feat: Add DrawGraph tool for graph visualization #7172

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

Merged

cccclai merged 1 commit into pytorch:main from Nick-Wei:draw_graph_util

Dec 6, 2024

Contributor

Nick-Wei commented Dec 4, 2024

Implemented the DrawGraph tool to visualize graphs.
Added test cases to validate the correctness of the generated graphs.

pytorch-bot bot commented Dec 4, 2024 •

edited

Loading

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/7172

📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 1 New Failure

As of commit 938d8f2 with merge base 8861b9a ():

NEW FAILURE - The following job has failed:

pull / unittest-arm / linux-job (gh)
RuntimeError: Command docker exec -t 83c41ba4f094be30cb613fa9c3cbe0f59f03168483594426fc47c376699ddcdf /exec failed with exit code 1

This comment was automatically generated by Dr. CI and updates every 15 minutes.

facebook-github-bot added the CLA Signed label

Contributor Author

Nick-Wei commented Dec 4, 2024

Hi @haowhsu-quic , this PR enables to draw graphs in the C++ backend. During preprocessing, we can obtain the op wrapper list, which can then be used to generate corresponding graphs. Hopefully we could have this merged, thanks.
Below are the generated graphs and the excel file containing scale data and offset data.

dbort requested review from cccclai and haowhsu-quic

December 4, 2024 21:57

dbort added the partner: qualcomm label

haowhsu-quic reviewed

View reviewed changes

backends/qualcomm/aot/python/PyQnnWrapperAdaptor.cpp Outdated

+                    .def(
+                        "GetInputTensors",
+                        &OpWrapper::GetInputTensors,
+                        "A function which get input tensors")

Collaborator

haowhsu-quic Dec 5, 2024

gets. Ditto for followings.

haowhsu-quic reviewed

View reviewed changes

backends/qualcomm/aot/python/PyQnnWrapperAdaptor.cpp Outdated

+                    );
+                py::enum_<Qnn_Definition_t>(m, "QnnDefinition")
+                    .value("IMPL_GENERATED", Qnn_Definition_t::QNN_DEFINITION_IMPL_GENERATED)

Collaborator

haowhsu-quic Dec 5, 2024

~~"IMPL_GENERATED"~~ -> "QNN_DEFINITION_IMPL_GENERATED" to align the style. Ditto for followings.

haowhsu-quic reviewed

View reviewed changes

backends/qualcomm/tests/models.py Outdated

@@ @@ -1079,3 +1079,19 @@ def forward(self, x, y): @@
                       x = x.view(new_shape)
                       x = x.permute(0, 2, 1, 3)
                       return torch.matmul(x, y.transpose(-1, -2))
+              class draw_graph_model(torch.nn.Module):

Collaborator

haowhsu-quic Dec 5, 2024

Camel form on class name would be better, please also put this in alphabetical order.

haowhsu-quic reviewed

View reviewed changes

backends/qualcomm/utils/utils.py Outdated

                   for node in gm.graph.nodes:
                       if dtype := get_quant_io_dtype_fn(node):
                           node.meta[QCOM_QUANTIZED_IO] = dtype
+              class DrawGraph:

Collaborator

haowhsu-quic Dec 5, 2024

Could you put class definition to the below of _AnnotationSkipper? thanks.

haowhsu-quic reviewed

View reviewed changes

backends/qualcomm/utils/utils.py Outdated

+                                              if input_node_name not in node_list:
+                                                  node_list[input_node_name] = {"node" : input_node, "input_list" : []}
+                                              input_list.append(input_node_name)
+                                              # ToDo: tensor v2

Collaborator

haowhsu-quic Dec 5, 2024

#TODO

haowhsu-quic reviewed

View reviewed changes

backends/qualcomm/utils/utils.py Outdated

+                                              input_list.append(input_node_name)
+                                              # ToDo: tensor v2
+                                          elif (op_wrapper.GetOpConfig()["outputTensors"][j].version == 2):
+                                              pass

Collaborator

haowhsu-quic Dec 5, 2024

Could we raise an exception here to prevent drawing the wrong graph?

haowhsu-quic reviewed

View reviewed changes

backends/qualcomm/utils/utils.py Outdated

+                                      node_list[node_name]["input_list"] = input_list
+                              # TODO: tensor v2
+                              elif (op_wrapper.GetOpConfig()["outputTensors"][i].version == 2):
+                                  pass

Collaborator

haowhsu-quic Dec 5, 2024

Ditto.

haowhsu-quic suggested changes

View reviewed changes

Collaborator

haowhsu-quic left a comment

Thank you for the contribution, I think this is definitely a great tool for us to debug real graph in QNN.

haowhsu-quic reviewed

View reviewed changes

backends/qualcomm/utils/utils.py Outdated

+                                  node_name = node.name
+                                  input_list = []
+                                  for j in range(op_wrapper.GetOpConfig()["numOfInputs"]):
+                                          if(op_wrapper.GetOpConfig()["inputTensors"][j].version == 1):

Collaborator

haowhsu-quic Dec 5, 2024

Looks like wrong indentation here.

Contributor Author

Nick-Wei commented Dec 5, 2024

@pytorchbot label "feature"

pytorch-bot bot added the feature label

haowhsu-quic reviewed

View reviewed changes

backends/qualcomm/utils/utils.py Outdated

@@ @@ -93,6 +93,12 @@ @@
               from torch.fx import passes
               from torch.fx.passes.operator_support import OperatorSupportBase
               from torch.library import Library
+              from graphviz import Digraph

Collaborator

haowhsu-quic Dec 5, 2024

Could you create a debugger folder under backends/qualcomm and move this implementation to there. Then we could get rid of dependency of python package in utils.py.


          feat: Add DrawGraph tool for graph visualization

938d8f2

- Implemented the DrawGraph tool to visualize graphs.
- Added test cases to validate the correctness of the generated graphs.

Nick-Wei force-pushed the draw_graph_util branch from eb5c334 to 938d8f2 Compare

December 5, 2024 04:24

Contributor Author

Nick-Wei commented Dec 5, 2024

@pytorchbot label "feature"

Contributor Author

Nick-Wei commented Dec 5, 2024

@haowhsu-quic The above modifications have been completed. Please help confirm, thanks.

haowhsu-quic approved these changes

View reviewed changes

Collaborator

haowhsu-quic left a comment

Looks good to me, thank you for contribution!
Hi @cccclai, it would be great to have this & #6591, mind taking a look? Thank you.

Contributor Author

Nick-Wei commented Dec 5, 2024

Thank you for taking the time to review it.

Contributor

facebook-github-bot commented Dec 5, 2024

@cccclai has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

cccclai added the release notes: backends [DO NOT USE] label

cccclai approved these changes

View reviewed changes

Contributor

cccclai left a comment

Thank you for contributing!

cccclai merged commit be0bd75 into pytorch:main

41 of 44 checks passed

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

CLA Signed partner: qualcomm release notes: backends [DO NOT USE]