan Apr 3, 2024

mpharrigan Apr 3, 2024

mpharrigan Apr 3, 2024

mpharrigan Apr 3, 2024

tanujkhattar Apr 3, 2024

tanujkhattar commented Apr 3, 2024

		sigma = _compute_sigma(bloq, graph)
		return t_counts_from_sigma(sigma)



		def _populate_flame_graph_data(
		bloq: Bloq, graph: nx.DiGraph, graph_t: nx.DiGraph, prefix: List[str]



		def get_flame_graph_svg_data(
		bloqs: Bloq, file_path: Union[None, pathlib.Path, str] = None, *kwargs

Add flamegraph visualizations to visualize T-complexity of algorithms #732

Add flamegraph visualizations to visualize T-complexity of algorithms #732

Conversation

tanujkhattar commented Mar 2, 2024 • edited Loading

Hubbard model Qubitization

THC

fdmalone commented Mar 2, 2024

tanujkhattar commented Apr 1, 2024

tanujkhattar commented Apr 2, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

tanujkhattar commented Apr 3, 2024

tanujkhattar commented Mar 2, 2024 •

edited

Loading